Intel Architecture

Software Developer’s
Manual
Volume 3:
System Programming

NOTE: The Intel Architecture Software Developer’s Manual consists of

three volumes: Basic Architecture, Order Number 243190; Instruction Set
Reference, Order Number 243191; and the System Programming Guide,
Order Number 243192.
Please refer to all three volumes when evaluating your design needs.

1999
Information in this document is provided in connection with Intel products. No license, express or implied, by estoppel
or otherwise, to any intellectual property rights is granted by this document. Except as provided in Intel’s Terms and
Conditions of Sale for such products, Intel assumes no liability whatsoever, and Intel disclaims any express or implied
warranty, relating to sale and/or use of Intel products including liability or warranties relating to fitness for a particular
purpose, merchantability, or infringement of any patent, copyright or other intellectual property right. Intel products are
not intended for use in medical, life saving, or life sustaining applications.

Intel may make changes to specifications and product descriptions at any time, without notice.

Designers must not rely on the absence or characteristics of any features or instructions marked “reserved” or
“undefined.” Intel reserves these for future definition and shall have no responsibility whatsoever for conflicts or
incompatibilities arising from future changes to them.

Intel’s Intel Architecture processors (e.g., Pentium®, Pentium® II, Pentium® III, and Pentium® Pro processors) may
contain design defects or errors known as errata which may cause the product to deviate from published
specifications. Current characterized errata are available on request.

Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your
product order.

Copies of documents which have an ordering number and are referenced in this document, or other Intel literature,
may be obtained by calling 1-800-548-4725, or by visiting Intel's literature center at http://www.intel.com.

COPYRIGHT © INTEL CORPORATION 1999

*THIRD-PARTY BRANDS AND NAMES ARE THE PROPERTY OF THEIR RESPECTIVE OWNERS.
 TABLE OF CONTENTS

CHAPTER 1
ABOUT THIS MANUAL
1.1. P6 FAMILY PROCESSOR TERMINOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1
1.2. OVERVIEW OF THE INTEL ARCHITECTURE SOFTWARE DEVELOPER’S MANUAL,
VOLUME 3: SYSTEM PROGRAMMING GUIDE . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1
1.3. OVERVIEW OF THE INTEL ARCHITECTURE SOFTWARE DEVELOPER’S MANUAL,
VOLUME 1: BASIC ARCHITECTURE 1-3
1.4. OVERVIEW OF THE INTEL ARCHITECTURE SOFTWARE DEVELOPER’S MANUAL,
VOLUME 2: INSTRUCTION SET REFERENCE 1-5
1.5. NOTATIONAL CONVENTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-5
1.5.1. Bit and Byte Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-6
1.5.2. Reserved Bits and Software Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-6
1.5.3. Instruction Operands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-7
1.5.4. Hexadecimal and Binary Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-7
1.5.5. Segmented Addressing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-7
1.5.6. Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-8
1.6. RELATED LITERATURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-9

CHAPTER 2
SYSTEM ARCHITECTURE OVERVIEW
2.1. OVERVIEW OF THE SYSTEM-LEVEL ARCHITECTURE . . . . . . . . . . . . . . . . . . . 2-1
2.1.1. Global and Local Descriptor Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-3
2.1.2. System Segments, Segment Descriptors, and Gates . . . . . . . . . . . . . . . . . . . . . .2-3
2.1.3. Task-State Segments and Task Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-4
2.1.4. Interrupt and Exception Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-4
2.1.5. Memory Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-5
2.1.6. System Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-5
2.1.7. Other System Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-6
2.2. MODES OF OPERATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6
2.3. SYSTEM FLAGS AND FIELDS IN THE EFLAGS REGISTER . . . . . . . . . . . . . . . . 2-8
2.4. MEMORY-MANAGEMENT REGISTERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-10
2.4.1. Global Descriptor Table Register (GDTR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-10
2.4.2. Local Descriptor Table Register (LDTR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-11
2.4.3. IDTR Interrupt Descriptor Table Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-11
2.4.4. Task Register (TR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-11
2.5. CONTROL REGISTERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-12
2.5.1. CPUID Qualification of Control Register Flags . . . . . . . . . . . . . . . . . . . . . . . . . .2-18
2.6. SYSTEM INSTRUCTION SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-18
2.6.1. Loading and Storing System Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-20
2.6.2. Verifying of Access Privileges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-20
2.6.3. Loading and Storing Debug Registers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-21
2.6.4. Invalidating Caches and TLBs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-21
2.6.5. Controlling the Processor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-22
2.6.6. Reading Performance-Monitoring and Time-Stamp Counters . . . . . . . . . . . . . .2-22
2.6.7. Reading and Writing Model-Specific Registers . . . . . . . . . . . . . . . . . . . . . . . . . .2-23
2.6.8. Loading and Storing the Streaming SIMD Extensions Control/Status Word . . . .2-23

iii
TABLE OF CONTENTS

CHAPTER 3
PROTECTED-MODE MEMORY MANAGEMENT
3.1. MEMORY MANAGEMENT OVERVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
3.2. USING SEGMENTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-3
3.2.1. Basic Flat Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-3
3.2.2. Protected Flat Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-4
3.2.3. Multisegment Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-5
3.2.4. Paging and Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-6
3.3. PHYSICAL ADDRESS SPACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-6
3.4. LOGICAL AND LINEAR ADDRESSES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-6
3.4.1. Segment Selectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-7
3.4.2. Segment Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-8
3.4.3. Segment Descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-9
3.4.3.1. Code- and Data-Segment Descriptor Types. . . . . . . . . . . . . . . . . . . . . . . . . .3-13
3.5. SYSTEM DESCRIPTOR TYPES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-15
3.5.1. Segment Descriptor Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-16
3.6. PAGING (VIRTUAL MEMORY) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-18
3.6.1. Paging Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-19
3.6.2. Page Tables and Directories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-20
3.6.2.1. Linear Address Translation (4-KByte Pages) . . . . . . . . . . . . . . . . . . . . . . . . .3-20
3.6.2.2. Linear Address Translation (4-MByte Pages). . . . . . . . . . . . . . . . . . . . . . . . .3-21
3.6.2.3. Mixing 4-KByte and 4-MByte Pages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-22
3.6.3. Base Address of the Page Directory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-23
3.6.4. Page-Directory and Page-Table Entries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-23
3.6.5. Not Present Page-Directory and Page-Table Entries . . . . . . . . . . . . . . . . . . . . .3-28
3.7. TRANSLATION LOOKASIDE BUFFERS (TLBS) . . . . . . . . . . . . . . . . . . . . . . . . . 3-28
3.8. PHYSICAL ADDRESS EXTENSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-29
3.8.1. Linear Address Translation With Extended
Addressing Enabled (4-KByte Pages) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-30
3.8.2. Linear Address Translation With Extended Addressing Enabled
(2-MByte or 4-MByte Pages) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-32
3.8.3. Accessing the Full Extended Physical Address Space With the
Extended Page-Table Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-32
3.8.4. Page-Directory and Page-Table Entries With Extended Addressing Enabled . .3-33
3.9. 36-BIT PAGE SIZE EXTENSION (PSE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-35
3.9.1. Description of the 36-bit PSE Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-36
3.9.2. Fault Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-39
3.10. MAPPING SEGMENTS TO PAGES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-40

CHAPTER 4
PROTECTION
4.1. ENABLING AND DISABLING SEGMENT AND PAGE PROTECTION . . . . . . . . . . 4-2
4.2. FIELDS AND FLAGS USED FOR SEGMENT-LEVEL AND
PAGE-LEVEL PROTECTION 4-2
4.3. LIMIT CHECKING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
4.4. TYPE CHECKING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
4.4.1. Null Segment Selector Checking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-7
4.5. PRIVILEGE LEVELS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8
4.6. PRIVILEGE LEVEL CHECKING WHEN ACCESSING DATA SEGMENTS . . . . . . 4-9
4.6.1. Accessing Data in Code Segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-12
4.7. PRIVILEGE LEVEL CHECKING WHEN LOADING THE SS REGISTER . . . . . . . 4-12

iv
 TABLE OF CONTENTS

4.8. PRIVILEGE LEVEL CHECKING WHEN TRANSFERRING PROGRAM CONTROL

BETWEEN CODE SEGMENTS 4-12
4.8.1. Direct Calls or Jumps to Code Segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-13
4.8.1.1. Accessing Nonconforming Code Segments . . . . . . . . . . . . . . . . . . . . . . . . . 4-14
4.8.1.2. Accessing Conforming Code Segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-15
4.8.2. Gate Descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16
4.8.3. Call Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16
4.8.4. Accessing a Code Segment Through a Call Gate . . . . . . . . . . . . . . . . . . . . . . . 4-17
4.8.5. Stack Switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-21
4.8.6. Returning from a Called Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-23
4.9. PRIVILEGED INSTRUCTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-25
4.10. POINTER VALIDATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-25
4.10.1. Checking Access Rights (LAR Instruction) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-26
4.10.2. Checking Read/Write Rights (VERR and VERW Instructions) . . . . . . . . . . . . . 4-27
4.10.3. Checking That the Pointer Offset Is Within Limits (LSL Instruction) . . . . . . . . . 4-28
4.10.4. Checking Caller Access Privileges (ARPL Instruction) . . . . . . . . . . . . . . . . . . . 4-28
4.10.5. Checking Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-30
4.11. PAGE-LEVEL PROTECTION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-30
4.11.1. Page-Protection Flags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-31
4.11.2. Restricting Addressable Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-31
4.11.3. Page Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-32
4.11.4. Combining Protection of Both Levels of Page Tables . . . . . . . . . . . . . . . . . . . . 4-32
4.11.5. Overrides to Page Protection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-32
4.12. COMBINING PAGE AND SEGMENT PROTECTION . . . . . . . . . . . . . . . . . . . . . . 4-33

CHAPTER 5
INTERRUPT AND EXCEPTION HANDLING
5.1. INTERRUPT AND EXCEPTION OVERVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
5.1.1. Sources of Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
5.1.1.1. External Interrupts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
5.1.1.2. Maskable Hardware Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
5.1.1.3. Software-Generated Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
5.1.2. Sources of Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
5.1.2.1. Program-Error Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
5.1.2.2. Software-Generated Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
5.1.2.3. Machine-Check Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
5.2. EXCEPTION AND INTERRUPT VECTORS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
5.3. EXCEPTION CLASSIFICATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
5.4. PROGRAM OR TASK RESTART. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-7
5.5. NONMASKABLE INTERRUPT (NMI). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-8
5.5.1. Handling Multiple NMIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-8
5.6. ENABLING AND DISABLING INTERRUPTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-8
5.6.1. Masking Maskable Hardware Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-8
5.6.2. Masking Instruction Breakpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-9
5.6.3. Masking Exceptions and Interrupts When Switching Stacks . . . . . . . . . . . . . . . 5-10
5.7. PRIORITY AMONG SIMULTANEOUS EXCEPTIONS AND INTERRUPTS . . . . . 5-10
5.8. INTERRUPT DESCRIPTOR TABLE (IDT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-11
5.9. IDT DESCRIPTORS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-13
5.10. EXCEPTION AND INTERRUPT HANDLING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
5.10.1. Exception- or Interrupt-Handler Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
5.10.1.1. Protection of Exception- and Interrupt-Handler Procedures . . . . . . . . . . . . . 5-17
5.10.1.2. Flag Usage By Exception- or Interrupt-Handler Procedure. . . . . . . . . . . . . . 5-18

v
TABLE OF CONTENTS

5.10.2. Interrupt Tasks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-18

5.11. ERROR CODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-20
5.12. EXCEPTION AND INTERRUPT REFERENCE . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-21

CHAPTER 6
TASK MANAGEMENT
6.1. TASK MANAGEMENT OVERVIEW. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1
6.1.1. Task Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-1
6.1.2. Task State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-2
6.1.3. Executing a Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-3
6.2. TASK MANAGEMENT DATA STRUCTURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-4
6.2.1. Task-State Segment (TSS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-4
6.2.2. TSS Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-6
6.2.3. Task Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-8
6.2.4. Task-Gate Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-8
6.3. TASK SWITCHING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-10
6.4. TASK LINKING. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-14
6.4.1. Use of Busy Flag To Prevent Recursive Task Switching . . . . . . . . . . . . . . . . . .6-16
6.4.2. Modifying Task Linkages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-16
6.5. TASK ADDRESS SPACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-17
6.5.1. Mapping Tasks to the Linear and Physical Address Spaces. . . . . . . . . . . . . . . .6-17
6.5.2. Task Logical Address Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-18
6.6. 16-BIT TASK-STATE SEGMENT (TSS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-19

CHAPTER 7
MULTIPLE-PROCESSOR MANAGEMENT
7.1. LOCKED ATOMIC OPERATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-2
7.1.1. Guaranteed Atomic Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-2
7.1.2. Bus Locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-3
7.1.2.1. Automatic Locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-3
7.1.2.2. Software Controlled Bus Locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-4
7.1.3. Handling Self- and Cross-Modifying Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-5
7.1.4. Effects of a LOCK Operation on Internal Processor Caches. . . . . . . . . . . . . . . . .7-6
7.2. MEMORY ORDERING. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-6
7.2.1. Memory Ordering in the Pentium ® and Intel486™ Processors. . . . . . . . . . . . . . .7-7
7.2.2. Memory Ordering in the P6 Family Processors. . . . . . . . . . . . . . . . . . . . . . . . . . .7-7
7.2.3. Out of Order Stores From String Operations in P6 Family Processors . . . . . . . . .7-9
7.2.4. Strengthening or Weakening the Memory Ordering Model . . . . . . . . . . . . . . . . . .7-9
7.3. PROPAGATION OF PAGE TABLE ENTRY CHANGES TO
MULTIPLE PROCESSORS 7-11
7.4. SERIALIZING INSTRUCTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-11
7.5. ADVANCED PROGRAMMABLE INTERRUPT CONTROLLER (APIC). . . . . . . . . 7-13
7.5.1. Presence of APIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-14
7.5.2. Enabling or Disabling the Local APIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-14
7.5.3. APIC Bus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-14
7.5.4. Valid Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-15
7.5.5. Interrupt Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-15
7.5.6. Bus Arbitration Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-15
7.5.7. The Local APIC Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-16
7.5.8. Relocation of the APIC Registers Base Address. . . . . . . . . . . . . . . . . . . . . . . . .7-19
7.5.9. Interrupt Destination and APIC ID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-20
7.5.9.1. Physical Destination Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-20

vi
 TABLE OF CONTENTS

7.5.9.2. Logical Destination Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-20

7.5.9.3. Flat Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-21
7.5.9.4. Cluster Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-21
7.5.9.5. Arbitration Priority . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-22
7.5.10. Interrupt Distribution Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-22
7.5.11. Local Vector Table. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-23
7.5.12. Interprocessor and Self-Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-25
7.5.13. Interrupt Acceptance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-30
7.5.13.1. Interrupt Acceptance Decision Flow Chart . . . . . . . . . . . . . . . . . . . . . . . . . . 7-30
7.5.13.2. Task Priority Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-31
7.5.13.3. Processor Priority Register (PPR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-32
7.5.13.4. Arbitration Priority Register (APR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-32
7.5.13.5. Spurious Interrupt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-33
7.5.13.6. End-Of-Interrupt (EOI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-33
7.5.14. Local APIC State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-33
7.5.14.1. Spurious-Interrupt Vector Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-34
7.5.14.2. Local APIC Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-35
7.5.14.3. Local APIC State After Power-Up Reset. . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-35
7.5.14.4. Local APIC State After an INIT Reset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-35
7.5.14.5. Local APIC State After INIT-Deassert Message . . . . . . . . . . . . . . . . . . . . . . 7-35
7.5.15. Local APIC Version Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-36
7.5.16. APIC Bus Arbitration Mechanism and Protocol . . . . . . . . . . . . . . . . . . . . . . . . . 7-36
7.5.16.1. Bus Message Formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-37
7.5.16.2. APIC Bus Status Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-40
7.5.17. Error Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-42
7.5.18. Timer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-43
7.5.19. Software Visible Differences Between the Local APIC and the 82489DX . . . . . 7-44
7.5.20. Performance Related Differences between the Local APIC and the 82489DX . 7-45
7.5.21. New Features Incorporated in the Pentium® and P6 Family
Processors Local APIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-45
7.6. DUAL-PROCESSOR (DP) INITIALIZATION PROTOCOL . . . . . . . . . . . . . . . . . . . 7-45
7.7. MULTIPLE-PROCESSOR (MP) INITIALIZATION PROTOCOL. . . . . . . . . . . . . . . 7-46
7.7.1. MP Initialization Protocol Requirements and Restrictions . . . . . . . . . . . . . . . . . 7-46
7.7.2. MP Protocol Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-47
7.7.3. Error Detection During the MP Initialization Protocol. . . . . . . . . . . . . . . . . . . . . 7-48
7.7.4. Error Handling During the MP Initialization Protocol . . . . . . . . . . . . . . . . . . . . . 7-48
7.7.5. MP Initialization Protocol Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-48

CHAPTER 8
PROCESSOR MANAGEMENT AND INITIALIZATION
8.1. INITIALIZATION OVERVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
8.1.1. Processor State After Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-2
8.1.2. Processor Built-In Self-Test (BIST) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-2
8.1.3. Model and Stepping Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-5
8.1.4. First Instruction Executed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-6
8.2. FPU INITIALIZATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-6
8.2.1. Configuring the FPU Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-6
8.2.2. Setting the Processor for FPU Software Emulation. . . . . . . . . . . . . . . . . . . . . . . 8-8
8.3. CACHE ENABLING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-8
8.4. MODEL-SPECIFIC REGISTERS (MSRS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-8
8.5. MEMORY TYPE RANGE REGISTERS (MTRRS) . . . . . . . . . . . . . . . . . . . . . . . . . . 8-9
8.6. SOFTWARE INITIALIZATION FOR REAL-ADDRESS MODE OPERATION . . . . 8-10

vii
TABLE OF CONTENTS

8.6.1. Real-Address Mode IDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-10

8.6.2. NMI Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-10
8.7. SOFTWARE INITIALIZATION FOR PROTECTED-MODE OPERATION . . . . . . . 8-11
8.7.1. Protected-Mode System Data Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-12
8.7.2. Initializing Protected-Mode Exceptions and Interrupts . . . . . . . . . . . . . . . . . . . .8-12
8.7.3. Initializing Paging. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-12
8.7.4. Initializing Multitasking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-13
8.8. MODE SWITCHING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-13
8.8.1. Switching to Protected Mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-14
8.8.2. Switching Back to Real-Address Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-15
8.9. INITIALIZATION AND MODE SWITCHING EXAMPLE . . . . . . . . . . . . . . . . . . . . . 8-16
8.9.1. Assembler Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-19
8.9.2. STARTUP.ASM Listing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-19
8.9.3. MAIN.ASM Source Code. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-29
8.9.4. Supporting Files. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-29
8.10. P6 FAMILY MICROCODE UPDATE FEATURE . . . . . . . . . . . . . . . . . . . . . . . . . . 8-31
8.10.1. Microcode Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-32
8.10.2. Microcode Update Loader . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-35
8.10.2.1. Update Loading Procedure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-36
8.10.2.2. Hard Resets in Update Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-36
8.10.2.3. Update in a Multiprocessor System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-37
8.10.2.4. Update Loader Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-37
8.10.3. Update Signature and Verification. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-37
8.10.3.1. Determining the Signature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-38
8.10.3.2. Authenticating the Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-38
8.10.4. P6 Family Processor Microcode Update Specifications . . . . . . . . . . . . . . . . . . .8-39
8.10.4.1. Responsibilities of the BIOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-39
8.10.4.2. Responsibilities of the Calling Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-40
8.10.4.3. Microcode Update Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-43
8.10.4.4. INT 15h-based Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-43
8.10.4.5. Return Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-50

CHAPTER 9
MEMORY CACHE CONTROL
9.1. INTERNAL CACHES, TLBS, AND BUFFERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-1
9.2. CACHING TERMINOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-4
9.3. METHODS OF CACHING AVAILABLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-5
9.3.1. Buffering of Write Combining Memory Locations . . . . . . . . . . . . . . . . . . . . . . . . .9-7
9.3.2. Choosing a Memory Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-8
9.4. CACHE CONTROL PROTOCOL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-9
9.5. CACHE CONTROL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-9
9.5.1. Precedence of Cache Controls (P6 Family Processor) . . . . . . . . . . . . . . . . . . . .9-13
9.5.2. Preventing Caching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-14
9.6. CACHE MANAGEMENT INSTRUCTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-15
9.7. SELF-MODIFYING CODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-15
9.8. IMPLICIT CACHING (P6 FAMILY PROCESSORS) . . . . . . . . . . . . . . . . . . . . . . . 9-16
9.9. EXPLICIT CACHING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-16
9.10. INVALIDATING THE TRANSLATION LOOKASIDE BUFFERS (TLBS) . . . . . . . . 9-17
9.11. WRITE BUFFER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-17
9.12. MEMORY TYPE RANGE REGISTERS (MTRRS) . . . . . . . . . . . . . . . . . . . . . . . . . 9-18
9.12.1. MTRR Feature Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-20
9.12.2. Setting Memory Ranges with MTRRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-21

viii
 TABLE OF CONTENTS

9.12.2.1. MTRRdefType Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-21

9.12.2.2. Fixed Range MTRRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-22
9.12.2.3. Variable Range MTRRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-23
9.12.3. Example Base and Mask Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-25
9.12.4. Range Size and Alignment Requirement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-26
9.12.4.1. MTRR Precedences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-26
9.12.5. MTRR Initialization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-27
9.12.6. Remapping Memory Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-27
9.12.7. MTRR Maintenance Programming Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-28
9.12.7.1. MemTypeGet() Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-28
9.12.7.2. MemTypeSet() Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-29
9.12.8. Multiple-Processor Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-31
9.12.9. Large Page Size Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-32
9.13. PAGE ATTRIBUTE TABLE (PAT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-33
9.13.1. Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-33
9.13.2. Detecting Support for the PAT Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-34
9.13.3. Technical Description of the PAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-34
9.13.4. Accessing the PAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-35
9.13.5. Programming the PAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-38

CHAPTER 10
MMX™ TECHNOLOGY SYSTEM PROGRAMMING
10.1. EMULATION OF THE MMX™ INSTRUCTION SET . . . . . . . . . . . . . . . . . . . . . . . 10-1
10.2. THE MMX™ STATE AND MMX™ REGISTER ALIASING . . . . . . . . . . . . . . . . . . 10-1
10.2.1. Effect of MMX™ and Floating-Point Instructions on the FPU Tag Word . . . . . . 10-3
10.3. SAVING AND RESTORING THE MMX™ STATE AND REGISTERS . . . . . . . . . . 10-4
10.4. DESIGNING OPERATING SYSTEM TASK AND CONTEXT
SWITCHING FACILITIES 10-5
10.4.1. Using the TS Flag in Control Register CR0 to Control MMX™/FPU
State Saving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-5
10.5. EXCEPTIONS THAT CAN OCCUR WHEN EXECUTING
MMX™ INSTRUCTIONS 10-7
10.5.1. Effect of MMX™ Instructions on Pending Floating-Point Exceptions . . . . . . . . 10-8
10.6. DEBUGGING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-8

CHAPTER 11
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING
11.1. EMULATION OF THE STREAMING SIMD EXTENSIONS . . . . . . . . . . . . . . . . . . 11-1
11.2. MMX™ STATE AND STREAMING SIMD EXTENSIONS . . . . . . . . . . . . . . . . . . . 11-1
11.3. NEW PENTIUM® III PROCESSOR REGISTERS . . . . . . . . . . . . . . . . . . . . . . . . . 11-1
11.3.1. SIMD Floating-point Registers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-2
11.3.2. SIMD Floating-point Control/Status Registers . . . . . . . . . . . . . . . . . . . . . . . . . . 11-2
11.3.2.1. Rounding Control Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-3
11.3.2.2. Flush-to-Zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-5
11.4. ENABLING STREAMING SIMD EXTENSIONS SUPPORT. . . . . . . . . . . . . . . . . . 11-6
11.4.1. Enabling Streaming SIMD Extensions Support . . . . . . . . . . . . . . . . . . . . . . . . . 11-6
11.4.2. Device Not Available (DNA) Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-6
11.4.3. FXSAVE/FXRSTOR as a Replacement for FSAVE/FRSTOR. . . . . . . . . . . . . . 11-7
11.4.4. Numeric Error flag and IGNNE# . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-7
11.5. SAVING AND RESTORING THE STREAMING SIMD EXTENSIONS STATE . . . 11-7
11.6. DESIGNING OPERATING SYSTEM TASK AND CONTEXT
SWITCHING FACILITIES 11-8

ix
TABLE OF CONTENTS

11.6.1. Using the TS Flag in Control Register CR0 to Control SIMD Floating-Point
State Saving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-8
11.7. EXCEPTIONS THAT CAN OCCUR WHEN EXECUTING STREAMING SIMD
EXTENSIONS INSTRUCTIONS 11-11
11.7.1. SIMD Floating-point Non-Numeric Exceptions . . . . . . . . . . . . . . . . . . . . . . . . .11-12
11.7.2. SIMD Floating-point Numeric Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-13
11.7.2.1. Exception Priority . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-13
11.7.2.2. Automatic Masked Exception Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-14
11.7.2.3. Software Exception Handling - Unmasked Exceptions. . . . . . . . . . . . . . . . .11-15
11.7.2.4. Interaction with x87 numeric exceptions. . . . . . . . . . . . . . . . . . . . . . . . . . . .11-16
11.7.3. SIMD Floating-point Numeric Exception Conditions and
Masked/Unmasked Responses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-16
11.7.3.1. Invalid Operation Exception(#IA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-17
11.7.3.2. Division-By-Zero Exception (#Z). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-18
11.7.3.3. Denormal Operand Exception (#D) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-19
11.7.3.4. Numeric Overflow Exception (#O) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-19
11.7.3.5. Numeric Underflow Exception (#U) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-20
11.7.3.6. Inexact Result (Precision) Exception (#P) . . . . . . . . . . . . . . . . . . . . . . . . . .11-21
11.7.4. Effect of Streaming SIMD Extensions Instructions on Pending
Floating-Point Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-22
11.8. DEBUGGING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-22

CHAPTER 12
SYSTEM MANAGEMENT MODE (SMM)
12.1. SYSTEM MANAGEMENT MODE OVERVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-1
12.2. SYSTEM MANAGEMENT INTERRUPT (SMI) . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
12.3. SWITCHING BETWEEN SMM AND THE OTHER PROCESSOR
OPERATING MODES 12-2
12.3.1. Entering SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-2
12.3.1.1. Exiting From SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-3
12.4. SMRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-4
12.4.1. SMRAM State Save Map. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-5
12.4.2. SMRAM Caching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-7
12.5. SMI HANDLER EXECUTION ENVIRONMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-8
12.6. EXCEPTIONS AND INTERRUPTS WITHIN SMM . . . . . . . . . . . . . . . . . . . . . . . 12-10
12.7. NMI HANDLING WHILE IN SMM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-11
12.8. SAVING THE FPU STATE WHILE IN SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-11
12.9. SMM REVISION IDENTIFIER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-12
12.10. AUTO HALT RESTART . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-13
12.10.1. Executing the HLT Instruction in SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-14
12.11. SMBASE RELOCATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-14
12.11.1. Relocating SMRAM to an Address Above 1 MByte. . . . . . . . . . . . . . . . . . . . . .12-15
12.12. I/O INSTRUCTION RESTART . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-15
12.12.1. Back-to-Back SMI Interrupts When I/O Instruction Restart Is Being Used . . . .12-16
12.13. SMM MULTIPLE-PROCESSOR CONSIDERATIONS. . . . . . . . . . . . . . . . . . . . . 12-17

CHAPTER 13
MACHINE-CHECK ARCHITECTURE
13.1. MACHINE-CHECK EXCEPTIONS AND ARCHITECTURE . . . . . . . . . . . . . . . . . . 13-1
13.2. COMPATIBILITY WITH PENTIUM® PROCESSOR . . . . . . . . . . . . . . . . . . . . . . . 13-1
13.3. MACHINE-CHECK MSRS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-2
13.3.1. Machine-Check Global Control MSRs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13-2

x
 TABLE OF CONTENTS

13.3.1.1. MCG_CAP MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-2

13.3.1.2. MCG_STATUS MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-3
13.3.1.3. MCG_CTL MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-4
13.3.2. Error-Reporting Register Banks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-4
13.3.2.1. MCi_CTL MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-4
13.3.2.2. MCi_STATUS MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-5
13.3.2.3. MCi_ADDR MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-6
13.3.2.4. MCi_MISC MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-7
13.3.3. Mapping of the Pentium® Processor Machine-Check Errors to the P6
Family Machine-Check Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-7
13.4. MACHINE-CHECK AVAILABILITY. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-7
13.5. MACHINE-CHECK INITIALIZATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-7
13.6. INTERPRETING THE MCA ERROR CODES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-8
13.6.1. Simple Error Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-9
13.6.2. Compound Error Codes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-9
13.6.3. Interpreting the Machine-Check Error Codes for External Bus Errors. . . . . . . 13-11
13.7. GUIDELINES FOR WRITING MACHINE-CHECK SOFTWARE . . . . . . . . . . . . . 13-14
13.7.1. Machine-Check Exception Handler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-14
13.7.2. Pentium® Processor Machine-Check Exception Handling . . . . . . . . . . . . . . . 13-16
13.7.3. Logging Correctable Machine-Check Errors . . . . . . . . . . . . . . . . . . . . . . . . . . 13-16

CHAPTER 14
CODE OPTIMIZATION
14.1. CODE OPTIMIZATION GUIDELINES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-1
14.1.1. General Code Optimization Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-1
14.1.2. Guidelines for Optimizing MMX™ Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-2
14.1.3. Guidelines for Optimizing Floating-Point Code . . . . . . . . . . . . . . . . . . . . . . . . . 14-2
14.1.4. Guidelines for Optimizing SIMD Floating-point Code . . . . . . . . . . . . . . . . . . . . 14-3
14.2. BRANCH PREDICTION OPTIMIZATION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-4
14.2.1. Branch Prediction Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-4
14.2.2. Optimizing Branch Predictions in Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-5
14.2.3. Eliminating and Reducing the Number of Branches . . . . . . . . . . . . . . . . . . . . . 14-5
14.3. REDUCING PARTIAL REGISTER STALLS ON P6 FAMILY PROCESSORS. . . . 14-7
14.4. ALIGNMENT RULES AND GUIDELINES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-9
14.4.1. Alignment Penalties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-9
14.4.2. Code Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-9
14.4.3. Data Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-9
14.4.3.1. Alignment of Data Structures and Arrays Greater Than 32 Bytes . . . . . . . 14-10
14.4.3.2. Alignment of Data in Memory and on the Stack . . . . . . . . . . . . . . . . . . . . . 14-10
14.5. INSTRUCTION SCHEDULING OVERVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-12
14.5.1. Instruction Pairing Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-12
14.5.1.1. General Pairing Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-12
14.5.1.2. Integer Pairing Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-13
14.5.1.3. MMX™ Instruction Pairing Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-17
14.5.2. Pipelining Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-18
14.5.2.1. MMX™ Instruction Pipelining Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . 14-18
14.5.2.2. Floating-Point Pipelining Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-18
14.5.3. Scheduling Rules for P6 Family Processors . . . . . . . . . . . . . . . . . . . . . . . . . . 14-22
14.6. ACCESSING MEMORY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-24
14.6.1. Using MMX™ Instructions That Access Memory. . . . . . . . . . . . . . . . . . . . . . . 14-24
14.6.2. Partial Memory Accesses With MMX™ Instructions . . . . . . . . . . . . . . . . . . . . 14-25
14.6.3. Write Allocation Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-27

xi
TABLE OF CONTENTS

14.7. ADDRESSING MODES AND REGISTER USAGE . . . . . . . . . . . . . . . . . . . . . . . 14-29

14.8. INSTRUCTION LENGTH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-30
14.9. PREFIXED OPCODES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-31
14.10. INTEGER INSTRUCTION SELECTION AND OPTIMIZATIONS. . . . . . . . . . . . . 14-32

CHAPTER 15
DEBUGGING AND PERFORMANCE MONITORING
15.1. OVERVIEW OF THE DEBUGGING SUPPORT FACILITIES . . . . . . . . . . . . . . . . 15-1
15.2. DEBUG REGISTERS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-2
15.2.1. Debug Address Registers (DR0-DR3). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-4
15.2.2. Debug Registers DR4 and DR5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-4
15.2.3. Debug Status Register (DR6) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-4
15.2.4. Debug Control Register (DR7) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-5
15.2.5. Breakpoint Field Recognition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-6
15.3. DEBUG EXCEPTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-7
15.3.1. Debug Exception (#DB)—Interrupt Vector 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-8
15.3.1.1. Instruction-Breakpoint Exception Condition . . . . . . . . . . . . . . . . . . . . . . . . . .15-8
15.3.1.2. Data Memory and I/O Breakpoint Exception Conditions . . . . . . . . . . . . . . . .15-9
15.3.1.3. General-Detect Exception Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-10
15.3.1.4. Single-Step Exception Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-10
15.3.1.5. Task-Switch Exception Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-11
15.3.2. Breakpoint Exception (#BP)—Interrupt Vector 3 . . . . . . . . . . . . . . . . . . . . . . . .15-11
15.4. LAST BRANCH, INTERRUPT, AND EXCEPTION RECORDING . . . . . . . . . . . . 15-11
15.4.1. DebugCtlMSR Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-11
15.4.2. Last Branch and Last Exception MSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-13
15.4.3. Monitoring Branches, Exceptions, and Interrupts . . . . . . . . . . . . . . . . . . . . . . .15-13
15.4.4. Single-Stepping on Branches, Exceptions, and Interrupts . . . . . . . . . . . . . . . .15-14
15.4.5. Initializing Last Branch or Last Exception/Interrupt Recording . . . . . . . . . . . . .15-14
15.5. TIME-STAMP COUNTER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-14
15.6. PERFORMANCE-MONITORING COUNTERS . . . . . . . . . . . . . . . . . . . . . . . . . . 15-15
15.6.1. P6 Family Processor Performance-Monitoring Counters . . . . . . . . . . . . . . . . .15-15
15.6.1.1. PerfEvtSel0 and PerfEvtSel1 MSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-16
15.6.1.2. PerfCtr0 and PerfCtr1 MSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-18
15.6.1.3. Starting and Stopping the Performance-Monitoring Counters . . . . . . . . . . .15-18
15.6.1.4. Event and Time-Stamp Monitoring Software . . . . . . . . . . . . . . . . . . . . . . . .15-18
15.6.2. Monitoring Counter Overflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-19
15.6.3. Pentium® Processor Performance-Monitoring Counters. . . . . . . . . . . . . . . . . .15-20
15.6.3.1. Control and Event Select Register (CESR) . . . . . . . . . . . . . . . . . . . . . . . . .15-20
15.6.3.2. Use of the Performance-Monitoring Pins . . . . . . . . . . . . . . . . . . . . . . . . . . .15-21
15.6.3.3. Events Counted . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-22

CHAPTER 16
8086 EMULATION
16.1. REAL-ADDRESS MODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-1
16.1.1. Address Translation in Real-Address Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . .16-3
16.1.2. Registers Supported in Real-Address Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . .16-4
16.1.3. Instructions Supported in Real-Address Mode . . . . . . . . . . . . . . . . . . . . . . . . . .16-4
16.1.4. Interrupt and Exception Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16-6
16.2. VIRTUAL-8086 MODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-9
16.2.1. Enabling Virtual-8086 Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16-9
16.2.2. Structure of a Virtual-8086 Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16-9
16.2.3. Paging of Virtual-8086 Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16-10

xii
 TABLE OF CONTENTS

16.2.4. Protection within a Virtual-8086 Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-11

16.2.5. Entering Virtual-8086 Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-11
16.2.6. Leaving Virtual-8086 Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-13
16.2.7. Sensitive Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-14
16.2.8. Virtual-8086 Mode I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-14
16.2.8.1. I/O-Port-Mapped I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-15
16.2.8.2. Memory-Mapped I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-15
16.2.8.3. Special I/O Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-15
16.3. INTERRUPT AND EXCEPTION HANDLING IN VIRTUAL-8086 MODE . . . . . . . 16-15
16.3.1. Class 1—Hardware Interrupt and Exception Handling in Virtual-8086 Mode . 16-17
16.3.1.1. Handling an Interrupt or Exception Through a Protected-Mode Trap or
Interrupt Gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-17
16.3.1.2. Handling an Interrupt or Exception With an 8086 Program Interrupt or
Exception Handler. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-19
16.3.1.3. Handling an Interrupt or Exception Through a Task Gate . . . . . . . . . . . . . 16-20
16.3.2. Class 2—Maskable Hardware Interrupt Handling in Virtual-8086
Mode Using the Virtual Interrupt Mechanism. . . . . . . . . . . . . . . . . . . . . . . . . . 16-20
16.3.3. Class 3—Software Interrupt Handling in Virtual-8086 Mode . . . . . . . . . . . . . . 16-23
16.3.3.1. Method 1: Software Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-25
16.3.3.2. Methods 2 and 3: Software Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . 16-26
16.3.3.3. Method 4: Software Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-26
16.3.3.4. Method 5: Software Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-26
16.3.3.5. Method 6: Software Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-27
16.4. PROTECTED-MODE VIRTUAL INTERRUPTS . . . . . . . . . . . . . . . . . . . . . . . . . . 16-27

CHAPTER 17
MIXING 16-BIT AND 32-BIT CODE
17.1. DEFINING 16-BIT AND 32-BIT PROGRAM MODULES . . . . . . . . . . . . . . . . . . . . 17-2
17.2. MIXING 16-BIT AND 32-BIT OPERATIONS WITHIN A CODE SEGMENT. . . . . . 17-2
17.3. SHARING DATA AMONG MIXED-SIZE CODE SEGMENTS . . . . . . . . . . . . . . . . 17-3
17.4. TRANSFERRING CONTROL AMONG MIXED-SIZE CODE SEGMENTS . . . . . . 17-4
17.4.1. Code-Segment Pointer Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-5
17.4.2. Stack Management for Control Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-5
17.4.2.1. Controlling the Operand-Size Attribute For a Call. . . . . . . . . . . . . . . . . . . . . 17-7
17.4.2.2. Passing Parameters With a Gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-7
17.4.3. Interrupt Control Transfers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-8
17.4.4. Parameter Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-8
17.4.5. Writing Interface Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-8

CHAPTER 18
INTEL ARCHITECTURE COMPATIBILITY
18.1. INTEL ARCHITECTURE FAMILIES AND CATEGORIES . . . . . . . . . . . . . . . . . . . 18-1
18.2. RESERVED BITS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-1
18.3. ENABLING NEW FUNCTIONS AND MODES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-2
18.4. DETECTING THE PRESENCE OF NEW FEATURES THROUGH SOFTWARE . 18-2
18.5. MMX™ TECHNOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-3
18.6. STREAMING SIMD EXTENSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-3
18.7. NEW INSTRUCTIONS IN THE PENTIUM® AND LATER INTEL
ARCHITECTURE PROCESSORS 18-3
18.7.1. Instructions Added Prior to the Pentium® Processor. . . . . . . . . . . . . . . . . . . . . 18-5
18.8. OBSOLETE INSTRUCTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-5
18.9. UNDEFINED OPCODES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-6

xiii
TABLE OF CONTENTS

18.10. NEW FLAGS IN THE EFLAGS REGISTER. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-6

18.10.1. Using EFLAGS Flags to Distinguish Between 32-Bit Intel
Architecture Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-6
18.11. STACK OPERATIONS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-7
18.11.1. PUSH SP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-7
18.11.2. EFLAGS Pushed on the Stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-7
18.12. FPU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-7
18.12.1. Control Register CR0 Flags. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-8
18.12.2. FPU Status Word. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-8
18.12.2.1. Condition Code Flags (C0 through C3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-8
18.12.2.2. Stack Fault Flag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-9
18.12.3. FPU Control Word . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-9
18.12.4. FPU Tag Word. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-9
18.12.5. Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-10
18.12.5.1. NaNs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-10
18.12.5.2. Pseudo-zero, Pseudo-NaN, Pseudo-infinity, and Unnormal Formats . . . . .18-10
18.12.6. Floating-Point Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-11
18.12.6.1. Denormal Operand Exception (#D) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-11
18.12.6.2. Numeric Overflow Exception (#O) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-11
18.12.6.3. Numeric Underflow Exception (#U) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-12
18.12.6.4. Exception Precedence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-12
18.12.6.5. CS and EIP For FPU Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-12
18.12.6.6. FPU Error Signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-12
18.12.6.7. Assertion of the FERR# Pin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-13
18.12.6.8. Invalid Operation Exception On Denormals . . . . . . . . . . . . . . . . . . . . . . . . .18-13
18.12.6.9. Alignment Check Exceptions (#AC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-13
18.12.6.10. Segment Not Present Exception During FLDENV . . . . . . . . . . . . . . . . . . . .18-14
18.12.6.11. Device Not Available Exception (#NM). . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-14
18.12.6.12. Coprocessor Segment Overrun Exception . . . . . . . . . . . . . . . . . . . . . . . . . .18-14
18.12.6.13. General Protection Exception (#GP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-14
18.12.6.14. Floating-Point Error Exception (#MF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-14
18.12.7. Changes to Floating-Point Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-14
18.12.7.1. FDIV, FPREM, and FSQRT Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . .18-15
18.12.7.2. FSCALE Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-15
18.12.7.3. FPREM1 Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-15
18.12.7.4. FPREM Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-15
18.12.7.5. FUCOM, FUCOMP, and FUCOMPP Instructions. . . . . . . . . . . . . . . . . . . . .18-15
18.12.7.6. FPTAN Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-15
18.12.7.7. Stack Overflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-16
18.12.7.8. FSIN, FCOS, and FSINCOS Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . .18-16
18.12.7.9. FPATAN Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-16
18.12.7.10. F2XM1 Instruction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-16
18.12.7.11. FLD Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-16
18.12.7.12. FXTRACT Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-17
18.12.7.13. Load Constant Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-17
18.12.7.14. FSETPM Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-17
18.12.7.15. FXAM Instruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-17
18.12.7.16. FSAVE and FSTENV Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-18
18.12.8. Transcendental Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-18
18.12.9. Obsolete Instructions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-18
18.12.10. WAIT/FWAIT Prefix Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-18
18.12.11. Operands Split Across Segments and/or Pages . . . . . . . . . . . . . . . . . . . . . . . .18-18

xiv
 TABLE OF CONTENTS

18.12.12. FPU Instruction Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-19

18.13. SERIALIZING INSTRUCTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-19
18.14. FPU AND MATH COPROCESSOR INITIALIZATION . . . . . . . . . . . . . . . . . . . . . 18-19
18.14.1. Intel 387 and Intel 287 Math Coprocessor Initialization . . . . . . . . . . . . . . . . . . 18-19
18.14.2. Intel486™ SX Processor and Intel 487 SX Math Coprocessor Initialization . . 18-20
18.15. CONTROL REGISTERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-21
18.16. MEMORY MANAGEMENT FACILITIES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-23
18.16.1. New Memory Management Control Flags . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-23
18.16.1.1. Physical Memory Addressing Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-23
18.16.1.2. Global Pages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-23
18.16.1.3. Larger Page Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-23
18.16.2. CD and NW Cache Control Flags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-23
18.16.3. Descriptor Types and Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-24
18.16.4. Changes in Segment Descriptor Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-24
18.17. DEBUG FACILITIES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-24
18.17.1. Differences in Debug Register DR6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-24
18.17.2. Differences in Debug Register DR7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-24
18.17.3. Debug Registers DR4 and DR5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-25
18.17.4. Recognition of Breakpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-25
18.18. TEST REGISTERS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-25
18.19. EXCEPTIONS AND/OR EXCEPTION CONDITIONS . . . . . . . . . . . . . . . . . . . . . 18-25
18.19.1. Machine-Check Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-27
18.19.2. Priority OF Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-27
18.20. INTERRUPTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-27
18.20.1. Interrupt Propagation Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-27
18.20.2. NMI Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
18.20.3. IDT Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
18.21. TASK SWITCHING AND TSS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
18.21.1. P6 Family and Pentium® Processor TSS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
18.21.2. TSS Selector Writes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
18.21.3. Order of Reads/Writes to the TSS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
18.21.4. Using A 16-Bit TSS with 32-Bit Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-29
18.21.5. Differences in I/O Map Base Addresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-29
18.22. CACHE MANAGEMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-30
18.22.1. Self-Modifying Code with Cache Enabled . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-31
18.23. PAGING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-31
18.23.1. Large Pages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-32
18.23.2. PCD and PWT Flags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-32
18.23.3. Enabling and Disabling Paging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-32
18.24. STACK OPERATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-33
18.24.1. Selector Pushes and Pops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-33
18.24.2. Error Code Pushes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-33
18.24.3. Fault Handling Effects on the Stack. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-33
18.24.4. Interlevel RET/IRET From a 16-Bit Interrupt or Call Gate . . . . . . . . . . . . . . . . 18-34
18.25. MIXING 16- AND 32-BIT SEGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-34
18.26. SEGMENT AND ADDRESS WRAPAROUND. . . . . . . . . . . . . . . . . . . . . . . . . . . 18-35
18.26.1. Segment Wraparound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-35
18.27. WRITE BUFFERS AND MEMORY ORDERING . . . . . . . . . . . . . . . . . . . . . . . . . 18-36
18.28. BUS LOCKING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-37
18.29. BUS HOLD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-37
18.30. TWO WAYS TO RUN INTEL 286 PROCESSOR TASKS . . . . . . . . . . . . . . . . . . 18-37
18.31. MODEL-SPECIFIC EXTENSIONS TO THE INTEL ARCHITECTURE . . . . . . . . 18-38

xv
TABLE OF CONTENTS

18.31.1. Model-Specific Registers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-38

18.31.2. RDMSR and WRMSR Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-38
18.31.3. Memory Type Range Registers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-39
18.31.4. Machine-Check Exception and Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . .18-39
18.31.5. Performance-Monitoring Counters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-40

APPENDIX A
PERFORMANCE-MONITORING EVENTS
A.1. P6 FAMILY PROCESSOR PERFORMANCE-MONITORING EVENTS . . . . . . . . . A-1
A.2. PENTIUM® PROCESSOR PERFORMANCE-MONITORING EVENTS . . . . . . . . A-12

APPENDIX B
MODEL-SPECIFIC REGISTERS

APPENDIX C ®
DUAL-PROCESSOR (DP) BOOTUP SEQUENCE EXAMPLE (SPECIFIC TO PENTIUM
PROCESSORS)
C.1. PRIMARY PROCESSOR’S SEQUENCE OF EVENTS . . . . . . . . . . . . . . . . . . . . . . C-1
C.2. SECONDARY PROCESSOR’S SEQUENCE OF EVENTS FOLLOWING
RECEIPT OF START-UP IPI C-4

APPENDIX D
MULTIPLE-PROCESSOR (MP) BOOTUP SEQUENCE EXAMPLE (SPECIFIC TO P6 FAMILY
PROCESSORS)
D.1. BSP’S SEQUENCE OF EVENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D-1
D.2. AP’S SEQUENCE OF EVENTS FOLLOWING RECEIPT OF START-UP IPI . . . . . D-3

APPENDIX E
PROGRAMMING THE LINT0 AND LINT1 INPUTS
E.1. CONSTANTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E-1
E.2. LINT[0:1] PINS PROGRAMMING PROCEDURE . . . . . . . . . . . . . . . . . . . . . . . . . . E-1

xvi
 TABLE OF FIGURES

Figure 1-1. Bit and Byte Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-6

Figure 2-1. System-Level Registers and Data Structures. . . . . . . . . . . . . . . . . . . . . . . . . .2-2
Figure 2-2. Transitions Among the Processor’s Operating Modes . . . . . . . . . . . . . . . . . . .2-7
Figure 2-3. System Flags in the EFLAGS Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-8
Figure 2-4. Memory Management Registers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-10
Figure 2-5. Control Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-12
Figure 3-1. Segmentation and Paging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-2
Figure 3-2. Flat Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-4
Figure 3-3. Protected Flat Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-4
Figure 3-4. Multisegment Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-5
Figure 3-5. Logical Address to Linear Address Translation . . . . . . . . . . . . . . . . . . . . . . . .3-7
Figure 3-6. Segment Selector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-8
Figure 3-7. Segment Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-9
Figure 3-8. Segment Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-11
Figure 3-9. Segment Descriptor When Segment-Present Flag Is Clear . . . . . . . . . . . . . .3-13
Figure 3-10. Global and Local Descriptor Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-17
Figure 3-11. Pseudo-Descriptor Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-18
Figure 3-12. Linear Address Translation (4-KByte Pages) . . . . . . . . . . . . . . . . . . . . . . . . .3-21
Figure 3-13. Linear Address Translation (4-MByte Pages). . . . . . . . . . . . . . . . . . . . . . . . .3-22
Figure 3-14. Format of Page-Directory and Page-Table Entries for 4-KByte Pages
and 32-Bit Physical Addresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-24
Figure 3-15. Format of Page-Directory Entries for 4-MByte Pages and 32-Bit Addresses .3-25
Figure 3-16. Format of a Page-Table or Page-Directory Entry for a Not-Present Page . . .3-28
Figure 3-17. Register CR3 Format When the Physical Address Extension is Enabled . . .3-30
Figure 3-18. Linear Address Translation With Extended Physical Addressing
Enabled (4-KByte Pages) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-31
Figure 3-19. Linear Address Translation With Extended Physical Addressing
Enabled (2-MByte or 4-MByte Pages) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-33
Figure 3-20. Format of Page-Directory-Pointer-Table, Page-Directory, and Page-Table
Entries for 4-KByte Pages and 36-Bit Extended Physical Addresses . . . . . .3-34
Figure 3-21. Format of Page-Directory-Pointer-Table and Page-Directory Entries for
2- or 4-MByte Pages and 36-Bit Extended Physical Addresses. . . . . . . . . . .3-35
Figure 3-22. PDE Format Differences between 36-bit and 32-bit addressing. . . . . . . . . . .3-38
Figure 3-23. Memory Management Convention That Assigns a Page Table to
Each Segment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-40
Figure 4-1. Descriptor Fields Used for Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-4
Figure 4-2. Protection Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-8
Figure 4-3. Privilege Check for Data Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-10
Figure 4-4. Examples of Accessing Data Segments From Various Privilege Levels . . . .4-11
Figure 4-5. Privilege Check for Control Transfer Without Using a Gate . . . . . . . . . . . . . .4-13
Figure 4-6. Examples of Accessing Conforming and Nonconforming Code
Segments From Various Privilege Levels. . . . . . . . . . . . . . . . . . . . . . . . . . . .4-14
Figure 4-7. Call-Gate Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-17
Figure 4-8. Call-Gate Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-18
Figure 4-9. Privilege Check for Control Transfer with Call Gate . . . . . . . . . . . . . . . . . . . .4-19
Figure 4-10. Example of Accessing Call Gates At Various Privilege Levels. . . . . . . . . . . .4-20
Figure 4-11. Stack Switching During an Interprivilege-Level Call . . . . . . . . . . . . . . . . . . . .4-23
Figure 4-12. Use of RPL to Weaken Privilege Level of Called Procedure . . . . . . . . . . . . .4-29
Figure 5-1. Relationship of the IDTR and IDT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-13

xvii
TABLE OF FIGURES

Figure 5-2. IDT Gate Descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-14

Figure 5-3. Interrupt Procedure Call . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-16
Figure 5-4. Stack Usage on Transfers to Interrupt and Exception-Handling Routines . . .5-17
Figure 5-5. Interrupt Task Switch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-19
Figure 5-6. Error Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-20
Figure 5-7. Page-Fault Error Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-45
Figure 6-1. Structure of a Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-2
Figure 6-2. 32-Bit Task-State Segment (TSS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-5
Figure 6-3. TSS Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-7
Figure 6-4. Task Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-9
Figure 6-5. Task-Gate Descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-9
Figure 6-6. Task Gates Referencing the Same Task . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-11
Figure 6-7. Nested Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-15
Figure 6-8. Overlapping Linear-to-Physical Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . .6-18
Figure 6-9. 16-Bit TSS Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-20
Figure 7-1. Example of Write Ordering in Multiple-Processor Systems . . . . . . . . . . . . . . .7-8
Figure 7-2. I/O APIC and Local APICs in Multiple-Processor Systems . . . . . . . . . . . . . .7-14
Figure 7-3. Local APIC Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-17
Figure 7-4. APIC_BASE_MSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-19
Figure 7-5. Local APIC ID Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-20
Figure 7-6. Logical Destination Register (LDR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-21
Figure 7-7. Destination Format Register (DFR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-21
Figure 7-8. Local Vector Table (LVT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-24
Figure 7-9. Interrupt Command Register (ICR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-26
Figure 7-10. IRR, ISR and TMR Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-30
Figure 7-11. Interrupt Acceptance Flow Chart for the Local APIC . . . . . . . . . . . . . . . . . . .7-31
Figure 7-12. Task Priority Register (TPR). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-32
Figure 7-13. EOI Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-33
Figure 7-14. Spurious-Interrupt Vector Register (SVR) . . . . . . . . . . . . . . . . . . . . . . . . . . .7-34
Figure 7-15. Local APIC Version Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-36
Figure 7-16. Error Status Register (ESR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-42
Figure 7-17. Divide Configuration Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-43
Figure 7-18. Initial Count and Current Count Registers . . . . . . . . . . . . . . . . . . . . . . . . . . .7-44
Figure 7-19. SMP System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-49
Figure 8-1. Contents of CR0 Register after Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-5
Figure 8-2. Processor Type and Signature in the EDX Register after Reset . . . . . . . . . . .8-5
Figure 8-3. Processor State After Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-17
Figure 8-4. Constructing Temporary GDT and Switching to Protected Mode
(Lines 162-172 of List File) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-26
Figure 8-5. Moving the GDT, IDT and TSS from ROM to RAM
(Lines 196-261 of List File) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-27
Figure 8-6. Task Switching (Lines 282-296 of List File) . . . . . . . . . . . . . . . . . . . . . . . . . .8-28
Figure 8-7. Integrating Processor Specific Updates . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-32
Figure 8-8. Format of the Microcode Update Data Block . . . . . . . . . . . . . . . . . . . . . . . . .8-35
Figure 8-9. Write Operation Flow Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-47
Figure 9-1. Intel Architecture Caches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-2
Figure 9-2. Cache-Control Mechanisms Available in the Intel Architecture Processors . .9-10
Figure 9-3. Mapping Physical Memory With MTRRs . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-20
Figure 9-4. MTRRcap Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-21
Figure 9-5. MTRRdefType Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-22
Figure 9-6. MTRRphysBasen and MTRRphysMaskn Variable-Range Register Pair . . . .9-24
Figure 9-7. Page Attribute Table Model Specific Register . . . . . . . . . . . . . . . . . . . . . . . .9-34

xviii
 TABLE OF FIGURES

Figure 9-8. Page Attribute Table Index Scheme for Paging Hierarchy . . . . . . . . . . . . . . 9-36
Figure 10-1. Mapping of MMX™ Registers to Floating-Point Registers . . . . . . . . . . . . . . 10-2
Figure 10-2. Example of MMX™/FPU State Saving During an
Operating System-Controlled Task Switch . . . . . . . . . . . . . . . . . . . . . . . . . . 10-6
Figure 10-3. Mapping of MMX™ Registers to Floating-Point (FP) Registers . . . . . . . . . . 10-9
Figure 11-1. Streaming SIMD Extensions Control/Status Register Format. . . . . . . . . . . . 11-3
Figure 11-2. Example of SIMD Floating-Point State Saving During an
Operating System-Controlled Task Switch . . . . . . . . . . . . . . . . . . . . . . . . . . 11-9
Figure 12-1. SMRAM Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-5
Figure 12-2. SMM Revision Identifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-13
Figure 12-3. Auto HALT Restart Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-13
Figure 12-4. SMBASE Relocation Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-15
Figure 12-5. I/O Instruction Restart Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-16
Figure 13-1. Machine-Check MSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-2
Figure 13-2. MCG_CAP Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-3
Figure 13-3. MCG_STATUS Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-3
Figure 13-4. MCi_CTL Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-4
Figure 13-5. MCi_STATUS Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-5
Figure 13-6. Machine-Check Bank Address Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-6
Figure 14-1. Stack and Memory Layout of Static Variables . . . . . . . . . . . . . . . . . . . . . . 14-11
Figure 14-2. Pipeline Example of AGI Stall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-29
Figure 15-1. Debug Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-3
Figure 15-2. DebugCtlMSR Register. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-12
Figure 15-3. PerfEvtSel0 and PerfEvtSel1 MSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-17
Figure 15-4. CESR MSR (Pentium® Processor Only) . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-21
Figure 16-1. Real-Address Mode Address Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . 16-4
Figure 16-2. Interrupt Vector Table in Real-Address Mode. . . . . . . . . . . . . . . . . . . . . . . . 16-7
Figure 16-3. Entering and Leaving Virtual-8086 Mode . . . . . . . . . . . . . . . . . . . . . . . . . . 16-12
Figure 16-4. Privilege Level 0 Stack After Interrupt or Exception in Virtual-8086 Mode . 16-18
Figure 16-5. Software Interrupt Redirection Bit Map in TSS . . . . . . . . . . . . . . . . . . . . . . 16-25
Figure 17-1. Stack after Far 16- and 32-Bit Calls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-6
Figure 18-1. I/O Map Base Address Differences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-30

xix
TABLE OF FIGURES

xx
 TABLE OF TABLES

Table 2-1. Action Taken for Combinations of EM, MP, TS, CR4.OSFXSR,
and CPUID.XMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-15
Table 2-2. Summary of System Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-19
Table 3-1. Code- and Data-Segment Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-14
Table 3-2. System-Segment and Gate-Descriptor Types . . . . . . . . . . . . . . . . . . . . . . . .3-16
Table 3-3. Page Sizes and Physical Address Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-20
Table 3-4. Paging Modes and Physical Address Size . . . . . . . . . . . . . . . . . . . . . . . . . . .3-37
Table 4-1. Privilege Check Rules for Call Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-19
Table 4-2. Combined Page-Directory and Page-Table Protection. . . . . . . . . . . . . . . . . .4-33
Table 5-1. Protected-Mode Exceptions and Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . .5-6
Table 5-2. SIMD Floating-Point Exceptions Priority. . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-11
Table 5-3. Priority Among Simultaneous Exceptions and Interrupts . . . . . . . . . . . . . . . .5-12
Table 5-4. Interrupt and Exception Classes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-32
Table 5-5. Conditions for Generating a Double Fault . . . . . . . . . . . . . . . . . . . . . . . . . . .5-33
Table 5-6. Invalid TSS Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-35
Table 5-7. Alignment Requirements by Data Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-50
Table 6-1. Exception Conditions Checked During a Task Switch . . . . . . . . . . . . . . . . . .6-13
Table 6-2. Effect of a Task Switch on Busy Flag, NT Flag, Previous Task Link Field,
and TS Flag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-15
Table 7-1. Local APIC Register Address Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-18
Table 7-2. Valid Combinations for the APIC Interrupt Command Register . . . . . . . . . . .7-29
Table 7-3. EOI Message (14 Cycles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-37
Table 7-4. Short Message (21 Cycles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-38
Table 7-5. Nonfocused Lowest Priority Message (34 Cycles) . . . . . . . . . . . . . . . . . . . .7-39
Table 7-6. APIC Bus Status Cycles Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-40
Table 7-7. Types of Boot Phase IPIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-47
Table 7-8. Boot Phase IPI Message Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-47
Table 8-1. 32-Bit Intel Architecture Processor States Following Power-up,
Reset, or INIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-3
Table 8-2. Recommended Settings of EM and MP Flags on Intel
Architecture Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-7
Table 8-3. Software Emulation Settings of EM, MP, and NE Flags . . . . . . . . . . . . . . . . . .8-8
Table 8-4. Main Initialization Steps in STARTUP.ASM Source Listing . . . . . . . . . . . . . .8-18
Table 8-5. Relationship Between BLD Item and ASM Source File . . . . . . . . . . . . . . . . .8-31
Table 8-6. P6 Family Processor MSR Register Components . . . . . . . . . . . . . . . . . . . . .8-33
Table 8-7. Microcode Update Encoding Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-34
Table 8-8. Microcode Update Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-43
Table 8-9. Parameters for the Presence Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-44
Table 8-10. Parameters for the Write Update Data Function. . . . . . . . . . . . . . . . . . . . . . .8-45
Table 8-11. Parameters for the Control Update Sub-function . . . . . . . . . . . . . . . . . . . . . .8-48
Table 8-12. Mnemonic Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-48
Table 8-13. Parameters for the Read Microcode Update Data Function. . . . . . . . . . . . . .8-49
Table 8-14. Return Code Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8-50
Table 9-1. Characteristics of the Caches, TLBs, and Write Buffer in
Intel Architecture Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-3
Table 9-2. Methods of Caching Available in P6 Family, Pentium ®,
and Intel486™ Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-6
Table 9-3. MESI Cache Line States. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-9
Table 9-4. Cache Operating Modes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-11

xxi
TABLE OF TABLES

Table 9-5. Effective Memory Type Depending on MTRR, PCD, and PWT Settings . . . .9-14
Table 9-6. MTRR Memory Types and Their Properties . . . . . . . . . . . . . . . . . . . . . . . . . .9-19
Table 9-7. Address Mapping for Fixed-Range MTRRs . . . . . . . . . . . . . . . . . . . . . . . . . .9-23
Table 9-8. PAT Indexing and Values After Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-35
Table 9-9. Effective Memory Type Depending on MTRRs and PAT . . . . . . . . . . . . . . . .9-37
Table 9-10. PAT Memory Types and Their Properties . . . . . . . . . . . . . . . . . . . . . . . . . . .9-38
Table 10-1. Effects of MMX™ Instructions on FPU State . . . . . . . . . . . . . . . . . . . . . . . . .10-3
Table 10-2. Effect of the MMX™ and Floating-Point Instructions on the
FPU Tag Word . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10-3
Table 11-1. SIMD Floating-point Register Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-2
Table 11-2. Rounding Control Field (RC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-4
Table 11-3. Rounding of Positive Numbers Greater than the
Maximum Positive Finite Value. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-5
Table 11-4. Rounding of Negative Numbers Smaller than the
Maximum Negative Finite Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-5
Table 11-5. CPUID Bits for Streaming SIMD Extensions Support . . . . . . . . . . . . . . . . . .11-6
Table 11-6. CR4 Bits for Streaming SIMD Extensions Support . . . . . . . . . . . . . . . . . . . .11-6
Table 11-7. Streaming SIMD Extensions Faults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11-12
Table 11-8. Invalid Arithmetic Operations and the Masked Responses to Them . . . . . .11-18
Table 11-9. Masked Responses to Numeric Overflow . . . . . . . . . . . . . . . . . . . . . . . . . .11-20
Table 12-1. SMRAM State Save Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-5
Table 12-2. Processor Register Initialization in SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-9
Table 12-3. Auto HALT Restart Flag Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-14
Table 12-4. I/O Instruction Restart Field Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-16
Table 13-1. Simple Error Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13-9
Table 13-2. General Forms of Compound Error Codes. . . . . . . . . . . . . . . . . . . . . . . . . . .13-9
Table 13-3. Encoding for TT (Transaction Type) Sub-Field. . . . . . . . . . . . . . . . . . . . . . .13-10
Table 13-4. Level Encoding for LL (Memory Hierarchy Level) Sub-Field . . . . . . . . . . . .13-10
Table 13-5. Encoding of Request (RRRR) Sub-Field . . . . . . . . . . . . . . . . . . . . . . . . . . .13-10
Table 13-6. Encodings of PP, T, and II Sub-Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13-11
Table 13-7. Encoding of the MCi_STATUS Register for External Bus Errors . . . . . . . .13-11
Table 14-1. Small and Large General-Purpose Register Pairs . . . . . . . . . . . . . . . . . . . . .14-7
Table 14-2. Pairable Integer Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14-14
Table 15-1. Breakpointing Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-7
Table 15-2. Debug Exception Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15-8
Table 16-1. Real-Address Mode Exceptions and Interrupts . . . . . . . . . . . . . . . . . . . . . .16-8
Table 16-2. Software Interrupt Handling Methods While in Virtual-8086 Mode . . . . . . . .16-24
Table 17-1. Characteristics of 16-Bit and 32-Bit Program Modules. . . . . . . . . . . . . . . . . .17-1
Table 18-1. New Instructions in the Pentium® and Later Intel Architecture Processors . .18-3
Table 18-1. Recommended Values of the FP Related Bits for Intel486™ SX
Microprocessor/Intel 487 SX Math Coprocessor System . . . . . . . . . . . . . . .18-20
Table 18-2. EM and MP Flag Interpretation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-20
Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-2
Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-12
Table B-1. Model-Specific Registers (MSRs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1

xxii
 1
About This Manual
 CHAPTER 1
ABOUT THIS MANUAL

The Intel Architecture Software Developer’s Manual, Volume 2: Instruction Set Reference
(Order Number 243191) is part of a three-volume set that describes the architecture and
programming environment of all Intel Architecture processors. The other two volumes in this
set are:
• The Intel Architecture Software Developer’s Manual, Volume 1: Basic Architecture (Order
Number 243190).
• The Intel Architecture Software Developer’s Manual, Volume 3: System Programing Guide
(Order Number 243192).
The Intel Architecture Software Developer’s Manual, Volume 1, describes the basic architecture
and programming environment of an Intel Architecture processor; the Intel Architecture Soft-
ware Developer’s Manual, Volume 2, describes the instructions set of the processor and the
opcode structure. These two volumes are aimed at application programmers who are writing
programs to run under existing operating systems or executives. The Intel Architecture Software
Developer’s Manual, Volume 3, describes the operating-system support environment of an Intel
Architecture processor, including memory management, protection, task management, interrupt
and exception handling, and system management mode. It also provides Intel Architecture
processor compatibility information. This volume is aimed at operating-system and BIOS
designers and programmers.

1.1. P6 FAMILY PROCESSOR TERMINOLOGY

This manual includes information pertaining primarily to the 32-bit Intel Architecture proces-
sors, which include the Intel386™, Intel486™, and Pentium® processors, and the P6 family
processors. The P6 family processors are those Intel Architecture processors based on the P6
family microarchitecture. This family includes the Pentium® Pro, Pentium® II, Pentium® III
processor, and any future processors based on the P6 family microarchitecture.

1.2. OVERVIEW OF THE INTEL ARCHITECTURE SOFTWARE

DEVELOPER’S MANUAL, VOLUME 3: SYSTEM
PROGRAMMING GUIDE
The contents of this manual are as follows:
Chapter 1 — About This Manual. Gives an overview of all three volumes of the Intel Archi-
tecture Software Developer’s Manual. It also describes the notational conventions in these
manuals and lists related Intel manuals and documentation of interest to programmers and hard-
ware designers.

1-1
ABOUT THIS MANUAL

Chapter 2 — System Architecture Overview. Describes the modes of operation of an Intel

Architecture processor and the mechanisms provided in the Intel Architecture to support oper-
ating systems and executives, including the system-oriented registers and data structures and the
system-oriented instructions. The steps necessary for switching between real-address and
protected modes are also identified.
Chapter 3 — Protected-Mode Memory Management. Describes the data structures, registers,
and instructions that support segmentation and paging and explains how they can be used to
implement a “flat” (unsegmented) memory model or a segmented memory model.
Chapter 4 — Protection. Describes the support for page and segment protection provided in
the Intel Architecture. This chapter also explains the implementation of privilege rules, stack
switching, pointer validation, user and supervisor modes.
Chapter 5 — Interrupt and Exception Handling. Describes the basic interrupt mechanisms
defined in the Intel Architecture, shows how interrupts and exceptions relate to protection, and
describes how the architecture handles each exception type. Reference information for each
Intel Architecture exception is given at the end of this chapter.
Chapter 6 — Task Management. Describes the mechanisms the Intel Architecture provides to
support multitasking and inter-task protection.
Chapter 7 — Multiple-Processor Management. Describes the instructions and flags that
support multiple processors with shared memory, memory ordering, and the advanced program-
mable interrupt controller (APIC).
Chapter 8 — Processor Management and Initialization. Defines the state of an Intel Archi-
tecture processor and its floating-point and SIMD floating-point units after reset initialization.
This chapter also explains how to set up an Intel Architecture processor for real-address mode
operation and protected- mode operation, and how to switch between modes.
Chapter 9 — Memory Cache Control. Describes the general concept of caching and the
caching mechanisms supported by the Intel Architecture. This chapter also describes the
memory type range registers (MTRRs) and how they can be used to map memory types of phys-
ical memory. MTRRs were introduced into the Intel Architecture with the Pentium® Pro
processor. It also presents information on using the new cache control and memory streaming
instructions introduced with the Pentium® III processor.
Chapter 10 — MMX™ Technology System Programming. Describes those aspects of the
Intel MMX™ technology that must be handled and considered at the system programming level,
including task switching, exception handling, and compatibility with existing system environ-
ments. The MMX™ technology was introduced into the Intel Architecture with the Pentium®
processor.
Chapter 11 — Streaming SIMD Extensions System Programming. Describes those aspects
of Streaming SIMD Extensions that must be handled and considered at the system programming
level, including task switching, exception handling, and compatibility with existing system
environments. Streaming SIMD Extensions were introduced into the Intel Architecture with the
Pentium® processor.
Chapter 12 — System Management Mode (SMM). Describes the Intel Architecture’s system
management mode (SMM), which can be used to implement power management functions.

1-2
 ABOUT THIS MANUAL

Chapter 13 — Machine-Check Architecture. Describes the machine-check architecture,

which was introduced into the Intel Architecture with the Pentium® processor.
Chapter 14 — Code Optimization. Discusses general optimization techniques for program-
ming an Intel Architecture processor.
Chapter 15 — Debugging and Performance Monitoring. Describes the debugging registers
and other debug mechanism provided in the Intel Architecture. This chapter also describes the
time-stamp counter and the performance-monitoring counters.
Chapter 16 — 8086 Emulation. Describes the real-address and virtual-8086 modes of the Intel
Architecture.
Chapter 17 — Mixing 16-Bit and 32-Bit Code. Describes how to mix 16-bit and 32-bit code
modules within the same program or task.
Chapter 18 — Intel Architecture Compatibility. Describes the programming differences
between the Intel 286, Intel386™, Intel486™, Pentium®, and P6 family processors. The differ-
ences among the 32-bit Intel Architecture processors (the Intel386™, Intel486™, Pentium®, and
P6 family processors) are described throughout the three volumes of the Intel Architecture Soft-
ware Developer’s Manual, as relevant to particular features of the architecture. This chapter
provides a collection of all the relevant compatibility information for all Intel Architecture
processors and also describes the basic differences with respect to the 16-bit Intel Architecture
processors (the Intel 8086 and Intel 286 processors).
Appendix A — Performance-Monitoring Events. Lists the events that can be counted with
the performance-monitoring counters and the codes used to select these events. Both Pentium®
processor and P6 family processor events are described.
Appendix B — Model-Specific Registers (MSRs). Lists the MSRs available in the Pentium®
and P6 family processors and their functions.
Appendix C — Dual-Processor (DP) Bootup Sequence Example (Specific to Pentium®
Processors). Gives an example of how to use the DP protocol to boot two Pentium® processors
(a primary processor and a secondary processor) in a DP system and initialize their APICs.
Appendix D — Multiple-Processor (MP) Bootup Sequence Example (Specific to P6 Family
Processors). Gives an example of how to use of the MP protocol to boot two P6 family proces-
sors in a MP system and initialize their APICs.
Appendix E — Programming the LINT0 and LINT1 Inputs. Gives an example of how to
program the LINT0 and LINT1 pins for specific interrupt vectors.

1.3. OVERVIEW OF THE INTEL ARCHITECTURE SOFTWARE

DEVELOPER’S MANUAL, VOLUME 1: BASIC
ARCHITECTURE
The contents of the Intel Architecture Software Developer’s Manual, Volume 1 are as follows:
Chapter 1 — About This Manual. Gives an overview of all three volumes of the Intel Archi-
tecture Software Developer’s Manual. It also describes the notational conventions in these

1-3
ABOUT THIS MANUAL

manuals and lists related Intel manuals and documentation of interest to programmers and hard-
ware designers.
Chapter 2 — Introduction to the Intel Architecture. Introduces the Intel Architecture and the
families of Intel processors that are based on this architecture. It also gives an overview of the
common features found in these processors and brief history of the Intel Architecture.
Chapter 3 — Basic Execution Environment. Introduces the models of memory organization
and describes the register set used by applications.
Chapter 4 — Procedure Calls, Interrupts, and Exceptions. Describes the procedure stack
and the mechanisms provided for making procedure calls and for servicing interrupts and
exceptions.
Chapter 5 — Data Types and Addressing Modes. Describes the data types and addressing
modes recognized by the processor.
Chapter 6 — Instruction Set Summary. Gives an overview of all the Intel Architecture
instructions except those executed by the processor’s floating-point unit. The instructions are
presented in functionally related groups.
Chapter 7 — Floating-Point Unit. Describes the Intel Architecture floating-point unit,
including the floating-point registers and data types; gives an overview of the floating-point
instruction set; and describes the processor’s floating-point exception conditions.
Chapter 8 — Programming with the Intel MMX™ Technology. Describes the Intel MMX™
technology, including MMX™ registers and data types, and gives an overview of the MMX™
instruction set.
Chapter 9 — Programming with the Streaming SIMD Extensions. Describes the Intel
Streaming SIMD Extensions, including the registers and data types.
Chapter 10— Input/Output. Describes the processor’s I/O architecture, including I/O port
addressing, the I/O instructions, and the I/O protection mechanism.
Chapter 11 — Processor Identification and Feature Determination. Describes how to deter-
mine the CPU type and the features that are available in the processor.
Appendix A — EFLAGS Cross-Reference. Summarizes how the Intel Architecture instruc-
tions affect the flags in the EFLAGS register.
Appendix B — EFLAGS Condition Codes. Summarizes how the conditional jump, move, and
byte set on condition code instructions use the condition code flags (OF, CF, ZF, SF, and PF) in
the EFLAGS register.
Appendix C — Floating-Point Exceptions Summary. Summarizes the exceptions that can be
raised by floating-point instructions.
Appendix D — SIMD Floating-Point Exceptions Summary. Provides the Streaming SIMD
Extensions mnemonics, and the exceptions that each instruction can cause.
Appendix E — Guidelines for Writing FPU Exception Handlers. Describes how to design
and write MS-DOS* compatible exception handling facilities for FPU and SIMD floating-point
exceptions, including both software and hardware requirements and assembly-language code

1-4
 ABOUT THIS MANUAL

examples. This appendix also describes general techniques for writing robust FPU exception
handlers.
Appendix F — Guidelines for Writing SIMD-FP Exception Handlers. Provides guidelines
for the Streaming SIMD Extensions instructions that can generate numeric (floating-point)
exceptions, and gives an overview of the necessary support for handling such exceptions.

1.4. OVERVIEW OF THE INTEL ARCHITECTURE SOFTWARE

DEVELOPER’S MANUAL, VOLUME 2: INSTRUCTION SET
REFERENCE
The contents of the Intel Architecture Software Developer’s Manual, Volume 2, are as follows:
Chapter 1 — About This Manual. Gives an overview of all three volumes of the Intel Archi-
tecture Software Developer’s Manual. It also describes the notational conventions in these
manuals and lists related Intel manuals and documentation of interest to programmers and hard-
ware designers.
Chapter 2 — Instruction Format. Describes the machine-level instruction format used for all
Intel Architecture instructions and gives the allowable encodings of prefixes, the operand-iden-
tifier byte (ModR/M byte), the addressing-mode specifier byte (SIB byte), and the displacement
and immediate bytes.
Chapter 3 — Instruction Set Reference. Describes each of the Intel Architecture instructions
in detail, including an algorithmic description of operations, the effect on flags, the effect of
operand- and address-size attributes, and the exceptions that may be generated. The instructions
are arranged in alphabetical order. The FPU, MMX™ Technology instructions, and Streaming
SIMD Extensions are included in this chapter.
Appendix A — Opcode Map. Gives an opcode map for the Intel Architecture instruction set.
Appendix B — Instruction Formats and Encodings. Gives the binary encoding of each form
of each Intel Architecture instruction.
Appendix C — Compiler Intrinsics and Functional Equivalents. Gives the Intel C/C++
compiler intrinsics and functional equivalents for the MMX™ Technology instructions and
Streaming SIMD Extensions.

1.5. NOTATIONAL CONVENTIONS

This manual uses special notation for data-structure formats, for symbolic representation of
instructions, and for hexadecimal numbers. A review of this notation makes the manual easier
to read.

1-5
ABOUT THIS MANUAL

1.5.1. Bit and Byte Order

In illustrations of data structures in memory, smaller addresses appear toward the bottom of the
figure; addresses increase toward the top. Bit positions are numbered from right to left. The
numerical value of a set bit is equal to two raised to the power of the bit position. Intel Archi-
tecture processors are “little endian” machines; this means the bytes of a word are numbered
starting from the least significant byte. Figure 1-1 illustrates these conventions.

1.5.2. Reserved Bits and Software Compatibility

In many register and memory layout descriptions, certain bits are marked as reserved. When
bits are marked as reserved, it is essential for compatibility with future processors that software
treat these bits as having a future, though unknown, effect. The behavior of reserved bits should
be regarded as not only undefined, but unpredictable. Software should follow these guidelines
in dealing with reserved bits:
• Do not depend on the states of any reserved bits when testing the values of registers which
contain such bits. Mask out the reserved bits before testing.
• Do not depend on the states of any reserved bits when storing to memory or to a register.
• Do not depend on the ability to retain information written into any reserved bits.
• When loading a register, always load the reserved bits with the values indicated in the
documentation, if any, or reload them with values previously read from the same register.

NOTE
Avoid any software dependence upon the state of reserved bits in Intel Archi-
tecture registers. Depending upon the values of reserved register bits will
make software dependent upon the unspecified manner in which the
processor handles these bits. Programs that depend upon reserved values risk
incompatibility with future processors.

Data Structure
Highest 24 23 8 7 Bit offset
31 16 15 0
Address
28
24
20
16
12
8
4
Lowest
Byte 3 Byte 2 Byte 1 Byte 0 0 Address

Byte Offset

Figure 1-1. Bit and Byte Order

1-6
 ABOUT THIS MANUAL

1.5.3. Instruction Operands

When instructions are represented symbolically, a subset of the Intel Architecture assembly
language is used. In this subset, an instruction has the following format:
label: mnemonic argument1, argument2, argument3
where:
• A label is an identifier which is followed by a colon.
• A mnemonic is a reserved name for a class of instruction opcodes which have the same
function.
• The operands argument1, argument2, and argument3 are optional. There may be from
zero to three operands, depending on the opcode. When present, they take the form of
either literals or identifiers for data items. Operand identifiers are either reserved names of
registers or are assumed to be assigned to data items declared in another part of the
program (which may not be shown in the example).
When two operands are present in an arithmetic or logical instruction, the right operand is the
source and the left operand is the destination.
For example:
LOADREG: MOV EAX, SUBTOTAL
In this example, LOADREG is a label, MOV is the mnemonic identifier of an opcode, EAX is
the destination operand, and SUBTOTAL is the source operand. Some assembly languages put
the source and destination in reverse order.

1.5.4. Hexadecimal and Binary Numbers

Base 16 (hexadecimal) numbers are represented by a string of hexadecimal digits followed by
the character H (for example, F82EH). A hexadecimal digit is a character from the following
set: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, A, B, C, D, E, and F.
Base 2 (binary) numbers are represented by a string of 1s and 0s, sometimes followed by the
character B (for example, 1010B). The “B” designation is only used in situations where confu-
sion as to the type of number might arise.

1.5.5. Segmented Addressing

The processor uses byte addressing. This means memory is organized and accessed as a
sequence of bytes. Whether one or more bytes are being accessed, a byte address is used to
locate the byte or bytes of memory. The range of memory that can be addressed is called an
address space.
The processor also supports segmented addressing. This is a form of addressing where a
program may have many independent address spaces, called segments. For example, a program
can keep its code (instructions) and stack in separate segments. Code addresses would always

1-7
ABOUT THIS MANUAL

refer to the code space, and stack addresses would always refer to the stack space. The following
notation is used to specify a byte address within a segment:
Segment-register:Byte-address
For example, the following segment address identifies the byte at address FF79H in the segment
pointed by the DS register:
DS:FF79H
The following segment address identifies an instruction address in the code segment. The CS
register points to the code segment and the EIP register contains the address of the instruction.
CS:EIP

1.5.6. Exceptions
An exception is an event that typically occurs when an instruction causes an error. For example,
an attempt to divide by zero generates an exception. However, some exceptions, such as break-
points, occur under other conditions. Some types of exceptions may provide error codes. An
error code reports additional information about the error. An example of the notation used to
show an exception and error code is shown below.
#PF(fault code)
This example refers to a page-fault exception under conditions where an error code naming a
type of fault is reported. Under some conditions, exceptions which produce error codes may not
be able to report an accurate code. In this case, the error code is zero, as shown below for a
general-protection exception.
#GP(0)
Refer to Chapter 5, Interrupt and Exception Handling, for a list of exception mnemonics and
their descriptions.

1-8
 ABOUT THIS MANUAL

1.6. RELATED LITERATURE

The following books contain additional material related to Intel processors:
• Intel Pentium® II Processor Specification Update, Order Number 243337-010.
• Intel Pentium® Pro Processor Specification Update, Order Number 242689-031.
• Intel Pentium® Processor Specification Update, Order Number 242480.
• AP-485, Intel Processor Identification and the CPUID Instruction, Order Number 241618-
006.
• AP-578, Software and Hardware Considerations for FPU Exception Handlers for Intel
Architecture Processors, Order Number 243291.
• Pentium® Pro Processor Data Book, Order Number 242690.
• Pentium® Pro BIOS Writer’s Guide, http://www.intel.com/procs/ppro/info/index.htm.
• Pentium® Processor Data Book, Order Number 241428.
• 82496 Cache Controller and 82491 Cache SRAM Data Book For Use With the Pentium®
Processor, Order Number 241429.
• Intel486™ Microprocessor Data Book, Order Number 240440.
• Intel486™ SX CPU/Intel487™ SX Math Coprocessor Data Book, Order Number 240950.
• Intel486™ DX2 Microprocessor Data Book, Order Number 241245.
• Intel486™ Microprocessor Product Brief Book, Order Number 240459.
• Intel386™ Processor Hardware Reference Manual, Order Number 231732.
• Intel386™ Processor System Software Writer's Guide, Order Number 231499.
• Intel386™ High-Performance 32-Bit CHMOS Microprocessor with Integrated Memory
Management, Order Number 231630.
• 376 Embedded Processor Programmer’s Reference Manual, Order Number 240314.
• 80387 DX User’s Manual Programmer’s Reference, Order Number 231917.
• 376 High-Performance 32-Bit Embedded Processor, Order Number 240182.
• Intel386™ SX Microprocessor, Order Number 240187.
• Intel Architecture Optimization Manual, Order Number 242816-002.

1-9
ABOUT THIS MANUAL

1-10
 2
System Architecture
Overview
 SYSTEM ARCHITECTURE OVERVIEW

CHAPTER 2
SYSTEM ARCHITECTURE OVERVIEW

The 32-bit members of the Intel Architecture family of processors provide extensive support for
operating-system and system-development software. This support is part of the processor’s
system-level architecture and includes features to assist in the following operations:
• Memory management
• Protection of software modules
• Multitasking
• Exception and interrupt handling
• Multiprocessing
• Cache management
• Hardware resource and power management
• Debugging and performance monitoring
This chapter provides a brief overview of the processor’s system-level architecture; a detailed
description of each part of this architecture given in the following chapters. This chapter also
describes the system registers that are used to set up and control the processor at the system level
and gives a brief overview of the processor’s system-level (operating system) instructions.
Many of the system-level architectural features of the processor are used only by system
programmers. Application programmers may need to read this chapter, and the following chap-
ters which describe the use of these features, in order to understand the hardware facilities used
by system programmers to create a reliable and secure environment for application programs.

NOTE
This overview and most of the subsequent chapters of this book focus on the
“native” or protected-mode operation of the 32-bit Intel Architecture
processors. As described in Chapter 8, Processor Management and Initial-
ization, all Intel Architecture processors enter real-address mode following a
power-up or reset. Software must then initiate a switch from real-address
mode to protected mode.

2.1. OVERVIEW OF THE SYSTEM-LEVEL ARCHITECTURE

The Intel Architecture’s system architecture consists of a set of registers, data structures, and
instructions designed to support basic system-level operations such as memory management,
interrupt and exception handling, task management, and control of multiple processors (multi-
processing). Figure 2-1 provides a generalized summary of the system registers and data
structures.

2-1
SYSTEM ARCHITECTURE OVERVIEW

EFLAGS Register Physical Address

Code, Data or
Linear Address Stack Segment
Control Registers
Task-State
CR4 Segment Selector Segment (TSS)
CR3 Task
CR2 Code
Register
CR1 Data
CR0 Stack
MXCSR1 Global Descriptor
Task Register Table (GDT)

Interrupt Handler
Segment Sel. Seg. Desc.
Code
Current
Interrupt TSS Seg. Sel. TSS Desc. TSS Stack
Vector
Seg. Desc.
Interrupt Descriptor Task-State
Table (IDT) Segment (TSS) Task
TSS Desc.
Code
Interrupt Gate LTD Desc. Data
Stack
Task Gate
GDTR
Trap Gate
Local Descriptor Exception Handler
Table (LDT) Code
Current
TSS Stack
IDTR Call-Gate Seg. Desc.
Segment Selector
Call Gate
Protected Procedure
Code
LDTR Current
TSS Stack

Linear Address Space Linear Address

Dir Table Offset

Linear Addr.
Page Directory Page Table Page

Physical Addr.
Pg. Dir. Entry Pg. Tbl. Entry

0 This page mapping example is for 4-KByte pages

CR3* and the normal 32-bit physical address size.
*Physical Address
1. MXCSR is new control/status register in the Pentium® III processor.

Figure 2-1. System-Level Registers and Data Structures

2-2
 SYSTEM ARCHITECTURE OVERVIEW

2.1.1. Global and Local Descriptor Tables

When operating in protected mode, all memory accesses pass through either the global
descriptor table (GDT) or the (optional) local descriptor table (LDT), shown in Figure 2-1.
These tables contain entries called segment descriptors. A segment descriptor provides the base
address of a segment and access rights, type, and usage information. Each segment descriptor
has a segment selector associated with it. The segment selector provides an index into the GDT
or LDT (to its associated segment descriptor), a global/local flag (that determines whether the
segment selector points to the GDT or the LDT), and access rights information.
To access a byte in a segment, both a segment selector and an offset must be supplied. The
segment selector provides access to the segment descriptor for the segment (in the GDT or
LDT). From the segment descriptor, the processor obtains the base address of the segment in the
linear address space. The offset then provides the location of the byte relative to the base
address. This mechanism can be used to access any valid code, data, or stack segment in the
GDT or LDT, provided the segment is accessible from the current privilege level (CPL) at which
the processor is operating. (The CPL is defined as the protection level of the currently executing
code segment.)
In Figure 2-1 the solid arrows indicate a linear address, the dashed lines indicate a segment
selector, and the dotted arrows indicate a physical address. For simplicity, many of the segment
selectors are shown as direct pointers to a segment. However, the actual path from a segment
selector to its associated segment is always through the GDT or LDT.
The linear address of the base of the GDT is contained in the GDT register (GDTR); the linear
address of the LDT is contained in the LDT register (LDTR).

2.1.2. System Segments, Segment Descriptors, and Gates

Besides the code, data, and stack segments that make up the execution environment of a program
or procedure, the system architecture also defines two system segments: the task-state segment
(TSS) and the LDT. (The GDT is not considered a segment because it is not accessed by means
of a segment selector and segment descriptor.) Each of these segment types has a segment
descriptor defined for it.
The system architecture also defines a set of special descriptors called gates (the call gate, inter-
rupt gate, trap gate, and task gate) that provide protected gateways to system procedures and
handlers that operate at different privilege levels than application programs and procedures.
For example, a CALL to a call gate provides access to a procedure in a code segment that is at
the same or numerically lower privilege level (more privileged) than the current code segment.
To access a procedure through a call gate, the calling procedure1 must supply the selector of the
call gate. The processor than performs an access rights check on the call gate, comparing the
CPL with the privilege level of the call gate and the destination code segment pointed to by the
call gate. If access to the destination code segment is allowed, the processor gets the segment
selector for the destination code segment and an offset into that code segment from the call gate.

1. The word “procedure” is commonly used in this document as a general term for a logical unit or block of
code (such as a program, procedure, function, or routine). The term is not restricted to the definition of a
procedure in the Intel Architecture assembly language.

2-3
SYSTEM ARCHITECTURE OVERVIEW

If the call requires a change in privilege level, the processor also switches to the stack for that
privilege level. (The segment selector for the new stack is obtained from the TSS for the
currently running task.) Gates also facilitate transitions between 16-bit and 32-bit code
segments, and vice versa.

2.1.3. Task-State Segments and Task Gates

The TSS (refer to Figure 2-1) defines the state of the execution environment for a task. It
includes the state of the general-purpose registers, the segment registers, the EFLAGS register,
the EIP register, and segment selectors and stack pointers for three stack segments (one stack
each for privilege levels 0, 1, and 2). It also includes the segment selector for the LDT associated
with the task and the page-table base address.
All program execution in protected mode happens within the context of a task, called the current
task. The segment selector for the TSS for the current task is stored in the task register. The
simplest method of switching to a task is to make a call or jump to the task. Here, the segment
selector for the TSS of the new task is given in the CALL or JMP instruction. In switching tasks,
the processor performs the following actions:
1. Stores the state of the current task in the current TSS.
2. Loads the task register with the segment selector for the new task.
3. Accesses the new TSS through a segment descriptor in the GDT.
4. Loads the state of the new task from the new TSS into the general-purpose registers, the
segment registers, the LDTR, control register CR3 (page-table base address), the EFLAGS
register, and the EIP register.
5. Begins execution of the new task.
A task can also be accessed through a task gate. A task gate is similar to a call gate, except that
it provides access (through a segment selector) to a TSS rather than a code segment.

2.1.4. Interrupt and Exception Handling

External interrupts, software interrupts, and exceptions are handled through the interrupt
descriptor table (IDT), refer to Figure 2-1. The IDT contains a collection of gate descriptors,
which provide access to interrupt and exception handlers. Like the GDT, the IDT is not a
segment. The linear address of the base of the IDT is contained in the IDT register (IDTR).
The gate descriptors in the IDT can be of the interrupt-, trap-, or task-gate type. To access an
interrupt or exception handler, the processor must first receive an interrupt vector (interrupt
number) from internal hardware, an external interrupt controller, or from software by means of
an INT, INTO, INT 3, or BOUND instruction. The interrupt vector provides an index into the
IDT to a gate descriptor. If the selected gate descriptor is an interrupt gate or a trap gate, the asso-
ciated handler procedure is accessed in a manner very similar to calling a procedure through a
call gate. If the descriptor is a task gate, the handler is accessed through a task switch.

2-4
 SYSTEM ARCHITECTURE OVERVIEW

2.1.5. Memory Management

The system architecture supports either direct physical addressing of memory or virtual memory
(through paging). When physical addressing is used, a linear address is treated as a physical
address. When paging is used, all the code, data, stack, and system segments and the GDT and
IDT can be paged, with only the most recently accessed pages being held in physical memory.
The location of pages (or page frames as they are sometimes called in the Intel Architecture) in
physical memory is contained in two types of system data structures (a page directory and a set
of page tables), both of which reside in physical memory (refer to Figure 2-1). An entry in a page
directory contains the physical address of the base of a page table, access rights, and memory
management information. An entry in a page table contains the physical address of a page frame,
access rights, and memory management information. The base physical address of the page
directory is contained in control register CR3.
To use this paging mechanism, a linear address is broken into three parts, providing separate
offsets into the page directory, the page table, and the page frame.
A system can have a single page directory or several. For example, each task can have its own
page directory.

2.1.6. System Registers

To assist in initializing the processor and controlling system operations, the system architecture
provides system flags in the EFLAGS register and several system registers:
• The system flags and IOPL field in the EFLAGS register control task and mode switching,
interrupt handling, instruction tracing, and access rights. Refer to Section 2.3., “System
Flags and Fields in the EFLAGS Register” for a description of these flags.
• The control registers (CR0, CR2, CR3, and CR4) contain a variety of flags and data fields
for controlling system-level operations. With the introduction of the Pentium® III
processor, CR4 now contains bits indicating support Pentium® III processor specific
capabilities within the OS. Refer to Section 2.5., “Control Registers” for a description of
these flags.
• The debug registers (not shown in Figure 2-1) allow the setting of breakpoints for use in
debugging programs and systems software. Refer to Chapter 15, Debugging and
Performance Monitoring, for a description of these registers.
• The GDTR, LDTR, and IDTR registers contain the linear addresses and sizes (limits) of
their respective tables. Refer to Section 2.4., “Memory-Management Registers” for a
description of these registers.
• The task register contains the linear address and size of the TSS for the current task. Refer
to Section 2.4., “Memory-Management Registers” for a description of this register.
• Model-specific registers (not shown in Figure 2-1).
The model-specific registers (MSRs) are a group of registers available primarily to operating-
system or executive procedures (that is, code running at privilege level 0). These registers
control items such as the debug extensions, the performance-monitoring counters, the machine-
check architecture, and the memory type ranges (MTRRs). The number and functions of these

2-5
SYSTEM ARCHITECTURE OVERVIEW

registers varies among the different members of the Intel Architecture processor families.
Section 8.4., “Model-Specific Registers (MSRs)” in Chapter 8, Processor Management and
Initialization for more information about the MSRs and Appendix B, Model-Specific Registers
for a complete list of the MSRs.
Most systems restrict access to all system registers (other than the EFLAGS register) by appli-
cation programs. Systems can be designed, however, where all programs and procedures run at
the most privileged level (privilege level 0), in which case application programs are allowed to
modify the system registers.

2.1.7. Other System Resources

Besides the system registers and data structures described in the previous sections, the system
architecture provides the following additional resources:
• Operating system instructions (refer to Section 2.6., “System Instruction Summary”).
• Performance-monitoring counters (not shown in Figure 2-1).
• Internal caches and buffers (not shown in Figure 2-1).
The performance-monitoring counters are event counters that can be programmed to count
processor events such as the number of instructions decoded, the number of interrupts received,
or the number of cache loads. Refer to Section 15.6., “Performance-Monitoring Counters”, in
Chapter 15, Debugging and Performance Monitoring, for more information about these
counters.
The processor provides several internal caches and buffers. The caches are used to store both
data and instructions. The buffers are used to store things like decoded addresses to system and
application segments and write operations waiting to be performed. Refer to Chapter 9, Memory
Cache Control, for a detailed discussion of the processor’s caches and buffers.

2.2. MODES OF OPERATION

The Intel Architecture supports three operating modes and one quasi-operating mode:
• Protected mode. This is the native operating mode of the processor. In this mode all
instructions and architectural features are available, providing the highest performance and
capability. This is the recommended mode for all new applications and operating systems.
• Real-address mode. This operating mode provides the programming environment of the
Intel 8086 processor, with a few extensions (such as the ability to switch to protected or
system management mode).
• System management mode (SMM). The system management mode (SMM) is a standard
architectural feature in all Intel Architecture processors, beginning with the Intel386™ SL
processor. This mode provides an operating system or executive with a transparent
mechanism for implementing power management and OEM differentiation features. SMM
is entered through activation of an external system interrupt pin (SMI#), which generates a

2-6
 SYSTEM ARCHITECTURE OVERVIEW

system management interrupt (SMI). In SMM, the processor switches to a separate address
space while saving the context of the currently running program or task. SMM-specific
code may then be executed transparently. Upon returning from SMM, the processor is
placed back into its state prior to the SMI.
• Virtual-8086 mode. In protected mode, the processor supports a quasi-operating mode
known as virtual-8086 mode. This mode allows the processor to execute 8086 software in
a protected, multitasking environment.
Figure 2-2 shows how the processor moves among these operating modes.

Real-Address
Mode
Reset SMI#
or
Reset or RSM
PE=1
PE=0

SMI# System
Reset Protected Mode Management
Mode
RSM

VM=0 VM=1

SMI#
Virtual-8086
Mode
RSM

Figure 2-2. Transitions Among the Processor’s Operating Modes

The processor is placed in real-address mode following power-up or a reset. Thereafter, the PE
flag in control register CR0 controls whether the processor is operating in real-address or
protected mode (refer to Section 2.5., “Control Registers”). Refer to Section 8.8., “Mode
Switching” in Chapter 8, Processor Management and Initialization for detailed information on
switching between real-address mode and protected mode.
The VM flag in the EFLAGS register determines whether the processor is operating in protected
mode or virtual-8086 mode. Transitions between protected mode and virtual-8086 mode are
generally carried out as part of a task switch or a return from an interrupt or exception handler
(refer to Section 16.2.5., “Entering Virtual-8086 Mode” in Chapter 16, 8086 Emulation).
The processor switches to SMM whenever it receives an SMI while the processor is in real-
address, protected, or virtual-8086 modes. Upon execution of the RSM instruction, the
processor always returns to the mode it was in when the SMI occurred.

2-7
SYSTEM ARCHITECTURE OVERVIEW

2.3. SYSTEM FLAGS AND FIELDS IN THE EFLAGS REGISTER

The system flags and IOPL field of the EFLAGS register control I/O, maskable hardware inter-
rupts, debugging, task switching, and the virtual-8086 mode (refer to Figure 2-3). Only privi-
leged code (typically operating system or executive code) should be allowed to modify these
bits.
The functions of the system flags and IOPL are as follows:
TF Trap (bit 8). Set to enable single-step mode for debugging; clear to disable single-step
mode. In single-step mode, the processor generates a debug exception after each
instruction, which allows the execution state of a program to be inspected after each
instruction. If an application program sets the TF flag using a POPF, POPFD, or IRET
instruction, a debug exception is generated after the instruction that follows the POPF,
POPFD, or IRET instruction.

31 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
I
V V
I I I A V R 0 N O O D I T S Z A P C
Reserved (set to 0) D C M F T P F F F F F F 0 F 0 F 1 F
P F
L

ID — Identification Flag
VIP — Virtual Interrupt Pending
VIF — Virtual Interrupt Flag
AC — Alignment Check
VM — Virtual-8086 Mode
RF — Resume Flag
NT — Nested Task Flag
IOPL— I/O Privilege Level
IF — Interrupt Enable Flag
TF — Trap Flag
Reserved

Figure 2-3. System Flags in the EFLAGS Register

IF Interrupt enable (bit 9). Controls the response of the processor to maskable hardware
interrupt requests (refer to Section 5.1.1.2., “Maskable Hardware Interrupts” in
Chapter 5, Interrupt and Exception Handling). Set to respond to maskable hardware
interrupts; cleared to inhibit maskable hardware interrupts. The IF flag does not affect
the generation of exceptions or nonmaskable interrupts (NMI interrupts). The CPL,
IOPL, and the state of the VME flag in control register CR4 determine whether the IF
flag can be modified by the CLI, STI, POPF, POPFD, and IRET instructions.
IOPL I/O privilege level field (bits 12 and 13). Indicates the I/O privilege level (IOPL) of
the currently running program or task. The CPL of the currently running program or
task must be less than or equal to the IOPL to access the I/O address space. This field
can only be modified by the POPF and IRET instructions when operating at a CPL of
0. Refer to Chapter 10, Input/Output, of the Intel Architecture Software Developer’s
Manual, Volume 1, for more information on the relationship of the IOPL to I/O opera-
tions.
2-8
 SYSTEM ARCHITECTURE OVERVIEW

The IOPL is also one of the mechanisms that controls the modification of the IF flag
and the handling of interrupts in virtual-8086 mode when the virtual mode extensions
are in effect (the VME flag in control register CR4 is set).
NT Nested task (bit 14). Controls the chaining of interrupted and called tasks. The
processor sets this flag on calls to a task initiated with a CALL instruction, an interrupt,
or an exception. It examines and modifies this flag on returns from a task initiated with
the IRET instruction. The flag can be explicitly set or cleared with the POPF/POPFD
instructions; however, changing to the state of this flag can generate unexpected excep-
tions in application programs. Refer to Section 6.4., “Task Linking” in Chapter 6, Task
Management for more information on nested tasks.
RF Resume (bit 16). Controls the processor’s response to instruction-breakpoint condi-
tions. When set, this flag temporarily disables debug exceptions (#DE) from being
generated for instruction breakpoints; although, other exception conditions can
cause an exception to be generated. When clear, instruction breakpoints will generate
debug exceptions.
The primary function of the RF flag is to allow the restarting of an instruction following
a debug exception that was caused by an instruction breakpoint condition. Here,
debugger software must set this flag in the EFLAGS image on the stack just prior to
returning to the interrupted program with the IRETD instruction, to prevent the instruc-
tion breakpoint from causing another debug exception. The processor then automati-
cally clears this flag after the instruction returned to has been successfully executed,
enabling instruction breakpoint faults again.
Refer to Section 15.3.1.1., “Instruction-Breakpoint Exception Condition”, in Chapter
15, Debugging and Performance Monitoring, for more information on the use of this
flag.
VM Virtual-8086 mode (bit 17). Set to enable virtual-8086 mode; clear to return to
protected mode. Refer to Section 16.2.1., “Enabling Virtual-8086 Mode” in Chapter
16, 8086 Emulation for a detailed description of the use of this flag to switch to virtual-
8086 mode.
AC Alignment check (bit 18). Set this flag and the AM flag in the CR0 register to enable
alignment checking of memory references; clear the AC flag and/or the AM flag to
disable alignment checking. An alignment-check exception is generated when refer-
ence is made to an unaligned operand, such as a word at an odd byte address or a
doubleword at an address which is not an integral multiple of four. Alignment-check
exceptions are generated only in user mode (privilege level 3). Memory references that
default to privilege level 0, such as segment descriptor loads, do not generate this
exception even when caused by instructions executed in user-mode.
The alignment-check exception can be used to check alignment of data. This is useful
when exchanging data with other processors, which require all data to be aligned. The
alignment-check exception can also be used by interpreters to flag some pointers as
special by misaligning the pointer. This eliminates overhead of checking each pointer
and only handles the special pointer when used.

2-9
 SYSTEM ARCHITECTURE OVERVIEW

VIF Virtual Interrupt (bit 19). Contains a virtual image of the IF flag. This flag is used in
conjunction with the VIP flag. The processor only recognizes the VIF flag when either
the VME flag or the PVI flag in control register CR4 is set and the IOPL is less than 3.
(The VME flag enables the virtual-8086 mode extensions; the PVI flag enables the
protected-mode virtual interrupts.) Refer to Section 16.3.3.5., “Method 6: Software
Interrupt Handling” and Section 16.4., “Protected-Mode Virtual Interrupts” in Chapter
16, 8086 Emulation for detailed information about the use of this flag.
VIP Virtual interrupt pending (bit 20). Set by software to indicate that an interrupt is
pending; cleared to indicate that no interrupt is pending. This flag is used in conjunc-
tion with the VIF flag. The processor reads this flag but never modifies it. The
processor only recognizes the VIP flag when either the VME flag or the PVI flag in
control register CR4 is set and the IOPL is less than 3. (The VME flag enables the
virtual-8086 mode extensions; the PVI flag enables the protected-mode virtual inter-
rupts.) Refer to Section 16.3.3.5., “Method 6: Software Interrupt Handling” and
Section 16.4., “Protected-Mode Virtual Interrupts” in Chapter 16, 8086 Emulation for
detailed information about the use of this flag.
ID Identification (bit 21). The ability of a program or procedure to set or clear this flag
indicates support for the CPUID instruction.

2.4. MEMORY-MANAGEMENT REGISTERS

The processor provides four memory-management registers (GDTR, LDTR, IDTR, and TR)
that specify the locations of the data structures which control segmented memory management
(refer to Figure 2-4). Special instructions are provided for loading and storing these registers.

System Table Registers

47 16 15 0
GDTR 32-bit Linear Base Address 16-Bit Table Limit
IDTR 32-bit Linear Base Address 16-Bit Table Limit

System Segment Segment Descriptor Registers (Automatically Loaded)

15 Registers 0 Attributes
Task
Register Seg. Sel. 32-bit Linear Base Address Segment Limit
LDTR Seg. Sel. 32-bit Linear Base Address Segment Limit

Figure 2-4. Memory Management Registers

2.4.1. Global Descriptor Table Register (GDTR)

The GDTR register holds the 32-bit base address and 16-bit table limit for the GDT. The base
address specifies the linear address of byte 0 of the GDT; the table limit specifies the number of
bytes in the table. The LGDT and SGDT instructions load and store the GDTR register, respec-
tively. On power up or reset of the processor, the base address is set to the default value of 0 and

2-10
 SYSTEM ARCHITECTURE OVERVIEW

the limit is set to FFFFH. A new base address must be loaded into the GDTR as part of the
processor initialization process for protected-mode operation. Refer to Section 3.5.1., “Segment
Descriptor Tables” in Chapter 3, Protected-Mode Memory Management for more information
on the base address and limit fields.

2.4.2. Local Descriptor Table Register (LDTR)

The LDTR register holds the 16-bit segment selector, 32-bit base address, 16-bit segment limit,
and descriptor attributes for the LDT. The base address specifies the linear address of byte 0 of
the LDT segment; the segment limit specifies the number of bytes in the segment. Refer to
Section 3.5.1., “Segment Descriptor Tables” in Chapter 3, Protected-Mode Memory Manage-
ment for more information on the base address and limit fields.
The LLDT and SLDT instructions load and store the segment selector part of the LDTR register,
respectively. The segment that contains the LDT must have a segment descriptor in the GDT.
When the LLDT instruction loads a segment selector in the LDTR, the base address, limit, and
descriptor attributes from the LDT descriptor are automatically loaded into the LDTR.
When a task switch occurs, the LDTR is automatically loaded with the segment selector and
descriptor for the LDT for the new task. The contents of the LDTR are not automatically saved
prior to writing the new LDT information into the register.
On power up or reset of the processor, the segment selector and base address are set to the default
value of 0 and the limit is set to FFFFH.

2.4.3. IDTR Interrupt Descriptor Table Register

The IDTR register holds the 32-bit base address and 16-bit table limit for the IDT. The base
address specifies the linear address of byte 0 of the IDT; the table limit specifies the number of
bytes in the table. The LIDT and SIDT instructions load and store the IDTR register, respec-
tively. On power up or reset of the processor, the base address is set to the default value of 0 and
the limit is set to FFFFH. The base address and limit in the register can then be changed as part
of the processor initialization process. Refer to Section 5.8., “Interrupt Descriptor Table (IDT)”
in Chapter 5, Interrupt and Exception Handling for more information on the base address and
limit fields.

2.4.4. Task Register (TR)

The task register holds the 16-bit segment selector, 32-bit base address, 16-bit segment limit,
and descriptor attributes for the TSS of the current task. It references a TSS descriptor in the
GDT. The base address specifies the linear address of byte 0 of the TSS; the segment limit spec-
ifies the number of bytes in the TSS. (Refer to Section 6.2.3., “Task Register” in Chapter 6, Task
Management for more information about the task register.)
The LTR and STR instructions load and store the segment selector part of the task register,
respectively. When the LTR instruction loads a segment selector in the task register, the base

2-11
SYSTEM ARCHITECTURE OVERVIEW

address, limit, and descriptor attributes from the TSS descriptor are automatically loaded into
the task register. On power up or reset of the processor, the base address is set to the default value
of 0 and the limit is set to FFFFH.
When a task switch occurs, the task register is automatically loaded with the segment selector
and descriptor for the TSS for the new task. The contents of the task register are not automati-
cally saved prior to writing the new TSS information into the register.

2.5. CONTROL REGISTERS

The control registers (CR0, CR1, CR2, CR3, and CR4) determine operating mode of the
processor and the characteristics of the currently executing task (refer to Figure 2-5).

31 10 9 8 7 6 5 4 3 2 1 0

P P M P P T P V
Reserved (set to 0) C G C A S D S V M CR4
E D I E
E E E E E

OSXMMEXCPT
OSFXSR
31 12 11 5 4 3 2 0

P P
C W CR3
Page-Directory Base
D T (PDBR)
31 0

Page-Fault Linear Address CR2

31 0

CR1

31 30 29 30 19 18 17 16 15 6 5 4 3 2 1 0

P C N A W N E T E M P
G D W M P E T S M P E CR0

Reserved

Figure 2-5. Control Registers

2-12
 SYSTEM ARCHITECTURE OVERVIEW

The control registers:

• CR0—Contains system control flags that control operating mode and states of the
processor.
• CR1—Reserved.
• CR2—Contains the page-fault linear address (the linear address that caused a page fault).
• CR3—Contains the physical address of the base of the page directory and two flags (PCD
and PWT). This register is also known as the page-directory base register (PDBR). Only
the 20 most-significant bits of the page-directory base address are specified; the lower 12
bits of the address are assumed to be 0. The page directory must thus be aligned to a page
(4-KByte) boundary. The PCD and PWT flags control caching of the page directory in the
processor’s internal data caches (they do not control TLB caching of page-directory
information).
When using the physical address extension, the CR3 register contains the base address of
the page-directory-pointer table (refer to Section 3.8., “Physical Address Extension” in
Chapter 3, Protected-Mode Memory Management).
• CR4—Contains a group of flags that enable several architectural extensions, as well as
indicating the level of OS support for the Streaming SIMD Extensions.
In protected mode, the move-to-or-from-control-registers forms of the MOV instruction allow
the control registers to be read (at privilege level 0 only) or loaded (at privilege level 0 only).
These restrictions mean that application programs (running at privilege levels 1, 2, or 3) are
prevented from reading or loading the control registers.
A program running at privilege level 1, 2, or 3 should not attempt to read or write the control
registers. An attempt to read or write these registers will result in a general protection fault
(GP(0)). The functions of the flags in the control registers are as follows:
PG Paging (bit 31 of CR0). Enables paging when set; disables paging when clear. When
paging is disabled, all linear addresses are treated as physical addresses. The PG flag
has no effect if the PE flag (bit 0 of register CR0) is not also set; in fact, setting the PG
flag when the PE flag is clear causes a general-protection exception (#GP) to be gener-
ated. Refer to Section 3.6., “Paging (Virtual Memory)” in Chapter 3, Protected-Mode
Memory Management for a detailed description of the processor’s paging mechanism.
CD Cache Disable (bit 30 of CR0). When the CD and NW flags are clear, caching of
memory locations for the whole of physical memory in the processor’s internal (and
external) caches is enabled. When the CD flag is set, caching is restricted as described
in Table 9-4, in Chapter 9, Memory Cache Control. To prevent the processor from
accessing and updating its caches, the CD flag must be set and the caches must be
invalidated so that no cache hits can occur (refer to Section 9.5.2., “Preventing
Caching”, in Chapter 9, Memory Cache Control). Refer to Section 9.5., “Cache
Control”, Chapter 9, Memory Cache Control, for a detailed description of the addi-
tional restrictions that can be placed on the caching of selected pages or regions of
memory.
NW Not Write-through (bit 29 of CR0). When the NW and CD flags are clear, write-back
(for Pentium® and P6 family processors) or write-through (for Intel486™ processors)
is enabled for writes that hit the cache and invalidation cycles are enabled. Refer to
2-13
SYSTEM ARCHITECTURE OVERVIEW

Table 9-4, in Chapter 9, Memory Cache Control, for detailed information about the
affect of the NW flag on caching for other settings of the CD and NW flags.
AM Alignment Mask (bit 18 of CR0). Enables automatic alignment checking when set;
disables alignment checking when clear. Alignment checking is performed only when
the AM flag is set, the AC flag in the EFLAGS register is set, the CPL is 3, and the
processor is operating in either protected or virtual-8086 mode.
WP Write Protect (bit 16 of CR0). Inhibits supervisor-level procedures from writing into
user-level read-only pages when set; allows supervisor-level procedures to write into
user-level read-only pages when clear. This flag facilitates implementation of the copy-
on-write method of creating a new process (forking) used by operating systems such as
UNIX*.
NE Numeric Error (bit 5 of CR0). Enables the native (internal) mechanism for reporting
FPU errors when set; enables the PC-style FPU error reporting mechanism when clear.
When the NE flag is clear and the IGNNE# input is asserted, FPU errors are ignored.
When the NE flag is clear and the IGNNE# input is deasserted, an unmasked FPU error
causes the processor to assert the FERR# pin to generate an external interrupt and to
stop instruction execution immediately before executing the next waiting floating-
point instruction or WAIT/FWAIT instruction. The FERR# pin is intended to drive an
input to an external interrupt controller (the FERR# pin emulates the ERROR# pin of
the Intel 287 and Intel 387 DX math coprocessors). The NE flag, IGNNE# pin, and
FERR# pin are used with external logic to implement PC-style error reporting. (Refer
to “Software Exception Handling” in Chapter 7, and Appendix D in the Intel Architec-
ture Software Developer’s Manual, Volume 1, for more information about FPU error
reporting and for detailed information on when the FERR# pin is asserted, which is
implementation dependent.)
ET Extension Type (bit 4 of CR0). Reserved in the P6 family and Pentium® processors.
(In the P6 family processors, this flag is hardcoded to 1.) In the Intel386™ and
Intel486™ processors, this flag indicates support of Intel 387 DX math coprocessor
instructions when set.
TS Task Switched (bit 3 of CR0). Allows the saving of FPU context on a task switch to
be delayed until the FPU is actually accessed by the new task. The processor sets this
flag on every task switch and tests it when interpreting floating-point arithmetic
instructions.
• If the TS flag is set, a device-not-available exception (#NM) is raised prior to the
execution of a floating-point instruction.
• If the TS flag and the MP flag (also in the CR0 register) are both set, an #NM
exception is raised prior to the execution of floating-point instruction or a
WAIT/FWAIT instruction.
Table 2-1 shows the actions taken for floating-point, WAIT/FWAIT, MMX™, and
Streaming SIMD Extensions based on the settings of the TS, EM, and MP flags.

2-14
 SYSTEM ARCHITECTURE OVERVIEW

Table 2-1. Action Taken for Combinations of EM, MP, TS, CR4.OSFXSR, and CPUID.XMM
CR0 Flags CR4 CPUID Instruction Type
EM MP TS OSFXSR XMM Floating-Point WAIT/FWAIT MMX™ Streaming
Technology SIMD
Extensions
0 0 0 - - Execute Execute Execute -
0 0 1 - - #NM Exception Execute #NM -
Exception
0 1 0 - - Execute Execute Execute -
0 1 1 - - #NM Exception #NM Exception #NM -
Exception
1 0 0 - - #NM Exception Execute #UD Exception -
1 0 1 - - #NM Exception Execute #UD Exception -
1 1 0 - - #NM Exception Execute #UD Exception -
EM MP TS OSFXSR XMM Floating-Point WAIT/FWAIT MMX™ Streaming
Technology SIMD
Extensions
1 1 1 - - #NM Exception #NM Exception #UD Exception -
1 - - - - - - - #UD Interrupt
6
0 - 1 1 1 - - - #NM Interrupt
7
- - - 0 - - - - #UD Interrupt
6
- - - - 0 - - - #UD Interrupt
6

The processor does not automatically save the context of the FPU on a task switch.
Instead it sets the TS flag, which causes the processor to raise an #NM exception when-
ever it encounters a floating-point instruction in the instruction stream for the new task.
The fault handler for the #NM exception can then be used to clear the TS flag (with the
CLTS instruction) and save the context of the FPU. If the task never encounters a
floating-point instruction, the FPU context is never saved.

EM Emulation (bit 2 of CR0). Indicates that the processor does not have an internal or
external FPU when set; indicates an FPU is present when clear. When the EM flag is
set, execution of a floating-point instruction generates a device-not-available exception
(#NM). This flag must be set when the processor does not have an internal FPU or is
not connected to a math coprocessor. If the processor does have an internal FPU,
setting this flag would force all floating-point instructions to be handled by software
emulation. Table 8-2 in Chapter 8, Processor Management and Initialization shows the
recommended setting of this flag, depending on the Intel Architecture processor and

2-15
SYSTEM ARCHITECTURE OVERVIEW

FPU or math coprocessor present in the system. Table 2-1 shows the interaction of the
EM, MP, and TS flags.
Note that the EM flag also affects the execution of the MMX™ instructions (refer to
Table 2-1). When this flag is set, execution of an MMX™ instruction causes an invalid
opcode exception (#UD) to be generated. Thus, if an Intel Architecture processor
incorporates MMX™ technology, the EM flag must be set to 0 to enable execution of
MMX™ instructions.
Similarly for the Streaming SIMD Extensions, when this flag is set, execution of a Streaming
SIMD Extensions instruction causes an invalid opcode exception (#UD) to be generated. Thus,
if an Intel Architecture processor incorporates Streaming SIMD Extensions, the EM flag must
be set to 0 to enable execution of Streaming SIMD Extensions. The exception to this is the
PREFETCH and SFENCE instructions. These instructions are not affected by the EM flag.
MP Monitor Coprocessor (bit 1 of CR0). Controls the interaction of the WAIT (or
FWAIT) instruction with the TS flag (bit 3 of CR0). If the MP flag is set, a WAIT
instruction generates a device-not-available exception (#NM) if the TS flag is set. If the
MP flag is clear, the WAIT instruction ignores the setting of the TS flag. Table 8-2 in
Chapter 8, Processor Management and Initialization shows the recommended setting
of this flag, depending on the Intel Architecture processor and FPU or math copro-
cessor present in the system. Table 2-1 shows the interaction of the MP, EM, and TS
flags.
PE Protection Enable (bit 0 of CR0). Enables protected mode when set; enables real-
address mode when clear. This flag does not enable paging directly. It only enables
segment-level protection. To enable paging, both the PE and PG flags must be set.
Refer to Section 8.8., “Mode Switching” in Chapter 8, Processor Management and
Initialization for information using the PE flag to switch between real and protected
mode.
PCD Page-level Cache Disable (bit 4 of CR3). Controls caching of the current page direc-
tory. When the PCD flag is set, caching of the page-directory is prevented; when the
flag is clear, the page-directory can be cached. This flag affects only the processor’s
internal caches (both L1 and L2, when present). The processor ignores this flag if
paging is not used (the PG flag in register CR0 is clear) or the CD (cache disable) flag
in CR0 is set. Refer to Chapter 9, Memory Cache Control, for more information about
the use of this flag. Refer to Section 3.6.4., “Page-Directory and Page-Table Entries”
in Chapter 3, Protected-Mode Memory Management for a description of a companion
PCD flag in the page-directory and page-table entries.
PWT Page-level Writes Transparent (bit 3 of CR3). Controls the write-through or write-
back caching policy of the current page directory. When the PWT flag is set, write-
through caching is enabled; when the flag is clear, write-back caching is enabled. This
flag affects only the internal caches (both L1 and L2, when present). The processor
ignores this flag if paging is not used (the PG flag in register CR0 is clear) or the CD
(cache disable) flag in CR0 is set. Refer to Section 9.5., “Cache Control”, in Chapter
9, Memory Cache Control, for more information about the use of this flag. Refer to
Section 3.6.4., “Page-Directory and Page-Table Entries” in Chapter 3, Protected-Mode

2-16
 SYSTEM ARCHITECTURE OVERVIEW

Memory Management for a description of a companion PCD flag in the page-directory

and page-table entries.
VME Virtual-8086 Mode Extensions (bit 0 of CR4). Enables interrupt- and exception-
handling extensions in virtual-8086 mode when set; disables the extensions when clear.
Use of the virtual mode extensions can improve the performance of virtual-8086 appli-
cations by eliminating the overhead of calling the virtual-8086 monitor to handle inter-
rupts and exceptions that occur while executing an 8086 program and, instead,
redirecting the interrupts and exceptions back to the 8086 program’s handlers. It also
provides hardware support for a virtual interrupt flag (VIF) to improve reliability of
running 8086 programs in multitasking and multiple-processor environments. Refer to
Section 16.3., “Interrupt and Exception Handling in Virtual-8086 Mode” in Chapter 16,
8086 Emulation for detailed information about the use of this feature.
PVI Protected-Mode Virtual Interrupts (bit 1 of CR4). Enables hardware support for a
virtual interrupt flag (VIF) in protected mode when set; disables the VIF flag in
protected mode when clear. Refer to Section 16.4., “Protected-Mode Virtual Inter-
rupts” in Chapter 16, 8086 Emulation for detailed information about the use of this
feature.
TSD Time Stamp Disable (bit 2 of CR4). Restricts the execution of the RDTSC instruction
to procedures running at privilege level 0 when set; allows RDTSC instruction to be
executed at any privilege level when clear.
DE Debugging Extensions (bit 3 of CR4). References to debug registers DR4 and DR5
cause an undefined opcode (#UD) exception to be generated when set; when clear,
processor aliases references to registers DR4 and DR5 for compatibility with software
written to run on earlier Intel Architecture processors. Refer to Section 15.2.2., “Debug
Registers DR4 and DR5”, in Chapter 15, Debugging and Performance Monitoring, for
more information on the function of this flag.
PSE Page Size Extensions (bit 4 of CR4). Enables 4-MByte pages when set; restricts pages
to 4 KBytes when clear. Refer to Section 3.6.1., “Paging Options” in Chapter 3,
Protected-Mode Memory Management for more information about the use of this flag.
PAE Physical Address Extension (bit 5 of CR4). Enables paging mechanism to reference
36-bit physical addresses when set; restricts physical addresses to 32 bits when clear.
Refer to Section 3.8., “Physical Address Extension” in Chapter 3, Protected-Mode
Memory Management for more information about the physical address extension.
MCE Machine-Check Enable (bit 6 of CR4). Enables the machine-check exception when
set; disables the machine-check exception when clear. Refer to Chapter 13, Machine-
Check Architecture, for more information about the machine-check exception and
machine- check architecture.
PGE Page Global Enable (bit 7 of CR4). (Introduced in the P6 family processors.) Enables
the global page feature when set; disables the global page feature when clear. The
global page feature allows frequently used or shared pages to be marked as global to
all users (done with the global flag, bit 8, in a page-directory or page-table entry).
Global pages are not flushed from the translation-lookaside buffer (TLB) on a task
switch or a write to register CR3. In addition, the bit must not be enabled before paging

2-17
SYSTEM ARCHITECTURE OVERVIEW

is enabled via CR0.PG. Program correctness may be affected by reversing this

sequence, and processor performance will be impacted. Refer to Section 3.7., “Trans-
lation Lookaside Buffers (TLBs)” in Chapter 3, Protected-Mode Memory Management
for more information on the use of this bit.
PCE Performance-Monitoring Counter Enable (bit 8 of CR4). Enables execution of the
RDPMC instruction for programs or procedures running at any protection level when
set; RDPMC instruction can be executed only at protection level 0 when clear.
OSFXSR
Operating Sytsem FXSAVE/FXRSTOR Support (bit 9 of CR4). The operating
system will set this bit if both the CPU and the OS support the use of
FXSAVE/FXRSTOR for use during context switches.
OSXMMEXCPT
Operating System Unmasked Exception Support (bit 10 of CR4). The operating
system will set this bit if it provides support for unmasked SIMD floating-point excep-
tions.

2.5.1. CPUID Qualification of Control Register Flags

The VME, PVI, TSD, DE, PSE, PAE, MCE, PGE, PCE, OSFXSR, and OSXMMCEPT flags in
control register CR4 are model specific. All of these flags (except PCE) can be qualified with
the CPUID instruction to determine if they are implemented on the processor before they are
used.

2.6. SYSTEM INSTRUCTION SUMMARY

The system instructions handle system-level functions such as loading system registers,
managing the cache, managing interrupts, or setting up the debug registers. Many of these
instructions can be executed only by operating-system or executive procedures (that is, proce-
dures running at privilege level 0). Others can be executed at any privilege level and are thus
available to application programs. Table 2-2 lists the system instructions and indicates whether
they are available and useful for application programs. These instructions are described in detail
in Chapter 3, Instruction Set Reference, of the Intel Architecture Software Developer’s Manual,
Volume 2.

2-18
 SYSTEM ARCHITECTURE OVERVIEW

:
Table 2-2. Summary of System Instructions
Useful to Protected from
Instruction Description Application? Application?
LLDT Load LDT Register No Yes
SLDT Store LDT Register No No
LGDT Load GDT Register No Yes
SGDT Store GDT Register No No
LTR Load Task Register No Yes
STR Store Task Register No No
LIDT Load IDT Register No Yes
SIDT Store IDT Register No No
MOV CRn Load and store control registers Yes Yes (load only)
SMSW Store MSW Yes No
LMSW Load MSW No Yes
CLTS Clear TS flag in CR0 No Yes
ARPL Adjust RPL Yes1 No
LAR Load Access Rights Yes No
LSL Load Segment Limit Yes No
VERR Verify for Reading Yes No
VERW Verify for Writing Yes No
MOV DBn Load and store debug registers No Yes
INVD Invalidate cache, no writeback No Yes
WBINVD Invalidate cache, with writeback No Yes
INVLPG Invalidate TLB entry No Yes
HLT Halt Processor No Yes
LOCK (Prefix) Bus Lock Yes No
RSM Return from system management mode No Yes
RDMSR3 Read Model-Specific Registers No Yes
3
WRMSR Write Model-Specific Registers No Yes
RDPMC4 Read Performance-Monitoring Counter Yes Yes2
3
RDTSC Read Time-Stamp Counter Yes Yes2
LDMXCSR5 Load MXCSR Register Yes No
5
STMXCSR Store MXCSR Resister Yes No

NOTES:
1. Useful to application programs running at a CPL of 1 or 2.
2. The TSD and PCE flags in control register CR4 control access to these instructions by application
programs running at a CPL of 3.
3. These instructions were introduced into the Intel Architecture with the Pentium® processor.
4. This instruction was introduced into the Intel Architecture with the Pentium® Pro processor and the Pen-
tium processor with MMX™ technology.
5. This instruction was introduced into the Intel Architecture with the Pentium® III processor.

2-19
SYSTEM ARCHITECTURE OVERVIEW

2.6.1. Loading and Storing System Registers

The GDTR, LDTR, IDTR, and TR registers each have a load and store instruction for loading
data into and storing data from the register:
LGDT (Load GDTR Register) Loads the GDT base address and limit from memory into the
GDTR register.
SGDT (Store GDTR Register) Stores the GDT base address and limit from the GDTR register
into memory.
LIDT (Load IDTR Register) Loads the IDT base address and limit from memory into the
IDTR register.
SIDT (Load IDTR Register Stores the IDT base address and limit from the IDTR register
into memory.
LLDT (Load LDT Register) Loads the LDT segment selector and segment descriptor from
memory into the LDTR. (The segment selector operand can also
be located in a general-purpose register.)
SLDT (Store LDT Register) Stores the LDT segment selector from the LDTR register into
memory or a general-purpose register.
LTR (Load Task Register) Loads segment selector and segment descriptor for a TSS from
memory into the task register. (The segment selector operand
can also be located in a general-purpose register.)
STR (Store Task Register) Stores the segment selector for the current task TSS from the
task register into memory or a general-purpose register.
The LMSW (load machine status word) and SMSW (store machine status word) instructions
operate on bits 0 through 15 of control register CR0. These instructions are provided for compat-
ibility with the 16-bit Intel 286 processor. Program written to run on 32-bit Intel Architecture
processors should not use these instructions. Instead, they should access the control register CR0
using the MOV instruction.
The CLTS (clear TS flag in CR0) instruction is provided for use in handling a device-not-avail-
able exception (#NM) that occurs when the processor attempts to execute a floating-point
instruction when the TS flag is set. This instruction allows the TS flag to be cleared after the
FPU context has been saved, preventing further #NM exceptions. Refer to Section 2.5., “Control
Registers” for more information about the TS flag.
The control registers (CR0, CR1, CR2, CR3, and CR4) are loaded with the MOV instruction.
This instruction can load a control register from a general-purpose register or store the contents
of the control register in a general-purpose register.

2.6.2. Verifying of Access Privileges

The processor provides several instructions for examining segment selectors and segment
descriptors to determine if access to their associated segments is allowed. These instructions

2-20
 SYSTEM ARCHITECTURE OVERVIEW

duplicate some of the automatic access rights and type checking done by the processor, thus
allowing operating-system or executive software to prevent exceptions from being generated.
The ARPL (adjust RPL) instruction adjusts the RPL (requestor privilege level) of a segment
selector to match that of the program or procedure that supplied the segment selector. Refer to
Section 4.10.4., “Checking Caller Access Privileges (ARPL Instruction)” in Chapter 4, Protec-
tion for a detailed explanation of the function and use of this instruction.
The LAR (load access rights) instruction verifies the accessibility of a specified segment and
loads the access rights information from the segment’s segment descriptor into a general-
purpose register. Software can then examine the access rights to determine if the segment type
is compatible with its intended use. Refer to Section 4.10.1., “Checking Access Rights (LAR
Instruction)” in Chapter 4, Protection for a detailed explanation of the function and use of this
instruction.
The LSL (load segment limit) instruction verifies the accessibility of a specified segment and
loads the segment limit from the segment’s segment descriptor into a general-purpose register.
Software can then compare the segment limit with an offset into the segment to determine
whether the offset lies within the segment. Refer to Section 4.10.3., “Checking That the Pointer
Offset Is Within Limits (LSL Instruction)” in Chapter 4, Protection for a detailed explanation of
the function and use of this instruction.
The VERR (verify for reading) and VERW (verify for writing) instructions verify if a selected
segment is readable or writable, respectively, at the CPL. Refer to Section 4.10.2., “Checking
Read/Write Rights (VERR and VERW Instructions)” in Chapter 4, Protection for a detailed
explanation of the function and use of this instruction.

2.6.3. Loading and Storing Debug Registers

The internal debugging facilities in the processor are controlled by a set of 8 debug registers
(DR0 through DR7). The MOV instruction allows setup data to be loaded into and stored from
these registers.

2.6.4. Invalidating Caches and TLBs

The processor provides several instructions for use in explicitly invalidating its caches and TLB
entries. The INVD (invalidate cache with no writeback) instruction invalidates all data and
instruction entries in the internal caches and TLBs and sends a signal to the external caches indi-
cating that they should be invalidated also.
The WBINVD (invalidate cache with writeback) instruction performs the same function as the
INVD instruction, except that it writes back any modified lines in its internal caches to memory
before it invalidates the caches. After invalidating the internal caches, it signals the external
caches to write back modified data and invalidate their contents.
The INVLPG (invalidate TLB entry) instruction invalidates (flushes) the TLB entry for a spec-
ified page.

2-21
SYSTEM ARCHITECTURE OVERVIEW

2.6.5. Controlling the Processor

The HLT (halt processor) instruction stops the processor until an enabled interrupt (such as NMI
or SMI, which are normally enabled), the BINIT# signal, the INIT# signal, or the RESET#
signal is received. The processor generates a special bus cycle to indicate that the halt mode has
been entered. Hardware may respond to this signal in a number of ways. An indicator light on
the front panel may be turned on. An NMI interrupt for recording diagnostic information may
be generated. Reset initialization may be invoked. (Note that the BINIT# pin was introduced
with the Pentium® Pro processor.)
The LOCK prefix invokes a locked (atomic) read-modify-write operation when modifying a
memory operand. This mechanism is used to allow reliable communications between processors
in multiprocessor systems. In the Pentium® and earlier Intel Architecture processors, the LOCK
prefix causes the processor to assert the LOCK# signal during the instruction, which always
causes an explicit bus lock to occur. In the P6 family processors, the locking operation is handled
with either a cache lock or bus lock. If a memory access is cacheable and affects only a single
cache line, a cache lock is invoked and the system bus and the actual memory location in system
memory are not locked during the operation. Here, other P6 family processors on the bus write-
back any modified data and invalidate their caches as necessary to maintain system memory
coherency. If the memory access is not cacheable and/or it crosses a cache line boundary, the
processor’s LOCK# signal is asserted and the processor does not respond to requests for bus
control during the locked operation.
The RSM (return from SMM) instruction restores the processor (from a context dump) to the
state it was in prior to an system management mode (SMM) interrupt.

2.6.6. Reading Performance-Monitoring and Time-Stamp

Counters
The RDPMC (read performance-monitoring counter) and RDTSC (read time-stamp counter)
instructions allow an application program to read the processors performance-monitoring and
time-stamp counters, respectively.
The P6 family processors have two 40-bit performance counters that record either the occur-
rence of events or the duration of events. The events that can be monitored include the number
of instructions decoded, number of interrupts received, of number of cache loads. Each counter
can be set up to monitor a different event, using the system instruction WRMSR to set up values
in the model-specific registers PerfEvtSel0 and PerfEvtSel1. The RDPMC instruction loads the
current count in counter 0 or 1 into the EDX:EAX registers.
The time-stamp counter is a model-specific 64-bit counter that is reset to zero each time the
processor is reset. If not reset, the counter will increment ~6.3 x 10 15 times per year when
the processor is operating at a clock rate of 200 MHz. At this clock frequency, it would take
over 2000 years for the counter to wrap around. The RDTSC instruction loads the current
count of the time-stamp counter into the EDX:EAX registers.

2-22
 SYSTEM ARCHITECTURE OVERVIEW

Refer to Section 15.5., “Time-Stamp Counter”, and Section 15.6., “Performance-Monitoring

Counters”, in Chapter 15, Debugging and Performance Monitoring, for more information about
the performance monitoring and time-stamp counters.
The RDTSC instruction was introduced into the Intel Architecture with the Pentium® processor.
The RDPMC instruction was introduced into the Intel Architecture with the Pentium® Pro
processor and the Pentium® processor with MMX™ technology. Earlier Pentium® processors
have two performance-monitoring counters, but they can be read only with the RDMSR instruc-
tion, and only at privilege level 0.

2.6.7. Reading and Writing Model-Specific Registers

The RDMSR (read model-specific register) and WRMSR (write model-specific register) allow
the processor’s 64-bit model-specific registers (MSRs) to be read and written to, respectively.
The MSR to be read or written to is specified by the value in the ECX register. The RDMSR
instruction reads the value from the specified MSR into the EDX:EAX registers; the WRMSR
writes the value in the EDX:EAX registers into the specified MSR. Refer to Section 8.4.,
“Model-Specific Registers (MSRs)” in Chapter 8, Processor Management and Initialization for
more information about the MSRs.
The RDMSR and WRMSR instructions were introduced into the Intel Architecture with the
Pentium® processor.

2.6.8. Loading and Storing the Streaming SIMD Extensions

Control/Status Word
The LDMXCSR (load Streaming SIMD Extensions control/status word from memory) and
STMXCSR (store Streaming SIMD Extensions control/status word to memory) allow the
Pentium® III processor’s 32-bit control/status word to be read and written to, respectively. The
MXCSR control/status register is used to enable masked/unmasked exception handling, to set
rounding modes, to set flush-to-zero mode, and to view exception status flags. For more infor-
mation on the LDMXCSR and STMXCSR instructions, refer to the Intel Architecture Software
Developer’s Manual, Vol 2, for a complete description of these instructions.

2-23
SYSTEM ARCHITECTURE OVERVIEW

2-24
 3
Protected-Mode
Memory
Management
 PROTECTED-MODE MEMORY MANAGEMENT

CHAPTER 3
PROTECTED-MODE MEMORY MANAGEMENT

This chapter describes the Intel Architecture’s protected-mode memory management facilities,
including the physical memory requirements, the segmentation mechanism, and the paging
mechanism. Refer to Chapter 4, Protection for a description of the processor’s protection mech-
anism. Refer to Chapter 16, 8086 Emulation for a description of memory addressing protection
in real-address and virtual-8086 modes.

3.1. MEMORY MANAGEMENT OVERVIEW

The memory management facilities of the Intel Architecture are divided into two parts: segmen-
tation and paging. Segmentation provides a mechanism of isolating individual code, data, and
stack modules so that multiple programs (or tasks) can run on the same processor without inter-
fering with one another. Paging provides a mechanism for implementing a conventional
demand-paged, virtual-memory system where sections of a program’s execution environment
are mapped into physical memory as needed. Paging can also be used to provide isolation
between multiple tasks. When operating in protected mode, some form of segmentation must be
used. There is no mode bit to disable segmentation. The use of paging, however, is optional.
These two mechanisms (segmentation and paging) can be configured to support simple single-
program (or single-task) systems, multitasking systems, or multiple-processor systems that used
shared memory.
As shown in Figure 3-1, segmentation provides a mechanism for dividing the processor’s
addressable memory space (called the linear address space) into smaller protected address
spaces called segments. Segments can be used to hold the code, data, and stack for a program
or to hold system data structures (such as a TSS or LDT). If more than one program (or task) is
running on a processor, each program can be assigned its own set of segments. The processor
then enforces the boundaries between these segments and insures that one program does not
interfere with the execution of another program by writing into the other program’s segments.
The segmentation mechanism also allows typing of segments so that the operations that may be
performed on a particular type of segment can be restricted.
All of the segments within a system are contained in the processor’s linear address space. To
locate a byte in a particular segment, a logical address (sometimes called a far pointer) must be
provided. A logical address consists of a segment selector and an offset. The segment selector
is a unique identifier for a segment. Among other things it provides an offset into a descriptor
table (such as the global descriptor table, GDT) to a data structure called a segment descriptor.
Each segment has a segment descriptor, which specifies the size of the segment, the access rights
and privilege level for the segment, the segment type, and the location of the first byte of the
segment in the linear address space (called the base address of the segment). The offset part of
the logical address is added to the base address for the segment to locate a byte within the
segment. The base address plus the offset thus forms a linear address in the processor’s linear

3-1
PROTECTED-MODE MEMORY MANAGEMENT

address space.

Logical Address
(or Far Pointer)

Segment
Selector Offset Linear Address
Space

Linear Address
Global Descriptor
Dir Table Offset Physical
Table (GDT)
Address
Space
Segment
Page Table Page
Segment
Descriptor
Page Directory Phy. Addr.
Lin. Addr.
Entry
Entry
Segment
Base Address

Page

Segmentation Paging

Figure 3-1. Segmentation and Paging

If paging is not used, the linear address space of the processor is mapped directly into the phys-
ical address space of processor. The physical address space is defined as the range of addresses
that the processor can generate on its address bus.
Because multitasking computing systems commonly define a linear address space much larger
than it is economically feasible to contain all at once in physical memory, some method of
“virtualizing” the linear address space is needed. This virtualization of the linear address space
is handled through the processor’s paging mechanism.
Paging supports a “virtual memory” environment where a large linear address space is simulated
with a small amount of physical memory (RAM and ROM) and some disk storage. When using
paging, each segment is divided into pages (ordinarily 4 KBytes each in size), which are stored
either in physical memory or on the disk. The operating system or executive maintains a page
directory and a set of page tables to keep track of the pages. When a program (or task) attempts
to access an address location in the linear address space, the processor uses the page directory

3-2
 PROTECTED-MODE MEMORY MANAGEMENT

and page tables to translate the linear address into a physical address and then performs the
requested operation (read or write) on the memory location. If the page being accessed is not
currently in physical memory, the processor interrupts execution of the program (by generating
a page-fault exception). The operating system or executive then reads the page into physical
memory from the disk and continues executing the program.
When paging is implemented properly in the operating-system or executive, the swapping of
pages between physical memory and the disk is transparent to the correct execution of a
program. Even programs written for 16-bit Intel Architecture processors can be paged (transpar-
ently) when they are run in virtual-8086 mode.

3.2. USING SEGMENTS

The segmentation mechanism supported by the Intel Architecture can be used to implement a
wide variety of system designs. These designs range from flat models that make only minimal
use of segmentation to protect programs to multisegmented models that employ segmentation
to create a robust operating environment in which multiple programs and tasks can be executed
reliably.
The following sections give several examples of how segmentation can be employed in a system
to improve memory management performance and reliability.

3.2.1. Basic Flat Model

The simplest memory model for a system is the basic “flat model,” in which the operating
system and application programs have access to a continuous, unsegmented address space. To
the greatest extent possible, this basic flat model hides the segmentation mechanism of the archi-
tecture from both the system designer and the application programmer.
To implement a basic flat memory model with the Intel Architecture, at least two segment
descriptors must be created, one for referencing a code segment and one for referencing a data
segment (refer to Figure 3-2). Both of these segments, however, are mapped to the entire linear
address space: that is, both segment descriptors have the same base address value of 0 and the
same segment limit of 4 GBytes. By setting the segment limit to 4 GBytes, the segmentation
mechanism is kept from generating exceptions for out of limit memory references, even if no
physical memory resides at a particular address. ROM (EPROM) is generally located at the top
of the physical address space, because the processor begins execution at FFFF_FFF0H. RAM
(DRAM) is placed at the bottom of the address space because the initial base address for the DS
data segment after reset initialization is 0.

3-3
PROTECTED-MODE MEMORY MANAGEMENT

Linear Address Space

(or Physical Memory)
Segment FFFFFFFFH
Registers Code
CS Code- and Data-Segment
Descriptors Not Present
SS

DS Access Limit Data and

ES Base Address Stack 0

FS

GS

Figure 3-2. Flat Model

3.2.2. Protected Flat Model

The protected flat model is similar to the basic flat model, except the segment limits are set to
include only the range of addresses for which physical memory actually exists (refer to Figure
3-3). A general-protection exception (#GP) is then generated on any attempt to access nonex-
istent memory. This model provides a minimum level of hardware protection against some kinds
of program bugs.

Segment Linear Address Space

Descriptors (or Physical Memory)
Segment Access Limit
Registers Code FFFFFFFFH
Base Address
CS
Not Present
ES
Memory I/O
SS Access Limit
DS Base Address
Data and
FS Stack

GS 0

Figure 3-3. Protected Flat Model

More complexity can be added to this protected flat model to provide more protection. For
example, for the paging mechanism to provide isolation between user and supervisor code and
data, four segments need to be defined: code and data segments at privilege level 3 for the user,
and code and data segments at privilege level 0 for the supervisor. Usually these segments all
overlay each other and start at address 0 in the linear address space. This flat segmentation

3-4
 PROTECTED-MODE MEMORY MANAGEMENT

model along with a simple paging structure can protect the operating system from applications,
and by adding a separate paging structure for each task or process, it can also protect applica-
tions from each other. Similar designs are used by several popular multitasking operating
systems.

3.2.3. Multisegment Model

A multisegment model (such as the one shown in Figure 3-4) uses the full capabilities of the
segmentation mechanism to provided hardware enforced protection of code, data structures, and
programs and tasks. Here, each program (or task) is given its own table of segment descriptors
and its own segments. The segments can be completely private to their assigned programs or
shared among programs. Access to all segments and to the execution environments of individual
programs running on the system is controlled by hardware.

Segment Segment Linear Address Space

Registers Descriptors (or Physical Memory)
Access Limit
CS
Base Address Stack
Access Limit
SS
Base Address

Access Limit
DS
Base Address Code
Access Limit
ES
Base Address
Data
Access Limit
FS
Base Address
Data
Access Limit
GS
Base Address
Data
Access Limit
Base Address

Access Limit
Base Address
Data
Access Limit
Base Address

Access Limit
Base Address

Figure 3-4. Multisegment Model

3-5
PROTECTED-MODE MEMORY MANAGEMENT

Access checks can be used to protect not only against referencing an address outside the limit
of a segment, but also against performing disallowed operations in certain segments. For
example, since code segments are designated as read-only segments, hardware can be used to
prevent writes into code segments. The access rights information created for segments can also
be used to set up protection rings or levels. Protection levels can be used to protect operating-
system procedures from unauthorized access by application programs.

3.2.4. Paging and Segmentation

Paging can be used with any of the segmentation models described in Figures 3-2, 3-3, and 3-4.
The processor’s paging mechanism divides the linear address space (into which segments are
mapped) into pages (as shown in Figure 3-1). These linear-address-space pages are then mapped
to pages in the physical address space. The paging mechanism offers several page-level protec-
tion facilities that can be used with or instead of the segment-protection facilities. For example,
it lets read-write protection be enforced on a page-by-page basis. The paging mechanism also
provides two-level user-supervisor protection that can also be specified on a page-by-page basis.

3.3. PHYSICAL ADDRESS SPACE

In protected mode, the Intel Architecture provides a normal physical address space of 4 GBytes
(232 bytes). This is the address space that the processor can address on its address bus. This
address space is flat (unsegmented), with addresses ranging continuously from 0 to
FFFFFFFFH. This physical address space can be mapped to read-write memory, read-only
memory, and memory mapped I/O. The memory mapping facilities described in this chapter can
be used to divide this physical memory up into segments and/or pages.
(Introduced in the Pentium® Pro processor.) The Intel Architecture also supports an extension of
the physical address space to 236 bytes (64 GBytes), with a maximum physical address of
FFFFFFFFFH. This extension is invoked with the physical address extension (PAE) flag,
located in bit 5 of control register CR4. (Refer to Section 3.8., “Physical Address Extension” for
more information about extended physical addressing.)

3.4. LOGICAL AND LINEAR ADDRESSES

At the system-architecture level in protected mode, the processor uses two stages of address
translation to arrive at a physical address: logical-address translation and linear address space
paging.
Even with the minimum use of segments, every byte in the processor’s address space is accessed
with a logical address. A logical address consists of a 16-bit segment selector and a 32-bit offset
(refer to Figure 3-5). The segment selector identifies the segment the byte is located in and the
offset specifies the location of the byte in the segment relative to the base address of the segment.
The processor translates every logical address into a linear address. A linear address is a 32-bit
address in the processor’s linear address space. Like the physical address space, the linear
address space is a flat (unsegmented), 232-byte address space, with addresses ranging from 0 to

3-6
 PROTECTED-MODE MEMORY MANAGEMENT

FFFFFFFH. The linear address space contains all the segments and system tables defined for a
system.
To translate a logical address into a linear address, the processor does the following:
1. Uses the offset in the segment selector to locate the segment descriptor for the segment in
the GDT or LDT and reads it into the processor. (This step is needed only when a new
segment selector is loaded into a segment register.)
2. Examines the segment descriptor to check the access rights and range of the segment to
insure that the segment is accessible and that the offset is within the limits of the segment.
3. Adds the base address of the segment from the segment descriptor to the offset to form a
linear address.

15 0 31 0
Logical Seg. Selector Offset
Address

Descriptor Table

Segment Base Address

Descriptor
+

31 0
Linear Address

Figure 3-5. Logical Address to Linear Address Translation

If paging is not used, the processor maps the linear address directly to a physical address (that
is, the linear address goes out on the processor’s address bus). If the linear address space is
paged, a second level of address translation is used to translate the linear address into a physical
address. Page translation is described in Section 3.6., “Paging (Virtual Memory)”

3.4.1. Segment Selectors

A segment selector is a 16-bit identifier for a segment (refer to Figure 3-6). It does not point
directly to the segment, but instead points to the segment descriptor that defines the segment. A
segment selector contains the following items:
Index (Bits 3 through 15). Selects one of 8192 descriptors in the GDT or LDT. The
processor multiplies the index value by 8 (the number of bytes in a segment
descriptor) and adds the result to the base address of the GDT or LDT (from
the GDTR or LDTR register, respectively).

3-7
PROTECTED-MODE MEMORY MANAGEMENT

TI (table indicator) flag

(Bit 2). Specifies the descriptor table to use: clearing this flag selects the GDT;
setting this flag selects the current LDT.

15 3 2 1 0

Index T RPL
I

Table Indicator
0 = GDT
1 = LDT
Requested Privilege Level (RPL)

Figure 3-6. Segment Selector

Requested Privilege Level (RPL)

(Bits 0 and 1). Specifies the privilege level of the selector. The privilege level
can range from 0 to 3, with 0 being the most privileged level. Refer to Section
4.5., “Privilege Levels” in Chapter 4, Protection for a description of the rela-
tionship of the RPL to the CPL of the executing program (or task) and the
descriptor privilege level (DPL) of the descriptor the segment selector points
to.
The first entry of the GDT is not used by the processor. A segment selector that points to this
entry of the GDT (that is, a segment selector with an index of 0 and the TI flag set to 0) is used
as a “null segment selector.” The processor does not generate an exception when a segment
register (other than the CS or SS registers) is loaded with a null selector. It does, however,
generate an exception when a segment register holding a null selector is used to access memory.
A null selector can be used to initialize unused segment registers. Loading the CS or SS register
with a null segment selector causes a general-protection exception (#GP) to be generated.
Segment selectors are visible to application programs as part of a pointer variable, but the values
of selectors are usually assigned or modified by link editors or linking loaders, not application
programs.

3.4.2. Segment Registers

To reduce address translation time and coding complexity, the processor provides registers for
holding up to 6 segment selectors (refer to Figure 3-7). Each of these segment registers support
a specific kind of memory reference (code, stack, or data). For virtually any kind of program
execution to take place, at least the code-segment (CS), data-segment (DS), and stack-segment
(SS) registers must be loaded with valid segment selectors. The processor also provides three
additional data-segment registers (ES, FS, and GS), which can be used to make additional data
segments available to the currently executing program (or task).
For a program to access a segment, the segment selector for the segment must have been loaded
in one of the segment registers. So, although a system can define thousands of segments, only 6

3-8
 PROTECTED-MODE MEMORY MANAGEMENT

can be available for immediate use. Other segments can be made available by loading their
segment selectors into these registers during program execution.

Visible Part Hidden Part

Segment Selector Base Address, Limit, Access Information CS
SS
DS
ES
FS
GS

Figure 3-7. Segment Registers

Every segment register has a “visible” part and a “hidden” part. (The hidden part is sometimes
referred to as a “descriptor cache” or a “shadow register.”) When a segment selector is loaded
into the visible part of a segment register, the processor also loads the hidden part of the segment
register with the base address, segment limit, and access control information from the segment
descriptor pointed to by the segment selector. The information cached in the segment register
(visible and hidden) allows the processor to translate addresses without taking extra bus cycles
to read the base address and limit from the segment descriptor. In systems in which multiple
processors have access to the same descriptor tables, it is the responsibility of software to reload
the segment registers when the descriptor tables are modified. If this is not done, an old segment
descriptor cached in a segment register might be used after its memory-resident version has been
modified.
Two kinds of load instructions are provided for loading the segment registers:
1. Direct load instructions such as the MOV, POP, LDS, LES, LSS, LGS, and LFS instruc-
tions. These instructions explicitly reference the segment registers.
2. Implied load instructions such as the far pointer versions of the CALL, JMP, and RET
instructions and the IRET, INTn, INTO and INT3 instructions. These instructions change
the contents of the CS register (and sometimes other segment registers) as an incidental
part of their operation.
The MOV instruction can also be used to store visible part of a segment register in a general-
purpose register.

3.4.3. Segment Descriptors

A segment descriptor is a data structure in a GDT or LDT that provides the processor with the
size and location of a segment, as well as access control and status information. Segment
descriptors are typically created by compilers, linkers, loaders, or the operating system or exec-

3-9
PROTECTED-MODE MEMORY MANAGEMENT

utive, but not application programs. Figure 3-8 illustrates the general descriptor format for all
types of segment descriptors.
The flags and fields in a segment descriptor are as follows:
Segment limit field
Specifies the size of the segment. The processor puts together the two segment
limit fields to form a 20-bit value. The processor interprets the segment limit
in one of two ways, depending on the setting of the G (granularity) flag:
• If the granularity flag is clear, the segment size can range from 1 byte to 1
MByte, in byte increments.
• If the granularity flag is set, the segment size can range from 4 KBytes to
4 GBytes, in 4-KByte increments.
The processor uses the segment limit in two different ways, depending on
whether the segment is an expand-up or an expand-down segment. Refer to
Section 3.4.3.1., “Code- and Data-Segment Descriptor Types” for more infor-
mation about segment types. For expand-up segments, the offset in a logical
address can range from 0 to the segment limit. Offsets greater than the segment
limit generate general-protection exceptions (#GP). For expand-down
segments, the segment limit has the reverse function; the offset can range from
the segment limit to FFFFFFFFH or FFFFH, depending on the setting of the B
flag. Offsets less than the segment limit generate general-protection excep-
tions. Decreasing the value in the segment limit field for an expand-down
segment allocates new memory at the bottom of the segment's address space,
rather than at the top. Intel Architecture stacks always grow downwards,
making this mechanism is convenient for expandable stacks.

3-10
 PROTECTED-MODE MEMORY MANAGEMENT

31 24 23 22 21 20 19 16 15 14 13 12 11 8 7 0

D A Seg. D
Base 31:24 G / 0 V Limit P P S Type Base 23:16 4
B L 19:16 L

31 16 15 0

Base Address 15:00 Segment Limit 15:00 0

AVL — Available for use by system software

BASE — Segment base address
D/B — Default operation size (0 = 16-bit segment; 1 = 32-bit segment)
DPL — Descriptor privilege level
G — Granularity
LIMIT — Segment Limit
P — Segment present
S — Descriptor type (0 = system; 1 = code or data)
TYPE — Segment type

Figure 3-8. Segment Descriptor

Base address fields

Defines the location of byte 0 of the segment within the 4-GByte linear address
space. The processor puts together the three base address fields to form a single
32-bit value. Segment base addresses should be aligned to 16-byte boundaries.
Although 16-byte alignment is not required, this alignment allows programs to
maximize performance by aligning code and data on 16-byte boundaries.
Type field Indicates the segment or gate type and specifies the kinds of access that can be
made to the segment and the direction of growth. The interpretation of this field
depends on whether the descriptor type flag specifies an application (code or
data) descriptor or a system descriptor. The encoding of the type field is
different for code, data, and system descriptors (refer to Figure 4-1 in Chapter
4, Protection). Refer to Section 3.4.3.1., “Code- and Data-Segment Descriptor
Types” for a description of how this field is used to specify code and data-
segment types.
S (descriptor type) flag
Specifies whether the segment descriptor is for a system segment (S flag is
clear) or a code or data segment (S flag is set).

3-11
PROTECTED-MODE MEMORY MANAGEMENT

DPL (descriptor privilege level) field

Specifies the privilege level of the segment. The privilege level can range from
0 to 3, with 0 being the most privileged level. The DPL is used to control access
to the segment. Refer to Section 4.5., “Privilege Levels” in Chapter 4, Protec-
tion for a description of the relationship of the DPL to the CPL of the executing
code segment and the RPL of a segment selector.
P (segment-present) flag
Indicates whether the segment is present in memory (set) or not present (clear).
If this flag is clear, the processor generates a segment-not-present exception
(#NP) when a segment selector that points to the segment descriptor is loaded
into a segment register. Memory management software can use this flag to
control which segments are actually loaded into physical memory at a given
time. It offers a control in addition to paging for managing virtual memory.
Figure 3-9 shows the format of a segment descriptor when the segment-present
flag is clear. When this flag is clear, the operating system or executive is free
to use the locations marked “Available” to store its own data, such as informa-
tion regarding the whereabouts of the missing segment.
D/B (default operation size/default stack pointer size and/or upper bound) flag
Performs different functions depending on whether the segment descriptor is
an executable code segment, an expand-down data segment, or a stack
segment. (This flag should always be set to 1 for 32-bit code and data segments
and to 0 for 16-bit code and data segments.)
• Executable code segment. The flag is called the D flag and it indicates the
default length for effective addresses and operands referenced by instruc-
tions in the segment. If the flag is set, 32-bit addresses and 32-bit or 8-bit
operands are assumed; if it is clear, 16-bit addresses and 16-bit or 8-bit
operands are assumed. The instruction prefix 66H can be used to select an
operand size other than the default, and the prefix 67H can be used select
an address size other than the default.
• Stack segment (data segment pointed to by the SS register). The flag is
called the B (big) flag and it specifies the size of the stack pointer used for
implicit stack operations (such as pushes, pops, and calls). If the flag is set,
a 32-bit stack pointer is used, which is stored in the 32-bit ESP register; if
the flag is clear, a 16-bit stack pointer is used, which is stored in the 16-bit
SP register. If the stack segment is set up to be an expand-down data
segment (described in the next paragraph), the B flag also specifies the
upper bound of the stack segment.
• Expand-down data segment. The flag is called the B flag and it specifies
the upper bound of the segment. If the flag is set, the upper bound is
FFFFFFFFH (4 GBytes); if the flag is clear, the upper bound is FFFFH (64
KBytes).

3-12
 PROTECTED-MODE MEMORY MANAGEMENT

31 16 15 14 13 12 11 8 7 0

D
Available 0 P S Type Available 4
L

31 0

Available 0

Figure 3-9. Segment Descriptor When Segment-Present Flag Is Clear

G (granularity) flag
Determines the scaling of the segment limit field. When the granularity flag is
clear, the segment limit is interpreted in byte units; when flag is set, the
segment limit is interpreted in 4-KByte units. (This flag does not affect the
granularity of the base address; it is always byte granular.) When the granu-
larity flag is set, the twelve least significant bits of an offset are not tested when
checking the offset against the segment limit. For example, when the granu-
larity flag is set, a limit of 0 results in valid offsets from 0 to 4095.
Available and reserved bits
Bit 20 of the second doubleword of the segment descriptor is available for use
by system software; bit 21 is reserved and should always be set to 0.

3.4.3.1. CODE- AND DATA-SEGMENT DESCRIPTOR TYPES

When the S (descriptor type) flag in a segment descriptor is set, the descriptor is for either a code
or a data segment. The highest order bit of the type field (bit 11 of the second double word of
the segment descriptor) then determines whether the descriptor is for a data segment (clear) or
a code segment (set).
For data segments, the three low-order bits of the type field (bits 8, 9, and 10) are interpreted as
accessed (A), write-enable (W), and expansion-direction (E). Refer to Table 3-1 for a descrip-
tion of the encoding of the bits in the type field for code and data segments. Data segments can
be read-only or read/write segments, depending on the setting of the write-enable bit.

3-13
PROTECTED-MODE MEMORY MANAGEMENT

Table 3-1. Code- and Data-Segment Types

Type Field
11 10 9 8 Descriptor
Decimal E W A Type Description
0 0 0 0 0 Data Read-Only
1 0 0 0 1 Data Read-Only, accessed
2 0 0 1 0 Data Read/Write
3 0 0 1 1 Data Read/Write, accessed
4 0 1 0 0 Data Read-Only, expand-down
5 0 1 0 1 Data Read-Only, expand-down, accessed
6 0 1 1 0 Data Read/Write, expand-down
7 0 1 1 1 Data Read/Write, expand-down, accessed
C R A
8 1 0 0 0 Code Execute-Only
9 1 0 0 1 Code Execute-Only, accessed
10 1 0 1 0 Code Execute/Read
11 1 0 1 1 Code Execute/Read, accessed
12 1 1 0 0 Code Execute-Only, conforming
13 1 1 0 1 Code Execute-Only, conforming, accessed
14 1 1 1 0 Code Execute/Read-Only, conforming
15 1 1 1 1 Code Execute/Read-Only, conforming, accessed

Stack segments are data segments which must be read/write segments. Loading the SS register
with a segment selector for a nonwritable data segment generates a general-protection exception
(#GP). If the size of a stack segment needs to be changed dynamically, the stack segment can be
an expand-down data segment (expansion-direction flag set). Here, dynamically changing the
segment limit causes stack space to be added to the bottom of the stack. If the size of a stack
segment is intended to remain static, the stack segment may be either an expand-up or expand-
down type.
The accessed bit indicates whether the segment has been accessed since the last time the oper-
ating-system or executive cleared the bit. The processor sets this bit whenever it loads a segment
selector for the segment into a segment register. The bit remains set until explicitly cleared. This
bit can be used both for virtual memory management and for debugging.
For code segments, the three low-order bits of the type field are interpreted as accessed (A), read
enable (R), and conforming (C). Code segments can be execute-only or execute/read, depending
on the setting of the read-enable bit. An execute/read segment might be used when constants or
other static data have been placed with instruction code in a ROM. Here, data can be read from
the code segment either by using an instruction with a CS override prefix or by loading a
segment selector for the code segment in a data-segment register (the DS, ES, FS, or GS regis-
ters). In protected mode, code segments are not writable.
Code segments can be either conforming or nonconforming. A transfer of execution into a more-
privileged conforming segment allows execution to continue at the current privilege level. A
transfer into a nonconforming segment at a different privilege level results in a general-protec-
tion exception (#GP), unless a call gate or task gate is used (refer to Section 4.8.1., “Direct Calls
or Jumps to Code Segments” in Chapter 4, Protection for more information on conforming and

3-14
 PROTECTED-MODE MEMORY MANAGEMENT

nonconforming code segments). System utilities that do not access protected facilities and
handlers for some types of exceptions (such as, divide error or overflow) may be loaded in
conforming code segments. Utilities that need to be protected from less privileged programs and
procedures should be placed in nonconforming code segments.

NOTE
Execution cannot be transferred by a call or a jump to a less-privileged
(numerically higher privilege level) code segment, regardless of whether the
target segment is a conforming or nonconforming code segment. Attempting
such an execution transfer will result in a general-protection exception.
All data segments are nonconforming, meaning that they cannot be accessed by less privileged
programs or procedures (code executing at numerically high privilege levels). Unlike code
segments, however, data segments can be accessed by more privileged programs or procedures
(code executing at numerically lower privilege levels) without using a special access gate.
The processor may update the Type field when a segment is accessed, even if the access is a read
cycle. If the descriptor tables have been put in ROM, it may be necessary for hardware to prevent
the ROM from being enabled onto the data bus during a write cycle. It also may be necessary to
return the READY# signal to the processor when a write cycle to ROM occurs, otherwise
the cycle will not terminate. These features of the hardware design are necessary for using
ROM-based descriptor tables with the Intel386™ DX processor, which always sets the
Accessed bit when a segment descriptor is loaded. The P6 family, Pentium®, and Intel486™
processors, however, only set the accessed bit if it is not already set. Writes to descriptor tables
in ROM can be avoided by setting the accessed bits in every descriptor.

3.5. SYSTEM DESCRIPTOR TYPES

When the S (descriptor type) flag in a segment descriptor is clear, the descriptor type is a system
descriptor. The processor recognizes the following types of system descriptors:
• Local descriptor-table (LDT) segment descriptor.
• Task-state segment (TSS) descriptor.
• Call-gate descriptor.
• Interrupt-gate descriptor.
• Trap-gate descriptor.
• Task-gate descriptor.
These descriptor types fall into two categories: system-segment descriptors and gate descriptors.
System-segment descriptors point to system segments (LDT and TSS segments). Gate descrip-
tors are in themselves “gates,” which hold pointers to procedure entry points in code segments
(call, interrupt, and trap gates) or which hold segment selectors for TSS’s (task gates). Table 3-2
shows the encoding of the type field for system-segment descriptors and gate descriptors.

3-15
PROTECTED-MODE MEMORY MANAGEMENT

Table 3-2. System-Segment and Gate-Descriptor Types

Type Field
Decimal 11 10 9 8 Description
0 0 0 0 0 Reserved
1 0 0 0 1 16-Bit TSS (Available)
2 0 0 1 0 LDT
3 0 0 1 1 16-Bit TSS (Busy)
4 0 1 0 0 16-Bit Call Gate
5 0 1 0 1 Task Gate
6 0 1 1 0 16-Bit Interrupt Gate
7 0 1 1 1 16-Bit Trap Gate
8 1 0 0 0 Reserved
9 1 0 0 1 32-Bit TSS (Available)
10 1 0 1 0 Reserved
11 1 0 1 1 32-Bit TSS (Busy)
12 1 1 0 0 32-Bit Call Gate
13 1 1 0 1 Reserved
14 1 1 1 0 32-Bit Interrupt Gate
15 1 1 1 1 32-Bit Trap Gate

For more information on the system-segment descriptors, refer to Section 3.5.1., “Segment
Descriptor Tables”, and Section 6.2.2., “TSS Descriptor” in Chapter 6, Task Management. For
more information on the gate descriptors, refer to Section 4.8.2., “Gate Descriptors” in Chapter
4, Protection; Section 5.9., “IDT Descriptors” in Chapter 5, Interrupt and Exception Handling;
and Section 6.2.4., “Task-Gate Descriptor” in Chapter 6, Task Management.

3.5.1. Segment Descriptor Tables

A segment descriptor table is an array of segment descriptors (refer to Figure 3-10). A descriptor
table is variable in length and can contain up to 8192 (213) 8-byte descriptors. There are two
kinds of descriptor tables:
• The global descriptor table (GDT)
• The local descriptor tables (LDT)

3-16
 PROTECTED-MODE MEMORY MANAGEMENT

Global Local
Descriptor Descriptor
Table (GDT) Table (LDT)

T
I TI = 0 TI = 1
Segment
Selector
56 56

48 48

40 40

32 32

24 24

16 16

8 8
First Descriptor in
GDT is Not Used 0 0

GDTR Register LDTR Register

Limit Limit
Base Address Base Address
Seg. Sel.

Figure 3-10. Global and Local Descriptor Tables

Each system must have one GDT defined, which may be used for all programs and tasks in the
system. Optionally, one or more LDTs can be defined. For example, an LDT can be defined for
each separate task being run, or some or all tasks can share the same LDT.
The GDT is not a segment itself; instead, it is a data structure in the linear address space. The
base linear address and limit of the GDT must be loaded into the GDTR register (refer to Section
2.4., “Memory-Management Registers” in Chapter 2, System Architecture Overview). The base
addresses of the GDT should be aligned on an eight-byte boundary to yield the best processor
performance. The limit value for the GDT is expressed in bytes. As with segments, the limit
value is added to the base address to get the address of the last valid byte. A limit value of 0
results in exactly one valid byte. Because segment descriptors are always 8 bytes long, the GDT
limit should always be one less than an integral multiple of eight (that is, 8N – 1).
The first descriptor in the GDT is not used by the processor. A segment selector to this “null
descriptor” does not generate an exception when loaded into a data-segment register (DS, ES,
FS, or GS), but it always generates a general-protection exception (#GP) when an attempt is

3-17
PROTECTED-MODE MEMORY MANAGEMENT

made to access memory using the descriptor. By initializing the segment registers with this
segment selector, accidental reference to unused segment registers can be guaranteed to generate
an exception.
The LDT is located in a system segment of the LDT type. The GDT must contain a segment
descriptor for the LDT segment. If the system supports multiple LDTs, each must have a sepa-
rate segment selector and segment descriptor in the GDT. The segment descriptor for an LDT
can be located anywhere in the GDT. Refer to Section 3.5., “System Descriptor Types” for infor-
mation on the LDT segment-descriptor type.
An LDT is accessed with its segment selector. To eliminate address translations when accessing
the LDT, the segment selector, base linear address, limit, and access rights of the LDT are stored
in the LDTR register (refer to Section 2.4., “Memory-Management Registers” in Chapter 2,
System Architecture Overview).
When the GDTR register is stored (using the SGDT instruction), a 48-bit “pseudo-descriptor”
is stored in memory (refer to Figure 3-11). To avoid alignment check faults in user mode (priv-
ilege level 3), the pseudo-descriptor should be located at an odd word address (that is, address
MOD 4 is equal to 2). This causes the processor to store an aligned word, followed by an aligned
doubleword. User-mode programs normally do not store pseudo-descriptors, but the possibility
of generating an alignment check fault can be avoided by aligning pseudo-descriptors in this
way. The same alignment should be used when storing the IDTR register using the SIDT instruc-
tion. When storing the LDTR or task register (using the SLTR or STR instruction, respectively),
the pseudo-descriptor should be located at a doubleword address (that is, address MOD 4 is
equal to 0).

47 16 15 0
Base Address Limit

Figure 3-11. Pseudo-Descriptor Format

3.6. PAGING (VIRTUAL MEMORY)

When operating in protected mode, the Intel Architecture permits the linear address space to be
mapped directly into a large physical memory (for example, 4 GBytes of RAM) or indirectly
(using paging) into a smaller physical memory and disk storage. This latter method of mapping
the linear address space is commonly referred to as virtual memory or demand-paged virtual
memory.
When paging is used, the processor divides the linear address space into fixed-size pages (gener-
ally 4 KBytes in length) that can be mapped into physical memory and/or disk storage. When a
program (or task) references a logical address in memory, the processor translates the address
into a linear address and then uses its paging mechanism to translate the linear address into a
corresponding physical address. If the page containing the linear address is not currently in
physical memory, the processor generates a page-fault exception (#PF). The exception handler
for the page-fault exception typically directs the operating system or executive to load the page
from disk storage into physical memory (perhaps writing a different page from physical memory

3-18
 PROTECTED-MODE MEMORY MANAGEMENT

out to disk in the process). When the page has been loaded in physical memory, a return from
the exception handler causes the instruction that generated the exception to be restarted. The
information that the processor uses to map linear addresses into the physical address space and
to generate page-fault exceptions (when necessary) is contained in page directories and page
tables stored in memory.
Paging is different from segmentation through its use of fixed-size pages. Unlike segments,
which usually are the same size as the code or data structures they hold, pages have a fixed size.
If segmentation is the only form of address translation used, a data structure present in physical
memory will have all of its parts in memory. If paging is used, a data structure can be partly in
memory and partly in disk storage.
To minimize the number of bus cycles required for address translation, the most recently
accessed page-directory and page-table entries are cached in the processor in devices called
translation lookaside buffers (TLBs). The TLBs satisfy most requests for reading the current
page directory and page tables without requiring a bus cycle. Extra bus cycles occur only when
the TLBs do not contain a page-table entry, which typically happens when a page has not been
accessed for a long time. Refer to Section 3.7., “Translation Lookaside Buffers (TLBs)” for
more information on the TLBs.

3.6.1. Paging Options

Paging is controlled by three flags in the processor’s control registers:
• PG (paging) flag, bit 31 of CR0 (available in all Intel Architecture processors beginning
with the Intel386™ processor).
• PSE (page size extensions) flag, bit 4 of CR4 (introduced in the Pentium® and Pentium®
Pro processors).
• PAE (physical address extension) flag, bit 5 of CR4 (introduced in the Pentium® Pro
processors).
The PG flag enables the page-translation mechanism. The operating system or executive usually
sets this flag during processor initialization. The PG flag must be set if the processor’s page-
translation mechanism is to be used to implement a demand-paged virtual memory system or if
the operating system is designed to run more than one program (or task) in virtual-8086 mode.
The PSE flag enables large page sizes: 4-MByte pages or 2-MByte pages (when the PAE flag is
set). When the PSE flag is clear, the more common page length of 4 KBytes is used. Refer to
Chapter 3.6.2.2., Linear Address Translation (4-MByte Pages) and Section 3.8.2., “Linear
Address Translation With Extended Addressing Enabled (2-MByte or 4-MByte Pages)” for
more information about the use of the PSE flag.
The PAE flag enables 36-bit physical addresses. This physical address extension can only be
used when paging is enabled. It relies on page directories and page tables to reference physical
addresses above FFFFFFFFH. Refer to Section 3.8., “Physical Address Extension” for more
information about the physical address extension.

3-19
PROTECTED-MODE MEMORY MANAGEMENT

3.6.2. Page Tables and Directories

The information that the processor uses to translate linear addresses into physical addresses
(when paging is enabled) is contained in four data structures:
• Page directory—An array of 32-bit page-directory entries (PDEs) contained in a 4-KByte
page. Up to 1024 page-directory entries can be held in a page directory.
• Page table—An array of 32-bit page-table entries (PTEs) contained in a 4-KByte page. Up
to 1024 page-table entries can be held in a page table. (Page tables are not used for 2-
MByte or 4-MByte pages. These page sizes are mapped directly from one or more page-
directory entries.)
• Page—A 4-KByte, 2-MByte, or 4-MByte flat address space.
• Page-Directory-Pointer Table—An array of four 64-bit entries, each of which points to a
page directory. This data structure is only used when the physical address extension is
enabled (refer to Section 3.8., “Physical Address Extension”).
These tables provide access to either 4-KByte or 4-MByte pages when normal 32-bit physical
addressing is being used and to either 4-KByte, 2-MByte, or 4-MByte pages when extended (36-
bit) physical addressing is being used. Table 3-3 shows the page size and physical address size
obtained from various settings of the paging control flags. Each page-directory entry contains a
PS (page size) flag that specifies whether the entry points to a page table whose entries in turn
point to 4-KByte pages (PS set to 0) or whether the page-directory entry points directly to a 4-
MByte or 2-MByte page (PSE or PAE set to 1 and PS set to 1).

Table 3-3. Page Sizes and Physical Address Sizes

PAE Flag, Physical
PG Flag, CR0 CR4 PSE Flag, CR4 PS Flag, PDE Page Size Address Size
0 X X X — Paging Disabled
1 0 0 X 4 KBytes 32 Bits
1 0 1 0 4 KBytes 32 Bits
1 0 1 1 4 MBytes 32 Bits
1 1 X 0 4 KBytes 36 Bits
1 1 X 1 2 MBytes 36 Bits

3.6.2.1. LINEAR ADDRESS TRANSLATION (4-KBYTE PAGES)

Figure 3-12 shows the page directory and page-table hierarchy when mapping linear addresses
to 4-KByte pages. The entries in the page directory point to page tables, and the entries in a page
table point to pages in physical memory. This paging method can be used to address up to 220
pages, which spans a linear address space of 232 bytes (4 GBytes).

3-20
 PROTECTED-MODE MEMORY MANAGEMENT

Linear Address
31 22 21 12 11 0
Directory Table Offset

12 4-KByte Page

10 10 Page Table Physical Address

Page Directory

Page-Table Entry

Directory Entry

32* 1024 PDE ∗ 1024 PTE = 220 Pages

CR3 (PDBR)

*32 bits aligned onto a 4-KByte boundary.

Figure 3-12. Linear Address Translation (4-KByte Pages)

To select the various table entries, the linear address is divided into three sections:
• Page-directory entry—Bits 22 through 31 provide an offset to an entry in the page
directory. The selected entry provides the base physical address of a page table.
• Page-table entry—Bits 12 through 21 of the linear address provide an offset to an entry in
the selected page table. This entry provides the base physical address of a page in physical
memory.
• Page offset—Bits 0 through 11 provides an offset to a physical address in the page.
Memory management software has the option of using one page directory for all programs and
tasks, one page directory for each task, or some combination of the two.

3.6.2.2. LINEAR ADDRESS TRANSLATION (4-MBYTE PAGES)

Figure 3-12 shows how a page directory can be used to map linear addresses to 4-MByte pages.
The entries in the page directory point to 4-MByte pages in physical memory. This paging
method can be used to map up to 1024 pages into a 4-GByte linear address space.

3-21
PROTECTED-MODE MEMORY MANAGEMENT

Linear Address
31 22 21 0
Directory Offset

22 4-MByte Page

10 Page Directory
Physical Address

Directory Entry

32* 1024 PDE = 1024 Pages

CR3 (PDBR)
*32 bits aligned onto a 4-KByte boundary.

Figure 3-13. Linear Address Translation (4-MByte Pages)

The 4-MByte page size is selected by setting the PSE flag in control register CR4 and setting
the page size (PS) flag in a page-directory entry (refer to Figure 3-14). With these flags set, the
linear address is divided into two sections:
• Page directory entry—Bits 22 through 31 provide an offset to an entry in the page
directory. The selected entry provides the base physical address of a 4-MByte page.
• Page offset—Bits 0 through 21 provides an offset to a physical address in the page.

NOTE
®
(For the Pentium processor only.) When enabling or disabling large page
sizes, the TLBs must be invalidated (flushed) after the PSE flag in control
register CR4 has been set or cleared. Otherwise, incorrect page translation
might occur due to the processor using outdated page translation information
stored in the TLBs. Refer to Section 9.10., “Invalidating the Translation
Lookaside Buffers (TLBs)”, in Chapter 9, Memory Cache Control, for
information on how to invalidate the TLBs.

3.6.2.3. MIXING 4-KBYTE AND 4-MBYTE PAGES

When the PSE flag in CR4 is set, both 4-MByte pages and page tables for 4-KByte pages can
be accessed from the same page directory. If the PSE flag is clear, only page tables for 4-KByte
pages can be accessed (regardless of the setting of the PS flag in a page-directory entry).
A typical example of mixing 4-KByte and 4-MByte pages is to place the operating system or
executive’s kernel in a large page to reduce TLB misses and thus improve overall system perfor-
mance. The processor maintains 4-MByte page entries and 4-KByte page entries in separate

3-22
 PROTECTED-MODE MEMORY MANAGEMENT

TLBs. So, placing often used code such as the kernel in a large page, frees up 4-KByte-page
TLB entries for application programs and tasks.

3.6.3. Base Address of the Page Directory

The physical address of the current page directory is stored in the CR3 register (also called the
page directory base register or PDBR). (Refer to Figure 2-5 and Section 2.5., “Control Regis-
ters” in Chapter 2, System Architecture Overview for more information on the PDBR.) If paging
is to be used, the PDBR must be loaded as part of the processor initialization process (prior to
enabling paging). The PDBR can then be changed either explicitly by loading a new value in
CR3 with a MOV instruction or implicitly as part of a task switch. (Refer to Section 6.2.1.,
“Task-State Segment (TSS)” in Chapter 6, Task Management for a description of how the
contents of the CR3 register is set for a task.)
There is no present flag in the PDBR for the page directory. The page directory may be not-
present (paged out of physical memory) while its associated task is suspended, but the operating
system must ensure that the page directory indicated by the PDBR image in a task's TSS is
present in physical memory before the task is dispatched. The page directory must also remain
in memory as long as the task is active.

3.6.4. Page-Directory and Page-Table Entries

Figure 3-14 shows the format for the page-directory and page-table entries when 4-KByte
pages and 32-bit physical addresses are being used. Figure 3-14 shows the format for the
page-directory entries when 4-MByte pages and 32-bit physical addresses are being used. Refer
to Section 3.8., “Physical Address Extension” for the format of page-directory and page-table
entries when the physical address extension is being used.

3-23
PROTECTED-MODE MEMORY MANAGEMENT

Page-Directory Entry (4-KByte Page Table)

31 12 11 9 8 7 6 5 4 3 2 1 0
P P U R
Page-Table Base Address Avail. G P 0 A C W / / P
S D T S W

Available for system programmer’s use

Global page (Ignored)
Page size (0 indicates 4 KBytes)
Reserved (set to 0)
Accessed
Cache disabled
Write-through
User/Supervisor
Read/Write
Present

Page-Table Entry (4-KByte Page)

31 12 11 9 8 7 6 5 4 3 2 1 0
P P U R
Page Base Address Avail. G 0 D A C W / / P
D T S W

Available for system programmer’s use

Global page
Reserved (set to 0)
Dirty
Accessed
Cache disabled
Write-through
User/Supervisor
Read/Write
Present

Figure 3-14. Format of Page-Directory and Page-Table Entries for 4-KByte Pages
and 32-Bit Physical Addresses

3-24
 PROTECTED-MODE MEMORY MANAGEMENT

Page-Directory Entry (4-MByte Page)

31 22 21 12 11 9 8 7 6 5 4 3 2 1 0
P P U R
Page Base Address Reserved Avail. G P D A C W / / P
S D T S W

Available for system programmer’s use

Global page
Page size (1 indicates 4 MBytes)
Dirty
Accessed
Cache disabled
Write-through
User/Supervisor
Read/Write
Present

Figure 3-15. Format of Page-Directory Entries for 4-MByte Pages and 32-Bit Addresses

The functions of the flags and fields in the entries in Figures 3-14 and 3-15 are as follows:
Page base address, bits 12 through 32
(Page-table entries for 4-KByte pages.) Specifies the physical address of the
first byte of a 4-KByte page. The bits in this field are interpreted as the 20 most-
significant bits of the physical address, which forces pages to be aligned on
4-KByte boundaries.
(Page-directory entries for 4-KByte page tables.) Specifies the physical
address of the first byte of a page table. The bits in this field are interpreted as
the 20 most-significant bits of the physical address, which forces page tables to
be aligned on 4-KByte boundaries.
(Page-directory entries for 4-MByte pages.) Specifies the physical address of
the first byte of a 4-MByte page. Only bits 22 through 31 of this field are used
(and bits 12 through 21 are reserved and must be set to 0, for Intel Architecture
processors through the Pentium® II processor). The base address bits are inter-
preted as the 10 most-significant bits of the physical address, which forces 4-
MByte pages to be aligned on 4-MByte boundaries.
Present (P) flag, bit 0
Indicates whether the page or page table being pointed to by the entry is
currently loaded in physical memory. When the flag is set, the page is in phys-
ical memory and address translation is carried out. When the flag is clear, the
page is not in memory and, if the processor attempts to access the page, it
generates a page-fault exception (#PF).
The processor does not set or clear this flag; it is up to the operating system or
executive to maintain the state of the flag.

3-25
PROTECTED-MODE MEMORY MANAGEMENT

The bit must be set to 1 whenever extended physical addressing mode is

enabled.
If the processor generates a page-fault exception, the operating system must
carry out the following operations in the order below:
1. Copy the page from disk storage into physical memory, if needed.
2. Load the page address into the page-table or page-directory entry and set
its present flag. Other bits, such as the dirty and accessed flags, may also
be set at this time.
3. Invalidate the current page-table entry in the TLB (refer to Section 3.7.,
“Translation Lookaside Buffers (TLBs)” for a discussion of TLBs and
how to invalidate them).
4. Return from the page-fault handler to restart the interrupted program or
task.
Read/write (R/W) flag, bit 1
Specifies the read-write privileges for a page or group of pages (in the case of
a page-directory entry that points to a page table). When this flag is clear, the
page is read only; when the flag is set, the page can be read and written into.
This flag interacts with the U/S flag and the WP flag in register CR0. Refer to
Section 4.11., “Page-Level Protection” and Table 4-2 in Chapter 4, Protection
for a detailed discussion of the use of these flags.
User/supervisor (U/S) flag, bit 2
Specifies the user-supervisor privileges for a page or group of pages (in the
case of a page-directory entry that points to a page table). When this flag is
clear, the page is assigned the supervisor privilege level; when the flag is set,
the page is assigned the user privilege level. This flag interacts with the R/W
flag and the WP flag in register CR0. Refer to Section 4.11., “Page-Level
Protection” and Table 4-2 in Chapter 4, Protection for a detail discussion of the
use of these flags.
Page-level write-through (PWT) flag, bit 3
Controls the write-through or write-back caching policy of individual pages or
page tables. When the PWT flag is set, write-through caching is enabled for the
associated page or page table; when the flag is clear, write-back caching is
enabled for the associated page or page table. The processor ignores this flag if
the CD (cache disable) flag in CR0 is set. Refer to Section 9.5., “Cache
Control”, in Chapter 9, Memory Cache Control, for more information about the
use of this flag. Refer to Section 2.5., “Control Registers” in Chapter 2, System
Architecture Overview for a description of a companion PWT flag in control
register CR3.
Page-level cache disable (PCD) flag, bit 4
Controls the caching of individual pages or page tables. When the PCD flag is
set, caching of the associated page or page table is prevented; when the flag is
clear, the page or page table can be cached. This flag permits caching to be

3-26
 PROTECTED-MODE MEMORY MANAGEMENT

disabled for pages that contain memory-mapped I/O ports or that do not
provide a performance benefit when cached. The processor ignores this flag
(assumes it is set) if the CD (cache disable) flag in CR0 is set. Refer to Chapter
9, Memory Cache Control, for more information about the use of this flag.
Refer to Section 2.5. in Chapter 2, System Architecture Overview for a descrip-
tion of a companion PCD flag in control register CR3.
Accessed (A) flag, bit 5
Indicates whether a page or page table has been accessed (read from or written
to) when set. Memory management software typically clears this flag when a
page or page table is initially loaded into physical memory. The processor then
sets this flag the first time a page or page table is accessed. This flag is a
“sticky” flag, meaning that once set, the processor does not implicitly clear it.
Only software can clear this flag. The accessed and dirty flags are provided for
use by memory management software to manage the transfer of pages and page
tables into and out of physical memory.
Dirty (D) flag, bit 6
Indicates whether a page has been written to when set. (This flag is not used in
page-directory entries that point to page tables.) Memory management soft-
ware typically clears this flag when a page is initially loaded into physical
memory. The processor then sets this flag the first time a page is accessed for
a write operation. This flag is “sticky,” meaning that once set, the processor
does not implicitly clear it. Only software can clear this flag. The dirty and
accessed flags are provided for use by memory management software to
manage the transfer of pages and page tables into and out of physical memory.
Page size (PS) flag, bit 7
Determines the page size. This flag is only used in page-directory entries.
When this flag is clear, the page size is 4 KBytes and the page-directory entry
points to a page table. When the flag is set, the page size is 4 MBytes for normal
32-bit addressing (and 2 MBytes if extended physical addressing is enabled)
and the page-directory entry points to a page. If the page-directory entry points
to a page table, all the pages associated with that page table will be 4-KByte
pages.
Global (G) flag, bit 8
(Introduced in the Pentium® Pro processor.) Indicates a global page when set.
When a page is marked global and the page global enable (PGE) flag in register
CR4 is set, the page-table or page-directory entry for the page is not invalidated
in the TLB when register CR3 is loaded or a task switch occurs. This flag is
provided to prevent frequently used pages (such as pages that contain kernel or
other operating system or executive code) from being flushed from the TLB.
Only software can set or clear this flag. For page-directory entries that point to
page tables, this flag is ignored and the global characteristics of a page are set
in the page-table entries. Refer to Section 3.7., “Translation Lookaside Buffers
(TLBs)” for more information about the use of this flag. (This bit is reserved in
Pentium® and earlier Intel Architecture processors.)

3-27
PROTECTED-MODE MEMORY MANAGEMENT

Reserved and available-to-software bits

In a page-table entry, bit 7 is reserved and should be set to 0; in a page-directory
entry that points to a page table, bit 6 is reserved and should be set to 0. For a
page-directory entry for a 4-MByte page, bits 12 through 21 are reserved and
must be set to 0, for Intel Architecture processors through the Pentium® II
processor. For both types of entries, bits 9, 10, and 11 are available for use by
software. (When the present bit is clear, bits 1 through 31 are available to soft-
ware—refer to Figure 3-16.) When the PSE and PAE flags in control register
CR4 are set, the processor generates a page fault if reserved bits are not set to 0.

3.6.5. Not Present Page-Directory and Page-Table Entries

When the present flag is clear for a page-table or page-directory entry, the operating system or
executive may use the rest of the entry for storage of information such as the location of the page
in the disk storage system (refer to ).

31 0

Available to Operating System or Executive 0

Figure 3-16. Format of a Page-Table or Page-Directory Entry for a Not-Present Page

3.7. TRANSLATION LOOKASIDE BUFFERS (TLBS)

The processor stores the most recently used page-directory and page-table entries in on-chip
caches called translation lookaside buffers or TLBs. The P6 family and Pentium® processors
have separate TLBs for the data and instruction caches. Also, the P6 family processors maintain
separate TLBs for 4-KByte and 4-MByte page sizes. The CPUID instruction can be used to
determine the sizes of the TLBs provided in the P6 family and Pentium® processors.
Most paging is performed using the contents of the TLBs. Bus cycles to the page directory and
page tables in memory are performed only when the TLBs do not contain the translation infor-
mation for a requested page.
The TLBs are inaccessible to application programs and tasks (privilege level greater than 0); that
is, they cannot invalidate TLBs. Only operating system or executive procedures running at priv-
ilege level of 0 can invalidate TLBs or selected TBL entries. Whenever a page-directory or
page-table entry is changed (including when the present flag is set to zero), the operating-system
must immediately invalidate the corresponding entry in the TLB so that it can be updated the
next time the entry is referenced. However, if the physical address extension (PAE) feature is
enabled to use 36-bit addressing, a new table is added to the paging hierarchy. This new table is
called the page directory pointer table (as described in Section 3.8., “Physical Address Exten-
sion”). If an entry is changed in this table (to point to another page directory), the TLBs must
then be flushed by writing to CR3.

3-28
 PROTECTED-MODE MEMORY MANAGEMENT

All (nonglobal) TLBs are automatically invalidated any time the CR3 register is loaded (unless
the G flag for a page or page-table entry is set, as describe later in this section). The CR3 register
can be loaded in either of two ways:
• Explicitly, using the MOV instruction, for example:
MOV CR3, EAX

where the EAX register contains an appropriate page-directory base address.

• Implicitly by executing a task switch, which automatically changes the contents of the CR3
register.
The INVLPG instruction is provided to invalidate a specific page-table entry in the TLB.
Normally, this instruction invalidates only an individual TLB entry; however, in some cases, it
may invalidate more than the selected entry and may even invalidate all of the TLBs. This
instruction ignores the setting of the G flag in a page-directory or page-table entry (refer to the
following paragraph).
(Introduced in the Pentium® Pro processor.) The page global enable (PGE) flag in register CR4
and the global (G) flag of a page-directory or page-table entry (bit 8) can be used to prevent
frequently used pages from being automatically invalidated in the TLBs on a task switch or a
load of register CR3. (Refer to Section 3.6.4., “Page-Directory and Page-Table Entries” for more
information about the global flag.) When the processor loads a page-directory or page-table
entry for a global page into a TLB, the entry will remain in the TLB indefinitely. The only way
to deterministically invalidate global page entries is to clear the PGE flag and then invalidate the
TLBs or to use the INVLPG instruction to invalidate individual page-directory or page-table
entries in the TLBs.
For additional information about invalidation of the TLBs, refer to Section 9.10., “Invalidating
the Translation Lookaside Buffers (TLBs)”, in Chapter 9, Memory Cache Control.

3.8. PHYSICAL ADDRESS EXTENSION

The physical address extension (PAE) flag in register CR4 enables an extension of physical
addresses from 32 bits to 36 bits. (This feature was introduced into the Intel Architecture in the
Pentium® Pro processors.) Here, the processor provides 4 additional address line pins to accom-
modate the additional address bits. This option can only be used when paging is enabled (that
is, when both the PG flag in register CR0 and the PAE flag in register CR4 are set).
When the physical address extension is enabled, the processor allows several sizes of pages:
4-KByte, 2-MByte, or 4-MByte. As with 32-bit addressing, these page sizes can be addressed
within the same set of paging tables (that is, a page-directory entry can point to either a 2-MByte
or 4-MByte page or a page table that in turn points to 4-KByte pages). To support the 36-bit
physical addresses, the following changes are made to the paging data structures:
• The paging table entries are increased to 64 bits to accommodate 36-bit base physical
addresses. Each 4-KByte page directory and page table can thus have up to 512 entries.

3-29
PROTECTED-MODE MEMORY MANAGEMENT

• A new table, called the page-directory-pointer table, is added to the linear-address

translation hierarchy. This table has 4 entries of 64-bits each, and it lies above the page
directory in the hierarchy. With the physical address extension mechanism enabled, the
processor supports up to 4 page directories.
• The 20-bit page-directory base address field in register CR3 (PDPR) is replaced with a
27-bit page-directory-pointer-table base address field (refer to Figure 3-17). (In this case,
register CR3 is called the PDPTR.) This field provides the 27 most-significant bits of the
physical address of the first byte of the page-directory-pointer table, which forces the table
to be located on a 32-byte boundary.
• Linear address translation is changed to allow mapping 32-bit linear addresses into the
larger physical address space.

31 0
P P
Page-Directory-Pointer-Table Base Address C W 0 0 0
D T

Figure 3-17. Register CR3 Format When the Physical Address Extension is Enabled

3.8.1. Linear Address Translation With Extended Addressing

Enabled (4-KByte Pages)
Figure 3-12 shows the page-directory-pointer, page-directory, and page-table hierarchy when
mapping linear addresses to 4-KByte pages with extended physical addressing enabled. This
paging method can be used to address up to 220 pages, which spans a linear address space of 232
bytes (4 GBytes).

3-30
 PROTECTED-MODE MEMORY MANAGEMENT

Linear Address
31 30 29 21 20 12 11 0
Directory Pointer Directory Table Offset

12 4-KByte Page

Page Table Physical Address

Page Directory 9
9
Page-Table Entry
2
Directory Entry

Page-Directory-
Pointer Table
4 PDPTE ∗ 512 PDE ∗ 512 PTE = 220 Pages

Dir. Pointer Entry

32*
CR3 (PDBR)
*32 bits aligned onto a 32-byte boundary

Figure 3-18. Linear Address Translation With Extended Physical Addressing Enabled
(4-KByte Pages)

To select the various table entries, the linear address is divided into three sections:
• Page-directory-pointer-table entry—Bits 30 and 31 provide an offset to one of the 4 entries
in the page-directory-pointer table. The selected entry provides the base physical address
of a page directory.
• Page-directory entry—Bits 21 through 29 provide an offset to an entry in the selected page
directory. The selected entry provides the base physical address of a page table.
• Page-table entry—Bits 12 through 20 provide an offset to an entry in the selected page
table. This entry provides the base physical address of a page in physical memory.
• Page offset—Bits 0 through 11 provide an offset to a physical address in the page.

3-31
PROTECTED-MODE MEMORY MANAGEMENT

3.8.2. Linear Address Translation With Extended Addressing

Enabled (2-MByte or 4-MByte Pages)
Figure 3-12 shows how a page-directory-pointer table and page directories can be used to map
linear addresses to 2-MByte or 4-MByte pages. This paging method can be used to map up to
2048 pages (4 page-directory-pointer-table entries times 512 page-directory entries) into a
4-GByte linear address space.
The 2-MByte or 4-MByte page size is selected by setting the PSE flag in control register CR4
and setting the page size (PS) flag in a page-directory entry (refer to Figure 3-14). With these
flags set, the linear address is divided into three sections:
• Page-directory-pointer-table entry—Bits 30 and 31 provide an offset to an entry in the
page-directory-pointer table. The selected entry provides the base physical address of a
page directory.
• Page-directory entry—Bits 21 through 29 provide an offset to an entry in the page
directory. The selected entry provides the base physical address of a 2-MByte or 4-MByte
page.
• Page offset—Bits 0 through 20 provides an offset to a physical address in the page.

3.8.3. Accessing the Full Extended Physical Address Space

With the Extended Page-Table Structure
The page-table structure described in the previous two sections allows up to 4 GBytes of
the 64-GByte extended physical address space to be addressed at one time. Additional 4-GByte
sections of physical memory can be addressed in either of two way:
• Change the pointer in register CR3 to point to another page-directory-pointer table, which
in turn points to another set of page directories and page tables.
• Change entries in the page-directory-pointer table to point to other page directories, which
in turn point to other sets of page tables.

3-32
 PROTECTED-MODE MEMORY MANAGEMENT

Linear Address
31 30 29 21 20 0
Directory Offset
Pointer Directory

21 2 or 4-MByte Pages

9
Page Directory Physical Address
Page-Directory-
Pointer Table
2
Directory Entry
Dir. Pointer Entry

32* 4 PDPTE ∗ 512 PDE = 2048 Pages

CR3 (PDBR)
*32 bits aligned onto a 32-byte boundary

Figure 3-19. Linear Address Translation With Extended Physical Addressing Enabled
(2-MByte or 4-MByte Pages)

3.8.4. Page-Directory and Page-Table Entries With Extended

Addressing Enabled
Figure 3-20 shows the format for the page-directory-pointer-table, page-directory, and
page-table entries when 4-KByte pages and 36-bit extended physical addresses are being
used. Figure 3-21 shows the format for the page-directory-pointer-table and page-directory
entries when 2-MByte or 4-MByte pages and 36-bit extended physical addresses are being
used. The functions of the flags in these entries are the same as described in Section 3.6.4.,
“Page-Directory and Page-Table Entries”. The major differences in these entries are as follows:
• A page-directory-pointer-table entry is added.
• The size of the entries are increased from 32 bits to 64 bits.
• The maximum number of entries in a page directory or page table is 512.
• The base physical address field in each entry is extended to 24 bits.

3-33
PROTECTED-MODE MEMORY MANAGEMENT

Page-Directory-Pointer-Table Entry
63 36 35 32
Base
Reserved (set to 0) Addr.

31 12 11 9 8 5 4 3 2 1 0
P P
Page-Directory Base Address Avail. Reserved C W Res. 1
D T

Page-Directory Entry (4-KByte Page Table)

63 36 35 32
Base
Reserved (set to 0) Addr.

31 12 11 9 8 7 6 5 4 3 2 1 0
P P U R
Page-Table Base Address Avail. 0 0 0 A C W / / P
D T S W

Page-Table Entry (4-KByte Page)

63 36 35 32
Base
Reserved (set to 0) Addr.

31 12 11 9 8 7 6 5 4 3 2 1 0
P P U R
Page Base Address Avail. G 0 D A C W / / P
D T S W

Figure 3-20. Format of Page-Directory-Pointer-Table, Page-Directory, and Page-Table

Entries for 4-KByte Pages and 36-Bit Extended Physical Addresses

The base physical address in an entry specifies the following, depending on the type of entry:
• Page-directory-pointer-table entry—the physical address of the first byte of a
4-KByte page directory.
• Page-directory entry—the physical address of the first byte of a 4-KByte page table or a
2-MByte page.
• Page-table entry—the physical address of the first byte of a 4-KByte page.
For all table entries (except for page-directory entries that point to 2-MByte or 4-MByte pages),
the bits in the page base address are interpreted as the 24 most-significant bits of a 36-bit phys-
ical address, which forces page tables and pages to be aligned on 4-KByte boundaries. When a
page-directory entry points to a 2-MByte or 4-MByte page, the base address is interpreted as the
15 most-significant bits of a 36-bit physical address, which forces pages to be aligned on 2-
MByte or 4-MByte boundaries.

3-34
 PROTECTED-MODE MEMORY MANAGEMENT

Page-Directory-Pointer-Table Entry
63 36 35 32
Base
Reserved (set to 0) Addr.

31 12 11 9 8 5 4 3 2 1 0
P P
Page Directory Base Address Avail. Reserved C W Res. 1
D T

Page-Directory Entry (2- or 4-MByte Pages)

63 36 35 32
Base
Reserved (set to 0) Addr.

31 21 20 12 11 9 8 7 6 5 4 3 2 1 0
P P U R
Page Base Address Reserved (set to 0) Avail. G 1 D A C W / / P
D T S W

Figure 3-21. Format of Page-Directory-Pointer-Table and Page-Directory Entries for

2- or 4-MByte Pages and 36-Bit Extended Physical Addresses

The present (P) flag (bit 0) in all page-directory-pointer-table entries must be set to 1 anytime
extended physical addressing mode is enabled; that is, whenever the PAE flag (bit 5 in register
CR4) and the PG flag (bit 31 in register CR0) are set. If the P flag is not set in all 4 page-direc-
tory-pointer-table entries in the page-directory-pointer table when extended physical addressing
is enabled, a general-protection exception (#GP) is generated.
The page size (PS) flag (bit 7) in a page-directory entry determines if the entry points to a page
table or a 2-MByte or 4-MByte page. When this flag is clear, the entry points to a page table;
when the flag is set, the entry points to a 2-MByte or 4-MByte page. This flag allows 4-KByte,
2-MByte, or 4-MByte pages to be mixed within one set of paging tables.
Access (A) and dirty (D) flags (bits 5 and 6) are provided for table entries that point to pages.
Bits 9, 10, and 11 in all the table entries for the physical address extension are available for use
by software. (When the present flag is clear, bits 1 through 63 are available to software.) All bits
in Figure 3-14 that are marked reserved or 0 should be set to 0 by software and not accessed by
software. When the PSE and/or PAE flags in control register CR4 are set, the processor gener-
ates a page fault (#PF) if reserved bits in page-directory and page-table entries are not set to 0,
and it generates a general-protection exception (#GP) if reserved bits in a page-directory-
pointer-table entry are not set to 0.

3.9. 36-BIT PAGE SIZE EXTENSION (PSE)

The 36-bit PSE extends 36-bit physical address support to 4-MByte pages while maintaining a
4-byte page-directory entry. This approach provides a simple mechanism for operating system

3-35
PROTECTED-MODE MEMORY MANAGEMENT

vendors to address physical memory above 4-GBytes without requiring major design changes,
but has practical limitations with respect to demand paging.
The P6 family of processors’ physical address extension (PAE) feature provides generic access
to a 36-bit physical address space. However, it requires expansion of the page-directory and
page-table entries to an 8-byte format (64 bit), and the addition of a page-directory-pointer table,
resulting in another level of indirection to address translation.
For P6-family processors that support the 36-bit PSE feature, the virtual memory architecture is
extended to support 4-MByte page size granularity in combination with 36-bit physical
addressing. Note that some P6-family processors do not support this feature. For information
about determining a processor’s feature support, refer to the following documents:
• AP-485, Intel Processor Identification and the CPUID Instruction
• Addendum—Intel Architecture Software Developer’s Manual, Volume1: Basic Archi-
tecture
For information about the virtual memory architecture features of P6-family processors, refer to
Chapter 3 of the Intel Architecture Software Developer’s Manual, Volume3: System Program-
ming Guide.

3.9.1. Description of the 36-bit PSE Feature

The 36-bit PSE feature (PSE-36) is detected by an operating system through the CPUID instruc-
tion. Specifically, the operating system executes the CPUID instruction with the value 1 in the
EAX register and then determines support for the feature by inspecting bit 17 of the EDX
register return value (see Addendum—Intel Architecture Software Developer’s Manual,
Volume1: Basic Architecture). If the PSE-36 feature is supported, an operating system is
permitted to utilize the feature, as well as use certain formerly reserved bits. To use the 36-bit
PSE feature, the PSE flag must be enabled by the operating system (bit 4 of CR4). Note that a
separate control bit in CR 4 does not exist to regulate the use of 36-bit MByte pages, because
this feature becomes the example for 4-MByte pages on processors that support it.
Table 3-8 shows the page size and physical address size obtained from various settings of the
page-control flags for the P6-family processors that support the 36-bit PSE feature. Shaded in
gray is the change to this table resulting from the 36-bit PSE feature.

3-36
 PROTECTED-MODE MEMORY MANAGEMENT

Table 3-4. Paging Modes and Physical Address Size

PG Flag PAE Flag PSE Flag PS Flag Page Physical
(in CR0) (in CR4) (in CR4) (in the PDE) Size Address Size
0 X X X — Paging Disabled
1 0 0 X 4 KB 32 bits
1 0 1 0 4 KB 32 bits
1 0 1 1 4 KB 36 bits
1 1 X 0 4 KB 36 bits
1 1 X 1 2 KB 36 bits

To use the 36-bit PSE feature, the PAE feature must be cleared (as indicated in Table 3-4).
However, the 36-bit PSE in no way affects the PAE feature. Existing operating systems and soft-
wware that use the PAE will continue to have compatible functionality and features with P6-
family processors that support 36-bit PSE. Specifically, the Page-Directory Entry (PDE) format
when PAE is enabled for 2-MByte or 4-MByte pages is exactly as depicted in Figure 3-21 of the
Intel Architecture Software Developer’s Manual, Volume3: System Programming Guide.
No matter which 36-bit addressing feature is used (PAE or 36-bit PSE), the linear address space
of the processor remains at 32 bits. Applications must partition the address space of their work
loads across multiple operating system process to take advantage of the additonal physical
memory provided in the system.
The 36-bit PSE feature estends the PDE format of the Intel Architecture for 4-MByte pages and
32-bit addresses by utilizing bits 16-13 (formerly reserved bits that were required to be zero) to
extend the physical address without requiring an 8-byte page-directory entry. Therefore, with
the 36-bit PSE feature, a page directory can contain up to 1024 entries, each pointing to a 4-
MByte page that can exist anywhere in the 36-bit physical address space of the processor.
Figure 3-22 shows the difference between PDE formats for 4-MByte pages on P6-family proces-
sors that support the 36-bit PSE feature compared to P6-family processors that do not support
the 36-bit PSE feature (i.e., 32-bit addressing).
Figure 3-22 also shows the linear address mapping to 4-MByte pages when the 36-bit PSE is
enabled. The base physical address of the 4-MByte page is contained in the PDE. PA-2 (bits 13-
16) is used to provide the upper four bits (bits 32-35) of the 36-bit physical address. PA-1 (bits
22-31) continues to provide the next ten bits (bits 22-31) of the physical address for the 4-MByte
page. The offset into the page is provided by the lower 22 bits of the linear address. This scheme
eliminates the second level of indirection caused by the use of 4-KByte page tables.

3-37
PROTECTED-MODE MEMORY MANAGEMENT

Page Directory Entry format for processors that support 36-bit addressing for 4-MByte pages
31 22 21 17 16 13 12 11 8 7 6 0
PA - 1 Reserved PA - 2 PAT PS=1
Page Directory Entry format for processors that support 32-bit addressing for 4-MByte pages
31 22 21 12 11 8 7 6 0
Base Page Address Reserved PS=1

Figure 3-22. PDE Format Differences between 36-bit and 32-bit addressing

Notes:
1. PA-2 = Bits 35-32 of thebase physical address for the 4-MByte page (correspond to bits 16-13)
2. PA-2 = Bits 31-22 of thebase physical address for the 4-MByte page
3. PAT = Bit 12 used as the Most Significant Bit of the index into Page Attribute Table (PAT); see Section
10.2.
4. PS = Bit 7 is the Page Size Bit—indicates 4-MByte page (must be set to 1)
5. Reserved = Bits 21-17 are reserved for future expansion
6. No change in format or meaning of bits 11-8 and 6-0; refer to Figure 3-15 for details.

The PSE-36 feature is transparent to existing operating systems that utilize 4-MByte pages,
because unused bits in PA-2 are currently enforced as zero by Intel processors. The feature
requires 4-MByte pages aligned on a 4-MByte boundary and 4 MBytes of physically contiguous
memory. Therefore, the ten bits of PA-1 are sufficient to specify the base physical address of any
4-MByte page below 4 GBytes. An operating system can easily support addresses greater than
4 GBytes simply by providing the upper 4 bits of the physical address in PA-2 when creating a
PDE for a 4-MByte page.
Figure 3-23 shows the linear address mapping to 4 MB pages when the 36-bit PSE is enabled.
The base physical address of the 4 MB page is contained in the PDE. PA-2 (bits 13-16) is used
to provide the upper four bits (bits 32-35) of the 36-bit physical address. PA-1 (bits 22-31)
continues to provide the next ten bits (bits 22-31) of the physical address for the 4 MB page. The
offset into the page is provided by the lower 22 bits of the linear address. This scheme eliminates
the second level of indirection caused by the use of 4 KB page tables.

3-38
 PROTECTED-MODE MEMORY MANAGEMENT

Linear Address 4 MB Page

31 22 21 0
Directory Index

Page Directory
31 2221 17 16 13 12 11 8 7 6 0

Page Frame Address Reserved PA-2 PAT PS=1

PA-1

CR3

Figure 3-23. Page Size Extension Linear to Physical Translation

The PSE-36 feature is transparent to existing operating systems that utilize 4 MB pages because
unused bits in PA-2 are currently enforced as zero by Intel processors. The feature requires 4
MB pages aligned on a 4 MB boundary and 4 MB of physically contiguous memory. Therefore,
the ten bits of PA-1 are sufficient to specify the base physical address of any 4 MB page below
4GB. An operating system easily can support addresses greater than 4 GB simply by providing
the upper 4 bits of the physical address in PA-2 when creating a PDE for a 4 MB page.

3.9.2. Fault Detection

There are several conditions that can cause P6-family processors that support this feature to
generate a page fault (PF) fault. These conditions are related to the use of, or switching between,
various memory management features:
• If the PSE feature is enabled, a nonzero value in any of the remaining reserved bits (17-21)
of a 4-MByte PDE causes a page fault, with the reserved bit (bit 3) set in the error code.
• If the PAE feature is enabled and set to use 2-MByte or 4-MByte pages (that is, 8-byte
page-directory table entries are being used), a nonzero value in any of the reserved bits 13-
20 causes a page fault, with the reserved bit (bit 3) set in the error code. Note that bit 12 is
now being used to support the Page Attribute Table feature (refer to Section 9.13., “Page
Attribute Table (PAT)”).

3-39
PROTECTED-MODE MEMORY MANAGEMENT

3.10. MAPPING SEGMENTS TO PAGES

The segmentation and paging mechanisms provide in the Intel Architecture support a wide
variety of approaches to memory management. When segmentation and paging is combined,
segments can be mapped to pages in several ways. To implement a flat (unsegmented)
addressing environment, for example, all the code, data, and stack modules can be mapped to
one or more large segments (up to 4-GBytes) that share same range of linear addresses (refer to
Figure 3-2). Here, segments are essentially invisible to applications and the operating-system or
executive. If paging is used, the paging mechanism can map a single linear address space
(contained in a single segment) into virtual memory. Or, each program (or task) can have its own
large linear address space (contained in its own segment), which is mapped into virtual memory
through its own page directory and set of page tables.
Segments can be smaller than the size of a page. If one of these segments is placed in a page
which is not shared with another segment, the extra memory is wasted. For example, a small data
structure, such as a 1-byte semaphore, occupies 4K bytes if it is placed in a page by itself. If
many semaphores are used, it is more efficient to pack them into a single page.
The Intel Architecture does not enforce correspondence between the boundaries of pages and
segments. A page can contain the end of one segment and the beginning of another. Likewise, a
segment can contain the end of one page and the beginning of another.
Memory-management software may be simpler and more efficient if it enforces some alignment
between page and segment boundaries. For example, if a segment which can fit in one page is
placed in two pages, there may be twice as much paging overhead to support access to that
segment.
One approach to combining paging and segmentation that simplifies memory-management soft-
ware is to give each segment its own page table, as shown in Figure 3-24. This convention gives
the segment a single entry in the page directory that provides the access control information for
paging the entire segment.

Page Frames

LDT Page Directory Page Tables

PTE
PTE
PTE
Seg. Descript. PDE
Seg. Descript. PDE

PTE
PTE

Figure 3-24. Memory Management Convention That Assigns a Page Table to Each
Segment

3-40
 4
Protection
 PROTECTION

CHAPTER 4
PROTECTION

In protected mode, the Intel Architecture provides a protection mechanism that operates at both
the segment level and the page level. This protection mechanism provides the ability to limit
access to certain segments or pages based on privilege levels (four privilege levels for segments
and two privilege levels for pages). For example, critical operating-system code and data can be
protected by placing them in more privileged segments than those that contain applications
code. The processor’s protection mechanism will then prevent application code from accessing
the operating-system code and data in any but a controlled, defined manner.
Segment and page protection can be used at all stages of software development to assist in local-
izing and detecting design problems and bugs. It can also be incorporated into end-products to
offer added robustness to operating systems, utilities software, and applications software.
When the protection mechanism is used, each memory reference is checked to verify that it
satisfies various protection checks. All checks are made before the memory cycle is started; any
violation results in an exception. Because checks are performed in parallel with address transla-
tion, there is no performance penalty. The protection checks that are performed fall into the
following categories:
• Limit checks.
• Type checks.
• Privilege level checks.
• Restriction of addressable domain.
• Restriction of procedure entry-points.
• Restriction of instruction set.
All protection violation results in an exception being generated. Refer to Chapter 5, Interrupt
and Exception Handling for an explanation of the exception mechanism. This chapter describes
the protection mechanism and the violations which lead to exceptions.
The following sections describe the protection mechanism available in protected mode. Refer to
Chapter 16, 8086 Emulation for information on protection in real-address and virtual-8086
mode.

4-1
PROTECTION

4.1. ENABLING AND DISABLING SEGMENT AND PAGE

PROTECTION
Setting the PE flag in register CR0 causes the processor to switch to protected mode, which in
turn enables the segment-protection mechanism. Once in protected mode, there is no control bit
for turning the protection mechanism on or off. The part of the segment-protection mechanism
that is based on privilege levels can essentially be disabled while still in protected mode by
assigning a privilege level of 0 (most privileged) to all segment selectors and segment descrip-
tors. This action disables the privilege level protection barriers between segments, but other
protection checks such as limit checking and type checking are still carried out.
Page-level protection is automatically enabled when paging is enabled (by setting the PG flag
in register CR0). Here again there is no mode bit for turning off page-level protection once
paging is enabled. However, page-level protection can be disabled by performing the following
operations:
• Clear the WP flag in control register CR0.
• Set the read/write (R/W) and user/supervisor (U/S) flags for each page-directory and page-
table entry.
This action makes each page a writable, user page, which in effect disables page-level
protection.

4.2. FIELDS AND FLAGS USED FOR SEGMENT-LEVEL AND

PAGE-LEVEL PROTECTION
The processor’s protection mechanism uses the following fields and flags in the system data
structures to control access to segments and pages:
• Descriptor type (S) flag—(Bit 12 in the second doubleword of a segment descriptor.)
Determines if the segment descriptor is for a system segment or a code or data segment.
• Type field—(Bits 8 through 11 in the second doubleword of a segment descriptor.)
Determines the type of code, data, or system segment.
• Limit field—(Bits 0 through 15 of the first doubleword and bits 16 through 19 of the
second doubleword of a segment descriptor.) Determines the size of the segment, along
with the G flag and E flag (for data segments).
• G flag—(Bit 23 in the second doubleword of a segment descriptor.) Determines the size of
the segment, along with the limit field and E flag (for data segments).
• E flag—(Bit 10 in the second doubleword of a data-segment descriptor.) Determines the
size of the segment, along with the limit field and G flag.
• Descriptor privilege level (DPL) field—(Bits 13 and 14 in the second doubleword of a
segment descriptor.) Determines the privilege level of the segment.
• Requested privilege level (RPL) field. (Bits 0 and 1 of any segment selector.) Specifies the
requested privilege level of a segment selector.

4-2
 PROTECTION

• Current privilege level (CPL) field. (Bits 0 and 1 of the CS segment register.) Indicates the
privilege level of the currently executing program or procedure. The term current privilege
level (CPL) refers to the setting of this field.
• User/supervisor (U/S) flag. (Bit 2 of a page-directory or page-table entry.) Determines the
type of page: user or supervisor.
• Read/write (R/W) flag. (Bit 1 of a page-directory or page-table entry.) Determines the type
of access allowed to a page: read only or read-write.
Figure 4-1 shows the location of the various fields and flags in the data, code, and system-
segment descriptors; Figure 3-6 in Chapter 3, Protected-Mode Memory Management shows the
location of the RPL (or CPL) field in a segment selector (or the CS register); and Figure 3-14 in
Chapter 3, Protected-Mode Memory Management shows the location of the U/S and R/W flags
in the page-directory and page-table entries.

4-3
PROTECTION

Data-Segment Descriptor
31 24 23 22 21 20 19 16 15 14 13 12 11 8 7 0

A Limit D Type
Base 31:24 G B 0 V P P Base 23:16 4
19:16
L L 1 0 E W A

31 16 15 0

Base Address 15:00 Segment Limit 15:00 0

Code-Segment Descriptor
31 24 23 22 21 20 19 16 15 14 13 12 11 8 7 0

A Limit D Type
Base 31:24 G D 0 V P P Base 23:16 4
19:16
L L 1 1 C R A

31 16 15 0

Base Address 15:00 Segment Limit 15:00 0

System-Segment Descriptor
31 24 23 22 21 20 19 16 15 14 13 12 11 8 7 0

Limit D
Base 31:24 G 0 P P 0 Type Base 23:16 4
19:16
L

31 16 15 0

Base Address 15:00 Segment Limit 15:00 0

A Accessed E Expansion Direction

AVL Available to Sys. Programmer’s G Granularity
B Big R Readable
C Conforming LIMIT Segment Limit
D Default W Writable
DPL Descriptor Privilege Level P Present
Reserved

Figure 4-1. Descriptor Fields Used for Protection

Many different styles of protection schemes can be implemented with these fields and flags.
When the operating system creates a descriptor, it places values in these fields and flags in
keeping with the particular protection style chosen for an operating system or executive. Appli-
cation program do not generally access or modify these fields and flags.
The following sections describe how the processor uses these fields and flags to perform the
various categories of checks described in the introduction to this chapter.

4-4
 PROTECTION

4.3. LIMIT CHECKING

The limit field of a segment descriptor prevents programs or procedures from addressing
memory locations outside the segment. The effective value of the limit depends on the setting
of the G (granularity) flag (refer to Figure 4-1). For data segments, the limit also depends on the
E (expansion direction) flag and the B (default stack pointer size and/or upper bound) flag. The
E flag is one of the bits in the type field when the segment descriptor is for a data-segment type.
When the G flag is clear (byte granularity), the effective limit is the value of the 20-bit limit field
in the segment descriptor. Here, the limit ranges from 0 to FFFFFH (1 MByte). When the G flag
is set (4-KByte page granularity), the processor scales the value in the limit field by a factor of
2^12 (4 KBytes). In this case, the effective limit ranges from FFFH (4 KBytes) to FFFFFFFFH
(4 GBytes). Note that when scaling is used (G flag is set), the lower 12 bits of a segment offset
(address) are not checked against the limit; for example, note that if the segment limit is 0,
offsets 0 through FFFH are still valid.
For all types of segments except expand-down data segments, the effective limit is the last
address that is allowed to be accessed in the segment, which is one less than the size, in bytes,
of the segment. The processor causes a general-protection exception any time an attempt is made
to access the following addresses in a segment:
• A byte at an offset greater than the effective limit
• A word at an offset greater than the (effective-limit – 1)
• A doubleword at an offset greater than the (effective-limit – 3)
• A quadword at an offset greater than the (effective-limit – 7)
For expand-down data segments, the segment limit has the same function but is interpreted
differently. Here, the effective limit specifies the last address that is not allowed to be accessed
within the segment; the range of valid offsets is from (effective-limit + 1) to FFFFFFFFH if the
B flag is set and from (effective-limit + 1) to FFFFH if the B flag is clear. An expand-down
segment has maximum size when the segment limit is 0.
Limit checking catches programming errors such as runaway code, runaway subscripts, and
invalid pointer calculations. These errors are detected when they occur, so identification of the
cause is easier. Without limit checking, these errors could overwrite code or data in another
segment.
In addition to checking segment limits, the processor also checks descriptor table limits. The
GDTR and IDTR registers contain 16-bit limit values that the processor uses to prevent
programs from selecting a segment descriptors outside the respective descriptor tables. The
LDTR and task registers contain 32-bit segment limit value (read from the segment descriptors
for the current LDT and TSS, respectively). The processor uses these segment limits to prevent
accesses beyond the bounds of the current LDT and TSS. Refer to Section 3.5.1., “Segment
Descriptor Tables” in Chapter 3, Protected-Mode Memory Management for more information
on the GDT and LDT limit fields; refer to Section 5.8., “Interrupt Descriptor Table (IDT)” in
Chapter 5, Interrupt and Exception Handling for more information on the IDT limit field; and
refer to Section 6.2.3., “Task Register” in Chapter 6, Task Management for more information on
the TSS segment limit field.

4-5
PROTECTION

4.4. TYPE CHECKING

Segment descriptors contain type information in two places:
• The S (descriptor type) flag.
• The type field.
The processor uses this information to detect programming errors that result in an attempt to use
a segment or gate in an incorrect or unintended manner.
The S flag indicates whether a descriptor is a system type or a code or data type. The type field
provides 4 additional bits for use in defining various types of code, data, and system descriptors.
Table 3-1 in Chapter 3, Protected-Mode Memory Management shows the encoding of the type
field for code and data descriptors; Table 3-2 in Chapter 3, Protected-Mode Memory Manage-
ment shows the encoding of the field for system descriptors.
The processor examines type information at various times while operating on segment selectors
and segment descriptors. The following list gives examples of typical operations where type
checking is performed. This list is not exhaustive.
• When a segment selector is loaded into a segment register. Certain segment registers
can contain only certain descriptor types, for example:
— The CS register only can be loaded with a selector for a code segment.
— Segment selectors for code segments that are not readable or for system segments
cannot be loaded into data-segment registers (DS, ES, FS, and GS).
— Only segment selectors of writable data segments can be loaded into the SS register.
• When a segment selector is loaded into the LDTR or task register.
— The LDTR can only be loaded with a selector for an LDT.
— The task register can only be loaded with a segment selector for a TSS.
• When instructions access segments whose descriptors are already loaded into
segment registers. Certain segments can be used by instructions only in certain predefined
ways, for example:
— No instruction may write into an executable segment.
— No instruction may write into a data segment if it is not writable.
— No instruction may read an executable segment unless the readable flag is set.
• When an instruction operand contains a segment selector. Certain instructions can
access segment or gates of only a particular type, for example:
— A far CALL or far JMP instruction can only access a segment descriptor for a
conforming code segment, nonconforming code segment, call gate, task gate, or TSS.
— The LLDT instruction must reference a segment descriptor for an LDT.
— The LTR instruction must reference a segment descriptor for a TSS.

4-6
 PROTECTION

— The LAR instruction must reference a segment or gate descriptor for an LDT, TSS,
call gate, task gate, code segment, or data segment.
— The LSL instruction must reference a segment descriptor for a LDT, TSS, code
segment, or data segment.
— IDT entries must be interrupt, trap, or task gates.
• During certain internal operations. For example:
— On a far call or far jump (executed with a far CALL or far JMP instruction), the
processor determines the type of control transfer to be carried out (call or jump to
another code segment, a call or jump through a gate, or a task switch) by checking the
type field in the segment (or gate) descriptor pointed to by the segment (or gate)
selector given as an operand in the CALL or JMP instruction. If the descriptor type is
for a code segment or call gate, a call or jump to another code segment is indicated; if
the descriptor type is for a TSS or task gate, a task switch is indicated.
— On a call or jump through a call gate (or on an interrupt- or exception-handler call
through a trap or interrupt gate), the processor automatically checks that the segment
descriptor being pointed to by the gate is for a code segment.
— On a call or jump to a new task through a task gate (or on an interrupt- or exception-
handler call to a new task through a task gate), the processor automatically checks that
the segment descriptor being pointed to by the task gate is for a TSS.
— On a call or jump to a new task by a direct reference to a TSS, the processor automati-
cally checks that the segment descriptor being pointed to by the CALL or JMP
instruction is for a TSS.
— On return from a nested task (initiated by an IRET instruction), the processor checks
that the previous task link field in the current TSS points to a TSS.

4.4.1. Null Segment Selector Checking

Attempting to load a null segment selector (refer to Section 3.4.1. in Chapter 3, Protected-Mode
Memory Management) into the CS or SS segment register generates a general-protection excep-
tion (#GP). A null segment selector can be loaded into the DS, ES, FS, or GS register, but any
attempt to access a segment through one of these registers when it is loaded with a null segment
selector results in a #GP exception being generated. Loading unused data-segment registers with
a null segment selector is a useful method of detecting accesses to unused segment registers
and/or preventing unwanted accesses to data segments.

4-7
PROTECTION

4.5. PRIVILEGE LEVELS

The processor’s segment-protection mechanism recognizes 4 privilege levels, numbered from 0
to 3. The greater numbers mean lesser privileges. Figure 4-2 shows how these levels of privilege
can be interpreted as rings of protection. The center (reserved for the most privileged code, data,
and stacks) is used for the segments containing the critical software, usually the kernel of an
operating system. Outer rings are used for less critical software. (Systems that use only 2 of the
4 possible privilege levels should use levels 0 and 3.)

Protection Rings

Operating
System
Kernel Level 0

Operating System
Services Level 1

Level 2
Applications Level 3

Figure 4-2. Protection Rings

The processor uses privilege levels to prevent a program or task operating at a lesser privilege
level from accessing a segment with a greater privilege, except under controlled situations.
When the processor detects a privilege level violation, it generates a general-protection excep-
tion (#GP).
To carry out privilege-level checks between code segments and data segments, the processor
recognizes the following three types of privilege levels:
• Current privilege level (CPL). The CPL is the privilege level of the currently executing
program or task. It is stored in bits 0 and 1 of the CS and SS segment registers. Normally,
the CPL is equal to the privilege level of the code segment from which instructions are
being fetched. The processor changes the CPL when program control is transferred to a
code segment with a different privilege level. The CPL is treated slightly differently when
accessing conforming code segments. Conforming code segments can be accessed from
any privilege level that is equal to or numerically greater (less privileged) than the DPL of
the conforming code segment. Also, the CPL is not changed when the processor accesses a
conforming code segment that has a different privilege level than the CPL.
• Descriptor privilege level (DPL). The DPL is the privilege level of a segment or gate. It is
stored in the DPL field of the segment or gate descriptor for the segment or gate. When the
currently executing code segment attempts to access a segment or gate, the DPL of the

4-8
 PROTECTION

segment or gate is compared to the CPL and RPL of the segment or gate selector (as
described later in this section). The DPL is interpreted differently, depending on the type of
segment or gate being accessed:
— Data segment. The DPL indicates the numerically highest privilege level that a
program or task can have to be allowed to access the segment. For example, if the DPL
of a data segment is 1, only programs running at a CPL of 0 or 1 can access the
segment.
— Nonconforming code segment (without using a call gate). The DPL indicates the
privilege level that a program or task must be at to access the segment. For example, if
the DPL of a nonconforming code segment is 0, only programs running at a CPL of 0
can access the segment.
— Call gate. The DPL indicates the numerically highest privilege level that the currently
executing program or task can be at and still be able to access the call gate. (This is the
same access rule as for a data segment.)
— Conforming code segment and nonconforming code segment accessed through a
call gate. The DPL indicates the numerically lowest privilege level that a program or
task can have to be allowed to access the segment. For example, if the DPL of a
conforming code segment is 2, programs running at a CPL of 0 or 1 cannot access the
segment.
— TSS. The DPL indicates the numerically highest privilege level that the currently
executing program or task can be at and still be able to access the TSS. (This is the
same access rule as for a data segment.)
• Requested privilege level (RPL). The RPL is an override privilege level that is assigned
to segment selectors. It is stored in bits 0 and 1 of the segment selector. The processor
checks the RPL along with the CPL to determine if access to a segment is allowed. Even if
the program or task requesting access to a segment has sufficient privilege to access the
segment, access is denied if the RPL is not of sufficient privilege level. That is, if the RPL
of a segment selector is numerically greater than the CPL, the RPL overrides the CPL, and
vice versa. The RPL can be used to insure that privileged code does not access a segment
on behalf of an application program unless the program itself has access privileges for that
segment. Refer to Section 4.10.4., “Checking Caller Access Privileges (ARPL
Instruction)” for a detailed description of the purpose and typical use of the RPL.
Privilege levels are checked when the segment selector of a segment descriptor is loaded into a
segment register. The checks used for data access differ from those used for transfers of program
control among code segments; therefore, the two kinds of accesses are considered separately in
the following sections.

4.6. PRIVILEGE LEVEL CHECKING WHEN ACCESSING

DATA SEGMENTS
To access operands in a data segment, the segment selector for the data segment must be loaded
into the data-segment registers (DS, ES, FS, or GS) or into the stack-segment register (SS).

4-9
PROTECTION

(Segment registers can be loaded with the MOV, POP, LDS, LES, LFS, LGS, and LSS instruc-
tions.) Before the processor loads a segment selector into a segment register, it performs a priv-
ilege check (refer to Figure 4-3) by comparing the privilege levels of the currently running
program or task (the CPL), the RPL of the segment selector, and the DPL of the segment’s
segment descriptor. The processor loads the segment selector into the segment register if the
DPL is numerically greater than or equal to both the CPL and the RPL. Otherwise, a general-
protection fault is generated and the segment register is not loaded.

CS Register
CPL

Segment Selector
For Data Segment
RPL

Data-Segment Descriptor Privilege

Check
DPL

Figure 4-3. Privilege Check for Data Access

Figure 4-4 shows four procedures (located in codes segments A, B, C, and D), each running at
different privilege levels and each attempting to access the same data segment.
• The procedure in code segment A is able to access data segment E using segment selector
E1, because the CPL of code segment A and the RPL of segment selector E1 are equal to
the DPL of data segment E.
• The procedure in code segment B is able to access data segment E using segment selector
E2, because the CPL of code segment A and the RPL of segment selector E2 are both
numerically lower than (more privileged) than the DPL of data segment E. A code segment
B procedure can also access data segment E using segment selector E1.
• The procedure in code segment C is not able to access data segment E using segment
selector E3 (dotted line), because the CPL of code segment C and the RPL of segment
selector E3 are both numerically greater than (less privileged) than the DPL of data
segment E. Even if a code segment C procedure were to use segment selector E1 or E2,
such that the RPL would be acceptable, it still could not access data segment E because its
CPL is not privileged enough.
• The procedure in code segment D should be able to access data segment E because code
segment D’s CPL is numerically less than the DPL of data segment E. However, the RPL
of segment selector E3 (which the code segment D procedure is using to access data
segment E) is numerically greater than the DPL of data segment E, so access is not

4-10
 PROTECTION

allowed. If the code segment D procedure were to use segment selector E1 or E2 to access
the data segment, access would be allowed.

Code
Segment C Segment Sel. E3
CPL=3 RPL=3
3 Lowest Privilege

Code Data
Segment A Segment Sel. E1
RPL=2 Segment E
CPL=2
2 DPL=2

Code
Segment B Segment Sel. E2
RPL=1
CPL=1
1
Code
Segment D
CPL=0
0 Highest Privilege

Figure 4-4. Examples of Accessing Data Segments From Various Privilege Levels

As demonstrated in the previous examples, the addressable domain of a program or task varies
as its CPL changes. When the CPL is 0, data segments at all privilege levels are accessible; when
the CPL is 1, only data segments at privilege levels 1 through 3 are accessible; when the CPL is
3, only data segments at privilege level 3 are accessible.
The RPL of a segment selector can always override the addressable domain of a program or task.
When properly used, RPLs can prevent problems caused by accidental (or intensional) use of
segment selectors for privileged data segments by less privileged programs or procedures.
It is important to note that the RPL of a segment selector for a data segment is under software
control. For example, an application program running at a CPL of 3 can set the RPL for a data-
segment selector to 0. With the RPL set to 0, only the CPL checks, not the RPL checks, will
provide protection against deliberate, direct attempts to violate privilege-level security for the
data segment. To prevent these types of privilege-level-check violations, a program or procedure
can check access privileges whenever it receives a data-segment selector from another proce-
dure (refer to Section 4.10.4., “Checking Caller Access Privileges (ARPL Instruction)”).

4-11
PROTECTION

4.6.1. Accessing Data in Code Segments

In some instances it may be desirable to access data structures that are contained in a code
segment. The following methods of accessing data in code segments are possible:
• Load a data-segment register with a segment selector for a nonconforming, readable, code
segment.
• Load a data-segment register with a segment selector for a conforming, readable, code
segment.
• Use a code-segment override prefix (CS) to read a readable, code segment whose selector
is already loaded in the CS register.
The same rules for accessing data segments apply to method 1. Method 2 is always valid because
the privilege level of a conforming code segment is effectively the same as the CPL, regardless
of its DPL. Method 3 is always valid because the DPL of the code segment selected by the CS
register is the same as the CPL.

4.7. PRIVILEGE LEVEL CHECKING WHEN LOADING THE SS

REGISTER
Privilege level checking also occurs when the SS register is loaded with the segment selector for
a stack segment. Here all privilege levels related to the stack segment must match the CPL; that
is, the CPL, the RPL of the stack-segment selector, and the DPL of the stack-segment descriptor
must be the same. If the RPL and DPL are not equal to the CPL, a general-protection exception
(#GP) is generated.

4.8. PRIVILEGE LEVEL CHECKING WHEN TRANSFERRING

PROGRAM CONTROL BETWEEN CODE SEGMENTS
To transfer program control from one code segment to another, the segment selector for the
destination code segment must be loaded into the code-segment register (CS). As part of this
loading process, the processor examines the segment descriptor for the destination code segment
and performs various limit, type, and privilege checks. If these checks are successful, the CS
register is loaded, program control is transferred to the new code segment, and program execu-
tion begins at the instruction pointed to by the EIP register.
Program control transfers are carried out with the JMP, CALL, RET, INT n, and IRET instruc-
tions, as well as by the exception and interrupt mechanisms. Exceptions, interrupts, and the
IRET instruction are special cases discussed in Chapter 5, Interrupt and Exception Handling.
This chapter discusses only the JMP, CALL, and RET instructions.
A JMP or CALL instruction can reference another code segment in any of four ways:
• The target operand contains the segment selector for the target code segment.
• The target operand points to a call-gate descriptor, which contains the segment selector for
the target code segment.

4-12
 PROTECTION

• The target operand points to a TSS, which contains the segment selector for the target code
segment.
• The target operand points to a task gate, which points to a TSS, which in turn contains the
segment selector for the target code segment.
The following sections describe first two types of references. Refer to Section 6.3., “Task
Switching” in Chapter 6, Task Management for information on transferring program control
through a task gate and/or TSS.

4.8.1. Direct Calls or Jumps to Code Segments

The near forms of the JMP, CALL, and RET instructions transfer program control within the
current code segment, so privilege-level checks are not performed. The far forms of the JMP,
CALL, and RET instructions transfer control to other code segments, so the processor does
perform privilege-level checks.
When transferring program control to another code segment without going through a call gate,
the processor examines four kinds of privilege level and type information (refer to Figure 4-5):
• The CPL. (Here, the CPL is the privilege level of the calling code segment; that is, the code
segment that contains the procedure that is making the call or jump.)

CS Register

CPL

Segment Selector
For Code Segment

RPL

Destination Code
Segment Descriptor Privilege
Check
DPL C

Figure 4-5. Privilege Check for Control Transfer Without Using a Gate

• The DPL of the segment descriptor for the destination code segment that contains the
called procedure.
• The RPL of the segment selector of the destination code segment.
• The conforming (C) flag in the segment descriptor for the destination code segment, which
determines whether the segment is a conforming (C flag is set) or nonconforming (C flag is
clear) code segment. (Refer to Section 3.4.3.1., “Code- and Data-Segment Descriptor

4-13
PROTECTION

Types” in Chapter 3, Protected-Mode Memory Management for more information about

this flag.)
The rules that the processor uses to check the CPL, RPL, and DPL depends on the setting of the
C flag, as described in the following sections.

4.8.1.1. ACCESSING NONCONFORMING CODE SEGMENTS

When accessing nonconforming code segments, the CPL of the calling procedure must be equal
to the DPL of the destination code segment; otherwise, the processor generates a general-protec-
tion exception (#GP).
For example, in Figure 4-6, code segment C is a nonconforming code segment. Therefore, a
procedure in code segment A can call a procedure in code segment C (using segment selector
C1), because they are at the same privilege level (the CPL of code segment A is equal to the DPL
of code segment C). However, a procedure in code segment B cannot call a procedure in code
segment C (using segment selector C2 or C1), because the two code segments are at different
privilege levels.

Segment Sel. D2
Code RPL=3
Segment B
CPL=3 Segment Sel. C2
RPL=3
3 Lowest Privilege

Segment Sel. C1 Code

Code RPL=2 Segment C
Segment A
Segment Sel. D1 DPL=2
CPL=2 RPL=2 Nonconforming
2 Code Segment

Code
Segment D
DPL=3
Conforming
1 Code Segment

0 Highest Privilege

Figure 4-6. Examples of Accessing Conforming and Nonconforming Code Segments

From Various Privilege Levels

4-14
 PROTECTION

The RPL of the segment selector that points to a nonconforming code segment has a limited
effect on the privilege check. The RPL must be numerically less than or equal to the CPL of the
calling procedure for a successful control transfer to occur. So, in the example in Figure 4-6, the
RPLs of segment selectors C1 and C2 could legally be set to 0, 1, or 2, but not to 3.
When the segment selector of a nonconforming code segment is loaded into the CS register, the
privilege level field is not changed; that is, it remains at the CPL (which is the privilege level of
the calling procedure). This is true, even if the RPL of the segment selector is different from the
CPL.

4.8.1.2. ACCESSING CONFORMING CODE SEGMENTS

When accessing conforming code segments, the CPL of the calling procedure may be numeri-
cally equal to or greater than (less privileged) the DPL of the destination code segment; the
processor generates a general-protection exception (#GP) only if the CPL is less than the DPL.
(The segment selector RPL for the destination code segment is not checked if the segment is a
conforming code segment.)
In the example in Figure 4-6, code segment D is a conforming code segment. Therefore, calling
procedures in both code segment A and B can access code segment D (using either segment
selector D1 or D2, respectively), because they both have CPLs that are greater than or equal to
the DPL of the conforming code segment. For conforming code segments, the DPL repre-
sents the numerically lowest privilege level that a calling procedure may be at to success-
fully make a call to the code segment.
(Note that segments selectors D1 and D2 are identical except for their respective RPLs. But
since RPLs are not checked when accessing conforming code segments, the two segment selec-
tors are essentially interchangeable.)
When program control is transferred to a conforming code segment, the CPL does not change,
even if the DPL of the destination code segment is less than the CPL. This situation is the only
one where the CPL may be different from the DPL of the current code segment. Also, since the
CPL does not change, no stack switch occurs.
Conforming segments are used for code modules such as math libraries and exception handlers,
which support applications but do not require access to protected system facilities. These
modules are part of the operating system or executive software, but they can be executed at
numerically higher privilege levels (less privileged levels). Keeping the CPL at the level of a
calling code segment when switching to a conforming code segment prevents an application
program from accessing nonconforming code segments while at the privilege level (DPL) of a
conforming code segment and thus prevents it from accessing more privileged data.
Most code segments are nonconforming. For these segments, program control can be transferred
only to code segments at the same level of privilege, unless the transfer is carried out through a
call gate, as described in the following sections.

4-15
PROTECTION

4.8.2. Gate Descriptors

To provide controlled access to code segments with different privilege levels, the processor
provides special set of descriptors called gate descriptors. There are four kinds of gate
descriptors:
• Call gates
• Trap gates
• Interrupt gates
• Task gates
Task gates are used for task switching and are discussed in Chapter 6, Task Management. Trap
and interrupt gates are special kinds of call gates used for calling exception and interrupt
handlers. The are described in Chapter 5, Interrupt and Exception Handling. This chapter is
concerned only with call gates.

4.8.3. Call Gates

Call gates facilitate controlled transfers of program control between different privilege levels.
They are typically used only in operating systems or executives that use the privilege-level
protection mechanism. Call gates are also useful for transferring program control between 16-bit
and 32-bit code segments, as described in Section 17.4., “Transferring Control Among Mixed-
Size Code Segments” in Chapter 17, Mixing 16-Bit and 32-Bit Code.
Figure 4-7 shows the format of a call-gate descriptor. A call-gate descriptor may reside in the
GDT or in an LDT, but not in the interrupt descriptor table (IDT). It performs six functions:
• It specifies the code segment to be accessed.
• It defines an entry point for a procedure in the specified code segment.
• It specifies the privilege level required for a caller trying to access the procedure.
• If a stack switch occurs, it specifies the number of optional parameters to be copied
between stacks.
• It defines the size of values to be pushed onto the target stack: 16-bit gates force 16-bit
pushes and 32-bit gates force 32-bit pushes.
• It specifies whether the call-gate descriptor is valid.

4-16
 PROTECTION

31 16 15 14 13 12 11 8 7 6 5 4 0

D Type Param.
Offset in Segment 31:16 P P 0 0 0
Count 4
L 0 1 1 0 0

31 16 15 0

Segment Selector Offset in Segment 15:00 0

DPL Descriptor Privilege Level

P Gate Valid

Figure 4-7. Call-Gate Descriptor

The segment selector field in a call gate specifies the code segment to be accessed. The offset
field specifies the entry point in the code segment. This entry point is generally to the first
instruction of a specific procedure. The DPL field indicates the privilege level of the call gate,
which in turn is the privilege level required to access the selected procedure through the gate.
The P flag indicates whether the call-gate descriptor is valid. (The presence of the code segment
to which the gate points is indicated by the P flag in the code segment’s descriptor.) The param-
eter count field indicates the number of parameters to copy from the calling procedures stack to
the new stack if a stack switch occurs (refer to Section 4.8.5., “Stack Switching”). The parameter
count specifies the number of words for 16-bit call gates and doublewords for 32-bit call gates.
Note that the P flag in a gate descriptor is normally always set to 1. If it is set to 0, a not present
(#NP) exception is generated when a program attempts to access the descriptor. The operating
system can use the P flag for special purposes. For example, it could be used to track the number
of times the gate is used. Here, the P flag is initially set to 0 causing a trap to the not-present
exception handler. The exception handler then increments a counter and sets the P flag to 1, so
that on returning from the handler, the gate descriptor will be valid.

4.8.4. Accessing a Code Segment Through a Call Gate

To access a call gate, a far pointer to the gate is provided as a target operand in a CALL or JMP
instruction. The segment selector from this pointer identifies the call gate (refer to Figure 4-8);
the offset from the pointer is required, but not used or checked by the processor. (The offset can
be set to any value.)
When the processor has accessed the call gate, it uses the segment selector from the call gate to
locate the segment descriptor for the destination code segment. (This segment descriptor can be
in the GDT or the LDT.) It then combines the base address from the code-segment descriptor
with the offset from the call gate to form the linear address of the procedure entry point in the
code segment.
As shown in Figure 4-9, four different privilege levels are used to check the validity of a
program control transfer through a call gate:

4-17
PROTECTION

• The CPL (current privilege level).

• The RPL (requestor's privilege level) of the call gate’s selector.
• The DPL (descriptor privilege level) of the call gate descriptor.
• The DPL of the segment descriptor of the destination code segment.
The C flag (conforming) in the segment descriptor for the destination code segment is also
checked.

Far Pointer to Call Gate

Segment Selector Offset
Required but not used by processor

Descriptor Table

Offset Call-Gate
Segment Selector Offset Descriptor

Base Base Code-Segment

+ Base Descriptor

Procedure
Entry Point

Figure 4-8. Call-Gate Mechanism

4-18
 PROTECTION

CS Register
CPL

Call-Gate Selector
RPL

Call Gate (Descriptor) Privilege

Check
DPL

Destination Code-
Segment Descriptor
DPL

Figure 4-9. Privilege Check for Control Transfer with Call Gate

The privilege checking rules are different depending on whether the control transfer was initi-
ated with a CALL or a JMP instruction, as shown in Table 4-1.

Table 4-1. Privilege Check Rules for Call Gates

Instruction Privilege Check Rules
CALL CPL ≤ call gate DPL; RPL ≤ call gate DPL
Destination conforming code segment DPL ≤ CPL
Destination nonconforming code segment DPL ≤ CPL
JMP CPL ≤ call gate DPL; RPL ≤ call gate DPL
Destination conforming code segment DPL ≤ CPL
Destination nonconforming code segment DPL = CPL

The DPL field of the call-gate descriptor specifies the numerically highest privilege level from
which a calling procedure can access the call gate; that is, to access a call gate, the CPL of a
calling procedure must be equal to or less than the DPL of the call gate. For example, in Figure
4-12, call gate A has a DPL of 3. So calling procedures at all CPLs (0 through 3) can access this
call gate, which includes calling procedures in code segments A, B, and C. Call gate B has a
DPL of 2, so only calling procedures at a CPL or 0, 1, or 2 can access call gate B, which includes
calling procedures in code segments B and C. The dotted line shows that a calling procedure in
code segment A cannot access call gate B.

4-19
PROTECTION

The RPL of the segment selector to a call gate must satisfy the same test as the CPL of the calling
procedure; that is, the RPL must be less than or equal to the DPL of the call gate. In the example
in Figure 4-12, a calling procedure in code segment C can access call gate B using gate selector
B2 or B1, but it could not use gate selector B3 to access call gate B.
If the privilege checks between the calling procedure and call gate are successful, the processor
then checks the DPL of the code-segment descriptor against the CPL of the calling procedure.
Here, the privilege check rules vary between CALL and JMP instructions. Only CALL instruc-
tions can use call gates to transfer program control to more privileged (numerically lower priv-
ilege level) nonconforming code segments; that is, to nonconforming code segments with a DPL
less than the CPL. A JMP instruction can use a call gate only to transfer program control to a
nonconforming code segment with a DPL equal to the CPL. CALL and JMP instruction can both
transfer program control to a more privileged conforming code segment; that is, to a conforming
code segment with a DPL less than or equal to the CPL.
If a call is made to a more privileged (numerically lower privilege level) nonconforming desti-
nation code segment, the CPL is lowered to the DPL of the destination code segment and a stack
switch occurs (refer to Section 4.8.5., “Stack Switching”). If a call or jump is made to a more
privileged conforming destination code segment, the CPL is not changed and no stack switch
occurs.

Code Gate Selector A Call

Segment A RPL=3 Gate A
CPL=3 Gate Selector B3 DPL=3
RPL=3
3 Lowest Privilege

Code Call
Segment B Gate Selector B1
RPL=2 Gate B
CPL=2 DPL=2

2
Code
Segment C Gate Selector B2
RPL=1
CPL=1
No Stack Stack Switch
1 Switch Occurs Occurs

Code Code
Segment D Segment E
DPL=0 DPL=0
Conforming Nonconforming
0 Highest Privilege Code Segment Code Segment

Figure 4-10. Example of Accessing Call Gates At Various Privilege Levels

4-20
 PROTECTION

Call gates allow a single code segment to have procedures that can be accessed at different priv-
ilege levels. For example, an operating system located in a code segment may have some
services which are intended to be used by both the operating system and application software
(such as procedures for handling character I/O). Call gates for these procedures can be set up
that allow access at all privilege levels (0 through 3). More privileged call gates (with DPLs of
0 or 1) can then be set up for other operating system services that are intended to be used only
by the operating system (such as procedures that initialize device drivers).

4.8.5. Stack Switching

Whenever a call gate is used to transfer program control to a more privileged nonconforming
code segment (that is, when the DPL of the nonconforming destination code segment is less than
the CPL), the processor automatically switches to the stack for the destination code segment’s
privilege level. This stack switching is carried out to prevent more privileged procedures from
crashing due to insufficient stack space. It also prevents less privileged procedures from inter-
fering (by accident or intent) with more privileged procedures through a shared stack.
Each task must define up to 4 stacks: one for applications code (running at privilege level 3) and
one for each of the privilege levels 2, 1, and 0 that are used. (If only two privilege levels are used
[3 and 0], then only two stacks must be defined.) Each of these stacks is located in a separate
segment and is identified with a segment selector and an offset into the stack segment (a stack
pointer).
The segment selector and stack pointer for the privilege level 3 stack is located in the SS and
ESP registers, respectively, when privilege-level-3 code is being executed and is automatically
stored on the called procedure’s stack when a stack switch occurs.
Pointers to the privilege level 0, 1, and 2 stacks are stored in the TSS for the currently running
task (refer to Figure 6-2 in Chapter 6, Task Management). Each of these pointers consists of a
segment selector and a stack pointer (loaded into the ESP register). These initial pointers are
strictly read-only values. The processor does not change them while the task is running. They
are used only to create new stacks when calls are made to more privileged levels (numerically
lower privilege levels). These stacks are disposed of when a return is made from the called
procedure. The next time the procedure is called, a new stack is created using the initial stack
pointer. (The TSS does not specify a stack for privilege level 3 because the processor does not
allow a transfer of program control from a procedure running at a CPL of 0, 1, or 2 to a procedure
running at a CPL of 3, except on a return.)
The operating system is responsible for creating stacks and stack-segment descriptors for all the
privilege levels to be used and for loading initial pointers for these stacks into the TSS. Each
stack must be read/write accessible (as specified in the type field of its segment descriptor) and
must contain enough space (as specified in the limit field) to hold the following items:
• The contents of the SS, ESP, CS, and EIP registers for the calling procedure.
• The parameters and temporary variables required by the called procedure.
• The EFLAGS register and error code, when implicit calls are made to an exception or
interrupt handler.

4-21
PROTECTION

The stack will need to require enough space to contain many frames of these items, because
procedures often call other procedures, and an operating system may support nesting of multiple
interrupts. Each stack should be large enough to allow for the worst case nesting scenario at its
privilege level.
(If the operating system does not use the processor’s multitasking mechanism, it still must create
at least one TSS for this stack-related purpose.)
When a procedure call through a call gate results in a change in privilege level, the processor
performs the following steps to switch stacks and begin execution of the called procedure at a
new privilege level:
1. Uses the DPL of the destination code segment (the new CPL) to select a pointer to the new
stack (segment selector and stack pointer) from the TSS.
2. Reads the segment selector and stack pointer for the stack to be switched to from the
current TSS. Any limit violations detected while reading the stack-segment selector, stack
pointer, or stack-segment descriptor cause an invalid TSS (#TS) exception to be generated.
3. Checks the stack-segment descriptor for the proper privileges and type and generates an
invalid TSS (#TS) exception if violations are detected.
4. Temporarily saves the current values of the SS and ESP registers.
5. Loads the segment selector and stack pointer for the new stack in the SS and ESP registers.
6. Pushes the temporarily saved values for the SS and ESP registers (for the calling
procedure) onto the new stack (refer to Figure 4-11).
7. Copies the number of parameter specified in the parameter count field of the call gate from
the calling procedure’s stack to the new stack. If the count is 0, no parameters are copied.
8. Pushes the return instruction pointer (the current contents of the CS and EIP registers) onto
the new stack.
9. Loads the segment selector for the new code segment and the new instruction pointer from
the call gate into the CS and EIP registers, respectively, and begins execution of the called
procedure.
Refer to the description of the CALL instruction in Chapter 3, Instruction Set Reference, in the
Intel Architecture Software Developer’s Manual, Volume 2, for a detailed description of the priv-
ilege level checks and other protection checks that the processor performs on a far call through
a call gate.

4-22
 PROTECTION

Calling Procedure’s Stack Called Procedure’s Stack

Calling SS

Parameter 1 Calling ESP

Parameter 2 Parameter 1

Parameter 3 ESP Parameter 2

Parameter 3

Calling CS

Calling EIP ESP

Figure 4-11. Stack Switching During an Interprivilege-Level Call

The parameter count field in a call gate specifies the number of data items (up to 31) that the
processor should copy from the calling procedure’s stack to the stack of the called procedure. If
more than 31 data items need to be passed to the called procedure, one of the parameters can be
a pointer to a data structure, or the saved contents of the SS and ESP registers may be used to
access parameters in the old stack space. The size of the data items passed to the called proce-
dure depends on the call gate size, as described in Section 4.8.3., “Call Gates”

4.8.6. Returning from a Called Procedure

The RET instruction can be used to perform a near return, a far return at the same privilege level,
and a far return to a different privilege level. This instruction is intended to execute returns from
procedures that were called with a CALL instruction. It does not support returns from a JMP
instruction, because the JMP instruction does not save a return instruction pointer on the stack.
A near return only transfers program control within the current code segment; therefore, the
processor performs only a limit check. When the processor pops the return instruction pointer
from the stack into the EIP register, it checks that the pointer does not exceed the limit of the
current code segment.
On a far return at the same privilege level, the processor pops both a segment selector for the
code segment being returned to and a return instruction pointer from the stack. Under normal
conditions, these pointers should be valid, because they were pushed on the stack by the CALL
instruction. However, the processor performs privilege checks to detect situations where the
current procedure might have altered the pointer or failed to maintain the stack properly.

4-23
PROTECTION

A far return that requires a privilege-level change is only allowed when returning to a less priv-
ileged level (that is, the DPL of the return code segment is numerically greater than the CPL).
The processor uses the RPL field from the CS register value saved for the calling procedure
(refer to Figure 4-11) to determine if a return to a numerically higher privilege level is required.
If the RPL is numerically greater (less privileged) than the CPL, a return across privilege levels
occurs.
The processor performs the following steps when performing a far return to a calling procedure
(refer to Figures 4-2 and 4-4 in the Intel Architecture Software Developer’s Manual, Volume 1,
for an illustration of the stack contents prior to and after a return):
1. Checks the RPL field of the saved CS register value to determine if a privilege level
change is required on the return.
2. Loads the CS and EIP registers with the values on the called procedure’s stack. (Type and
privilege level checks are performed on the code-segment descriptor and RPL of the code-
segment selector.)
3. (If the RET instruction includes a parameter count operand and the return requires a
privilege level change.) Adds the parameter count (in bytes obtained from the RET
instruction) to the current ESP register value (after popping the CS and EIP values), to step
past the parameters on the called procedure’s stack. The resulting value in the ESP register
points to the saved SS and ESP values for the calling procedure’s stack. (Note that the byte
count in the RET instruction must be chosen to match the parameter count in the call gate
that the calling procedure referenced when it made the original call multiplied by the size
of the parameters.)
4. (If the return requires a privilege level change.) Loads the SS and ESP registers with the
saved SS and ESP values and switches back to the calling procedure’s stack. The SS and
ESP values for the called procedure’s stack are discarded. Any limit violations detected
while loading the stack-segment selector or stack pointer cause a general-protection
exception (#GP) to be generated. The new stack-segment descriptor is also checked for
type and privilege violations.
5. (If the RET instruction includes a parameter count operand.) Adds the parameter count (in
bytes obtained from the RET instruction) to the current ESP register value, to step past the
parameters on the calling procedure’s stack. The resulting ESP value is not checked against
the limit of the stack segment. If the ESP value is beyond the limit, that fact is not
recognized until the next stack operation.
6. (If the return requires a privilege level change.) Checks the contents of the DS, ES, FS, and
GS segment registers. If any of these registers refer to segments whose DPL is less than the
new CPL (excluding conforming code segments), the segment register is loaded with a null
segment selector.
Refer to the description of the RET instruction in Chapter 3, Instruction Set Reference, of the
Intel Architecture Software Developer’s Manual, Volume 2, for a detailed description of the priv-
ilege level checks and other protection checks that the processor performs on a far return.

4-24
 PROTECTION

4.9. PRIVILEGED INSTRUCTIONS

Some of the system instructions (called “privileged instructions” are protected from use by
application programs. The privileged instructions control system functions (such as the loading
of system registers). They can be executed only when the CPL is 0 (most privileged). If one of
these instructions is executed when the CPL is not 0, a general-protection exception (#GP) is
generated. The following system instructions are privileged instructions:
• LGDT—Load GDT register.
• LLDT—Load LDT register.
• LTR—Load task register.
• LIDT—Load IDT register.
• MOV (control registers)—Load and store control registers.
• LMSW—Load machine status word.
• CLTS—Clear task-switched flag in register CR0.
• MOV (debug registers)—Load and store debug registers.
• INVD—Invalidate cache, without writeback.
• WBINVD—Invalidate cache, with writeback.
• INVLPG—Invalidate TLB entry.
• HLT—Halt processor.
• RDMSR—Read Model-Specific Registers.
• WRMSR—Write Model-Specific Registers.
• RDPMC—Read Performance-Monitoring Counter.
• RDTSC—Read Time-Stamp Counter.
Some of the privileged instructions are available only in the more recent families of Intel Archi-
tecture processors (refer to Section 18.7., “New Instructions In the Pentium® and Later Intel
Architecture Processors”, in Chapter 18, Intel Architecture Compatibility).
The PCE and TSD flags in register CR4 (bits 4 and 2, respectively) enable the RDPMC and
RDTSC instructions, respectively, to be executed at any CPL.

4.10. POINTER VALIDATION

When operating in protected mode, the processor validates all pointers to enforce protection
between segments and maintain isolation between privilege levels. Pointer validation consists
of the following checks:
1. Checking access rights to determine if the segment type is compatible with its use.
2. Checking read/write rights

4-25
PROTECTION

3. Checking if the pointer offset exceeds the segment limit.

4. Checking if the supplier of the pointer is allowed to access the segment.
5. Checking the offset alignment.
The processor automatically performs first, second, and third checks during instruction execu-
tion. Software must explicitly request the fourth check by issuing an ARPL instruction. The fifth
check (offset alignment) is performed automatically at privilege level 3 if alignment checking is
turned on. Offset alignment does not affect isolation of privilege levels.

4.10.1. Checking Access Rights (LAR Instruction)

When the processor accesses a segment using a far pointer, it performs an access rights check
on the segment descriptor pointed to by the far pointer. This check is performed to determine if
type and privilege level (DPL) of the segment descriptor are compatible with the operation to be
performed. For example, when making a far call in protected mode, the segment-descriptor type
must be for a conforming or nonconforming code segment, a call gate, a task gate, or a TSS.
Then, if the call is to a nonconforming code segment, the DPL of the code segment must be equal
to the CPL, and the RPL of the code segment’s segment selector must be less than or equal to
the DPL. If type or privilege level are found to be incompatible, the appropriate exception is
generated.
To prevent type incompatibility exceptions from being generated, software can check the access
rights of a segment descriptor using the LAR (load access rights) instruction. The LAR instruc-
tion specifies the segment selector for the segment descriptor whose access rights are to be
checked and a destination register. The instruction then performs the following operations:
1. Check that the segment selector is not null.
2. Checks that the segment selector points to a segment descriptor that is within the descriptor
table limit (GDT or LDT).
3. Checks that the segment descriptor is a code, data, LDT, call gate, task gate, or TSS
segment-descriptor type.
4. If the segment is not a conforming code segment, checks if the segment descriptor is
visible at the CPL (that is, if the CPL and the RPL of the segment selector are less than or
equal to the DPL).
5. If the privilege level and type checks pass, loads the second doubleword of the segment
descriptor into the destination register (masked by the value 00FXFF00H, where X
indicates that the corresponding 4 bits are undefined) and sets the ZF flag in the EFLAGS
register. If the segment selector is not visible at the current privilege level or is an invalid
type for the LAR instruction, the instruction does not modify the destination register and
clears the ZF flag.
Once loaded in the destination register, software can preform additional checks on the access
rights information.

4-26
 PROTECTION

4.10.2. Checking Read/Write Rights (VERR and VERW

Instructions)
When the processor accesses any code or data segment it checks the read/write privileges
assigned to the segment to verify that the intended read or write operation is allowed. Software
can check read/write rights using the VERR (verify for reading) and VERW (verify for writing)
instructions. Both these instructions specify the segment selector for the segment being checked.
The instructions then perform the following operations:
1. Check that the segment selector is not null.
2. Checks that the segment selector points to a segment descriptor that is within the descriptor
table limit (GDT or LDT).
3. Checks that the segment descriptor is a code or data-segment descriptor type.
4. If the segment is not a conforming code segment, checks if the segment descriptor is
visible at the CPL (that is, if the CPL and the RPL of the segment selector are less than or
equal to the DPL).
5. Checks that the segment is readable (for the VERR instruction) or writable (for the
VERW) instruction.
The VERR instruction sets the ZF flag in the EFLAGS register if the segment is visible at the
CPL and readable; the VERW sets the ZF flag if the segment is visible and writable. (Code
segments are never writable.) The ZF flag is cleared if any of these checks fail.

4-27
PROTECTION

4.10.3. Checking That the Pointer Offset Is Within Limits (LSL

Instruction)
When the processor accesses any segment it performs a limit check to insure that the offset is
within the limit of the segment. Software can perform this limit check using the LSL (load
segment limit) instruction. Like the LAR instruction, the LSL instruction specifies the segment
selector for the segment descriptor whose limit is to be checked and a destination register. The
instruction then performs the following operations:
1. Check that the segment selector is not null.
2. Checks that the segment selector points to a segment descriptor that is within the descriptor
table limit (GDT or LDT).
3. Checks that the segment descriptor is a code, data, LDT, or TSS segment-descriptor type.
4. If the segment is not a conforming code segment, checks if the segment descriptor is
visible at the CPL (that is, if the CPL and the RPL of the segment selector less than or
equal to the DPL).
5. If the privilege level and type checks pass, loads the unscrambled limit (the limit scaled
according to the setting of the G flag in the segment descriptor) into the destination register
and sets the ZF flag in the EFLAGS register. If the segment selector is not visible at the
current privilege level or is an invalid type for the LSL instruction, the instruction does not
modify the destination register and clears the ZF flag.
Once loaded in the destination register, software can compare the segment limit with the offset
of a pointer.

4.10.4. Checking Caller Access Privileges (ARPL Instruction)

The requestor’s privilege level (RPL) field of a segment selector is intended to carry the privi-
lege level of a calling procedure (the calling procedure’s CPL) to a called procedure. The called
procedure then uses the RPL to determine if access to a segment is allowed. The RPL is said to
“weaken” the privilege level of the called procedure to that of the RPL.
Operating-system procedures typically use the RPL to prevent less privileged application
programs from accessing data located in more privileged segments. When an operating-system
procedure (the called procedure) receives a segment selector from an application program (the
calling procedure), it sets the segment selector’s RPL to the privilege level of the calling proce-
dure. Then, when the operating system uses the segment selector to access its associated
segment, the processor performs privilege checks using the calling procedure’s privilege level
(stored in the RPL) rather than the numerically lower privilege level (the CPL) of the operating-
system procedure. The RPL thus insures that the operating system does not access a segment on
behalf of an application program unless that program itself has access to the segment.
Figure 4-12 shows an example of how the processor uses the RPL field. In this example, an
application program (located in code segment A) possesses a segment selector (segment selector
D1) that points to a privileged data structure (that is, a data structure located in a data segment
D at privilege level 0). The application program cannot access data segment D, because it does

4-28
 PROTECTION

not have sufficient privilege, but the operating system (located in code segment C) can. So, in
an attempt to access data segment D, the application program executes a call to the operating
system and passes segment selector D1 to the operating system as a parameter on the stack.
Before passing the segment selector, the (well behaved) application program sets the RPL of the
segment selector to its current privilege level (which in this example is 3). If the operating
system attempts to access data segment D using segment selector D1, the processor compares
the CPL (which is now 0 following the call), the RPL of segment selector D1, and the DPL of
data segment D (which is 0). Since the RPL is greater than the DPL, access to data segment D
is denied. The processor’s protection mechanism thus protects data segment D from access by
the operating system, because application program’s privilege level (represented by the RPL of
segment selector B) is greater than the DPL of data segment D.

Passed as a
parameter on
the stack.
Application Program
Code Call
Segment A Gate Selector B Segment Sel. D1
RPL=3 Gate B RPL=3
CPL=3 DPL=3
3 Lowest Privilege

2
Access
not
allowed

1
Code Data
Operating Segment C Segment Sel. D2 Segment D
System RPL=0
DPL=0 DPL=0
Access
0 Highest Privilege allowed

Figure 4-12. Use of RPL to Weaken Privilege Level of Called Procedure

Now assume that instead of setting the RPL of the segment selector to 3, the application program
sets the RPL to 0 (segment selector D2). The operating system can now access data segment D,
because its CPL and the RPL of segment selector D2 are both equal to the DPL of data segment
D. Because the application program is able to change the RPL of a segment selector to any value,
it can potentially use a procedure operating at a numerically lower privilege level to access a

4-29
PROTECTION

protected data structure. This ability to lower the RPL of a segment selector breaches the
processor’s protection mechanism.
Because a called procedure cannot rely on the calling procedure to set the RPL correctly, oper-
ating-system procedures (executing at numerically lower privilege-levels) that receive segment
selectors from numerically higher privilege-level procedures need to test the RPL of the segment
selector to determine if it is at the appropriate level. The ARPL (adjust requested privilege level)
instruction is provided for this purpose. This instruction adjusts the RPL of one segment selector
to match that of another segment selector.
The example in Figure 4-12 demonstrates how the ARPL instruction is intended to be used.
When the operating-system receives segment selector D2 from the application program, it uses
the ARPL instruction to compare the RPL of the segment selector with the privilege level of the
application program (represented by the code-segment selector pushed onto the stack). If the
RPL is less than application program’s privilege level, the ARPL instruction changes the RPL
of the segment selector to match the privilege level of the application program (segment
selector D1). Using this instruction thus prevents a procedure running at a numerically higher
privilege level from accessing numerically lower privilege-level (more privileged) segments by
lowering the RPL of a segment selector.
Note that the privilege level of the application program can be determined by reading the RPL
field of the segment selector for the application-program’s code segment. This segment selector
is stored on the stack as part of the call to the operating system. The operating system can copy
the segment selector from the stack into a register for use as an operand for the ARPL
instruction.

4.10.5. Checking Alignment

When the CPL is 3, alignment of memory references can be checked by setting the AM flag in
the CR0 register and the AC flag in the EFLAGS register. Unaligned memory references
generate alignment exceptions (#AC). The processor does not generate alignment exceptions
when operating at privilege level 0, 1, or 2. Refer to Table 5-7 in Chapter 5, Interrupt and Excep-
tion Handling for a description of the alignment requirements when alignment checking is
enabled.

4.11. PAGE-LEVEL PROTECTION

Page-level protection can be used alone or applied to segments. When page-level protection is
used with the flat memory model, it allows supervisor code and data (the operating system or
executive) to be protected from user code and data (application programs). It also allows pages
containing code to be write protected. When the segment- and page-level protection are
combined, page-level read/write protection allows more protection granularity within segments.
With page-level protection (as with segment-level protection) each memory reference is
checked to verify that protection checks are satisfied. All checks are made before the memory
cycle is started, and any violation prevents the cycle from starting and results in a page-fault

4-30
 PROTECTION

exception being generated. Because checks are performed in parallel with address translation,
there is no performance penalty.
The processor performs two page-level protection checks:
• Restriction of addressable domain (supervisor and user modes).
• Page type (read only or read/write).
Violations of either of these checks results in a page-fault exception being generated. Refer to
Chapter 5, Interrupt and Exception Handling for an explanation of the page-fault exception
mechanism. This chapter describes the protection violations which lead to page-fault excep-
tions.

4.11.1. Page-Protection Flags

Protection information for pages is contained in two flags in a page-directory or page-table entry
(refer to Figure 3-14 in Chapter 3, Protected-Mode Memory Management): the read/write flag
(bit 1) and the user/supervisor flag (bit 2). The protection checks are applied to both first- and
second-level page tables (that is, page directories and page tables).

4.11.2. Restricting Addressable Domain

The page-level protection mechanism allows restricting access to pages based on two privilege
levels:
• Supervisor mode (U/S flag is 0)—(Most privileged) For the operating system or executive,
other system software (such as device drivers), and protected system data (such as page
tables).
• User mode (U/S flag is 1)—(Least privileged) For application code and data.
The segment privilege levels map to the page privilege levels as follows. If the processor is
currently operating at a CPL of 0, 1, or 2, it is in supervisor mode; if it is operating at a CPL of
3, it is in user mode. When the processor is in supervisor mode, it can access all pages; when in
user mode, it can access only user-level pages. (Note that the WP flag in control register CR0
modifies the supervisor permissions, as described in Section 4.11.3., “Page Type”)
Note that to use the page-level protection mechanism, code and data segments must be set up
for at least two segment-based privilege levels: level 0 for supervisor code and data segments
and level 3 for user code and data segments. (In this model, the stacks are placed in the data
segments.) To minimize the use of segments, a flat memory model can be used (refer to Section
3.2.1., “Basic Flat Model” in Section 3, “Protected-Mode Memory Management”). Here, the
user and supervisor code and data segments all begin at address zero in the linear address space
and overlay each other. With this arrangement, operating-system code (running at the supervisor
level) and application code (running at the user level) can execute as if there are no segments.
Protection between operating-system and application code and data is provided by the
processor’s page-level protection mechanism.

4-31
PROTECTION

4.11.3. Page Type

The page-level protection mechanism recognizes two page types:
• Read-only access (R/W flag is 0).
• Read/write access (R/W flag is 1).
When the processor is in supervisor mode and the WP flag in register CR0 is clear (its state
following reset initialization), all pages are both readable and writable (write-protection is
ignored). When the processor is in user mode, it can write only to user-mode pages that are
read/write accessible. User-mode pages which are read/write or read-only are readable; super-
visor-mode pages are neither readable nor writable from user mode. A page-fault exception is
generated on any attempt to violate the protection rules.
The P6 family, Pentium®, and Intel486™ processors allow user-mode pages to be write-
protected against supervisor-mode access. Setting the WP flag in register CR0 to 1 enables
supervisor-mode sensitivity to user-mode, write-protected pages. This supervisor write-protect
feature is useful for implementing a “copy-on-write” strategy used by some operating systems,
such as UNIX*, for task creation (also called forking or spawning). When a new task is created,
it is possible to copy the entire address space of the parent task. This gives the child task a
complete, duplicate set of the parent's segments and pages. An alternative copy-on-write
strategy saves memory space and time by mapping the child's segments and pages to the same
segments and pages used by the parent task. A private copy of a page gets created only when
one of the tasks writes to the page. By using the WP flag and marking the shared pages as read-
only, the supervisor can detect an attempt to write to a user-level page, and can copy the page at
that time.

4.11.4. Combining Protection of Both Levels of Page Tables

For any one page, the protection attributes of its page-directory entry (first-level page table) may
differ from those of its page-table entry (second-level page table). The processor checks the
protection for a page in both its page-directory and the page-table entries. Table 4-2 shows the
protection provided by the possible combinations of protection attributes when the WP flag is
clear.

4.11.5. Overrides to Page Protection

The following types of memory accesses are checked as if they are privilege-level 0 accesses,
regardless of the CPL at which the processor is currently operating:
• Access to segment descriptors in the GDT, LDT, or IDT.
• Access to an inner-privilege-level stack during an inter-privilege-level call or a call to in
exception or interrupt handler, when a change of privilege level occurs.

4-32
 PROTECTION

4.12. COMBINING PAGE AND SEGMENT PROTECTION

When paging is enabled, the processor evaluates segment protection first, then evaluates page
protection. If the processor detects a protection violation at either the segment level or the page
level, the memory access is not carried out and an exception is generated. If an exception is
generated by segmentation, no paging exception is generated.
Page-level protections cannot be used to override segment-level protection. For example, a code
segment is by definition not writable. If a code segment is paged, setting the R/W flag for the
pages to read-write does not make the pages writable. Attempts to write into the pages will be
blocked by segment-level protection checks.
Page-level protection can be used to enhance segment-level protection. For example, if a large
read-write data segment is paged, the page-protection mechanism can be used to write-protect
individual pages.

Table 4-2. Combined Page-Directory and Page-Table Protection

Page-Directory Entry Page-Table Entry Combined Effect
Privilege Access Type Privilege Access Type Privilege Access Type
User Read-Only User Read-Only User Read-Only
User Read-Only User Read-Write User Read-Only
User Read-Write User Read-Only User Read-Only
User Read-Write User Read-Write User Read/Write
User Read-Only Supervisor Read-Only Supervisor Read/Write*
User Read-Only Supervisor Read-Write Supervisor Read/Write*
User Read-Write Supervisor Read-Only Supervisor Read/Write*
User Read-Write Supervisor Read-Write Supervisor Read/Write
Supervisor Read-Only User Read-Only Supervisor Read/Write*
Supervisor Read-Only User Read-Write Supervisor Read/Write*
Supervisor Read-Write User Read-Only Supervisor Read/Write*
Supervisor Read-Write User Read-Write Supervisor Read/Write
Supervisor Read-Only Supervisor Read-Only Supervisor Read/Write*
Supervisor Read-Only Supervisor Read-Write Supervisor Read/Write*
Supervisor Read-Write Supervisor Read-Only Supervisor Read/Write*
Supervisor Read-Write Supervisor Read-Write Supervisor Read/Write

NOTE:
* If the WP flag of CR0 is set, the access type is determined by the R/W flags of the page-directory and
page-table entries.

4-33
PROTECTION

4-34
 5
Interrupt and
Exception Handling
 INTERRUPT AND EXCEPTION HANDLING

CHAPTER 5
INTERRUPT AND EXCEPTION HANDLING

This chapter describes the processor’s interrupt and exception-handling mechanism, when oper-
ating in protected mode. Most of the information provided here also applies to the interrupt and
exception mechanism used in real-address or virtual-8086 mode. Refer to Chapter 16, 8086
Emulation for a description of the differences in the interrupt and exception mechanism for real-
address and virtual-8086 mode.

5.1. INTERRUPT AND EXCEPTION OVERVIEW

Interrupts and exceptions are forced transfers of execution from the currently running program
or task to a special procedure or task called a handler. Interrupts typically occur at random times
during the execution of a program, in response to signals from hardware. They are used to handle
events external to the processor, such as requests to service peripheral devices. Software can also
generate interrupts by executing the INT n instruction. Exceptions occur when the processor
detects an error condition while executing an instruction, such as division by zero. The processor
detects a variety of error conditions including protection violations, page faults, and internal
machine faults. The machine-check architecture of the P6 family and Pentium® processors
also permits a machine-check exception to be generated when internal hardware errors and bus
errors are detected.
The processor’s interrupt and exception-handling mechanism allows interrupts and exceptions
to be handled transparently to application programs and the operating system or executive.
When an interrupt is received or an exception is detected, the currently running procedure or
task is automatically suspended while the processor executes an interrupt or exception handler.
When execution of the handler is complete, the processor resumes execution of the interrupted
procedure or task. The resumption of the interrupted procedure or task happens without loss of
program continuity, unless recovery from an exception was not possible or an interrupt caused
the currently running program to be terminated.
This chapter describes the processor’s interrupt and exception-handling mechanism, when oper-
ating in protected mode. A detailed description of the exceptions and the conditions that cause
them to be generated is given at the end of this chapter. Refer to Chapter 16, 8086 Emulation for
a description of the interrupt and exception mechanism for real-address and virtual-8086 mode.

5.1.1. Sources of Interrupts

The processor receives interrupts from two sources:
• External (hardware generated) interrupts.
• Software-generated interrupts.

5-1
INTERRUPT AND EXCEPTION HANDLING

5.1.1.1. EXTERNAL INTERRUPTS

External interrupts are received through pins on the processor or through the local APIC serial
bus. The primary interrupt pins on a P6 family or Pentium® processor are the LINT[1:0] pins,
which are connected to the local APIC (refer to Section 7.5., “Advanced Programmable Inter-
rupt Controller (APIC)” in Chapter 7, Multiple-Processor Management). When the local APIC
is disabled, these pins are configured as INTR and NMI pins, respectively. Asserting the INTR
pin signals the processor that an external interrupt has occurred, and the processor reads from
the system bus the interrupt vector number provided by an external interrupt controller, such as
an 8259A (refer to Section 5.2., “Exception and Interrupt Vectors”). Asserting the NMI pin
signals a nonmaskable interrupt (NMI), which is assigned to interrupt vector 2.
When the local APIC is enabled, the LINT[1:0] pins can be programmed through the APIC’s
vector table to be associated with any of the processor’s exception or interrupt vectors.
The processor’s local APIC can be connected to a system-based I/O APIC. Here, external inter-
rupts received at the I/O APIC’s pins can be directed to the local APIC through the APIC serial
bus (pins PICD[1:0]). The I/O APIC determines the vector number of the interrupt and sends
this number to the local APIC. When a system contains multiple processors, processors can also
send interrupts to one another by means of the APIC serial bus.
The LINT[1:0] pins are not available on the Intel486™ processor and the earlier Pentium®
processors that do not contain an on-chip local APIC. Instead these processors have dedicated
NMI and INTR pins. With these processors, external interrupts are typically generated by a
system-based interrupt controller (8259A), with the interrupts being signaled through the INTR
pin.
Note that several other pins on the processor cause a processor interrupt to occur; however, these
interrupts are not handled by the interrupt and exception mechanism described in this chapter.
These pins include the RESET#, FLUSH#, STPCLK#, SMI#, R/S#, and INIT# pins. Which of
these pins are included on a particular Intel Architecture processor is implementation dependent.
The functions of these pins are described in the data books for the individual processors. The
SMI# pin is also described in Chapter 12, System Management Mode (SMM).

5.1.1.2. MASKABLE HARDWARE INTERRUPTS

Any external interrupt that is delivered to the processor by means of the INTR pin or through
the local APIC is called a maskable hardware interrupt. The maskable hardware interrupts
that can be delivered through the INTR pin include all Intel Architecture defined interrupt
vectors from 0 through 255; those that can be delivered through the local APIC include interrupt
vectors 16 through 255.
All maskable hardware interrupts can be masked as a group. Use the single IF flag in the
EFLAGS register (refer to Section 5.6.1., “Masking Maskable Hardware Interrupts”) to mask
these maskable interrupts. Note that when interrupts 0 through 15 are delivered through the local
APIC, the APIC indicates the receipt of an illegal vector.

5-2
 INTERRUPT AND EXCEPTION HANDLING

5.1.1.3. SOFTWARE-GENERATED INTERRUPTS

The INT n instruction permits interrupts to be generated from within software by supplying the
interrupt vector number as an operand. For example, the INT 35 instruction forces an implicit
call to the interrupt handler for interrupt 35.
Any of the interrupt vectors from 0 to 255 can be used as a parameter in this instruction. If the
processor’s predefined NMI vector is used, however, the response of the processor will not be
the same as it would be from an NMI interrupt generated in the normal manner. If vector number
2 (the NMI vector) is used in this instruction, the NMI interrupt handler is called, but the
processor’s NMI-handling hardware is not activated.
Note that interrupts generated in software with the INT n instruction cannot be masked by the
IF flag in the EFLAGS register.

5.1.2. Sources of Exceptions

The processor receives exceptions from three sources:
• Processor-detected program-error exceptions.
• Software-generated exceptions.
• Machine-check exceptions.

5.1.2.1. PROGRAM-ERROR EXCEPTIONS

The processor generates one or more exceptions when it detects program errors during the
execution in an application program or the operating system or executive. The Intel Architecture
defines a vector number for each processor-detectable exception. The exceptions are further
classified as faults, traps, and aborts (refer to Section 5.3., “Exception Classifications”).

5.1.2.2. SOFTWARE-GENERATED EXCEPTIONS

The INTO, INT 3, and BOUND instructions permit exceptions to be generated in software.
These instructions allow checks for specific exception conditions to be performed at specific
points in the instruction stream. For example, the INT 3 instruction causes a breakpoint excep-
tion to be generated.
The INT n instruction can be used to emulate a specific exception in software, with one limita-
tion. If the n operand in the INT n instruction contains a vector for one of the Intel Architecture
exceptions, the processor will generate an interrupt to that vector, which will in turn invoke the
exception handler associated with that vector. Because this is actually an interrupt, however, the
processor does not push an error code onto the stack, even if a hardware-generated exception for
that vector normally produces one. For those exceptions that produce an error code, the excep-
tion handler will attempt to pop an error code from the stack while handling the exception. If the
INT n instruction was used to emulate the generation of an exception, the handler will pop off
and discard the EIP (in place of the missing error code), sending the return to the wrong location.

5-3
INTERRUPT AND EXCEPTION HANDLING

5.1.2.3. MACHINE-CHECK EXCEPTIONS

The P6 family and Pentium® processors provide both internal and external machine-check
mechanisms for checking the operation of the internal chip hardware and bus transactions.
These mechanisms constitute extended (implementation dependent) exception mechanisms.
When a machine-check error is detected, the processor signals a machine-check exception
(vector 18) and returns an error code. Refer to “Interrupt 18—Machine Check Exception
(#MC)” at the end of this chapter and Chapter 13, Machine-Check Architecture, for a detailed
description of the machine-check mechanism.

5.2. EXCEPTION AND INTERRUPT VECTORS

The processor associates an identification number, called a vector, with each exception and
interrupt. Table 5-1 shows the assignment of exception and interrupt vectors. This table also
gives the exception type for each vector, indicates whether an error code is saved on the stack
for an exception, and gives the source of the exception or interrupt.
The vectors in the range 0 through 31 are assigned to the exceptions and the NMI interrupt. Not
all of these vectors are currently used by the processor. Unassigned vectors in this range are
reserved for possible future uses. Do not use the reserved vectors.
The vectors in the range 32 to 255 are designated as user-defined interrupts. These interrupts are
not reserved by the Intel Architecture and are generally assigned to external I/O devices and to
permit them to signal the processor through one of the external hardware interrupt mechanisms
described in Section 5.1.1., “Sources of Interrupts”

5.3. EXCEPTION CLASSIFICATIONS

Exceptions are classified as faults, traps, or aborts depending on the way they are reported and
whether the instruction that caused the exception can be restarted with no loss of program or task
continuity.
Faults A fault is an exception that can generally be corrected and that, once corrected,
allows the program to be restarted with no loss of continuity. When a fault is
reported, the processor restores the machine state to the state prior to the begin-
ning of execution of the faulting instruction. The return address (saved contents
of the CS and EIP registers) for the fault handler points to the faulting instruc-
tion, rather than the instruction following the faulting instruction.
Note: There are a small subset of exceptions that are normally reported as
faults, but under architectural corner cases, they are not restartable and some
processor context will be lost. An example of these cases is the execution of the
POPAD instruction where the stack frame crosses over the the end of the stack
segment. The exception handler will see that the CS:EIP has been restored as
if the POPAD instruction had not executed however internal processor state
(general purpose registers) will have been modified. These corner cases are

5-4
 INTERRUPT AND EXCEPTION HANDLING

considered programming errors and an application causeing this class of

exceptions will likely be terminated by the operating system.
Traps A trap is an exception that is reported immediately following the execution of
the trapping instruction. Traps allow execution of a program or task to be
continued without loss of program continuity. The return address for the trap
handler points to the instruction to be executed after the trapping instruction.
Aborts An abort is an exception that does not always report the precise location of the
instruction causing the exception and does not allow restart of the program or
task that caused the exception. Aborts are used to report severe errors, such as
hardware errors and inconsistent or illegal values in system tables.

5-5
INTERRUPT AND EXCEPTION HANDLING

Table 5-1. Protected-Mode Exceptions and Interrupts

Vector Mne- Error
No. monic Description Type Code Source
0 #DE Divide Error Fault No DIV and IDIV instructions.
1 #DB Debug Fault/ No Any code or data reference or the
Trap INT 1 instruction.
2 — NMI Interrupt Interrupt No Nonmaskable external interrupt.
3 #BP Breakpoint Trap No INT 3 instruction.
4 #OF Overflow Trap No INTO instruction.
5 #BR BOUND Range Exceeded Fault No BOUND instruction.
6 #UD Invalid Opcode (Undefined Fault No UD2 instruction or reserved
Opcode) opcode.1
7 #NM Device Not Available (No Fault No Floating-point or WAIT/FWAIT
Math Coprocessor) instruction.
8 #DF Double Fault Abort Yes Any instruction that can generate
(Zero) an exception, an NMI, or an INTR.
9 Coprocessor Segment Fault No Floating-point instruction.2
Overrun (reserved)
10 #TS Invalid TSS Fault Yes Task switch or TSS access.
11 #NP Segment Not Present Fault Yes Loading segment registers or
accessing system segments.
12 #SS Stack-Segment Fault Fault Yes Stack operations and SS register
loads.
13 #GP General Protection Fault Yes Any memory reference and other
protection checks.
14 #PF Page Fault Fault Yes Any memory reference.
15 — (Intel reserved. Do not use.) No
16 #MF Floating-Point Error (Math Fault No Floating-point or WAIT/FWAIT
Fault) instruction.
17 #AC Alignment Check Fault Yes Any data reference in memory.3
(Zero)
18 #MC Machine Check Abort No Error codes (if any) and source
are model dependent.4
19 #XF Streaming SIMD Extensions Fault No SIMD floating-point instructions5
20-31 — Intel reserved. Do not use.
32- — User Defined (Nonreserved) Interrupt External interrupt or INT n
255 Interrupts instruction.

NOTES:
1. The UD2 instruction was introduced in the Pentium® Pro processor.
2. Intel Architecture processors after the Intel386™ processor do not generate this exception.
3. This exception was introduced in the Intel486™ processor.
4. This exception was introduced in the Pentium® processor and enhanced in the P6 family processors.
5. This exception was introduced in the Pentium® III processor.

5-6
 INTERRUPT AND EXCEPTION HANDLING

5.4. PROGRAM OR TASK RESTART

To allow restarting of program or task following the handling of an exception or an interrupt, all
exceptions except aborts are guaranteed to report the exception on a precise instruction
boundary, and all interrupts are guaranteed to be taken on an instruction boundary.
For fault-class exceptions, the return instruction pointer that the processor saves when it gener-
ates the exception points to the faulting instruction. So, when a program or task is restarted
following the handling of a fault, the faulting instruction is restarted (re-executed). Restarting
the faulting instruction is commonly used to handle exceptions that are generated when access
to an operand is blocked. The most common example of a fault is a page-fault exception (#PF)
that occurs when a program or task references an operand in a page that is not in memory. When
a page-fault exception occurs, the exception handler can load the page into memory and resume
execution of the program or task by restarting the faulting instruction. To insure that this instruc-
tion restart is handled transparently to the currently executing program or task, the processor
saves the necessary registers and stack pointers to allow it to restore itself to its state prior to the
execution of the faulting instruction.
For trap-class exceptions, the return instruction pointer points to the instruction following the
trapping instruction. If a trap is detected during an instruction which transfers execution, the
return instruction pointer reflects the transfer. For example, if a trap is detected while executing
a JMP instruction, the return instruction pointer points to the destination of the JMP instruction,
not to the next address past the JMP instruction. All trap exceptions allow program or task restart
with no loss of continuity. For example, the overflow exception is a trapping exception. Here,
the return instruction pointer points to the instruction following the INTO instruction that tested
the OF (overflow) flag in the EFLAGS register. The trap handler for this exception resolves the
overflow condition. Upon return from the trap handler, program or task execution continues at
the next instruction following the INTO instruction.
The abort-class exceptions do not support reliable restarting of the program or task. Abort
handlers generally are designed to collect diagnostic information about the state of the processor
when the abort exception occurred and then shut down the application and system as gracefully
as possible.
Interrupts rigorously support restarting of interrupted programs and tasks without loss of conti-
nuity. The return instruction pointer saved for an interrupt points to the next instruction to be
executed at the instruction boundary where the processor took the interrupt. If the instruction
just executed has a repeat prefix, the interrupt is taken at the end of the current iteration with the
registers set to execute the next iteration.
The ability of a P6 family processor to speculatively execute instructions does not affect the
taking of interrupts by the processor. Interrupts are taken at instruction boundaries located
during the retirement phase of instruction execution; so they are always taken in the “in-order”
instruction stream. Refer to Chapter 2, Introduction to the Intel Architecture, in the Intel Archi-
tecture Software Developer’s Manual, Volume 1, for more information about the P6 family
processors’ microarchitecture and its support for out-of-order instruction execution.
Note that the Pentium® processor and earlier Intel Architecture processors also perform varying
amounts of prefetching and preliminary decoding of instructions; however, here also exceptions
and interrupts are not signaled until actual “in-order” execution of the instructions. For a given

5-7
INTERRUPT AND EXCEPTION HANDLING

code sample, the signaling of exceptions will occur uniformly when the code is executed on any
family of Intel Architecture processors (except where new exceptions or new opcodes have been
defined).

5.5. NONMASKABLE INTERRUPT (NMI)

The nonmaskable interrupt (NMI) can be generated in either of two ways:
• External hardware asserts the NMI pin.
• The processor receives a message on the APIC serial bus of delivery mode NMI.
When the processor receives a NMI from either of these sources, the processor handles it imme-
diately by calling the NMI handler pointed to by interrupt vector number 2. The processor also
invokes certain hardware conditions to insure that no other interrupts, including NMI interrupts,
are received until the NMI handler has completed executing (refer to Section 5.5.1., “Handling
Multiple NMIs”).
Also, when an NMI is received from either of the above sources, it cannot be masked by the IF
flag in the EFLAGS register.
It is possible to issue a maskable hardware interrupt (through the INTR pin) to vector 2 to invoke
the NMI interrupt handler; however, this interrupt will not truly be an NMI interrupt. A true NMI
interrupt that activates the processor’s NMI-handling hardware can only be delivered through
one of the mechanisms listed above.

5.5.1. Handling Multiple NMIs

While an NMI interrupt handler is executing, the processor disables additional calls to the NMI
handler until the next IRET instruction is executed. This blocking of subsequent NMIs prevents
stacking up calls to the NMI handler. It is recommended that the NMI interrupt handler be
accessed through an interrupt gate to disable maskable hardware interrupts (refer to Section
5.6.1., “Masking Maskable Hardware Interrupts”).

5.6. ENABLING AND DISABLING INTERRUPTS

The processor inhibits the generation of some interrupts, depending on the state of the processor
and of the IF and RF flags in the EFLAGS register, as described in the following sections.

5.6.1. Masking Maskable Hardware Interrupts

The IF flag can disable the servicing of maskable hardware interrupts received on the
processor’s INTR pin or through the local APIC (refer to Section 5.1.1.2., “Maskable Hardware
Interrupts”). When the IF flag is clear, the processor inhibits interrupts delivered to the INTR
pin or through the local APIC from generating an internal interrupt request; when the IF flag is
set, interrupts delivered to the INTR or through the local APIC pin are processed as normal

5-8
 INTERRUPT AND EXCEPTION HANDLING

external interrupts. The IF flag does not affect nonmaskable interrupts (NMIs) delivered to the
NMI pin or delivery mode NMI messages delivered through the APIC serial bus, nor does it
affect processor generated exceptions. As with the other flags in the EFLAGS register, the
processor clears the IF flag in response to a hardware reset.
The fact that the group of maskable hardware interrupts includes the reserved interrupt and
exception vectors 0 through 32 can potentially cause confusion. Architecturally, when the IF
flag is set, an interrupt for any of the vectors from 0 through 32 can be delivered to the processor
through the INTR pin and any of the vectors from 16 through 32 can be delivered through the
local APIC. The processor will then generate an interrupt and call the interrupt or exception
handler pointed to by the vector number. So for example, it is possible to invoke the page-fault
handler through the INTR pin (by means of vector 14); however, this is not a true page-fault
exception. It is an interrupt. As with the INT n instruction (refer to Section 5.1.2.2., “Software-
Generated Exceptions”), when an interrupt is generated through the INTR pin to an exception
vector, the processor does not push an error code on the stack, so the exception handler may not
operate correctly.
The IF flag can be set or cleared with the STI (set interrupt-enable flag) and CLI (clear interrupt-
enable flag) instructions, respectively. These instructions may be executed only if the CPL is
equal to or less than the IOPL. A general-protection exception (#GP) is generated if they are
executed when the CPL is greater than the IOPL. (The effect of the IOPL on these instructions
is modified slightly when the virtual mode extension is enabled by setting the VME flag in
control register CR4, refer to Section 16.3., “Interrupt and Exception Handling in Virtual-8086
Mode” in Chapter 16, 8086 Emulation.)
The IF flag is also affected by the following operations:
• The PUSHF instruction stores all flags on the stack, where they can be examined and
modified. The POPF instruction can be used to load the modified flags back into the
EFLAGS register.
• Task switches and the POPF and IRET instructions load the EFLAGS register; therefore,
they can be used to modify the setting of the IF flag.
• When an interrupt is handled through an interrupt gate, the IF flag is automatically cleared,
which disables maskable hardware interrupts. (If an interrupt is handled through a trap
gate, the IF flag is not cleared.)
Refer to the descriptions of the CLI, STI, PUSHF, POPF, and IRET instructions in Chapter 3,
Instruction Set Reference, of the Intel Architecture Software Developer’s Manual, Volume 2, for
a detailed description of the operations these instructions are allowed to perform on the IF flag.

5.6.2. Masking Instruction Breakpoints

The RF (resume) flag in the EFLAGS register controls the response of the processor to instruc-
tion-breakpoint conditions (refer to the description of the RF flag in Section 2.3., “System Flags
and Fields in the EFLAGS Register” in Chapter 2, System Architecture Overview). When set, it
prevents an instruction breakpoint from generating a debug exception (#DB); when clear,
instruction breakpoints will generate debug exceptions. The primary function of the RF flag is

5-9
INTERRUPT AND EXCEPTION HANDLING

to prevent the processor from going into a debug exception loop on an instruction-breakpoint.
Refer to Section 15.3.1.1., “Instruction-Breakpoint Exception Condition”, in Chapter 15,
Debugging and Performance Monitoring, for more information on the use of this flag.

5.6.3. Masking Exceptions and Interrupts When Switching

Stacks
To switch to a different stack segment, software often uses a pair of instructions, for example:
MOV SS, AX
MOV ESP, StackTop

If an interrupt or exception occurs after the segment selector has been loaded into the SS register
but before the ESP register has been loaded, these two parts of the logical address into the stack
space are inconsistent for the duration of the interrupt or exception handler.
To prevent this situation, the processor inhibits interrupts, debug exceptions, and single-step trap
exceptions after either a MOV to SS instruction or a POP to SS instruction, until the instruction
boundary following the next instruction is reached. All other faults may still be generated. If the
LSS instruction is used to modify the contents of the SS register (which is the recommended
method of modifying this register), this problem does not occur.

5.7. PRIORITY AMONG SIMULTANEOUS EXCEPTIONS AND

INTERRUPTS
If more than one exception or interrupt is pending at an instruction boundary, the processor
services them in a predictable order. Table 5-3 shows the priority among classes of exception
and interrupt sources. While priority among these classes is consistent throughout the architec-
ture, exceptions within each class are implementation-dependent and may vary from processor
to processor. The processor first services a pending exception or interrupt from the class which
has the highest priority, transferring execution to the first instruction of the handler. Lower
priority exceptions are discarded; lower priority interrupts are held pending. Discarded excep-
tions are re-generated when the interrupt handler returns execution to the point in the program
or task where the exceptions and/or interrupts occurred.
The Pentium® III processor added the SIMD floating-point execution unit. The SIMD floating-
point execution unit can generate exceptions as well. Since the SIMD floating-point execution
unit utilizes a 4-wide register set an exception may result from more than one operand within a
SIMD floating-point register. Hence the Pentium® III processor handles these exceptions
according to a predetermined precedence. When a sub-operand of a packed instruction generates
two or more exception conditions, the exception precedence sometimes results in the higher
priority exception being handled and the lower priority exceptions being ignored. Prioritization
of exceptions is performed only on a sub-operand basis, and not between suboperands. For
example, an invalid exception generated by one sub-operand will not prevent the reporting of a
divide-by-zero exception generated by another sub-operand. Table 5-2 shows the precedence for
Streaming SIMD Extensions numeric exceptions. The table reflects the order in which interrupts
are handled upon simultaneous recognition by the processor (for example, when multiple inter-
rupts are pending at an instruction boundary). However, the table does not necessarily reflect the

5-10
 INTERRUPT AND EXCEPTION HANDLING

order in which interrupts will be recognized by the processor if received simultaneously at the
processor pins.

Table 5-2. SIMD Floating-Point Exceptions Priority

Priority Description

1(Highest) Invalid operation exception due to SNaN

operand (or any NaN operand for max, min, or
certain compare and convert operations)

2 QNaN operand1

3 Any other invalid operation exception not

mentioned above or a divide-by-zero
exception2

4 Denormal operand exception2

5 Numeric overflow and underflow exceptions

possibly in conjunction with the inexact result
exception2

6(Lowest) Inexact result exception

1. Though this is not an exception, the handling of a QNaN operand has precedence over lower priority
exceptions. For example, a QNaN divided by zero results in a QNaN, not a zero-divide exception.
2. If masked, then instruction execution continues, and a lower priority exception can occur as well.

5.8. INTERRUPT DESCRIPTOR TABLE (IDT)

The interrupt descriptor table (IDT) associates each exception or interrupt vector with a gate
descriptor for the procedure or task used to service the associated exception or interrupt. Like
the GDT and LDTs, the IDT is an array of 8-byte descriptors (in protected mode). Unlike the
GDT, the first entry of the IDT may contain a descriptor. To form an index into the IDT, the
processor scales the exception or interrupt vector by eight (the number of bytes in a gate
descriptor). Because there are only 256 interrupt or exception vectors, the IDT need not contain
more than 256 descriptors. It can contain fewer than 256 descriptors, because descriptors are
required only for the interrupt and exception vectors that may occur. All empty descriptor slots
in the IDT should have the present flag for the descriptor set to 0.

5-11
INTERRUPT AND EXCEPTION HANDLING

Table 5-3. Priority Among Simultaneous Exceptions and Interrupts

Priority Descriptions
1 (Highest) Hardware Reset and Machine Checks
- RESET
- Machine Check
2 Trap on Task Switch
- T flag in TSS is set
3 External Hardware Interventions
- FLUSH
- STOPCLK
- SMI
- INIT
4 Traps on the Previous Instruction
- Breakpoints
- Debug Trap Exceptions (TF flag set or data/I-O breakpoint)
5 External Interrupts
- NMI Interrupts
- Maskable Hardware Interrupts
6 Faults from Fetching Next Instruction
- Code Breakpoint Fault
- Code-Segment Limit Violation1
- Code Page Fault1
7 Faults from Decoding the Next Instruction
- Instruction length > 15 bytes
- Illegal Opcode
- Coprocessor Not Available
8 (Lowest) Faults on Executing an Instruction
- Floating-point exception
- Overflow
- Bound error
- Invalid TSS
- Segment Not Present
- Stack fault
- General Protection
- Data Page Fault
- Alignment Check
- SIMD floating-point exception

NOTE:
1. For the Pentium® and Intel486™ processors, the Code Segment Limit Violation and the Code Page Fault
exceptions are assigned to the priority 7.

The base addresses of the IDT should be aligned on an 8-byte boundary to maximize perfor-
mance of cache line fills. The limit value is expressed in bytes and is added to the base address
to get the address of the last valid byte. A limit value of 0 results in exactly 1 valid byte. Because
IDT entries are always eight bytes long, the limit should always be one less than an integral
multiple of eight (that is, 8N – 1).

5-12
 INTERRUPT AND EXCEPTION HANDLING

The IDT may reside anywhere in the linear address space. As shown in Figure 5-1, the processor
locates the IDT using the IDTR register. This register holds both a 32-bit base address and 16-bit
limit for the IDT.

IDTR Register
47 16 15 0
IDT Base Address IDT Limit

Interrupt
Descriptor Table (IDT)
+ Gate for
Interrupt #n (n−1)∗8

Gate for
Interrupt #3 16
Gate for
Interrupt #2 8
Gate for
Interrupt #1 0
31 0

Figure 5-1. Relationship of the IDTR and IDT

The LIDT (load IDT register) and SIDT (store IDT register) instructions load and store the
contents of the IDTR register, respectively. The LIDT instruction loads the IDTR register with
the base address and limit held in a memory operand. This instruction can be executed only
when the CPL is 0. It normally is used by the initialization code of an operating system when
creating an IDT. An operating system also may use it to change from one IDT to another. The
SIDT instruction copies the base and limit value stored in IDTR to memory. This instruction can
be executed at any privilege level.
If a vector references a descriptor beyond the limit of the IDT, a general-protection exception
(#GP) is generated.

5.9. IDT DESCRIPTORS

The IDT may contain any of three kinds of gate descriptors:
• Task-gate descriptor
• Interrupt-gate descriptor
• Trap-gate descriptor

5-13
INTERRUPT AND EXCEPTION HANDLING

Figure 5-2 shows the formats for the task-gate, interrupt-gate, and trap-gate descriptors. The
format of a task gate used in an IDT is the same as that of a task gate used in the GDT or an LDT
(refer to Section 6.2.4., “Task-Gate Descriptor” in Chapter 6, Task Management). The task gate
contains the segment selector for a TSS for an exception and/or interrupt handler task.

Task Gate
31 16 15 14 13 12 8 7 0

D
P P 0 0 1 0 1 4
L

31 16 15 0

TSS Segment Selector 0

Interrupt Gate
31 16 15 14 13 12 8 7 5 4 0

D
Offset 31..16 P P 0 D 1 1 0 0 0 0 4
L

31 16 15 0

Segment Selector Offset 15..0 0

Trap Gate
31 16 15 14 13 12 8 7 5 4 0

D
Offset 31..16 P P 0 D 1 1 1 0 0 0 4
L

31 16 15 0

Segment Selector Offset 15..0 0

DPL Descriptor Privilege Level

Offset Offset to procedure entry point
P Segment Present flag
Selector Segment Selector for destination code segment
D Size of gate: 1 = 32 bits; 0 = 16 bits
Reserved

Figure 5-2. IDT Gate Descriptors

Interrupt and trap gates are very similar to call gates (refer to Section 4.8.3., “Call Gates” in
Chapter 4, Protection). They contain a far pointer (segment selector and offset) that the
processor uses to transfer execution to a handler procedure in an exception- or interrupt-handler

5-14
 INTERRUPT AND EXCEPTION HANDLING

code segment. These gates differ in the way the processor handles the IF flag in the EFLAGS
register (refer to Section 5.10.1.2., “Flag Usage By Exception- or Interrupt-Handler Proce-
dure”).

5.10. EXCEPTION AND INTERRUPT HANDLING

The processor handles calls to exception- and interrupt-handlers similar to the way it handles
calls with a CALL instruction to a procedure or a task. When responding to an exception or inter-
rupt, the processor uses the exception or interrupt vector as an index to a descriptor in the IDT.
If the index points to an interrupt gate or trap gate, the processor calls the exception or interrupt
handler in a manner similar to a CALL to a call gate (refer to Section 4.8.2., “Gate Descriptors”
through Section 4.8.6., “Returning from a Called Procedure” in Chapter 4, Protection). If index
points to a task gate, the processor executes a task switch to the exception- or interrupt-handler
task in a manner similar to a CALL to a task gate (refer to Section 6.3., “Task Switching” in
Chapter 6, Task Management).

5.10.1. Exception- or Interrupt-Handler Procedures

An interrupt gate or trap gate references an exception- or interrupt-handler procedure that runs
in the context of the currently executing task (refer to Figure 5-3). The segment selector for the
gate points to a segment descriptor for an executable code segment in either the GDT or the
current LDT. The offset field of the gate descriptor points to the beginning of the exception- or
interrupt-handling procedure.
When the processor performs a call to the exception- or interrupt-handler procedure, it saves the
current states of the EFLAGS register, CS register, and EIP register on the stack (refer to Figure
5-4). (The CS and EIP registers provide a return instruction pointer for the handler.) If an excep-
tion causes an error code to be saved, it is pushed on the stack after the EIP value.
If the handler procedure is going to be executed at the same privilege level as the interrupted
procedure, the handler uses the current stack.
If the handler procedure is going to be executed at a numerically lower privilege level, a stack
switch occurs. When a stack switch occurs, a stack pointer for the stack to be returned to is also
saved on the stack. (The SS and ESP registers provide a return stack pointer for the handler.)
The segment selector and stack pointer for the stack to be used by the handler is obtained from
the TSS for the currently executing task. The processor copies the EFLAGS, SS, ESP, CS, EIP,
and error code information from the interrupted procedure’s stack to the handler’s stack.
To return from an exception- or interrupt-handler procedure, the handler must use the IRET (or
IRETD) instruction. The IRET instruction is similar to the RET instruction except that it restores
the saved flags into the EFLAGS register. The IOPL field of the EFLAGS register is restored
only if the CPL is 0. The IF flag is changed only if the CPL is less than or equal to the IOPL.
Refer to “IRET/IRETD—Interrupt Return” in Chapter 3 of the Intel Architecture Software
Developer’s Manual, Volume 2, for the complete operation performed by the IRET instruction.
If a stack switch occurred when calling the handler procedure, the IRET instruction switches
back to the interrupted procedure’s stack on the return.

5-15
INTERRUPT AND EXCEPTION HANDLING

Destination
IDT Code Segment

Interrupt
Offset Procedure
Interrupt
Vector
Interrupt or
Trap Gate
+

Segment Selector

GDT or LDT
Base
Address

Segment
Descriptor

Figure 5-3. Interrupt Procedure Call

5-16
 INTERRUPT AND EXCEPTION HANDLING

Stack Usage with No

Privilege-Level Change
Interrupted Procedure’s
and Handler’s Stack

ESP Before
EFLAGS Transfer to Handler
CS
EIP
Error Code ESP After
Transfer to Handler

Stack Usage with

Privilege-Level Change
Interrupted Procedure’s Handler’s Stack
Stack

ESP Before
Transfer to Handler SS
ESP
EFLAGS
CS
EIP
ESP After Error Code
Transfer to Handler

Figure 5-4. Stack Usage on Transfers to Interrupt and Exception-Handling Routines

5.10.1.1. PROTECTION OF EXCEPTION- AND INTERRUPT-HANDLER

PROCEDURES
The privilege-level protection for exception- and interrupt-handler procedures is similar to that
used for ordinary procedure calls when called through a call gate (refer to Section 4.8.4.,
“Accessing a Code Segment Through a Call Gate” in Chapter 4, Protection). The processor does
not permit transfer of execution to an exception- or interrupt-handler procedure in a less privi-
leged code segment (numerically greater privilege level) than the CPL. An attempt to violate this
rule results in a general-protection exception (#GP). The protection mechanism for exception-
and interrupt-handler procedures is different in the following ways:
• Because interrupt and exception vectors have no RPL, the RPL is not checked on implicit
calls to exception and interrupt handlers.
• The processor checks the DPL of the interrupt or trap gate only if an exception or interrupt
is generated with an INT n, INT 3, or INTO instruction. Here, the CPL must be less than or
equal to the DPL of the gate. This restriction prevents application programs or procedures
running at privilege level 3 from using a software interrupt to access critical exception

5-17
INTERRUPT AND EXCEPTION HANDLING

handlers, such as the page-fault handler, providing that those handlers are placed in more
privileged code segments (numerically lower privilege level). For hardware-generated
interrupts and processor-detected exceptions, the processor ignores the DPL of interrupt
and trap gates.
Because exceptions and interrupts generally do not occur at predictable times, these privilege
rules effectively impose restrictions on the privilege levels at which exception and interrupt-
handling procedures can run. Either of the following techniques can be used to avoid privilege-
level violations.
• The exception or interrupt handler can be placed in a conforming code segment. This
technique can be used for handlers that only need to access data available on the stack (for
example, divide error exceptions). If the handler needs data from a data segment, the data
segment needs to be accessible from privilege level 3, which would make it unprotected.
• The handler can be placed in a nonconforming code segment with privilege level 0. This
handler would always run, regardless of the CPL that the interrupted program or task is
running at.

5.10.1.2. FLAG USAGE BY EXCEPTION- OR INTERRUPT-HANDLER

PROCEDURE
When accessing an exception or interrupt handler through either an interrupt gate or a trap gate,
the processor clears the TF flag in the EFLAGS register after it saves the contents of the
EFLAGS register on the stack. (On calls to exception and interrupt handlers, the processor also
clears the VM, RF, and NT flags in the EFLAGS register, after they are saved on the stack.)
Clearing the TF flag prevents instruction tracing from affecting interrupt response. A subsequent
IRET instruction restores the TF (and VM, RF, and NT) flags to the values in the saved contents
of the EFLAGS register on the stack.
The only difference between an interrupt gate and a trap gate is the way the processor handles
the IF flag in the EFLAGS register. When accessing an exception- or interrupt-handling proce-
dure through an interrupt gate, the processor clears the IF flag to prevent other interrupts from
interfering with the current interrupt handler. A subsequent IRET instruction restores the IF flag
to its value in the saved contents of the EFLAGS register on the stack. Accessing a handler
procedure through a trap gate does not affect the IF flag.

5.10.2. Interrupt Tasks

When an exception or interrupt handler is accessed through a task gate in the IDT, a task switch
results. Handling an exception or interrupt with a separate task offers several advantages:
• The entire context of the interrupted program or task is saved automatically.
• A new TSS permits the handler to use a new privilege level 0 stack when handling the
exception or interrupt. If an exception or interrupt occurs when the current privilege level 0
stack is corrupted, accessing the handler through a task gate can prevent a system crash by
providing the handler with a new privilege level 0 stack.

5-18
 INTERRUPT AND EXCEPTION HANDLING

• The handler can be further isolated from other tasks by giving it a separate address space.
This is done by giving it a separate LDT.
The disadvantage of handling an interrupt with a separate task is that the amount of machine
state that must be saved on a task switch makes it slower than using an interrupt gate, resulting
in increased interrupt latency.
A task gate in the IDT references a TSS descriptor in the GDT (refer to Figure 5-5). A switch to
the handler task is handled in the same manner as an ordinary task switch (refer to Section 6.3.,
“Task Switching” in Chapter 6, Task Management). The link back to the interrupted task is
stored in the previous task link field of the handler task’s TSS. If an exception caused an error
code to be generated, this error code is copied to the stack of the new task.

TSS for Interrupt-

IDT Handling Task

Interrupt
Vector Task Gate

TSS Selector TSS

Base
Address
GDT

TSS Descriptor

Figure 5-5. Interrupt Task Switch

5-19
INTERRUPT AND EXCEPTION HANDLING

When exception- or interrupt-handler tasks are used in an operating system, there are actually
two mechanisms that can be used to dispatch tasks: the software scheduler (part of the operating
system) and the hardware scheduler (part of the processor’s interrupt mechanism). The software
scheduler needs to accommodate interrupt tasks that may be dispatched when interrupts are
enabled.

5.11. ERROR CODE

When an exception condition is related to a specific segment, the processor pushes an error code
onto the stack of the exception handler (whether it is a procedure or task). The error code has
the format shown in Figure 5-6. The error code resembles a segment selector; however, instead
of a TI flag and RPL field, the error code contains 3 flags:
EXT External event (bit 0). When set, indicates that an event external to the
program caused the exception, such as a hardware interrupt.
IDT Descriptor location (bit 1). When set, indicates that the index portion of the
error code refers to a gate descriptor in the IDT; when clear, indicates that the
index refers to a descriptor in the GDT or the current LDT.
TI GDT/LDT (bit 2). Only used when the IDT flag is clear. When set, the TI flag
indicates that the index portion of the error code refers to a segment or gate
descriptor in the LDT; when clear, it indicates that the index refers to a
descriptor in the current GDT.

31 3 2 1 0
T I E
Reserved Segment Selector Index I D X
T T

Figure 5-6. Error Code

The segment selector index field provides an index into the IDT, GDT, or current LDT to the
segment or gate selector being referenced by the error code. In some cases the error code is null
(that is, all bits in the lower word are clear). A null error code indicates that the error was not
caused by a reference to a specific segment or that a null segment descriptor was referenced in
an operation.
The format of the error code is different for page-fault exceptions (#PF), refer to “Interrupt
14—Page-Fault Exception (#PF)” in this chapter.
The error code is pushed on the stack as a doubleword or word (depending on the default inter-
rupt, trap, or task gate size). To keep the stack aligned for doubleword pushes, the upper half of
the error code is reserved. Note that the error code is not popped when the IRET instruction is
executed to return from an exception handler, so the handler must remove the error code before
executing a return.

5-20
 INTERRUPT AND EXCEPTION HANDLING

Error codes are not pushed on the stack for exceptions that are generated externally (with the
INTR or LINT[1:0] pins) or the INT n instruction, even if an error code is normally produced
for those exceptions.

5.12. EXCEPTION AND INTERRUPT REFERENCE

The following sections describe conditions which generate exceptions and interrupts. They are
arranged in the order of vector numbers. The information contained in these sections are as
follows:
Exception Class Indicates whether the exception class is a fault, trap, or abort type.
Some exceptions can be either a fault or trap type, depending on
when the error condition is detected. (This section is not applicable
to interrupts.)
Description Gives a general description of the purpose of the exception or inter-
rupt type. It also describes how the processor handles the exception
or interrupt.
Exception Error Code Indicates whether an error code is saved for the exception. If one is
saved, the contents of the error code are described. (This section is
not applicable to interrupts.)
Saved Instruction Pointer Describes which instruction the saved (or return) instruction pointer
points to. It also indicates whether the pointer can be used to restart
a faulting instruction.
Program State Change Describes the effects of the exception or interrupt on the state of the
currently running program or task and the possibilities of restarting
the program or task without loss of continuity.

5-21
INTERRUPT AND EXCEPTION HANDLING

Interrupt 0—Divide Error Exception (#DE)

Exception Class Fault.

Description
Indicates the divisor operand for a DIV or IDIV instruction is 0 or that the result cannot be repre-
sented in the number of bits specified for the destination operand.

Exception Error Code

None.

Saved Instruction Pointer

Saved contents of CS and EIP registers point to the instruction that generated the exception.

Program State Change

A program-state change does not accompany the divide error, because the exception occurs
before the faulting instruction is executed.

5-22
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 1—Debug Exception (#DB)

Exception Class Trap or Fault. The exception handler can distinguish between traps or
faults by examining the contents of DR6 and the other debug registers.

Description
Indicates that one or more of several debug-exception conditions has been detected. Whether the
exception is a fault or a trap depends on the condition, as shown below:

Exception Condition Exception Class

Instruction fetch breakpoint Fault
Data read or write breakpoint Trap
I/O read or write breakpoint Trap
General detect condition (in conjunction with in-circuit emulation) Fault
Single-step Trap
Task-switch Trap
Execution of INT 1 instruction Trap

Refer to Chapter 15, Debugging and Performance Monitoring, for detailed information about
the debug exceptions.

Exception Error Code

None. An exception handler can examine the debug registers to determine which condition
caused the exception.

Saved Instruction Pointer

Fault—Saved contents of CS and EIP registers point to the instruction that generated the
exception.
Trap—Saved contents of CS and EIP registers point to the instruction following the instruction
that generated the exception.

Program State Change

Fault—A program-state change does not accompany the debug exception, because the excep-
tion occurs before the faulting instruction is executed. The program can resume normal execu-
tion upon returning from the debug exception handler
Trap—A program-state change does accompany the debug exception, because the instruction or
task switch being executed is allowed to complete before the exception is generated. However,
the new state of the program is not corrupted and execution of the program can continue reliably.

5-23
INTERRUPT AND EXCEPTION HANDLING

Interrupt 2—NMI Interrupt

Exception Class Not applicable.

Description
The nonmaskable interrupt (NMI) is generated externally by asserting the processor’s NMI pin
or through an NMI request set by the I/O APIC to the local APIC on the APIC serial bus. This
interrupt causes the NMI interrupt handler to be called.

Exception Error Code

Not applicable.

Saved Instruction Pointer

The processor always takes an NMI interrupt on an instruction boundary. The saved contents of
CS and EIP registers point to the next instruction to be executed at the point the interrupt is
taken. Refer to Section 5.4., “Program or Task Restart” for more information about when the
processor takes NMI interrupts.

Program State Change

The instruction executing when an NMI interrupt is received is completed before the NMI is
generated. A program or task can thus be restarted upon returning from an interrupt handler
without loss of continuity, provided the interrupt handler saves the state of the processor before
handling the interrupt and restores the processor’s state prior to a return.

5-24
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 3—Breakpoint Exception (#BP)

Exception Class Trap.

Description
Indicates that a breakpoint instruction (INT 3) was executed, causing a breakpoint trap to be
generated. Typically, a debugger sets a breakpoint by replacing the first opcode byte of an
instruction with the opcode for the INT 3 instruction. (The INT 3 instruction is one byte long,
which makes it easy to replace an opcode in a code segment in RAM with the breakpoint
opcode.) The operating system or a debugging tool can use a data segment mapped to the same
physical address space as the code segment to place an INT 3 instruction in places where it is
desired to call the debugger.
With the P6 family, Pentium®, Intel486™, and Intel386™ processors, it is more convenient to
set breakpoints with the debug registers. (Refer to Section 15.3.2., “Breakpoint Exception
(#BP)—Interrupt Vector 3”, in Chapter 15, Debugging and Performance Monitoring, for infor-
mation about the breakpoint exception.) If more breakpoints are needed beyond what the debug
registers allow, the INT 3 instruction can be used.
The breakpoint (#BP) exception can also be generated by executing the INT n instruction with
an operand of 3. The action of this instruction (INT 3) is slightly different than that of the INT
3 instruction (refer to “INTn/INTO/INT3—Call to Interrupt Procedure” in Chapter 3 of the Intel
Architecture Software Developer’s Manual, Volume 2).

Exception Error Code

None.

Saved Instruction Pointer

Saved contents of CS and EIP registers point to the instruction following the INT 3 instruction.

Program State Change

Even though the EIP points to the instruction following the breakpoint instruction, the state of
the program is essentially unchanged because the INT 3 instruction does not affect any register
or memory locations. The debugger can thus resume the suspended program by replacing the
INT 3 instruction that caused the breakpoint with the original opcode and decrementing the
saved contents of the EIP register. Upon returning from the debugger, program execution
resumes with the replaced instruction.

5-25
INTERRUPT AND EXCEPTION HANDLING

Interrupt 4—Overflow Exception (#OF)

Exception Class Trap.

Description
Indicates that an overflow trap occurred when an INTO instruction was executed. The INTO
instruction checks the state of the OF flag in the EFLAGS register. If the OF flag is set, an over-
flow trap is generated.
Some arithmetic instructions (such as the ADD and SUB) perform both signed and unsigned
arithmetic. These instructions set the OF and CF flags in the EFLAGS register to indicate signed
overflow and unsigned overflow, respectively. When performing arithmetic on signed operands,
the OF flag can be tested directly or the INTO instruction can be used. The benefit of using the
INTO instruction is that if the overflow exception is detected, an exception handler can be called
automatically to handle the overflow condition.

Exception Error Code

None.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the instruction following the INTO
instruction.

Program State Change

Even though the EIP points to the instruction following the INTO instruction, the state of the
program is essentially unchanged because the INTO instruction does not affect any register or
memory locations. The program can thus resume normal execution upon returning from the
overflow exception handler.

5-26
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 5—BOUND Range Exceeded Exception (#BR)

Exception Class Fault.

Description
Indicates that a BOUND-range-exceeded fault occurred when a BOUND instruction was
executed. The BOUND instruction checks that a signed array index is within the upper and
lower bounds of an array located in memory. If the array index is not within the bounds of the
array, a BOUND-range-exceeded fault is generated.

Exception Error Code

None.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the BOUND instruction that generated the
exception.

Program State Change

A program-state change does not accompany the bounds-check fault, because the operands for
the BOUND instruction are not modified. Returning from the BOUND-range-exceeded excep-
tion handler causes the BOUND instruction to be restarted.

5-27
INTERRUPT AND EXCEPTION HANDLING

Interrupt 6—Invalid Opcode Exception (#UD)

Exception Class Fault.

Description
Indicates that the processor did one of the following things:
• Attempted to execute a Streaming SIMD Extensions instruction in an Intel Architecture
processor that does not support the Streaming SIMD Extensions.
• Attempted to execute a Streaming SIMD Extensions instruction when the OSFXSR bit is
not set (0) in CR4. Note this does not include the following Streaming SIMD Extensions:
PAVGB, PAVGW, PEXTRW, PINSRW, PMAXSW, PMAXUB, PMINSW, PMINUB,
PMOVMSKB, PMULHUW, PSADBW, PSHUFW, MASKMOVQ, MOVNTQ,
PREFETCH and SFENCE.
• Attempted to execute a Streaming SIMD Extensions instruction in an Intel Architecture
processor which causes a numeric exception when the OSXMMEXCPT bit is not set (0) in
CR4.
• Attempted to execute an invalid or reserved opcode, including any MMX™ instruction in
an Intel Architecture processor that does not support the MMX™ architecture.
• Attempted to execute an MMX™ instruction or SIMD floating-point instruction when the
EM flag in register CR0 is set. Note this does not include the following Streaming SIMD
Extensions: SFENCE and PREFETCH.
• Attempted to execute an instruction with an operand type that is invalid for its accompa-
nying opcode; for example, the source operand for a LES instruction is not a memory
location.
• Executed a UD2 instruction.
• Detected a LOCK prefix that precedes an instruction that may not be locked or one that
may be locked but the destination operand is not a memory location.
• Attempted to execute an LLDT, SLDT, LTR, STR, LSL, LAR, VERR, VERW, or ARPL
instruction while in real-address or virtual-8086 mode.
• Attempted to execute the RSM instruction when not in SMM mode.
In the P6 family processors, this exception is not generated until an attempt is made to retire the
result of executing an invalid instruction; that is, decoding and speculatively attempting to
execute an invalid opcode does not generate this exception. Likewise, in the Pentium® processor
and earlier Intel Architecture processors, this exception is not generated as the result of
prefetching and preliminary decoding of an invalid instruction. (Refer to Section 5.4., “Program
or Task Restart” for general rules for taking of interrupts and exceptions.)
The opcodes D6 and F1 are undefined opcodes that are reserved by Intel. These opcodes, even
though undefined, do not generate an invalid opcode exception.

5-28
 INTERRUPT AND EXCEPTION HANDLING

The UD2 instruction is guaranteed to generate an invalid opcode exception.

Exception Error Code

None.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the instruction that generated the exception.

Program State Change

A program-state change does not accompany an invalid-opcode fault, because the invalid
instruction is not executed.

5-29
INTERRUPT AND EXCEPTION HANDLING

Interrupt 7—Device Not Available Exception (#NM)

Exception Class Fault.

Description
Indicates one of the following things:
The device-not-available fault is generated by either of three conditions:
• The processor executed a floating-point instruction while the EM flag of register CR0 was
set.
• The processor executed a floating-point, MMX™ or SIMD floating-point (excluding
prefetch, sfence or streaming store instructions) instruction while the TS flag of register
CR0 was set.
• The processor executed a WAIT or FWAIT instruction while the MP and TS flags of
register CR0 were set.
The EM flag is set when the processor does not have an internal floating-point unit. An excep-
tion is then generated each time a floating-point instruction is encountered, allowing an excep-
tion handler to call floating-point instruction emulation routines.
The TS flag indicates that a context switch (task switch) has occurred since the last time a
floating-point, MMX™ or SIMD floating-point (excluding prefetch, sfence or streaming store
instructions) instruction was executed, but that the context of the FPU was not saved. When the
TS flag is set, the processor generates a device-not-available exception each time a floating-
point, MMX™ or SIMD floating-point (excluding prefetch, sfence or streaming store instruc-
tions) instruction is encountered. The exception handler can then save the context of the FPU
before it executes the instruction. Refer to Section 2.5., “Control Registers”, in Chapter 2,
System Architecture Overview, for more information about the TS flag.
The MP flag in control register CR0 is used along with the TS flag to determine if WAIT or
FWAIT instructions should generate a device-not-available exception. It extends the function of
the TS flag to the WAIT and FWAIT instructions, giving the exception handler an opportunity
to save the context of the FPU before the WAIT or FWAIT instruction is executed. The MP flag
is provided primarily for use with the Intel286 and Intel386™ DX processors. For programs
running on the P6 family, Pentium®, or Intel486™ DX processors, or the Intel 487 SX coproces-
sors, the MP flag should always be set; for programs running on the Intel486™ SX processor,
the MP flag should be clear.

Exception Error Code

None.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the floating-point instruction or the
WAIT/FWAIT instruction that generated the exception.

5-30
 INTERRUPT AND EXCEPTION HANDLING

Program State Change

A program-state change does not accompany a device-not-available fault, because the instruc-
tion that generated the exception is not executed.
If the EM flag is set, the exception handler can then read the floating-point instruction pointed
to by the EIP and call the appropriate emulation routine.
If the MP and TS flags are set or the TS flag alone is set, the exception handler can save the
context of the FPU, clear the TS flag, and continue execution at the interrupted floating-point or
WAIT/FWAIT instruction.

5-31
INTERRUPT AND EXCEPTION HANDLING

Interrupt 8—Double Fault Exception (#DF)

Exception Class Abort.

Description
Indicates that the processor detected a second exception while calling an exception handler for
a prior exception. Normally, when the processor detects another exception while trying to call
an exception handler, the two exceptions can be handled serially. If, however, the processor
cannot handle them serially, it signals the double-fault exception. To determine when two faults
need to be signaled as a double fault, the processor divides the exceptions into three classes:
benign exceptions, contributory exceptions, and page faults (refer to Table 5-4).

Table 5-4. Interrupt and Exception Classes

Class Vector Number Description
Benign Exceptions and Interrupts 1 Debug Exception
2 NMI Interrupt
3 Breakpoint
4 Overflow
5 BOUND Range Exceeded
6 Invalid Opcode
7 Device Not Available
9 Coprocessor Segment Overrun
16 Floating-Point Error
17 Alignment Check
18 Machine Check
19 SIMD floating-point extensions
All INT n
All INTR
Contributory Exceptions 0 Divide Error
10 Invalid TSS
11 Segment Not Present
12 Stack Fault
13 General Protection
Page Faults 14 Page Fault

Table 5-5 shows the various combinations of exception classes that cause a double fault to be
generated. A double-fault exception falls in the abort class of exceptions. The program or task
cannot be restarted or resumed. The double-fault handler can be used to collect diagnostic infor-
mation about the state of the machine and/or, when possible, to shut the application and/or
system down gracefully or restart the system.
A segment or page fault may be encountered while prefetching instructions; however, this
behavior is outside the domain of Table 5-5. Any further faults generated while the processor is
attempting to transfer control to the appropriate fault handler could still lead to a double-fault
sequence.

5-32
 INTERRUPT AND EXCEPTION HANDLING

Table 5-5. Conditions for Generating a Double Fault

Second Exception
First Exception Benign Contributory Page Fault
Benign Handle Exceptions Handle Exceptions Handle Exceptions
Serially Serially Serially
Contributory Handle Exceptions Generate a Double Fault Handle Exceptions
Serially Serially
Page Fault Handle Exceptions Generate a Double Fault Generate a Double Fault
Serially

If another exception occurs while attempting to call the double-fault handler, the processor
enters shutdown mode. This mode is similar to the state following execution of an HLT instruc-
tion. In this mode, the processor stops executing instructions until an NMI interrupt, SMI inter-
rupt, hardware reset, or INIT# is received. The processor generates a special bus cycle to
indicate that it has entered shutdown mode. Software designers may need to be aware of the
response of hardware to receiving this signal. For example, hardware may turn on an indicator
light on the front panel, generate an NMI interrupt to record diagnostic information, invoke reset
initialization, generate an INIT initialization, or generate an SMI.
If the shutdown occurs while the processor is executing an NMI interrupt handler, then only a
hardware reset can restart the processor.

Exception Error Code

Zero. The processor always pushes an error code of 0 onto the stack of the double-fault handler.

Saved Instruction Pointer

The saved contents of CS and EIP registers are undefined.

Program State Change

A program-state following a double-fault exception is undefined. The program or task cannot
be resumed or restarted. The only available action of the double-fault exception handler is to
collect all possible context information for use in diagnostics and then close the application
and/or shut down or reset the processor.

5-33
INTERRUPT AND EXCEPTION HANDLING

Interrupt 9—Coprocessor Segment Overrun

Exception Class Abort. (Intel reserved; do not use. Recent Intel Architecture proces-
sors do not generate this exception.)

Description
Indicates that an Intel386™ CPU-based systems with an Intel 387 math coprocessor detected a
page or segment violation while transferring the middle portion of an Intel 387 math copro-
cessor operand. The P6 family, Pentium®, and Intel486™ processors do not generate this excep-
tion; instead, this condition is detected with a general protection exception (#GP), interrupt 13.

Exception Error Code

None.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the instruction that generated the exception.

Program State Change

A program-state following a coprocessor segment-overrun exception is undefined. The program
or task cannot be resumed or restarted. The only available action of the exception handler is to
save the instruction pointer and reinitialize the FPU using the FNINIT instruction.

5-34
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 10—Invalid TSS Exception (#TS)

Exception Class Fault.

Description
Indicates that a task switch was attempted and that invalid information was detected in the TSS
for the target task. Table 5-6 shows the conditions that will cause an invalid-TSS exception to
be generated. In general, these invalid conditions result from protection violations for the TSS
descriptor; the LDT pointed to by the TSS; or the stack, code, or data segments referenced by
the TSS.

Table 5-6. Invalid TSS Conditions

Error Code Index Invalid Condition
TSS segment selector index TSS segment limit less than 67H for 32-bit TSS or less than 2CH for 16-
bit TSS.
LDT segment selector index Invalid LDT or LDT not present
Stack-segment selector index Stack-segment selector exceeds descriptor table limit
Stack-segment selector index Stack segment is not writable
Stack-segment selector index Stack segment DPL ≠ CPL
Stack-segment selector index Stack-segment selector RPL ≠ CPL
Code-segment selector index Code-segment selector exceeds descriptor table limit
Code-segment selector index Code segment is not executable
Code-segment selector index Nonconforming code segment DPL ≠ CPL
Code-segment selector index Conforming code segment DPL greater than CPL
Data-segment selector index Data-segment selector exceeds descriptor table limit
Data-segment selector index Data segment not readable

This exception can generated either in the context of the original task or in the context of the
new task (refer to Section 6.3., “Task Switching” in Chapter 6, Task Management). Until the
processor has completely verified the presence of the new TSS, the exception is generated in the
context of the original task. Once the existence of the new TSS is verified, the task switch is
considered complete. Any invalid-TSS conditions detected after this point are handled in the
context of the new task. (A task switch is considered complete when the task register is loaded
with the segment selector for the new TSS and, if the switch is due to a procedure call or inter-
rupt, the previous task link field of the new TSS references the old TSS.)
To insure that a valid TSS is available to process the exception, the invalid-TSS exception
handler must be a task called using a task gate.

5-35
INTERRUPT AND EXCEPTION HANDLING

Exception Error Code

An error code containing the segment selector index for the segment descriptor that caused the
violation is pushed onto the stack of the exception handler. If the EXT flag is set, it indicates that
the exception was caused by an event external to the currently running program (for example, if
an external interrupt handler using a task gate attempted a task switch to an invalid TSS).

Saved Instruction Pointer

If the exception condition was detected before the task switch was carried out, the saved
contents of CS and EIP registers point to the instruction that invoked the task switch. If the
exception condition was detected after the task switch was carried out, the saved contents of CS
and EIP registers point to the first instruction of the new task.

Program State Change

The ability of the invalid-TSS handler to recover from the fault depends on the error condition
than causes the fault. Refer to Section 6.3., “Task Switching” in Chapter 6, Task Management
for more information on the task switch process and the possible recovery actions that can be
taken.
If an invalid TSS exception occurs during a task switch, it can occur before or after the commit-
to-new-task point. If it occurs before the commit point, no program state change occurs. If it
occurs after the commit point (when the segment descriptor information for the new segment
selectors have been loaded in the segment registers), the processor will load all the state infor-
mation from the new TSS before it generates the exception. During a task switch, the processor
first loads all the segment registers with segment selectors from the TSS, then checks their
contents for validity. If an invalid TSS exception is discovered, the remaining segment registers
are loaded but not checked for validity and therefore may not be usable for referencing memory.
The invalid TSS handler should not rely on being able to use the segment selectors found in the
CS, SS, DS, ES, FS, and GS registers without causing another exception. The exception handler
should load all segment registers before trying to resume the new task; otherwise, general-
protection exceptions (#GP) may result later under conditions that make diagnosis more diffi-
cult. The Intel recommended way of dealing situation is to use a task for the invalid TSS excep-
tion handler. The task switch back to the interrupted task from the invalid-TSS exception-
handler task will then cause the processor to check the registers as it loads them from the TSS.

5-36
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 11—Segment Not Present (#NP)

Exception Class Fault.

Description
Indicates that the present flag of a segment or gate descriptor is clear. The processor can generate
this exception during any of the following operations:
• While attempting to load CS, DS, ES, FS, or GS registers. [Detection of a not-present
segment while loading the SS register causes a stack fault exception (#SS) to be
generated.] This situation can occur while performing a task switch.
• While attempting to load the LDTR using an LLDT instruction. Detection of a not-present
LDT while loading the LDTR during a task switch operation causes an invalid-TSS
exception (#TS) to be generated.
• When executing the LTR instruction and the TSS is marked not present.
• While attempting to use a gate descriptor or TSS that is marked segment-not-present, but is
otherwise valid.
An operating system typically uses the segment-not-present exception to implement virtual
memory at the segment level. If the exception handler loads the segment and returns, the inter-
rupted program or task resumes execution.
A not-present indication in a gate descriptor, however, does not indicate that a segment is not
present (because gates do not correspond to segments). The operating system may use the
present flag for gate descriptors to trigger exceptions of special significance to the operating
system.

Exception Error Code

An error code containing the segment selector index for the segment descriptor that caused the
violation is pushed onto the stack of the exception handler. If the EXT flag is set, it indicates that
the exception resulted from an external event (NMI or INTR) that caused an interrupt, which
subsequently referenced a not-present segment. The IDT flag is set if the error code refers to an
IDT entry (e.g., an INT instruction referencing a not-present gate).

Saved Instruction Pointer

The saved contents of CS and EIP registers normally point to the instruction that generated the
exception. If the exception occurred while loading segment descriptors for the segment selectors
in a new TSS, the CS and EIP registers point to the first instruction in the new task. If the excep-
tion occurred while accessing a gate descriptor, the CS and EIP registers point to the instruction
that invoked the access (for example a CALL instruction that references a call gate).

5-37
INTERRUPT AND EXCEPTION HANDLING

Program State Change

If the segment-not-present exception occurs as the result of loading a register (CS, DS, SS, ES,
FS, GS, or LDTR), a program-state change does accompany the exception, because the register
is not loaded. Recovery from this exception is possible by simply loading the missing segment
into memory and setting the present flag in the segment descriptor.
If the segment-not-present exception occurs while accessing a gate descriptor, a program-state
change does not accompany the exception. Recovery from this exception is possible merely by
setting the present flag in the gate descriptor.
If a segment-not-present exception occurs during a task switch, it can occur before or after the
commit-to-new-task point (refer to Section 6.3., “Task Switching” in Chapter 6, Task Manage-
ment). If it occurs before the commit point, no program state change occurs. If it occurs after the
commit point, the processor will load all the state information from the new TSS (without
performing any additional limit, present, or type checks) before it generates the exception. The
segment-not-present exception handler should thus not rely on being able to use the segment
selectors found in the CS, SS, DS, ES, FS, and GS registers without causing another exception.
(Refer to the Program State Change description for “Interrupt 10—Invalid TSS Exception
(#TS)” in this chapter for additional information on how to handle this situation.)

5-38
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 12—Stack Fault Exception (#SS)

Exception Class Fault.

Description
Indicates that one of the following stack related conditions was detected:
• A limit violation is detected during an operation that refers to the SS register. Operations
that can cause a limit violation include stack-oriented instructions such as POP, PUSH,
CALL, RET, IRET, ENTER, and LEAVE, as well as other memory references which
implicitly or explicitly use the SS register (for example, MOV AX, [BP+6] or MOV AX,
SS:[EAX+6]). The ENTER instruction generates this exception when there is not enough
stack space for allocating local variables.
• A not-present stack segment is detected when attempting to load the SS register. This
violation can occur during the execution of a task switch, a CALL instruction to a different
privilege level, a return to a different privilege level, an LSS instruction, or a MOV or POP
instruction to the SS register.
Recovery from this fault is possible by either extending the limit of the stack segment (in the
case of a limit violation) or loading the missing stack segment into memory (in the case of a not-
present violation.

Exception Error Code

If the exception is caused by a not-present stack segment or by overflow of the new stack during
an inter-privilege-level call, the error code contains a segment selector for the segment that
caused the exception. Here, the exception handler can test the present flag in the segment
descriptor pointed to by the segment selector to determine the cause of the exception. For a
normal limit violation (on a stack segment already in use) the error code is set to 0.

Saved Instruction Pointer

The saved contents of CS and EIP registers generally point to the instruction that generated the
exception. However, when the exception results from attempting to load a not-present stack
segment during a task switch, the CS and EIP registers point to the first instruction of the new
task.

Program State Change

A program-state change does not generally accompany a stack-fault exception, because the
instruction that generated the fault is not executed. Here, the instruction can be restarted after
the exception handler has corrected the stack fault condition.
If a stack fault occurs during a task switch, it occurs after the commit-to-new-task point (refer
to Section 6.3., “Task Switching” Chapter 6, Task Management). Here, the processor loads all
the state information from the new TSS (without performing any additional limit, present, or

5-39
INTERRUPT AND EXCEPTION HANDLING

type checks) before it generates the exception. The stack fault handler should thus not rely on
being able to use the segment selectors found in the CS, SS, DS, ES, FS, and GS registers
without causing another exception. The exception handler should check all segment registers
before trying to resume the new task; otherwise, general protection faults may result later under
conditions that are more difficult to diagnose. (Refer to the Program State Change description
for “Interrupt 10—Invalid TSS Exception (#TS)” in this chapter for additional information on
how to handle this situation.)

5-40
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 13—General Protection Exception (#GP)

Exception Class Fault.

Description
Indicates that the processor detected one of a class of protection violations called “general-
protection violations.” The conditions that cause this exception to be generated comprise all the
protection violations that do not cause other exceptions to be generated (such as, invalid-TSS,
segment-not-present, stack-fault, or page-fault exceptions). The following conditions cause
general-protection exceptions to be generated:
• Exceeding the segment limit when accessing the CS, DS, ES, FS, or GS segments.
• Exceeding the segment limit when referencing a descriptor table (except during a task
switch or a stack switch).
• Transferring execution to a segment that is not executable.
• Writing to a code segment or a read-only data segment.
• Reading from an execute-only code segment.
• Loading the SS register with a segment selector for a read-only segment (unless the
selector comes from a TSS during a task switch, in which case an invalid-TSS exception
occurs).
• Loading the SS, DS, ES, FS, or GS register with a segment selector for a system segment.
• Loading the DS, ES, FS, or GS register with a segment selector for an execute-only code
segment.
• Loading the SS register with the segment selector of an executable segment or a null
segment selector.
• Loading the CS register with a segment selector for a data segment or a null segment
selector.
• Accessing memory using the DS, ES, FS, or GS register when it contains a null segment
selector.
• Switching to a busy task during a call or jump to a TSS.
• Switching to an available (nonbusy) task during the execution of an IRET instruction.
• Using a segment selector on task switch that points to a TSS descriptor in the current LDT.
TSS descriptors can only reside in the GDT.
• Violating any of the privilege rules described in Chapter 4, Protection.
• Exceeding the instruction length limit of 15 bytes (this only can occur when redundant
prefixes are placed before an instruction).

5-41
INTERRUPT AND EXCEPTION HANDLING

• Loading the CR0 register with a set PG flag (paging enabled) and a clear PE flag
(protection disabled).
• Loading the CR0 register with a set NW flag and a clear CD flag.
• Referencing an entry in the IDT (following an interrupt or exception) that is not an
interrupt, trap, or task gate.
• Attempting to access an interrupt or exception handler through an interrupt or trap gate
from virtual-8086 mode when the handler’s code segment DPL is greater than 0.
• Attempting to write a 1 into a reserved bit of CR4.
• Attempting to execute a privileged instruction when the CPL is not equal to 0 (refer to
Section 4.9., “Privileged Instructions” in Chapter 4, Protection for a list of privileged
instructions).
• Writing to a reserved bit in an MSR.
• Accessing a gate that contains a null segment selector.
• Executing the INT n instruction when the CPL is greater than the DPL of the referenced
interrupt, trap, or task gate.
• The segment selector in a call, interrupt, or trap gate does not point to a code segment.
• The segment selector operand in the LLDT instruction is a local type (TI flag is set) or
does not point to a segment descriptor of the LDT type.
• The segment selector operand in the LTR instruction is local or points to a TSS that is not
available.
• The target code-segment selector for a call, jump, or return is null.
• If the PAE and/or PSE flag in control register CR4 is set and the processor detects any
reserved bits in a page-directory-pointer-table entry set to 1. These bits are checked during
a write to control registers CR0, CR3, or CR4 that causes a reloading of the page-
directory-pointer-table entry.
A program or task can be restarted following any general-protection exception. If the exception
occurs while attempting to call an interrupt handler, the interrupted program can be restartable,
but the interrupt may be lost.

Exception Error Code

The processor pushes an error code onto the exception handler’s stack. If the fault condition was
detected while loading a segment descriptor, the error code contains a segment selector to or IDT
vector number for the descriptor; otherwise, the error code is 0. The source of the selector in an
error code may be any of the following:
• An operand of the instruction.
• A selector from a gate which is the operand of the instruction.
• A selector from a TSS involved in a task switch.

5-42
 INTERRUPT AND EXCEPTION HANDLING

• IDT vector number.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the instruction that generated the exception.

Program State Change

In general, a program-state change does not accompany a general-protection exception, because
the invalid instruction or operation is not executed. An exception handler can be designed to
correct all of the conditions that cause general-protection exceptions and restart the program or
task without any loss of program continuity.
If a general-protection exception occurs during a task switch, it can occur before or after the
commit-to-new-task point (refer to Section 6.3., “Task Switching” in Chapter 6, Task Manage-
ment). If it occurs before the commit point, no program state change occurs. If it occurs after the
commit point, the processor will load all the state information from the new TSS (without
performing any additional limit, present, or type checks) before it generates the exception. The
general-protection exception handler should thus not rely on being able to use the segment selec-
tors found in the CS, SS, DS, ES, FS, and GS registers without causing another exception. (Refer
to the Program State Change description for “Interrupt 10—Invalid TSS Exception (#TS)” in
this chapter for additional information on how to handle this situation.)

5-43
INTERRUPT AND EXCEPTION HANDLING

Interrupt 14—Page-Fault Exception (#PF)

Exception Class Fault.

Description
Indicates that, with paging enabled (the PG flag in the CR0 register is set), the processor detected
one of the following conditions while using the page-translation mechanism to translate a linear
address to a physical address:
• The P (present) flag in a page-directory or page-table entry needed for the address
translation is clear, indicating that a page table or the page containing the operand is not
present in physical memory.
• The procedure does not have sufficient privilege to access the indicated page (that is, a
procedure running in user mode attempts to access a supervisor-mode page).
• Code running in user mode attempts to write to a read-only page. In the Intel486™ and
later processors, if the WP flag is set in CR0, the page fault will also be triggered by code
running in supervisor mode that tries to write to a read-only user-mode page.
The exception handler can recover from page-not-present conditions and restart the program or
task without any loss of program continuity. It can also restart the program or task after a privi-
lege violation, but the problem that caused the privilege violation may be uncorrectable.

Exception Error Code

Yes (special format). The processor provides the page-fault handler with two items of informa-
tion to aid in diagnosing the exception and recovering from it:
• An error code on the stack. The error code for a page fault has a format different from that
for other exceptions (refer to Figure 5-7). The error code tells the exception handler four
things:
— The P flag indicates whether the exception was due to a not-present page (0) or to
either an access rights violation or the use of a reserved bit (1).
— The W/R flag indicates whether the memory access that caused the exception was a
read (0) or write (1).
— The U/S flag indicates whether the processor was executing at user mode (1) or
supervisor mode (0) at the time of the exception.
— The RSVD flag indicates that the processor detected 1s in reserved bits of the page
directory, when the PSE or PAE flags in control register CR4 are set to 1. (The PSE
flag is only available in the P6 family and Pentium® processors, and the PAE flag is
only available on the P6 family processors. In earlier Intel Architecture processor
families, the bit position of the RSVD flag is reserved.)

5-44
 INTERRUPT AND EXCEPTION HANDLING

31 4 3 2 1 0
R U R
S
Reserved V / / P
D S W

P 0 The fault was caused by a nonpresent page.

1 The fault was caused by a page-level protection violation.

W/R 0 The access causing the fault was a read.

1 The access causing the fault was a write.

U/S 0 The access causing the fault originated when the processor
was executing in supervisor mode.
1 The access causing the fault originated when the processor
was executing in user mode.
RSVD 0 The fault was not caused by a reserved bit violation.
1 The page fault occured because a 1 was detected in one of the
reserved bit positions of a page table entry or directory entry
that was marked present.

Figure 5-7. Page-Fault Error Code

• The contents of the CR2 register. The processor loads the CR2 register with the 32-bit
linear address that generated the exception. The page-fault handler can use this address to
locate the corresponding page directory and page-table entries. If another page fault can
potentially occur during execution of the page-fault handler, the handler must push the
contents of the CR2 register onto the stack before the second page fault occurs.
If a page fault is caused by a page-level protection violation, the access flag in the page-directory
entry is set when the fault occurs. The behavior of Intel Architecture processors regarding the
access flag in the corresponding page-table entry is model specific and not architecturally
defined.

Saved Instruction Pointer

The saved contents of CS and EIP registers generally point to the instruction that generated the
exception. If the page-fault exception occurred during a task switch, the CS and EIP registers
may point to the first instruction of the new task (as described in the following “Program State
Change” section).

Program State Change

A program-state change does not normally accompany a page-fault exception, because the
instruction that causes the exception to be generated is not executed. After the page-fault excep-
tion handler has corrected the violation (for example, loaded the missing page into memory),
execution of the program or task can be resumed.

5-45
INTERRUPT AND EXCEPTION HANDLING

When a page-fault exception is generated during a task switch, the program-state may change,
as follows. During a task switch, a page-fault exception can occur during any of following
operations:
• While writing the state of the original task into the TSS of that task.
• While reading the GDT to locate the TSS descriptor of the new task.
• While reading the TSS of the new task.
• While reading segment descriptors associated with segment selectors from the new task.
• While reading the LDT of the new task to verify the segment registers stored in the new
TSS.
In the last two cases the exception occurs in the context of the new task. The instruction pointer
refers to the first instruction of the new task, not to the instruction which caused the task switch
(or the last instruction to be executed, in the case of an interrupt). If the design of the operating
system permits page faults to occur during task-switches, the page-fault handler should be called
through a task gate.
If a page fault occurs during a task switch, the processor will load all the state information from
the new TSS (without performing any additional limit, present, or type checks) before it gener-
ates the exception. The page-fault handler should thus not rely on being able to use the segment
selectors found in the CS, SS, DS, ES, FS, and GS registers without causing another exception.
(Refer to the Program State Change description for “Interrupt 10—Invalid TSS Exception
(#TS)” in this chapter for additional information on how to handle this situation.)

Additional Exception-Handling Information

Special care should be taken to ensure that an exception that occurs during an explicit stack
switch does not cause the processor to use an invalid stack pointer (SS:ESP). Software written
for 16-bit Intel Architecture processors often use a pair of instructions to change to a new stack,
for example:
MOV SS, AX
MOV SP, StackTop

When executing this code on one of the 32-bit Intel Architecture processors, it is possible to get
a page fault, general-protection fault (#GP), or alignment check fault (#AC) after the segment
selector has been loaded into the SS register but before the ESP register has been loaded. At this
point, the two parts of the stack pointer (SS and ESP) are inconsistent. The new stack segment
is being used with the old stack pointer.
The processor does not use the inconsistent stack pointer if the exception handler switches to a
well defined stack (that is, the handler is a task or a more privileged procedure). However, if the
exception handler is called at the same privilege level and from the same task, the processor will
attempt to use the inconsistent stack pointer.
In systems that handle page-fault, general-protection, or alignment check exceptions within the
faulting task (with trap or interrupt gates), software executing at the same privilege level as the
exception handler should initialize a new stack by using the LSS instruction rather than a pair

5-46
 INTERRUPT AND EXCEPTION HANDLING

of MOV instructions, as described earlier in this note. When the exception handler is running at
privilege level 0 (the normal case), the problem is limited to procedures or tasks that run at priv-
ilege level 0, typically the kernel of the operating system.

5-47
INTERRUPT AND EXCEPTION HANDLING

Interrupt 16—Floating-Point Error Exception (#MF)

Exception Class Fault.

Description
Indicates that the FPU has detected a floating-point-error exception. The NE flag in the register
CR0 must be set and the appropriate exception must be unmasked (clear mask bit in the control
register) for an interrupt 16, floating-point-error exception to be generated. (Refer to Section
2.5., “Control Registers” in Chapter 2, System Architecture Overview for a detailed description
of the NE flag.)
While executing floating-point instructions, the FPU detects and reports six types of floating-
point errors:
• Invalid operation (#I)
— Stack overflow or underflow (#IS)
— Invalid arithmetic operation (#IA)
• Divide-by-zero (#Z)
• Denormalized operand (#D)
• Numeric overflow (#O)
• Numeric underflow (#U)
• Inexact result (precision) (#P)
For each of these error types, the FPU provides a flag in the FPU status register and a mask bit
in the FPU control register. If the FPU detects a floating-point error and the mask bit for the error
is set, the FPU handles the error automatically by generating a predefined (default) response and
continuing program execution. The default responses have been designed to provide a reason-
able result for most floating-point applications.
If the mask for the error is clear and the NE flag in register CR0 is set, the FPU does the
following:
1. Sets the necessary flag in the FPU status register.
2. Waits until the next “waiting” floating-point instruction or WAIT/FWAIT instruction is
encountered in the program’s instruction stream. (The FPU checks for pending floating-
point exceptions on “waiting” instructions prior to executing them. All the floating-point
instructions except the FNINIT, FNCLEX, FNSTSW, FNSTSW AX, FNSTCW,
FNSTENV, and FNSAVE instructions are “waiting” instructions.)
3. Generates an internal error signal that causes the processor to generate a floating-point-
error exception.

5-48
 INTERRUPT AND EXCEPTION HANDLING

All of the floating-point-error conditions can be recovered from. The floating-point-error excep-
tion handler can determine the error condition that caused the exception from the settings of the
flags in the FPU status word. Refer to “Software Exception Handling” in Chapter 7 of the Intel
Architecture Software Developer’s Manual, Volume 1, for more information on handling
floating-point-error exceptions.

Exception Error Code

None. The FPU provides its own error information.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the floating-point or WAIT/FWAIT instruc-
tion that was about to be executed when the floating-point-error exception was generated. This
is not the faulting instruction in which the error condition was detected. The address of the
faulting instruction is contained in the FPU instruction pointer register. Refer to “The FPU
Instruction and Operand (Data) Pointers” in Chapter 7 of the Intel Architecture Software Devel-
oper’s Manual, Volume 1, for more information about information the FPU saves for use in
handling floating-point-error exceptions.

Program State Change

A program-state change generally accompanies a floating-point-error exception because the
handling of the exception is delayed until the next waiting floating-point or WAIT/FWAIT
instruction following the faulting instruction. The FPU, however, saves sufficient information
about the error condition to allow recovery from the error and re-execution of the faulting
instruction if needed.
In situations where nonfloating-point instructions depend on the results of a floating-point
instruction, a WAIT or FWAIT instruction can be inserted in front of a dependent instruction to
force a pending floating-point-error exception to be handled before the dependent instruction is
executed. Refer to “Floating-Point Exception Synchronization” in Chapter 7 of the Intel Archi-
tecture Software Developer’s Manual, Volume 1, for more information about synchronization of
floating-point-error exceptions.

5-49
INTERRUPT AND EXCEPTION HANDLING

Interrupt 17—Alignment Check Exception (#AC)

Exception Class Fault.

Description
Indicates that the processor detected an unaligned memory operand when alignment checking
was enabled. Alignment checks are only carried out in data (or stack) segments (not in code or
system segments). An example of an alignment-check violation is a word stored at an odd byte
address, or a doubleword stored at an address that is not an integer multiple of 4. Table 5-7 lists
the alignment requirements various data types recognized by the processor.
Table 5-7. Alignment Requirements by Data Type
Data Type Address Must Be Divisible By
Word 2
Doubleword 4
Single Real 4
Double Real 8
Extended Real 8
Segment Selector 2
32-bit Far Pointer 2
48-bit Far Pointer 4
32-bit Pointer 4
GDTR, IDTR, LDTR, or Task Register Contents 4
FSTENV/FLDENV Save Area 4 or 2, depending on operand size
FSAVE/FRSTOR Save Area 4 or 2, depending on operand size
Bit String 2 or 4 depending on the operand-size attribute.
1
128-bit 16

1. 128-bit datatype introduced with the Pentium® III processor. This type of alignment check is done for
operands less than 128-bits in size: 32-bit scalar single and 16-bit/32-bit/64-bit integer MMX™ technol-
ogy; 2, 4, or 8 byte alignments checks are possible when #AC is enabled. Some exceptional cases are:
• The MOVUPS instruction, which performs a 128-bit unaligned load or store. In this case, 2/4/8-byte
misalignments will be detected, but detection of 16-byte misalignment is not guaranteed and may
vary with implementation.
• The FXSAVE/FXRSTOR instructions - refer to instruction descriptions
To enable alignment checking, the following conditions must be true:
• AM flag in CR0 register is set.
• AC flag in the EFLAGS register is set.
• The CPL is 3 (protected mode or virtual-8086 mode).

5-50
 INTERRUPT AND EXCEPTION HANDLING

Alignment-check faults are generated only when operating at privilege level 3 (user mode).
Memory references that default to privilege level 0, such as segment descriptor loads, do not
generate alignment-check faults, even when caused by a memory reference made from privilege
level 3.
Storing the contents of the GDTR, IDTR, LDTR, or task register in memory while at privilege
level 3 can generate an alignment-check fault. Although application programs do not normally
store these registers, the fault can be avoided by aligning the information stored on an even
word-address.
FSAVE and FRSTOR instructions generate unaligned references which can cause alignment-
check faults. These instructions are rarely needed by application programs.

Exception Error Code

Yes (always zero).

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the instruction that generated the exception.

Program State Change

A program-state change does not accompany an alignment-check fault, because the instruction
is not executed.

5-51
INTERRUPT AND EXCEPTION HANDLING

Interrupt 18—Machine-Check Exception (#MC)

Exception Class Abort.

Description
Indicates that the processor detected an internal machine error or a bus error, or that an external
agent detected a bus error. The machine-check exception is model-specific, available only on
the P6 family and Pentium® processors. The implementation of the machine-check exception is
different between the P6 family and Pentium® processors, and these implementations may not
be compatible with future Intel Architecture processors. (Use the CPUID instruction to deter-
mine whether this feature is present.)
Bus errors detected by external agents are signaled to the processor on dedicated pins: the
BINIT# pin on the P6 family processors and the BUSCHK# pin on the Pentium® processor.
When one of these pins is enabled, asserting the pin causes error information to be loaded into
machine-check registers and a machine-check exception is generated.
The machine-check exception and machine-check architecture are discussed in detail in Chapter
13, Machine-Check Architecture. Also, refer to the data books for the individual processors for
processor-specific hardware information.

Exception Error Code

None. Error information is provide by machine-check MSRs.

Saved Instruction Pointer

For the P6 family processors, if the EIPV flag in the MCG_STATUS MSR is set, the saved
contents of CS and EIP registers are directly associated with the error that caused the machine-
check exception to be generated; if the flag is clear, the saved instruction pointer may not be
associated with the error (refer to Section 13.3.1.2., “MCG_STATUS MSR”, in Chapter 13,
Machine-Check Architecture).
For the Pentium® processor, contents of the CS and EIP registers may not be associated with the
error.

Program State Change

A program-state change always accompanies a machine-check exception. If the machine-check
mechanism is enabled (the MCE flag in control register CR4 is set), a machine-check exception
results in an abort; that is, information about the exception can be collected from the machine-
check MSRs, but the program cannot be restarted. If the machine-check mechanism is not
enabled, a machine-check exception causes the processor to enter the shutdown state.

5-52
 INTERRUPT AND EXCEPTION HANDLING

Interrupt 19—SIMD Floating-Point Exception (#XF)

Exception Class Fault.

Description
Indicates the processor has detected a SIMD floating-point execution unit exception. The appro-
priate status flag in the MXCSR register must be set and the particular exception unmasked for
this interrupt to be generated.
There are six classes of numeric exception conditions that can occur while executing Streaming
SIMD Extensions:

1. Invalid operation (#I)

2. Divide-by-zero (#Z)
3. Denormalized operand (#D)
4. Numeric overflow (#O)
5. Numeric underflow (#U)
6. Inexact result (Precision) (#P)
Invalid, Divide-by-zero, and Denormal exceptions are pre-computation exceptions, i.e., they are
detected before any arithmetic operation occurs. Underflow, Overflow, and Precision exceptions
are post-computational exceptions.
When numeric exceptions occur, a processor supporting Streaming SIMD Extensions takes one
of two possible courses of action:

• The processor can handle the exception by itself, producing the most reasonable result and
allowing numeric program execution to continue undisturbed (i.e., masked exception
response).
• A software exception handler can be invoked to handle the exception (i.e., unmasked
exception response).
Each of the six exception conditions described above has corresponding flag and mask bits in
the MXCSR. If an exception is masked (the corresponding mask bit in MXCSR = 1), the
processor takes an appropriate default action and continues with the computation. If the excep-
tion is unmasked (mask bit = 0) and the OS supports SIMD floating-point exceptions (i.e.
CR4.OSXMMEXCPT = 1), a software exception handler is invoked immediately through
SIMD floating-point exception interrupt vector 19. If the exception is unmasked (mask bit = 0)
and the OS does not support SIMD floating-point exceptions (i.e. CR4.OSXMMEXCPT = 0),
an invalid opcode exception is signaled instead of a SIMD floating-point exception.
Note that because SIMD floating-point exceptions are precise and occur immediately, the situ-
ation does not arise where an x87-FP instruction, an FWAIT instruction, or another Streaming
SIMD Extensions instruction will catch a pending unmasked SIMD floating-point exception.

5-53
INTERRUPT AND EXCEPTION HANDLING

Exception Error Code

None. The Streaming SIMD Extensions provide their own error information.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the Streaming SIMD Extensions instruction
that was executed when the SIMD floating-point exception was generated. This is the faulting
instruction in which the error condition was detected.

Program State Change

A program-state change generally accompanies a SIMD floating-point exception because the
handling of the exception is immediate unless the particular exception is masked. The Pentium®
III processor contains sufficient information about the error condition to allow recovery from
the error and re-execution of the faulting instruction if needed.
In situations where a SIMD floating-point exception occurred while the SIMD floating-point
exceptions were masked, SIMD floating-point exceptions were then unmasked, and a Streaming
SIMD Extensions instruction was executed, then no exception is raised.

5-54
 INTERRUPT AND EXCEPTION HANDLING

Interrupts 32 to 255—User Defined Interrupts

Exception Class Not applicable.

Description
Indicates that the processor did one of the following things:
• Executed an INT n instruction where the instruction operand is one of the vector numbers
from 32 through 255.
• Responded to an interrupt request at the INTR pin or from the local APIC when the
interrupt vector number associated with the request is from 32 through 255.

Exception Error Code

Not applicable.

Saved Instruction Pointer

The saved contents of CS and EIP registers point to the instruction that follows the INT n
instruction or instruction following the instruction on which the INTR signal occurred.

Program State Change

A program-state change does not accompany interrupts generated by the INT n instruction or
the INTR signal. The INT n instruction generates the interrupt within the instruction stream.
When the processor receives an INTR signal, it commits all state changes for all previous
instructions before it responds to the interrupt; so, program execution can resume upon returning
from the interrupt handler.

5-55
INTERRUPT AND EXCEPTION HANDLING

5-56
 6
Task Management
 TASK MANAGEMENT

CHAPTER 6
TASK MANAGEMENT

This chapter describes the Intel Architecture’s task management facilities. These facilities are
only available when the processor is running in protected mode.

6.1. TASK MANAGEMENT OVERVIEW

A task is a unit of work that a processor can dispatch, execute, and suspend. It can be used to
execute a program, a task or process, an operating-system service utility, an interrupt or excep-
tion handler, or a kernel or executive utility.
The Intel Architecture provides a mechanism for saving the state of a task, for dispatching tasks
for execution, and for switching from one task to another. When operating in protected mode,
all processor execution takes place from within a task. Even simple systems must define at least
one task. More complex systems can use the processor’s task management facilities to support
multitasking applications.

6.1.1. Task Structure

A task is made up of two parts: a task execution space and a task-state segment (TSS). The task
execution space consists of a code segment, a stack segment, and one or more data segments
(refer to Figure 6-1). If an operating system or executive uses the processor’s privilege-level
protection mechanism, the task execution space also provides a separate stack for each privilege
level.
The TSS specifies the segments that make up the task execution space and provides a storage
place for task state information. In multitasking systems, the TSS also provides a mechanism for
linking tasks.

NOTE
This chapter describes primarily 32-bit tasks and the 32-bit TSS structure.
For information on 16-bit tasks and the 16-bit TSS structure, refer to Section
6.6., “16-Bit Task-State Segment (TSS)”.
A task is identified by the segment selector for its TSS. When a task is loaded into the processor
for execution, the segment selector, base address, limit, and segment descriptor attributes for the
TSS are loaded into the task register (refer to Section 2.4.4., “Task Register (TR)” in Chapter 2,
System Architecture Overview).
If paging is implemented for the task, the base address of the page directory used by the task is
loaded into control register CR3.

6-1
TASK MANAGEMENT

Code
Segment

Task-State Data
Segment Segment
(TSS)
Stack
Segment
(Current Priv.
Level)

Stack Seg.
Priv. Level 0

Stack Seg.
Priv. Level 1
Task Register Stack
Segment
CR3 (Priv. Level 2)

Figure 6-1. Structure of a Task

6.1.2. Task State

The following items define the state of the currently executing task:
• The task’s current execution space, defined by the segment selectors in the segment
registers (CS, DS, SS, ES, FS, and GS).
• The state of the general-purpose registers.
• The state of the EFLAGS register.
• The state of the EIP register.
• The state of control register CR3.
• The state of the task register.
• The state of the LDTR register.
• The I/O map base address and I/O map (contained in the TSS).
• Stack pointers to the privilege 0, 1, and 2 stacks (contained in the TSS).
• Link to previously executed task (contained in the TSS).
Prior to dispatching a task, all of these items are contained in the task’s TSS, except the state of
the task register. Also, the complete contents of the LDTR register are not contained in the TSS,
only the segment selector for the LDT.

6-2
 TASK MANAGEMENT

6.1.3. Executing a Task

Software or the processor can dispatch a task for execution in one of the following ways:
• A explicit call to a task with the CALL instruction.
• A explicit jump to a task with the JMP instruction.
• An implicit call (by the processor) to an interrupt-handler task.
• An implicit call to an exception-handler task.
• A return (initiated with an IRET instruction) when the NT flag in the EFLAGS register is
set.
All of these methods of dispatching a task identify the task to be dispatched with a segment
selector that points either to a task gate or the TSS for the task. When dispatching a task with a
CALL or JMP instruction, the selector in the instruction may select either the TSS directly or a
task gate that holds the selector for the TSS. When dispatching a task to handle an interrupt or
exception, the IDT entry for the interrupt or exception must contain a task gate that holds the
selector for the interrupt- or exception-handler TSS.
When a task is dispatched for execution, a task switch automatically occurs between the
currently running task and the dispatched task. During a task switch, the execution environment
of the currently executing task (called the task’s state or context) is saved in its TSS and execu-
tion of the task is suspended. The context for the dispatched task is then loaded into the processor
and execution of that task begins with the instruction pointed to by the newly loaded EIP
register. If the task has not been run since the system was last initialized, the EIP will point to
the first instruction of the task’s code; otherwise, it will point to the next instruction after the last
instruction that the task executed when it was last active.
If the currently executing task (the calling task) called the task being dispatched (the called task),
the TSS segment selector for the calling task is stored in the TSS of the called task to provide a
link back to the calling task.
For all Intel Architecture processors, tasks are not recursive. A task cannot call or jump to itself.
Interrupts and exceptions can be handled with a task switch to a handler task. Here, the processor
not only can perform a task switch to handle the interrupt or exception, but it can automatically
switch back to the interrupted task upon returning from the interrupt- or exception-handler task.
This mechanism can handle interrupts that occur during interrupt tasks.
As part of a task switch, the processor can also switch to another LDT, allowing each task to have
a different logical-to-physical address mapping for LDT-based segments. The page-directory base
register (CR3) also is reloaded on a task switch, allowing each task to have its own set of page
tables. These protection facilities help isolate tasks and prevent them from interfering with one
another. If one or both of these protection mechanisms are not used, the processor provides no
protection between tasks. This is true even with operating systems that use multiple privilege
levels for protection. Here, a task running at privilege level 3 that uses the same LDT and page
tables as other privilege-level-3 tasks can access code and corrupt data and the stack of other
tasks.

6-3
TASK MANAGEMENT

Use of task management facilities for handling multitasking applications is optional. Multi-
tasking can be handled in software, with each software defined task executed in the context of
a single Intel Architecture task.

6.2. TASK MANAGEMENT DATA STRUCTURES

The processor defines five data structures for handling task-related activities:
• Task-state segment (TSS).
• Task-gate descriptor.
• TSS descriptor.
• Task register.
• NT flag in the EFLAGS register.
When operating in protected mode, a TSS and TSS descriptor must be created for at least one
task, and the segment selector for the TSS must be loaded into the task register (using the LTR
instruction).

6.2.1. Task-State Segment (TSS)

The processor state information needed to restore a task is saved in a system segment called the
task-state segment (TSS). Figure 6-2 shows the format of a TSS for tasks designed for 32-bit
CPUs. (Compatibility with 16-bit Intel 286 processor tasks is provided by a different kind of
TSS, refer to Figure 6-9.) The fields of a TSS are divided into two main categories: dynamic
fields and static fields.
The processor updates the dynamic fields when a task is suspended during a task switch. The
following are dynamic fields:
General-purpose register fields
State of the EAX, ECX, EDX, EBX, ESP, EBP, ESI, and EDI registers prior to
the task switch.
Segment selector fields
Segment selectors stored in the ES, CS, SS, DS, FS, and GS registers prior to
the task switch.
EFLAGS register field
State of the EFAGS register prior to the task switch.
EIP (instruction pointer) field
State of the EIP register prior to the task switch.
Previous task link field
Contains the segment selector for the TSS of the previous task (updated on a
task switch that was initiated by a call, interrupt, or exception). This field

6-4
 TASK MANAGEMENT

(which is sometimes called the back link field) permits a task switch back to
the previous task to be initiated with an IRET instruction.
The processor reads the static fields, but does not normally change them. These fields are set up
when a task is created. The following are static fields:
LDT segment selector field
Contains the segment selector for the task’s LDT.

31 15 0
I/O Map Base Address T 100
LDT Segment Selector 96
GS 92
FS 88
DS 84
SS 80
CS 76
ES 72
EDI 68
ESI 64
EBP 60
ESP 56
EBX 52
EDX 48
ECX 44
EAX 40
EFLAGS 36
EIP 32
CR3 (PDBR) 28
SS2 24
ESP2 20
SS1 16
ESP1 12
SS0 8
ESP0 4
Previous Task Link 0

Reserved bits. Set to 0.

Figure 6-2. 32-Bit Task-State Segment (TSS)

6-5
TASK MANAGEMENT

CR3 control register field

Contains the base physical address of the page directory to be used by the task.
Control register CR3 is also known as the page-directory base register (PDBR).
Privilege level-0, -1, and -2 stack pointer fields
These stack pointers consist of a logical address made up of the segment
selector for the stack segment (SS0, SS1, and SS2) and an offset into the stack
(ESP0, ESP1, and ESP2). Note that the values in these fields are static for a
particular task; whereas, the SS and ESP values will change if stack switching
occurs within the task.
T (debug trap) flag (byte 100, bit 0)
When set, the T flag causes the processor to raise a debug exception when a
task switch to this task occurs (refer to Section 15.3.1.5., “Task-Switch Excep-
tion Condition”, in Chapter 15, Debugging and Performance Monitoring).
I/O map base address field
Contains a 16-bit offset from the base of the TSS to the I/O permission bit map
and interrupt redirection bitmap. When present, these maps are stored in the
TSS at higher addresses. The I/O map base address points to the beginning of
the I/O permission bit map and the end of the interrupt redirection bit map.
Refer to Chapter 9, Input/Output, in the Intel Architecture Software Devel-
oper’s Manual, Volume 1, for more information about the I/O permission bit
map. Refer to Section 16.3., “Interrupt and Exception Handling in Virtual-
8086 Mode” in Chapter 16, 8086 Emulation for a detailed description of the
interrupt redirection bit map.
If paging is used, care should be taken to avoid placing a page boundary within the part of the
TSS that the processor reads during a task switch (the first 104 bytes). If a page boundary is
placed within this part of the TSS, the pages on either side of the boundary must be present at
the same time and contiguous in physical memory. The reason for this restriction is that when
accessing a TSS during a task switch, the processor reads and writes into the first 104 bytes of
each TSS from contiguous physical addresses beginning with the physical address of the first
byte of the TSS. It may not perform address translations at a page boundary if one occurs within
this area. So, after the TSS access begins, if a part of the 104 bytes is not both present and phys-
ically contiguous, the processor will access incorrect TSS information, without generating a
page-fault exception. The reading of this incorrect information will generally lead to an unre-
coverable exception later in the task switch process.
Also, if paging is used, the pages corresponding to the previous task’s TSS, the current task’s
TSS, and the descriptor table entries for each should be marked as read/write. The task switch
will be carried out faster if the pages containing these structures are also present in memory
before the task switch is initiated.

6.2.2. TSS Descriptor

The TSS, like all other segments, is defined by a segment descriptor. Figure 6-3 shows the
format of a TSS descriptor. TSS descriptors may only be placed in the GDT; they cannot be
placed in an LDT or the IDT. An attempt to access a TSS using a segment selector with its TI

6-6
 TASK MANAGEMENT

flag set (which indicates the current LDT) causes a general-protection exception (#GP) to be
generated. A general-protection exception is also generated if an attempt is made to load a
segment selector for a TSS into a segment register.
The busy flag (B) in the type field indicates whether the task is busy. A busy task is currently
running or is suspended. A type field with a value of 1001B indicates an inactive task; a value
of 1011B indicates a busy task. Tasks are not recursive. The processor uses the busy flag to
detect an attempt to call a task whose execution has been interrupted. To insure that there is only
one busy flag is associated with a task, each TSS should have only one TSS descriptor that points
to it.

TSS Descriptor
31 24 23 22 21 20 19 16 15 14 13 12 11 8 7 0

A Limit D Type
Base 31:24 G 0 0 V P P Base 23:16 4
19:16
L L 0 1 0 B 1

31 16 15 0

Base Address 15:00 Segment Limit 15:00 0

AVL Available for use by system software

B Busy flag
BASE Segment Base Address
DPL Descriptor Privilege Level
G Granularity
LIMIT Segment Limit
P Segment Present
TYPE Segment Type

Figure 6-3. TSS Descriptor

The base, limit, and DPL fields and the granularity and present flags have functions similar to
their use in data-segment descriptors (refer to Section 3.4.3., “Segment Descriptors” in Chapter
3, Protected-Mode Memory Management). The limit field must have a value equal to or greater
than 67H (for a 32-bit TSS), one byte less than the minimum size of a TSS. Attempting to switch
to a task whose TSS descriptor has a limit less than 67H generates an invalid-TSS exception
(#TS). A larger limit is required if an I/O permission bit map is included in the TSS. An even
larger limit would be required if the operating system stores additional data in the TSS. The
processor does not check for a limit greater than 67H on a task switch; however, it does when
accessing the I/O permission bit map or interrupt redirection bit map.
Any program or procedure with access to a TSS descriptor (that is, whose CPL is numerically
equal to or less than the DPL of the TSS descriptor) can dispatch the task with a call or a jump.
In most systems, the DPLs of TSS descriptors should be set to values less than 3, so that only
privileged software can perform task switching. However, in multitasking applications, DPLs
for some TSS descriptors can be set to 3 to allow task switching at the application (or user) priv-
ilege level.

6-7
TASK MANAGEMENT

6.2.3. Task Register

The task register holds the 16-bit segment selector and the entire segment descriptor (32-bit base
address, 16-bit segment limit, and descriptor attributes) for the TSS of the current task (refer to
Figure 2-4 in Chapter 2, System Architecture Overview). This information is copied from the
TSS descriptor in the GDT for the current task. Figure 6-4 shows the path the processor uses to
accesses the TSS, using the information in the task register.
The task register has both a visible part (that can be read and changed by software) and an invis-
ible part (that is maintained by the processor and is inaccessible by software). The segment
selector in the visible portion points to a TSS descriptor in the GDT. The processor uses the
invisible portion of the task register to cache the segment descriptor for the TSS. Caching these
values in a register makes execution of the task more efficient, because the processor does not
need to fetch these values from memory to reference the TSS of the current task.
The LTR (load task register) and STR (store task register) instructions load and read the visible
portion of the task register. The LTR instruction loads a segment selector (source operand) into
the task register that points to a TSS descriptor in the GDT, and then loads the invisible portion
of the task register with information from the TSS descriptor. This instruction is a privileged
instruction that may be executed only when the CPL is 0. The LTR instruction generally is used
during system initialization to put an initial value in the task register. Afterwards, the contents
of the task register are changed implicitly when a task switch occurs.
The STR (store task register) instruction stores the visible portion of the task register in a
general-purpose register or memory. This instruction can be executed by code running at any
privilege level, to identify the currently running task; however, it is normally used only by oper-
ating system software.
On power up or reset of the processor, the segment selector and base address are set to the default
value of 0 and the limit is set to FFFFH.

6.2.4. Task-Gate Descriptor

A task-gate descriptor provides an indirect, protected reference to a task. Figure 6-5 shows the
format of a task-gate descriptor. A task-gate descriptor can be placed in the GDT, an LDT, or the
IDT.
The TSS segment selector field in a task-gate descriptor points to a TSS descriptor in the GDT.
The RPL in this segment selector is not used.
The DPL of a task-gate descriptor controls access to the TSS descriptor during a task switch.
When a program or procedure makes a call or jump to a task through a task gate, the CPL and
the RPL field of the gate selector pointing to the task gate must be less than or equal to the DPL
of the task-gate descriptor. (Note that when a task gate is used, the DPL of the destination TSS
descriptor is not used.)

6-8
 TASK MANAGEMENT

TSS
+

Visible Part Invisible Part

Task
Register Selector Base Address Segment Limit

GDT

TSS Descriptor

Figure 6-4. Task Register

31 16 15 14 13 12 11 8 7 0

D Type
P P 4
L 0 0 1 0 1

31 16 15 0

TSS Segment Selector 0

DPL Descriptor Privilege Level

P Segment Present
TYPE Segment Type

Reserved

Figure 6-5. Task-Gate Descriptor

6-9
TASK MANAGEMENT

A task can be accessed either through a task-gate descriptor or a TSS descriptor. Both of these
structures are provided to satisfy the following needs:
• The need for a task to have only one busy flag. Because the busy flag for a task is stored in
the TSS descriptor, each task should have only one TSS descriptor. There may, however,
be several task gates that reference the same TSS descriptor.
• The need to provide selective access to tasks. Task gates fill this need, because they can
reside in an LDT and can have a DPL that is different from the TSS descriptor’s DPL. A
program or procedure that does not have sufficient privilege to access the TSS descriptor
for a task in the GDT (which usually has a DPL of 0) may be allowed access to the task
through a task gate with a higher DPL. Task gates give the operating system greater
latitude for limiting access to specific tasks.
• The need for an interrupt or exception to be handled by an independent task. Task gates
may also reside in the IDT, which allows interrupts and exceptions to be handled by
handler tasks. When an interrupt or exception vector points to a task gate, the processor
switches to the specified task.
Figure 6-6 illustrates how a task gate in an LDT, a task gate in the GDT, and a task gate in the
IDT can all point to the same task.

6.3. TASK SWITCHING

The processor transfers execution to another task in any of four cases:
• The current program, task, or procedure executes a JMP or CALL instruction to a TSS
descriptor in the GDT.
• The current program, task, or procedure executes a JMP or CALL instruction to a task-gate
descriptor in the GDT or the current LDT.
• An interrupt or exception vector points to a task-gate descriptor in the IDT.
• The current task executes an IRET when the NT flag in the EFLAGS register is set.
The JMP, CALL, and IRET instructions, as well as interrupts and exceptions, are all generalized
mechanisms for redirecting a program. The referencing of a TSS descriptor or a task gate (when
calling or jumping to a task) or the state of the NT flag (when executing an IRET instruction)
determines whether a task switch occurs.
The processor performs the following operations when switching to a new task:
1. Obtains the TSS segment selector for the new task as the operand of the JMP or CALL
instruction, from a task gate, or from the previous task link field (for a task switch initiated
with an IRET instruction).

6-10
 TASK MANAGEMENT

LDT GDT TSS

Task Gate

Task Gate TSS Descriptor

IDT

Task Gate

Figure 6-6. Task Gates Referencing the Same Task

2. Checks that the current (old) task is allowed to switch to the new task. Data-access
privilege rules apply to JMP and CALL instructions. The CPL of the current (old) task and
the RPL of the segment selector for the new task must be less than or equal to the DPL of
the TSS descriptor or task gate being referenced. Exceptions, interrupts (except for
interrupts generated by the INT n instruction), and the IRET instruction are permitted to
switch tasks regardless of the DPL of the destination task-gate or TSS descriptor. For
interrupts generated by the INT n instruction, the DPL is checked.
3. Checks that the TSS descriptor of the new task is marked present and has a valid limit
(greater than or equal to 67H).
4. Checks that the new task is available (call, jump, exception, or interrupt) or busy (IRET
return).

6-11
TASK MANAGEMENT

5. Checks that the current (old) TSS, new TSS, and all segment descriptors used in the task
switch are paged into system memory.
6. If the task switch was initiated with a JMP or IRET instruction, the processor clears the
busy (B) flag in the current (old) task’s TSS descriptor; if initiated with a CALL
instruction, an exception, or an interrupt, the busy (B) flag is left set. (Refer to Table 6-2.)
7. If the task switch was initiated with an IRET instruction, the processor clears the NT flag
in a temporarily saved image of the EFLAGS register; if initiated with a CALL or JMP
instruction, an exception, or an interrupt, the NT flag is left unchanged in the saved
EFLAGS image.
8. Saves the state of the current (old) task in the current task’s TSS. The processor finds the
base address of the current TSS in the task register and then copies the states of the
following registers into the current TSS: all the general-purpose registers, segment
selectors from the segment registers, the temporarily saved image of the EFLAGS register,
and the instruction pointer register (EIP).

NOTE
At this point, if all checks and saves have been carried out successfully, the
processor commits to the task switch. If an unrecoverable error occurs in
steps 1 through 8, the processor does not complete the task switch and insures
that the processor is returned to its state prior to the execution of the
instruction that initiated the task switch. If an unrecoverable error occurs after
the commit point (in steps 9 through 14), the processor completes the task
switch (without performing additional access and segment availability
checks) and generates the appropriate exception prior to beginning execution
of the new task. If exceptions occur after the commit point, the exception
handler must finish the task switch itself before allowing the processor to
begin executing the task. Refer to Chapter 5, Interrupt and Exception
Handling for more information about the affect of exceptions on a task when
they occur after the commit point of a task switch.

9. If the task switch was initiated with a CALL instruction, an exception, or an interrupt, the
processor sets the NT flag in the EFLAGS image stored in the new task’s TSS; if initiated
with an IRET instruction, the processor restores the NT flag from the EFLAGS image
stored on the stack. If initiated with a JMP instruction, the NT flag is left unchanged.
(Refer to Table 6-2.)
10. If the task switch was initiated with a CALL instruction, JMP instruction, an exception, or
an interrupt, the processor sets the busy (B) flag in the new task’s TSS descriptor; if
initiated with an IRET instruction, the busy (B) flag is left set.
11. Sets the TS flag in the control register CR0 image stored in the new task’s TSS.
12. Loads the task register with the segment selector and descriptor for the new task's TSS.

6-12
 TASK MANAGEMENT

13. Loads the new task’s state from its TSS into processor. Any errors associated with the
loading and qualification of segment descriptors in this step occur in the context of the new
task. The task state information that is loaded here includes the LDTR register, the PDBR
(control register CR3), the EFLAGS register, the EIP register, the general-purpose
registers, and the segment descriptor parts of the segment registers.
14. Begins executing the new task. (To an exception handler, the first instruction of the new
task appears not to have been executed.)
The state of the currently executing task is always saved when a successful task switch occurs.
If the task is resumed, execution starts with the instruction pointed to by the saved EIP value,
and the registers are restored to the values they held when the task was suspended.
When switching tasks, the privilege level of the new task does not inherit its privilege level from
the suspended task. The new task begins executing at the privilege level specified in the CPL
field of the CS register, which is loaded from the TSS. Because tasks are isolated by their sepa-
rate address spaces and TSSs and because privilege rules control access to a TSS, software does
not need to perform explicit privilege checks on a task switch.
Table 6-1 shows the exception conditions that the processor checks for when switching tasks. It
also shows the exception that is generated for each check if an error is detected and the segment
that the error code references. (The order of the checks in the table is the order used in the P6
family processors. The exact order is model specific and may be different for other Intel Archi-
tecture processors.) Exception handlers designed to handle these exceptions may be subject to
recursive calls if they attempt to reload the segment selector that generated the exception. The
cause of the exception (or the first of multiple causes) should be fixed before reloading the
selector.

Table 6-1. Exception Conditions Checked During a Task Switch

Error Code
Condition Checked Exception1 Reference2
Segment selector for a TSS descriptor references #GP New Task’s TSS
the GDT and is within the limits of the table.
TSS descriptor is present in memory. #NP New Task’s TSS
TSS descriptor is not busy (for task switch initiated by a #GP (for JMP, CALL, Task’s back-link TSS
call, interrupt, or exception). INT)
TSS descriptor is not busy (for task switch initiated by #TS (for IRET) New Task’s TSS
an IRET instruction).
TSS segment limit greater than or equal to 108 (for 32- #TS New Task’s TSS
bit TSS) or 44 (for 16-bit TSS).
Registers are loaded from the values in the TSS.
LDT segment selector of new task is valid 3. #TS New Task’s LDT
Code segment DPL matches segment selector RPL. #TS New Code Segment
2
SS segment selector is valid . #TS New Stack Segment
Stack segment is present in memory. #SF New Stack Segment

6-13
TASK MANAGEMENT

Table 6-1. Exception Conditions Checked During a Task Switch (Contd.)

Stack segment DPL matches CPL. #TS New stack segment
LDT of new task is present in memory. #TS New Task’s LDT
3
CS segment selector is valid . #TS New Code Segment
Code segment is present in memory. #NP New Code Segment
Stack segment DPL matches selector RPL. #TS New Stack Segment
3
DS, ES, FS, and GS segment selectors are valid . #TS New Data Segment
DS, ES, FS, and GS segments are readable. #TS New Data Segment
DS, ES, FS, and GS segments are present in memory. #NP New Data Segment
DS, ES, FS, and GS segment DPL greater than or #TS New Data Segment
equal to CPL (unless these are conforming segments).

NOTES:
1. #NP is segment-not-present exception, #GP is general-protection exception, #TS is invalid-TSS excep-
tion, and #SF is stack-fault exception.
2. The error code contains an index to the segment descriptor referenced in this column.
3. A segment selector is valid if it is in a compatible type of table (GDT or LDT), occupies an address within
the table’s segment limit, and refers to a compatible type of descriptor (for example, a segment selector in
the CS register only is valid when it points to a code-segment descriptor).

The TS (task switched) flag in the control register CR0 is set every time a task switch occurs.
System software uses the TS flag to coordinate the actions of floating-point unit when gener-
ating floating-point exceptions with the rest of the processor. The TS flag indicates that the
context of the floating-point unit may be different from that of the current task. Refer to Section
2.5., “Control Registers” in Chapter 2, System Architecture Overview for a detailed description
of the function and use of the TS flag.

6.4. TASK LINKING

The previous task link field of the TSS (sometimes called the “backlink”) and the NT flag in the
EFLAGS register are used to return execution to the previous task. The NT flag indicates
whether the currently executing task is nested within the execution of another task, and the
previous task link field of the current task's TSS holds the TSS selector for the higher-level task
in the nesting hierarchy, if there is one (refer to Figure 6-7).
When a CALL instruction, an interrupt, or an exception causes a task switch, the processor
copies the segment selector for the current TSS into the previous task link field of the TSS for
the new task, and then sets the NT flag in the EFLAGS register. The NT flag indicates that the
previous task link field of the TSS has been loaded with a saved TSS segment selector. If soft-
ware uses an IRET instruction to suspend the new task, the processor uses the value in the
previous task link field and the NT flag to return to the previous task; that is, if the NT flag is
set, the processor performs a task switch to the task specified in the previous task link field.

6-14
 TASK MANAGEMENT

NOTE
When a JMP instruction causes a task switch, the new task is not nested; that
is, the NT flag is set to 0 and the previous task link field is not used. A JMP
instruction is used to dispatch a new task when nesting is not desired.

Top Level Nested More Deeply Currently Executing

Task Task Nested Task Task
TSS TSS TSS EFLAGS
NT=1

NT=0 NT=1 NT=1

Prev. Task Link Prev. Task Link Prev. Task Link Task Register

Figure 6-7. Nested Tasks

Table 6-2 summarizes the uses of the busy flag (in the TSS segment descriptor), the NT flag, the
previous task link field, and TS flag (in control register CR0) during a task switch. Note that the
NT flag may be modified by software executing at any privilege level. It is possible for a
program to set its NT flag and execute an IRET instruction, which would have the effect of
invoking the task specified in the previous link field of the current task’s TSS. To keep spurious
task switches from succeeding, the operating system should initialize the previous task link field
for every TSS it creates to 0.

Table 6-2. Effect of a Task Switch on Busy Flag, NT Flag, Previous Task Link Field,
and TS Flag
Effect of CALL
Effect of JMP Instruction or Effect of IRET
Flag or Field instruction Interrupt Instruction
Busy (B) flag of new Flag is set. Must have Flag is set. Must have No change. Must have
task. been clear before. been clear before. been set.
Busy flag of old task. Flag is cleared. No change. Flag is Flag is cleared.
currently set.
NT flag of new task. No change. Flag is set. Restored to value from
TSS of new task.
NT flag of old task. No change. No change. Flag is cleared.
Previous task link field of No change. Loaded with selector No change.
new task. for old task’s TSS.
Previous task link field of No change. No change. No change.
old task.
TS flag in control Flag is set. Flag is set. Flag is set.
register CR0.

6-15
TASK MANAGEMENT

6.4.1. Use of Busy Flag To Prevent Recursive Task Switching

A TSS allows only one context to be saved for a task; therefore, once a task is called
(dispatched), a recursive (or re-entrant) call to the task would cause the current state of the task
to be lost. The busy flag in the TSS segment descriptor is provided to prevent re-entrant task
switching and subsequent loss of task state information. The processor manages the busy flag as
follows:
1. When dispatching a task, the processor sets the busy flag of the new task.
2. If during a task switch, the current task is placed in a nested chain (the task switch is being
generated by a CALL instruction, an interrupt, or an exception), the busy flag for the
current task remains set.
3. When switching to the new task (initiated by a CALL instruction, interrupt, or exception),
the processor generates a general-protection exception (#GP) if the busy flag of the new
task is already set. (If the task switch is initiated with an IRET instruction, the exception is
not raised because the processor expects the busy flag to be set.)
4. When a task is terminated by a jump to a new task (initiated with a JMP instruction in the
task code) or by an IRET instruction in the task code, the processor clears the busy flag,
returning the task to the “not busy” state.
In this manner the processor prevents recursive task switching by preventing a task from
switching to itself or to any task in a nested chain of tasks. The chain of nested suspended tasks
may grow to any length, due to multiple calls, interrupts, or exceptions. The busy flag prevents
a task from being invoked if it is in this chain.
The busy flag may be used in multiprocessor configurations, because the processor follows a
LOCK protocol (on the bus or in the cache) when it sets or clears the busy flag. This lock keeps
two processors from invoking the same task at the same time. (Refer to Section 7.1.2.1., “Auto-
matic Locking” in Chapter 7, Multiple-Processor Management for more information about
setting the busy flag in a multiprocessor applications.)

6.4.2. Modifying Task Linkages

In a uniprocessor system, in situations where it is necessary to remove a task from a chain of
linked tasks, use the following procedure to remove the task:
1. Disable interrupts.
2. Change the previous task link field in the TSS of the pre-empting task (the task that
suspended the task to be removed). It is assumed that the pre-empting task is the next task
(newer task) in the chain from the task to be removed. Change the previous task link field
should to point to the TSS of the next oldest or to an even older task in the chain.
3. Clear the busy (B) flag in the TSS segment descriptor for the task being removed from the
chain. If more than one task is being removed from the chain, the busy flag for each task
being remove must be cleared.
4. Enable interrupts.

6-16
 TASK MANAGEMENT

In a multiprocessing system, additional synchronization and serialization operations must be

added to this procedure to insure that the TSS and its segment descriptor are both locked when
the previous task link field is changed and the busy flag is cleared.

6.5. TASK ADDRESS SPACE

The address space for a task consists of the segments that the task can access. These segments
include the code, data, stack, and system segments referenced in the TSS and any other segments
accessed by the task code. These segments are mapped into the processor’s linear address space,
which is in turn mapped into the processor’s physical address space (either directly or through
paging).
The LDT segment field in the TSS can be used to give each task its own LDT. Giving a task its
own LDT allows the task address space to be isolated from other tasks by placing the segment
descriptors for all the segments associated with the task in the task’s LDT.
It also is possible for several tasks to use the same LDT. This is a simple and memory-efficient
way to allow some tasks to communicate with or control each other, without dropping the
protection barriers for the entire system.
Because all tasks have access to the GDT, it also is possible to create shared segments accessed
through segment descriptors in this table.
If paging is enabled, the CR3 register (PDBR) field in the TSS allows each task can also have
its own set of page tables for mapping linear addresses to physical addresses. Or, several tasks
can share the same set of page tables.

6.5.1. Mapping Tasks to the Linear and Physical Address

Spaces
Tasks can be mapped to the linear address space and physical address space in either of two
ways:
• One linear-to-physical address space mapping is shared among all tasks. When paging is
not enabled, this is the only choice. Without paging, all linear addresses map to the same
physical addresses. When paging is enabled, this form of linear-to-physical address space
mapping is obtained by using one page directory for all tasks. The linear address space
may exceed the available physical space if demand-paged virtual memory is supported.
• Each task has its own linear address space that is mapped to the physical address space.
This form of mapping is accomplished by using a different page directory for each task.
Because the PDBR (control register CR3) is loaded on each task switch, each task may
have a different page directory.
The linear address spaces of different tasks may map to completely distinct physical addresses.
If the entries of different page directories point to different page tables and the page tables point
to different pages of physical memory, then the tasks do not share any physical addresses.

6-17
TASK MANAGEMENT

With either method of mapping task linear address spaces, the TSSs for all tasks must lie in a
shared area of the physical space, which is accessible to all tasks. This mapping is required so
that the mapping of TSS addresses does not change while the processor is reading and updating
the TSSs during a task switch. The linear address space mapped by the GDT also should be
mapped to a shared area of the physical space; otherwise, the purpose of the GDT is defeated.
Figure 6-8 shows how the linear address spaces of two tasks can overlap in the physical space
by sharing page tables.

TSS Page Directories Page Tables Page Frames

Task A
Task A TSS Page

Task A
PTE Page
PTE
PDBR PDE PTE Task A
PDE Page
Shared PT
Shared
Page
PTE
PTE Shared
Task B TSS Page

Task B
Page
PDBR PDE PTE
PDE PTE Task B
Page

Figure 6-8. Overlapping Linear-to-Physical Mappings

6.5.2. Task Logical Address Space

To allow the sharing of data among tasks, use any of the following techniques to create shared
logical-to-physical address-space mappings for data segments:
• Through the segment descriptors in the GDT. All tasks must have access to the segment
descriptors in the GDT. If some segment descriptors in the GDT point to segments in the
linear-address space that are mapped into an area of the physical-address space common to
all tasks, then all tasks can share the data and code in those segments.
• Through a shared LDT. Two or more tasks can use the same LDT if the LDT fields in their
TSSs point to the same LDT. If some segment descriptors in a shared LDT point to
segments that are mapped to a common area of the physical address space, the data and
code in those segments can be shared among the tasks that share the LDT. This method of
sharing is more selective than sharing through the GDT, because the sharing can be limited

6-18
 TASK MANAGEMENT

to specific tasks. Other tasks in the system may have different LDTs that do not give them
access to the shared segments.
• Through segment descriptors in distinct LDTs that are mapped to common addresses in the
linear address space. If this common area of the linear address space is mapped to the same
area of the physical address space for each task, these segment descriptors permit the tasks
to share segments. Such segment descriptors are commonly called aliases. This method of
sharing is even more selective than those listed above, because, other segment descriptors
in the LDTs may point to independent linear addresses which are not shared.

6.6. 16-BIT TASK-STATE SEGMENT (TSS)

The 32-bit Intel Architecture processors also recognize a 16-bit TSS format like the one used in
Intel 286 processors (refer to Figure 6-9). It is supported for compatibility with software written
to run on these earlier Intel Architecture processors.
The following additional information is important to know about the 16-bit TSS.
• Do not use a 16-bit TSS to implement a virtual-8086 task.
• The valid segment limit for a 16-bit TSS is 2CH.
• The 16-bit TSS does not contain a field for the base address of the page directory, which is
loaded into control register CR3. Therefore, a separate set of page tables for each task is
not supported for 16-bit tasks. If a 16-bit task is dispatched, the page-table structure for the
previous task is used.
• The I/O base address is not included in the 16-bit TSS, so none of the functions of the I/O
map are supported.
• When task state is saved in a 16-bit TSS, the upper 16 bits of the EFLAGS register and the
EIP register are lost.
• When the general-purpose registers are loaded or saved from a 16-bit TSS, the upper 16
bits of the registers are modified and not maintained.

6-19
TASK MANAGEMENT

15 0
Task LDT Selector 42
DS Selector 40
SS Selector 38
CS Selector 36
ES Selector 34
DI 32
SI 30
BP 28
SP 26
BX 24
DX 22
CX 20
AX 18
FLAG Word 16
IP (Entry Point) 14
SS2 12
SP2 10
SS1 8
SP1 6
SS0 4
SP0 2
Previous Task Link 0

Figure 6-9. 16-Bit TSS Format

6-20
 7
Multiple-Processor
Management
 CHAPTER 7
MULTIPLE-PROCESSOR MANAGEMENT

The Intel Architecture provides several mechanisms for managing and improving the perfor-
mance of multiple processors connected to the same system bus. These mechanisms include:
• Bus locking and/or cache coherency management for performing atomic operations on
system memory.
• Serializing instructions. (These instructions apply only to the Pentium® and P6 family
processors.)
• Advance programmable interrupt controller (APIC) located on the processor chip. (The
APIC architecture was introduced into the Intel Architecture with the Pentium ® processor.)
• A secondary (level 2, L2) cache. For the P6 family processors, the L2 cache is included in
the processor package and is tightly coupled to the processor. For the Pentium® and
Intel486™ processors, pins are provided to support an external L2 cache.
These mechanisms are particularly useful in symmetric-multiprocessing systems; however, they
can also be used in applications where a Intel Architecture processor and a special-purpose
processor (such as a communications, graphics, or video processor) share the system bus.
The main goals of these multiprocessing mechanisms are as follows:
• To maintain system memory coherency—When two or more processors are attempting
simultaneously to access the same address in system memory, some communication
mechanism or memory access protocol must be available to promote data coherency and,
in some instances, to allow one processor to temporarily lock a memory location.
• To maintain cache consistency—When one processor accesses data cached in another
processor, it must not receive incorrect data. If it modifies data, all other processors that
access that data must receive the modified data.
• To allow predictable ordering of writes to memory—In some circumstances, it is important
that memory writes be observed externally in precisely the same order as programmed.
• To distribute interrupt handling among a group of processors—When several processors
are operating in a system in parallel, it is useful to have a centralized mechanism for
receiving interrupts and distributing them to available processors for servicing.
The Intel Architecture’s caching mechanism and cache consistency are discussed in Chapter 9,
Memory Cache Control. Bus and memory locking, serializing instructions, memory ordering,
and the processor’s internal APIC are discussed in the following sections.

7-1
MULTIPLE-PROCESSOR MANAGEMENT

7.1. LOCKED ATOMIC OPERATIONS

The 32-bit Intel Architecture processors support locked atomic operations on locations in
system memory. These operations are typically used to manage shared data structures (such as
semaphores, segment descriptors, system segments, or page tables) in which two or more
processors may try simultaneously to modify the same field or flag. The processor uses three
interdependent mechanisms for carrying out locked atomic operations:
• Guaranteed atomic operations.
• Bus locking, using the LOCK# signal and the LOCK instruction prefix.
• Cache coherency protocols that insure that atomic operations can be carried out on cached
data structures (cache lock). This mechanism is present in the P6 family processors.
These mechanisms are interdependent in the following ways. Certain basic memory transactions
(such as reading or writing a byte in system memory) are always guaranteed to be handled atom-
ically. That is, once started, the processor guarantees that the operation will be completed before
another processor or bus agent is allowed access to the memory location. The processor also
supports bus locking for performing selected memory operations (such as a read-modify-write
operation in a shared area of memory) that typically need to be handled atomically, but are not
automatically handled this way. Because frequently used memory locations are often cached in
a processor’s L1 or L2 caches, atomic operations can often be carried out inside a processor’s
caches without asserting the bus lock. Here the processor’s cache coherency protocols insure
that other processors that are caching the same memory locations are managed properly while
atomic operations are performed on cached memory locations.
Note that the mechanisms for handling locked atomic operations have evolved as the complexity
of Intel Architecture processors has evolved. As such, more recent Intel Architecture processors
(such as the P6 family processors) provide a more refined locking mechanism than earlier Intel
Architecture processors, as is described in the following sections.

7.1.1. Guaranteed Atomic Operations

The Intel386™, Intel486™, Pentium®, and P6 family processors guarantee that the following
basic memory operations will always be carried out atomically:
• Reading or writing a byte.
• Reading or writing a word aligned on a 16-bit boundary.
• Reading or writing a doubleword aligned on a 32-bit boundary.
The P6 family processors guarantee that the following additional memory operations will
always be carried out atomically:
• Reading or writing a quadword aligned on a 64-bit boundary. (This operation is also
guaranteed on the Pentium® processor.)
• 16-bit accesses to uncached memory locations that fit within a 32-bit data bus.
• 16-, 32-, and 64-bit accesses to cached memory that fit within a 32-Byte cache line.

7-2
 MULTIPLE-PROCESSOR MANAGEMENT

Accesses to cacheable memory that are split across bus widths, cache lines, and page boundaries
are not guaranteed to be atomic by the Intel486™, Pentium®, or P6 family processors. The P6
family processors provide bus control signals that permit external memory subsystems to make
split accesses atomic; however, nonaligned data accesses will seriously impact the performance
of the processor and should be avoided where possible.

7.1.2. Bus Locking

Intel Architecture processors provide a LOCK# signal that is asserted automatically during
certain critical memory operations to lock the system bus. While this output signal is asserted,
requests from other processors or bus agents for control of the bus are blocked. Software can
specify other occasions when the LOCK semantics are to be followed by prepending the LOCK
prefix to an instruction.
In the case of the Intel386™, Intel486™, and Pentium® processors, explicitly locked instruc-
tions will result in the assertion of the LOCK# signal. It is the responsibility of the hardware
designer to make the LOCK# signal available in system hardware to control memory accesses
among processors.
For the P6 family processors, if the memory area being accessed is cached internally in the
processor, the LOCK# signal is generally not asserted; instead, locking is only applied to the
processor’s caches (refer to Section 7.1.4., “Effects of a LOCK Operation on Internal Processor
Caches”).

7.1.2.1. AUTOMATIC LOCKING

The operations on which the processor automatically follows the LOCK semantics are as
follows:
• When executing an XCHG instruction that references memory.
• When setting the B (busy) flag of a TSS descriptor. The processor tests and sets the busy
flag in the type field of the TSS descriptor when switching to a task. To insure that two
processors do not switch to the same task simultaneously, the processor follows the LOCK
semantics while testing and setting this flag.
• When updating segment descriptors. When loading a segment descriptor, the processor
will set the accessed flag in the segment descriptor if the flag is clear. During this
operation, the processor follows the LOCK semantics so that the descriptor will not be
modified by another processor while it is being updated. For this action to be effective,
operating-system procedures that update descriptors should use the following steps:
— Use a locked operation to modify the access-rights byte to indicate that the segment
descriptor is not-present, and specify a value for the type field that indicates that the
descriptor is being updated.
— Update the fields of the segment descriptor. (This operation may require several
memory accesses; therefore, locked operations cannot be used.)

7-3
MULTIPLE-PROCESSOR MANAGEMENT

— Use a locked operation to modify the access-rights byte to indicate that the segment
descriptor is valid and present.
Note that the Intel386™ processor always updates the accessed flag in the segment
descriptor, whether it is clear or not. The P6 family, Pentium®, and Intel486™ processors
only update this flag if it is not already set.
• When updating page-directory and page-table entries. When updating page-directory
and page-table entries, the processor uses locked cycles to set the accessed and dirty flag in
the page-directory and page-table entries.
• Acknowledging interrupts. After an interrupt request, an interrupt controller may use the
data bus to send the interrupt vector for the interrupt to the processor. The processor
follows the LOCK semantics during this time to ensure that no other data appears on the
data bus when the interrupt vector is being transmitted.

7.1.2.2. SOFTWARE CONTROLLED BUS LOCKING

To explicitly force the LOCK semantics, software can use the LOCK prefix with the following
instructions when they are used to modify a memory location. An invalid-opcode exception
(#UD) is generated when the LOCK prefix is used with any other instruction or when no write
operation is made to memory (that is, when the destination operand is in a register).
• The bit test and modify instructions (BTS, BTR, and BTC).
• The exchange instructions (XADD, CMPXCHG, and CMPXCHG8B).
• The LOCK prefix is automatically assumed for XCHG instruction.
• The following single-operand arithmetic and logical instructions: INC, DEC, NOT, and
NEG.
• The following two-operand arithmetic and logical instructions: ADD, ADC, SUB, SBB,
AND, OR, and XOR.
A locked instruction is guaranteed to lock only the area of memory defined by the destination
operand, but may be interpreted by the system as a lock for a larger memory area.
Software should access semaphores (shared memory used for signaling between multiple
processors) using identical addresses and operand lengths. For example, if one processor
accesses a semaphore using a word access, other processors should not access the semaphore
using a byte access.
The integrity of a bus lock is not affected by the alignment of the memory field. The LOCK
semantics are followed for as many bus cycles as necessary to update the entire operand.
However, it is recommend that locked accesses be aligned on their natural boundaries for better
system performance:
• Any boundary for an 8-bit access (locked or otherwise).
• 16-bit boundary for locked word accesses.
• 32-bit boundary for locked doubleword access.

7-4
 MULTIPLE-PROCESSOR MANAGEMENT

• 64-bit boundary for locked quadword access.

Locked operations are atomic with respect to all other memory operations and all externally
visible events. Only instruction fetch and page table accesses can pass locked instructions.
Locked instructions can be used to synchronize data written by one processor and read by
another processor.
For the P6 family processors, locked operations serialize all outstanding load and store opera-
tions (that is, wait for them to complete).
Locked instructions should not be used to insure that data written can be fetched as instructions.

NOTE
The locked instructions for the current versions of the Intel486™, Pentium®,
and P6 family processors will allow data written to be fetched as instructions.
However, Intel recommends that developers who require the use of self-
modifying code use a different synchronizing mechanism, described in the
following sections.

7.1.3. Handling Self- and Cross-Modifying Code

The act of a processor writing data into a currently executing code segment with the intent of
executing that data as code is called self-modifying code. Intel Architecture processors exhibit
model-specific behavior when executing self-modified code, depending upon how far ahead of
the current execution pointer the code has been modified. As processor architectures become
more complex and start to speculatively execute code ahead of the retirement point (as in the P6
family processors), the rules regarding which code should execute, pre- or post-modification,
become blurred. To write self-modifying code and ensure that it is compliant with current and
future Intel Architectures one of the following two coding options should be chosen.
(* OPTION 1 *)
Store modified code (as data) into code segment;
Jump to new code or an intermediate location;
Execute new code;

(* OPTION 2 *)
Store modified code (as data) into code segment;
Execute a serializing instruction; (* For example, CPUID instruction *)
Execute new code;
(The use of one of these options is not required for programs intended to run on the Pentium® or
Intel486™ processors, but are recommended to insure compatibility with the P6 family proces-
sors.)
It should be noted that self-modifying code will execute at a lower level of performance than
nonself-modifying or normal code. The degree of the performance deterioration will depend
upon the frequency of modification and specific characteristics of the code.

7-5
MULTIPLE-PROCESSOR MANAGEMENT

The act of one processor writing data into the currently executing code segment of a second
processor with the intent of having the second processor execute that data as code is called
cross-modifying code. As with self-modifying code, Intel Architecture processors exhibit
model-specific behavior when executing cross-modifying code, depending upon how far ahead
of the executing processors current execution pointer the code has been modified. To write
cross-modifying code and insure that it is compliant with current and future Intel Architectures,
the following processor synchronization algorithm should be implemented.
; Action of Modifying Processor
Store modified code (as data) into code segment;
Memory_Flag ← 1;

; Action of Executing Processor

WHILE (Memory_Flag ≠ 1)
Wait for code to update;
ELIHW;
Execute serializing instruction; (* For example, CPUID instruction *)
Begin executing modified code;
(The use of this option is not required for programs intended to run on the Intel486™ processor,
but is recommended to insure compatibility with the Pentium®, and P6 family processors.)
Like self-modifying code, cross-modifying code will execute at a lower level of performance
than noncross-modifying (normal) code, depending upon the frequency of modification and
specific characteristics of the code.

7.1.4. Effects of a LOCK Operation on Internal Processor

Caches
For the Intel486™ and Pentium® processors, the LOCK# signal is always asserted on the bus
during a LOCK operation, even if the area of memory being locked is cached in the processor.
For the P6 family processors, if the area of memory being locked during a LOCK operation is
cached in the processor that is performing the LOCK operation as write-back memory and is
completely contained in a cache line, the processor may not assert the LOCK# signal on the bus.
Instead, it will modify the memory location internally and allow it’s cache coherency mecha-
nism to insure that the operation is carried out atomically. This operation is called “cache
locking.” The cache coherency mechanism automatically prevents two or more processors that
have cached the same area of memory from simultaneously modifying data in that area.

7.2. MEMORY ORDERING

The term memory ordering refers to the order in which the processor issues reads (loads) and
writes (stores) out onto the bus to system memory. The Intel Architecture supports several
memory ordering models depending on the implementation of the architecture. For example, the
Intel386™ processor enforces program ordering (generally referred to as strong ordering),

7-6
 MULTIPLE-PROCESSOR MANAGEMENT

where reads and writes are issued on the system bus in the order they occur in the instruction
stream under all circumstances.
To allow optimizing of instruction execution, the Intel Architecture allows departures from
strong-ordering model called processor ordering in P6-family processors. These processor-
ordering variations allow performance enhancing operations such as allowing reads to go ahead
of writes by buffering writes. The goal of any of these variations is to increase instruction execu-
tion speeds, while maintaining memory coherency, even in multiple-processor systems.
The following sections describe the memory ordering models used by the Intel486™, Pentium®,
and P6 family processors.

7.2.1. Memory Ordering in the Pentium® and Intel486™

Processors
The Pentium® and Intel486™ processors follow the processor-ordered memory model;
however, they operate as strongly-ordered processors under most circumstances. Reads and
writes always appear in programmed order at the system bus—except for the following situation
where processor ordering is exhibited. Read misses are permitted to go ahead of buffered writes
on the system bus when all the buffered writes are cache hits and, therefore, are not directed to
the same address being accessed by the read miss.
In the case of I/O operations, both reads and writes always appear in programmed order.
Software intended to operate correctly in processor-ordered processors (such as the P6 family
processors) should not depend on the relatively strong ordering of the Pentium® or Intel486™
processors. Instead, it should insure that accesses to shared variables that are intended to control
concurrent execution among processors are explicitly required to obey program ordering
through the use of appropriate locking or serializing operations (refer to Section 7.2.4.,
“Strengthening or Weakening the Memory Ordering Model”).

7.2.2. Memory Ordering in the P6 Family Processors

The P6 family processors also use a processor-ordered memory ordering model that can be
further refined defined as “write ordered with store-buffer forwarding.” This model can be char-
acterized as follows.
In a single-processor system for memory regions defined as write-back cacheable, the following
ordering rules apply:
1. Reads can be carried out speculatively and in any order.
2. Reads can pass buffered writes, but the processor is self-consistent.
3. Writes to memory are always carried out in program order.
4. Writes can be buffered.
5. Writes are not performed speculatively; they are only performed for instructions that have
actually been retired.

7-7
MULTIPLE-PROCESSOR MANAGEMENT

6. Data from buffered writes can be forwarded to waiting reads within the processor.
7. Reads or writes cannot pass (be carried out ahead of) I/O instructions, locked instructions,
or serializing instructions.
The second rule allows a read to pass a write. However, if the write is to the same memory loca-
tion as the read, the processor’s internal “snooping” mechanism will detect the conflict and
update the already cached read before the processor executes the instruction that uses the value.
The sixth rule constitutes an exception to an otherwise write ordered model.
In a multiple-processor system, the following ordering rules apply:
• Individual processors use the same ordering rules as in a single-processor system.
• Writes by a single processor are observed in the same order by all processors.
• Writes from the individual processors on the system bus are globally observed and are
NOT ordered with respect to each other.
The latter rule can be clarified by the example in Figure 7-1. Consider three processors in a
system and each processor performs three writes, one to each of three defined locations (A, B,
and C). Individually, the processors perform the writes in the same program order, but because
of bus arbitration and other memory access mechanisms, the order that the three processors write
the individual memory locations can differ each time the respective code sequences are executed
on the processors. The final values in location A, B, and C would possibly vary on each execu-
tion of the write sequence.

Order of Writes From Individual Processors

Processor #1 Processor #2 Processor #3
Each processor
Write A.1 Write A.2 Write A.3
is guaranteed to
perform writes Write B.1 Write B.2 Write B.3
in program order. Write C.1 Write C.2 Write C.3

Example of Order of Actual Writes

From All Processors to Memory
Writes are in order Write A.1
with respect to Write B.1
individual processors. Write A.2 Writes from all
Write A.3 processors are
Write C.1 not guaranteed
Write B.2 to occur in a
Write C.2 particular order.
Write B.3
Write C.3

Figure 7-1. Example of Write Ordering in Multiple-Processor Systems

7-8
 MULTIPLE-PROCESSOR MANAGEMENT

The processor-ordering model described in this section is virtually identical to that used by the
Pentium® and Intel486™ processors. The only enhancements in the P6 family processors are:
• Added support for speculative reads.
• Store-buffer forwarding, when a read passes a write to the same memory location.
• Out of order store from long string store and string move operations (refer to Section
7.2.3., “Out of Order Stores From String Operations in P6 Family Processors” below).

7.2.3. Out of Order Stores From String Operations in P6 Family

Processors
The P6 family processors modify the processors operation during the string store operations
(initiated with the MOVS and STOS instructions) to maximize performance. Once the “fast
string” operations initial conditions are met (as described below), the processor will essentially
operate on, from an external perspective, the string in a cache line by cache line mode. This
results in the processor looping on issuing a cache-line read for the source address and an inval-
idation on the external bus for the destination address, knowing that all bytes in the destination
cache line will be modified, for the length of the string. In this mode interrupts will only be
accepted by the processor on cache line boundaries. It is possible in this mode that the destina-
tion line invalidations, and therefore stores, will be issued on the external bus out of order.
Code dependent upon sequential store ordering should not use the string operations for the entire
data structure to be stored. Data and semaphores should be separated. Order dependent code
should use a discrete semaphore uniquely stored to after any string operations to allow correctly
ordered data to be seen by all processors.
Initial conditions for “fast string” operations:
• Source and destination addresses must be 8-byte aligned.
• String operation must be performed in ascending address order.
• The initial operation counter (ECX) must be equal to or greater than 64.
• Source and destination must not overlap by less than a cache line (32 bytes).
• The memory type for both source and destination addresses must be either WB or WC.

7.2.4. Strengthening or Weakening the Memory Ordering Model

The Intel Architecture provides several mechanisms for strengthening or weakening the
memory ordering model to handle special programming situations. These mechanisms include:
• The I/O instructions, locking instructions, the LOCK prefix, and serializing instructions
force stronger ordering on the processor.
• The memory type range registers (MTRRs) can be used to strengthen or weaken memory
ordering for specific area of physical memory (refer to Section 9.12., “Memory Type

7-9
MULTIPLE-PROCESSOR MANAGEMENT

Range Registers (MTRRs)”, in Chapter 9, Memory Cache Control). MTRRs are available
only in the P6 family processors.
These mechanisms can be used as follows.
Memory mapped devices and other I/O devices on the bus are often sensitive to the order of
writes to their I/O buffers. I/O instructions can be used to (the IN and OUT instructions) impose
strong write ordering on such accesses as follows. Prior to executing an I/O instruction, the
processor waits for all previous instructions in the program to complete and for all buffered
writes to drain to memory. Only instruction fetch and page tables walks can pass I/O instruc-
tions. Execution of subsequent instructions do not begin until the processor determines that the
I/O instruction has been completed.
Synchronization mechanisms in multiple-processor systems may depend upon a strong
memory-ordering model. Here, a program can use a locking instruction such as the XCHG
instruction or the LOCK prefix to insure that a read-modify-write operation on memory is
carried out atomically. Locking operations typically operate like I/O operations in that they wait
for all previous instructions to complete and for all buffered writes to drain to memory (refer to
Section 7.1.2., “Bus Locking”).
Program synchronization can also be carried out with serializing instructions (refer to Section
7.4., “Serializing Instructions”). These instructions are typically used at critical procedure or
task boundaries to force completion of all previous instructions before a jump to a new section
of code or a context switch occurs. Like the I/O and locking instructions, the processor waits
until all previous instructions have been completed and all buffered writes have been drained to
memory before executing the serializing instruction.
The MTRRs were introduced in the P6 family processors to define the cache characteristics for
specified areas of physical memory. The following are two examples of how memory types set
up with MTRRs can be used strengthen or weaken memory ordering for the P6 family proces-
sors:
• The uncached (UC) memory type forces a strong-ordering model on memory accesses.
Here, all reads and writes to the UC memory region appear on the bus and out-of-order or
speculative accesses are not performed. This memory type can be applied to an address
range dedicated to memory mapped I/O devices to force strong memory ordering.
• For areas of memory where weak ordering is acceptable, the write back (WB) memory
type can be chosen. Here, reads can be performed speculatively and writes can be buffered
and combined. For this type of memory, cache locking is performed on atomic (locked)
operations that do not split across cache lines, which helps to reduce the performance
penalty associated with the use of the typical synchronization instructions, such as XCHG,
that lock the bus during the entire read-modify-write operation. With the WB memory
type, the XCHG instruction locks the cache instead of the bus if the memory access is
contained within a cache line.
It is recommended that software written to run on P6 family processors assume the processor-
ordering model or a weaker memory-ordering model. The P6 family processors do not imple-
ment a strong memory-ordering model, except when using the UC memory type. Despite the
fact that P6 family processors support processor ordering, Intel does not guarantee that future
processors will support this model. To make software portable to future processors, it is recom-

7-10
 MULTIPLE-PROCESSOR MANAGEMENT

mended that operating systems provide critical region and resource control constructs and API’s
(application program interfaces) based on I/O, locking, and/or serializing instructions be used to
synchronize access to shared areas of memory in multiple-processor systems. Also, software
should not depend on processor ordering in situations where the system hardware does not vsup-
port this memory-ordering model.

7.3. PROPAGATION OF PAGE TABLE ENTRY CHANGES TO

MULTIPLE PROCESSORS
In a multiprocessor system, when one processor changes a page table entry or mapping, the
changes must also be propagated to all the other processors. This process is also known as “TLB
Shootdown.” Propagation may be done by memory-based semaphores and/or interprocessor
interrupts between processors. One naive but algorithmically correct TLB Shootdown sequence
for the Intel Architecture is:
1. Begin barrier: Stop all processors. Cause all but one to HALT or stop in a spinloop.
2. Let the active processor change the PTE(s).
3. Let all processors invalidate the PTE(s) modified in their TLBs.
4. End barrier: Resume all processors.
Alternate, performance-optimized, TBL Shootdown algorithms may be developed; however,
care must be taken by the developers to ensure that either:
• The differing TLB mappings are not actually used on different processors during the
update process.
OR
• The operating system is prepared to deal with the case where processor(s) is/are using the
stale mapping during the update process.

7.4. SERIALIZING INSTRUCTIONS

The Intel Architecture defines several serializing instructions. These instructions force the
processor to complete all modifications to flags, registers, and memory by previous instructions
and to drain all buffered writes to memory before the next instruction is fetched and executed.
For example, when a MOV to control register instruction is used to load a new value into control
register CR0 to enable protected mode, the processor must perform a serializing operation
before it enters protected mode. This serializing operation insures that all operations that were
started while the processor was in real-address mode are completed before the switch to
protected mode is made.
The concept of serializing instructions was introduced into the Intel Architecture with the
Pentium® processor to support parallel instruction execution. Serializing instructions have no
meaning for the Intel486™ and earlier processors that do not implement parallel instruction
execution.

7-11
MULTIPLE-PROCESSOR MANAGEMENT

It is important to note that executing of serializing instructions on P6 family processors constrain

speculative execution, because the results of speculatively executed instructions are discarded.
The following instructions are serializing instructions:
• Privileged serializing instructions—MOV (to control register), MOV (to debug register),
WRMSR, INVD, INVLPG, WBINVD, LGDT, LLDT, LIDT, and LTR.
• Nonprivileged serializing instructions—CPUID, IRET, and RSM.
The CPUID instruction can be executed at any privilege level to serialize instruction execu-
tion with no effect on program flow, except that the EAX, EBX, ECX, and EDX registers
are modified.
Nothing can pass a serializing instruction, and serializing instructions cannot pass any other
instruction (read, write, instruction fetch, or I/O).
When the processor serializes instruction execution, it ensures that all pending memory transac-
tions are completed, including writes stored in its store buffer, before it executes the next
instruction.
The following additional information is worth noting regarding serializing instructions:
• The processor does not writeback the contents of modified data in its data cache to external
memory when it serializes instruction execution. Software can force modified data to be
written back by executing the WBINVD instruction, which is a serializing instruction. It
should be noted that frequent use of the WBINVD instruction will seriously reduce system
performance.
• When an instruction is executed that enables or disables paging (that is, changes the PG
flag in control register CR0), the instruction should be followed by a jump instruction. The
target instruction of the jump instruction is fetched with the new setting of the PG flag (that
is, paging is enabled or disabled), but the jump instruction itself is fetched with the
previous setting. The P6 family processors do not require the jump operation following the
move to register CR0 (because any use of the MOV instruction in a P6 family processor to
write to CR0 is completely serializing). However, to maintain backwards and forward
compatibility with code written to run on other Intel Architecture processors, it is
recommended that the jump operation be performed.
• Whenever an instruction is executed to change the contents of CR3 while paging is
enabled, the next instruction is fetched using the translation tables that correspond to the
new value of CR3. Therefore the next instruction and the sequentially following instruc-
tions should have a mapping based upon the new value of CR3. (Global entries in the
TLBs are not invalidated, refer to Section 9.10., “Invalidating the Translation Lookaside
Buffers (TLBs)”, Chapter 9, Memory Cache Control.)
• The Pentium® and P6 family processors use branch-prediction techniques to improve
performance by prefetching the destination of a branch instruction before the branch
instruction is executed. Consequently, instruction execution is not deterministically
serialized when a branch instruction is executed.

7-12
 MULTIPLE-PROCESSOR MANAGEMENT

7.5. ADVANCED PROGRAMMABLE INTERRUPT CONTROLLER

(APIC)
The Advanced Programmable Interrupt Controller (APIC), referred to in the following sections
as the local APIC, was introduced into the Intel Architecture with the Pentium® processor
(beginning with the 735/90 and 815/100 models) and is included in all P6 family processors. The
local APIC performs two main functions for the processor:
• It processes local external interrupts that the processor receives at its interrupt pins and
local internal interrupts that software generates.
• In multiple-processor systems, it communicates with an external I/O APIC chip. The
external I/O APIC receives external interrupt events from the system and interprocessor
interrupts from the processors on the system bus and distributes them to the processors on
the system bus. The I/O APIC is part of Intel’s system chip set.
Figure 7-2 shows the relationship of the local APICs on the processors in a multiple-processor
(MP) system and the I/O APIC. The local APIC controls the dispatching of interrupts (to its
associated processor) that it receives either locally or from the I/O APIC. It provides facilities
for queuing, nesting and masking of interrupts. It handles the interrupt delivery protocol with its
local processor and accesses to APIC registers, and also manages interprocessor interrupts and
remote APIC register reads. A timer on the local APIC allows local generation of interrupts, and
local interrupt pins permit local reception of processor-specific interrupts. The local APIC can
be disabled and used in conjunction with a standard 8259A-style interrupt controller. (Disabling
the local APIC can be done in hardware for the Pentium® processors or in software for the P6
family processors.)
The I/O APIC is responsible for receiving interrupts generated by I/O devices and distributing
them among the local APICs by means of the APIC Bus. The I/O APIC manages interrupts using
either static or dynamic distribution schemes. Dynamic distribution of interrupts allows routing
of interrupts to the lowest priority processors. It also handles the distribution of interprocessor
interrupts and system-wide control functions such as NMI, INIT, SMI and start-up-interpro-
cessor interrupts. Individual pins on the I/O APIC can be programmed to generate a specific,
prioritized interrupt vector when asserted. The I/O APIC also has a “virtual wire mode” that
allows it to cooperate with an external 8259A in the system.
The APIC in the Pentium® and P6 family processors is an architectural subset of the Intel
82489DX external APIC. The differences are described in Section 7.5.19., “Software Visible
Differences Between the Local APIC and the 82489DX”
The following sections focus on the local APIC, and its implementation in the P6 family proces-
sors. Contact Intel for the information on I/O APIC.

7-13
MULTIPLE-PROCESSOR MANAGEMENT

Processor #1 Processor #2 Processor #3

CPU CPU CPU

Local APIC Local APIC Local APIC

Local Local Local
Interrupts Interrupts Interrupts

APIC Bus

I/O APIC
External
Interrupts I/O Chip Set

Figure 7-2. I/O APIC and Local APICs in Multiple-Processor Systems

7.5.1. Presence of APIC

Beginning with the P6 family processors, the presence or absence of an on-chip APIC can be
detected using the CPUID instruction. When the CPUID instruction is executed, bit 9 of the
feature flags returned in the EDX register indicates the presence (set) or absence (clear) of an
on-chip local APIC.

7.5.2. Enabling or Disabling the Local APIC

For the P6 family processors, a flag (the E flag, bit 11) in the APIC_BASE_MSR register
permits the local APIC to be explicitly enabled or disabled. Refer to Section 7.5.8., “Relocation
of the APIC Registers Base Address” for a description of this flag. For the Pentium® processor,
the APICEN pin (which is shared with the PICD1 pin) is used during reset to enable or disable
the local APIC.

7.5.3. APIC Bus

All I/O APIC and local APICs communicate through the APIC bus (a 3-line inter-APIC bus).
Two of the lines are open-drain (wired-OR) and are used for data transmission; the third line is
a clock. The bus and its messages are invisible to software and are not classed as architec-
tural (that is, the APIC bus and message format may change in future implementations
without having any effect on software compatibility).

7-14
 MULTIPLE-PROCESSOR MANAGEMENT

7.5.4. Valid Interrupts

The local and I/O APICs support 240 distinct vectors in the range of 16 to 255. Interrupt priority
is implied by its vector, according to the following relationship:
priority = vector / 16
One is the lowest priority and 15 is the highest. Vectors 16 through 31 are reserved for exclusive
use by the processor. The remaining vectors are for general use. The processor’s local APIC
includes an in-service entry and a holding entry for each priority level. To avoid losing inter-
rupts, software should allocate no more than 2 interrupt vectors per priority.

7.5.5. Interrupt Sources

The local APIC can receive interrupts from the following sources:
• Interrupt pins on the processor chip, driven by locally connected I/O devices.
• A bus message from the I/O APIC, originated by an I/O device connected to the I/O APIC.
• A bus message from another processor’s local APIC, originated as an interprocessor
interrupt.
• The local APIC’s programmable timer or the error register, through the self-interrupt
generating mechanism.
• Software, through the self-interrupt generating mechanism.
• (P6 family processors.) The performance-monitoring counters.
The local APIC services the I/O APIC and interprocessor interrupts according to the information
included in the bus message (such as vector, trigger type, interrupt destination, etc.). Interpreta-
tion of the processor’s interrupt pins and the timer-generated interrupts is programmable, by
means of the local vector table (LVT). To generate an interprocessor interrupt, the source
processor programs its interrupt command register (ICR). The programming of the ICR causes
generation of a corresponding interrupt bus message. Refer to Section 7.5.11., “Local Vector
Table” and Section 7.5.12., “Interprocessor and Self-Interrupts” for detailed information on
programming the LVT and ICR, respectively.

7.5.6. Bus Arbitration Overview

Being connected on a common bus (the APIC bus), the local and I/O APICs have to arbitrate for
permission to send a message on the APIC bus. Logically, the APIC bus is a wired-OR connec-
tion, enabling more than one local APIC to send messages simultaneously. Each APIC issues its
arbitration priority at the beginning of each message, and one winner is collectively selected
following an arbitration round. At any given time, a local APIC’s the arbitration priority is a
unique value from 0 to 15. The arbitration priority of each local APIC is dynamically modified
after each successfully transmitted message to preserve fairness. Refer to Section 7.5.16., “APIC
Bus Arbitration Mechanism and Protocol” for a detailed discussion of bus arbitration.

7-15
MULTIPLE-PROCESSOR MANAGEMENT

Section 7.5.3., “APIC Bus” describes the existing arbitration protocols and bus message
formats, while Section 7.5.12., “Interprocessor and Self-Interrupts” describes the INIT level de-
assert message, used to resynchronize all local APICs’ arbitration IDs. Note that except for start-
up (refer to Section 7.5.11., “Local Vector Table”), all bus messages failing during delivery are
automatically retried. The software should avoid situations in which interrupt messages may be
“ignored” by disabled or nonexistent “target” local APICs, and messages are being resent
repeatedly.

7.5.7. The Local APIC Block Diagram

Figure 7-3 gives a functional block diagram for the local APIC. Software interacts with the local
APIC by reading and writing its registers. The registers are memory-mapped to the processor’s
physical address space, and for each processor they have an identical address space of 4 KBytes
starting at address FEE00000H. (Refer to Section 7.5.8., “Relocation of the APIC Registers
Base Address” for information on relocating the APIC registers base address for the P6 family
processors.)

NOTE
For P6 family processors, the APIC handles all memory accesses to addresses
within the 4-KByte APIC register space and no external bus cycles are
produced. For the Pentium® processors with an on-chip APIC, bus cycles are
produced for accesses to the 4-KByte APIC register space. Thus, for software
intended to run on Pentium® processors, system software should explicitly
not map the APIC register space to regular system memory. Doing so can
result in an invalid opcode exception (#UD) being generated or unpredictable
execution.
The 4-KByte APIC register address space should be mapped as uncacheable (UC), refer to
Section 9, “Memory Cache Control”, in Chapter 9, Memory Cache Control.

7-16
 MULTIPLE-PROCESSOR MANAGEMENT

DATA/ADDR INTA EXTINT INTR

Version Register

Timer EOI Register

Current Count
Register
Task Priority
Initial Count Register
Register
Divide Configuration
Register
Prioritizer
Local Vec Table

Timer
15 1
T S R V T S R V
LINT0/1 Local
Interrupts 0,1 TMR, ISR, IRR Registers

Performance
Monitoring Counters*
T R V T R V
Error Software Transparent Registers

Vec[3:0] Register
& TMR Bit Select

Interrupt Command Arb. ID Vector

Register Register Decode

INIT,
Processor Acceptance
APIC ID Priority Logic
NMI,
SMI
Register
Logical Destination Dest. Mode
& Vector
Register
Destination Format APIC Bus
Send/Receive Logic
Register

APIC Serial Bus

* Available only in P6 family processors

Figure 7-3. Local APIC Structure

Within the 4-KByte APIC register area, the register address allocation scheme is shown in Table
7-1. Register offsets are aligned on 128-bit boundaries. All registers must be accessed using 32-
bit loads and stores. Wider registers (64-bit or 256-bit) are defined and accessed as independent
multiple 32-bit registers. If a LOCK prefix is used with a MOV instruction that accesses the
APIC address space, the prefix is ignored; that is, a locking operation does not take place.

7-17
MULTIPLE-PROCESSOR MANAGEMENT

Table 7-1. Local APIC Register Address Map

Address Register Name Software Read/Write
FEE0 0000H Reserved
FEE0 0010H Reserved
FEE0 0020H Local APIC ID Register Read/write
FEE0 0030H Local APIC Version Register Read only
FEE0 0040H Reserved
FEE0 0050H Reserved
FEE0 0060H Reserved
FEE0 0070H Reserved
FEE0 0080H Task Priority Register Read/Write
FEE0 0090H Arbitration Priority Register Read only
FEE0 00A0H Processor Priority Register Read only
FEE0 00B0H EOI Register Write only
FEE0 00C0H Reserved
FEE0 00D0H Logical Destination Register Read/Write
FEE0 00E0H Destination Format Register Bits 0-27 Read only. Bits
28-31 Read/Write
FEE0 00F0H Spurious-Interrupt Vector Register Bits 0-3 Read only. Bits
4-9 Read/Write
FEE0 0100H through ISR 0-255 Read only
FEE0 0170H
FEE0 0180H through TMR 0-255 Read only
FEE0 01F0H
FEE0 0200H through IRR 0-255 Read only
FEE0 0270H
FEE0 0280H Error Status Register Read only
FEE0 0290H through Reserved
FEE0 02F0H
FEE0 0300H Interrupt Command Reg. 0-31 Read/Write
FEE0 0310H Interrupt Command Reg. 32-63 Read/Write
FEE0 0320H Local Vector Table (Timer) Read/Write
FEE0 0330H Reserved
FEE0 0340H Performance Counter LVT1 Read/Write
FEE0 0350H Local Vector Table (LINT0) Read/Write
FEE0 0360H Local Vector Table (LINT1) Read/Write
FEE0 0370H Local Vector Table (Error)2 Read/Write
FEE0 0380H Initial Count Register for Timer Read/Write

7-18
 MULTIPLE-PROCESSOR MANAGEMENT

Table 7-1. Local APIC Register Address Map (Contd.)

Address Register Name Software Read/Write
FEE0 0390H Current Count Register for Timer Read only
FEE0 03A0H through Reserved
FEE0 03D0H
FEE0 03E0H Timer Divide Configuration Register Read/Write
FEE0 03F0H Reserved

NOTES:
1. Introduced into the APIC Architecture in the Pentium® Pro processor.
2. Introduced into the APIC Architecture in the Pentium® processor.

7.5.8. Relocation of the APIC Registers Base Address

The P6 family processors permit the starting address of the APIC registers to be relocated from
FEE00000H to another physical address. This extension of the APIC architecture is provided to
help resolve conflicts with memory maps of existing systems. The P6 family processors also
provide the ability to enable or disable the local APIC.
An alternate APIC base address is specified through the APIC_BASE_MSR register. This MSR
is located at MSR address 27 (1BH). Figure 7-4 shows the encoding of the bits in this register.
This register also provides the flag for enabling or disabling the local APIC.
The functions of the bits in the APIC_BASE_MSR register are as follows:
BSP flag, bit 8 Indicates if the processor is the bootstrap processor (BSP), determined during
the MP initialization (refer to Section 7.7., “Multiple-Processor (MP) Initial-
ization Protocol”). Following a power-up or reset, this flag is clear for all the
processors in the system except the single BSP.

63 36 35 12 11 10 9 8 7 0

Reserved APIC Base

APIC Base—Base physical address

E—APIC enable/disable
BSP—Processor is BSP

Reserved

Figure 7-4. APIC_BASE_MSR

E (APIC Enabled) flag, bit 11

Permits the local APIC to be enabled (set) or disabled (clear). Following a
power-up or reset, this flag is set, enabling the local APIC. When this flag is

7-19
MULTIPLE-PROCESSOR MANAGEMENT

clear, the processor is functionally equivalent to an Intel Architecture processor

without an on-chip APIC (for example, an Intel486™ processor). This flag is
implementation dependent and in not guaranteed to be available or available at
the same location in future Intel Architecture processors.
APIC Base field, bits 12 through 35
Specifies the base address of the APIC registers. This 24-bit value is extended
by 12 bits at the low end to form the base address, which automatically aligns
the address on a 4-KByte boundary. Following a power-up or reset, this field is
set to FEE00000H.
Bits 0 through 7, bits 9 and 10, and bits 36 through 63 in the APIC_BASE_MSR register are
reserved.

7.5.9. Interrupt Destination and APIC ID

The destination of an interrupt can be one, all, or a subset of the processors in the system. The
sender specifies the destination of an interrupt in one of two destination modes: physical or
logical.

7.5.9.1. PHYSICAL DESTINATION MODE

In physical destination mode, the destination processor is specified by its local APIC ID. This
ID is matched against the local APIC’s actual physical ID, which is stored in the local APIC ID
register (refer to Figure 7-5). Either a single destination (the ID is 0 through 14) or a broadcast
to all (the ID is 15) can be specified in physical destination mode. Note that in this mode, up to
15 the local APICs can be individually addressed. An ID of all 1s denotes a broadcast to all local
APICs. The APIC ID register is loaded at power up by sampling configuration data that is driven
onto pins of the processor. For the P6 family processors, pins A11# and A12# and pins BR0#
through BR3# are sampled; for the Pentium® processor, pins BE0# through BE3# are sampled.
The ID portion can be read and modified by software.

31 28 27 24 23 0

Reserved APIC ID Reserved

Address: 0FEE0 0020H

Value after reset: 0000 0000H

Figure 7-5. Local APIC ID Register

7.5.9.2. LOGICAL DESTINATION MODE

In logical destination mode, message destinations are specified using an 8-bit message destina-
tion address (MDA). The MDA is compared against the 8-bit logical APIC ID field of the APIC
logical destination register (LDR), refer to Figure 7-6.

7-20
 MULTIPLE-PROCESSOR MANAGEMENT

31 24 23 0

Logical APIC ID Reserved

Address: 0FEE0 00D0H

Value after reset: 0000 0000H

Figure 7-6. Logical Destination Register (LDR)

Destination format register (DFR) defines the interpretation of the logical destination informa-
tion (refer to Figure 7-7). The DFR register can be programmed for flat model or cluster model
interrupt delivery modes.

31 28 0

Model Reserved (All 1s)

Address: 0FEE0 00E0H

Value after reset: FFFF FFFFH

Figure 7-7. Destination Format Register (DFR)

7.5.9.3. FLAT MODEL

For the flat model, bits 28 through 31 of the DFR must be programmed to 1111. The MDA is
interpreted as a decoded address. This scheme allows the specification of arbitrary groups of
local APICs simply by setting each APIC’s bit to 1 in the corresponding LDR. In the flat model,
up to 8 local APICs can coexist in the system. Broadcast to all APICs is achieved by setting all
8 bits of the MDA to ones.

7.5.9.4. CLUSTER MODEL

For the cluster model, the DFR bits 28 through 31 should be programmed to 0000. In this model,
there are two basic connection schemes: flat cluster and hierarchical cluster.
In the flat cluster connection model, all clusters are assumed to be connected on a single APIC
bus. Bits 28 through 31 of the MDA contains the encoded address of the destination cluster.
These bits are compared with bits 28 through 31 of the LDR to determine if the local APIC is
part of the cluster. Bits 24 through 27 of the MDA are compared with Bits 24 through 27 of the
LDR to identify individual local APIC unit within the cluster. Arbitrary sets of processors within
a cluster can be specified by writing the target cluster address in bits 28 through 31 of the MDA
and setting selected bits in bits 24 through 27 of the MDA, corresponding to the chosen members
of the cluster. In this mode, 15 clusters (with cluster addresses of 0 through 14) each having 4
processors can be specified in the message. The APIC arbitration ID, however, supports only
15 agents, and hence the total number of processors supported in this mode is limited to 15.

7-21
MULTIPLE-PROCESSOR MANAGEMENT

Broadcast to all local APICs is achieved by setting all destination bits to one. This guarantees a
match on all clusters, and selects all APICs in each cluster.
In the hierarchical cluster connection model, an arbitrary hierarchical network can be created by
connecting different flat clusters via independent APIC buses. This scheme requires a cluster
manager within each cluster, responsible for handling message passing between APIC buses.
One cluster contains up to 4 agents. Thus 15 cluster managers, each with 4 agents, can form a
network of up to 60 APIC agents. Note that hierarchical APIC networks requires a special
cluster manager device, which is not part of the local or the I/O APIC units.

7.5.9.5. ARBITRATION PRIORITY

Each local APIC is given an arbitration priority of from 0 to 15 upon reset. The I/O APIC uses
this priority during arbitration rounds to determine which local APIC should be allowed to
transmit a message on the APIC bus when multiple local APICs are issuing messages. The local
APIC with the highest arbitration priority wins access to the APIC bus. Upon completion of an
arbitration round, the winning local APIC lowers its arbitration priority to 0 and the losing local
APICs each raise theirs by 1. In this manner, the I/O APIC distributes message bus-cycles
among the contesting local APICs.
The current arbitration priority for a local APIC is stored in a 4-bit, software-transparent arbi-
tration ID (Arb ID) register. During reset, this register is initialized to the APIC ID number
(stored in the local APIC ID register). The INIT-deassert command resynchronizes the arbitra-
tion priorities of the local APICs by resetting Arb ID register of each agent to its current APIC
ID value.

7.5.10. Interrupt Distribution Mechanisms

The APIC supports two mechanisms for selecting the destination processor for an interrupt:
static and dynamic. Static distribution is used to access a specific processor in the network.
Using this mechanism, the interrupt is unconditionally delivered to all local APICs that match
the destination information supplied with the interrupt. The following delivery modes fall into
the static distribution category: fixed, SMI, NMI, EXTINT, and start-up.
Dynamic distribution assigns incoming interrupts to the lowest priority processor, which is
generally the least busy processor. It can be programmed in the LVT for local interrupt delivery
or the ICR for bus messages. Using dynamic distribution, only the “lowest priority” delivery
mode is allowed. From all processors listed in the destination, the processor selected is the one
whose current arbitration priority is the lowest. The latter is specified in the arbitration priority
register (APR), refer to Section 7.5.13.4., “Arbitration Priority Register (APR)” If more than one
processor shares the lowest priority, the processor with the highest arbitration priority (the
unique value in the Arb ID register) is selected.
In lowest priority mode, if a focus processor exists, it may accept the interrupt, regardless of its
priority. A processor is said to be the focus of an interrupt if it is currently servicing that interrupt
or if it has a pending request for that interrupt.

7-22
 MULTIPLE-PROCESSOR MANAGEMENT

7.5.11. Local Vector Table

The local APIC contains a local vector table (LVT), specifying interrupt delivery and status
information for the local interrupts. The information contained in this table includes the inter-
rupt’s associated vector, delivery mode, status bits and other data as shown in Figure 7-8. The
LVT incorporates five 32-bit entries: one for the timer, one each for the two local interrupt
(LINT0 and LINT1) pins, one for the error interrupt, and (in the P6 family processors) one for
the performance-monitoring counter interrupt.
The fields in the LVT are as follows:
Vector Interrupt vector number.
Delivery Mode Defined only for local interrupt entries 1 and 2 and the performance-
monitoring counter. The timer and the error status register (ESR)
generate only edge triggered maskable hardware interrupts to the
local processor. The delivery mode field does not exist for the timer
and error interrupts. The performance-monitoring counter LVT may
be programmed with a Deliver Mode equal to Fixed or NMI only.
Note that certain delivery modes will only operate as intended when
used in conjunction with a specific Trigger Mode. The allowable
delivery modes are as follows:
000 (Fixed) Delivers the interrupt, received on the local
interrupt pin, to this processor as specified in the
corresponding LVT entry. The trigger mode can be
edge or level. Note, if the processor is not used in
conjunction with an I/O APIC, the fixed delivery
mode may be software programmed for an edge-
triggered interrupt, but the P6 family processors
implementation will always operate in a level-
triggered mode.
100 (NMI) Delivers the interrupt, received on the local inter-
rupt pin, to this processor as an NMI interrupt. The
vector information is ignored. The NMI interrupt
is treated as edge-triggered, even if programmed
otherwise. Note that the NMI may be masked. It is
the software's responsibility to program the LVT
mask bit according to the desired behavior of
NMI.
111 (ExtINT) Delivers the interrupt, received on the local inter-
rupt pin, to this processor and responds as if the
interrupt originated in an externally connected
(8259A-compatible) interrupt controller. A spe-
cial INTA bus cycle corresponding to ExtINT, is
routed to the external controller. The latter is ex-
pected to supply the vector information. When the
delivery mode is ExtINT, the trigger-mode is

7-23
MULTIPLE-PROCESSOR MANAGEMENT

level-triggered, regardless of how the APIC trig-

gering mode is programmed. The APIC architec-
ture supports only one ExtINT source in a system,
usually contained in the compatibility bridge.

31 18 17 16 15 13 12 11 8 7 0

Timer Vector

Address: FEE0 0320H

Timer Mode
Value after Reset: 0001 0000H
0: One-shot
1: Periodic
Delivery Status
0: Idle
1: Send Pending
Mask
0: Not Masked
1: Masked

Interrupt Input Delivery Mode

Pin Polarity 000: Fixed
100: NMI
111: ExtlNT
Remote All other combinations
IRR are Reserved

Trigger Mode
0: Edge
1: Level

31 17 11 10 8 7 0

LINT0 Vector

LINT1 Vector

ERROR Vector

PCINT Vector

16 15 14 13 12
Address: FEE0 0350H
Reserved
Address: FEE0 0360H
Address: FEE0 0370H
Address: FEE0 0340H
Value After Reset: 0001 0000H

Figure 7-8. Local Vector Table (LVT)

7-24
 MULTIPLE-PROCESSOR MANAGEMENT

Delivery Status (read only)

Holds the current status of interrupt delivery. Two states are defined:
0 (Idle) There is currently no activity for this interrupt, or
the previous interrupt from this source has com-
pleted.
1 (Send Pending)
Indicates that the interrupt transmission has start-
ed, but has not yet been completely accepted.
Interrupt Input Pin Polarity
Specifies the polarity of the corresponding interrupt pin: (0) active
high or (1) active low.
Remote Interrupt Request Register (IRR) Bit
Used for level triggered interrupts only; its meaning is undefined for
edge triggered interrupts. For level triggered interrupts, the bit is set
when the logic of the local APIC accepts the interrupt. The remote
IRR bit is reset when an EOI command is received from the
processor.
Trigger Mode Selects the trigger mode for the local interrupt pins when the delivery
mode is Fixed: (0) edge sensitive and (1) level sensitive. When the
delivery mode is NMI, the trigger mode is always level sensitive;
when the delivery mode is ExtINT, the trigger mode is always level
sensitive. The timer and error interrupts are always treated as edge
sensitive.
Mask Interrupt mask: (0) enables reception of the interrupt and (1) inhibits
reception of the interrupt.
Timer Mode Selects the timer mode: (0) one-shot and (1) periodic (refer to Section
7.5.18., “Timer”).

7.5.12. Interprocessor and Self-Interrupts

A processor generates interprocessor interrupts by writing into the interrupt command register
(ICR) of its local APIC (refer to Figure 7-9). The processor may use the ICR for self interrupts
or for interrupting other processors (for example, to forward device interrupts originally
accepted by it to other processors for service). In addition, special inter-processor interrupts
(IPI) such as the start-up IPI message, can only be delivered using the ICR mechanism. ICR-
based interrupts are treated as edge triggered even if programmed otherwise. Note that not all
combinations of options for ICR generated interrupts are valid (refer to Table 7-2).

7-25
MULTIPLE-PROCESSOR MANAGEMENT

63 56 55 32
Destination Field Reserved

31 20 19 18 17 16 15 14 13 12 11 10 8 7 0
Reserved Vector

Destination Shorthand Delivery Mode

00: Dest. Field 000: Fixed
01: Self 001: Lowest Priority
10: All Incl. Self 010: SMI
11: All Excl. Self 011: Reserved
100: NMI
101: INIT
110: Start Up
Reserved 111: Reserved

Destination Mode
Address: FEE0 0310H 0: Physical
Value after Reset: 0H 1: Logical

Delivery Status
0: Idle
1: Send Pending

Level
0 = De-assert
1 = Assert

Trigger Mode
0: Edge
1: Level

Figure 7-9. Interrupt Command Register (ICR)

All fields of the ICR are read-write by software with the exception of the delivery status field,
which is read-only. Writing to the 32-bit word that contains the interrupt vector causes the inter-
rupt message to be sent. The ICR consists of the following fields.
Vector The vector identifying the interrupt being sent. The localAPIC
register addresses are summarized in Table 7-1.
Delivery Mode Specifies how the APICs listed in the destination field should act
upon reception of the interrupt. Note that all interprocessor interrupts
behave as edge triggered interrupts (except for INIT level de-assert
message) even if they are programmed as level triggered interrupts.
000 (Fixed) Deliver the interrupt to all processors listed in the
destination field according to the information pro-
vided in the ICR. The fixed interrupt is treated as

7-26
 MULTIPLE-PROCESSOR MANAGEMENT

an edge-triggered interrupt even if programmed

otherwise.
001 (Lowest Priority)
Same as fixed mode, except that the interrupt is
delivered to the processor executing at the lowest
priority among the set of processors listed in the
destination.
010 (SMI) Only the edge trigger mode is allowed. The vector
field must be programmed to 00B.
011 (Reserved)
100 (NMI) Delivers the interrupt as an NMI interrupt to all
processors listed in the destination field. The vec-
tor information is ignored. NMI is treated as an
edge triggered interrupt even if programmed oth-
erwise.
101 (INIT) Delivers the interrupt as an INIT signal to all pro-
cessors listed in the destination field. As a result,
all addressed APICs will assume their INIT state.
As in the case of NMI, the vector information is
ignored, and INIT is treated as an edge triggered
interrupt even if programmed otherwise.
101 (INIT Level De-assert)
(The trigger mode must also be set to 1 and level
mode to 0.) Sends a synchronization message to
all APIC agents to set their arbitration IDs to the
values of their APIC IDs. Note that the INIT inter-
rupt is sent to all agents, regardless of the destina-
tion field value. However, at least one valid
destination processor should be specified. For fu-
ture compatibility, the software is requested to use
a broadcast-to-all (“all-incl-self” shorthand, as de-
scribed below).
110 (Start-Up) Sends a special message between processors in a
multiple-processor system. For details refer to the
Pentium® Pro Family Developer’s Manual, Vol-
ume 1. The Vector information contains the start-
up address for the multiple-processor boot-up pro-
tocol. Start-up is treated as an edge triggered inter-
rupt even if programmed otherwise. Note that
interrupts are not automatically retried by the
source APIC upon failure in delivery of the mes-
sage. It is up to the software to decide whether a

7-27
MULTIPLE-PROCESSOR MANAGEMENT

retry is needed in the case of failure, and issue a

retry message accordingly.
Destination Mode Selects either (0) physical or (1) logical destination mode.
Delivery Status Indicates the delivery status:
0 (Idle) There is currently no activity for this interrupt, or
the previous interrupt from this source has com-
pleted.
1 (Send Pending)
Indicates that the interrupt transmission has start-
ed, but has not yet been completely accepted.
Level For INIT level de-assert delivery mode the level is 0. For all other
modes the level is 1.
Trigger Mode Used for the INIT level de-assert delivery mode only.
Destination Shorthand
Indicates whether a shorthand notation is used to specify the destina-
tion of the interrupt and, if so, which shorthand is used. Destination
shorthands do not use the 8-bit destination field, and can be sent by
software using a single write to the lower 32-bit part of the APIC
interrupt command register. Shorthands are defined for the following
cases: software self interrupt, interrupt to all processors in the system
including the sender, interrupts to all processors in the system
excluding the sender.
00: (destination field, no shorthand)
The destination is specified in bits 56 through 63
of the ICR.
01: (self) The current APIC is the single destination of the
interrupt. This is useful for software self inter-
rupts. The destination field is ignored. Refer to Ta-
ble 7-2 for description of supported modes. Note
that self interrupts do not generate bus messages.
10: (all including self)
The interrupt is sent to all processors in the system
including the processor sending the interrupt. The
APIC will broadcast a message with the destina-
tion field set to FH. Refer to Table 7-2 for descrip-
tion of supported modes.
11: (all excluding self)
The interrupt is sent to all processors in the system
with the exception of the processor sending the in-
terrupt. The APIC will broadcast a message using

7-28
 MULTIPLE-PROCESSOR MANAGEMENT

the physical destination mode and destination

field set to FH.
Destination This field is only used when the destination shorthand field is set to
“dest field”. If the destination mode is physical, then bits 56 through
59 contain the APIC ID. In logical destination mode, the interpreta-
tion of the 8-bit destination field depends on the DFR and LDR of the
local APIC Units.
Table 7-2 shows the valid combinations for the fields in the interrupt control register.

Table 7-2. Valid Combinations for the APIC Interrupt Command Register
Trigger Valid/ Destination
Mode Destination Mode Delivery Mode Invalid Shorthand
Edge Physical or Logical Fixed, Lowest Priority, NMI, Valid Dest. Field
SMI, INIT, Start-Up
Level Physical or Logical Fixed, Lowest Priority, NMI 1 Dest. field
Level Physical or Logical INIT 2 Dest. Field
Level x4 SMI, Start-Up Invalid3 x
Edge x Fixed Valid Self
Level x Fixed 1 Self
x x Lowest Priority, NMI, INIT, Invalid3 Self
SMI, Start-Up
Edge x Fixed Valid All inc Self
Level x Fixed 1 All inc Self
3
x x Lowest Priority, NMI, INIT, Invalid All inc Self
SMI, Start-Up
Edge x Fixed, Lowest Priority, NMI, Valid All excl Self
INIT, SMI, Start-Up
Level x Fixed, Lowest Priority, NMI 1 All excl Self
3
Level x SMI, Start-Up Invalid All excl Self
Level x INIT 2 All excl Self

NOTES:
1. Valid. Treated as edge triggered if Level = 1 (assert), otherwise ignored.
2. Valid. Treated as edge triggered when Level = 1 (assert); when Level = 0 (deassert), treated as “INIT
Level Deassert” message. Only INIT level deassert messages are allowed to have level = deassert. For
all other messages the level must be “assert.”
3. Invalid. The behavior of the APIC is undefined.
4. X—Don’t care.

7-29
MULTIPLE-PROCESSOR MANAGEMENT

7.5.13. Interrupt Acceptance

Three 256-bit read-only registers (the IRR, ISR, and TMR registers) are involved in the interrupt
acceptance logic (refer to Figure 7-10). The 256 bits represents the 256 possible vectors.
Because vectors 0 through 15 are reserved, so are bits 0 through 15 in these registers. The func-
tions of the three registers are as follows:
TMR (trigger mode register)
Upon acceptance of an interrupt, the corresponding TMR bit is
cleared for edge triggered interrupts and set for level interrupts. If the
TMR bit is set, the local APIC sends an EOI message to all I/O
APICs as a result of software issuing an EOI command (refer to
Section 7.5.13.6., “End-Of-Interrupt (EOI)” for a description of the
EOI register).

255 16 15 0

Reserved IRR

Reserved ISR

Reserved TMR

Addresses: IRR FEE0 0200H - FEE0 0270H

ISR FEE0 0100H - FEE0 0170H
TMR FEE0 0180H - FEE0 01F0H
Value after reset: 0H

Figure 7-10. IRR, ISR and TMR Registers

IRR (interrupt request register)

Contains the active interrupt requests that have been accepted, but
not yet dispensed by the current local APIC. A bit in IRR is set when
the APIC accepts the interrupt. The IRR bit is cleared, and a corre-
sponding ISR bit is set when the INTA cycle is issued.
ISR (in-service register)
Marks the interrupts that have been delivered to the processor, but
have not been fully serviced yet, as an EOI has not yet been received
from the processor. The ISR reflects the current state of the processor
interrupt queue. The ISR bit for the highest priority IRR is set during
the INTA cycle. During the EOI cycle, the highest priority ISR bit is
cleared, and if the corresponding TMR bit was set, an EOI message
is sent to all I/O APICs.

7.5.13.1. INTERRUPT ACCEPTANCE DECISION FLOW CHART

The process that the APIC uses to accept an interrupt is shown in the flow chart in Figure 7-11.
The response of the local APIC to the start-up IPI is explained in the Pentium® Pro Family
Developer’s Manual, Volume 1.

7-30
 MULTIPLE-PROCESSOR MANAGEMENT

Wait to Receive
Bus Message

Belong
Discard No to
Message Destination?

Yes
Is it
NMI/SMI/INIT Yes Accept
/ Message
ExtINT?

No
Lowest
Fixed Delivery Priority
Mode?

No Is Interrupt Am Yes
Set Status I Accept
to Retry Slot Available? Message
Focus?

Yes No

Yes Is Status No Other Yes Discard

a Retry? Focus? Message

No Accept
Message

No Is Yes
Set Status Interrupt Slot Arbitrate
to Retry Available?

No Am I Yes Accept
Winner? Message

Figure 7-11. Interrupt Acceptance Flow Chart for the Local APIC

7.5.13.2. TASK PRIORITY REGISTER

Task priority register (TPR) provides a priority threshold mechanism for interrupting the
processor (refer to Figure 7-12). Only interrupts whose priority is higher than that specified in
the TPR will be serviced. Other interrupts are recorded and are serviced as soon as the TPR value
is decreased enough to allow that. This enables the operating system to block temporarily
specific interrupts (generally low priority) from disturbing high-priority tasks execution. The
priority threshold mechanism is not applicable for delivery modes excluding the vector infor-
mation (that is, for ExtINT, NMI, SMI, INIT, INIT-Deassert, and Start-Up delivery modes).

7-31
MULTIPLE-PROCESSOR MANAGEMENT

31 8 7 0
Task
Reserved
Priority
Address: FEE0 0080H
Value after reset: 0H

Figure 7-12. Task Priority Register (TPR)

The Task Priority is specified in the TPR. The 4 most-significant bits of the task priority corre-
spond to the 16 interrupt priorities, while the 4 least-significant bits correspond to the sub-class
priority. The TPR value is generally denoted as x:y, where x is the main priority and y provides
more precision within a given priority class. When the x-value of the TPR is 15, the APIC will
not accept any interrupts.

7.5.13.3. PROCESSOR PRIORITY REGISTER (PPR)

The processor priority register (PPR) is used to determine whether a pending interrupt can be
dispensed to the processor. Its value is computed as follows:
IF TPR[7:4] ≥ ISRV[7:4]
THEN
PPR[7:0] = TPR[7:0]
ELSE
PPR[7:4] = ISRV[7:4] AND PPR[3:0] = 0
Where ISRV is the vector of the highest priority ISR bit set, or zero if no ISR bit is set. The PPR
format is identical to that of the TPR. The PPR address is FEE000A0H, and its value after reset
is zero.

7.5.13.4. ARBITRATION PRIORITY REGISTER (APR)

Arbitration priority register (APR) holds the current, lowest-priority of the processor, a value
used during lowest priority arbitration (refer to Section 7.5.16., “APIC Bus Arbitration Mecha-
nism and Protocol”). The APR format is identical to that of the TPR. The APR value is
computed as the following.
IF (TPR[7:4] ≥ IRRV[7:4]) AND (TPR[7:4] > ISRV[7:4])
THEN
APR[7:0] = TPR[7:0]
ELSE
APR[7:4] = max(TPR[7:4] AND ISRV[7:4], IRRV[7:4]), APR[3:0]=0.
Here, IRRV is the interrupt vector with the highest priority IRR bit set or cleared (if no IRR bit
is set). The APR address is FEE0 0090H, and its value after reset is 0.

7-32
 MULTIPLE-PROCESSOR MANAGEMENT

7.5.13.5. SPURIOUS INTERRUPT

A special situation may occur when a processor raises its task priority to be greater than or equal
to the level of the interrupt for which the processor INTR signal is currently being asserted. If
at the time the INTA cycle is issued, the interrupt that was to be dispensed has become masked
(programmed by software), the local APIC will return a spurious-interrupt vector to the
processor. Dispensing the spurious-interrupt vector does not affect the ISR, so the handler for
this vector should return without an EOI.

7.5.13.6. END-OF-INTERRUPT (EOI)

During the interrupt serving routine, software should indicate acceptance of lowest-priority,
fixed, timer, and error interrupts by writing an arbitrary value into its local APIC end-of-inter-
rupt (EOI) register (refer to Figure 7-13). This is an indication for the local APIC it can issue the
next interrupt, regardless of whether the current interrupt service has been terminated or not.
Note that interrupts whose priority is higher than that currently in service, do not wait for the
EOI command corresponding to the interrupt in service.

31 0

Address: 0FEE0 00B0H

Value after reset: 0H

Figure 7-13. EOI Register

Upon receiving end-of-interrupt, the APIC clears the highest priority bit in the ISR and selects
the next highest priority interrupt for posting to the CPU. If the terminated interrupt was a level-
triggered interrupt, the local APIC sends an end-of-interrupt message to all I/O APICs. Note that
EOI command is supplied for the above two interrupt delivery modes regardless of the interrupt
source (that is, as a result of either the I/O APIC interrupts or those issued on local pins or using
the ICR). For future compatibility, the software is requested to issue the end-of-interrupt
command by writing a value of 0H into the EOI register.

7.5.14. Local APIC State

In P6 family processors, all local APICs are initialized in a software-disabled state after power-
up. A software-disabled local APIC unit responds only to self-interrupts and to INIT, NMI, SMI,
and start-up messages arriving on the APIC Bus. The operation of local APICs during the
disabled state is as follows:
• For the INIT, NMI, SMI, and start-up messages, the APIC behaves normally, as if fully
enabled.

7-33
MULTIPLE-PROCESSOR MANAGEMENT

• Pending interrupts in the IRR and ISR registers are held and require masking or handling
by the CPU.
• A disabled local APIC does not affect the sending of APIC messages. It is software’s
responsibility to avoid issuing ICR commands if no sending of interrupts is desired.
• Disabling a local APIC does not affect the message in progress. The local APIC will
complete the reception/transmission of the current message and then enter the disabled
state.
• A disabled local APIC automatically sets all mask bits in the LVT entries. Trying to reset
these bits in the local vector table will be ignored.
• A software-disabled local APIC listens to all bus messages in order to keep its arbitration
ID synchronized with the rest of the system, in the event that it is re-enabled.
For the Pentium® processor, the local APIC is enabled and disabled through a hardware mecha-
nism. (Refer to the Pentium® Processor Data Book for a description of this mechanism.)

7.5.14.1. SPURIOUS-INTERRUPT VECTOR REGISTER

Software can enable or disable a local APIC at any time by programming bit 8 of the spurious-
interrupt vector register (SVR), refer to Figure 7-14. The functions of the fields in the SVR are
as follows:

31 10 9 8 7 4 3 0

Reserved 1111

Focus Processor Checking

0: Enabled
1: Disabled APIC Enabled Spurious Vector
0: APIC SW Disabled
1: APIC SW Enabled
Address: FEE0 00F0H
Value after reset: 0000 00FFH
Figure 7-14. Spurious-Interrupt Vector Register (SVR)

Spurious Vector Released during an INTA cycle when all pending interrupts are
masked or when no interrupt is pending. Bits 4 through 7 of the this
field are programmable by software, and bits 0 through 3 are hard-
wired to logical ones. Software writes to bits 0 through 3 have no
effect.
APIC Enable Allows software to enable (1) or disable (0) the local APIC. To
bypass APIC completely, use the APIC_BASE_MSR in Figure 7-4.
Focus Processor Determines if focus processor checking is enabled during the lowest
Checking Priority delivery: (0) enabled and (1) disabled.

7-34
 MULTIPLE-PROCESSOR MANAGEMENT

7.5.14.2. LOCAL APIC INITIALIZATION

On a hardware reset, the processor and its local APIC are initialized simultaneously. For the P6
family processors, the local APIC obtains its initial physical ID from system hardware at the
falling edge of the RESET# signal by sampling 6 lines on the system bus (the BR[3:0]) and
cluster ID[1:0] lines) and storing this value into the APIC ID register; for the Pentium®
processor, four lines are sampled (BE0# through BE3#). Refer to the Pentium® Pro & Pentium
II Processors Data Book and the Pentium® Processor Data Book for descriptions of this mech-
anism.

7.5.14.3. LOCAL APIC STATE AFTER POWER-UP RESET

The state of local APIC registers and state machines after a power-up reset are as follows:
• The following registers are all reset to 0: the IRR, ISR, TMR, ICR, LDR, and TPR
registers; the holding registers; the timer initial count and timer current count registers; the
remote register; and the divide configuration register.
• The DFR register is reset to all 1s.
• The LVT register entries are reset to 0 except for the mask bits, which are set to 1s.
• The local APIC version register is not affected.
• The local APIC ID and Arb ID registers are loaded from processor input pins (the Arb ID
register is set to the APIC ID value for the local APIC).
• All internal state machines are reset.
• APIC is software disabled (that is, bit 8 of the SVR register is set to 0).
• The spurious-interrupt vector register is initialized to FFH.

7.5.14.4. LOCAL APIC STATE AFTER AN INIT RESET

An INIT reset of the processor can be initiated in either of two ways:
• By asserting the processor’s INIT# pin.
• By sending the processor an INIT IPI (sending an APIC bus-based interrupt with the
delivery mode set to INIT).
Upon receiving an INIT via either of these two mechanisms, the processor responds by begin-
ning the initialization process of the processor core and the local APIC. The state of the local
APIC following an INIT reset is the same as it is after a power-up reset, except that the APIC
ID and Arb ID registers are not affected.

7.5.14.5. LOCAL APIC STATE AFTER INIT-DEASSERT MESSAGE

An INIT-disassert message has no affect on the state of the APIC, other than to reload the arbi-
tration ID register with the value in the APIC ID register.

7-35
MULTIPLE-PROCESSOR MANAGEMENT

7.5.15. Local APIC Version Register

The local APIC contains a hardwired version register, which software can use to identify the
APIC version (refer to Figure 7-16). In addition, the version register specifies the size of LVT
used in the specific implementation. The fields in the local APIC version register are as follows:

Version The version numbers of the local APIC or an external 82489DX

APIC controller:
1XH Local APIC.
0XH 82489DX.
20H through FFHReserved.
Max LVT Entry Shows the number of the highest order LVT entry. For the P6 family
processors, having 5 LVT entries, the Max LVT number is 4; for the
Pentium® processor, having 4 LVT entries, the Max LVT number is 3.

31 24 23 16 15 8 7 0

Reserved Max. LVT Reserved Version

Entry
Value after reset: 000N 00VVH
V = Version, N = # of LVT entries
Address: FEE0 0030H

Figure 7-15. Local APIC Version Register

7.5.16. APIC Bus Arbitration Mechanism and Protocol

Because only one message can be sent at a time on the APIC bus, the I/O APIC and local APICs
employ a “rotating priority” arbitration protocol to gain permission to send a message on the
APIC bus. One or more APICs may start sending their messages simultaneously. At the begin-
ning of every message, each APIC presents the type of the message it is sending and its current
arbitration priority on the APIC bus. This information is used for arbitration. After each arbitra-
tion cycle (within an arbitration round, only the potential winners keep driving the bus. By the
time all arbitration cycles are completed, there will be only one APIC left driving the bus. Once
a winner is selected, it is granted exclusive use of the bus, and will continue driving the bus to
send its actual message.
After each successfully transmitted message, all APICs increase their arbitration priority by 1.
The previous winner (that is, the one that has just successfully transmitted its message) assumes
a priority of 0 (lowest). An agent whose arbitration priority was 15 (highest) during arbitration,
but did not send a message, adopts the previous winner’s arbitration priority, incremented by 1.
Note that the arbitration protocol described above is slightly different if one of the APICs issues
a special End-Of-Interrupt (EOI). This high-priority message is granted the bus regardless of its
sender’s arbitration priority, unless more than one APIC issues an EOI message simultaneously.
In the latter case, the APICs sending the EOI messages arbitrate using their arbitration priorities.

7-36
 MULTIPLE-PROCESSOR MANAGEMENT

If the APICs are set up to use “lowest priority” arbitration (refer to Section 7.5.10., “Interrupt
Distribution Mechanisms”) and multiple APICs are currently executing at the lowest priority
(the value in the APR register), the arbitration priorities (unique values in the Arb ID register)
are used to break ties. All 8 bits of the APR are used for the lowest priority arbitration.

7.5.16.1. BUS MESSAGE FORMATS

The APICs use three types of messages: EOI message, short message, and non-focused lowest
priority message. The purpose of each type of message and its format are described below.
EOI Message. Local APICs send 14-cycle EOI messages to the I/O APIC to indicate that a level
triggered interrupt has been accepted by the processor. This interrupt, in turn, is a result of soft-
ware writing into the EOI register of the local APIC. Table 7-3 shows the cycles in an EOI
message.
The checksum is computed for cycles 6 through 9. It is a cumulative sum of the 2-bit (Bit1:Bit0)
logical data values. The carry out of all but the last addition is added to the sum. If any APIC
computes a different checksum than the one appearing on the bus in cycle 10, it signals an error,
driving 11 on the APIC bus during cycle 12. In this case, the APICs disregard the message. The
sending APIC will receive an appropriate error indication (refer to Section 7.5.17., “Error
Handling”) and resend the message. The status cycles are defined in Table 7-6.
Short Message. Short messages (21-cycles) are used for sending fixed, NMI, SMI, INIT, start-
up, ExtINT and lowest-priority-with-focus interrupts. Table 7-4 shows the cycles in a short
message.

Table 7-3. EOI Message (14 Cycles)

Cycle Bit1 Bit0
1 1 1 11 = EOI
2 ArbID3 0 Arbitration ID bits 3 through 0
3 ArbID2 0
4 ArbID1 0
5 ArbID0 0
6 V7 V6 Interrupt vector V7 - V0
7 V5 V4
8 V3 V2
9 V1 V0
10 C C Checksum for cycles 6 - 9
11 0 0
12 A A Status Cycle 0
13 A1 A1 Status Cycle 1
14 0 0 Idle

7-37
MULTIPLE-PROCESSOR MANAGEMENT

If the physical delivery mode is being used, then cycles 15 and 16 represent the APIC ID and
cycles 13 and 14 are considered don’t care by the receiver. If the logical delivery mode is being
used, then cycles 13 through 16 are the 8-bit logical destination field. For shorthands of “all-
incl-self” and “all-excl-self,” the physical delivery mode and an arbitration priority of 15
(D0:D3 = 1111) are used. The agent sending the message is the only one required to distinguish
between the two cases. It does so using internal information.
When using lowest priority delivery with an existing focus processor, the focus processor iden-
tifies itself by driving 10 during cycle 19 and accepts the interrupt. This is an indication to other
APICs to terminate arbitration. If the focus processor has not been found, the short message is
extended on-the-fly to the non-focused lowest-priority message. Note that except for the EOI
message, messages generating a checksum or an acceptance error (refer to Section 7.5.17.,
“Error Handling”) terminate after cycle 21.

Table 7-4. Short Message (21 Cycles)

Cycle Bit1 Bit0
1 0 1 0 1 = normal
2 ArbID3 0 Arbitration ID bits 3 through 0
3 ArbID2 0
4 ArbID1 0
5 ArbID0 0
6 DM M2 DM = Destination Mode
7 M1 M0 M2-M0 = Delivery mode
Cycle Bit1 Bit0
8 L TM L = Level, TM = Trigger Mode
9 V7 V6 V7-V0 = Interrupt Vector
10 V5 V4
11 V3 V2
12 V1 V0
13 D7 D6 D7-D0 = Destination
14 D5 D4
15 D3 D2
16 D1 D0
17 C C Checksum for cycles 6-16
18 0 0
19 A A Status cycle 0
20 A1 A1 Status cycle 1
21 0 0 Idle

Nonfocused Lowest Priority Message. These 34-cycle messages (refer to Table 7-5) are used
in the lowest priority delivery mode when a focus processor is not present. Cycles 1 through 20

7-38
 MULTIPLE-PROCESSOR MANAGEMENT

are same as for the short message. If during the status cycle (cycle 19) the state of the (A:A) flags
is 10B, a focus processor has been identified, and the short message format is used (refer to
Table 7-4). If the (A:A) flags are set to 00B, lowest priority arbitration is started and the 34-
cycles of the nonfocused lowest priority message are competed. For other combinations of status
flags, refer to Section 7.5.16.2., “APIC Bus Status Cycles”

Table 7-5. Nonfocused Lowest Priority Message (34 Cycles)

Cycle Bit0 Bit1
1 0 1 0 1 = normal
2 ArbID3 0 Arbitration ID bits 3 through 0
3 ArbID2 0
4 ArbID1 0
5 ArbID0 0
6 DM M2 DM = Destination mode
7 M1 M0 M2-M0 = Delivery mode
8 L TM L = Level, TM = Trigger Mode
9 V7 V6 V7-V0 = Interrupt Vector
10 V5 V4
11 V3 V2
12 V1 V0
13 D7 D6 D7-D0 = Destination
Cycle Bit0 Bit1
14 D5 D4
15 D3 D2
16 D1 D0
17 C C Checksum for cycles 6-16
18 0 0
19 A A Status cycle 0
20 A1 A1 Status cycle 1
21 P7 0 P7 - P0 = Inverted Processor Priority
22 P6 0
23 P5 0
24 P4 0
25 P3 0
26 P2 0
27 P1 0
28 P0 0
29 ArbID3 0 Arbitration ID 3 -0
30 ArbID2 0
31 ArbID1 0
32 ArbID0 0
33 A2 A2 Status Cycle
34 0 0 Idle

7-39
MULTIPLE-PROCESSOR MANAGEMENT

Cycles 21 through 28 are used to arbitrate for the lowest priority processor. The processors
participating in the arbitration drive their inverted processor priority on the bus. Only the local
APICs having free interrupt slots participate in the lowest priority arbitration. If no such APIC
exists, the message will be rejected, requiring it to be tried at a later time.
Cycles 29 through 32 are also used for arbitration in case two or more processors have the same
lowest priority. In the lowest priority delivery mode, all combinations of errors in cycle 33 (A2
A2) will set the “accept error” bit in the error status register (refer to Figure 7-16). Arbitration
priority update is performed in cycle 20, and is not affected by errors detected in cycle 33. Only
the local APIC that wins in the lowest priority arbitration, drives cycle 33. An error in cycle 33
will force the sender to resend the message.

7.5.16.2. APIC BUS STATUS CYCLES

Certain cycles within an APIC bus message are status cycles. During these cycles the status flags
(A:A) and (A1:A1) are examined. Table 7-6 shows how these status flags are interpreted,
depending on the current delivery mode and existence of a focus processor.

Table 7-6. APIC Bus Status Cycles Interpretation

Update
Delivery ArbID and Message
Mode A Status A1 Status A2 Status Cycle# Length Retry
EOI 00: CS_OK 10: Accept XX: Yes, 13 14 Cycle No
00: CS_OK 11: Retry XX: Yes, 13 14 Cycle Yes
00: CS_OK 0X: Accept Error XX: No 14 Cycle Yes
11: CS_Error XX: XX: No 14 Cycle Yes
10: Error XX: XX: No 14 Cycle Yes
01: Error XX: XX: No 14 Cycle Yes
Fixed 00: CS_OK 10: Accept XX: Yes, 20 21 Cycle No
00: CS_OK 11: Retry XX: Yes, 20 21 Cycle Yes
00: CS_OK 0X: Accept Error XX: No 21 Cycle Yes
11: CS_Error XX: XX: No 21 Cycle Yes
10: Error XX: XX: No 21 Cycle Yes
01: Error XX: XX: No 21 Cycle Yes
NMI, SMI, 00: CS_OK 10: Accept XX: Yes, 20 21 Cycle No
INIT, ExtINT,
Start-Up 00: CS_OK 11: Retry XX: Yes, 20 21 Cycle Yes
00: CS_OK 0X: Accept Error XX: No 21 Cycle Yes
11: CS_Error XX: XX: No 21 Cycle Yes
10: Error XX: XX: No 21 Cycle Yes
01: Error XX: XX: No 21 Cycle Yes

7-40
 MULTIPLE-PROCESSOR MANAGEMENT

Table 7-6. APIC Bus Status Cycles Interpretation (Contd.)

Lowest 00: CS_OK, NoFocus 11: Do Lowest 10: Accept Yes, 20 34 Cycle No
00: CS_OK, NoFocus 11: Do Lowest 11: Error Yes, 20 34 Cycle Yes

00: CS_OK, NoFocus 11: Do Lowest 0X: Error Yes, 20 34 Cycle Yes

00: CS_OK, NoFocus 10: End and Retry XX: Yes, 20 34 Cycle Yes
00: CS_OK, NoFocus 0X: Error XX: No 34 Cycle Yes
10: CS_OK, Focus XX: XX: Yes, 20 34 Cycle No
11: CS_Error XX: XX: No 21 Cycle Yes
01: Error XX: XX: No 21 Cycle Yes

7-41
MULTIPLE-PROCESSOR MANAGEMENT

7.5.17. Error Handling

The local APIC sets flags in the error status register (ESR) to record all the errors that is detects
(refer to Figure 7-16). The ESR is a read/write register and is reset after being written to by the
processor. A write to the ESR must be done just prior to reading the ESR to allow the register to
be updated. An error interrupt is generated when one of the error bits is set. Error bits are cumu-
lative. The ESR must be cleared by software after unmasking of the error interrupt entry in the
LVT is performed (by executing back-to-back a writes). If the software, however, wishes to
handle errors set in the register prior to unmasking, it should write and then read the ESR prior
or immediately after the unmasking.

31 8 7 6 5 4 3 2 1 0
Reserved

Illegal Register Address

Received Illegal Vector
Send Illegal Vector
Reserved
Receive Accept Error
Send Accept Error
Receive CS Error
Send CS Error

Address: FEE0 0280H

Value after reset: 0H

Figure 7-16. Error Status Register (ESR)

7-42
 MULTIPLE-PROCESSOR MANAGEMENT

The functions of the ESR flags are as follows:

Send CS Error Set when the local APIC detects a check sum error for a message
that was sent by it.
Receive CS Error Set when the local APIC detects a check sum error for a message
that was received by it.
Send Accept Error Set when the local APIC detects that a message it sent was not
accepted by any APIC on the bus.
Receive Accept Error Set when the local APIC detects that the message it received was not
accepted by any APIC on the bus, including itself.
Send Illegal Vector Set when the local APIC detects an illegal vector in the message that
it is sending on the bus.
Receive Illegal Vector Set when the local APIC detects an illegal vector in the message it
received, including an illegal vector code in the local vector table
interrupts and self-interrupts from ICR.
Illegal Reg. Address Set when the processor is trying to access a register that is not
(P6 Family Processors implemented in the P6 family processors’ local APIC register
Only) address space; that is, within FEE00000H (the APICBase MSR)
through FEE003FFH (the APICBase MSR plus 4K Bytes).

7.5.18. Timer
The local APIC unit contains a 32-bit programmable timer for use by the local processor. This
timer is configured through the timer register in the local vector table (refer to Figure 7-8). The
time base is derived from the processor’s bus clock, divided by a value specified in the divide
configuration register (refer to Figure 7-17). After reset, the timer is initialized to zero. The timer
supports one-shot and periodic modes. The timer can be configured to interrupt the local
processor with an arbitrary vector.

31 4 3 2 1 0
Reserved 0

Address: FEE0 03E0H

Value after reset: 0H
Divide Value (bits 0, 1 and 3)
000: Divide by 2
001: Divide by 4
010: Divide by 8
011: Divide by 16
100: Divide by 32
101: Divide by 64
110: Divide by 128
111: Divide by 1

Figure 7-17. Divide Configuration Register

7-43
MULTIPLE-PROCESSOR MANAGEMENT

The timer is started by programming its initial-count register, refer to Figure 7-18. The initial
count value is copied into the current-count register and count-down is begun. After the timer
reaches zero in one-shot mode, an interrupt is generated and the timer remains at its 0 value until
reprogrammed. In periodic mode, the current-count register is automatically reloaded from the
initial-count register when the count reaches 0 and the count-down is repeated. If during the
count-down process the initial-count register is set, the counting will restart and the new value
will be used. The initial-count register is read-write by software, while the current-count register
is read only.

31 0

Initial Count

Current Count

Address: Initial Count FEE0 0380H

Current Count FEE0 0390H
Value after reset: 0H

Figure 7-18. Initial Count and Current Count Registers

7.5.19. Software Visible Differences Between the Local APIC and

the 82489DX
The following local APIC features differ in their definitions from the 82489DX features:
• When the local APIC is disabled, its internal registers are not cleared. Instead, setting the
mask bits in the local vector table to disable the local APIC merely causes it to cease
accepting the bus messages except for INIT, SMI, NMI, and start-up. In the 82489DX,
when the local unit is disabled by resetting the bit 8 of the spurious vector register, all the
internal registers including the IRR, ISR and TMR are cleared and the mask bits in the
local vector tables are set to logical ones. In the disabled mode, 82489DX local unit will
accept only the reset deassert message.
• In the local APIC, NMI and INIT (except for INIT deassert) are always treated as edge
triggered interrupts, even if programmed otherwise. In the 82489DX these interrupts are
always level triggered.
• In the local APIC, interrupts generated through ICR messages are always treated as edge
triggered (except INIT Deassert). In the 82489DX, the ICR can be used to generate either
edge or level triggered interrupts.
• Logical Destination register the local APIC supports 8 bits, where it supports 32 bits for
the 82489DX.
• APIC ID register is 4 bits wide for the local APIC and 8 bits wide for the 82489DX.
• The remote read delivery mode provided in the 82489DX is not supported in the Intel
Architecture local APIC.

7-44
 MULTIPLE-PROCESSOR MANAGEMENT

7.5.20. Performance Related Differences between the Local APIC

and the 82489DX
For the 82489DX, in the lowest priority mode, all the target local APICs specified by the desti-
nation field participate in the lowest priority arbitration. Only those local APICs which have free
interrupt slots will participate in the lowest priority arbitration.

7.5.21. New Features Incorporated in the Pentium® and P6 Family

Processors Local APIC
The local APIC in the Pentium® and P6 family processors have the following new features not
found in the 82489DX.
• The local APIC supports cluster addressing in logical destination mode.
• Focus processor checking can be enabled/disabled in the local APIC.
• Interrupt input signal polarity can be programmed in the local APIC.
• The local APIC supports SMI through the ICR and I/O redirection table.
• The local APIC incorporates an error status register to log and report errors to the
processor.
In the P6 family processors, the local APIC incorporates an additional local vector table entry
to handle performance monitoring counter interrupts.

7.6. DUAL-PROCESSOR (DP) INITIALIZATION PROTOCOL

The Pentium® processor contains an internal dual-processing (DP) mechanism that permits two
processors to be initialized and configured for tightly coupled symmetric multiprocessing
(SMP). The DP initialization protocol supports the controlled booting and configuration of the
two Pentium® processors. When configuration has been completed, the two Pentium® processors
can share the processing load for the system and share the handling of interrupts received from
the system’s I/O APIC.
The Pentium® DP initialization protocol defines two processors:
• Primary processor (also called the bootstrap processor, BSP)—This processor boots itself,
configures the APIC environment, and starts the second processor.
• Secondary processor (also called the dual processor, DP)—This processor boots itself then
waits for a startup signal from the primary processor. Upon receiving the startup signal, it
completes its configuration.
Appendix C, Dual-Processor (DP) Bootup Sequence Example (Specific to Pentium® Proces-
sors) gives an example (with code) of the bootup sequence for two Pentium® processors oper-
ating in a DP configuration.

7-45
MULTIPLE-PROCESSOR MANAGEMENT

Appendix E, Programming the LINT0 and LINT1 Inputs describes (with code) how to program
the LINT[0:1] pins of the processor’s local APICs after a dual-processor configuration has been
completed.

7.7. MULTIPLE-PROCESSOR (MP) INITIALIZATION PROTOCOL

The Intel Architecture (beginning with the Pentium® Pro processors) defines a multiple-
processor (MP) initialization protocol, for use with both single- and multiple-processor systems.
(Here, multiple processors is defined as two or more processors.) The primary goals of this
protocol are as follows:
• To permit sequential or controlled booting of multiple processors (from 2 to 4) with no
dedicated system hardware. The initialization algorithm is not limited to 4 processors; it
can support supports from 1 to 15 processors in a multiclustered system when the APIC
busses are tied together. Larger systems are not supported.
• To be able to initiate the MP protocol without the need for a dedicated signal or BSP.
• To provide fault tolerance. No single processor is geographically designated the BSP. The
BSP is determined dynamically during initialization.
The following sections describe an MP initialization protocol.
Appendix D, Multiple-Processor (MP) Bootup Sequence Example (Specific to P6 Family
Processors) gives an example (with code) of the bootup sequence for two P6 family processors
operating in an MP configuration.
Appendix E, Programming the LINT0 and LINT1 Inputs describes (with code) how to program
the LINT[0:1] pins of the processor’s local APICs after an MP configuration has been
completed.

7.7.1. MP Initialization Protocol Requirements and Restrictions

The MP protocol imposes the following requirements and restrictions on the system:
• An APIC clock (APICLK) must be provided on all systems based on the P6 family
processors (excluding mobile processors and modules).
• All interrupt mechanisms must be disabled for the duration of the MP protocol algorithm,
including the window of time between the assertion of INIT# or receipt of an INIT IPI by
the application processors and the receipt of a STARTUP IPI by the application processors.
That is, requests generated by interrupting devices must not be seen by the local APIC unit
(on board the processor) until the completion of the algorithm. Failure to disable the
interrupt mechanisms may result in processor shutdown.
• The MP protocol should be initiated only after a hardware reset. After completion of the
protocol algorithm, a flag is set in the APIC base MSR of the BSP (APIC_BASE.BSP) to
indicate that it is the BSP. This flag is cleared for all other processors. If a processor or the
complete system is subject to an INIT sequence (either through the INIT# pin or an INIT

7-46
 MULTIPLE-PROCESSOR MANAGEMENT

IPI), then the MP protocol is not re-executed. Instead, each processor examines its BSP
flag to determine whether the processor should boot or wait for a STARTUP IPI.

7.7.2. MP Protocol Nomenclature

The MP initialization protocol defines two classes of processors:
• The bootstrap processor (BSP)—This primary processor is dynamically selected by the
MP initialization algorithm. After the BSP has been selected, it configures the APIC
environment, and starts the secondary processors, under software control.
• Application processors (APs)—These secondary processors are the remainder of the
processors in a MP system that were not selected as the BSP. The APs complete a minimal
self-configuration, then wait for a startup signal from the BSP processor. Upon receiving a
startup signal, an AP completes its configuration.
Table 7-7 describes the interrupt-style abbreviations that will be used through out the remaining
description of the MP initialization protocol. These IPIs do not define new interrupt messages.
They are messages that are special only by virtue of the time that they exist (that is, before the
RESET sequence is complete).

Table 7-7. Types of Boot Phase IPIs

Message Type Abbreviation Description
Boot Inter- BIPI An APIC serial bus message that Symmetric Multiprocessing
Processor Interrupt (SMP) agents use to dynamically determine a BSP after reset.
Final Boot Inter- FIPI An APIC serial bus message that the BSP issues before it fetches
Processor Interrupt from the reset vector. This message has the lowest priority of all
boot phase IPIs. When a BSP sees an FIPI that it issued, it
fetches the reset vector because no other boot phase IPIs can
follow an FIPI.
Startup Inter- SIPI Used to send a new reset vector to a Application Processor (non-
Processor Interrupt BSP) processor in an MP system.

Table 7-8 describes the various fields of each boot phase IPI.

Table 7-8. Boot Phase IPI Message Format

Destination Destination Trigger Destination Delivery Vector
Type Field Shorthand Mode Level Mode Mode (Hex)
BIPI Not used All including Edge Deassert Don’t Care Fixed 40 to 4E*
self (000)
FIPI Not used All including Edge Deassert Don’t Care Fixed 10 to 1E
self (000)
SIPI Used All allowed Edge Assert Physical or StartUp 00 to FF
Logical (110)

NOTE:
* For all P6 family processors.

7-47
MULTIPLE-PROCESSOR MANAGEMENT

For BIPI and FIPI messages, the lower 4 bits of the vector field are equal to the APIC ID of the
processor issuing the message. The upper 4 bits of the vector field of a BIPI or FIPI can be
thought of as the “generation ID” of the message. All processors that run symmetric to a P6
family processor will have a generation ID of 0100B or 4H. BIPIs in a system based on the P6
family processors will therefore use vector values ranging from 40H to 4EH (4FH can not be
used because FH is not a valid APIC ID).

7.7.3. Error Detection During the MP Initialization Protocol

Errors may occur on the APIC bus during the MP initialization phase. These errors may be tran-
sient or permanent and can be caused by a variety of failure mechanisms (for example, broken
traces, soft errors during bus usage, etc.). All serial bus related errors will result in an APIC
checksum or acceptance error.
The occurrence of an APIC error causes a processor shutdown.

7.7.4. Error Handling During the MP Initialization Protocol

The MP initialization protocol makes the following assumptions:
• If any errors are detected on the APIC bus during execution of the MP initialization
protocol, all processors will shutdown.
• In a system that conforms to Intel Architecture guidelines, a likely error (broken trace,
check sum error during transmission) will result in no more than one processor booting.
• The MP initialization protocol will be executed by processors even if they fail their BIST
sequences.

7.7.5. MP Initialization Protocol Algorithm

The MP initialization protocol uses the message passing capabilities of the processor’s local
APIC to dynamically determine a boot strap processor (BSP). The algorithm used essentially
implements a “race for the flag” mechanism using the APIC bus for atomicity.
The MP initialization algorithm is based on the fact that one and only one message is allowed
to exist on the APIC bus at a given time and that once the message is issued, it will complete
(APIC messages are atomic). Another feature of the APIC architecture that is used in the initial-
ization algorithm is the existence of a round-robin priority mechanism between all agents that
use the APIC bus.
The MP initialization protocol algorithm performs the following operations in a SMP system
(refer to Figure 7-19):
1. After completing their internal BISTs, all processors start their MP initialization protocol
sequence by issuing BIPIs to “all including self” (at time t=0). The four least significant
bits of the vector field of the IPI contain each processor's APIC ID. The APIC hardware

7-48
 MULTIPLE-PROCESSOR MANAGEMENT

observes the BNR# (block next request) pin to guarantee that the initial BIPI is not issued
on the APIC bus until the BIST sequence is completed for all processors in the system.
2. When the first BIPI completes (at time t=1), the APIC hardware (in each processor)
propagates an interrupt to the processor core to indicate the arrival of the BIPI.
3. The processor compares the four least significant bits of the BIPI’s vector field to the
processor's APIC ID. A match indicates that the processor should be the BSP and continue
the initialization sequence. If the APIC ID fails to match the BIPIs vector field, the
processor is essentially the “loser” or not the BSP. The processor then becomes an
application processor and should enter a “wait for SIPI” loop.
4. The winner (the BSP) issues an FIPI. The FIPI is issued to “all including self” and is
guaranteed to be the last IPI on the APIC bus during the initialization sequence. This is due
to the fact that the round-robin priority mechanism forces the winning APIC agent's (the
BSPs) arbitration priority to 0. The FIPI is therefore issued by a priority 0 agent and has to
wait until all other agents have issued their BIPI's. When the BSP receives the FIPI that it
issued (t=5), it will start fetching code at the reset vector (Intel Architecture address).

System (CPU) Bus

P6 Family P6 Family P6 Family P6 Family

Processor A Processor B Processor C Processor D

APIC Bus
t=0 t=1 t=2 t=3 t=4 t=5

BIPI.A BIPI.B BIPI.C BIPI.D FIPI

Serial Bus Activity

Figure 7-19. SMP System

5. All application processors (non-BSP processors) remain in a “halted” state and can only be
woken up by SIPIs issued by another processor (note an AP in the startup IPI loop will also
respond to BINIT and snoops).

7-49
 8
Processor
Management and
Initialization
 PROCESSOR MANAGEMENT AND INITIALIZATION

CHAPTER 8
PROCESSOR MANAGEMENT AND
INITIALIZATION

This chapter describes the facilities provided for managing processor wide functions and for
initializing the processor. The subjects covered include: processor initialization, FPU initializa-
tion, processor configuration, feature determination, mode switching, the MSRs (in the
Pentium® and P6 family processors), and the MTRRs (in the P6 family processors).

8.1. INITIALIZATION OVERVIEW

Following power-up or an assertion of the RESET# pin, each processor on the system bus
performs a hardware initialization of the processor (known as a hardware reset) and an optional
built-in self-test (BIST). A hardware reset sets each processor’s registers to a known state and
places the processor in real-address mode. It also invalidates the internal caches, translation
lookaside buffers (TLBs) and the branch target buffer (BTB). At this point, the action taken
depends on the processor family:
• P6 family processors—All the processors on the system bus (including a single processor
in a uniprocessor system) execute the multiple processor (MP) initialization protocol
across the APIC bus. The processor that is selected through this protocol as the bootstrap
processor (BSP) then immediately starts executing software-initialization code in the
current code segment beginning at the offset in the EIP register. The application (non-BSP)
processors (AP) go into a halt state while the BSP is executing initialization code. Refer to
Section 7.7., “Multiple-Processor (MP) Initialization Protocol” in Chapter 7, Multiple-
Processor Management for more details. Note that in a uniprocessor system, the single P6
family processor automatically becomes the BSP.
• Pentium® processors—In either a single- or dual- processor system, a single Pentium®
processor is always pre-designated as the primary processor. Following a reset, the primary
processor behaves as follows in both single- and dual-processor systems. Using the dual-
processor (DP) ready initialization protocol, the primary processor immediately starts
executing software-initialization code in the current code segment beginning at the offset
in the EIP register. The secondary processor (if there is one) goes into a halt state. (Refer to
Section 7.6., “Dual-Processor (DP) Initialization Protocol” in Chapter 7, Multiple-
Processor Management for more details.)
• Intel486™ processor—The primary processor (or single processor in a uniprocessor
system) immediately starts executing software-initialization code in the current code
segment beginning at the offset in the EIP register. (The Intel486™ does not automatically
execute a DP or MP initialization protocol to determine which processor is the primary
processor.)
The software-initialization code performs all system-specific initialization of the BSP or
primary processor and the system logic.

8-1
PROCESSOR MANAGEMENT AND INITIALIZATION

At this point, for MP (or DP) systems, the BSP (or primary) processor wakes up each AP (or
secondary) processor to enable those processors to execute self-configuration code.
When all processors are initialized, configured, and synchronized, the BSP or primary processor
begins executing an initial operating-system or executive task.
The floating-point unit (FPU) is also initialized to a known state during hardware reset. FPU
software initialization code can then be executed to perform operations such as setting the preci-
sion of the FPU and the exception masks. No special initialization of the FPU is required to
switch operating modes.
Asserting the INIT# pin on the processor invokes a similar response to a hardware reset. The
major difference is that during an INIT, the internal caches, MSRs, MTRRs, and FPU state are
left unchanged (although, the TLBs and BTB are invalidated as with a hardware reset). An INIT
provides a method for switching from protected to real-address mode while maintaining the
contents of the internal caches.

8.1.1. Processor State After Reset

Table 8-1 shows the state of the flags and other registers following power-up for the Pentium®
Pro, Pentium®, and Intel486™ processors. The state of control register CR0 is 60000010H (refer
to Figure 8-1), which places the processor is in real-address mode with paging disabled.

8.1.2. Processor Built-In Self-Test (BIST)

Hardware may request that the BIST be performed at power-up. The EAX register is cleared
(0H) if the processor passes the BIST. A nonzero value in the EAX register after the BIST indi-
cates that a processor fault was detected. If the BIST is not requested, the contents of the EAX
register after a hardware reset is 0H.
The overhead for performing a BIST varies between processor families. For example, the BIST
takes approximately 5.5 million processor clock periods to execute on the Pentium® Pro
processor. (This clock count is model-specific, and Intel reserves the right to change the exact
number of periods, for any of the Intel Architecture processors, without notification.)

8-2
 PROCESSOR MANAGEMENT AND INITIALIZATION

Table 8-1. 32-Bit Intel Architecture Processor States

Following Power-up, Reset, or INIT
Register P6 Family Processors Pentium® Processor Intel486™ Processor
EFLAGS1 00000002H 00000002H 00000002H
EIP 0000FFF0H 0000FFF0H 0000FFF0H
2 2
CR0 60000010H 60000010H 60000010H2
CR2, CR3, CR4 00000000H 00000000H 00000000H
®
MXCSR Pentium III processor only- NA NA
Pwr up or Reset: 1F80H
FINIT/FNINIT: Unchanged
CS Selector = F000H Selector = F000H Selector = F000H
Base = FFFF0000H Base = FFFF0000H Base = FFFF0000H
Limit = FFFFH Limit = FFFFH Limit = FFFFH
AR = Present, R/W, AR = Present, R/W, AR = Present, R/W,
Accessed Accessed Accessed
SS, DS, ES, FS, Selector = 0000H Selector = 0000H Selector = 0000H
GS Base = 00000000H Base = 00000000H Base = 00000000H
Limit = FFFFH Limit = FFFFH Limit = FFFFH
AR = Present, R/W, AR = Present, R/W, AR = Present, R/W,
Accessed Accessed Accessed
EDX 000006xxH 000005xxH 000004xxH
3 3
EAX 0 0 03
EBX, ECX, ESI, 00000000H 00000000H 00000000H
EDI, EBP, ESP
MM0 through Pentium® Pro processor - Pwr up or Reset: NA
MM74 NA 0000000000000000H
Pentium® II and Pentium® III FINIT/FNINIT: Unchanged
processor -
Pwr up or Reset:
0000000000000000H
FINIT/FNINIT: Unchanged
XMM0 through Pentium® III processor only- NA NA
XMM75 Pwr up or Reset:
0000000000000000H
FINIT/FNINIT: Unchanged
ST0 through Pwr up or Reset: +0.0 Pwr up or Reset: +0.0 Pwr up or Reset: +0.0
ST74 FINIT/FNINIT: Unchanged FINIT/FNINIT: Unchanged FINIT/FNINIT: Unchanged
FPU Control Pwr up or Reset: 0040H Pwr up or Reset: 0040H Pwr up or Reset: 0040H
Word4 FINIT/FNINIT: 037FH FINIT/FNINIT: 037FH FINIT/FNINIT: 037FH
FPU Status Pwr up or Reset: 0000H Pwr up or Reset: 0000H Pwr up or Reset: 0000H
Word4 FINIT/FNINIT: 0000H FINIT/FNINIT: 0000H FINIT/FNINIT: 0000H
FPU Tag Word4 Pwr up or Reset: 5555H Pwr up or Reset: 5555H Pwr up or Reset: 5555H
FINIT/FNINIT: FFFFH FINIT/FNINIT: FFFFH FINIT/FNINIT: FFFFH
FPU Data Pwr up or Reset: 0000H Pwr up or Reset: 0000H Pwr up or Reset: 0000H
Operand and CS FINIT/FNINIT: 0000H FINIT/FNINIT: 0000H FINIT/FNINIT: 0000H
Seg. Selectors4

8-3
PROCESSOR MANAGEMENT AND INITIALIZATION

Table 8-1. 32-Bit Intel Architecture Processor States

Following Power-up, Reset, or INIT (Contd.)
Register P6 Family Processors Pentium® Processor Intel486™ Processor
FPU Data Pwr up or Reset: Pwr up or Reset: Pwr up or Reset:
Operand and 00000000H 00000000H 00000000H
Inst. Pointers4 FINIT/FNINIT: 00000000H FINIT/FNINIT: 00000000H FINIT/FNINIT: 00000000H
GDTR,IDTR Base = 00000000H Base = 00000000H Base = 00000000H
Limit = FFFFH Limit = FFFFH Limit = FFFFH
AR = Present, R/W AR = Present, R/W AR = Present, R/W
LDTR, Task Selector = 0000H Selector = 0000H Selector = 0000H
Register Base = 00000000H Base = 00000000H Base = 00000000H
Limit = FFFFH Limit = FFFFH Limit = FFFFH
AR = Present, R/W AR = Present, R/W AR = Present, R/W
DR0, DR1, DR2, 00000000H 00000000H 00000000H
DR3
DR6 FFFF0FF0H FFFF0FF0H FFFF1FF0H
DR7 00000400H 00000400H 00000000H
Time-Stamp Power up or Reset: 0H Power up or Reset: 0H Not Implemented
Counter INIT: Unchanged INIT: Unchanged
Perf. Counters Power up or Reset: 0H Power up or Reset: 0H Not Implemented
and Event INIT: Unchanged INIT: Unchanged
Select
All Other MSRs Pwr up or Reset: Pwr up or Reset: Not Implemented
Undefined Undefined
INIT: Unchanged INIT: Unchanged
Data and Code Invalid Invalid Invalid
Cache, TLBs
Fixed MTRRs Pwr up or Reset: Disabled Not Implemented Not Implemented
INIT: Unchanged
Variable MTRRs Pwr up or Reset: Disabled Not Implemented Not Implemented
INIT: Unchanged
Machine-Check Pwr up or Reset: Not Implemented Not Implemented
Architecture Undefined
INIT: Unchanged
APIC Pwr up or Reset: Enabled Pwr up or Reset: Enabled Not Implemented
INIT: Unchanged INIT: Unchanged

NOTES:
1. The 10 most-significant bits of the EFLAGS register are undefined following a reset. Software should not
depend on the states of any of these bits.
2. The CD and NW flags are unchanged, bit 4 is set to 1, all other bits are cleared.
3. If Built-In Self-Test (BIST) is invoked on power up or reset, EAX is 0 only if all tests passed. (BIST cannot
be invoked during an INIT.)
4. The state of the FPU state and MMX™ registers is not changed by the execution of an INIT.
5. Available in the Pentium® III processor and Pentium® III Xeon™ processor only. The state of the SIMD
floating-point registers is not changed by the execution of an INIT.

8-4
 PROCESSOR MANAGEMENT AND INITIALIZATION

Paging disabled: 0
Caching disabled: 1
Not write-through disabled: 1

Alignment check disabled: 0

Write-protect disabled: 0

31 30 29 28 19 18 17 16 15 6 5 4 3 2 1 0

P C N A W N T E M P
1
G DW M P E S M P E

External FPU error reporting: 0

(Not used): 1
No task switch: 0
FPU instructions not trapped: 0
WAIT/FWAIT instructions not trapped: 0
Real-address mode: 0
Reserved

Figure 8-1. Contents of CR0 Register after Reset

8.1.3. Model and Stepping Information

Following a hardware reset, the EDX register contains component identification and revision
information (refer to Figure 8-2). The device ID field is set to the value 6H, 5H, 4H, or 3H to
indicate a Pentium® Pro, Pentium®, Intel486™, or Intel386™ processor, respectively. Different
values may be returned for the various members of these Intel Architecture families. For
example the Intel386™ SX processor returns 23H in the device ID field. Binary object code can
be made compatible with other Intel processors by using this number to select the correct initial-
ization software.

31 14 13 12 11 8 7 4 3 0

Stepping
EDX Family Model
ID

Processor Type
Family (0110B for the Pentium® Pro Processor Family)
Model (Beginning with 0001B)
Reserved

Figure 8-2. Processor Type and Signature in the EDX Register after Reset

8-5
PROCESSOR MANAGEMENT AND INITIALIZATION

The stepping ID field contains a unique identifier for the processor’s stepping ID or revision
level. The upper word of EDX is reserved following reset.

8.1.4. First Instruction Executed

The first instruction that is fetched and executed following a hardware reset is located at physical
address FFFFFFF0H. This address is 16 bytes below the processor’s uppermost physical
address. The EPROM containing the software-initialization code must be located at this address.
The address FFFFFFF0H is beyond the 1-MByte addressable range of the processor while in
real-address mode. The processor is initialized to this starting address as follows. The CS
register has two parts: the visible segment selector part and the hidden base address part. In real-
address mode, the base address is normally formed by shifting the 16-bit segment selector value
4 bits to the left to produce a 20-bit base address. However, during a hardware reset, the segment
selector in the CS register is loaded with F000H and the base address is loaded with
FFFF0000H. The starting address is thus formed by adding the base address to the value in the
EIP register (that is, FFFF0000 + FFF0H = FFFFFFF0H).
The first time the CS register is loaded with a new value after a hardware reset, the processor
will follow the normal rule for address translation in real-address mode (that is, [CS base address
= CS segment selector * 16]). To insure that the base address in the CS register remains
unchanged until the EPROM based software-initialization code is completed, the code must not
contain a far jump or far call or allow an interrupt to occur (which would cause the CS selector
value to be changed).

8.2. FPU INITIALIZATION

Software-initialization code can determine the whether the processor contains or is attached to
an FPU by using the CPUID instruction. The code must then initialize the FPU and set flags in
control register CR0 to reflect the state of the FPU environment.
A hardware reset places the Pentium® processor FPU in the state shown in Table 8-1. This state
is different from the state the processor is placed in when executing an FINIT or FNINIT instruc-
tion (also shown in Table 8-1). If the FPU is to be used, the software-initialization code should
execute an FINIT/FNINIT instruction following a hardware reset. These instructions, tag all
data registers as empty, clear all the exception masks, set the TOP-of-stack value to 0, and select
the default rounding and precision controls setting (round to nearest and 64-bit precision).
If the processor is reset by asserting the INIT# pin, the FPU state is not changed.

8.2.1. Configuring the FPU Environment

Initialization code must load the appropriate values into the MP, EM, and NE flags of control
register CR0. These bits are cleared on hardware reset of the processor. Figure 8-2 shows the
suggested settings for these flags, depending on the Intel Architecture processor being initial-

8-6
 PROCESSOR MANAGEMENT AND INITIALIZATION

ized. Initialization code can test for the type of processor present before setting or clearing these
flags.

Table 8-2. Recommended Settings of EM and MP Flags on Intel Architecture Processors

EM MP NE Intel Architecture Processor
1 0 1 Intel486™ SX, Intel386™ DX, and Intel386™ SX processors
only, without the presence of a math coprocessor.
0 1 1 or 0* Pentium® Pro, Pentium®, Intel486™ DX, and Intel 487 SX
processors, and also Intel386™ DX and Intel386™ SX
processors when a companion math coprocessor is present.

NOTE:
* The setting of the NE flag depends on the operating system being used.

The EM flag determines whether floating-point instructions are executed by the FPU (EM is
cleared) or generate a device-not-available exception (#NM) so that an exception handler can
emulate the floating-point operation (EM = 1). Ordinarily, the EM flag is cleared when an FPU
or math coprocessor is present and set if they are not present. If the EM flag is set and no FPU,
math coprocessor, or floating-point emulator is present, the system will hang when a floating-
point instruction is executed.
The MP flag determines whether WAIT/FWAIT instructions react to the setting of the TS flag.
If the MP flag is clear, WAIT/FWAIT instructions ignore the setting of the TS flag; if the MP
flag is set, they will generate a device-not-available exception (#NM) if the TS flag is set. Gener-
ally, the MP flag should be set for processors with an integrated FPU and clear for processors
without an integrated FPU and without a math coprocessor present. However, an operating
system can choose to save the floating-point context at every context switch, in which case there
would be no need to set the MP bit.
Table 2-1 in Chapter 2, System Architecture Overview shows the actions taken for floating-point
and WAIT/FWAIT instructions based on the settings of the EM, MP, and TS flags.
The NE flag determines whether unmasked floating-point exceptions are handled by generating
a floating-point error exception internally (NE is set, native mode) or through an external inter-
rupt (NE is cleared). In systems where an external interrupt controller is used to invoke numeric
exception handlers (such as MS-DOS-based systems), the NE bit should be cleared.

8-7
PROCESSOR MANAGEMENT AND INITIALIZATION

8.2.2. Setting the Processor for FPU Software Emulation

Setting the EM flag causes the processor to generate a device-not-available exception (#NM)
and trap to a software exception handler whenever it encounters a floating-point instruction.
(Table 8-2 shows when it is appropriate to use this flag.) Setting this flag has two functions:
• It allows floating-point code to run on an Intel processor that neither has an integrated FPU
nor is connected to an external math coprocessor, by using a floating-point emulator.
• It allows floating-point code to be executed using a special or nonstandard floating-point
emulator, selected for a particular application, regardless of whether an FPU or math
coprocessor is present.
To emulate floating-point instructions, the EM, MP, and NE flag in control register CR0 should
be set as shown in Table 8-3.

Table 8-3. Software Emulation Settings of EM, MP, and NE Flags

CR0 Bit Value
EM 1
MP 0
NE 1

Regardless of the value of the EM bit, the Intel486™ SX processor generates a device-not-avail-
able exception (#NM) upon encountering any floating-point instruction.

8.3. CACHE ENABLING

The Intel Architecture processors (beginning with the Intel486™ processor) contain internal
instruction and data caches. These caches are enabled by clearing the CD and NW flags in
control register CR0. (They are set during a hardware reset.) Because all internal cache lines are
invalid following reset initialization, it is not necessary to invalidate the cache before enabling
caching. Any external caches may require initialization and invalidation using a system-specific
initialization and invalidation code sequence.
Depending on the hardware and operating system or executive requirements, additional config-
uration of the processor’s caching facilities will probably be required. Beginning with the
Intel486™ processor, page-level caching can be controlled with the PCD and PWT flags in
page-directory and page-table entries. For P6 family processors, the memory type range regis-
ters (MTRRs) control the caching characteristics of the regions of physical memory. (For the
Intel486™ and Pentium® processors, external hardware can be used to control the caching char-
acteristics of regions of physical memory.) Refer to Chapter 9, Memory Cache Control, for
detailed information on configuration of the caching facilities in the P6 family processors and
system memory.

8.4. MODEL-SPECIFIC REGISTERS (MSRS)

The P6 family processors and Pentium® processors contain model-specific registers (MSRs).
These registers are by definition implementation specific; that is, they are not guaranteed to be
8-8
 PROCESSOR MANAGEMENT AND INITIALIZATION

supported on future Intel Architecture processors and/or to have the same functions. The MSRs
are provided to control a variety of hardware- and software-related features, including:
• The performance-monitoring counters (refer to Section 15.6., “Performance-Monitoring
Counters”, in Chapter 15, Debugging and Performance Monitoring).
• (P6 family processors only.) Debug extensions (refer to Section 15.4., “Last Branch,
Interrupt, and Exception Recording”, in Chapter 15, Debugging and Performance
Monitoring).
• (P6 family processors only.) The machine-check exception capability and its accompa-
nying machine-check architecture (refer to Chapter 13, Machine-Check Architecture).
• (P6 family processors only.) The MTRRs (refer to Section 9.12., “Memory Type Range
Registers (MTRRs)”, in Chapter 9, Memory Cache Control).
The MSRs can be read and written to using the RDMSR and WRMSR instructions, respectively.
When performing software initialization of a Pentium® Pro or Pentium® processor, many of the
MSRs will need to be initialized to set up things like performance-monitoring events, run-time
machine checks, and memory types for physical memory.
Systems configured to implement FRC mode must write all of the processors’ internal MSRs to
deterministic values before performing either a read or read-modify-write operation using these
registers. The following is a list of MSRs that are not initialized by the processors’ reset
sequences.
• All fixed and variable MTRRs.
• All Machine Check Architecture (MCA) status registers.
• Microcode update signature register.
• All L2 cache initialization MSRs.
The list of available performance-monitoring counters for the Pentium® Pro and Pentium®
processors is given in Appendix A, Performance-Monitoring Events, and the list of available
MSRs for the Pentium® Pro processor is given in Appendix B, Model-Specific Registers. The
references earlier in this section show where the functions of the various groups of MSRs are
described in this manual.

8.5. MEMORY TYPE RANGE REGISTERS (MTRRS)

Memory type range registers (MTRRs) were introduced into the Intel Architecture with the
Pentium® Pro processor. They allow the type of caching (or no caching) to be specified in system
memory for selected physical address ranges. They allow memory accesses to be optimized for
various types of memory such as RAM, ROM, frame buffer memory, and memory-mapped I/O
devices.
In general, initializing the MTRRs is normally handled by the software initialization code or
BIOS and is not an operating system or executive function. At the very least, all the MTRRs
must be cleared to 0, which selects the uncached (UC) memory type. Refer to Section 9.12.,

8-9
PROCESSOR MANAGEMENT AND INITIALIZATION

“Memory Type Range Registers (MTRRs)”, in Chapter 9, Memory Cache Control, for detailed
information on the MTRRs.

8.6. SOFTWARE INITIALIZATION FOR REAL-ADDRESS MODE

OPERATION
Following a hardware reset (either through a power-up or the assertion of the RESET# pin) the
processor is placed in real-address mode and begins executing software initialization code from
physical address FFFFFFF0H. Software initialization code must first set up the necessary data
structures for handling basic system functions, such as a real-mode IDT for handling interrupts
and exceptions. If the processor is to remain in real-address mode, software must then load addi-
tional operating-system or executive code modules and data structures to allow reliable execu-
tion of application programs in real-address mode.
If the processor is going to operate in protected mode, software must load the necessary data
structures to operate in protected mode and then switch to protected mode. The protected-mode
data structures that must be loaded are described in Section 8.7., “Software Initialization for
Protected-Mode Operation”.

8.6.1. Real-Address Mode IDT

In real-address mode, the only system data structure that must be loaded into memory is the IDT
(also called the “interrupt vector table”). By default, the address of the base of the IDT is phys-
ical address 0H. This address can be changed by using the LIDT instruction to change the base
address value in the IDTR. Software initialization code needs to load interrupt- and exception-
handler pointers into the IDT before interrupts can be enabled.
The actual interrupt- and exception-handler code can be contained either in EPROM or RAM;
however, the code must be located within the 1-MByte addressable range of the processor in
real-address mode. If the handler code is to be stored in RAM, it must be loaded along with the
IDT.

8.6.2. NMI Interrupt Handling

The NMI interrupt is always enabled (except when multiple NMIs are nested). If the IDT and
the NMI interrupt handler need to be loaded into RAM, there will be a period of time following
hardware reset when an NMI interrupt cannot be handled. During this time, hardware must
provide a mechanism to prevent an NMI interrupt from halting code execution until the IDT and
the necessary NMI handler software is loaded.

8-10
 PROCESSOR MANAGEMENT AND INITIALIZATION

Here are two examples of how NMIs can be handled during the initial states of processor initial-
ization:
• A simple IDT and NMI interrupt handler can be provided in EPROM. This allows an NMI
interrupt to be handled immediately after reset initialization.
• The system hardware can provide a mechanism to enable and disable NMIs by passing the
NMI# signal through an AND gate controlled by a flag in an I/O port. Hardware can clear
the flag when the processor is reset, and software can set the flag when it is ready to handle
NMI interrupts.

8.7. SOFTWARE INITIALIZATION FOR PROTECTED-MODE

OPERATION
The processor is placed in real-address mode following a hardware reset. At this point in the
initialization process, some basic data structures and code modules must be loaded into physical
memory to support further initialization of the processor, as described in Section 8.6., “Software
Initialization for Real-Address Mode Operation”. Before the processor can be switched to
protected mode, the software initialization code must load a minimum number of protected
mode data structures and code modules into memory to support reliable operation of the
processor in protected mode. These data structures include the following:
• A protected-mode IDT.
• A GDT.
• A TSS.
• (Optional.) An LDT.
• If paging is to be used, at least one page directory and one page table.
• A code segment that contains the code to be executed when the processor switches to
protected mode.
• One or more code modules that contain the necessary interrupt and exception handlers.
Software initialization code must also initialize the following system registers before the
processor can be switched to protected mode:
• The GDTR.
• (Optional.) The IDTR. This register can also be initialized immediately after switching to
protected mode, prior to enabling interrupts.
• Control registers CR1 through CR4.
• (Pentium® Pro processor only.) The memory type range registers (MTRRs).
With these data structures, code modules, and system registers initialized, the processor can be
switched to protected mode by loading control register CR0 with a value that sets the PE flag
(bit 0).

8-11
PROCESSOR MANAGEMENT AND INITIALIZATION

8.7.1. Protected-Mode System Data Structures

The contents of the protected-mode system data structures loaded into memory during software
initialization, depend largely on the type of memory management the protected-mode operating-
system or executive is going to support: flat, flat with paging, segmented, or segmented with
paging.
To implement a flat memory model without paging, software initialization code must at a
minimum load a GDT with one code and one data-segment descriptor. A null descriptor in the
first GDT entry is also required. The stack can be placed in a normal read/write data segment,
so no dedicated descriptor for the stack is required. A flat memory model with paging also
requires a page directory and at least one page table (unless all pages are 4 MBytes in which case
only a page directory is required). Refer to Section 8.7.3., “Initializing Paging”
Before the GDT can be used, the base address and limit for the GDT must be loaded into the
GDTR register using an LGDT instruction.
A multisegmented model may require additional segments for the operating system, as well as
segments and LDTs for each application program. LDTs require segment descriptors in the
GDT. Some operating systems allocate new segments and LDTs as they are needed. This
provides maximum flexibility for handling a dynamic programming environment. However,
many operating systems use a single LDT for all tasks, allocating GDT entries in advance. An
embedded system, such as a process controller, might pre-allocate a fixed number of segments
and LDTs for a fixed number of application programs. This would be a simple and efficient way
to structure the software environment of a real-time system.

8.7.2. Initializing Protected-Mode Exceptions and Interrupts

Software initialization code must at a minimum load a protected-mode IDT with gate descriptor
for each exception vector that the processor can generate. If interrupt or trap gates are used, the
gate descriptors can all point to the same code segment, which contains the necessary exception
handlers. If task gates are used, one TSS and accompanying code, data, and task segments are
required for each exception handler called with a task gate.
If hardware allows interrupts to be generated, gate descriptors must be provided in the IDT for
one or more interrupt handlers.
Before the IDT can be used, the base address and limit for the IDT must be loaded into the IDTR
register using an LIDT instruction. This operation is typically carried out immediately after
switching to protected mode.

8.7.3. Initializing Paging

Paging is controlled by the PG flag in control register CR0. When this flag is clear (its state
following a hardware reset), the paging mechanism is turned off; when it is set, paging is
enabled. Before setting the PG flag, the following data structures and registers must be initial-
ized:

8-12
 PROCESSOR MANAGEMENT AND INITIALIZATION

• Software must load at least one page directory and one page table into physical memory.
The page table can be eliminated if the page directory contains a directory entry pointing to
itself (here, the page directory and page table reside in the same page), or if only 4-MByte
pages are used.
• Control register CR3 (also called the PDBR register) is loaded with the physical base
address of the page directory.
• (Optional) Software may provide one set of code and data descriptors in the GDT or in an
LDT for supervisor mode and another set for user mode.
With this paging initialization complete, paging is enabled and the processor is switched to
protected mode at the same time by loading control register CR0 with an image in which the PG
and PE flags are set. (Paging cannot be enabled before the processor is switched to protected
mode.)

8.7.4. Initializing Multitasking

If the multitasking mechanism is not going to be used and changes between privilege levels are
not allowed, it is not necessary load a TSS into memory or to initialize the task register.
If the multitasking mechanism is going to be used and/or changes between privilege levels are
allowed, software initialization code must load at least one TSS and an accompanying TSS
descriptor. (A TSS is required to change privilege levels because pointers to the privileged-level
0, 1, and 2 stack segments and the stack pointers for these stacks are obtained from the TSS.)
TSS descriptors must not be marked as busy when they are created; they should be marked busy
by the processor only as a side-effect of performing a task switch. As with descriptors for LDTs,
TSS descriptors reside in the GDT.
After the processor has switched to protected mode, the LTR instruction can be used to load a
segment selector for a TSS descriptor into the task register. This instruction marks the TSS
descriptor as busy, but does not perform a task switch. The processor can, however, use the TSS
to locate pointers to privilege-level 0, 1, and 2 stacks. The segment selector for the TSS must be
loaded before software performs its first task switch in protected mode, because a task switch
copies the current task state into the TSS.
After the LTR instruction has been executed, further operations on the task register are
performed by task switching. As with other segments and LDTs, TSSs and TSS descriptors can
be either pre-allocated or allocated as needed.

8.8. MODE SWITCHING

To use the processor in protected mode, a mode switch must be performed from real-address
mode. Once in protected mode, software generally does not need to return to real-address mode.
To run software written to run in real-address mode (8086 mode), it is generally more convenient
to run the software in virtual-8086 mode, than to switch back to real-address mode.

8-13
PROCESSOR MANAGEMENT AND INITIALIZATION

8.8.1. Switching to Protected Mode

Before switching to protected mode, a minimum set of system data structures and code modules
must be loaded into memory, as described in Section 8.7., “Software Initialization for Protected-
Mode Operation”. Once these tables are created, software initialization code can switch into
protected mode.
Protected mode is entered by executing a MOV CR0 instruction that sets the PE flag in the CR0
register. (In the same instruction, the PG flag in register CR0 can be set to enable paging.)
Execution in protected mode begins with a CPL of 0.
The 32-bit Intel Architecture processors have slightly different requirements for switching to
protected mode. To insure upwards and downwards code compatibility with all 32-bit Intel
Architecture processors, it is recommended that the following steps be performed:
1. Disable interrupts. A CLI instruction disables maskable hardware interrupts. NMI
interrupts can be disabled with external circuitry. (Software must guarantee that no
exceptions or interrupts are generated during the mode switching operation.)
2. Execute the LGDT instruction to load the GDTR register with the base address of the
GDT.
3. Execute a MOV CR0 instruction that sets the PE flag (and optionally the PG flag) in
control register CR0.
4. Immediately following the MOV CR0 instruction, execute a far JMP or far CALL
instruction. (This operation is typically a far jump or call to the next instruction in the
instruction stream.)
The JMP or CALL instruction immediately after the MOV CR0 instruction changes the
flow of execution and serializes the processor.
If paging is enabled, the code for the MOV CR0 instruction and the JMP or CALL
instruction must come from a page that is identity mapped (that is, the linear address before
the jump is the same as the physical address after paging and protected mode is enabled).
The target instruction for the JMP or CALL instruction does not need to be identity
mapped.
5. If a local descriptor table is going to be used, execute the LLDT instruction to load the
segment selector for the LDT in the LDTR register.
6. Execute the LTR instruction to load the task register with a segment selector to the initial
protected-mode task or to a writable area of memory that can be used to store TSS
information on a task switch.
7. After entering protected mode, the segment registers continue to hold the contents they had
in real-address mode. The JMP or CALL instruction in step 4 resets the CS register.
Perform one of the following operations to update the contents of the remaining segment
registers.
— Reload segment registers DS, SS, ES, FS, and GS. If the ES, FS, and/or GS registers
are not going to be used, load them with a null selector.

8-14
 PROCESSOR MANAGEMENT AND INITIALIZATION

— Perform a JMP or CALL instruction to a new task, which automatically resets the
values of the segment registers and branches to a new code segment.
8. Execute the LIDT instruction to load the IDTR register with the address and limit of the
protected-mode IDT.
9. Execute the STI instruction to enable maskable hardware interrupts and perform the
necessary hardware operation to enable NMI interrupts.
Random failures can occur if other instructions exist between steps 3 and 4 above. Failures will
be readily seen in some situations, such as when instructions that reference memory are inserted
between steps 3 and 4 while in System Management mode.

8.8.2. Switching Back to Real-Address Mode

The processor switches back to real-address mode if software clears the PE bit in the CR0
register with a MOV CR0 instruction. A procedure that re-enters real-address mode should
perform the following steps:
1. Disable interrupts. A CLI instruction disables maskable hardware interrupts. NMI
interrupts can be disabled with external circuitry.
2. If paging is enabled, perform the following operations:
— Transfer program control to linear addresses that are identity mapped to physical
addresses (that is, linear addresses equal physical addresses).
— Insure that the GDT and IDT are in identity mapped pages.
— Clear the PG bit in the CR0 register.
— Move 0H into the CR3 register to flush the TLB.
3. Transfer program control to a readable segment that has a limit of 64 KBytes (FFFFH).
This operation loads the CS register with the segment limit required in real-address mode.
4. Load segment registers SS, DS, ES, FS, and GS with a selector for a descriptor containing
the following values, which are appropriate for real-address mode:
— Limit = 64 KBytes (0FFFFH)
— Byte granular (G = 0)
— Expand up (E = 0)
— Writable (W = 1)
— Present (P = 1)
— Base = any value
The segment registers must be loaded with nonnull segment selectors or the segment
registers will be unusable in real-address mode. Note that if the segment registers are not

8-15
PROCESSOR MANAGEMENT AND INITIALIZATION

reloaded, execution continues using the descriptor attributes loaded during protected
mode.
5. Execute an LIDT instruction to point to a real-address mode interrupt table that is within
the 1-MByte real-address mode address range.
6. Clear the PE flag in the CR0 register to switch to real-address mode.
7. Execute a far JMP instruction to jump to a real-address mode program. This operation
flushes the instruction queue and loads the appropriate base and access rights values in the
CS register.
8. Load the SS, DS, ES, FS, and GS registers as needed by the real-address mode code. If any
of the registers are not going to be used in real-address mode, write 0s to them.
9. Execute the STI instruction to enable maskable hardware interrupts and perform the
necessary hardware operation to enable NMI interrupts.

NOTE
All the code that is executed in steps 1 through 9 must be in a single page and
the linear addresses in that page must be identity mapped to physical
addresses.

8.9. INITIALIZATION AND MODE SWITCHING EXAMPLE

This section provides an initialization and mode switching example that can be incorporated into
an application. This code was originally written to initialize the Intel386™ processor, but it will
execute successfully on the Pentium® Pro, Pentium®, and Intel486™ processors. The code in this
example is intended to reside in EPROM and to run following a hardware reset of the processor.
The function of the code is to do the following:
• Establish a basic real-address mode operating environment.
• Load the necessary protected-mode system data structures into RAM.
• Load the system registers with the necessary pointers to the data structures and the
appropriate flag settings for protected-mode operation.
• Switch the processor to protected mode.
Figure 8-3 shows the physical memory layout for the processor following a hardware reset and
the starting point of this example. The EPROM that contains the initialization code resides at the
upper end of the processor’s physical memory address range, starting at address FFFFFFFFH
and going down from there. The address of the first instruction to be executed is at FFFFFFF0H,
the default starting address for the processor following a hardware reset.
The main steps carried out in this example are summarized in Table 8-4. The source listing for
the example (with the filename STARTUP.ASM) is given in Example 8-1. The line numbers
given in Table 8-4 refer to the source listing.

8-16
 PROCESSOR MANAGEMENT AND INITIALIZATION

The following are some additional notes concerning this example:

• When the processor is switched into protected mode, the original code segment base-
address value of FFFF0000H (located in the hidden part of the CS register) is retained and
execution continues from the current offset in the EIP register. The processor will thus
continue to execute code in the EPROM until a far jump or call is made to a new code
segment, at which time, the base address in the CS register will be changed.
• Maskable hardware interrupts are disabled after a hardware reset and should remain
disabled until the necessary interrupt handlers have been installed. The NMI interrupt is
not disabled following a reset. The NMI# pin must thus be inhibited from being asserted
until an NMI handler has been loaded and made available to the processor.
• The use of a temporary GDT allows simple transfer of tables from the EPROM to
anywhere in the RAM area. A GDT entry is constructed with its base pointing to address 0
and a limit of 4 GBytes. When the DS and ES registers are loaded with this descriptor, the
temporary GDT is no longer needed and can be replaced by the application GDT.
• This code loads one TSS and no LDTs. If more TSSs exist in the application, they must be
loaded into RAM. If there are LDTs they may be loaded as well.

After Reset
FFFF FFFFH
[CS.BASE+EIP] FFFF FFF0H

64K EPROM

EIP = 0000 FFF0H

CS.BASE = FFFF 0000H FFFF 0000H
DS.BASE = 0H
ES.BASE = 0H
SS.BASE = 0H
ESP = 0H

[SP, DS, SS, ES] 0

Figure 8-3. Processor State After Reset

8-17
PROCESSOR MANAGEMENT AND INITIALIZATION

Table 8-4. Main Initialization Steps in STARTUP.ASM Source Listing

STARTUP.ASM
Line Numbers
From To Description
157 157 Jump (short) to the entry code in the EPROM
162 169 Construct a temporary GDT in RAM with one entry:
0 - null
1 - R/W data segment, base = 0, limit = 4 GBytes
171 172 Load the GDTR to point to the temporary GDT
174 177 Load CR0 with PE flag set to switch to protected mode
179 181 Jump near to clear real mode instruction queue
184 186 Load DS, ES registers with GDT[1] descriptor, so both point to the entire
physical memory space
188 195 Perform specific board initialization that is imposed by the new protected
mode
196 218 Copy the application’s GDT from ROM into RAM
220 238 Copy the application’s IDT from ROM into RAM
241 243 Load application’s GDTR
244 245 Load application’s IDTR
247 261 Copy the application’s TSS from ROM into RAM
263 267 Update TSS descriptor and other aliases in GDT (GDT alias or IDT alias)
277 277 Load the task register (without task switch) using LTR instruction
282 286 Load SS, ESP with the value found in the application’s TSS
287 287 Push EFLAGS value found in the application’s TSS
288 288 Push CS value found in the application’s TSS
289 289 Push EIP value found in the application’s TSS
290 293 Load DS, ES with the value found in the application’s TSS
296 296 Perform IRET; pop the above values and enter the application code

8-18
 PROCESSOR MANAGEMENT AND INITIALIZATION

8.9.1. Assembler Usage

In this example, the Intel assembler ASM386 and build tools BLD386 are used to assemble and
build the initialization code module. The following assumptions are used when using the Intel
ASM386 and BLD386 tools.
• The ASM386 will generate the right operand size opcodes according to the code-segment
attribute. The attribute is assigned either by the ASM386 invocation controls or in the
code-segment definition.
• If a code segment that is going to run in real-address mode is defined, it must be set to a
USE 16 attribute. If a 32-bit operand is used in an instruction in this code segment (for
example, MOV EAX, EBX), the assembler automatically generates an operand prefix for
the instruction that forces the processor to execute a 32-bit operation, even though its
default code-segment attribute is 16-bit.
• Intel’s ASM386 assembler allows specific use of the 16- or 32-bit instructions, for
example, LGDTW, LGDTD, IRETD. If the generic instruction LGDT is used, the default-
segment attribute will be used to generate the right opcode.

8.9.2. STARTUP.ASM Listing

The source code listing to move the processor into protected mode is provided in Example 8-1.
This listing does not include any opcode and offset information.

Example 8-1. STARTUP.ASM

MS-DOS* 5.0(045-N) 386(TM) MACRO ASSEMBLER STARTUP 09:44:51 08/19/92 PAGE 1

MS-DOS 5.0(045-N) 386(TM) MACRO ASSEMBLER V4.0, ASSEMBLY OF MODULE

STARTUP
OBJECT MODULE PLACED IN startup.obj
ASSEMBLER INVOKED BY: f:\386tools\ASM386.EXE startup.a58 pw (132 )

LINE SOURCE

1 NAME STARTUP
2
3 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
4 ;
5 ; ASSUMPTIONS:
6 ;
7 ; 1. The bottom 64K of memory is ram, and can be used for
8 ; scratch space by this module.
9 ;
10 ; 2. The system has sufficient free usable ram to copy the
11 ; initial GDT, IDT, and TSS

8-19
PROCESSOR MANAGEMENT AND INITIALIZATION

12 ;
13 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
14
15 ; configuration data - must match with build definition
16
17 CS_BASE EQU 0FFFF0000H
18
19 ; CS_BASE is the linear address of the segment STARTUP_CODE
20 ; - this is specified in the build language file
21
22 RAM_START EQU 400H
23
24 ; RAM_START is the start of free, usable ram in the linear
25 ; memory space. The GDT, IDT, and initial TSS will be
26 ; copied above this space, and a small data segment will be
27 ; discarded at this linear address. The 32-bit word at
28 ; RAM_START will contain the linear address of the first
29 ; free byte above the copied tables - this may be useful if
30 ; a memory manager is used.
31
32 TSS_INDEX EQU 10
33
34 ; TSS_INDEX is the index of the TSS of the first task to
35 ; run after startup
36
37
38 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
39
40 ; ------------------------- STRUCTURES and EQU ---------------
41 ; structures for system data
42
43 ; TSS structure
44 TASK_STATE STRUC
45 link DW ?
46 link_h DW ?
47 ESP0 DD ?
48 SS0 DW ?
49 SS0_h DW ?
50 ESP1 DD ?
51 SS1 DW ?
52 SS1_h DW ?
53 ESP2 DD ?
54 SS2 DW ?
55 SS2_h DW ?
56 CR3_reg DD ?
57 EIP_reg DD ?
58 EFLAGS_reg DD ?

8-20
 PROCESSOR MANAGEMENT AND INITIALIZATION

59 EAX_reg DD ?
60 ECX_reg DD ?
61 EDX_reg DD ?
62 EBX_reg DD ?
63 ESP_reg DD ?
64 EBP_reg DD ?
65 ESI_reg DD ?
66 EDI_reg DD ?
67 ES_reg DW ?
68 ES_h DW ?
69 CS_reg DW ?
70 CS_h DW ?
71 SS_reg DW ?
72 SS_h DW ?
73 DS_reg DW ?
74 DS_h DW ?
75 FS_reg DW ?
76 FS_h DW ?
77 GS_reg DW ?
78 GS_h DW ?
79 LDT_reg DW ?
80 LDT_h DW ?
81 TRAP_reg DW ?
82 IO_map_base DW ?
83 TASK_STATE ENDS
84
85 ; basic structure of a descriptor
86 DESC STRUC
87 lim_0_15 DW ?
88 bas_0_15 DW ?
89 bas_16_23 DB ?
90 access DB ?
91 gran DB ?
92 bas_24_31 DB ?
93 DESC ENDS
94
95 ; structure for use with LGDT and LIDT instructions
96 TABLE_REG STRUC
97 table_lim DW ?
98 table_linear DD ?
99 TABLE_REG ENDS
100
101 ; offset of GDT and IDT descriptors in builder generated GDT
102 GDT_DESC_OFF EQU 1*SIZE(DESC)
103 IDT_DESC_OFF EQU 2*SIZE(DESC)
104
105 ; equates for building temporary GDT in RAM

8-21
PROCESSOR MANAGEMENT AND INITIALIZATION

106 LINEAR_SEL EQU 1*SIZE (DESC)

107 LINEAR_PROTO_LO EQU 00000FFFFH ; LINEAR_ALIAS
108 LINEAR_PROTO_HI EQU 000CF9200H
109
110 ; Protection Enable Bit in CR0
111 PE_BIT EQU 1B
112
113 ; ------------------------------------------------------------
114
115 ; ------------------------- DATA SEGMENT----------------------
116
117 ; Initially, this data segment starts at linear 0, according
118 ; to the processor’s power-up state.
119
120 STARTUP_DATA SEGMENT RW
121
122 free_mem_linear_base LABEL DWORD
123 TEMP_GDT LABEL BYTE ; must be first in segment
124 TEMP_GDT_NULL_DESC DESC <>
125 TEMP_GDT_LINEAR_DESC DESC <>
126
127 ; scratch areas for LGDT and LIDT instructions
128 TEMP_GDT_SCRATCH TABLE_REG <>
129 APP_GDT_RAM TABLE_REG <>
130 APP_IDT_RAM TABLE_REG <>
131 ; align end_data
132 fill DW ?
133
134 ; last thing in this segment - should be on a dword boundary
135 end_data LABEL BYTE
136
137 STARTUP_DATA ENDS
138 ; ------------------------------------------------------------
139
140
141 ; ------------------------- CODE SEGMENT----------------------
142 STARTUP_CODE SEGMENT ER PUBLIC USE16
143
144 ; filled in by builder
145 PUBLIC GDT_EPROM
146 GDT_EPROM TABLE_REG <>
147
148 ; filled in by builder
149 PUBLIC IDT_EPROM
150 IDT_EPROM TABLE_REG <>
151
152 ; entry point into startup code - the bootstrap will vector

8-22
 PROCESSOR MANAGEMENT AND INITIALIZATION

153 ; here with a near JMP generated by the builder. This

154 ; label must be in the top 64K of linear memory.
155
156 PUBLIC STARTUP
157 STARTUP:
158
159 ; DS,ES address the bottom 64K of flat linear memory
160 ASSUME DS:STARTUP_DATA, ES:STARTUP_DATA
161 ; See Figure 8-4
162 ; load GDTR with temporary GDT
163 LEA EBX,TEMP_GDT ; build the TEMP_GDT in low ram,
164 MOV DWORD PTR [EBX],0 ; where we can address
165 MOV DWORD PTR [EBX]+4,0
166 MOV DWORD PTR [EBX]+8, LINEAR_PROTO_LO
167 MOV DWORD PTR [EBX]+12, LINEAR_PROTO_HI
168 MOV TEMP_GDT_scratch.table_linear,EBX
169 MOV TEMP_GDT_scratch.table_lim,15
170
171 DB 66H ; execute a 32 bit LGDT
172 LGDT TEMP_GDT_scratch
173
174 ; enter protected mode
175 MOV EBX,CR0
176 OR EBX,PE_BIT
177 MOV CR0,EBX
178

179 ; clear prefetch queue

180 JMP CLEAR_LABEL
181 CLEAR_LABEL:
182
183 ; make DS and ES address 4G of linear memory
184 MOV CX,LINEAR_SEL
185 MOV DS,CX
186 MOV ES,CX
187
188 ; do board specific initialization
189 ;
190 ;
191 ; ......
192 ;
193
194
195 ; See Figure 8-5
196 ; copy EPROM GDT to ram at:
197 ; RAM_START + size (STARTUP_DATA)
198 MOV EAX,RAM_START

8-23
PROCESSOR MANAGEMENT AND INITIALIZATION

199 ADD EAX,OFFSET (end_data)

200 MOV EBX,RAM_START
201 MOV ECX, CS_BASE
202 ADD ECX, OFFSET (GDT_EPROM)
203 MOV ESI, [ECX].table_linear
204 MOV EDI,EAX
205 MOVZX ECX, [ECX].table_lim
206 MOV APP_GDT_ram[EBX].table_lim,CX
207 INC ECX
208 MOV EDX,EAX
209 MOV APP_GDT_ram[EBX].table_linear,EAX
210 ADD EAX,ECX
211 REP MOVS BYTE PTR ES:[EDI],BYTE PTR DS:[ESI]
212
213 ; fixup GDT base in descriptor
214 MOV ECX,EDX
215 MOV [EDX].bas_0_15+GDT_DESC_OFF,CX
216 ROR ECX,16
217 MOV [EDX].bas_16_23+GDT_DESC_OFF,CL
218 MOV [EDX].bas_24_31+GDT_DESC_OFF,CH
219
220 ; copy EPROM IDT to ram at:
221 ; RAM_START+size(STARTUP_DATA)+SIZE (EPROM GDT)
222 MOV ECX, CS_BASE
223 ADD ECX, OFFSET (IDT_EPROM)
224 MOV ESI, [ECX].table_linear
225 MOV EDI,EAX
226 MOVZX ECX, [ECX].table_lim
227 MOV APP_IDT_ram[EBX].table_lim,CX
228 INC ECX
229 MOV APP_IDT_ram[EBX].table_linear,EAX
230 MOV EBX,EAX
231 ADD EAX,ECX
232 REP MOVS BYTE PTR ES:[EDI],BYTE PTR DS:[ESI]
233
234 ; fixup IDT pointer in GDT
235 MOV [EDX].bas_0_15+IDT_DESC_OFF,BX
236 ROR EBX,16
237 MOV [EDX].bas_16_23+IDT_DESC_OFF,BL
238 MOV [EDX].bas_24_31+IDT_DESC_OFF,BH
239
240 ; load GDTR and IDTR
241 MOV EBX,RAM_START
242 DB 66H ; execute a 32 bit LGDT
243 LGDT APP_GDT_ram[EBX]
244 DB 66H ; execute a 32 bit LIDT
245 LIDT APP_IDT_ram[EBX]
246
247 ; move the TSS

8-24
 PROCESSOR MANAGEMENT AND INITIALIZATION

248 MOV EDI,EAX

249 MOV EBX,TSS_INDEX*SIZE(DESC)
250 MOV ECX,GDT_DESC_OFF ;build linear address for TSS
251 MOV GS,CX
252 MOV DH,GS:[EBX].bas_24_31
253 MOV DL,GS:[EBX].bas_16_23
254 ROL EDX,16
255 MOV DX,GS:[EBX].bas_0_15
256 MOV ESI,EDX
257 LSL ECX,EBX
258 INC ECX
259 MOV EDX,EAX
260 ADD EAX,ECX
261 REP MOVS BYTE PTR ES:[EDI],BYTE PTR DS:[ESI]
262
263 ; fixup TSS pointer
264 MOV GS:[EBX].bas_0_15,DX
265 ROL EDX,16
266 MOV GS:[EBX].bas_24_31,DH
267 MOV GS:[EBX].bas_16_23,DL
268 ROL EDX,16
269 ;save start of free ram at linear location RAMSTART
270 MOV free_mem_linear_base+RAM_START,EAX
271
272 ;assume no LDT used in the initial task - if necessary,
273 ;code to move the LDT could be added, and should resemble
274 ;that used to move the TSS
275
276 ; load task register
277 LTR BX ; No task switch, only descriptor loading
278 ; See Figure 8-6
279 ; load minimal set of registers necessary to simulate task
280 ; switch
281
282
283 MOV AX,[EDX].SS_reg ; start loading registers
284 MOV EDI,[EDX].ESP_reg
285 MOV SS,AX
286 MOV ESP,EDI ; stack now valid
287 PUSH DWORD PTR [EDX].EFLAGS_reg
288 PUSH DWORD PTR [EDX].CS_reg
289 PUSH DWORD PTR [EDX].EIP_reg
290 MOV AX,[EDX].DS_reg
291 MOV BX,[EDX].ES_reg
292 MOV DS,AX ; DS and ES no longer linear memory
293 MOV ES,BX
294

8-25
PROCESSOR MANAGEMENT AND INITIALIZATION

295 ; simulate far jump to initial task

296 IRETD
297
298 STARTUP_CODE ENDS
*** WARNING #377 IN 298, (PASS 2) SEGMENT CONTAINS PRIVILEGED INSTRUCTION(S)
299
300 END STARTUP, DS:STARTUP_DATA, SS:STARTUP_DATA
301
302

ASSEMBLY COMPLETE, 1 WARNING, NO ERRORS.

FFFF FFFFH

START: [CS.BASE+EIP] FFFF 0000H

• Jump near start

• Construct TEMP_GDT
• LGDT
• Move to protected mode

DS, ES = GDT[1] 4GB

Base
Limit GDT_SCRATCH

GDT [1] Base=0, Limit=4G

GDT [0] 0 TEMP_GDT

Figure 8-4. Constructing Temporary GDT and Switching to Protected Mode (Lines
162-172 of List File)

8-26
 PROCESSOR MANAGEMENT AND INITIALIZATION

FFFF FFFFH

TSS
IDT
GDT

• Move the GDT, IDT, TSS

from ROM to RAM
• Fix Aliases
• LTR

TSS RAM
IDT RAM
GDT RAM
RAM_START

Figure 8-5. Moving the GDT, IDT and TSS from ROM to RAM (Lines 196-261 of List File)

8-27
PROCESSOR MANAGEMENT AND INITIALIZATION

•
•
EIP
SS = TSS.SS EFLAGS
ESP = TSS.ESP •
PUSH TSS.EFLAG •
•
PUSH TSS.CS
ESP
PUSH TSS.EIP
ES = TSS.ES •
DS = TSS.DS
IRET ES
CS
SS
DS

GDT

TSS RAM
IDT Alias IDT RAM
GDT Alias GDT RAM
RAM_START
0

Figure 8-6. Task Switching (Lines 282-296 of List File)

8-28
 PROCESSOR MANAGEMENT AND INITIALIZATION

8.9.3. MAIN.ASM Source Code

The file MAIN.ASM shown in Example 8-2 defines the data and stack segments for this appli-
cation and can be substituted with the main module task written in a high-level language that is
invoked by the IRET instruction executed by STARTUP.ASM.

Example 8-2. MAIN.ASM

NAME main_module
data SEGMENT RW
dw 1000 dup(?)
DATA ENDS
stack stackseg 800
CODE SEGMENT ER use32 PUBLIC
main_start:
nop
nop
nop
CODE ENDS
END main_start, ds:data, ss:stack

8.9.4. Supporting Files

The batch file shown in Example 8-3 can be used to assemble the source code files
STARTUP.ASM and MAIN.ASM and build the final application.

Example 8-3. Batch File to Assemble and Build the Application

ASM386 STARTUP.ASM
ASM386 MAIN.ASM
BLD386 STARTUP.OBJ, MAIN.OBJ buildfile(EPROM.BLD) bootstrap(STARTUP) Bootload

BLD386 performs several operations in this example:

• It allocates physical memory location to segments and tables.
• It generates tables using the build file and the input files.
• It links object files and resolves references.
• It generates a boot-loadable file to be programmed into the EPROM.
Example 8-4 shows the build file used as an input to BLD386 to perform the above functions.

8-29
PROCESSOR MANAGEMENT AND INITIALIZATION

Example 8-4. Build File

INIT_BLD_EXAMPLE;

SEGMENT
*SEGMENTS(DPL = 0)
, startup.startup_code(BASE = 0FFFF0000H)
;

TASK
BOOT_TASK(OBJECT = startup, INITIAL,DPL = 0,
NOT INTENABLED)
, PROTECTED_MODE_TASK(OBJECT = main_module,DPL = 0,
NOT INTENABLED)
;

TABLE
GDT (
LOCATION = GDT_EPROM
, ENTRY = (
10: PROTECTED_MODE_TASK
, startup.startup_code
, startup.startup_data
, main_module.data
, main_module.code
, main_module.stack
)
),

IDT (
LOCATION = IDT_EPROM
);

MEMORY
(
RESERVE = (0..3FFFH
-- Area for the GDT, IDT, TSS copied from ROM
, 60000H..0FFFEFFFFH)
, RANGE = (ROM_AREA = ROM (0FFFF0000H..0FFFFFFFFH))
-- Eprom size 64K
, RANGE = (RAM_AREA = RAM (4000H..05FFFFH))
);

END
Table 8-5 shows the relationship of each build item with an ASM source file.

8-30
 PROCESSOR MANAGEMENT AND INITIALIZATION

Table 8-5. Relationship Between BLD Item and ASM Source File
BLD386 Controls and
Item ASM386 and Startup.A58 BLD file Effect
Bootstrap public startup bootstrap Near jump at
startup: start(startup) 0FFFFFFF0H to start
GDT location public GDT_EPROM TABLE The location of the GDT
GDT_EPROM TABLE_REG GDT(location = will be programmed into
<> GDT_EPROM) the GDT_EPROM
location
IDT location public IDT_EPROM TABLE The location of the IDT
IDT_EPROM TABLE_REG IDT(location = will be programmed into
<> IDT_EPROM the IDT_EPROM
location
RAM start RAM_START equ 400H memory (reserve = RAM_START is used as
(0..3FFFH)) the ram destination for
moving the tables. It
must be excluded from
the application’s
segment area.
Location of the TSS_INDEX EQU 10 TABLE GDT( Put the descriptor of the
application TSS ENTRY=( 10: application TSS in GDT
in the GDT PROTECTED_MODE_TA entry 10
SK))
EPROM size size and location of the SEGMENT startup.code Initialization code size
and location initialization code (base= 0FFFF0000H) must be less than 64K
...memory (RANGE( and resides at upper
ROM_AREA = most 64K of the 4GB
ROM(x..y)) memory space.

8.10. P6 FAMILY MICROCODE UPDATE FEATURE

P6 family processors have the capability to correct specific errata through the loading of an
Intel-supplied data block. This data block is referred to as a microcode update. This chapter
describes the underlying mechanisms the BIOS needs to provide in order to utilize this feature
during system initialization. It also describes a specification that provides for incorporating
future releases of the microcode update into a system BIOS.
Intel considers the combination of a particular silicon revision and the microcode update as the
equivalent stepping of the processor. Intel does not validate processors without the microcode
update loaded. Intel completes a full-stepping level validation and testing for new releases of
microcode updates.
A microcode update is used to correct specific errata in the processor. The BIOS, which incor-
porates an update loader, is responsible for loading the appropriate update on all processors
during system initialization (refer to Figure 8-7). There are effectively two steps to this process.
The first is to incorporate the necessary microcode updates into the BIOS, the second is to actu-
ally load the appropriate microcode update into the processor.

8-31
PROCESSOR MANAGEMENT AND INITIALIZATION

UPDATE
LOADER

Update
Blocks

New P6 Family CPU

Update
BIOS

Figure 8-7. Integrating Processor Specific Updates

8.10.1. Microcode Update

A microcode update consists of an Intel-supplied binary that contains a descriptive header and
data. No executable code resides within the update. This section describes the update and the
structure of its data format.
Each microcode update is tailored for a particular stepping of a P6 family processor. It is
designed such that a mismatch between a stepping of the processor and the update will result in
a failure to load. Thus, a given microcode update is associated with a particular type, family,
model, and stepping of the processor as returned by the CPUID instruction. In addition, the
intended processor platform type must be determined to properly target the microcode update.
The intended processor platform type is determined by reading a model-specific register MSR
(17h) (refer to Table 8-6) within the P6 family processor. This is a 64-bit register that may be
read using the RDMSR instruction (refer to Section 3.2., “Instruction Reference” Chapter 3,
Instruction Set Reference, Volume 1 of the Programmer’s Reference Manual). The three plat-
form ID bits, when read as a binary coded decimal (BCD) number indicate the bit position in the
microcode update header’s, Processor Flags field, that is associated with the installed processor.

8-32
 PROCESSOR MANAGEMENT AND INITIALIZATION

Register Name:BBL_CR_OVRD
MSR Address:017h
Access:Read Only
BBL_CR_OVRD is a 64-bit register accessed only when referenced as a Qword through a
RDMSR instruction.

Table 8-6. P6 Family Processor MSR Register Components

Bit Descriptions
63:53 Reserved
52:50 Platform ID bits (RO). The field gives information concerning the intended platform for the
processor.
52 51 50
0 0 0 Processor Flag 0 (See Processor Flags in Microcode Update Header)
0 0 1 Processor Flag 1
0 1 0 Processor Flag 2
0 1 1 Processor Flag 3
1 0 0 Processor Flag 4
1 0 1 Processor Flag 5
1 1 0 Processor Flag 6
1 1 1 Processor Flag 7
49:0 Reserved

The microcode update is a data block that is exactly 2048 bytes in length. The initial 48 bytes
of the update contain a header with information used to identify the update. The update header
and its reserved fields are interpreted by software based upon the header version. The initial
version of the header is 00000001h. An encoding scheme also guards against tampering of the
update data and provides a means for determining the authenticity of any given update. Table
8-7 defines each of the fields and Figure 8-8 shows the format of the microcode update data
block.

8-33
PROCESSOR MANAGEMENT AND INITIALIZATION

Table 8-7. Microcode Update Encoding Format

Offset Length
Field Name (in bytes) (in bytes) Description
Header Version 0 4 Version number of the update header.
Update Revision 4 4 Unique version number for the update, the basis for the
update signature provided by the processor to indicate
the current update functioning within the processor.
Used by the BIOS to authenticate the update and verify
that it is loaded successfully by the processor. The value
in this field cannot be used for processor stepping
identification alone.
Date 8 4 Date of the update creation in binary format: mmddyyyy
(e.g. 07/18/98 is 07181998h).
Processor 12 4 Processor type, family, model, and stepping of processor
that requires this particular update revision (e.g.,
00000650h). Each microcode update is designed
specifically for a given processor type, family, model, and
stepping of processor. The BIOS uses the Processor
field in conjunction with the CPUID instruction to
determine whether or not an update is appropriate to load
on a processor. The information encoded within this field
exactly corresponds to the bit representations returned
by the CPUID instruction.
Checksum 16 4 Checksum of update data and header. Used to verify the
integrity of the update header and data. Checksum is
correct when the summation of the 512 double words of
the update result in the value zero.
Loader Revision 20 4 Version number of the loader program needed to
correctly load this update. The initial version is
00000001h.
Processor Flags 24 4 Platform type information is encoded in the lower 8 bits of
this 4-byte field. Each bit represents a particular platform
type for a given CPUID. The BIOS uses the Processor
Flags field in conjunction with the platform ID bits in MSR
(17h) to determine whether or not an update is
appropriate to load on a processor.
Reserved 28 20 Reserved Fields for future expansion.
Update Data 48 2000 Update data.

8-34
 PROCESSOR MANAGEMENT AND INITIALIZATION

32 24 16 8 0

Update Data (2000 Bytes)

Reserved (20 Bytes)

Processor Flags
Reserved: 24 P7: I P6: I P5: I P4: I P3: I P2: I P1: I

Loader Revision
Checksum
Processor

Reserved: 18 ProcType: 2 Family: 4 Model: 4 Stepping: 4

Date
Month: 8 Day: 8 Year: 16

Update Revision
Header Revision
32 24 16 8 0

Figure 8-8. Format of the Microcode Update Data Block

8.10.2. Microcode Update Loader

This section describes the update loader used to load a microcode update into a P6 family
processor. It also discusses the requirements placed upon the BIOS to ensure proper loading of
an update.
The update loader contains the minimal instructions needed to load an update. The specific
instruction sequence that is required to load an update is dependent upon the loader revision field
contained within the update header. The revision of the update loader is expected to change very
infrequently, potentially only when new processor models are introduced.

8-35
PROCESSOR MANAGEMENT AND INITIALIZATION

The code below represents the update loader with a loader revision of 00000001h:
mov ecx,79h ; MSR to read in ECX
xor eax,eax ; clear EAX
xor ebx,ebx ; clear EBX
movax,cs ; Segment of microcode update
shl eax,4
movbx,offset Update ; Offset of microcode update
addeax,ebx ; Linear Address of Update in EAX
addeax,48d ; Offset of the Update Data within the Update
xor edx,edx ; Zero in EDX
WRMSR ; microcode update trigger

8.10.2.1. UPDATE LOADING PROCEDURE

The simple loader previously described assumes that Update is the address of a microcode
update (header and data) embedded within the code segment of the BIOS. It also assumes that
the processor is operating in real mode. The data may reside anywhere in memory that is acces-
sible by the processor within its current operating mode (real, protected).
Before the BIOS executes the microcode update trigger (WRMSR) instruction the following
must be true:
• EAX contains the linear address of the start of the update data
• EDX contains zero
• ECX contains 79h
Other requirements to keep in mind are:
• The microcode update must be loaded to the processor early on in the POST, and always
prior to the initialization of the P6 family processors L2 cache controller.
• If the update is loaded while the processor is in real mode, then the update data may not
cross a segment boundary.
• If the update is loaded while the processor is in real mode, then the update data may not
exceed a segment limit.
• If paging is enabled, pages that are currently present must map the update data.
• The microcode update data does not require any particular byte or word boundary
alignment.

8.10.2.2. HARD RESETS IN UPDATE LOADING

The effects of a loaded update are cleared from the processor upon a hard reset. Therefore, each
time a hard reset is asserted during the BIOS POST, the update must be reloaded on all proces-
sors that observed the reset. The effects of a loaded update are, however, maintained across a
processor INIT. There are no side effects caused by loading an update into a processor multiple
times.

8-36
 PROCESSOR MANAGEMENT AND INITIALIZATION

8.10.2.3. UPDATE IN A MULTIPROCESSOR SYSTEM

A multiprocessor (MP) system requires loading each processor with update data appropriate for
its CPUID and platform ID bits. The BIOS is responsible for ensuring that this requirement is
met, and that the loader is located in a module that is executed by all processors in the system.
If a system design permits multiple steppings of P6 family processors to exist concurrently, then
the BIOS must verify each individual processor against the update header information to ensure
appropriate loading. Given these considerations, it is most practical to load the update during
MP initialization.

8.10.2.4. UPDATE LOADER ENHANCEMENTS

The update loader presented in Section 8.10.2.1., “Update Loading Procedure” is a minimal
implementation that can be enhanced to provide additional functionality and features. Some
potential enhancements are described below:
• The BIOS can incorporate multiple updates to support multiple steppings of the P6 family
processor. This feature provides for operating in a mixed stepping environment on an MP
system and enables a user to upgrade to a later version of the processor. In this case,
modify the loader to check the CPUID and platform ID bits of the processor that it is
running on against the available headers before loading a particular update. The number of
updates is only limited by the available space in the BIOS.
• A loader can load the update and test the processor to determine if the update was loaded
correctly. This can be done as described in the Section 8.10.3., “Update Signature and
Verification”.
• A loader can verify the integrity of the update data by performing a checksum on the
double words of the update summing to zero, and can reject the update.
• A loader can provide power-on messages indicating successful loading of an update.

8.10.3. Update Signature and Verification

The P6 family processor provides capabilities to verify the authenticity of a particular update
and to identify the current update revision. This section describes the model-specific extensions
of the processor that support this feature. The update verification method below assumes that
the BIOS will only verify an update that is more recent than the revision currently loaded into
the processor.
The CPUID instruction returns a value in a model specific register in addition to its usual
register return values. The semantics of the CPUID instruction cause it to deposit an update ID
value in the 64-bit model-specific register (MSR) at address 08Bh. If no update is present in the
processor, the value in the MSR remains unmodified. Normally a zero value is preloaded into
the MSR by software before executing the CPUID instruction. If the MSR still contains zero
after executing CPUID, this indicates that no update is present.
The update ID value returned in the EDX register after a RDMSR instruction indicates the revi-
sion of the update loaded in the processor. This value, in combination with the normal CPUID

8-37
PROCESSOR MANAGEMENT AND INITIALIZATION

value returned in the EAX register, uniquely identifies a particular update. The signature ID can
be directly compared with the update revision field in the microcode update header for verifica-
tion of a correct update load. No consecutive updates released for a given stepping of the P6
family processor may share the same signature. Updates for different steppings are differenti-
ated by the CPUID value.

8.10.3.1. DETERMINING THE SIGNATURE

An update that is successfully loaded into the processor provides a signature that matches the
update revision of the currently functioning revision. This signature is available any time after
the actual update has been loaded, and requesting this signature does not have any negative
impact upon any currently loaded update. The procedure for determining this signature is:
mov ecx, 08Bh;Model Specific Register to Read in ECX
xor eax,eax ;clear EAX
xor edx,edx ;clear EDX
WRMSR ;Load 0 to MSR at 8Bh
mov eax,1
CPUID
mov ecx, 08BH;Model Specific Register to Read
RDMSR ;Read Model Specific Register
If there is an update currently active in the processor, its update revision is returned in the EDX
register after the RDMSR instruction has completed.

8.10.3.2. AUTHENTICATING THE UPDATE

An update may be authenticated by the BIOS using the signature primitive, described above,
with the following algorithm:
Z = Update revision from the update header to be authenticated;
X = Current Update Signature from MSR 8Bh;
If (Z > X) Then
Load Update that is to be authenticated;
Y = New Signature from MSR 8Bh;
If (Z == Y) then Success
Else Fail
Else Fail
The algorithm requires that the BIOS only authenticate updates that contain a numerically larger
revision than the currently loaded revision, where Current Signature (X) < New Update Revi-
sion (Z). A processor with no update loaded should be considered to have a revision equal to
zero. This authentication procedure relies upon the decoding provided by the processor to verify
an update from a potentially hostile source. As an example, this mechanism in conjunction with
other safeguards provides security for dynamically incorporating field updates into the BIOS.

8-38
 PROCESSOR MANAGEMENT AND INITIALIZATION

8.10.4. P6 Family Processor Microcode Update Specifications

This section describes the interface that an application can use to dynamically integrate
processor-specific updates into the system BIOS. In this discussion, the application is referred
to as the calling program or caller.
The real mode INT15 call specification described here is an Intel extension to an OEM BIOS.
This extension allows an application to read and modify the contents of the microcode update
data in NVRAM. The update loader, which is part of the system BIOS, cannot be updated by
the interface. All of the functions defined in the specification must be implemented for a system
to be considered compliant with the specification. The INT15 functions are accessible only
from real mode.

8.10.4.1. RESPONSIBILITIES OF THE BIOS

If a BIOS passes the presence test (INT 15h, AX=0D042h, BL=0h) it must implement all of the
sub-functions defined in the INT 15h, AX= 0D042h specification. There are no optional func-
tions. The BIOS must load the appropriate update for each processor during system initializa-
tion.
A header version of an update block containing the value 0FFFFFFFFh indicates that the update
block is unused and available for storing a new update.
The BIOS is responsible for providing a 2048 byte region of non-volatile storage (NVRAM) for
each potential processor stepping within a system. This storage unit is referred to as an update
block. The BIOS for a single processor system need only provide one update block to store the
microcode update data. The BIOS for a multiple processor capable system needs to provide one
update block for each unique processor stepping supported by the OEM’s system. The BIOS is
responsible for managing the NVRAM update blocks. This includes garbage collection, such
as removing update blocks that exist in NVRAM for which a corresponding processor does not
exist in the system. This specification only provides the mechanism for ensuring security, the
uniqueness of an entry, and that stale entries are not loaded. The actual update block manage-
ment is implementation specific on a per-BIOS basis. As an example, the BIOS may use update
blocks sequentially in ascending order with CPU signatures sorted versus the first available
block. In addition, garbage collection may be implemented as a setup option to clear all
NVRAM slots or as BIOS code that searches and eliminates unused entries during boot.
The following algorithm describes the steps performed during BIOS initialization used to load
the updates into the processor(s). It assumes that the BIOS ensures that no update contained
within NVRAM has a header version or loader version that does not match one currently
supported by the BIOS and that the update block contains a correct checksum. It also assumes
that the BIOS ensures that at most one update exists for each processor stepping and that older
update revisions are not allowed to overwrite more recent ones. These requirements are checked
by the BIOS during the execution of the write update function of this interface. The BIOS
sequentially scans through all of the update blocks in NVRAM starting with index 0. The BIOS
scans until it finds an update where the processor fields in the header match the family, model,
and stepping as well as the platform ID bits of the current processor.

8-39
PROCESSOR MANAGEMENT AND INITIALIZATION

For each processor in the system {

Determine the ProcType, Family, Model and Stepping via CPUID;
Determine the Platform ID Bits by reading the BBL_CR_OVRD[52:50] MSR;
for (I = UpdateBlock 0, I < NumOfUpdates; I++) {
If ((UpdateHeader.Processor ==
ProcType, Family, Model and Stepping) &&
(UpdateHeader.ProcessorFlags == Platform ID Bits)) {
Load UpdateHeader.UpdateData into the Processor;
Verify that update was correctly loaded into the processor
Go on to next processor
Break;
}
}
Programmer’s Note: The platform ID bits in the BBL_CR_OVRD MSR are
encoded as a three-bit binary coded decimal field. The platform ID bits in the
microcode update header are individually bit encoded. The algorithm must
do a translation from one format to the other prior to doing the comparison.
When performing the INT 15h, 0D042h functions, the BIOS must assume that the caller has no
knowledge about platform specific requirements. It is the responsibility of the BIOS calls to
manage all chipset and platform specific prerequisites for managing the NVRAM device. When
writing the update data via the write update sub-function, the BIOS must maintain implementa-
tion specific data requirements, such as the update of NVRAM checksum. The BIOS should
also attempt to verify the success of write operations on the storage device used to record the
update.

8.10.4.2. RESPONSIBILITIES OF THE CALLING PROGRAM

This section of the document lists the responsibilities of the calling program using the interface
specifications to load microcode update(s) into BIOS NVRAM.
The calling program should call the INT 15h, 0D042h functions from a pure real mode program
and should be executing on a system that is running in pure real mode. The caller should issue
the presence test function (sub function 0) and verify the signature and return codes of that func-
tion. It is important that the calling program provides the required scratch RAM buffers for the
BIOS and the proper stack size as specified in the interface definition.
The calling program should read any update data that already exists in the BIOS in order to make
decisions about the appropriateness of loading the update. The BIOS refuses to overwrite a
newer update with an older version. The update header contains information about version and
processor specifics for the calling program to make an intelligent decision about loading.
There can be no ambiguous updates. The BIOS refuses to allow multiple updates for the same
CPUID to exist at the same time. The BIOS also refuses to load an update for a processor that
does not exist in the system.
The calling application should implement a verify function that is run after the update write
function successfully completes. This function reads back the update and verifies that the BIOS

8-40
 PROCESSOR MANAGEMENT AND INITIALIZATION

returned an image identical to the one that was written. The following pseudo-code represents
a calling program.
INT 15 D042 Calling Program Pseudo-code
//
// We must be in real mode
//
If the system is not in Real mode
then Exit
//
// Detect the presence of Genuine Intel processor(s) that can be updated (CPUID)
//
If no Intel processors exist that can be updated
then Exit
//
// Detect the presence of the Intel microcode update extensions
//
If the BIOS fails the PresenceTest
then Exit
//
// If the APIC is enabled, see if any other processors are out there
//
Read APICBaseMSR
If APIC enabled {
Send Broadcast Message to all processors except self via APIC;
Have all processors execute CPUID and record Type, Family, Model, Stepping
Have all processors read BBL_CR_OVRD[52:50] and record platform ID bits
If current processor is not updatable
then Exit
}
//
// Determine the number of unique update slots needed for this system
//
NumSlots = 0;
For each processor {
If ((this is a unique processor stepping) and
(we have an update in the database for this processor)) {
Checksum the update from the database;
If Checksum fails
then Exit;
Increment NumSlots;
}
}
//
// Do we have enough update slots for all CPUs?
//
If there are more unique processor steppings than update slots provided by the BIOS
then Exit

8-41
PROCESSOR MANAGEMENT AND INITIALIZATION

//
// Do we need any update slots at all? If not, then we’re all done
//
If (NumSlots == 0)
then Exit

//
// Record updates for processors in NVRAM.
//
For (I=0; I<NumSlots; I++) {
//
// Load each Update
//
Issue the WriteUpdate function

If (STORAGE_FULL) returned {
Display Error -- BIOS is not managing NVRAM appropriately
exit
}
If (INVALID_REVISION) returned {
Display Message: More recent update already loaded in NVRAM for this stepping
continue;
}

If any other error returned {

Display Diagnostic
exit
}
//
// Verify the update was loaded correctly
//
Issue the ReadUpdate function

If an error occurred {
Display Diagnostic
exit
}
//
// Compare the Update read to that written
//
if (Update read != Update written) {
Display Diagnostic
exit
}
}
//

8-42
 PROCESSOR MANAGEMENT AND INITIALIZATION

// Enable Update Loading, and inform user

//
Issue the ControlUpdate function with Task=Enable.

8.10.4.3. MICROCODE UPDATE FUNCTIONS

Table 8-8 defines the current P6 family Processor microcode update functions.

Table 8-8. Microcode Update Functions

Microcode Update Function
Function Number Description Required/Optional
Presence test 00h Returns information about the supported Required
functions.
Write update data 01h Writes one of the update data areas (slots). Required
Update control 02h Globally controls the loading of updates. Required
Read update data 03h Reads one of the update data areas (slots). Required

8.10.4.4. INT 15H-BASED INTERFACE

Intel recommends that a BIOS interface be provided that allows additional microcode updates
to be added to the system flash. The INT15 interface is an Intel-defined method for doing this.
The program that calls this interface is responsible for providing three 64-kilobyte RAM areas
for BIOS use during calls to the read and write functions. These RAM scratch pads can be used
by the BIOS for any purpose, but only for the duration of the function call. The calling routine
places real mode segments pointing to the RAM blocks in the CX, DX and SI registers. Calls
to functions in this interface must be made with a minimum of 32 kilobytes of stack available to
the BIOS.
In general, each function returns with CF cleared and AH contains the returned status. The
general return codes and other constant definitions are listed in Section 8.10.4.5., “Return
Codes”.
The OEM Error (AL) is provided for the OEM to return additional error information specific to
the platform. If the BIOS provides no additional information about the error, the OEM Error
must be set to SUCCESS. The OEM Error field is undefined if AH contains either SUCCESS
(00) or NOT_IMPLEMENTED (86h). In all other cases it must be set with either SUCCESS or
a value meaningful to the OEM.
The following text details the functions provided by the INT15h-based interface.

8-43
PROCESSOR MANAGEMENT AND INITIALIZATION

Function 00h - Presence Test

This function verifies that the BIOS has implemented the required microcode update functions.
Table 8-3 lists the parameters and return codes for the function.

Table 8-9. Parameters for the Presence Test

Input:
AX Function Code 0D042h
BL Sub-function 00h - Presence Test
Output:
CF Carry Flag Carry Set - Failure - AH Contains Status.
Carry Clear - All return values are valid.
AH Return Code
AL OEM Error Additional OEM Information.
EBX Signature Part 1 ’INTE’ - Part one of the signature.
ECX Signature Part 2 ’LPEP’- Part two of the signature.
EDX Loader Version Version number of the microcode update loader.
SI Update Count Number of update blocks the system can record in NVRAM.
Return Codes: (See Table 8-8 for code definitions)
SUCCESS Function completed successfully.
NOT_IMPLEMENTED Function not implemented.

In order to assure that the BIOS function is present, the caller must verify the Carry Flag, the
Return Code, and the 64-bit signature. Each update block is exactly 2048 bytes in length. The
update count reflects the number of update blocks available for storage within non-volatile
RAM. The update count must return with a value greater than or equal to the number of unique
processor steppings currently installed within the system.
The loader version number refers to the revision of the update loader program that is included
in the system BIOS image.

8-44
 PROCESSOR MANAGEMENT AND INITIALIZATION

Function 01h - Write Microcode Update Data

This function integrates a new microcode update into the BIOS storage device. Table 8-4 lists
the parameters and return codes for the function.

Table 8-10. Parameters for the Write Update Data Function

Input:
AX Function Code 0D042h
BL Sub-function 01h - Write Update
ED:DI Update Address Real Mode pointer to the Intel Update structure. This buffer is
2048 bytes in length
CX Scratch Pad1 Real Mode Segment address of 64 kilobytes of RAM Block.
DX Scratch Pad2 Real Mode Segment address of 64 kilobytes of RAM Block.
SI Scratch Pad3 Real Mode Segment address of 64 kilobytes of RAM Block.
SS:SP Stack pointer 32 kilobytes of Stack Minimum.
Output:
CF Carry Flag Carry Set - Failure - AH Contains Status.
Carry Clear - All return values are valid.
AH Return Code Status of the Call
AL OEM Error Additional OEM Information.
Return Codes: (See Table 8-8 for code definitions)
SUCCESS Function completed successfully.
WRITE_FAILURE A failure because of the inability to write the storage device.
ERASE_FAILURE A failure because of the inability to erase the storage device.
READ_FAILURE A failure because of the inability to read the storage device.
STORAGE_FULL The BIOS non-volatile storage area is unable to accommodate
the update because all available update blocks are filled with
updates that are needed for processors in the system.
CPU_NOT_PRESENT The processor stepping does not currently exist in the system.
INVALID_HEADER The update header contains a header or loader version that is
not recognized by the BIOS.
INVALID_HEADER_CS The update does not checksum correctly.
SECURITY_FAILURE The processor rejected the update.
INVALID_REVISION The same or more recent revision of the update exists in the
storage device.

The BIOS is responsible for selecting an appropriate update block in the non-volatile storage for
storing the new update. This BIOS is also responsible for ensuring the integrity of the informa-
tion provided by the caller, including authenticating the proposed update before incorporating it
into storage.

8-45
PROCESSOR MANAGEMENT AND INITIALIZATION

Before writing the update block into NVRAM, the BIOS should ensure that the update structure
meets the following criteria in the following order:
1. The update header version should be equal to an update header version recognized by the
BIOS.
2. The update loader version in the update header should be equal to the update loader
version contained within the BIOS image.
3. The update block should checksum to zero. This checksum is computed as a 32-bit
summation of all 512 double words in the structure, including the header.
The BIOS selects an update block in non-volatile storage for storing the candidate update. The
BIOS can select any available update block as long as it guarantees that only a single update
exists for any given processor stepping in non-volatile storage. If the update block selected
already contains an update, the following additional criteria apply to overwrite it:
• The processor signature in the proposed update should be equal to the processor signature
in the header of the current update in NVRAM (CPUID + platform ID bits).
• The update revision in the proposed update should be greater than the update revision in
the header of the current update in NVRAM.
If no unused update blocks are available and the above criteria are not met, the BIOS can over-
write an update block for a processor stepping that is no longer present in the system. This can
be done by scanning the update blocks and comparing the processor steppings, identified in the
MP Specification table, to the processor steppings that currently exist in the system.
Finally, before storing the proposed update into NVRAM, the BIOS should verify the authen-
ticity of the update via the mechanism described in Section 8.10.2., “Microcode Update
Loader”. This includes loading the update into the current processor, executing the CPUID
instruction, reading MSR 08Bh, and comparing a calculated value with the update revision in
the proposed update header for equality.
When performing the write update function, the BIOS should record the entire update, including
the header and the update data. When writing an update, the original contents may be over-
written, assuming the above criteria have been met. It is the responsibility of the BIOS to ensure
that more recent updates are not overwritten through the use of this BIOS call, and that only a
single update exists within the NVRAM for any processor stepping.
Figure 8-9 shows the process the BIOS follows to choose an update block and ensure the integ-
rity of the data when it stores the new microcode update.

8-46
 PROCESSOR MANAGEMENT AND INITIALIZATION

Write Microcode Update

Does Update Match a Return

No
CPU in the System? CPU_NOT_PRESENT

Yes

Valid
Return
Update Header No
INVALID_HEADER
Version?

Yes

Does Loader
Return
Revision Match BIOS’s No
INVALID_HEADER
Loader?

Yes

Does Update Return

No
Checksum Correctly? INVALID_HEADER_CS

Yes

Update
Space Available
Matching CPU Already No Yes
in NVRAM?
In NVRAM?

No

Yes Return
STORAGE_FULL

Update
Return
Revision Newer Than No
INVALID_REVISION
NVRAM Update?

Yes

Update Pass Return

No
Authenticity Test? SECURITY_FAILURE

Yes

Update NMRAM Record

Return
SUCCESS

Figure 8-9. Write Operation Flow Chart

8-47
PROCESSOR MANAGEMENT AND INITIALIZATION

Function 02h - Microcode Update Control

This function enables loading of binary updates into the processor. Table 8-5 lists the parameters
and return codes for the function.

Table 8-11. Parameters for the Control Update Sub-function

Input:
AX Function Code 0D042h
BL Sub-function 02h - Control Update
BH Task See Description.
CX Scratch Pad1 Real Mode Segment of 64 kilobytes of RAM Block.
DX Scratch Pad2 Real Mode Segment of 64 kilobytes of RAM Block.
SI Scratch Pad3 Real Mode Segment of 64 kilobytes of RAM Block.
SS:SP Stack pointer 32 kilobytes of Stack Minimum.
Output:
CF Carry Flag Carry Set - Failure - AH contains Status.
Carry Clear - All return values are valid.
AH Return Code Status of the Call.
AL OEM Error Additional OEM Information.
BL Update Status Either Enable or Disable indicator.
Return Codes: (See Table 8-8 for code definitions)
SUCCESS Function completed successfully.
READ_FAILURE A failure because of the inability to read the storage device.

This control is provided on a global basis for all updates and processors. The caller can deter-
mine the current status of update loading (enabled or disabled) without changing the state. The
function does not allow the caller to disable loading of binary updates, as this poses a security
risk.
The caller specifies the requested operation by placing one of the values from Table 8-6 in the
BH register. After successfully completing this function the BL register contains either the
enable or the disable designator. Note that if the function fails, the update status return value is
undefined.

Table 8-12. Mnemonic Values

Mnemonic Value Meaning
Enable 1 Enable the Update loading at initialization time
Query 2 Determine the current state of the update control without changing
its status.

The READ_FAILURE error code returned by this function has meaning only if the control func-
tion is implemented in the BIOS NVRAM. The state of this feature (enabled/disabled) can also
be implemented using CMOS RAM bits where READ failure errors cannot occur.

8-48
 PROCESSOR MANAGEMENT AND INITIALIZATION

Function 03h - Read Microcode Update Data

This function reads a currently installed microcode update from the BIOS storage into a caller-
provided RAM buffer. Section 8-13, “Parameters for the Read Microcode Update Data Func-
tion” lists the parameters and return codes for the function.
Table 8-13. Parameters for the Read Microcode Update Data Function
Input:
AX Function Code 0D042h
BL Sub-function 03h - Read Update
ES:DI Buffer Address Real Mode pointer to the Intel Update structure that will be
written with the binary data.
ECX Scratch Pad1 Real Mode Segment address of 64 kilobytes of RAM Block
(lower 16 bits).
ECX Scratch Pad2 Real Mode Segment address of 64 kilobytes of RAM Block
(upper 16 bits).
DX Scratch Pad3 Real Mode Segment address of 64 kilobytes of RAM Block.
SS:SP Stack pointer 32 kilobytes of Stack Minimum.
SI Update Number The index number of the update block to be read. This value is
zero based and must be less than the update count returned
from the presence test function.
Output:
CF Carry Flag Carry Set - Failure - AH contains Status.
Carry Clear - All
return values
are valid.
AH Return Code Status of the Call.
AL OEM Error Additional OEM Information.
Return Codes: (See Table 8-8 for code definitions)
SUCCESS Function completed successfully.
READ_FAILURE A failure because of the inability to read the storage device.
UPDATE_NUM_INVALID Update number exceeds the maximum number of update
blocks implemented by the BIOS.

The read function enables the caller to read any update data that already exists in a BIOS and
make decisions about the addition of new updates. As a result of a successful call, the BIOS
copies exactly 2048 bytes into the location pointed to by ES:DI, with the contents of the update
block represented by update number.
An update block is considered unused and available for storing a new update if its header version
contains the value 0FFFFFFFFh after return from this function call. The actual implementation
of NVRAM storage management is not specified here and is BIOS dependent. As an example,
the actual data value used to represent an empty block by the BIOS may be zero, rather than

8-49
PROCESSOR MANAGEMENT AND INITIALIZATION

0FFFFFFFFh. The BIOS is responsible for translating this information into the header provided
by this function.

8.10.4.5. RETURN CODES

After the call has been made, the return codes listed in Table 8-8 are available in the AH register.

Table 8-14. Return Code Definitions

Return Code Value Description
SUCCESS 00h Function completed successfully
NOT_IMPLEMENTED 86h Function not implemented
ERASE_FAILURE 90h A failure because of the inability to erase the storage
device
WRITE_FAILURE 91h A failure because of the inability to write the storage device
READ_FAILURE 92h A failure because of the inability to read the storage device
STORAGE_FULL 93h The BIOS non-volatile storage area is unable to
accommodate the update because all available update
blocks are filled with updates that are needed for
processors in the system
CPU_NOT_PRESENT 94h The processor stepping does not currently exist in the
system
INVALID_HEADER 95h The update header contains a header or loader version
that is not recognized by the BIOS
INVALID_HEADER_CS 96h The update does not checksum correctly
SECURITY_FAILURE 97h The update was rejected by the processor
INVALID_REVISION 98h The same or more recent revision of the update exists in
the storage device
UPDATE_NUM_INVALID 99h The update number exceeds the maximum number of
update blocks implemented by the BIOS

8-50
 9
Memory Cache
Control
 MEMORY CACHE CONTROL

CHAPTER 9
MEMORY CACHE CONTROL

This chapter describes the Intel Architecture’s memory cache and cache control mechanisms,
the TLBs, and the write buffer. It also describes the memory type range registers (MTRRs)
found in the P6 family processors and how they are used to control caching of physical memory
locations.

9.1. INTERNAL CACHES, TLBS, AND BUFFERS

The Intel Architecture supports caches, translation look aside buffers (TLBs), and write buffers
for temporary on-chip (and external) storage of instructions and data (see Figure 9-1). Table 9-1
shows the characteristics of these caches and buffers for the P6 family, Pentium®, and Intel486™
processors. The sizes and characteristics of these units are machine specific and may
change in future versions of the processor. The CPUID instruction returns the sizes and char-
acteristics of the caches and buffers for the processor on which the instruction is executed. For
more information, see “CPUID—CPU Identification” in Chapter 3 of the Intel Architecture Soft-
ware Developer’s Manual, Volume 2.

9-1
MEMORY CACHE CONTROL

Physical
Memory

L2 Cache2,3 Data Cache

System Bus Unit (L11)
(External)
Cache Bus
Inst. TLBs

Bus Interface Unit

Data TLBs

Instruction Fetch Unit Instruction Cache (L11) Write Buffer

1
For the Intel486™ processor, the L1 Cache is a unified
instruction and data cache.
2
For the Pentium® and Intel486™ processors, the L2 Cache
is external to the processor package and there is
no cache bus (that is, the L2 cache interfaces with
the system bus).
3
For the Pentium® Pro, Pentium® II and Pentium® III processors,
the L2 Cache is internal to the processor package and there is
a separate cache bus.

Figure 9-1. Intel Architecture Caches

The Intel Architecture defines two separate caches: the level 1 (L1) cache and the level 2 (L2)
cache (see Figure 9-1). The L1 cache is closely coupled to the instruction fetch unit and execu-
tion units of the processor. For the Pentium® and P6 family processors, the L1 cache is divided
into two sections: one dedicated to caching instructions and one to caching data. For the
Intel486™ processor, the L1 cache is a unified instruction and data cache.

9-2
 MEMORY CACHE CONTROL

Table 9-1. Characteristics of the Caches, TLBs, and Write Buffer in

Intel Architecture Processors
Cache or Buffer Characteristics
®
L1 Instruction - P6 family and Pentium processors: 8 or 16 KBytes, 4-way set associative,
Cache1 32-byte cache line size; 2-way set associative for earlier Pentium® processors.
- Intel486™ processor: 8 or 16 KBytes, 4-way set associative, 16-byte cache line
size, instruction and data cache combined.
L1 Data Cache1 - P6 family processors: 16 KBytes, 4-way set associative, 32-byte cache line size;
8 KBytes, 2-way set associative for earlier P6 family processors.
- Pentium® processors: 16 KBytes, 4-way set associative, 32-byte cache line size;
8 KBytes, 2-way set associative for earlier Pentium® processors.
- Intel486™ processor: (see L1 instruction cache).
L2 Unified Cache2,3 - P6 family processors: 128 KBytes, 256 KBytes, 512 KBytes, 1 MByte, or 2
MByte, 4-way set associative, 32-byte cache line size.
- Pentium® processor: System specific, typically 256 or 512 KBytes, 4-way set
associative, 32-byte cache line size.
- Intel486™ processor: System specific.
Instruction TLB (4- - P6 family processors: 32 entries, 4-way set associative.
KByte Pages)1 - Pentium® processor: 32 entries, 4-way set associative; fully set
associative for Pentium® processors with MMX™ technology.
- Intel486™ processor: 32 entries, 4-way set associative, instruction
and data TLB combined.
Data TLB (4-KByte - Pentium® and P6 family processors: 64 entries, 4-way set associative; fully set
Pages)1 associative for Pentium® processors with MMX™ technology.
- Intel486™ processor: (see Instruction TLB).
Instruction TLB - P6 family processors: 2 entries, fully associative
(Large Pages) - Pentium® processor: Uses same TLB as used for 4-KByte pages.
- Intel486™ processor: None (large pages not supported).
Data TLB (Large - P6 family processors: 8 entries, 4-way set associative.
Pages) - Pentium® processor: 8 entries, 4-way set associative; uses same TLB as used
for 4-KByte pages in Pentium® processors with MMX™ technology.
- Intel486™ processor: None (large pages not supported).
Write Buffer - P6 family processors: 12 entries.
- Pentium® processor: 2 buffers, 1 entry each (Pentium® processors
with MMX™ technology have 4 buffers for 4 entries).
- Intel486™ processor: 4 entries.

NOTES:
1. In the Intel486™ processor, the L1 cache is a unified instruction and data cache, and the TLB is a unified
instruction and data TLB.
2. In the Intel486™ and Pentium® processors, the L2 cache is external to the processor package and
optional.
3. In the Pentium® Pro, Pentium® II, and Pentium® III processors, the L2 cache is internal to the processor
package.

9-3
MEMORY CACHE CONTROL

The L2 cache is a unified cache for storage of both instructions and data. It is closely coupled to
the L1 cache through the processor’s cache bus (for the P6 family processors) or the system bus
(for the Pentium® and Intel486™ processors).
The cache lines for the P6 family and Pentium® processors’ L1 and L2 caches are 32 bytes wide.
The processor always reads a cache line from system memory beginning on a 32-byte boundary.
(A 32-byte aligned cache line begins at an address with its 5 least-significant bits clear.) A cache
line can be filled from memory with a 4-transfer burst transaction. The caches do not support
partially-filled cache lines, so caching even a single doubleword requires caching an entire line.
(The cache line size for the Intel486™ processor is 16 bytes.)
The L1 and L2 caches are available in all execution modes. Using these caches greatly improves
the performance of the processor both in single- and multiple-processor systems. Caching can
also be used in system management mode (SMM); however, it must be handled carefully. For
more information, see Section 12.4.2., “SMRAM Caching”, in Chapter 12, System Management
Mode (SMM).
The TLBs store the most recently used page-directory and page-table entries. They speed up
memory accesses when paging is enabled by reducing the number of memory accesses that are
required to read the page tables stored in system memory. The TLBs are divided into four
groups: instruction TLBs for 4-KByte pages, data TLBs for 4-KByte pages; instruction TLBs
for large pages (2-MByte or 4-MByte pages), and data TLBs for large pages. (Only 4-KByte
pages are supported for Intel386™ and Intel486™ processors.) The TLBs are normally active
only in protected mode with paging enabled. When paging is disabled or the processor is in real-
address mode, the TLBs maintain their contents until explicitly or implicitly flushed. For more
information, see Section 9.10., “Invalidating the Translation Lookaside Buffers (TLBs)”.
The write buffer is associated with the processors instruction execution units. It allows writes to
system memory and/or the internal caches to be saved and in some cases combined to optimize
the processor’s bus accesses. The write buffer is always enabled in all execution modes.
The processor’s caches are for the most part transparent to software. When enabled, instructions
and data flow through these caches without the need for explicit software control. However,
knowledge of the behavior of these caches may be useful in optimizing software performance.
For example, knowledge of cache dimensions and replacement algorithms gives an indication
of how large of a data structure can be operated on at once without causing cache thrashing.
In multiprocessor systems, maintenance of cache consistency may, in rare circumstances,
require intervention by system software. For these rare cases, the processor provides privileged
cache control instructions for use in flushing caches.

9.2. CACHING TERMINOLOGY

The Intel Architecture (beginning with the Pentium® processor) uses the MESI (modified, exclu-
sive, shared, invalid) cache protocol to maintain consistency with internal caches and caches in
other processors. For more information, see Section 9.4., “Cache Control Protocol”. (The
Intel486™ processor uses an implementation defined caching protocol that operates in a similar
manner to the MESI protocol.)

9-4
 MEMORY CACHE CONTROL

When the processor recognizes that an operand being read from memory is cacheable, the
processor reads an entire cache line into the appropriate cache (L1, L2, or both). This operation
is called a cache line fill. If the memory location containing that operand is still cached the next
time the processor attempts to access the operand, the processor can read the operand from the
cache instead of going back to memory. This operation is called a cache hit.
When the processor attempts to write an operand to a cacheable area of memory, it first checks
if a cache line for that memory location exists in the cache. If a valid cache line does exist, the
processor (depending on the write policy currently in force) can write the operand into the cache
instead of writing it out to system memory. This operation is called a write hit. If a write misses
the cache (that is, a valid cache line is not present for the area of memory being written to), the
processor performs a cache line fill, write allocation. Then it writes the operand into the cache
line and (depending on the write policy currently in force) can also write it out to memory. If the
operand is to be written out to memory, it is written first into the write buffer, and then written
from the write buffer to memory when the system bus is available. (Note that for the Intel486™
and Pentium® processors, write misses do not result in a cache line fill; they always result in a
write to memory. For these processors, only read misses result in cache line fills.)
When operating in a multiple-processor system, Intel Architecture processors (beginning with
the Intel486™ processor) have the ability to snoop other processor’s accesses to system
memory and to their internal caches. They use this snooping ability to keep their internal caches
consistent both with system memory and with the caches in other processors on the bus. For
example, in the Pentium® and P6 family processors, if through snooping one processor detects
that another processor intends to write to a memory location that it currently has cached in
shared state, the snooping processor will invalidate its cache line forcing it to perform a cache
line fill the next time it accesses the same memory location.
Beginning with the P6 family processors, if a processor detects (through snooping) that another
processor is trying to access a memory location that it has modified in its cache, but has not yet
written back to system memory, the snooping processor will signal the other processor (by
means of the HITM# signal) that the cache line is held in modified state and will preform an
implicit write-back of the modified data. The implicit write-back is transferred directly to the
initial requesting processor and snooped by the memory controller to assure that system memory
has been updated. Here, the processor with the valid data may pass the data to the other proces-
sors without actually writing it to system memory; however, it is the responsibility of the
memory controller to snoop this operation and update memory.

9.3. METHODS OF CACHING AVAILABLE

The processor allows any area of system memory to be cached in the L1 and L2 caches. Within
individual pages or regions of system memory, it also allows the type of caching (also called
memory type) to be specified, using a variety of system flags and registers. For more informa-
tion, see Section 9.5., “Cache Control”. The caching methods currently defined for the Intel
Architecture are as follows. (Table 9-2 lists which types of caching are available on specific Intel
Architecture processors.)
• Uncacheable (UC)—System memory locations are not cached. All reads and writes appear
on the system bus and are executed in program order, without reordering. No speculative

9-5
MEMORY CACHE CONTROL

memory accesses, page-table walks, or prefetches of speculated branch targets are made.
This type of cache-control is useful for memory-mapped I/O devices. When used with
normal RAM, it greatly reduces processor performance.

Table 9-2. Methods of Caching Available in P6 Family, Pentium®,

and Intel486™ Processors
Caching Method P6 Family Pentium® Processor Intel486™ Processor
Processors
Uncacheable (UC) Yes Yes Yes
Write Combining (WC) Yes1 No No
2
Write Through (WT) Yes Yes Yes2
Write Back (WB) Yes Yes2 No
1
Write Protected (WP) Yes No No

NOTES:
1. Requires programming of MTRRs to implement.
2. Speculative reads not supported.

• Write Combining (WC)—System memory locations are not cached (as with uncacheable
memory) and coherency is not enforced by the processor’s bus coherency protocol.
Speculative reads are allowed. Writes may be delayed and combined in the write buffer to
reduce memory accesses. The writes may be delayed until the next occurrence of a buffer
or processor serialization event, e.g., CPUID execution, a read or write to uncached
memory, interrupt occurrence, LOCKed instruction execution, etc. if the WC buffer is
partially filled. This type of cache-control is appropriate for video frame buffers, where the
order of writes is unimportant as long as the writes update memory so they can be seen on
the graphics display. See Section 9.3.1., “Buffering of Write Combining Memory
Locations”, for more information about caching the WC memory type. The preferred
method is to use the new SFENCE (store fence) instruction introduced in the Pentium® III
processor. The SFENCE instruction ensures weakly ordered writes are written to memory
in order, i.e., it serializes only the store operations.
• Write-through (WT)—Writes and reads to and from system memory are cached. Reads
come from cache lines on cache hits; read misses cause cache fills. Speculative reads are
allowed. All writes are written to a cache line (when possible) and through to system
memory. When writing through to memory, invalid cache lines are never filled, and valid
cache lines are either filled or invalidated. Write combining is allowed. This type of cache-
control is appropriate for frame buffers or when there are devices on the system bus that
access system memory, but do not perform snooping of memory accesses. It enforces
coherency between caches in the processors and system memory.
• Write-back (WB)—Writes and reads to and from system memory are cached. Reads come
from cache lines on cache hits; read misses cause cache fills. Speculative reads are
allowed. Write misses cause cache line fills (in the P6 family processors), and writes are
performed entirely in the cache, when possible. Write combining is allowed. The write-
back memory type reduces bus traffic by eliminating many unnecessary writes to system
memory. Writes to a cache line are not immediately forwarded to system memory; instead,

9-6
 MEMORY CACHE CONTROL

they are accumulated in the cache. The modified cache lines are written to system memory
later, when a write-back operation is performed. Write-back operations are triggered when
cache lines need to be deallocated, such as when new cache lines are being allocated in a
cache that is already full. They also are triggered by the mechanisms used to maintain
cache consistency. This type of cache-control provides the best performance, but it requires
that all devices that access system memory on the system bus be able to snoop memory
accesses to insure system memory and cache coherency.
• Write protected (WP)—Reads come from cache lines when possible, and read misses
cause cache fills. Writes are propagated to the system bus and cause corresponding cache
lines on all processors on the bus to be invalidated. Speculative reads are allowed. This
caching option is available in the P6 family processors by programming the MTRRs
(seeTable 9-5).

9.3.1. Buffering of Write Combining Memory Locations

Writes to WC memory are not cached in the typical sense of the word cached. They are retained
in an internal buffer that is separate from the internal L1 and L2 caches. The buffer is not
snooped and thus does not provide data coherency. The write buffering is done to allow software
a small window of time to supply more modified data to the buffer while remaining as nonintru-
sive to software as possible. The size of the buffer is not architecturally defined, However the
Pentium® Pro and Pentium® II processors implement a single concurrent 32-byte buffer. The size
of this buffer was chosen by implementation convenience. In the Pentium® III processor there
are 4 write combine buffers. The size is the same as for the Pentium® Pro and Pentium® II proces-
sors. Buffer size and quantity changes may occur in future generations of the P6 family proces-
sors and so software should not rely upon the current 32-byte WC buffer size or the existence of
a single concurrent buffer or the 4 buffers in the Penitum III processor. The WC buffering of
writes also causes data to be collapsed (for example, multiple writes to the same location will
leave the last data written in the location and the other writes will be lost).
For the Pentium® Pro and Pentium® II processors, once software writes to a region of memory
that is addressed outside of the range of the current 32-byte buffer, the data in the buffer is auto-
matically forwarded to the system bus and written to memory. Therefore software that writes
more than one 32-byte buffers worth of data will ensure that the data from the first buffers
address range is forwarded to memory. The last buffer written in the sequence may be delayed
by the processor longer unless the buffers are deliberately emptied. Software developers should
not rely on the fact that there is only one active WC buffer at a time. Software developers
creating software that is sensitive to data being delayed must deliberately empty the WC buffers
and not assume the hardware will.
Once the processor has started to move data into the WC buffer, it will make a bus transaction
style decision based on how much of the buffer contains valid data. If the buffer is full (for
example, all 32 bytes are valid) the processor will execute a burst write transaction on the bus
that will result in all 32 bytes being transmitted on the data bus in a single transaction. If one or
more of the WC buffer’s bytes are invalid (for example, have not been written by software) then
the processor will start to move the data to memory using “partial write” transactions on the
system bus. There will be a maximum of 4 partial write transactions for one WC buffer of data
sent to memory. Once data in the WC buffer has started to be propagated to memory, the data is

9-7
MEMORY CACHE CONTROL

subject to the weak ordering semantics of its definition. Ordering is not maintained between the
successive allocation/deallocation of WC buffers (for example, writes to WC buffer 1 followed
by writes to WC buffer 2 may appear as buffer 2 followed by buffer 1 on the system bus. When
a WC buffer is propagated to memory as partial writes there is no guaranteed ordering between
successive partial writes (for example, a partial write for chunk 2 may appear on the bus before
the partial write for chunk 1 or vice versa). The only elements of WC propagation to the system
bus that are guaranteed are those provided by transaction atomicity. For the P6 family proces-
sors, a completely full WC buffer will always be propagated as a single burst transaction using
any of the chunk orders. In a WC buffer propagation where the data will be propagated as
partials, all data contained in the same chunk (0 mod 8 aligned) will be propagated simulta-
neously.

9.3.2. Choosing a Memory Type

The simplest system memory model does not use memory-mapped I/O with read or write side
effects, does not include a frame buffer, and uses the write-back memory type for all memory.
An I/O agent can perform direct memory access (DMA) to write-back memory and the cache
protocol maintains cache coherency.
A system can use uncacheable memory for other memory-mapped I/O, and should always use
uncacheable memory for memory-mapped I/O with read side effects.
Dual-ported memory can be considered a write side effect, making relatively prompt writes
desirable, because those writes cannot be observed at the other port until they reach the memory
agent. A system can use uncacheable, write-through, or write-combining memory for frame
buffers or dual-ported memory that contains pixel values displayed on a screen. Frame buffer
memory is typically large (a few megabytes) and is usually written more than it is read by the
processor. Using uncacheable memory for a frame buffer generates very large amounts of bus
traffic, because operations on the entire buffer are implemented using partial writes rather than
line writes. Using write-through memory for a frame buffer can displace almost all other useful
cached lines in the processor’s L2 cache and L1 data cache. Therefore, systems should use write-
combining memory for frame buffers whenever possible.
Software can use page-level cache control, to assign appropriate effective memory types when
software will not access data structures in ways that benefit from write-back caching. For
example, software may read a large data structure once and not access the structure again until
the structure is rewritten by another agent. Such a large data structure should be marked as
uncacheable, or reading it will evict cached lines that the processor will be referencing again. A
similar example would be a write-only data structure that is written to (to export the data to
another agent), but never read by software. Such a structure can be marked as uncacheable,
because software never reads the values that it writes (though as uncacheable memory, it will be
written using partial writes, while as write-back memory, it will be written using line writes,
which may not occur until the other agent reads the structure and triggers implicit write-backs).
On the Pentium® III processor, new capabilities exist that may allow the programmer to perform
similar functions with the prefetch and streaming store instructions. For more information on
these instructions, see Section 3.2., “Instruction Reference” in Chapter 3, Instruction Set Refer-
ence.

9-8
 MEMORY CACHE CONTROL

9.4. CACHE CONTROL PROTOCOL

The following section describes the cache control protocol currently defined for the Intel Archi-
tecture processors. This protocol is used by the P6 family and Pentium® processors. The
Intel486™ processor uses an implementation defined protocol that does not support the MESI
four-state protocol, but instead uses a two-state protocol with valid and invalid states defined.
In the L1 data cache and the P6 family processors’ L2 cache, the MESI (modified, exclusive,
shared, invalid) cache protocol maintains consistency with caches of other processors. The L1
data cache and the L2 cache has two MESI status flags per cache line. Each line can thus be
marked as being in one of the states defined in Table 9-3. In general, the operation of the MESI
protocol is transparent to programs.
The L1 instruction cache implements only the “SI” part of the MESI protocol, because the
instruction cache is not writable. The instruction cache monitors changes in the data cache to
maintain consistency between the caches when instructions are modified. See Section 9.7.,
“Self-Modifying Code”, for more information on the implications of caching instructions.

Table 9-3. MESI Cache Line States

Cache Line State M (Modified) E (Exclusive) S (Shared) I (Invalid)
This cache line is valid? Yes Yes Yes No
The memory copy is… …out of date …valid …valid —
Copies exist in caches of No No Maybe Maybe
other processors?
A write to this line … …does not go to …does not go to …causes the …goes directly
bus bus processor to to bus
gain exclusive
ownership of the
line

9.5. CACHE CONTROL

The current Intel Architecture provides the following cache-control mechanisms for use in
enabling caching and/or restricting caching to various pages or regions in memory (see Figure
9-2):
• CD flag, bit 30 of control register CR0—Controls caching of system memory locations.
For more information, see Section 2.5., “Control Registers”, in Chapter 2, System Archi-
tecture Overview. If the CD flag is clear, caching is enabled for the whole of system
memory, but may be restricted for individual pages or regions of memory by other cache-
control mechanisms. When the CD flag is set, caching is restricted in the L1 and L2 caches
for the P6 family processors and prevented for the Pentium ® and Intel486™ processors
(see note below). With the CD flag set, however, the caches will still respond to snoop
traffic. Caches should be explicitly flushed to insure memory coherency. For highest
processor performance, both the CD and the NW flags in control register CR0 should be
cleared. Table 9-4 shows the interaction of the CD and NW flags.

9-9
MEMORY CACHE CONTROL

NOTE
The effect of setting the CD flag is somewhat different for the P6 family, Pentium®,
and Intel486™ processors (see Table 9-4). To insure memory coherency after the
CD flag is set, the caches should be explicitly flushed. For more information, see
Section 9.5.2., “Preventing Caching”. Setting the CD flag for the P6 family
processors modifies cache line fill and update behaviour. Also for the P6 family
processors, setting the CD flag does not force strict ordering of memory accesses
unless the MTRRs are disabled and/or all memory is referenced as uncached. For
more information, see Section 7.2.4., “Strengthening or Weakening the Memory
Ordering Model”, in Chapter 7, Multiple-Processor Management.

CR4
P
G
E
Enables global pages
CR3 designated with G flag Physical Memory
P P
FFFFFFFFH2
C W
D T
Control caching of
page directory

Page-Directory or
CR0 Page-Table Entry
C N P P MTRRs3
D W G1 C W
D T

MTRRs control caching

of selected regions of
CD and NW Flags PCD and PWT flags
physical memory
control overall caching control page-level
of system memory caching Memory Types Allowed:
—Uncacheable (UC)
G flag controls page- —Write-Protected (WP)
level flushing of TLBs —Write-Combining (WC)
0
—Write-Through (WT)
Write Buffer —Write-Back (WB)
TLBs

1. G flag only available in P6 family processors.

2. If 36-bit physical addressing is being used, the maximum
physical address size is FFFFFFFFFH.
3. MTRRs available only in P6 family processors;
similar control available in Pentium® processor with
KEN# and WB/WT# pins, and in Intel486™ processor.

Figure 9-2. Cache-Control Mechanisms Available in the Intel Architecture Processors

9-10
 MEMORY CACHE CONTROL

Table 9-4. Cache Operating Modes

CD NW Caching and Read/Write Policy L1 L21
0 0 Normal highest performance cache operation.
- Read hits access the cache; read misses may cause replacement. Yes Yes
- Write hits update the cache. Yes Yes
- (Pentium® and P6 family processors.) Only writes to shared lines Yes Yes
and write misses update system memory.
- (P6 family processors.) Write misses cause cache line fills; write Yes Yes
hits can change shared lines to exclusive under control of the MTRRs
- (Pentium® processor.) Write misses do not cause cache line fills; write Yes
hits can change shared lines to exclusive under control of WB/WT#.
- (Intel486™ processor.) All writes update system memory; write misses Yes
do not cause cache line fills.
- Invalidation is allowed. Yes Yes
- External snoop traffic is supported. Yes Yes
0 1 Invalid setting.
A general-protection exception (#GP) with an error code of 0 is NA NA
generated.
1 0 Memory coherency is maintained.
- Read hits access the cache; read misses do not cause replacement. Yes Yes
- Write hits update the cache. Yes Yes
- (Pentium® and P6 family processors.) Only writes to shared lines Yes Yes
and write misses update system memory.
- (Intel486™ processor.) All writes update system memory Yes
- (Pentium® processor.) Write hits can change shared lines to exclusive Yes
under control of the WB/WT#.
- (P6 family processors.) Strict memory ordering is not enforced Yes Yes
unless the MTRRs are disabled and/or all memory is referenced as
uncached. For more information, see Section 7.2.4., “Strengthening or
Weakening the Memory Ordering Model”.
- Invalidation is allowed. Yes Yes
- External snoop traffic is supported. Yes Yes
1 1 Memory coherency is not maintained. This is the state of the processor
after a power up or reset.
- Read hits access the cache; read misses do not cause replacement. Yes Yes
- Write hits update the cache. Yes Yes
- (Pentium® and P6 family processors.) Write hits change exclusive Yes Yes
lines to modified.
- (Pentium® and P6 family processors.) Shared lines remain shared Yes Yes
after write hit.
- Write misses access memory. Yes Yes
- (P6 family processors.) Strict memory ordering is not enforced Yes Yes
unless the MTRRs are disabled and/or all memory is referenced as
uncached. For more information, see Section 7.2.4., “Strengthening or
Weakening the Memory Ordering Model”.
- Invalidation is inhibited when snooping; but is allowed with INVD and Yes Yes
WBINVD instructions. No Yes
- External snoop traffic is supported.

NOTE:
1. The P6 family processors are the only Intel Architecture processors that contain an integrated L2 cache.
The L2 column in this table is definitive for the P6 family processors. It is intended to represent what could
be implemented in a Pentium® processor based system with a platform specific write-back L2 cache.

9-11
MEMORY CACHE CONTROL

• NW flag, bit 29 of control register CR0—Controls the write policy for system memory
locations. For more information, see Section 2.5., “Control Registers”, in Chapter 2,
System Architecture Overview. If the NW and CD flags are clear, write-back is enabled for
the whole of system memory (write-through for the Intel486™ processor), but may be
restricted for individual pages or regions of memory by other cache-control mechanisms.
Table 9-4 shows how the other combinations of CD and NW flags affects caching.

NOTE
®
For the Pentium processor, when the L1 cache is disabled (the CD and NW
flags in control register CR0 are set), external snoops are accepted in DP
(dual-processor) systems and inhibited in uniprocessor systems. When
snoops are inhibited, address parity is not checked and APCHK# is not
asserted for a corrupt address; however, when snoops are accepted, address
parity is checked and APCHK# is asserted for corrupt addresses.
• PCD flag in the page-directory and page-table entries—Controls caching for individual
page tables and pages, respectively. For more information, see Section 3.6.4., “Page-
Directory and Page-Table Entries”, in Chapter 3, Protected-Mode Memory Management.
This flag only has effect when paging is enabled and the CD flag in control register CR0 is
clear. The PCD flag enables caching of the page table or page when clear and prevents
caching when set.
• PWT flag in the page-directory and page-table entries—Controls the write policy for
individual page tables and pages, respectively. For more information, see Section 3.6.4.,
“Page-Directory and Page-Table Entries”, in Chapter 3, Protected-Mode Memory
Management. This flag only has effect when paging is enabled and the NW flag in control
register CR0 is clear. The PWT flag enables write-back caching of the page table or page
when clear and write-through caching when set.
• PCD and PWT flags in control register CR3. Control the global caching and write policy
for the page directory. For more information, see Section 2.5., “Control Registers”, in
Chapter 2, System Architecture Overview. The PCD flag enables caching of the page
directory when clear and prevents caching when set. The PWT flag enables write-back
caching of the page directory when clear and write-through caching when set. These flags
do not affect the caching and write policy for individual page tables. These flags only have
effect when paging is enabled and the CD flag in control register CR0 is clear.
• G (global) flag in the page-directory and page-table entries (introduced to the Intel Archi-
tecture in the P6 family processors)—Controls the flushing of TLB entries for individual
pages. See Section 3.7., “Translation Lookaside Buffers (TLBs)”, in Chapter 3, Protected-
Mode Memory Management, for more information about this flag.
• PGE (page global enable) flag in control register CR4—Enables the establishment of
global pages with the G flag. See Section 3.7., “Translation Lookaside Buffers (TLBs)”, in
Chapter 3, Protected-Mode Memory Management, for more information about this flag.
• Memory type range registers (MTRRs) (introduced in the P6 family processors)—Control
the type of caching used in specific regions of physical memory. Any of the caching types
described in Section 9.3., “Methods of Caching Available”, can be selected. See Section

9-12
 MEMORY CACHE CONTROL

9.12., “Memory Type Range Registers (MTRRs)”, for a detailed description of the
MTRRs.
• KEN# and WB/WT# pins on Pentium® processor and KEN# pin alone on the Intel486™
processor—These pins allow external hardware to control the caching method used for
specific areas of memory. They perform similar (but not identical) functions to the MTRRs
in the P6 family processors.
• PCD and PWT pins on the Pentium® and Intel486™ processors—These pins (which are
associated with the PCD and PWT flags in control register CR3 and in the page-directory
and page-table entries) permit caching in an external L2 cache to be controlled on a page-
by-page basis, consistent with the control exercised on the L1 cache of these processors.
The P6 family processors do not provide these pins because the L2 cache in internal to the
chip package.

9.5.1. Precedence of Cache Controls (P6 Family Processor)

In the P6 family processors, the cache control flags and MTRRs operate hierarchically for
restricting caching. That is, if the CD flag is set, caching is prevented globally (see Table 9-4).
If the CD flag is clear, either the PCD flags and/or the MTRRs can be used to restrict caching.
If there is an overlap of page-level caching control and MTRR caching control, the mechanism
that prevents caching has precedence. For example, if an MTRR makes a region of system
memory uncachable, a PCD flag cannot be used to enable caching for a page in that region. The
converse is also true; that is, if the PCD flag is set, an MTRR cannot be used to make a region
of system memory cacheable.
In cases where there is a overlap in the assignment of the write-back and write-through caching
policies to a page and a region of memory, the write-through policy takes precedence. The write-
combining policy (which can only be assigned through an MTRR) takes precedence over either
write-through or write-back.
Table 9-5 describes the mapping from MTRR memory types and page-level caching attributes
to effective memory types, when normal caching is in effect (the CD and NW flags in control
register CR0 are clear). Combinations that appear in gray are implementation-defined and may
be implemented differently on future Intel Architecture processors. System designers are
encouraged to avoid these implementation-defined combinations.
When normal caching is in effect, the effective memory type is determined using the following
rules:
1. If the PCD and PWT attributes for the page are both 0, then the effective memory type is
identical to the MTRR-defined memory type.
2. If the PCD flag is set, then the effective memory type is UC.
3. If the PCD flag is clear and the PWT flag is set, the effective memory type is WT for the
WB memory type and the MTRR-defined memory type for all other memory types.
4. Setting the PCD and PWT flags to opposite values is considered model-specific for the WP
and WC memory types and architecturally-defined for the WB, WT, and UC memory
types.

9-13
MEMORY CACHE CONTROL

Table 9-5. Effective Memory Type Depending on MTRR, PCD, and PWT Settings
MTRR Memory Type PCD Value PWT Value Effective Memory Type
UC X X UC
WC 0 0 WC
0 1 WC
1 0 WC
1 1 UC
WT 0 X WT
1 X UC
WP 0 0 WP
0 1 WP
1 0 WC
1 1 UC
WB 0 0 WB
0 1 WT
1 X UC

NOTE:
This table assumes that the CD and NW flags in register CR0 are set to 0. The effective memory types in the
grey areas are implementation defined and may be different in future Intel Architecture processors.

9.5.2. Preventing Caching

To prevent the L1 and L2 caches from performing caching operations after they have been
enabled and have received cache fills, perform the following steps:
1. Enter the no-fill cache mode. (Set the CD flag in control register CR0 to 1 and the NW flag
to 0.
2. Flush all caches using the WBINVD instruction.
3. Disable the MTRRs and set the default memory type to uncached or set all MTRRs for the
uncached memory type. For more information, see the discussion of the TYPE field and
the E flag in Section 9.12.2.1., “MTRRdefType Register”.
The caches must be flushed when the CD flag is cleared to insure system memory coherency. If
the caches are not flushed in step 2, cache hits on reads will still occur and data will be read from
valid cache lines.

9-14
 MEMORY CACHE CONTROL

9.6. CACHE MANAGEMENT INSTRUCTIONS

The INVD and WBINVD instructions are used to invalidate the contents of the L1 and L2
caches. The INVD instruction invalidates all internal cache entries, then generates a special-
function bus cycle that indicates that external caches also should be invalidated. The INVD
instruction should be used with care. It does not force a write-back of modified cache lines;
therefore, data stored in the caches and not written back to system memory will be lost. Unless
there is a specific requirement or benefit to invalidating the caches without writing back the
modified lines (such as, during testing or fault recovery where cache coherency with main
memory is not a concern), software should use the WBINVD instruction.
The WBINVD instruction first writes back any modified lines in all the internal caches, then
invalidates the contents of both L1 and L2 caches. It ensures that cache coherency with main
memory is maintained regardless of the write policy in effect (that is, write-through or write-
back). Following this operation, the WBINVD instruction generates one (P6 family processors)
or two (Pentium® and Intel486™ processors) special-function bus cycles to indicate to external
cache controllers that write-back of modified data followed by invalidation of external caches
should occur.

9.7. SELF-MODIFYING CODE

A write to a memory location in a code segment that is currently cached in the processor causes
the associated cache line (or lines) to be invalidated. This check is based on the physical address
of the instruction. In addition, the P6 family and Pentium® processors check whether a write to
a code segment may modify an instruction that has been prefetched for execution. If the write
affects a prefetched instruction, the prefetch queue is invalidated. This latter check is based on
the linear address of the instruction.
In practice, the check on linear addresses should not create compatibility problems among Intel
Architecture processors. Applications that include self-modifying code use the same linear
address for modifying and fetching the instruction. Systems software, such as a debugger, that
might possibly modify an instruction using a different linear address than that used to fetch the
instruction, will execute a serializing operation, such as a CPUID instruction, before the modi-
fied instruction is executed, which will automatically resynchronize the instruction cache and
prefetch queue. See Section 7.1.3., “Handling Self- and Cross-Modifying Code”, in Chapter 7,
Multiple-Processor Management, for more information about the use of self-modifying code.
For Intel486™ processors, a write to an instruction in the cache will modify it in both the cache
and memory, but if the instruction was prefetched before the write, the old version of the instruc-
tion could be the one executed. To prevent the old instruction from being executed, flush the
instruction prefetch unit by coding a jump instruction immediately after any write that modifies
an instruction.

9-15
MEMORY CACHE CONTROL

9.8. IMPLICIT CACHING (P6 FAMILY PROCESSORS)

Implicit caching occurs when a memory element is made potentially cacheable, although the
element may never have been accessed in the normal von Neumann sequence. Implicit caching
occurs on the P6 family processors due to aggressive prefetching, branch prediction, and TLB
miss handling. Implicit caching is an extension of the behavior of existing Intel386™,
Intel486™, and Pentium® processor systems, since software running on these processor families
also has not been able to deterministically predict the behavior of instruction prefetch.
To avoid problems related to implicit caching, the operating system must explicitly invalidate
the cache when changes are made to cacheable data that the cache coherency mechanism does
not automatically handle. This includes writes to dual-ported or physically aliased memory
boards that are not detected by the snooping mechanisms of the processor, and changes to page-
table entries in memory.
The code in Example 9-1 shows the effect of implicit caching on page-table entries. The linear
address F000H points to physical location B000H (the page-table entry for F000H contains the
value B000H), and the page-table entry for linear address F000 is PTE_F000.

Example 9-1. Effect of Implicit Caching on Page-Table Entries

mov EAX, CR3 ; Invalidate the TLB
mov CR3, EAX ; by copying CR3 to itself
mov PTE_F000, A000H; Change F000H to point to A000H
mov EBX, [F000H];

Because of speculative execution in the P6 family processors, the last MOV instruction
performed would place the value at physical location B000H into EBX, rather than the value at
the new physical address A000H. This situation is remedied by placing a TLB invalidation
between the load and the store.

9.9. EXPLICIT CACHING

The Pentium® III processor introduced a new instruction designed to provide some control over
caching of data. The prefetch instruction is a “hint” to the processor that the data requested by
the prefetch instruction should be read into cache, even though it is not needed yet. The
processor assumes it will be needed soon.
Explicit caching occurs when the application program executes a prefetch instruction. The
programmer must be judicious in the use of the prefetch instruction. Overuse can lead to
resource conflicts and hence reduce the performance of an application. For more detailed infor-
mation on the proper use of the prefetch instruction, refer to Chapter 6, “Optimizing Cache Utili-
zation for Pentium® III Processors”, in the Intel Architecture Optimization Reference Manual
(Order Number 245127-001).
Prefetch can be used to read data into the cache prior to the application actually requiring it. This
helps to reduce the long latency typically associated with reading data from memory and causing
the processor to “stall”. It is important to remember that prefetch is only a hint to the processor

9-16
 MEMORY CACHE CONTROL

to fetch the data now or as soon as possible. It will be used soon. The prefetch instruction has
different variations that allow the programmer to control into which cache level the data will be
read. For more information on the variations of the prefetch instruction refer to Section 9.5.3.1.,
“Cacheability Hint Instructions”, Chapter 9, Programming with the Streaming SIMD Exten-
sions, if the Intel Architecture Software Developer’s Manual, Volume 2.

9.10. INVALIDATING THE TRANSLATION LOOKASIDE BUFFERS

(TLBS)
The processor updates its address translation caches (TLBs) transparently to software. Several
mechanisms are available, however, that allow software and hardware to invalidate the TLBs
either explicitly or as a side effect of another operation.
The INVLPG instruction invalidates the TLB for a specific page. This instruction is the most
efficient in cases where software only needs to invalidate a specific page, because it improves
performance over invalidating the whole TLB. This instruction is not affected by the state of the
G flag in a page-directory or page-table entry.
The following operations invalidate all TLB entries except global entries. (A global entry is one
for which the G (global) flag is set in its corresponding page-directory or page-table entry. The
global flag was introduced into the Intel Architecture in the P6 family processors, see Section
9.5., “Cache Control”.)
• Writing to control register CR3.
• A task switch that changes control register CR3.
The following operations invalidate all TLB entries, irrespective of the setting of the G flag:
• Asserting or de-asserting the FLUSH# pin.
• (P6 family processors only.) Writing to an MTRR (with a WRMSR instruction).
• Writing to control register CR0 to modify the PG or PE flag.
• (P6 family processors only.) Writing to control register CR4 to modify the PSE, PGE, or
PAE flag.
See Section 3.7., “Translation Lookaside Buffers (TLBs)”, in Chapter 3, Protected-Mode
Memory Management, for additional information about the TLBs.

9.11. WRITE BUFFER

Intel Architecture processors temporarily store each write (store) to memory in a write buffer.
The write buffer improves processor performance by allowing the processor to continue
executing instructions without having to wait until a write to memory and/or to a cache is
complete. It also allows writes to be delayed for more efficient use of memory-access bus cycles.

9-17
MEMORY CACHE CONTROL

In general, the existence of the write buffer is transparent to software, even in systems that use
multiple processors. The processor ensures that write operations are always carried out in
program order. It also insures that the contents of the write buffer are always drained to memory
in the following situations:
• When an exception or interrupt is generated.
• (P6 family processors only.) When a serializing instruction is executed.
• When an I/O instruction is executed.
• When a LOCK operation is performed.
• (P6 family processors only.) When a BINIT operation is performed.
• (Pentium® III processors only.) When using SFENCE to order stores.
The discussion of write ordering in Section 7.2., “Memory Ordering”, in Chapter 7, Multiple-
Processor Management, gives a detailed description of the operation of the write buffer.

9.12. MEMORY TYPE RANGE REGISTERS (MTRRS)

The following section pertains only to the P6 family processors.
The memory type range registers (MTRRs) provide a mechanism for associating the memory
types with physical-address ranges in system memory. For more information, see Section 9.3.,
“Methods of Caching Available”. They allow the processor to optimize operations for different
types of memory such as RAM, ROM, frame-buffer memory, and memory-mapped I/O devices.
They also simplify system hardware design by eliminating the memory control pins used for this
function on earlier Intel Architecture processors and the external logic needed to drive them.
The MTRR mechanism allows up to 96 memory ranges to be defined in physical memory, and
it defines a set of model-specific registers (MSRs) for specifying the type of memory that is
contained in each range. Table 9-6 shows the memory types that can be specified and their prop-
erties; Figure 9-3 shows the mapping of physical memory with MTRRs. See Section 9.3.,
“Methods of Caching Available”, for a more detailed description of each memory type.
Following a hardware reset, a P6 family processor disables all the fixed and variable MTRRs,
which in effect makes all of physical memory uncachable. Initialization software should then set
the MTRRs to a specific, system-defined memory map. Typically, the BIOS (basic input/output
system) software configures the MTRRs. The operating system or executive is then free to
modify the memory map using the normal page-level cacheability attributes.
In a multiprocessor system, different P6 family processors MUST use the identical MTRR
memory map so that software has a consistent view of memory, independent of the processor
executing a program.

9-18
 MEMORY CACHE CONTROL

Table 9-6. MTRR Memory Types and Their Properties

Cacheable in Allows
Encoding in L1 and L2 Writeback Speculative Memory Ordering
Mnemonic MTRR Caches Cacheable Reads Model
Uncacheable 0 No No No Strong Ordering
(UC)
Write Combining 1 No No Yes Weak Ordering
(WC)
Write-through 4 Yes No Yes Speculative
(WT) Processor Ordering
Write-protected 5 Yes for reads, No Yes Speculative
(WP) no for writes Processor Ordering
Writeback (WB) 6 Yes Yes Yes Speculative
Processor Ordering
Reserved 2, 3,
Encodings* 7 through 255

NOTE:
* Using these encoding result in a general-protection exception (#GP) being generated.

9-19
MEMORY CACHE CONTROL

Physical Memory
FFFFFFFFH

Address ranges not

mapped by an MTRR
are set to a default type

8 variable ranges
(from 4 KBytes to
maximum size of
physical memory)

100000H
64 fixed ranges FFFFFH
(4 KBytes each) 256 KBytes
C0000H
16 fixed ranges BFFFFH
256 KBytes
(16 KBytes each) 80000H
7FFFFH
8 fixed ranges
(64-KBytes each) 512 KBytes
0

Figure 9-3. Mapping Physical Memory With MTRRs

9.12.1. MTRR Feature Identification

The availability of the MTRR feature is model-specific. Software can determine if MTRRs are
supported on a processor by executing the CPUID instruction and reading the state of the MTRR
flag (bit 12) in the feature information register (EDX).
If the MTRR flag is set (indicating that the processor implements MTRRs), additional informa-
tion about MTRRs can be obtained from the 64-bit MTRRcap register. The MTRRcap register
is a read-only MSR that can be read with the RDMSR instruction. Figure 9-4 shows the contents
of the MTRRcap register. The functions of the flags and field in this register are as follows:
VCNT (variable range registers count) field, bits 0 through 7
Indicates the number of variable ranges implemented on the processor. The P6
family processors have eight pairs of MTRRs for setting up eight variable
ranges.

9-20
 MEMORY CACHE CONTROL

63 11 10 9 8 7 0
W F
Reserved C I VCNT
X

WC—Write-combining memory type supported

FIX—Fixed range registers supported
VCNT—Number of variable range registers
Reserved

Figure 9-4. MTRRcap Register

FIX (fixed range registers supported) flag, bit 8

Fixed range MTRRs (MTRRfix64K_00000 through MTRRfix4K_0F8000)
are supported when set; no fixed range registers are supported when clear.
WC (write combining) flag, bit 10
The write-combining (WC) memory type is supported when set; the WC type
is not supported when clear.
Bit 9 and bits 11 through 63 in the MTRRcap register are reserved. If software attempts to write
to the MTRRcap registers, a general-protection exception (#GP) is generated.
For the P6 family processors, the MTRRcap register always contains the value 508H.

9.12.2. Setting Memory Ranges with MTRRs

The memory ranges and the types of memory specified in each range are set by three groups of
registers: the MTRRdefType register, the fixed-range MTRRs, and the variable range MTRRs.
These registers can be read and written to using the RDMSR and WRMSR instructions, respec-
tively. The MTRRcap register indicates the availability of these registers on the processor. For
more information, see Section 9.12.1., “MTRR Feature Identification”.

9.12.2.1. MTRRDEFTYPE REGISTER

The MTRRdefType register (see Figure 9-4) sets the default properties of the regions of physical
memory that are not encompassed by MTRRs. For more information, see Section 9.4., “Cache
Control Protocol”. The functions of the flags and field in this register are as follows:
Type field, bits 0 through 7
Indicates the default memory type used for those physical memory address
ranges that do not have a memory type specified for them by an MTRR. See
Table 9-6 for the encoding of this field. If the MTRRs are disabled, this field
defines the memory type for all of physical memory. The legal values for this
field are 0, 1, 4, 5, and 6. All other values result in a general-protection excep-
tion (#GP) being generated.

9-21
MEMORY CACHE CONTROL

Intel recommends the use of the UC (uncached) memory type for all physical
memory addresses where memory does not exist. To assign the UC type to
nonexistent memory locations, it can either be specified as the default type in
the Type field or be explicitly assigned with the fixed and variable MTRRs.

63 12 11 10 9 8 7 0
F
Reserved E E Type

E—MTRR enable/disable
FE—Fixed-range MTRRs enable/disable
Type—Default memory type

Reserved

Figure 9-5. MTRRdefType Register

FE (fixed MTRRs enabled) flag, bit 10

Fixed-range MTRRs are enabled when set; fixed-range MTRRs are disabled
when clear. When the fixed-range MTRRs are enabled, they take priority over
the variable-range MTRRs when overlaps in ranges occur. If the fixed-range
MTRRs are disabled, the variable-range MTRRs can still be used and can map
the range ordinarily covered by the fixed-range MTRRs.
E (MTRRs enabled) flag, bit 11
MTRRs are enabled when set; all MTRRs are disabled when clear, and the UC
memory type is applied to all of physical memory. When this flag is set, the FE
flag can disable the fixed-range MTRRs; when the flag is clear, the FE flag has
no affect. When the E flag is set, the type specified in the default memory type
field is used for areas of memory not already mapped by either a fixed or vari-
able MTRR.
Bits 8 and 9, and bits 12 through 63, in the MTRRdefType register are reserved; the processor
generates a general-protection exception (#GP) if software attempts to write nonzero values to
them.

9.12.2.2. FIXED RANGE MTRRS

The fixed memory ranges are mapped with 8 fixed-range registers of 64 bits each. Each of these
registers is divided into 8-bit fields that are used to specify the memory type for each of the sub-
ranges the register controls. Table 9-7 shows the relationship between the fixed physical-address
ranges and the corresponding fields of the fixed-range MTRRs; Table 9-6 shows the encoding
of these field:
• Register MTRRfix64K_00000. Maps the 512-KByte address range from 0H to 7FFFFH.
This range is divided into eight 64-KByte sub-ranges.

9-22
 MEMORY CACHE CONTROL

• Registers MTRRfix16K_80000 and MTRRfix16K_A0000. Maps the two 128-KByte

address ranges from 80000H to BFFFFH. This range is divided into sixteen 16-KByte sub-
ranges, 8 ranges per register.
• Registers MTRRfix4K_C0000. and MTRRfix4K_F8000. Maps eight 32-KByte address
ranges from C0000H to FFFFFH. This range is divided into sixty-four 4-KByte sub-
ranges, 8 ranges per register.
See the Pentium® Pro BIOS Writer’s Guide for examples of assigning memory types with fixed-
range MTRRs.

Table 9-7. Address Mapping for Fixed-Range MTRRs

Address Range (hexadecimal) Register
63 56 55 48 47 40 39 32 31 24 23 16 15 8 7 0
70000- 60000- 50000- 40000- 30000- 20000- 10000- 00000- MTRRfix64K
7FFFF 6FFFF 5FFFF 4FFFF 3FFFF 2FFFF 1FFFF 0FFFF _00000
9C000 98000- 94000- 90000- 8C000- 88000- 84000- 80000- MTRRfix16K
9FFFF 98FFF 97FFF 93FFF 8FFFF 8BFFF 87FFF 83FFF _80000
BC000 B8000- B4000- B0000- AC000- A8000- A4000- A0000- MTRRfix16K
BFFFF BBFFF B7FFF B3FFF AFFFF ABFFF A7FFF A3FFF _A0000
C7000 C6000- C5000- C4000- C3000- C2000- C1000- C0000- MTRRfix4K_
C7FFF C6FFF C5FFF C4FFF C3FFF C2FFF C1FFF C0FFF C0000
CF000 CE000- CD000- CC000- CB000- CA000- C9000- C8000- MTRRfix4K_
CFFFF CEFFF CDFFF CCFFF CBFFF CAFFF C9FFF C8FFF C8000
D7000 D6000- D5000- D4000- D3000- D2000- D1000- D0000- MTRRfix4K_
D7FFF D6FFF D5FFF D4FFF D3FFF D2FFF D1FFF D0FFF D0000
DF000 DE000- DD000- DC000- DB000- DA000- D9000- D8000- MTRRfix4K_
DFFFF DEFFF DDFFF DCFFF DBFFF DAFFF D9FFF D8FFF D8000
E7000 E6000- E5000- E4000- E3000- E2000- E1000- E0000- MTRRfix4K_
E7FFF E6FFF E5FFF E4FFF E3FFF E2FFF E1FFF E0FFF E0000
EF000 EE000- ED000- EC000- EB000- EA000- E9000- E8000- MTRRfix4K_
EFFFF EEFFF EDFFF ECFFF EBFFF EAFFF E9FFF E8FFF E8000
F7000 F6000- F5000- F4000- F3000- F2000- F1000- F0000- MTRRfix4K_
F7FFF F6FFF F5FFF F4FFF F3FFF F2FFF F1FFF F0FFF F0000
FF000 FE000- FD000- FC000- FB000- FA000- F9000- F8000- MTRRfix4K_
FFFFF FEFFF FDFFF FCFFF FBFFF FAFFF F9FFF F8FFF F8000

9.12.2.3. VARIABLE RANGE MTRRS

The P6 family processors permit software to specify the memory type for eight variable-size
address ranges, using a pair of MTRRs for each range. The first of each pair (MTRRphysBasen)
defines the base address and memory type for the range, and the second (MTRRphysMaskn)
contains a mask that is used to determine the address range. The “n” suffix indicates registers
pairs 0 through 7. Figure 9-6 shows flags and fields in these registers. The functions of the flags
and fields in these registers are as follows:

9-23
MEMORY CACHE CONTROL

Type field, bits 0 through 7

Specifies the memory type for the range. See Table 9-6 for the encoding of this
field.

MTRRphysBasen Register
63 36 35 12 11 8 7 0

Reserved PhysBase Type

PhysBase—Base address of range

Type—Memory type for range

MTRRphysMaskn Register
63 36 35 12 11 10 0

Reserved PhysMask V Reserved

PhysMask—Sets range mask

V—Valid
Reserved

Figure 9-6. MTRRphysBasen and MTRRphysMaskn Variable-Range Register Pair

PhysBase field, bits 12 through 35

Specifies the base address of the address range. This 24-bit value is extended
by 12 bits at the low end to form the base address, which automatically aligns
the address on a 4-KByte boundary.
PhysMask field, bits 12 through 35
Specifies a 24-bit mask that determines the range of the region being mapped,
according to the following relationship:
Address_Within_Range AND PhysMask = PhysBase AND PhysMask
This 24-bit value is extended by 12 bits at the low end to form the mask value.
See Section 9.12.3., “Example Base and Mask Calculations”, for more infor-
mation and some examples of base address and mask computations.
V (valid) flag, bit 11
Enables the register pair when set; disables register pair when clear.
All other bits in the MTRRphysBasen and MTRRphysMaskn registers are reserved; the
processor generates a general-protection exception (#GP) if software attempts to write to them.
Overlapping variable MTRR ranges are not supported generically. However, two variable
ranges are allowed to overlap, if the following conditions are present:
• If both of them are UC (uncached).

9-24
 MEMORY CACHE CONTROL

• If one range is of type UC and the other is of type WB (write back).

In both cases above, the effective type for the overlapping region is UC. The processor’s
behavior is undefined for all other cases of overlapping variable ranges.
A variable range can overlap a fixed range (provided the fixed range MTRR’s are enabled).
Here, the memory type specified in the fixed range register overrides the one specified in vari-
able-range register pair.

NOTE
Some mask values can result in discontinuous ranges. In a discontinuous
range, the area not mapped by the mask value is set to the default memory
type. Intel does not encourage the use of discontinuous ranges, because they
could require physical memory to be present throughout the entire 4-GByte
physical memory map. If memory is not provided for the complete memory
map, the behaviour of the processor is undefined.

9.12.3. Example Base and Mask Calculations

The base and mask values entered into the variable-range MTRR pairs are 24-bit values that the
processor extends to 36-bits. For example, to enter a base address of 2 MBytes (200000H) to the
MTRRphysBase3 register, the 12 least-significant bits are truncated and the value 000200H is
entered into the PhysBase field. The same operation must be performed on mask values. For
instance, to map the address range from 200000H to 3FFFFFH (2 MBytes to 4 MBytes), a mask
value of FFFE00000H is required. Here again, the 12 least-significant bits of this mask value
are truncated, so that the value entered in the PhysMask field of the MTRRphysMask3 register
is FFFE00H. This mask is chosen so that when any address in the 200000H to 3FFFFFH range
is ANDed with the mask value it will return the same value as when the base address is ANDed
with the mask value (which is 200000H).
To map the address range from 400000H 7FFFFFH (4 MBytes to 8 MBytes), a base value of
000400H is entered in the PhysBase field and a mask value of FFFC00H is entered in the Phys-
Mask field.
Here is a real-life example of setting up the MTRRs for an entire system. Assume that the system
has the following characteristics:
• 96 MBytes of system memory is mapped as write-back memory (WB) for highest system
performance.
• A custom 4-MByte I/O card is mapped to uncached memory (UC) at a base address of 64
MBytes. This restriction forces the 96 MBytes of system memory to be addressed from 0
to 64 MBytes and from 68 MBytes to 100 MBytes, leaving a 4-MByte hole for the I/O
card.
• An 8-MByte graphics card is mapped to write-combining memory (WC) beginning at
address A0000000H.
• The BIOS area from 15 MBytes to 16 MBytes is mapped to UC memory.

9-25
MEMORY CACHE CONTROL

The following settings for the MTRRs will yield the proper mapping of the physical address
space for this system configuration. The x0_0x notation is used below to add clarity to the large
numbers represented.
MTRRPhysBase0 = 0000_0000_0000_0006h
MTRRPhysMask0 = 0000_000F_FC00_0800h Caches 0-64 MB as WB cache type.
MTRRPhysBase1 = 0000_0000_0400_0006h
MTRRPhysMask1 = 0000_000F_FE00_0800h Caches 64-96 MB as WB cache type.
MTRRPhysBase2 = 0000_0000_0600_0006h
MTRRPhysMask2 = 0000_000F_FFC0_0800h Caches 96-100 MB as WB cache type.
MTRRPhysBase3 = 0000_0000_0400_0000h
MTRRPhysMask3 = 0000_000F_FFC0_0800h Caches 64-68 MB as UC cache type.
MTRRPhysBase4 = 0000_0000_00F0_0000h
MTRRPhysMask4 = 0000_000F_FFF0_0800h Caches 15-16 MB as UC cache type
MTRRPhysBase5 = 0000_0000_A000_0001h
MTRRPhysMask5 = 0000_000F_FF80_0800h Cache A0000000h-A0800000 as WC type.
This MTRR setup uses the ability to overlap any two memory ranges (as long as the ranges are
mapped to WB and UC memory types) to minimize the number of MTRR registers that are
required to configure the memory environment. This setup also fulfills the requirement that two
register pairs are left for operating system usage.

9.12.4. Range Size and Alignment Requirement

The range that is to be mapped to a variable-range MTRR must meet the following “power of
2” size and alignment rules:
1. The minimum range size is 4 KBytes, and the base address of this range must be on at least
a 4-KByte boundary.
2. For ranges greater than 4 KBytes, each range must be of length 2n and its base address
must be aligned on a 2n boundary, where n is a value equal to or greater than 12. The base-
address alignment value cannot be less than its length. For example, an 8-KByte range
cannot be aligned on a 4-KByte boundary. It must be aligned on at least an 8-KByte
boundary.

9.12.4.1. MTRR PRECEDENCES

If the MTRRs are not enabled (by setting the E flag in the MTRRdefType register), then all
memory accesses are of the UC memory type. If the MTRRs are enabled, then the memory type
used for a memory access is determined as follows:
1. If the physical address falls within the first 1 MByte of physical memory and fixed MTRRs
are enabled, the processor uses the memory type stored for the appropriate fixed-range
MTRR.

9-26
 MEMORY CACHE CONTROL

2. Otherwise, the processor attempts to match the physical address with a memory type range
set with a pair of variable-range MTRRs:
a. If one variable memory range matches, the processor uses the memory type stored in
the MTRRphysBasen register for that range.
b. If two or more variable memory ranges match and the memory types are identical,
then that memory type is used.
c. If two or more variable memory ranges match and one of the memory types is UC, the
UC memory type used.
d. If two or more variable memory ranges match and the memory types are WT and WB,
the WT memory type is used.
e. If two or more variable memory ranges match and the memory types are other than UC
and WB, the behaviour of the processor is undefined.
3. If no fixed or variable memory range matches, the processor uses the default memory type.

9.12.5. MTRR Initialization

On a hardware reset, a P6 family processor clears the valid flags in the variable-range MTRRs
and clears the E flag in the MTRRdefType register to disable all MTRRs. All other bits in the
MTRRs are undefined. Prior to initializing the MTRRs, software (normally the system BIOS)
must initialize all fixed-range and variable-range MTRR registers fields to 0. Software can then
initialize the MTRRs according to the types of memory known to it, including memory on
devices that it auto-configures. This initialization is expected to occur prior to booting the oper-
ating system.
See Section 9.12.8., “Multiple-Processor Considerations”, for information on initializing
MTRRs in multiple-processor systems.

9.12.6. Remapping Memory Types

A system designer may re-map memory types to tune performance or because a future processor
may not implement all memory types supported by the P6 family processors. The following
rules support coherent memory-type re-mappings:
1. A memory type should not be mapped into another memory type that has a weaker
memory ordering model. For example, the uncacheable type cannot be mapped into any
other type, and the write-back, write-through, and write-protected types cannot be mapped
into the weakly ordered write-combining type.
2. A memory type that does not delay writes should not be mapped into a memory type that
does delay writes, because applications of such a memory type may rely on its write-
through behavior. Accordingly, the write-back type cannot be mapped into the write-
through type.

9-27
MEMORY CACHE CONTROL

3. A memory type that views write data as not necessarily stored and read back by a
subsequent read, such as the write-protected type, can only be mapped to another type with
the same behaviour (and there are no others for the P6 family processors) or to the
uncacheable type.
In many specific cases, a system designer can have additional information about how a memory
type is used, allowing additional mappings. For example, write-through memory with no asso-
ciated write side effects can be mapped into write-back memory.

9.12.7. MTRR Maintenance Programming Interface

The operating system maintains the MTRRs after booting and sets up or changes the memory
types for memory-mapped devices. The operating system should provide a driver and applica-
tion programming interface (API) to access and set the MTRRs. The function calls
MemTypeGet() and MemTypeSet() define this interface.

9.12.7.1. MEMTYPEGET() FUNCTION

The MemTypeGet() function returns the memory type of the physical memory range specified
by the parameters base and size. The base address is the starting physical address and the size is
the number of bytes for the memory range. The function automatically aligns the base address
and size to 4-KByte boundaries. Pseudocode for the MemTypeGet() function is given in
Example 9-2.

Example 9-2. MemTypeGet() Pseudocode

#define MIXED_TYPES -1 /* 0 < MIXED_TYPES || MIXED_TYPES > 256 */

IF CPU_FEATURES.MTRR /* processor supports MTRRs */

THEN
Align BASE and SIZE to 4-KByte boundary;
IF (BASE + SIZE) wrap 64-GByte address space
THEN return INVALID;
FI;
IF MTRRdefType.E = 0
THEN return UC;
FI;
FirstType ← Get4KMemType (BASE);
/* Obtains memory type for first 4-KByte range */
/* See Get4KMemType (4KByteRange) in Example 9-3 */
FOR each additional 4-KByte range specified in SIZE
NextType ← Get4KMemType (4KByteRange);
IF NextType ≠ FirstType
THEN return MixedTypes;
FI;
ROF;
return FirstType;

9-28
 MEMORY CACHE CONTROL

ELSE return UNSUPPORTED;

FI;
If the processor does not support MTRRs, the function returns UNSUPPORTED. If the MTRRs
are not enabled, then the UC memory type is returned. If more than one memory type corre-
sponds to the specified range, a status of MIXED_TYPES is returned. Otherwise, the memory
type defined for the range (UC, WC, WT, WB, or WP) is returned.
The pseudocode for the Get4KMemType() function in Example 9-3 obtains the memory type
for a single 4-KByte range at a given physical address. The sample code determines whether an
PHY_ADDRESS falls within a fixed range by comparing the address with the known fixed
ranges: 0 to 7FFFFH (64-KByte regions), 80000H to BFFFFH (16-KByte regions), and C0000H
to FFFFFH (4-KByte regions). If an address falls within one of these ranges, the appropriate bits
within one of its MTRRs determine the memory type.

Example 9-3. Get4KMemType() Pseudocode

IF MTRRcap.FIX AND MTRRdefType.FE /* fixed registers enabled */
THEN IF PHY_ADDRESS is within a fixed range
return MTRRfixed.Type;
FI;
FOR each variable-range MTRR in MTRRcap.VCNT
IF MTRRphysMask.V = 0
THEN continue;
FI;
IF (PHY_ADDRESS AND MTRRphysMask.Mask) = (MTRRphysBase.Base
AND MTRRphysMask.Mask)
THEN
return MTRRphysBase.Type;
FI;
ROF;
return MTRRdefType.Type;

9.12.7.2. MEMTYPESET() FUNCTION

The MemTypeSet() function in Example 9-4 sets a MTRR for the physical memory range spec-
ified by the parameters base and size to the type specified by type. The base address and size are
multiples of 4 KBytes and the size is not 0.

Example 9-4. MemTypeSet Pseudocode

IF CPU_FEATURES.MTRR (* processor supports MTRRs *)
THEN
IF BASE and SIZE are not 4-KByte aligned or size is 0
THEN return INVALID;
FI;
IF (BASE + SIZE) wrap 4-GByte address space
THEN return INVALID;

9-29
MEMORY CACHE CONTROL

FI;
IF TYPE is invalid for P6 family processors
THEN return UNSUPPORTED;
FI;
IF TYPE is WC and not supported
THEN return UNSUPPORTED;
FI;
IF MTRRcap.FIX is set AND range can be mapped using a fixed-range MTRR
THEN
pre_mtrr_change();
update affected MTRR;
post_mtrr_change();
FI;

ELSE (* try to map using a variable MTRR pair *)

IF MTRRcap.VCNT = 0
THEN return UNSUPPORTED;
FI;
IF conflicts with current variable ranges
THEN return RANGE_OVERLAP;
FI;
IF no MTRRs available
THEN return VAR_NOT_AVAILABLE;
FI;
IF BASE and SIZE do not meet the power of 2 requirements for variable MTRRs
THEN return INVALID_VAR_REQUEST;
FI;
pre_mtrr_change();
Update affected MTRRs;
post_mtrr_change();
FI;

pre_mtrr_change()
BEGIN
disable interrupts;
Save current value of CR4;
disable and flush caches;
flush TLBs;
disable MTRRs;
IF multiprocessing
THEN maintain consistency through IPIs;
FI;
END
post_mtrr_change()
BEGIN
flush caches and TLBs;
enable MTRRs;

9-30
 MEMORY CACHE CONTROL

enable caches;
restore value of CR4;
enable interrupts;
END
The physical address to variable range mapping algorithm in the MemTypeSet function detects
conflicts with current variable range registers by cycling through them and determining whether
the physical address in question matches any of the current ranges. During this scan, the algo-
rithm can detect whether any current variable ranges overlap and can be concatenated into a
single range.
The pre_mtrr_change() function disables interrupts prior to changing the MTRRs, to avoid
executing code with a partially valid MTRR setup. The algorithm disables caching by setting
the CD flag and clearing the NW flag in control register CR0. The caches are invalidated using
the WBINVD instruction. The algorithm disables the page global flag (PGE) in control register
CR4, if necessary, then flushes all TLB entries by updating control register CR3. Finally, it
disables MTRRs by clearing the E flag in the MTRRdefType register.
After the memory type is updated, the post_mtrr_change() function re-enables the MTRRs and
again invalidates the caches and TLBs. This second invalidation is required because of the
processor’s aggressive prefetch of both instructions and data. The algorithm restores interrupts
and re-enables caching by setting the CD flag.
An operating system can batch multiple MTRR updates so that only a single pair of cache inval-
idations occur.

9.12.8. Multiple-Processor Considerations

In multiple-processor systems, the operating systems must maintain MTRR consistency
between all the processors in the system. The P6 family processors provide no hardware support
to maintain this consistency. In general, all processors must have the same MTRR values.
This requirement implies that when the operating system initializes a multiple-processor system,
it must load the MTRRs of the boot processor while the E flag in register MTRRdefType is 0.
The operating system then directs other processors to load their MTRRs with the same memory
map. After all the processors have loaded their MTRRs, the operating system signals them to
enable their MTRRs. Barrier synchronization is used to prevent further memory accesses until
all processors indicate that the MTRRs are enabled. This synchronization is likely to be a shoot-
down style algorithm, with shared variables and interprocessor interrupts.
Any change to the value of the MTRRs in a multiple-processor system requires the operating
system to repeat the loading and enabling process to maintain consistency, using the following
procedure:
1. Broadcast to all processors to execute the following code sequence.
2. Disable interrupts.
3. Wait for all processors to reach this point.

9-31
MEMORY CACHE CONTROL

4. Enter the no-fill cache mode. (Set the CD flag in control register CR0 to 1 and the NW flag
to 0.)
5. Flush all caches using the WBINVD instruction.
6. Clear the PGE flag in control register CR4 (if set).
7. Flush all TLBs. (Execute a MOV from control register CR3 to another register and then a
MOV from that register back to CR3.)
8. Disable all range registers (by clearing the E flag in register MTRRdefType). If only
variable ranges are being modified, software may clear the valid bits for the affected
register pairs instead.
9. Update the MTRRs.
10. Enable all range registers (by setting the E flag in register MTRRdefType). If only
variable-range registers were modified and their individual valid bits were cleared, then set
the valid bits for the affected ranges instead.
11. Flush all caches and all TLBs a second time. (The TLB flush is required for P6 family
processors. Executing the WBINVD instruction is not needed when using P6 family
processors, but it may be needed in future systems.)
12. Enter the normal cache mode to re-enable caching. (Set the CD and NW flags in control
register CR0 to 0.)
13. Set PGE flag in control register CR4, if previously cleared.
14. Wait for all processors to reach this point.
15. Enable interrupts.

9.12.9. Large Page Size Considerations

The MTRRs provide memory typing for a limited number of regions that have a 4 KByte gran-
ularity (the same granularity as 4-KByte pages). The memory type for a given page is cached in
the processor’s TLBs. When using large pages (2 or 4 MBytes), a single page-table entry covers
multiple 4-KByte granules, each with a single memory type. Because the memory type for a
large page is cached in the TLB, the processor can behave in an undefined manner if a large page
is mapped to a region of memory that MTRRs have mapped with multiple memory types.
Undefined behavior can be avoided by insuring that all MTRR memory-type ranges within a
large page are of the same type. If a large page maps to a region of memory containing different
MTRR-defined memory types, the PCD and PWT flags in the page-table entry should be set for
the most conservative memory type for that range. For example, a large page used for memory
mapped I/O and regular memory is mapped as UC memory. Alternatively, the operating system
can map the region using multiple 4-KByte pages each with its own memory type. The require-
ment that all 4-KByte ranges in a large page are of the same memory type implies that large
pages with different memory types may suffer a performance penalty, since they must be marked
with the lowest common denominator memory type.

9-32
 MEMORY CACHE CONTROL

The P6 family processors provide special support for the physical memory range from 0 to 4
MBytes, which is potentially mapped by both the fixed and variable MTRRs. This support is
invoked when a P6 family processor detects a large page overlapping the first 1 MByte of this
memory range with a memory type that conflicts with the fixed MTRRs. Here, the processor
maps the memory range as multiple 4-KByte pages within the TLB. This operation insures
correct behavior at the cost of performance. To avoid this performance penalty, operating-
system software should reserve the large page option for regions of memory at addresses greater
than or equal to 4 MBytes.

9.13. PAGE ATTRIBUTE TABLE (PAT)

The Page Attribute Table (PAT) is an extension to Intel’s 32-bit processor virtual memory archi-
tecture for certain P6 family processors. Specifically, the PAT is an extension of the page-table
format, which allows the specification of memory types to regions of physical memory based
on linear address mappings. The PAT provides the equivalent functionality of an unlimited
number of Memory Type Range Registers (MTRRs).
Using the PAT in conjunction with the MTRRs of the P6 family of processors extends the
memory type information present in the current Intel Architecture page-table format. It
combines the extendable and programmable qualities of the MTRRs with the flexibility of the
page tables, allowing operating systems or applications to select the best memory type for their
needs. The ability to apply the best memory type in a flexible way enables higher levels of
performance.

NOTE
In multiple processor systems, the operating system(s) must maintain MTRR
consistency between all the processors in the system. The P6 family
processors provide no hardware support for maintaining this consistency. In
general, all processors must have the same MTRR values.

9.13.1. Background
The P6 family of processors support the assignment of specific memory types to physical
addresses. Memory type support is provided through the use of Memory Type Range Registers
(MTRRs). Currently there are two interacting mechanisms that work together to set the effective
memory type: the MTRRs and the page tables. Refer to the Intel Architecture Software Devel-
oper’s Manual, Volume 3: System Programming Guide.
The MTRRs define the memory types for physical address ranges. MTRRs have specific align-
ment and length requirements for the memory regions they describe. Therefore, they are useful
for statically describing memory types for physical ranges, and are typically set up by the system
BIOS. However, they are incapable of describing memory types for the dynamic, linearly
addressed data structures of programs. The MTRRs are an expandable and programmable way
to encode memory types, but are inflexible because they can only apply those memory types to
physical address ranges.

9-33
MEMORY CACHE CONTROL

The page tables allow memory types to be assigned dynamically to linearly addressed pages of
memory. This gives the operating system the maximum amount of flexibility in applying
memory types to any data structure. However, the page tables only offer three of the five basic
P6 processor family memory type encodings: Write-back (WB), Write-through (WT) and
Uncached (UC). The PAT extends the existing page-table format to enable the specification of
additional memory types.

9.13.2. Detecting Support for the PAT Feature

The page attribute table (PAT) feature is detected by an operating system through the use of the
CPUID instruction. Specifically, the operating system executes the CPUID instruction with the
value 1 in the EAX register, and then determines support for the feature by inspecting bit 16 of
the EDX register return value. If the PAT is supported, an operating system is permitted to utilize
the model specific register (MSR) specified for programming the PAT, as well as make use of
the PAT-index bit (PATi), which was formerly a reserved bit in the page tables.
Note that there is not a separate flag or control bit in any of the control registers that enables the
use of this feature. The PAT is always enabled on all processors that support it, and the table
lookup always occurs whenever paging is enabled and for all paging modes (e.g., PSE, PAE).

9.13.3. Technical Description of the PAT

The Page Attribute Table is a Model Specific Register (MSR) at address 277H (for information
about the MSRs, refer to Appendix B, Model-Specific Registers. The model specific register
address for the PAT is defined and will remain at the same address on future Intel processors that
support this feature. Figure 9-7 shows the format of the 64-bit register containing the PAT.

31 27 26 24 23 19 18 16 15 11 10 8 7 3 2 0
Rsvd PA3 Rsvd PA 2 Rsvd PA 1 Rsvd PA 0

63 59 58 56 55 51 50 48 47 43 42 40 39 35 34 32
Rsvd PA7 Rsvd PA6 Rsvd PA5 Rsvd PA4

NOTES:
1. PA0-7 = Specifies the eight page attribute locations contained within the PAT
2. Rsvd = Most significant bits for each Page Attribute are reserved for future expansion

Figure 9-7. Page Attribute Table Model Specific Register

Each of the eight page attribute fields can contain any of the available memory type encodings,
or indexes, as specified in Table 9-1.

9-34
 MEMORY CACHE CONTROL

9.13.4. Accessing the PAT

Access to the memory types that have been programmed into the PAT register fields is accom-
plished with a 3-bit index consisting of the PATi, PCD, and PWT bits. Table 9-8 shows how the
PAT register fields are indexed. The last column of the table shows which memory type the
processor assigns to each PAT field at processor reset and initialization. These initial values
provide complete backward compatibility with previous Intel processors and existing software
that use the previously existing page-table memory types and MTRRs.

Table 9-8. PAT Indexing and Values After Reset

1
PATi PCD PWT PAT Entry Memory Type at Reset
0 0 0 0 WB
0 0 1 1 WT
0 1 0 2 UC-2
0 1 1 3 UC3
1 0 0 4 WB
1 0 1 5 WT
1 1 0 6 UC-2
1 1 1 7 UC3

NOTES:
1. PATi bit is defined as bit 7 for 4 KB PTEs, bit 12 for PDEs mapping 2 MB/4 MB pages.
2. UC- is the page encoding PCD, PWT = 10 on P6 family processors that do not support this feature. UC-
in the page table is overridden by WC in the MTRRs.
3. UC is the page encoding PCD, PWT = 11 on P6 family processors that do not support this feature. UC in
the page-table overrides WC in the MTRRs.

In P6 family processors that do not support the PAT, the PCD and PWT bits are used to deter-
mine the page-table memory types of a given physical page. The PAT feature redefines these two
bits and combines them with a newly defined PAT-index bit (PATi) in the page-directory and
page-table entries. These three bits create an index into the 8-entry Page Attribute Table. The
memory type from the PAT is used in place of PCD and PWT for computing the effective
memory type.
The bit used for PATi differs depending upon the level of the paging hierarchy. PATi is bit 7 for
page-table entries, and bit 12 for page-directory entries that map to large pages. Reserved bit
faults are disabled for nonzero values for PATi, but remain present for all other reserved bits.
This is true for 4 KB/2 MB pages when PAE is enabled. The PAT index scheme for each level
of the paging hierarchy is shown in Figure 9-8.

9-35
MEMORY CACHE CONTROL

31 4 3
PCD PWT
Page-Directory Base Register (CR3)
31 4 3
PCD and PWT provide 2 bit
PCD PWT index into the PAT, allowing use
of first 4 entries
Page-Directory Pointer Table Entry
31 4 3
PCD PWT
4 KB Page-Directory Entry

31 13 12 4 3
PATi PCD PWT
2 MB/4 MB Page-Directory Entry PATi, PCD, and PWT provide 3 bit
index into the PAT, allowing use of
31 8 7 4 3 all 8 entries
PATi PCD PWT
4 KB Page-Table Entry

Figure 9-8. Page Attribute Table Index Scheme for Paging Hierarchy

NOTE:
This figure only shows the format of the lower 32 bits of the PDE, PDEPTR, and PTEs when in PAE mode
Refer to Figure 3-21 from Chapter 3, Protected-Mode Memory Management of the Intel Architecture Soft-
ware Developer’s Manual, Volume 3: System Programming Guide. Additionally, the formats shown in this
figure are not meant to accurately represent the entire structure, but only the labeled bits.

Figure 9-8 shows that the PAT bit is not defined in CR3, the Page-Directory-Pointer Tables when
PAE is enabled, or the Page Directory when it doesn’t describe a large page. In these cases, only
PCD and PWT are used to index into the PAT, limiting the operating system to using only the
first 4 entries of PAT for describing the memory attributes of the paging hierarchy. Note that all
8 PAT entries are available for describing a 4 KB/2 MB/4 MB page.
The memory type as now defined by PAT interacts with the MTRR memory type to determine
the effective memory type as outlined in Table 9-9. Compare this to Table 9-5.

9-36
 MEMORY CACHE CONTROL

Table 9-9. Effective Memory Type Depending on MTRRs and PAT

PAT Memory Type MTRR Memory Type Effective Memory Type
UC- WB, WT UC_PAGE
WC WC
UC UC_MTRR
WP Undefined
UC WB, WT, WP, WC UC_PAGE
UC UC_MTRR
WC X WC
WT WB, WT WT
UC UC_MTRR
WC Undefined
WP Undefined
WP WB, WP WP
UC UC_MTRR
WC, WT Undefined
WB WB WB
UC UC_MTRR
WC WC
WT WT
WP WP

NOTES:
• This table assumes that the CD and NW flags in register CR0 are set to 0. If CR0.CD = 1, then the effec-
tive memory type returned is UC, regardless of what is indicated in the table. However, this does not force
strict ordering. To ensure strict ordering, the MTRRs also must be disabled.
• The effective memory types in the gray areas are implementation dependent and may be different
between implementations of Intel Architecture processors.
• UC_MTRR indicates that the UC attribute came from the MTRRs and the processor(s) are not required to
snoop their caches since the data could never have been cached. This is preferred for performance rea-
sons.
• UC_PAGE indicates that the UC attribute came from the page tables and processors are required to
check their caches because the data may be cached due to page aliasing, which is not recommended.
• UC- is the page encoding PCD, PWT = 10 on P6 family processors that do not support this feature. UC- in the
PTE/PDE is overridden by WC in the MTRRs.
• UC is the page encoding PCD, PWT = 11 on P6 family processors that do not support this feature. UC in the
PTE/PDE overrides WC in the MTRRs.

Whenever the MTRRs are disabled, via bit 11 (E) in the MTRRDefType register, the effective
memory type is UC for all memory ranges.
An operating system can program the PAT and select the 8 most useful attribute combinations.
The PAT allows an operating system to offer performance-enhancing memory types to applica-
tions.

9-37
MEMORY CACHE CONTROL

The page attribute for addresses containing a page directory or page table supports only the first
four entries in the PAT, since a PAT-index bit is not defined for these mappings. The page
attribute is determined by using the two-bit value specified by PCD and PWT in CR3 (for page
directory) or the page-directory entry (for page tables). The same applies to Page-Directory-
Pointer Tables when PAE is enabled.

9.13.5. Programming the PAT

The Page Attribute Table is read/write accessible to software operating at ring 0 through the use
of the rdmsr and wrmsr instructions. Accesses are directed to the PAT through use of model
specific register address 277H. Refer to Figure 9-7 for the format of the 64-bit register
containing the PAT.
The PAT implementation on processors that support the feature defines only the 3 least signifi-
cant bits for page attributes. These bits are used to specify the memory type with the same
encoding as used for the P6 family MTRRs as shown in Table 9-6. Processors that support the
PAT feature modify those encodings slightly, in that encoding 0 is UC and encoding 7 is UC-,
as indicated in the Table 9-10. Encoding 7 remains undefined for the fixed and variable MTRRs,
and any attempt to write an undefined memory type encoding continues to generate a GP fault.
Attempting to write an undefined memory type encoding into the PAT generates a GP fault.

Table 9-10. PAT Memory Types and Their Properties

Allows
Writeback Speculative Memory Ordering
Mnemonic Encoding Cacheable Cacheable Reads Model
Uncacheable 0 No No No Strong
(UC) Ordering
Write Combining 1 No No Yes Weak
(WC) Ordering
Write-through 4 Yes No Yes Speculative
(WT) Processor
Ordering
Write-protect 5 Yes for No Yes Speculative
(WP) reads, no for Processor
writes Ordering
Write-back (WB) 6 Yes Yes Yes Speculative
Processor
Ordering
Uncached (UC-) 7 No No No Strong Ordered,
but can be
overridden by WC
in the MTRRs
Reserved 2, 3, 87-255

The operating system is responsible for ensuring that changes to a PAT entry occur in a manner
that maintains the consistency of the processor caches and translation lookaside buffers (TLB).
This is accomplished by following the procedure as specified in the Intel Architecture Software

9-38
 MEMORY CACHE CONTROL

Developer’s Manual, Volume 3: System Programming Guide, for changing the value of an
MTRR. It involves a specific sequence of operations that includes flushing the processor(s)
caches and TLBs. An operating system must ensure that the PAT of all processors in a multipro-
cessing system have the same values.
The PAT allows any memory type to be specified in the page tables, and therefore it is possible
to have a single physical page mapped by two different linear addresses with differing memory
types. This practice is strongly discouraged by Intel and should be avoided as it may lead to
undefined results. In particular, a WC page must never be aliased to a cacheable page because
WC writes may not check the processor caches. When remapping a page that was previously
mapped as a cacheable memory type to a WC page, an operating system can avoid this type of
aliasing by:
• Removing the previous mapping to a cacheable memory type in the page tables; that is,
make them not present.
• Flushing the TLBs of processors that may have used the mapping, even speculatively.
• Creating a new mapping to the same physical address with a new memory type, for
instance, WC.
• Flushing the caches on all processors that may have used the mapping previously.
Operating systems that use a Page Directory as a Page Table and enable Page Size Extensions
must carefully scrutinize the use of the PATi index bit for the 4 KB Page-Table Entries. The PATi
index bit for a PTE (bit 7) corresponds to the page size bit in a PDE. Therefore, the operating
system can only utilize PAT entries PA0-3 when setting the caching type for a page table that is
also used as a page directory. If the operating system attempts to use PAT entries PA4-7 when
using this memory as a page table, it effectively sets the PS bit for the access to this memory as
a page directory.

9-39
MEMORY CACHE CONTROL

9-40
 10
MMX™ Technology
System Programming
 MMX™ TECHNOLOGY SYSTEM PROGRAMMING

CHAPTER 10
MMX™ TECHNOLOGY SYSTEM PROGRAMMING

This chapter describes those features of the MMX™ technology that must be considered when
designing or enhancing an operating system to support MMX™ technology. It covers MMX™
instruction set emulation, the MMX™ state, aliasing of MMX™ registers, saving MMX™ state,
task and context switching considerations, exception handling, and debugging.

10.1. EMULATION OF THE MMX™ INSTRUCTION SET

The Intel Architecture does not support emulation of the MMX™ technology, as it does for
floating-point instructions. The EM flag in control register CR0 (provided to invoke emulation
of floating-point instructions) cannot be used for MMX™ technology emulation. If an MMX™
instruction is executed when the EM flag is set, an invalid opcode (UD#) exception is generated.

10.2. THE MMX™ STATE AND MMX™ REGISTER ALIASING

The MMX™ state consists of eight 64-bit registers (MM0 through MM7). These registers are
aliased to the 64-bit mantissas (bits 0 through 63) of floating-point registers R0 through R7 (see
Figure 10-2). Note that the MMX™ registers are mapped to the physical locations of the
floating-point registers (R0 through R7), not to the relative locations of the registers in the
floating-point register stack (ST0 through ST7). As a result, the MMX™ register mapping is
fixed and is not affected by value in the Top Of Stack (TOS) field in the floating-point status
word (bits 11 through 13).
When a value is written into an MMX™ register using an MMX™ instruction, the value also
appears in the corresponding floating-point register in bits 0 through 63. Likewise, when a
floating-point value written into a floating-point register by a floating-point instruction, the
mantissa of that value also appears in a the corresponding MMX™ register.
The execution of MMX™ instructions have several side effects on the FPU state contained in
the floating-point registers, the FPU tag word, and the FPU the status word. These side effects
are as follows:
• When an MMX™ instruction writes a value into an MMX™ register, at the same time, bits
64 through 79 of the corresponding floating-point register (the exponent field and the sign
bit) are set to all 1s.
• When an MMX™ instruction (other than the EMMS instruction) is executed, each of the
tag fields in the FPU tag word is set to 00B (valid). (See also Section 10.2.1., “Effect of
MMX™ and Floating-Point Instructions on the FPU Tag Word”.)
• When the EMMS instruction is executed, each tag field in the FPU tag word is set to 11B
(empty).

10-1
 MMX™ TECHNOLOGY SYSTEM PROGRAMMING

• Each time an MMX™ instruction is executed, the TOS value is set to 000B.

FPU Tag Floating-Point Registers

Register 79 64 63 0
00 Mantissa R7
00 R6
00 R5
00 R4
00 R3
00 R2
00 R1
00 R0

FPU Status Register

13 11
000
63 MMXTM Registers 0
TOS
MM7
MM6
MM5
MM4
MM3
MM2
MM1
TOS = 0
MM0

Figure 10-1. Mapping of MMX™ Registers to Floating-Point Registers

Execution of MMX™ instructions does not affect the other bits in the FPU status word (bits 0
through 10 and bits 14 and 15) or the contents of the other FPU registers that comprise the FPU
state (the FPU control word, instruction pointer, data pointer, or opcode registers).
Table 10-1 summarizes the effects of the MMX™ instructions on the FPU state.

10-2
 MMX™ TECHNOLOGY SYSTEM PROGRAMMING

Table 10-1. Effects of MMX™ Instructions on FPU State

MMX™ TOS Field of Exponent Bits
Instruction FPU Status Other FPU and Sign Bit of
Type FPU Tag Word Word Registers Rn Mantissa of Rn
Read from All tags set to 000B Unchanged Unchanged Unchanged
MMn register 00B (Valid)
Write to MMn All tags set to 000B Unchanged Set to all 1s Overwritten with
register 00B (Valid) MMX™ data
EMMS All fields set to 000B Unchanged Unchanged Unchanged
11B (Empty)

NOTE:
MMn refers to one MMX™ register; Rn refers to corresponding floating-point register.

10.2.1. Effect of MMX™ and Floating-Point Instructions on the

FPU Tag Word
Table 10-2 summarizes the effect of MMX™ and floating-point instructions on the tags in the
FPU tag word and the corresponding tags in an image of the tag word stored in memory.

Table 10-2. Effect of the MMX™ and Floating-Point Instructions on the

FPU Tag Word
Instruction Image of FPU Tag Word
Type Instruction FPU Tag Word Stored in Memory
MMX™ All (except All tags are set to 00B (valid). Not affected.
Instruction EMMS)
MMX™ EMMS All tags are set to 11B (empty). Not affected.
Instruction
Floating-Point All (except Tag for modified floating-point Not affected.
Instruction FXSAVE/FSAVE, register is set to 00B or 11B.
FSTENV,
FXRSTOR/FRST
OR, FLDENV)
Floating-Point FXSAVE/FSAVE, Tags and register values are read Tags are set according to the
Instruction FSTENV and interpreted; then all tags are set actual values in the floating-
to 11B. point registers; that is, empty
registers are marked 11B
and valid registers are
marked 00B (nonzero), 01B
(zero), or 10B (special).
Floating-Point FXRSTOR/FRST All tags marked 11B in memory are Tags are read and
Instruction OR, FLDENV set to 11B; interpreted, but not modified.
all other tags are set according to the
value in the corresponding floating-
point register: 00B (nonzero), 01B
(zero), or 10B (special).

10-3
MMX™ TECHNOLOGY SYSTEM PROGRAMMING

The values in the fields of the FPU tag word do not affect the contents of the MMX™ registers
or the execution of MMX™ instructions. However, the MMX™ instructions do modify the
contents of the FPU tag word, as is described in Section 10.2., “The MMX™ State and MMX™
Register Aliasing”. These modifications may affect the operation of the FPU when executing
floating-point instructions, if the FPU state is not initialized or restored prior to beginning
floating-point instruction execution.
Note that the FXSAVE/FSAVE and FSTENV instructions (which save FPU state information)
read the FPU tag register and contents of each of the floating-point registers, determine the
actual tag values for each register (empty, nonzero, zero, or special), and store the updated tag
word in memory. After executing these instructions, all the tags in the FPU tag word are set to
empty (11B). Likewise, the EMMS instruction clears MMX™ state from the MMX™/floating-
point registers by setting all the tags in the FPU tag word to 11B.

10.3. SAVING AND RESTORING THE MMX™ STATE AND

REGISTERS
The recommended method of saving and restoring the MMX™ technology state is as follows:
• Execute an FXSAVE/FSAVE/FNSAVE instruction to write the entire state of the
MMX™/FPU, the SIMD floating-point registers and the SIMD floating-point MXCSR to
memory.
• Execute an FXRSTOR/FRSTOR instruction to read the entire saved state of the
MMX™/FPU, the SIMD floating-point registers and the SIMD floating-point MXCSR
from memory into the FPU registers, the aliased MMX™ registers, the SIMD floating-
point registers and the SIMD floating-point MXCSR.
This save and restore method is required for operating systems (refer to Section 10.4.,
“Designing Operating System Task and Context Switching Facilities”).
Applications can in some cases save and restore only the MMX™ registers, in the following
way:
• Execute eight MOVQ instructions to write the contents of the MMX™ registers MM0
through MM7 to memory. An EMMS instruction may then (optionally) be executed to
clear the MMX™ state in the FPU.
• Execute eight MOVQ instructions to read the saved contents of the MMX™ registers from
memory into the MM0 through MM7 registers.

NOTE
Intel does not support scanning the FPU tag word and then only saving valid
entries.

10-4
 MMX™ TECHNOLOGY SYSTEM PROGRAMMING

10.4. DESIGNING OPERATING SYSTEM TASK AND CONTEXT

SWITCHING FACILITIES
When switching from one task or context to another, it is often necessary to save the MMX™
state (just as it is often necessary to save the state of the FPU). As a general rule, if the existing
task switching code for an operating system includes facilities for saving the state of the FPU,
these facilities can also be relied upon to save the MMX™ state, without rewriting the task
switch code. This reliance is possible because the MMX™ state is aliased to the FPU state (refer
to Section 10.2., “The MMX™ State and MMX™ Register Aliasing”).
When designing new MMX™ (and/or FPU) state saving facilities for an operating system,
several approaches are available:
• The operating system can require that applications (which will be run as tasks) take
responsibility for saving the state of the MMX™/FPU prior to a task suspension during a
task switch and for restoring the MMX™/FPU state when the task is resumed. The
application can use either of the state saving and restoring techniques given in Section
10.3., “Saving and Restoring the MMX™ State and Registers”. This approach to saving
MMX™/FPU state is appropriate for cooperative multitasking operating systems, where
the application has control over (or is able to determine) when a task switch is about to
occur and can save state prior to the task switch.
• The operating system can take the responsibility for automatically saving the
MMX™/FPU state as part of the task switch process (using an FXSAVE/FSAVE
instruction) and automatically restoring the MMX™/FPU state when a suspended task is
resumed (using an FXRSTOR/FRSTOR instruction). Here, the MMX™/FPU state must be
saved as part of the task state. This approach is appropriate for preemptive multitasking
operating systems, where the application cannot know when it is going to be preempted
and cannot prepare in advance for task switching. The operating system is responsible for
saving and restoring the task and MMX™/FPU state when necessary.
• The operating system can take the responsibility for saving the MMX™/FPU state as part
of the task switch process, but delay the saving of the MMX™/FPU state until an MMX™
or floating-point instruction is actually executed by the new task. Using this approach, the
MMX™/FPU state is saved only if an MMX™ or floating-point instruction needs to be
executed in the new task. (Refer to Section 10.4.1., “Using the TS Flag in Control Register
CR0 to Control MMX™/FPU State Saving”, for more information on this MMX™/FPU
state saving technique.)

10.4.1. Using the TS Flag in Control Register CR0 to Control

MMX™/FPU State Saving
Saving the MMX™/FPU state using the FXSAVE/FSAVE instruction is a relatively high-over-
head operation. If a task being switched to will not access the FPU (by executing an MMX™ or
a floating-point instruction), this overhead can be avoided by not automatically saving the
MMX™/FPU state on a task switch.

10-5
MMX™ TECHNOLOGY SYSTEM PROGRAMMING

The TS flag in control register CR0 is provided to allow the operating system to delay saving
the MMX™/FPU state until the FPU is actually accessed in the new task. When this flag is set,
the processor monitors the instruction stream for MMX™ or floating-point instructions. When
the processor detects an MMX™ or floating-point instruction, it raises a device-not-available
exception (#NM) prior to executing the instruction. The device-not-available exception handler
can then be used to save the MMX™/FPU state for the previous task (using an FXSAVE/FSAVE
instruction) and load the MMX™/FPU state for the current task (using an FXRSTOR/FRSTOR
instruction). If the task never encounters an MMX™ or floating-point instruction, the device-
not-available exception will not be raised and the MMX™/FPU state will not be saved unnec-
essarily.
The TS flag can be set either explicitly (by executing a MOV instruction to control register CR0)
or implicitly (using the processors native task switching mechanism). When the native task
switching mechanism is used, the processor automatically sets the TS flag on a task switch.
After the device-not-available handler has saved the MMX™/FPU state, it should execute the
CLTS instruction to clear the TS flag in CR0.
Figure 10-2 gives an example of an operating system that implements MMX™/FPU state saving
using the TS flag. In this example, task A is the currently running task and task B is the task
being switched to.

Task A Task B
MMX™/FPU
Application State Owner

Operating System
CR0.TS=1 and
Task A Floating-point or Task B
MMX™/FPU MMX™ Instruction MMX™/FPU
State Save Area is encountered. State Save Area

Operating System
Saves Task A Task Switching Code Loads Task B
MMX™/FPU State MMX™/FPU State
Device-Not-Available
Exception Handler

Figure 10-2. Example of MMX™/FPU State Saving During an Operating

System-Controlled Task Switch

The operating system maintains an MMX™/FPU save area for each task and defines a variable
(MMX™/FPUStateOwner) that indicates which task “owns” the MMX™/FPU state. In this
example, task A is the current MMX™/FPU state owner.
On a task switch, the operating system task switching code must execute the following pseudo-
code to set the TS flag according to who is the current MMX™/FPU state owner. If the new task

10-6
 MMX™ TECHNOLOGY SYSTEM PROGRAMMING

(task B in this example) is not the current MMX™/FPU state owner, the TS flag is set to 1; other-
wise, it is set to 0.
IF Task_Being_Switched_To ≠ MMX/FPUStateOwner
THEN
CR0.TS ← 1;
ELSE
CR0.TS ← 0;
FI;
If a new task attempts to use an MMX™ or floating-point instruction while the TS flag is set to
1, a device-not-available exception (#NM) is generated and the device-not-available exception
handler executes the following pseudo-code.
CR0.TS ← 0;
FSAVE “To MMX/FPU State Save Area for Current MMX/FPU State Owner”;
FRSTOR “MMX/FPU State From Current Task’s MMX/FPU State Save Area”;
MMX/FPUStateOwner ← Current_Task;
This handler code performs the following tasks:
• Clears the TS flag.
• Saves the MMX™/FPU state in the state save area for the current MMX™/FPU state
owner.
• Restores the MMX™/FPU state from the new task’s MMX™/FPU state save area.
• Updates the current MMX™/FPU state owner to be the current task.

10.5. EXCEPTIONS THAT CAN OCCUR WHEN EXECUTING

MMX™ INSTRUCTIONS
MMX™ instructions do not generate floating-point exceptions, nor do they affect the
processor’s status flags in the EFLAGS register or the FPU status word. The following excep-
tions can be generated during the execution of an MMX™ instruction:
• Exceptions during memory accesses:
— Stack-segment fault (#SS).
— General protection (#GP).
— Page fault (#PF).
— Alignment check (#AC), if alignment checking is enabled.
• System exceptions:
— Invalid Opcode (#UD), if the EM flag in control register CR0 is set when an MMX™
instruction is executed. (Refer to Section 10.1., “Emulation of the MMX™ Instruction
Set”).

10-7
MMX™ TECHNOLOGY SYSTEM PROGRAMMING

— Device not available (#NM), if an MMX™ instruction is executed when the TS flag in
control register CR0 is set. (See Refer to Section 10.4.1., “Using the TS Flag in
Control Register CR0 to Control MMX™/FPU State Saving”.)
• Floating-point error (#MF). (See Refer to Section 10.5.1., “Effect of MMX™ Instructions
on Pending Floating-Point Exceptions”.)
• Other exceptions can occur indirectly due to the faulty execution of the exception handlers
for the above exceptions. For example, if a stack-segment fault (#SS) occurs due to
MMX™ instructions, the interrupt gate for the stack-segment fault can direct the processor
to invalid TSS, causing an invalid TSS exception (#TS) to be generated.

10.5.1. Effect of MMX™ Instructions on Pending Floating-Point

Exceptions
If a floating-point exception is pending and the processor encounters an MMX™ instruction, the
processor generates a floating-point error (#MF) prior to executing the MMX™ instruction, to
allow the exception to be handled by the floating-point error exception handler. While the
handler is executing, the FPU state is maintained and is visible to the handler. Upon returning
from the exception handler, the MMX™ instruction is executed, which will alter the FPU state,
as described in Section 10.2., “The MMX™ State and MMX™ Register Aliasing”.

10.6. DEBUGGING
The debug facilities of the Intel Architecture operate in the same manner when executing
MMX™ instructions as when executing other Intel Architecture instructions. These facilities
enable debuggers to debug MMX™ technology code.
To correctly interpret the contents of the MMX™ or FPU registers from the FXSAVE/FSAVE
image in memory, a debugger needs to take account of the relationship between the floating-
point register’s logical locations relative to TOS and the MMX™ register’s physical locations.
In the floating-point context, STn refers to a floating-point register at location n relative to the
TOS. However, the tags in the FPU tag word are associated with the physical locations of the
floating-point registers (R0 through R7). The MMX™ registers always refer to the physical
locations of the registers (with MM0 through MM7 being mapped to R0 through R7).
In Figure 10-2, the inner circle refers to the physical location of the floating-point and MMX™
registers. The outer circle refers to the floating-point registers’s relative location to the current
TOS.

10-8
 MMX™ TECHNOLOGY SYSTEM PROGRAMMING

FP “push” FP “pop” FP “push

ST0 ST6

ST7 ST1 ST7

MM0 MM0
(R0) (R0)
MM7 MM1 MM7 MM1

TOS MM2 MM2

MM6 (R2) ST2 MM6 TOS ST0
(R2)

MM5 MM3 MM5 MM3 FP “pop”

MM4 MM4 ST1

Case A: TOS=0 Case B: TOS=2

Outer circle = FP register’s logical location relative to TOS

Inner circle = FPU tags = MMX™ register’s location = FP registers’s physical location

Figure 10-3. Mapping of MMX™ Registers to Floating-Point (FP) Registers

When the TOS equals 0 (case A in Figure 10-2), ST0 points to the physical location R0 on the
floating-point stack. MM0 maps to ST0, MM1 maps to ST1, and so on.
When the TOS equals 2 (case B in Figure 10-2), ST0 points to the physical location R2. MM0
maps to ST6, MM1 maps to ST7, MM2 maps to ST0, and so on.

10-9
MMX™ TECHNOLOGY SYSTEM PROGRAMMING

10-10
 11
Streaming SIMD
Extensions System
Programming
 CHAPTER 11
STREAMING SIMD EXTENSIONS SYSTEM
PROGRAMMING

This chapter describes those features of the Streaming SIMD Extensions that must be considered
when designing or enhancing an operating system to support the Pentium® III processor. It
covers extensions emulation, the new SIMD floating-point architectural state, similarities to
MMX™ technology, task and context switching considerations, exception handling, and debug-
ging.

11.1. EMULATION OF THE STREAMING SIMD EXTENSIONS

The Intel Architecture does not support emulation of the Streaming SIMD Extensions, as it does
for floating-point instructions. The EM flag in control register CR0 (provided to invoke emula-
tion of floating-point instructions) cannot be used for Streaming SIMD Extensions emulation.
If a Streaming SIMD Extensions instruction is executed when the EM flag is set (CR0.EM), an
invalid opcode (UD#/INT6) exception is generated instead of a device not available exception
(NM#/INT7).

11.2. MMX™ STATE AND STREAMING SIMD EXTENSIONS

The SIMD-integer instructions of the Streaming SIMD Extensions use the same registers as the
MMX™ technology instructions. In addition they have been implemented so the same rules for
MMX™ technology instructions apply to the Streaming SIMD Extensions. Hence everything
referenced in chapter 10 relating to MMX™ technology and system programming is applicable
to the SIMD-integer instructions in the Streaming SIMD Extensions.

11.3. NEW PENTIUM® III PROCESSOR REGISTERS

The Pentium® III Processor introduced a set of 128-bit general-purpose registers. These registers
are directly addressable and can be used to hold data only. In addition, the Pentium® III
Processor also introduced a new control/status register (MXCSR) that is used to flag exceptions
resulting from computations involving the SIMD floating-point registers, mask/unmask excep-
tions, and control the rounding and flush-to-zero modes. These registers are described more
completely in the following sections.

11-1
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

11.3.1. SIMD Floating-point Registers

Streaming SIMD Extensions provides eight 128-bit general-purpose registers, each of which
can be directly addressed. These registers are new state, and require support from the operating
system to use them.
The SIMD floating-point registers can hold packed 128-bit data. The SIMD floating-point
instructions access the SIMD floating-point registers directly using the register names XMM0
to XMM7 (Table 11-1). These registers can be used to perform calculations on data. They cannot
be used to address memory; addressing is accomplished by using the integer registers and
existing IA addressing modes.
The contents of SIMD floating-point registers are cleared upon reset.
There is a new control/status register MXCSR which is used to mask/unmask numerical excep-
tion handling, to set rounding modes, to set the flush-to-zero mode, and to view status flags.

Table 11-1. SIMD Floating-point Register Set

128 97 96 64 63 32 31 0
XMM0
XMM1
XMM2
XMM3
XMM4
XMM5
XMM6
XMM7

11.3.2. SIMD Floating-point Control/Status Registers

The control/status register is used to enable masked/unmasked numerical exception handling, to
set rounding modes, to set the flush-to-zero mode, and to view status flags. The contents of this
register can be loaded with the LDMXCSR and FXRSTOR instructions and stored in memory
with the STMXCSR and FXSAVE instructions. Figure 11-1 shows the format and encoding of
the fields in the MXCSR.

11-2
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

31-16 15 10 5 0
Reserved F R R P U O Z D I R P U O Z D I
Z C C M M M M M M s E E E E E E
v
d

Figure 11-1. Streaming SIMD Extensions Control/Status Register Format

Bits 5-0 indicate whether a Streaming SIMD Extensions numerical exception has been detected.
They are “sticky” flags, and can be cleared by using the LDMXCSR instruction to write zeroes
to these fields. If a LDMXCSR instruction clears a mask bit and sets the corresponding excep-
tion flag bit, an exception will not be generated because of this change. This type of exception
will occur only upon the next Streaming SIMD Extensions instruction to cause it. Streaming
SIMD Extensions use only one exception flag for each exception. There is no provision for indi-
vidual exception reporting within a packed data type. In situations where multiple identical
exceptions occur within the same instruction, the associated exception flag is updated and indi-
cates that at least one of these conditions happened. These flags are cleared upon reset.
Bits 12-7 configure numerical exception masking; an exception type is masked if the corre-
sponding bit is set and it is unmasked if the bit is clear. These bits are set upon reset, meaning
that all numerical exceptions are masked.
Bits 14-13 encode the rounding control, which provides for the common round to nearest mode,
as well as directed rounding and true chop (refer to Section 11.3.2.1., “Rounding Control
Field”). The rounding control is set to round to nearest upon reset.
Bit 15 (FZ) is used to turn on the flush-to-zero mode (refer to Section 11.3.2.2., “Flush-to-
Zero”). This bit is cleared upon reset, disabling the flush-to-zero mode.
The other bits of MXCSR (bits 31-16 and bit 6) are defined as reserved and cleared; attempting
to write a non-zero value to these bits, using either the FXRSTOR or LDMXCSR instructions,
will result in a general protection exception.

11.3.2.1. ROUNDING CONTROL FIELD

The rounding control (RC) field of MXCSR (bits 13 and 14) controls how the results of floating-
point instructions are rounded. Four rounding modes are supported: round to nearest, round up,
round down, and round toward zero (see Table 11-2). Round to nearest is the default rounding
mode and is suitable for most applications. It provides the most accurate and statistically unbi-
ased estimate of the true result.

11-3
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

Table 11-2. Rounding Control Field (RC)

Rounding RC Field
Mode Setting Description
Round to 00B Rounded result is the closest to the infinitely precise result. If two values
nearest (even) are equally close, the result is the even value (that is, the one with the
least-significant bit of zero).
Round down 01B Rounded result is closest to, but no greater than the infinitely precise
(toward −∞) result.
Round up 10B Rounded result is closest to, but no less than the infinitely precise result.
(toward +∞)
Round toward 11B Rounded result is closest to, but no greater in absolute value than the
zero (truncate) infinitely precise result.

The round up and round down modes are termed directed rounding and can be used to imple-
ment interval arithmetic. Interval arithmetic is used to determine upper and lower bounds for the
true result of a multistep computation, when the intermediate results of the computation are
subject to rounding.
The round toward zero mode (sometimes called the “chop” mode) is commonly used when
performing integer arithmetic with the processor.
Whenever possible, the processor produces an infinitely precise result. However, it is often the
case that the infinitely precise result of an arithmetic or store operation cannot be encoded
exactly in the format of the destination operand. For example, the following value (a) has a 24-
bit fraction. The least-significant bit of this fraction (the underlined bit) cannot be encoded
exactly in the single-real format (which has only a 23-bit fraction):
(a) 1.0001 0000 1000 0011 1001 0111E2 101
To round this result (a), the processor first selects two representable fractions b and c that most
closely bracket a in value (b < a < c).
(b) 1.0001 0000 1000 0011 1001 011E2 101
(c) 1.0001 0000 1000 0011 1001 100E2 101
The processor then sets the result to b or to c according to the rounding mode selected in the RC
field. Rounding introduces an error in a result that is less than one unit in the last place to which
the result is rounded.
The rounded result is called the inexact result. When the processor produces an inexact result,
the floating-point precision (inexact) flag (PE) is set in MXCSR.
When the infinitely precise result is between the largest positive finite value allowed in a partic-
ular format and +∞, the processor rounds the result as shown in Table 11-3.

11-4
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

Table 11-3. Rounding of Positive Numbers Greater than the

Maximum Positive Finite Value
Rounding Mode Result
Rounding to nearest (even) +∞
Rounding down (toward −∞) Maximum, positive finite value
Rounding up (toward +∞) +∞
Rounding toward zero (Truncate) Maximum, positive finite value

When the infinitely precise result is between the largest negative finite value allowed in a partic-
ular format and −∞, the processor rounds the result as shown in Table 11-4.

Table 11-4. Rounding of Negative Numbers Smaller than the

Maximum Negative Finite Value
Rounding Mode Result
Rounding to nearest (even) -∞
Rounding toward zero (Truncate) Maximum, negative finite value
Rounding up (toward +∞) Maximum, negative finite value
Rounding down (toward −∞) -∞

The rounding modes have no effect on comparison operations, operations that produce exact
results, or operations that produce NaN results.

11.3.2.2. FLUSH-TO-ZERO
Turning on the Flush-To-Zero mode has the following effects when tiny results occur (i.e. when
the infinitely precise result rounded to the destination precision with an unbounded exponent, is
smaller in absolute value than the smallest normal number that can be represented; this is similar
to the underflow condition when underflow traps are unmasked):
• Zero results are returned with the sign of the true result
• Precision and underflow exception flags are set
The IEEE mandated masked response to underflow is to deliver the denormalized result (i.e.,
gradual underflow); consequently, the flush-to-zero mode is not compatible with IEEE Standard
754. It is provided primarily for performance reasons. At the cost of a slight precision loss, faster
execution can be achieved for applications where underflow is common. Underflow for flush-
to-zero is defined to occur when the exponent for a computed result, prior to denormalization
scaling, falls in the denormal range; this is regardless of whether a loss of accuracy has occurred.
Unmasking the underflow exception takes precedence over flush-to-zero mode; this means that
an exception handler will be invoked for a Streaming SIMD Extensions instruction that gener-
ates an underflow condition while this exception is unmasked, regardless of whether flush-to-
zero is enabled.

11-5
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

11.4. ENABLING STREAMING SIMD EXTENSIONS SUPPORT

This section describes the interface of the Intel Architecture Streaming SIMD Extensions with
the operating system.

11.4.1. Enabling Streaming SIMD Extensions Support

Certain steps must be taken in both the application and the OS to check if the CPU supports
Streaming SIMD Extensions and associated unmasked exceptions. This section describes this
process, which is conducted using the bits described in Table 11-5 and Table 11-6.
If the OS wants to use FXSAVE/FXRSTOR, it will first check CPUID.FXSR to determine if the
CPU supports these instructions. If the CPU does support FXSAVE/FXRSTOR, then the OS can
set CR4.OSFXSR without faulting and enable code for context switching that utilizes
FXSAVE/FXRSTOR instead of FSAVE/FRSTOR.
At this point, if the OS also supports unmasked SIMD floating-point exceptions, it should check
CPUID.XMM to see if this is a Streaming SIMD Extensions-enabled processor. If
CPUID.XMM is set, this verifies that the OS can set CR4.OSXMMEXCPT without faulting.
The process by which an application detects the existence of Streaming SIMD Extensions as
discussed in Section 9.5.1., “Detecting Support for Streaming SIMD Extensions Using the
CPUID Instruction” Chapter 9, Programming with the Streaming SIMD Extensions, in the Intel
Architecture Software Developer’s Manual, Volume 1. For additional information and examples,
see AP-900, Identifying Support for Streaming SIMD Extensions in the Processor and Oper-
ating System.

Table 11-5. CPUID Bits for Streaming SIMD Extensions Support

CPUID bit (EAX = 1) Meaning
FXSR If set, CPU supports FXSAVE/FXRSTOR. The OS can read this bit
(EDX bit24) to determine if it can use FXSAVE/FXRSTOR in place of
FSAVE/FRSTOR for context switches.
XMM If set, the Streaming SIMD Extensions set is supported by the
(EDX bit25) processor.

Table 11-6. CR4 Bits for Streaming SIMD Extensions Support

CR4 bit Meaning
OSFXSR Defaults to clear. If both the CPU and the OS support FXSAVE/FXRSTOR for
(bit9) use during context switches, then the OS will set this bit.
OSXMMEXCPT Defaults to clear. The OS will set this bit if it supports unmasked SIMD floating-
(bit10) point exceptions.

11.4.2. Device Not Available (DNA) Exceptions

Streaming SIMD Extensions will cause a DNA Exception (#NM) if the processor attempts to
execute a SIMD floating-point instruction while CR0.TS is set. If CPUID.XMM is clear, execu-

11-6
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

tion of any Streaming SIMD Extensions instruction will cause an invalid opcode fault regardless
of the state of CR0.TS.

11.4.3. FXSAVE/FXRSTOR as a Replacement for FSAVE/FRSTOR

The FXSAVE and FXRSTOR instructions are designed to be a replacement for
FSAVE/FRSTOR, to be used by the OS for context switches. These have been optimized to be
faster than FSAVE/FRSTOR, while still saving/restoring the additional SIMD floating-point
state. To meet this goal, FXSAVE differs from FSAVE in that it does not cause an FINIT to be
performed, nor does FXSAVE initialize the SIMD floating-point registers in any way. While
FXSAVE/FXRSTOR does save/restore the x87-FP state, FSAVE/FRSTOR does not affect the
SIMD floating-point state. This allows for FXSAVE/FXRSTOR and FSAVE/FRSTOR to be
nested. State saved with FXSAVE and restored with FRSTOR (and vice versa) will result in
incorrect restoration of state in the processor. FXSAVE will not save the SIMD floating-point
state (SIMD floating-point registers and MXCSR register) if the CR4.OSFXSR bit is not set.

11.4.4. Numeric Error flag and IGNNE#

Streaming SIMD Extensions ignore CR0.NE (treats it as if it were always set) and the IGNNE#
pin and always use the vector 19 software exception for error reporting.

11.5. SAVING AND RESTORING THE STREAMING SIMD

EXTENSIONS STATE
The recommended method of saving and restoring the Streaming SIMD Extensions state is as
follows:
• Execute an FXSAVE instruction to write the entire state of the MMX™/FPU, the SIMD
floating-point registers, and the SIMD floating-point MXCSR to memory.
• Execute an FXRSTOR instruction to read the entire saved state of the MMX™/FPU, the
SIMDP floating-point registers and the SIMD floating-point MXCSR from memory into
the FPU registers and the aliased MMX™ registers.
This save and restore method is required for operating systems (see Section 10.6., “Designing
Operating System Task and Context Switching Facilities”).
Applications can in some cases save and restore only the SIMD floating-point registers, in the
following way:
• Execute eight MOVAPS instructions to write the contents of the SIMD floating-point
registers XMM0 through XMM7 to memory. Execute a STMXCSR instruction to save the
MXCSR register to memory.
• Execute eight MOVAPS instructions to read the saved contents of the SIMD floating-point
registers from memory into the XMM0 through XMM7 registers. Execute a LDMXCSR

11-7
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

instruction to read the saved contents of the MXCSR register from memory into the
MXCSR register.

11.6. DESIGNING OPERATING SYSTEM TASK AND CONTEXT

SWITCHING FACILITIES
When switching from one task or context to another, it is often necessary to save the SIMD
floating-point state (just as it is often necessary to save the state of the FPU). As mentioned in
the previous chapter, the MMX™ state is aliased on the FPU state. The SIMD floating-point
registers in the Pentium® III processor introduce a new state. When designing new SIMD
floating-point state saving facilities for an operating system, several approaches are available:
• The operating system can require that applications (which will be run as tasks) take
responsibility for saving the SIMD floating-point state prior to a task suspension during a
task switch and for restoring the SIMD floating-point state when the task is resumed. The
application can use either of the state saving and restoring techniques given in Section
10.5., “Saving and Restoring the Streaming SIMD Extensions state”. This approach to
saving the SIMD floating-point state is appropriate for cooperative multitasking operating
systems, where the application has control over (or is able to determine) when a task
switch is about to occur and can save state prior to the task switch.
• The operating system can take the responsibility for automatically saving the SIMD
floating-point state as part of the task switch process (using an FXSAVE instruction) and
automatically restoring the SIMD floating-point state when a suspended task is resumed
(using an FXRSTOR instruction). Here, the SIMD floating-point state must be saved as
part of the task state. This approach is appropriate for preemptive multitasking operating
systems, where the application cannot know when it is going to be preempted and cannot
prepare in advance for task switching. The operating system is responsible for saving and
restoring the task and SIMD floating-point state when necessary.
• The operating system can take the responsibility for saving the SIMD floating-point state
as part of the task switch process, but delay the saving of the SIMD floating-point state
until a Streaming SIMD Extensions instruction is actually executed by the new task. Using
this approach, the SIMD floating-point state is saved only if a Streaming SIMD Extensions
instruction needs to be executed in the new task. (See Section 10.6.1., “Using the TS Flag
in Control Register CR0 to Control SIMD Floating-Point State Saving”, for more
information on this SIMD floating-point state saving technique.)

11.6.1. Using the TS Flag in Control Register CR0 to Control SIMD

Floating-Point State Saving
Saving the SIMD floating-point state using the FXSAVE instruction is not as high-overhead
operation as FSAVE. However an operating system may choose to wait to save the SIMD
floating-point state to avoid this overhead. If a task being switched to will not access the SIMD
floating-point registers (by executing a Streaming SIMD Extensions instruction), this overhead
can be avoided by not automatically saving the SIMD floating-point state on a task switch.

11-8
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

The TS flag in control register CR0 is provided to allow the operating system to delay saving
the SIMD floating-point state until the SIMD floating-point registers are actually accessed in the
new task. When this flag is set, the processor monitors the instruction stream for Streaming
SIMD Extensions instructions. When the processor detects a Streaming SIMD Extensions
instruction, it raises a device-not-available exception (#NM) prior to executing the instruction.
The device-not-available exception handler can then be used to save the SIMD floating-point
state for the previous task (using an FXSAVE instruction) and load the SIMD floating-point state
for the current task (using an FXRSTOR instruction). If the task never encounters a Streaming
SIMD Extensions instruction, the device-not-available exception will not be raised and the
SIMD floating-point state will not be saved unnecessarily.
The TS flag can be set either explicitly (by executing a MOV instruction to control register CR0)
or implicitly (using the processor’s native task switching mechanism). When the native task
switching mechanism is used, the processor automatically sets the TS flag on a task switch.
After the device-not-available handler has saved the SIMD floating-point state, it should execute
the CLTS instruction to clear the TS flag in CR0.
Figure 10-2 gives an example of an operating system that implements SIMD floating-point state
saving using the TS flag. In this example, task A is the currently running task and task B is the
task being switched to.

Task A Task B
SIMD floating-point
Application State Owner

Operating System
CR0.TS=1 and
Task A extensions Task B
SIMD floating-point instruction SIMD floating-point
State Save Area is encountered. State Save Area

Operating System
Saves Task A Task Switching Code Loads Task B
SIMD floating-point State SIMD floating-point State
Device-Not-Available
Exception Handler

Figure 11-2. Example of SIMD Floating-Point State Saving During an Operating System-
Controlled Task Switch

The operating system maintains a SIMD floating-point save area for each task and defines a
variable (SIMD-fpStateOwner) that indicates which task “owns” the SIMD floating-point state.
In this example, task A is the current SIMD floating-point state owner.
On a task switch, the operating system task switching code must execute the following pseudo-
code to set the TS flag according to the current SIMD floating-point state owner. If the new task

11-9
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

(task B in this example) is not the current SIMD floating-point state owner, the TS flag is set to
1; otherwise, it is set to 0.
IF Task_Being_Switched_To ≠ SIMD-fpStateOwner
THEN
CR0.TS ← 1;
ELSE
CR0.TS ← 0;
FI;
If a new task attempts to use a Streaming SIMD Extensions instruction while the TS flag is set
to 1, a device-not-available exception (#NM) is generated and the device-not-available excep-
tion handler executes the following pseudo-code.
CR0.TS ← 0;
FXSAVE “To SIMD floating-point State Save Area for Current SIMD Floating-point State
Owner”;
FXRSTOR “SIMD floating-point State From Current Task’s SIMD Floating-point State Save
Area”;
SIMF-fpStateOwner ← Current_Task;

11-10
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

This handler code performs the following tasks:

• Clears the TS flag.
• Saves the SIMD floating-point state in the state save area for the current SIMD floating-
point state owner.
• Restores the SIMD floating-point state from the new task’s SIMD floating-point state save
area.
• Updates the current SIMD floating-point state owner to be the current task.

11.7. EXCEPTIONS THAT CAN OCCUR WHEN EXECUTING

STREAMING SIMD EXTENSIONS INSTRUCTIONS
Streaming SIMD Extensions can generate two kinds of exceptions:
• Non-numeric exceptions
• Numeric exceptions
Streaming SIMD Extensions can generate the same type of memory access exceptions as the
Intel Architecture instructions do. Some examples are: page fault, segment not present, and limit
violations. Existing exception handlers can handle these types of exceptions without any code
modification. The SIMD floating-point PREFETCH instruction hints will not generate any kind
of exception and instead will be ignored.
Streaming SIMD Extensions can generate the same six numeric exceptions that x87-FP instruc-
tions can generate. All Streaming SIMD Extensions numeric exceptions are reported indepen-
dently of x87-FP numeric exceptions. Independent masking and unmasking of Streaming SIMD
Extensions numeric exceptions is achieved by setting/resetting specific bits in the MXCSR
register.
The application must ensure that the OS can support unmasked SIMD floating-point exceptions
before unmasking them. For more details, refer to Section 9.5.1., “Detecting Support for
Streaming SIMD Extensions Using the CPUID Instruction” Chapter 9, Programming with the
Streaming SIMD Extensions, in the Intel Architecture Software Developer’s Manual, Volume 1
and AP-900, Identifying Support for Streaming SIMD Extensions in the Processor and Oper-
ating System. If an application unmasks exceptions using either FXRSTOR or LDMXCSR
without the required OS support being enabled, then an invalid opcode fault, instead of a SIMD
floating-point exception, will be generated on the first faulting SIMD floating-point instruction.

11-11
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

11.7.1. SIMD Floating-point Non-Numeric Exceptions

• Exceptions during memory accesses:
— Invalid opcode (#UD).
— Stack exception (#SS).
— General protection (#GP).
— Page fault (#PF).
— Alignment check (#AC), if alignment checking is enabled.
• System exceptions:
— Invalid Opcode (#UD), if the EM flag in control register CR0 is set, the CPUID.XMM
bit is not set, or the CR4.OSFXSR* bit is not set, when a Streaming SIMD Extensions
instruction is executed (see Section 10.1., “Emulation of the Streaming SIMD Exten-
sions”).
— Device not available (#NM), if a Streaming SIMD Extensions instruction is executed
when the TS flag in control register CR0 is set. (See Section 10.6.1., “Using the TS
Flag in Control Register CR0 to Control SIMD Floating-Point State Saving”.)
• Other exceptions can occur indirectly due to the faulty execution of the exception handlers
for the above exceptions. For example, if a stack-segment fault (#SS) occurs due to
Streaming SIMD Extensions instructions, the interrupt gate for the stack-segment fault can
direct the processor to invalid TSS, causing an invalid TSS exception (#TS) to be
generated.
Table 11-7 lists the causes for Interrupt 6 and Interrupt 7 with Streaming SIMD Extensions.

Table 11-7. Streaming SIMD Extensions Faults

CR0.EM CR4.OSFXSR CPUID.XMM CR0.TS EXCEPTION
1 - - - #UD Interrupt 6
- 0 - - #UD Interrupt 6
- - 0 - #UD Interrupt 6
0 1 1 1 #NM Interrupt 7

11-12
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

11.7.2. SIMD Floating-point Numeric Exceptions

There are six classes of numeric exception conditions that can occur while executing Streaming
SIMD Extensions:
• Invalid operation (#I)
• Divide-by-zero (#Z)
• Denormal operand (#D)
• Numeric overflow (#O)
• Numeric underflow (#U)
• Inexact result (Precision) (#P)
Invalid, Divide-by-zero and Denormal exceptions are pre-computation exceptions, i.e., they are
detected before any arithmetic operation occurs. Underflow, Overflow and Precision exceptions
are post-computation exceptions.
When numeric exceptions occur, a processor supporting Streaming SIMD Extensions takes one
of two possible courses of action:
• The processor can handle the exception by itself, producing the most reasonable result and
allowing numeric program execution to continue undisturbed (i.e., masked exception
response).
• A software exception handler can be invoked to handle the exception (i.e., unmasked
exception response).
Each of the six exception conditions described above has corresponding flag and mask bits in
the MXCSR. If an exception is masked (the corresponding mask bit in MXCSR = 1), the
processor takes an appropriate default action and continues with the computation. If the excep-
tion is unmasked (mask bit = 0) and the OS supports SIMD floating-point exceptions (i.e.
CR4.OSXMMEXCPT = 1), a software exception handler is invoked immediately through
SIMD floating-point exception interrupt vector 19. If the exception is unmasked (mask bit = 0)
and the OS does not support SIMD floating-point exceptions (i.e. CR4.OSXMMEXCPT = 0),
an invalid opcode exception is signaled instead of a SIMD floating-point exception.
Note that because SIMD floating-point exceptions are precise and occur immediately, the situ-
ation does not arise where an x87-FP instruction, an FWAIT instruction, or another Streaming
SIMD Extensions instruction will catch a pending unmasked SIMD floating-point exception.

11.7.2.1. EXCEPTION PRIORITY

The processor handles exceptions according to a predetermined precedence. When a sub-
operand of a packed instruction generates two or more exception conditions, the exception
precedence sometimes results in the higher-priority exception being handled and the lower-
priority exceptions being ignored. For example, dividing an SNaN by zero could potentially
signal an invalid-arithmetic-operand exception (due to the SNaN operand) and a divide-by-zero
exception. Here, if both exceptions are masked, the processor handles the higher-priority excep-
tion only (the invalid-arithmetic-operand exception), returning the quiet version of the SNaN to

11-13
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

the destination. The prioritization policy also applies for unmasked exceptions; if both invalid
and divide-by-zero are unmasked for the previous example, only the invalid flag will be set.
Prioritization of exceptions is performed only on an individual sub-operand basis, and not
between suboperands; for example, an invalid exception generated by one sub-operand will not
prevent the reporting of a divide-by-zero exception generated by another sub-operand.
The precedence for SIMD floating-point numeric exceptions is as follows:
1. Invalid operation exception due to NaN operands (refer to Table 11-8).
2. QNaN operand. Though this is not an exception, the handling of a QNaN operand has
precedence over lower-priority exceptions. For example, a QNaN divided by zero results
in a QNaN, not a zero-divide exception.
3. Any other invalid operation exception not mentioned above or a divide-by-zero exception
(refer to Table 11-8).
4. Denormal operand exception. If masked, then instruction execution continues, and a
lower-priority exception can occur as well.
5. Numeric overflow and underflow exceptions possibly in conjunction with the inexact
result exception.
6. Inexact result exception.

11.7.2.2. AUTOMATIC MASKED EXCEPTION HANDLING

If the processor detects an exception condition for a masked exception (an exception with its
mask bit set), it delivers a predefined (default) response and continues executing instructions.
The masked (default) responses to exceptions have been chosen to deliver a reasonable result
for each exception condition and are generally satisfactory for most application code. By
masking or unmasking specific floating-point exceptions in the MXCSR, programmers can
delegate responsibility for most exceptions to the processor and reserve the most severe excep-
tion conditions for software exception handlers.
Because the exception flags are “sticky,” they provide a cumulative record of the exceptions that
have occurred since they were last cleared. A programmer can thus mask all exceptions, run a
calculation, and then inspect the exception flags to see if any exceptions were detected during
the calculation.
Note that when exceptions are masked, the processor may detect multiple exceptions in a single
instruction, because:
• It continues executing the instruction after performing its masked response; for example,
the processor could detect a denormalized operand, perform its masked response to this
exception, and then detect an underflow
• Exceptions may occur naturally in pairs, such as numeric underflow and inexact result
(precision)
• Packed instructions can produce independent exceptions for each pair of operands.

11-14
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

Updating of exception flags is generated by a logical-OR of exception conditions for all sub-
operand computations, where the OR is done independently for each type of exception; for
packed computations this means 4 sub-operands and for scalar computations this means 1 sub-
operand (the lowest one).

11.7.2.3. SOFTWARE EXCEPTION HANDLING - UNMASKED EXCEPTIONS

An application must ensure that the operating system supports unmasked exceptions before
unmasking any of the exceptions in the MXCSR (refer to Section 9.5.1., “Detecting Support for
Streaming SIMD Extensions Using the CPUID Instruction” Chapter 9, Programming with the
Streaming SIMD Extensions, Volume 1 of the Programmer’s Reference Manual).
If the processor detects a condition for an unmasked SIMD floating-point application exception,
a software handler is invoked immediately at the end of the excepting instruction. The handler
is invoked through the SIMD floating-point exception interrupt (vector 19), irrespective of the
state of the CR0.NE flag. If an exception is unmasked, but SIMD floating-point unmasked
exceptions are not enabled (CR4.OSXMMEXCPT = 0), an invalid opcode fault is generated.
However, the corresponding exception bit will still be set in the MXCSR, as it would be if
CR4.OSXMMEXCPT =1, since the invalid opcode handler or the user needs to determine the
cause of the exception.
A typical action of the exception handler is to store x87-FP and SIMD floating-point state infor-
mation in memory (with the FXSAVE/FXRSTOR instructions) so that it can evaluate the excep-
tion and formulate an appropriate response. Other typical exception handler actions can include:
• Examine stored x87-FP and SIMD floating-point state information (control/status) to
determine the nature of the error.
• Taking action to correct the condition that caused the error.
• Clear the exception bits in the x87-FP status word (FSW) or the SIMD floating-point
control register (MXCSR)
• Return to the interrupted program and resume normal execution.
In lieu of writing recovery procedures, the exception handler can do one or more of the
following:
• Increment in software an exception counter for later display or printing.
• Print or display diagnostic information (such as the SIMD floating-point register state).
• Halt further program execution.
When an unmasked exception occurs, the processor will not alter the contents of the source
register operands prior to invoking the unmasked handler. Similarly, the integer EFLAGS will
also not be modified if an unmasked exception occurs while executing the COMISS or
UCOMISS instructions. Exception flags will be updated according to the following rules:
• Updating of exception flags is generated by a logical-OR of exception conditions for all
sub-operand computations, where the OR is done independently for each type of

11-15
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

exception; for packed computations this means 4 sub-operands and for scalar computations
this means 1 sub-operand (the lowest one).
• In the case of only masked exception conditions, all flags will be updated,
• In the case of an unmasked pre-computation type of exception condition (e.g., denormal
input), all flags relating to all pre-computation conditions (masked or unmasked) will be
updated, and no subsequent computation is performed (i.e., no post-computation condition
can occur if there is an unmasked pre-computation condition).
• In the case of an unmasked post-computation exception condition, all flags relating to all
post-computation conditions (masked or unmasked) will be updated; all pre-computation
conditions, which must be masked-only will also be reported.

11.7.2.4. INTERACTION WITH X87 NUMERIC EXCEPTIONS

The Streaming SIMD Extensions control/status register was separated from its x87-FP counter-
parts to allow for maximum flexibility. Consequently, the Streaming SIMD Extensions architec-
ture is independent of the x87-FP architecture, but has the following implications for x87-FP
applications that call Streaming SIMD Extensions-enabled libraries:
• The x87-FP rounding mode specified in FCW will not apply to calls in a Streaming SIMD
Extensions library (unless the rounding control in MXCSR is explicitly set to the same
mode).
• x87-FP exception observability may not apply to a Streaming SIMD Extensions library.
• An application that expects to catch x87-FP exceptions that occur in an x87-FP
library will not be notified if an exception occurs in a Streaming SIMD Extensions
library, unless the exception masks enabled in FCW have also been enabled in
MXCSR.
• An application will not be able to unmask exceptions after returning from a
Streaming SIMD Extensions library call to detect if an error occurred. A SIMD
floating-point exception flag that is already set when the corresponding exception
is unmasked will not generate a fault; only the next occurrence of that exception
will generate an unmasked fault.
• An application which checks FSW to determine if any masked exception flags
were set during an x87-FP library call will also need to check MXCSR in order to
observe a similar occurrence of a masked exception within a Streaming SIMD
Extensions library.

11.7.3. SIMD Floating-point Numeric Exception Conditions and

Masked/Unmasked Responses
The following sections describe the various conditions that cause a SIMD floating-point
numeric exception to be generated and the masked response of the processor when these condi-
tions are detected.

11-16
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

11.7.3.1. INVALID OPERATION EXCEPTION(#IA)

The invalid operation exception occurs in response to an invalid arithmetic operand, or to an
invalid combination of operands.
If the invalid operation exception is masked, the processor sets the IE flag in MXCSR and
returns the single-precision QNaN indefinite value or another QNaN value (derived from a NaN
input operand) to the destination operand. This value overwrites the destination register speci-
fied by the instruction.
If the invalid operation exception is not masked, the processor sets the IE flag in MXCSR and
an exception handler is invoked (see Section 11.7.2.3., “Software Exception Handling -
Unmasked Exceptions”) and the operands remain unchanged.
The processor can detect a variety of invalid arithmetic operations that can be coded in a
program. These operations generally indicate a programming error, such as dividing ∞ by ∞.
Table 11-8 lists the SIMD floating-point invalid arithmetic operations that the processor detects.
This group includes the invalid operations defined in IEEE Std. 854.
The flag (IE) for this exception is bit 0 of MXCSR, and the mask bit (IM) is bit 7 of MXCSR.
The invalid operation exception is not affected by the flush-to-zero mode.

11-17
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

Table 11-8. Invalid Arithmetic Operations and the Masked Responses to Them
Condition Masked Response
ADDPS/ADDSS/DIVPS/DIVSS/ Return the Signaling NaN converted to a quiet
MULPS/MULSS/SUBPS/SUBSS with a SNaN NaN; Refer to Table 7-18, in Chapter 7,
operand. Floating-Point Unit, for more details; set #IA
flag.
CMPPS/CMPSS with QNaN/SNaN operands Return a mask of all 0’s for predicates "eq", "lt",
(QNaN applies only for predicates "lt", "le", "nlt", "le", and "ord", and a mask of all 1’s for
"nle") predicates "neq", "nlt", "nle", and "unord"; set
#IA flag.
COMISS with QNaN/SNaN operand(s). Set EFLAGS values to ’not comparable’; set
#IA flag.
UCOMISS with SNaN operand(s). Set EFLAGS values to ’not comparable’; set
#IA flag.
SQRTPS/SQRTSS with SNaN operand(s). Return the SNan converted to a QNaN; set #IA
flag;
Addition of opposite signed infinities or Return the QNaN Indefinite; set #IA flag.
subtraction of like-signed infinities.
Multiplication of infinity by zero. Return the QNaN Indefinite; set #IA flag.
Divide of (0/0) or( ∞ / ∞ .) Return the QNaN Indefinite; set #IA flag.
SQRTPS/SQRTSS of negative operands (except Return the QNaN Indefinite; set #IA flag.
negative zero).
Conversion to integer when the source register is Return the Integer Indefinite; set #IA flag.
a NaN, Infinity or exceeds the representable
range.

NOTE:
RCPPS/RCPSS/RSQRTPS/RSQRTSS with QNaN/SNaN operand(s) do not raise an invalid exception.
They return either the SNaN operand converted to QNaN, or the original QNaN operand.
RSQRTPS/RSQRTSS with negative operands (but not for negative zero) do not raise an invalid excep-
tion, and return QNaN Indefinite.

11.7.3.2. DIVISION-BY-ZERO EXCEPTION (#Z)

The processor reports a divide-by-zero exception whenever an instruction attempts to divide a
finite non-zero operand by 0. This is possible with DIVPS, DIVSS.
The masked response for DIVPS, DIVSS is to set the ZE flag in MXCSR and return an infinity
signed with the exclusive OR of the signs of the operands. If the divide-by-zero exception is not
masked, the ZE flag is set, a software exception handler is invoked (see Section 11.7.2.3., “Soft-
ware Exception Handling - Unmasked Exceptions”) and the source operands remain unchanged.
Note that the response for RCPPS, RSQRTPS, RCPSS and RSQRTSS is to return an infinity of
the same sign as the operand. These instructions do not set any exception flags and thus are not
affected by the exception masks.
The flag (ZE) for the divide-by-zero exception is bit 2 of MXCSR, and the mask bit (ZM) is bit
9 of MXCSR.
The divide-by-zero exception is not affected by the flush-to-zero mode.

11-18
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

11.7.3.3. DENORMAL OPERAND EXCEPTION (#D)

The processor signals the denormal operand exception if an arithmetic instruction attempts to
operate on a denormal operand.
When a denormal operand exception occurs and the exception is masked, the processor sets the
DE flag in MXCSR, then proceeds with the instruction. Operating on denormal numbers will
produce results at least as good as, and often better than, what can be obtained when denormal
numbers are flushed to zero. Programmers can mask this exception so that a computation may
proceed, then analyze any loss of accuracy when the final result is delivered.
When a denormal operand exception occurs and the exception is not masked, the processor sets
the DE bit in MXCSR and a software exception handler is invoked (see Section 11.7.2.3., “Soft-
ware Exception Handling - Unmasked Exceptions”). The source operands remain unchanged.
When denormal operands have reduced significance due to loss of low-order bits, it may be
advisable to not operate on them. Precluding denormal operands from computations can be
accomplished by an exception handler that responds to unmasked denormal operand exceptions.
Note that the response for RCPPS, RSQRTPS, RCPSS and RSQRTSS is to return an infinity of
the same sign as the operand. These instructions do not set any exception flags and thus are not
affected by the exception masks.
Conversion instructions (CVTPI2PS, CVTPS2PI, CVTTPS2PI, CVTSI2SS, CVTSS2SI,
CVTTSS2SI) do not signal denormal exceptions.
The flag (DE) for this exception is bit 1 of MXCSR, and the mask bit (DM) is bit 8 of MXCSR.
The denormal operand exception is not affected by the flush-to-zero mode.

11.7.3.4. NUMERIC OVERFLOW EXCEPTION (#O)

The processor reports a floating-point numeric overflow exception whenever the result of an
instruction rounded to the destination precision with unbounded exponent exceeds the largest
allowable finite value that will fit into the destination operand. This is possible with ADDPS,
ADDSS, SUBPS, SUBSS, MULPS, MULSS, DIVPS, DIVSS.
When a numeric overflow exception occurs and the exception is masked, the processor sets the
MXCSR.OE and MXCSR.PE flags and returns one of the values shown in Table 11-9 according
to the current rounding mode of the processor (see Section 11.3.2.1., “Rounding Control Field”).
When a numeric overflow exception occurs and the exception is unmasked, the operands are left
unaltered and a software exception handler is invoked (see Section 11.7.2.3., “Software Excep-
tion Handling - Unmasked Exceptions”). The MXCSR.OE flag is set; the MXCSR.PE flag is
only set if a loss of accuracy has occurred in addition to overflow when rounding the result to
the destination precision, with unbounded exponent.
The flag (OE) for the numeric overflow exception is bit 3 of MXCSR, and the mask bit (OM) is
bit 10 of MXCSR.
The numeric overflow exception is not affected by the flush-to-zero mode.

11-19
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

Note that the overflow status flag is not set by RCPPS/RCPSS, since these instructions are
combinatorial and are not affected by exception masks.
.
Table 11-9. Masked Responses to Numeric Overflow
Rounding Mode Sign of True Result Result
To nearest + +∞
– –∞
Toward –∞ + Largest finite positive number
– –∞
Toward +∞ + +∞
– Largest finite negative number
Toward zero + Largest finite positive number
– Largest finite negative number

11.7.3.5. NUMERIC UNDERFLOW EXCEPTION (#U)

The processor might report a floating-point numeric underflow exception whenever the rounded
result of an arithmetic instruction is tiny; that is, the result rounded to the destination precision
with unbounded exponent is less than the smallest possible normalized, finite value that will fit
into the destination operand. The Underflow exception can occur in the execution of the instruc-
tions ADDPS, ADDSS, SUBPS, SUBSS, MULPS, MULSS, DIVPS and DIVSS.
Two related events contribute to underflow:
• Creation of a tiny result which, because it is so small, may cause some other exception
later (such as overflow upon division).
• Creation of an inexact result; i.e. the delivered result differs from what would have been
computed were both the exponent and precision unbounded.
Which of these events triggers the underflow exception depends on whether the underflow
exception is masked:
• Underflow exceptions masked. The underflow exception is signaled when the result is both
tiny and inexact.
• Underflow exceptions not masked: The underflow exception is signaled when the result is
tiny, regardless of inexactness.
The response to an underflow exception also depends on whether the exception is masked:
• Masked response: The result is normal, denormal or zero. The precision exception is also
triggered. The OE and PE flags are set in MXCSR.
• Unmasked response: The UE flag is set in MXCSR. If the original computation generated
an imprecise mantissa, the inexact (#P) status flag PE will also be set in the MXCSR. In
either case (result imprecise or not), the underflow (#U) status flag is set, the operands are

11-20
 STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

left unaltered, and a software exception handler is invoked (see Section 11.7.2.3.,
“Software Exception Handling - Unmasked Exceptions”).
If underflow is masked and flush-to-zero mode is enabled, an underflow condition will set the
underflow (#U) and inexact (#P) status flags UE and PE in MXCSR and a correctly signed zero
result will be returned; this will avoid the performance penalty associated with generating a
denormalized result. If underflow is unmasked, the flush-to-zero mode is ignored and an under-
flow condition will be handled as described above.
Note that the underflow status flag is not set by RCPPS/RCPSS, since these instructions are
combinatorial and are not affected by exception masks.
The flag (UE) for the numeric underflow exception is bit 4 of MXCSR and the mask bit (UM)
is bit 11 of MXCSR.

11.7.3.6. INEXACT RESULT (PRECISION) EXCEPTION (#P)

The inexact result exception (also called the precision exception) occurs if the result of an oper-
ation is not exactly representable in the destination format. For example, the fraction 1/3 cannot
be precisely represented in binary form. This exception occurs frequently and indicates that
some (normally acceptable) accuracy has been lost. The exception is supported for applications
that need to perform exact arithmetic only. Because the rounded result is generally satisfactory
for most applications, this exception is commonly masked.
If the inexact result exception is masked when an inexact result condition occurs and a numeric
overflow or underflow condition has not occurred, the processor sets the inexact (#P) status flag
(PE flag) and stores the rounded result in the destination operand. The current rounding mode
determines the method used to round the result (refer to Section 11.3.2.1., “Rounding Control
Field”).
If the inexact result exception is not masked when an inexact result occurs and numeric overflow
or underflow has not occurred, the operands are left unaltered, the PE flag is set in MXCSR, the
inexact (#P) status flag is set, and a software exception handler is invoked (see Section 11.7.2.3.,
“Software Exception Handling - Unmasked Exceptions”).
If an inexact result occurs in conjunction with numeric overflow or underflow, one of the
following operations is carried out:
• If an inexact result occurs along with masked overflow or underflow, the OE or UE flag
and the PE flag are set in MXCSR and the result is stored as described for the overflow or
underflow exceptions (see Section 11.7.3.4., “Numeric Overflow Exception (#O)”. or
Section 11.7.3.5., “Numeric Underflow Exception (#U)”). If the inexact result exception is
unmasked, the processor also invokes the software exception handler.
• If an inexact result occurs along with unmasked overflow or underflow, the OE or UE flag
and the PE flag are set and the software exception handler is invoked.
Note that the inexact result flag is not set by RCPPS, RSQRTPS, RCPSS and RSQRTSS, since
these instructions are combinatorial and are not affected by the exception masks.
The inexact result exception flag (PE) is bit 5 of MXCSR, and the mask bit (PM) is bit 12 of
MXCSR.

11-21
STREAMING SIMD EXTENSIONS SYSTEM PROGRAMMING

In flush-to-zero mode, the inexact result exception is reported along with the underflow excep-
tion (the latter must be masked).

11.7.4. Effect of Streaming SIMD Extensions Instructions on

Pending Floating-Point Exceptions
Unlike MMX™ instructions which will generate a floating-point error (#MF) prior to executing
the MMX™ instruction, execution of a Streaming SIMD Extensions instruction does not
generate a floating-point error (#MF) prior to executing the instruction. Hence they will not
catch pending x87 floating-point exceptions. In addition, they will not cause assertion of FERR#
(independent of the value of CR0.NE) and they ignore the assertion/de-assertion of IGNNE#.

11.8. DEBUGGING
The debug facilities of the Intel Architecture operate in the same manner when executing
Streaming SIMD Extensions as when executing other Intel Architecture instructions. These
facilities enable debuggers to debug code utilizing these instructions.
To correctly interpret the contents of the Pentium® III processor registers from the FXSAVE
image in memory, a debugger needs to take account of the relationship between the floating-
point register’s logical locations relative to TOS and the MMX™ register’s physical locations
(refer to Section 10.6., “Debugging”, Chapter 10, MMX™ Technology System Programming).
In addition it needs to have knowledge of the SIMD floating-point registers and the state save
data area used by the FXSAVE instruction.
Comparisons of the Streaming SIMD Extensions and x87 results can be performed within the
Pentium® III processor at the internal single precision format and/or externally at the memory
single precision format. The internal format comparison is required to allow the partitioning of
the data space to reduce test time.

11-22
 12
System Management
Mode
 SYSTEM MANAGEMENT MODE (SMM)

CHAPTER 12
SYSTEM MANAGEMENT MODE (SMM)

This chapter describes the Intel Architecture’s System Management Mode (SMM) architecture.
SMM was introduced into the Intel Architecture in the Intel386™ SL processor (a mobile
specialized version of the Intel386™ processor). It is also available in the Intel486™ processors
(beginning with the Intel486™ SL and Intel486™ enhanced versions) and in the Intel Pentium®
and P6 family processors. For a detailed description of the hardware that supports SMM, refer
to the developer’s manuals for each of the Intel Architecture processors.

12.1. SYSTEM MANAGEMENT MODE OVERVIEW

SMM is a special-purpose operating mode provided for handling system-wide functions like
power management, system hardware control, or proprietary OEM-designed code. It is intended
for use only by system firmware, not by applications software or general-purpose systems soft-
ware. The main benefit of SMM is that it offers a distinct and easily isolated processor environ-
ment that operates transparently to the operating system or executive and software applications.
When SMM is invoked through a system management interrupt (SMI), the processor saves the
current state of the processor (the processor’s context), then switches to a separate operating
environment contained in system management RAM (SMRAM). While in SMM, the processor
executes SMI handler code to perform operations such as powering down unused disk drives or
monitors, executing proprietary code, or placing the whole system in a suspended state. When
the SMI handler has completed its operations, it executes a resume (RSM) instruction. This
instruction causes the processor to reload the saved context of the processor, switch back to
protected or real mode, and resume executing the interrupted application or operating-system
program or task.
The following SMM mechanisms make it transparent to applications programs and operating
systems:
• The only way to enter SMM is by means of an SMI.
• The processor executes SMM code in a separate address space (SMRAM) that can be
made inaccessible from the other operating modes.
• Upon entering SMM, the processor saves the context of the interrupted program or task.
• All interrupts normally handled by the operating system are disabled upon entry into
SMM.
• The RSM instruction can be executed only in SMM.
SMM is similar to real-address mode in that there are no privilege levels or address mapping.
An SMM program can address up to 4 GBytes of memory and can execute all I/O and applicable
system instructions. Refer to Section 12.5., “SMI Handler Execution Environment” for more
information about the SMM execution environment.

12-1
SYSTEM MANAGEMENT MODE (SMM)

NOTE
The physical address extension (PAE) mechanism available in the P6 family
processors is not supported when a processor is in SMM.

12.2. SYSTEM MANAGEMENT INTERRUPT (SMI)

The only way to enter SMM is by signaling an SMI through the SMI# pin on the processor or
through an SMI message received through the APIC bus. The SMI is a nonmaskable external
interrupt that operates independently from the processor’s interrupt- and exception-handling
mechanism and the local APIC. The SMI takes precedence over an NMI and a maskable inter-
rupt. SMM is nonreentrant; that is, the SMI is disabled while the processor is in SMM.

NOTE
In the P6 family processors, when a processor that is designated as the
application processor during an MP initialization protocol is waiting for a
startup IPI, it is in a mode where SMIs are masked.

12.3. SWITCHING BETWEEN SMM AND THE OTHER PROCESSOR

OPERATING MODES
Figure 2-2 in Chapter 2, System Architecture Overview shows how the processor moves between
SMM and the other processor operating modes (protected, real-address, and virtual-8086).
Signaling an SMI while the processor is in real-address, protected, or virtual-8086 modes always
causes the processor to switch to SMM. Upon execution of the RSM instruction, the processor
always returns to the mode it was in when the SMI occurred.

12.3.1. Entering SMM

The processor always handles an SMI on an architecturally defined “interruptible” point in
program execution (which is commonly at an Intel Architecture instruction boundary). When
the processor receives an SMI, it waits for all instructions to retire and for all stores to complete.
The processor then saves its current context in SMRAM (refer to Section 12.4., “SMRAM”),
enters SMM, and begins to execute the SMI handler.
Upon entering SMM, the processor signals external hardware that SMM handling has begun.
The signaling mechanism used is implementation dependent. For the P6 family processors, an
SMI acknowledge transaction is generated on the system bus and the multiplexed status signal
EXF4 is asserted each time a bus transaction is generated while the processor is in SMM. For
the Pentium® and Intel486™ processors, the SMIACT# pin is asserted.
An SMI has a greater priority than debug exceptions and external interrupts. Thus, if an NMI,
maskable hardware interrupt, or a debug exception occurs at an instruction boundary along with
an SMI, only the SMI is handled. Subsequent SMI requests are not acknowledged while the
processor is in SMM. The first SMI interrupt request that occurs while the processor is in SMM

12-2
 SYSTEM MANAGEMENT MODE (SMM)

(that is, after SMM has been acknowledged to external hardware) is latched and serviced when
the processor exits SMM with the RSM instruction. The processor will latch only one SMI while
in SMM.
Refer to Section 12.5., “SMI Handler Execution Environment” for a detailed description of the
execution environment when in SMM.

12.3.1.1. EXITING FROM SMM

The only way to exit SMM is to execute the RSM instruction. The RSM instruction is only avail-
able to the SMI handler; if the processor is not in SMM, attempts to execute the RSM instruction
result in an invalid-opcode exception (#UD) being generated.
The RSM instruction restores the processor’s context by loading the state save image from
SMRAM back into the processor’s registers. The processor then returns an SMIACK transaction
on the system bus and returns program control back to the interrupted program.
Upon successful completion of the RSM instruction, the processor signals external hardware
that SMM has been exited. For the P6 family processors, an SMI acknowledge transaction is
generated on the system bus and the multiplexed status signal EXF4 is no longer generated on
bus cycles. For the Pentium® and Intel486™ processors, the SMIACT# pin is deserted.
If the processor detects invalid state information saved in the SMRAM, it enters the shutdown
state and generates a special bus cycle to indicate it has entered shutdown state. Shutdown
happens only in the following situations:
• A reserved bit in control register CR4 is set to 1 on a write to CR4. This error should not
happen unless SMI handler code modifies reserved areas of the SMRAM saved state map
(refer to Section 12.4.1., “SMRAM State Save Map”). Note that CR4 is not distinctly part
of the saved state map.
• An illegal combination of bits is written to control register CR0, in particular PG set to 1
and PE set to 0, or NW set to 1 and CD set to 0.
• (For the Pentium® and Intel486™ processors only.) If the address stored in the SMBASE
register when an RSM instruction is executed is not aligned on a 32-KByte boundary. This
restriction does not apply to the P6 family processors.
In shutdown state, the processor stops executing instructions until a RESET#, INIT# or NMI#
is asserted. The processor also recognizes the FLUSH# signal while in the shutdown state. In
addition, the Pentium® processor recognizes the SMI# signal while in shutdown state, but the P6
family and Intel486™ processors do not. (It is not recommended that the SMI# pin be asserted
on a Pentium® processor to bring the processor out of shutdown state, because the action of the
processor in this circumstance is not well defined.)
If the processor is in the HALT state when the SMI is received, the processor handles the return
from SMM slightly differently (refer to Section 12.10., “Auto HALT Restart”). Also, the
SMBASE address can be changed on a return from SMM (refer to Section 12.11., “SMBASE
Relocation”).

12-3
SYSTEM MANAGEMENT MODE (SMM)

12.4. SMRAM
While in SMM, the processor executes code and stores data in the SMRAM space. The SMRAM
space is mapped to the physical address space of the processor and can be up to 4 GBytes in size.
The processor uses this space to save the context of the processor and to store the SMI handler
code, data and stack. It can also be used to store system management information (such as the
system configuration and specific information about powered-down devices) and OEM-specific
information.
The default SMRAM size is 64 KBytes beginning at a base physical address in physical memory
called the SMBASE (refer to Figure 12-1). The SMBASE default value following a hardware
reset is 30000H. The processor looks for the first instruction of the SMI handler at the address
[SMBASE + 8000H]. It stores the processor’s state in the area from [SMBASE + FE00H] to
[SMBASE + FFFFH]. Refer to Section 12.4.1., “SMRAM State Save Map” for a description of
the mapping of the state save area.
The system logic is minimally required to decode the physical address range for the SMRAM
from [SMBASE + 8000H] to [SMBASE + FFFFH]. A larger area can be decoded if needed. The
size of this SMRAM can be between 32 KBytes and 4 GBytes.
The location of the SMRAM can be changed by changing the SMBASE value (refer to Section
12.11., “SMBASE Relocation”). It should be noted that all processors in a multiple-processor
system are initialized with the same SMBASE value (30000H). Initialization software must
sequentially place each processor in SMM and change its SMBASE so that it does not overlap
those of other processors.
The actual physical location of the SMRAM can be in system memory or in a separate RAM
memory. The processor generates an SMI acknowledge transaction (P6 family processors) or
asserts the SMIACT# pin (Pentium® and Intel486™ processors) when the processor receives an
SMI (refer to Section 12.3.1., “Entering SMM”). System logic can use the SMI acknowledge
transaction or the assertion of the SMIACT# pin to decode accesses to the SMRAM and redirect
them (if desired) to specific SMRAM memory. If a separate RAM memory is used for SMRAM,
system logic should provide a programmable method of mapping the SMRAM into system
memory space when the processor is not in SMM. This mechanism will enable start-up proce-
dures to initialize the SMRAM space (that is, load the SMI handler) before executing the SMI
handler during SMM.

12-4
 SYSTEM MANAGEMENT MODE (SMM)

SMRAM
SMBASE + FFFFH
Start of State Save Area

SMI Handler Entry Point

SMBASE + 8000H

SMBASE
Figure 12-1. SMRAM Usage

12.4.1. SMRAM State Save Map

When the processor initially enters SMM, it writes its state to the state save area of the SMRAM.
The state save area begins at [SMBASE + 8000H + 7FFFH] and extends down to [SMBASE +
8000H + 7E00H]. Table 12-1 shows the state save map. The offset in column 1 is relative to the
SMBASE value plus 8000H. Reserved spaces should not be used by software.
Some of the registers in the SMRAM state save area (marked YES in column 3) may be read
and changed by the SMI handler, with the changed values restored to the processor registers by
the RSM instruction. Some register images are read-only, and must not be modified (modifying
these registers will result in unpredictable behavior). An SMI handler should not rely on any
values stored in an area that is marked as reserved.

Table 12-1. SMRAM State Save Map

Offset
(Added to SMBASE + 8000H) Register Writable?
7FFCH CR0 No
7FF8H CR3 No
7FF4H EFLAGS Yes
7FF0H EIP Yes
7FECH EDI Yes
7FE8H ESI Yes
7FE4H EBP Yes
7FE0H ESP Yes
7FDCH EBX Yes
7FD8H EDX Yes
7FD4H ECX Yes

12-5
SYSTEM MANAGEMENT MODE (SMM)

Table 12-1. SMRAM State Save Map (Contd.)

Offset
(Added to SMBASE + 8000H) Register Writable?
7FD0H EAX Yes
7FCCH DR6 No
7FC8H DR7 No
7FC4H TR* No
7FC0H LDT Base* No
7FBCH GS* No
7FB8H FS* No
7FB4H DS* No
7FB0H SS* No
7FACH CS* No
7FA8H ES* No
7FA7H - 7F04H Reserved No
7F02H Auto HALT Restart Field (Word) Yes
7F00H I/O Instruction Restart Field (Word) Yes
7EFCH SMM Revision Identifier Field (Doubleword) No
7EF8H SMBASE Field (Doubleword) Yes
7EF7H - 7E00H Reserved No

NOTE:
* Upper two bytes are reserved.

The following registers are saved (but not readable) and restored upon exiting SMM:
• Control register CR4 (CR4 is set to “0” while in the SMM handler).
• The hidden segment descriptor information stored in segment registers CS, DS, ES, FS,
GS, and SS.
If an SMI request is issued for the purpose of powering down the processor, the values of all
reserved locations in the SMM state save must be saved to nonvolatile memory.
The following state is not automatically saved and restored following an SMI and the RSM
instruction, respectively:
• Debug registers DR0 through DR3.
• The FPU registers.
• The MTRRs.
• Control register CR2.
• The model-specific registers (for the P6 family and Pentium® processors) or test registers
TR3 through TR7 (for the Pentium® and Intel486™ processors).
12-6
 SYSTEM MANAGEMENT MODE (SMM)

• The state of the trap controller.

• The machine-check architecture registers.
• The APIC internal interrupt state (ISR, IRR, etc.).
• The microcode update state.
If an SMI is used to power down the processor, a power-on reset will be required before
returning to SMM, which will reset much of this state back to its default values. So an SMI
handler that is going to trigger power down should first read these registers listed above directly,
and save them (along with the rest of RAM) to nonvolatile storage. After the power-on reset, the
continuation of the SMI handler should restore these values, along with the rest of the system’s
state. Anytime the SMI handler changes these registers in the processor, it must also save and
restore them.

NOTE
A small subset of the MSRs (such as, the time-stamp counter and
performance-monitoring counter) are not arbitrarily writable and therefore
cannot be saved and restored. SMM-based power-down and restoration
should only be performed with operating systems that do not use or rely on
the values of these registers. Operating system developers should be aware of
this fact and ensure that their operating-system assisted power-down and
restoration software is immune to unexpected changes in these register
values.

12.4.2. SMRAM Caching

An Intel Architecture processor supporting SMM does not unconditionally write back and inval-
idate its cache before entering SMM. Therefore, if SMRAM is in a location that is “shadowed”
by any existing system memory that is visible to the application or operating system, then it is
necessary for the system to flush the cache upon entering SMM. This may be accomplished by
asserting the FLUSH# pin at the same time as the request to enter SMM. The priorities of the
FLUSH# pin and the SMI# are such that the FLUSH# will be serviced first. To guarantee this
behavior, the processor requires that the following constraints on the interaction of SMI# and
FLUSH# be met.
In a system where the FLUSH# pin and SMI# pins are synchronous and the set up and hold times
are met, then the FLUSH# and SMI# pins may be asserted in the same clock. In asynchronous
systems, the FLUSH# pin must be asserted at least one clock before the SMI# pin to guarantee
that the FLUSH# pin is serviced first. Note that in Pentium® processor systems that use the
FLUSH# pin to write back and invalidate cache contents before entering SMM, the processor
will prefetch at least one cache line in between when the Flush Acknowledge cycle is run, and
the subsequent recognition of SMI# and the assertion of SMIACT#. It is the obligation of the
system to ensure that these lines are not cached by returning KEN# inactive to the Pentium®
processor.

12-7
SYSTEM MANAGEMENT MODE (SMM)

Intel Architecture processors do not write back or invalidate their internal caches upon leaving
SMM. For this reason, references to the SMRAM area must not be cached if any part of the
SMRAM shadows (overlays) non-SMRAM memory; that is, system DRAM or video RAM. It
is the obligation of the system to ensure that all memory references to overlapped areas are
uncached; that is, the KEN# pin is sampled inactive during all references to the SMRAM area
for the Pentium® processor. The WBINVD instruction should be used to ensure cache coherency
at the end of a cached SMM execution in systems that have a protected SMM memory region
provided by the chipset.
The P6 family of processors have no external equivalent of the KEN# pin. All memory accesses
are typed via the MTRRs. It is not practical therefore to have memory access to a certain address
be cached in one access and not cached in another. Intel does not recommend the caching of
SMM space in any overlapping memory environment on the P6 family of processors.

12.5. SMI HANDLER EXECUTION ENVIRONMENT

After saving the current context of the processor, the processor initializes its core registers to the
values shown in Table 12-2. Upon entering SMM, the PE and PG flags in control register CR0
are cleared, which places the processor is in an environment similar to real-address mode. The
differences between the SMM execution environment and the real-address mode execution
environment are as follows:
• The addressable SMRAM address space ranges from 0 to FFFFFFFFH (4 GBytes). (The
physical address extension (enabled with the PAE flag in control register CR4) is not
supported in SMM.)
• The normal 64-KByte segment limit for real-address mode is increased to 4 GBytes.
• The default operand and address sizes are set to 16 bits, which restricts the addressable
SMRAM address space to the 1-MByte real-address mode limit for native real-address-
mode code. However, operand-size and address-size override prefixes can be used to
access the address space beyond the 1-MByte.
• Near jumps and calls can be made to anywhere in the 4-GByte address space if a 32-bit
operand-size override prefix is used. Due to the real-address-mode style of base-address
formation, a far call or jump cannot transfer control to a segment with a base address of
more than 20 bits (1 MByte). However, since the segment limit in SMM is 4 GBytes,
offsets into a segment that go beyond the 1-MByte limit are allowed when using 32-bit
operand-size override prefixes. Any program control transfer that does not have a 32-bit
operand-size override prefix truncates the EIP value to the 16 low-order bits.

12-8
 SYSTEM MANAGEMENT MODE (SMM)

Table 12-2. Processor Register Initialization in SMM

Register Contents
General-purpose registers Undefined
EFLAGS 00000002H
EIP 00008000H
CS selector SMM Base shifted right 4 bits (default 3000H)
CS base SMM Base (default 30000H)
DS, ES, FS, GS, SS Selectors 0000H
DS, ES, FS, GS, SS Bases 000000000H
DS, ES, FS, GS, SS Limits 0FFFFFFFFH
CR0 PE, EM, TS and PG flags set to 0; others unmodified
DR6 Undefined
DR7 00000400H

• Data and the stack can be located anywhere in the 4-GByte address space, but can be
accessed only with a 32-bit address-size override if they are located above 1 MByte. As
with the code segment, the base address for a data or stack segment cannot be more than 20
bits.
The value in segment register CS is automatically set to the default of 30000H for the SMBASE
shifted 4 bits to the right; that is, 3000H. The EIP register is set to 8000H. When the EIP value
is added to shifted CS value (the SMBASE), the resulting linear address points to the first
instruction of the SMI handler.
The other segment registers (DS, SS, ES, FS, and GS) are cleared to 0 and their segment limits
are set to 4 GBytes. In this state, the SMRAM address space may be treated as a single flat 4-
Gbyte linear address space. If a segment register is loaded with a 16-bit value, that value is then
shifted left by 4 bits and loaded into the segment base (hidden part of the segment register). The
limits and attributes are not modified.
Maskable hardware interrupts, exceptions, NMI interrupts, SMI interrupts, A20M interrupts,
single-step traps, breakpoint traps, and INIT operations are inhibited when the processor enters
SMM. Maskable hardware interrupts, exceptions, single-step traps, and breakpoint traps can be
enabled in SMM if the SMM execution environment provides and initializes an interrupt table
and the necessary interrupt and exception handlers (refer to Section 12.6., “Exceptions and
Interrupts Within SMM”).

12-9
SYSTEM MANAGEMENT MODE (SMM)

12.6. EXCEPTIONS AND INTERRUPTS WITHIN SMM

When the processor enters SMM, all hardware interrupts are disabled in the following manner:
• The IF flag in the EFLAGS register is cleared, which inhibits maskable hardware
interrupts from being generated.
• The TF flag in the EFLAGS register is cleared, which disables single-step traps
• Debug register DR7 is cleared, which disables breakpoint traps. (This action prevents a
debugger from accidentally breaking into an SMM handler if a debug breakpoint is set in
normal address space that overlays code or data in SMRAM.)
• NMI, SMI, and A20M interrupts are blocked by internal SMM logic. (Refer to Section
12.7., “NMI Handling While in SMM” for further information about how NMIs are
handled in SMM.)
Software-invoked interrupts and exceptions can still occur, and maskable hardware interrupts
can be enabled by setting the IF flag. Intel recommends that SMM code be written in so that it
does not invoke software interrupts (with the INT n, INTO, INT 3, or BOUND instructions) or
generate exceptions.
If the SMM handler requires interrupt and exception handling, an SMM interrupt table and the
necessary exception and interrupt handlers must be created and initialized from within SMM.
Until the interrupt table is correctly initialized (using the LIDT instruction), exceptions and soft-
ware interrupts will result in unpredictable processor behavior.
The following restrictions apply when designing SMM interrupt and exception-handling
facilities:
• The interrupt table should be located at linear address 0 and must contain real-address
mode style interrupt vectors (4 bytes containing CS and IP).
• Due to the real-address mode style of base address formation, an interrupt or exception
cannot transfer control to a segment with a base address of more that 20 bits.
• An interrupt or exception cannot transfer control to a segment offset of more than 16 bits
(64 KBytes).
• When an exception or interrupt occurs, only the 16 least-significant bits of the return
address (EIP) are pushed onto the stack. If the offset of the interrupted procedure is greater
than 64 KBytes, it is not possible for the interrupt/exception handler to return control to
that procedure. (One solution to this problem is for a handler to adjust the return address on
the stack.)
• The SMBASE relocation feature affects the way the processor will return from an interrupt
or exception generated while the SMI handler is executing. For example, if the SMBASE
is relocated to above 1 MByte, but the exception handlers are below 1 MByte, a normal
return to the SMI handler is not possible. One solution is to provide the exception handler
with a mechanism for calculating a return address above 1 MByte from the 16-bit return
address on the stack, then use a 32-bit far call to return to the interrupted procedure.

12-10
 SYSTEM MANAGEMENT MODE (SMM)

• If an SMI handler needs access to the debug trap facilities, it must insure that an SMM
accessible debug handler is available and save the current contents of debug registers DR0
through DR3 (for later restoration). Debug registers DR0 through DR3 and DR7 must then
be initialized with the appropriate values.
• If an SMI handler needs access to the single-step mechanism, it must insure that an SMM
accessible single-step handler is available, and then set the TF flag in the EFLAGS
register.
• If the SMI design requires the processor to respond to maskable hardware interrupts or
software-generated interrupts while in SMM, it must ensure that SMM accessible interrupt
handlers are available and then set the IF flag in the EFLAGS register (using the STI
instruction). Software interrupts are not blocked upon entry to SMM, so they do not need
to be enabled.

12.7. NMI HANDLING WHILE IN SMM

NMI interrupts are blocked upon entry to the SMI handler. If an NMI request occurs during the
SMI handler, it is latched and serviced after the processor exits SMM. Only one NMI request
will be latched during the SMI handler. If an NMI request is pending when the processor
executes the RSM instruction, the NMI is serviced before the next instruction of the interrupted
code sequence.
Although NMI requests are blocked when the CPU enters SMM, they may be enabled through
software by executing an IRET/IRETD instruction. If the SMM handler requires the use of NMI
interrupts, it should invoke a dummy interrupt service routine for the purpose of executing an
IRET/IRETD instruction. Once an IRET/IRETD instruciton is executed, NMI interrupt requr-
ests are serviced in the same “real mode” manner in which they are handled outside of SMM.
A special case can occur if an SMI handler nests inside an NMI handler and then another NMI
occurs. During NMI interrupt handling, NMI interrupts are disabled, so normally NMI inter-
rupts are serviced and completed with an IRET instruction one at a time. When the processor
enters SMM while executing an NMI handler, the processor saves the SMRAM state save map
but does not save the attribute to keep NMI interrupts disabled. Potentially, an NMI could be
latched (while in SMM or upon exit) and serviced upon exit of SMM even though the previous
NMI handler has still not completed. One or more NMIs could thus be nested inside the first
NMI handler. The NMI interrupt handler should take this possibility into consideration.
Also, for the Pentium® processor, exceptions that invoke a trap or fault handler will enable NMI
interrupts from inside of SMM. This behavior is implementation specific for the Pentium®
processor and is not part the Intel Architecture.

12.8. SAVING THE FPU STATE WHILE IN SMM

In some instances (for example prior to powering down system memory when entering a 0-volt
suspend state), it is necessary to save the state of the FPU while in SMM. Care should be taken
when performing this operation to insure that relevant FPU state information is not lost. The

12-11
SYSTEM MANAGEMENT MODE (SMM)

safest way to perform this task is to place the processor in 32-bit protected mode before saving
the FPU state. The reason for this is as follows.
The FSAVE instruction saves the FPU context in any of four different formats, depending on
which mode the processor is in when FSAVE is executed (refer to Figures 7-13 through 7-16 in
the Intel Architecture Software Developer’s Manual, Volume 1). When in SMM, by default, the
16-bit real-address mode format is used (shown in Figure 7-16). If an SMI interrupt occurs while
the processor is in a mode other than 16-bit real-address mode, FSAVE and FRSTOR will be
unable to save and restore all the relevant FPU information, and this situation may result in a
malfunction when the interrupted program is resumed. To avoid this problem, the processor
should be in 32-bit protected mode when executing the FSAVE and FRSTOR instructions.
The following guidelines should be used when going into protected mode from an SMI handler
to save and restore the FPU state:
• Use the CPUID instruction to insure that the processor contains an FPU.
• Create a 32-bit code segment in SMRAM space that contains procedures or routines to
save and restore the FPU using the FSAVE and FRSTOR instructions, respectively. A
GDT with an appropriate code-segment descriptor (D bit is set to 1) for the 32-bit code
segment must also be placed in SMRAM.
• Write a procedure or routine that can be called by the SMI handler to save and restore the
FPU state. This procedure should do the following:
— Place the processor in 32-bit protected mode as describe in Section 8.8.1., “Switching
to Protected Mode” in Chapter 8, Processor Management and Initialization.
— Execute a far JMP to the 32-bit code segment that contains the FPU save and restore
procedures.
— Place the processor back in 16-bit real-address mode before returning to the SMI
handler (refer to Section 8.8.2., “Switching Back to Real-Address Mode” in Chapter 8,
Processor Management and Initialization).
The SMI handler may continue to execute in protected mode after the FPU state has been saved
and return safely to the interrupted program from protected mode. However, it is recommended
that the handler execute primarily in 16- or 32-bit real-address mode.

12.9. SMM REVISION IDENTIFIER

The SMM revision identifier field is used to indicate the version of SMM and the SMM exten-
sions that are supported by the processor (refer to Figure 12-2). The SMM revision identifier
is written during SMM entry and can be examined in SMRAM space at offset 7EFCH. The
lower word of the SMM revision identifier refers to the version of the base SMM architecture.

12-12
 SYSTEM MANAGEMENT MODE (SMM)

Register Offset
7EFCH
31 18 17 16 15 0

Reserved SMM Revision Identifier

SMBASE Relocation
I/O Instruction Restart
Reserved

Figure 12-2. SMM Revision Identifier

The upper word of the SMM revision identifier refers to the extensions available. If the I/O
instruction restart flag (bit 16) is set, the processor supports the I/O instruction restart (refer to
Section 12.12., “I/O Instruction Restart”); if the SMBASE relocation flag (bit 17) is set,
SMRAM base address relocation is supported (refer to Section 12.11., “SMBASE Relocation”).

12.10. AUTO HALT RESTART

If the processor is in a HALT state (due to the prior execution of a HLT instruction) when it
receives an SMI, the processor records the fact in the auto HALT restart flag in the saved
processor state (refer to Figure 12-3). (This flag is located at offset 7F02H and bit 0 in the state
save area of the SMRAM.)
If the processor sets the auto HALT restart flag upon entering SMM (indicating that the SMI
occurred when the processor was in the HALT state), the SMI handler has two options:
• It can leave the auto HALT restart flag set, which instructs the RSM instruction to return
program control to the HLT instruction. This option in effect causes the processor to re-
enter the HALT state after handling the SMI. (This is the default operation.)
• It can clear the auto HALT restart flag, with instructs the RSM instruction to return
program control to the instruction following the HLT instruction.

15 1 0
Register Offset
7F02H

Reserved
Auto HALT Restart

Figure 12-3. Auto HALT Restart Field

12-13
SYSTEM MANAGEMENT MODE (SMM)

These options are summarized in Table 12-3. Note that if the processor was not in a HALT state
when the SMI was received (the auto HALT restart flag is cleared), setting the flag to 1 will
cause unpredictable behavior when the RSM instruction is executed.

Table 12-3. Auto HALT Restart Flag Values

Value of Flag After Entry Value of Flag When
to SMM Exiting SMM Action of Processor When Exiting SMM

0 0 Returns to next instruction in interrupted program

or task
Unpredictable
0 1
Returns to next instruction after HLT instruction
1 0
Returns to HALT state
1 1

If the HLT instruction is restarted, the processor will generate a memory access to fetch the HLT
instruction (if it is not in the internal cache), and execute a HLT bus transaction. This behavior
results in multiple HLT bus transactions for the same HLT instruction.

12.10.1. Executing the HLT Instruction in SMM

The HLT instruction should not be executed during SMM, unless interrupts have been enabled
by setting the IF flag in the EFLAGS register. If the processor is halted in SMM, the only event
that can remove the processor from this state is a maskable hardware interrupt or a hardware
reset.

12.11. SMBASE RELOCATION

The default base address for the SMRAM is 30000H. This value is contained in an internal
processor register called the SMBASE register. The operating system or executive can relocate
the SMRAM by setting the SMBASE field in the saved state map (at offset 7EF8H) to a new
value (refer to Figure 12-4). The RSM instruction reloads the internal SMBASE register with
the value in the SMBASE field each time it exits SMM. All subsequent SMI requests will use
the new SMBASE value to find the starting address for the SMI handler (at SMBASE + 8000H)
and the SMRAM state save area (from SMBASE + FE00H to SMBASE + FFFFH). (The
processor resets the value in its internal SMBASE register to 30000H on a RESET, but does not
change it on an INIT.) In multiple-processor systems, initialization software must adjust the

12-14
 SYSTEM MANAGEMENT MODE (SMM)

SMBASE value for each processor so that the SMRAM state save areas for each processor do
not overlap. (For Pentium® and Intel486™ processors, the SMBASE values must be aligned on
a 32-KByte boundary or the processor will enter shutdown state during the execution of a RSM
instruction.)

31 0
Register Offset
SMM Base
7EF8H

Figure 12-4. SMBASE Relocation Field

If the SMBASE relocation flag in the SMM revision identifier field is set, it indicates the ability
to relocate the SMBASE (refer to Section 12.9., “SMM Revision Identifier”).

12.11.1. Relocating SMRAM to an Address Above 1 MByte

In SMM, the segment base registers can only be updated by changing the value in the segment
registers. The segment registers contain only 16 bits, which allows only 20 bits to be used for a
segment base address (the segment register is shifted left 4 bits to determine the segment base
address). If SMRAM is relocated to an address above 1 MByte, software operating in real-
address mode can no longer initialize the segment registers to point to the SMRAM base address
(SMBASE).
The SMRAM can still be accessed by using 32-bit address-size override prefixes to generate an
offset to the correct address. For example, if the SMBASE has been relocated to FFFFFFH
(immediately below the 16-MByte boundary) and the DS, ES, FS, and GS registers are still
initialized to 0H, data in SMRAM can be accessed by using 32-bit displacement registers, as in
the following example:
mov esi,00FFxxxxH; 64K segment immediately below 16M
mov ax,ds:[esi]

A stack located above the 1-MByte boundary can be accessed in the same manner.

12.12. I/O INSTRUCTION RESTART

If the I/O instruction restart flag in the SMM revision identifier field is set (refer to Section 12.9.,
“SMM Revision Identifier”), the I/O instruction restart mechanism is present on the processor.
This mechanism allows an interrupted I/O instruction to be re-executed upon returning from
SMM mode. For example, if an I/O instruction is used to access a powered-down I/O device, a
chip set supporting this device can intercept the access and respond by asserting SMI#. This
action invokes the SMI handler to power-up the device. Upon returning from the SMI handler,
the I/O instruction restart mechanism can be used to re-execute the I/O instruction that caused
the SMI.

12-15
SYSTEM MANAGEMENT MODE (SMM)

The I/O instruction restart field (at offset 7F00H in the SMM state-save area, refer to Figure
12-5) controls I/O instruction restart. When an RSM instruction is executed, if this field contains
the value FFH, then the EIP register is modified to point to the I/O instruction that received the
SMI request. The processor will then automatically re-execute the I/O instruction that the SMI
trapped. (The processor saves the necessary machine state to insure that re-execution of the
instruction is handled coherently.)

15 0

I/O Instruction Restart Field Register Offset

7F00H

Figure 12-5. I/O Instruction Restart Field

If the I/O instruction restart field contains the value 00H when the RSM instruction is executed,
then the processor begins program execution with the instruction following the I/O instruction.
(When a repeat prefix is being used, the next instruction may be the next I/O instruction in the
repeat loop.) Not re-executing the interrupted I/O instruction is the default behavior; the
processor automatically initializes the I/O instruction restart field to 00H upon entering SMM.
Table 12-4 summarizes the states of the I/O instruction restart field.

Table 12-4. I/O Instruction Restart Field Values

Value of Flag After Value of Flag When
Entry to SMM Exiting SMM Action of Processor When Exiting SMM

00H 00H Does not re-execute trapped I/O instruction.

00H FFH Re-executes trapped I/O instruction.

Note that the I/O instruction restart mechanism does not indicate the cause of the SMI. It is the
responsibility of the SMI handler to examine the state of the processor to determine the cause of
the SMI and to determine if an I/O instruction was interrupted and should be restarted upon
exiting SMM. If an SMI interrupt is signaled on a non-I/O instruction boundary, setting the I/O
instruction restart field to FFH prior to executing the RSM instruction will likely result in a
program error.

12.12.1. Back-to-Back SMI Interrupts When I/O Instruction Restart

Is Being Used
If an SMI interrupt is signaled while the processor is servicing an SMI interrupt that occurred
on an I/O instruction boundary, the processor will service the new SMI request before restarting
the originally interrupted I/O instruction. If the I/O instruction restart field is set to FFH prior to
returning from the second SMI handler, the EIP will point to an address different from the orig-
inally interrupted I/O instruction, which will likely lead to a program error. To avoid this situa-
tion, the SMI handler must be able to recognize the occurrence of back-to-back SMI interrupts

12-16
 SYSTEM MANAGEMENT MODE (SMM)

when I/O instruction restart is being used and insure that the handler sets the I/O instruction
restart field to 00H prior to returning from the second invocation of the SMI handler.

12.13. SMM MULTIPLE-PROCESSOR CONSIDERATIONS

The following should be noted when designing multiple-processor systems:
• Any processor in a multiprocessor system can respond to an SMM.
• Each processor needs its own SMRAM space. This space can be in system memory or in a
separate RAM.
• The SMRAMs for different processors can be overlapped in the same memory space. The
only stipulation is that each processor needs its own state save area and its own dynamic
data storage area. (Also, for the Pentium® and Intel486™ processors, the SMBASE
address must be located on a 32-KByte boundary.) Code and static data can be shared
among processors. Overlapping SMRAM spaces can be done more efficiently with the P6
family processors because they do not require that the SMBASE address be on a 32-KByte
boundary.
• The SMI handler will need to initialize the SMBASE for each processor.
• Processors can respond to local SMIs through their SMI# pins or to SMIs received through
the APIC interface. The APIC interface can distribute SMIs to different processors.
• Two or more processors can be executing in SMM at the same time.
• When operating Pentium® processors in dual processing (DP) mode, the SMIACT# pin is
driven only by the MRM processor and should be sampled with ADS#. For additional
details, refer to Chapter 14 of the Pentium® Processor Family User’s Manual, Volume 1.
SMM is not re-entrant, because the SMRAM State Save Map is fixed relative to the SMBASE.
If there is a need to support two or more processors in SMM mode at the same time then each
processor should have dedicated SMRAM spaces. This can be done by using the SMBASE
Relocation feature (refer to Section 12.11., “SMBASE Relocation”).

12-17
SYSTEM MANAGEMENT MODE (SMM)

12-18
 13
Machine-Check
Architecture
 MACHINE-CHECK ARCHITECTURE

CHAPTER 13
MACHINE-CHECK ARCHITECTURE

This chapter describes the P6 family’s machine-check architecture and machine-check excep-
tion mechanism. Refer to Chapter 5, Interrupt and Exception Handling for more information on
the machine-check exception. A brief description of the Pentium® processor’s machine check
capability is also given.

13.1. MACHINE-CHECK EXCEPTIONS AND ARCHITECTURE

The P6 family of processors implement a machine-check architecture that provides a mecha-
nism for detecting and reporting hardware (machine) errors, such as system bus errors, ECC
errors, parity errors, cache errors, and TLB errors. It consists of a set of model-specific registers
(MSRs) that are used to set up machine checking and additional banks of MSRs for recording
the errors that are detected. The processor signals the detection of a machine-check error by
generating a machine-check exception (#MC). A machine-check exception is generally an abort
class exception. The implementation of the machine-check architecture, does not ordinarily
permit the processor to be restarted reliably after generating a machine-check exception;
however, the machine-check-exception handler can collect information about the machine-
check error from the machine-check MSRs.

13.2. COMPATIBILITY WITH PENTIUM® PROCESSOR

The P6 family processors support and extend the machine-check exception mechanism used in
the Pentium® processor. The Pentium® processor reports the following machine-check errors:
• Data parity errors during read cycles.
• Unsuccessful completion of a bus cycle.
These errors are reported through the P5_MC_TYPE and P5_MC_ADDR MSRs, which are
implementation specific for the Pentium® processor. These MSRs can be read with the RDMSR
instruction. Refer to Table B-1 in Appendix B, Model-Specific Registers for the register
addresses for these MSRs.
The machine-check error reporting mechanism that the Pentium ® processors use is similar to
that used in the P6 family processors. That is, when an error is detected, it is recorded in the
P5_MC_TYPE and P5_MC_ADDR MSRs and then the processor generates a machine-check
exception (#MC).
Refer to Section 13.3.3., “Mapping of the Pentium® Processor Machine-Check Errors to the P6
Family Machine-Check Architecture” and Section 13.7.2., “Pentium® Processor Machine-
Check Exception Handling” for information on compatibility between machine-check code
written to run on the Pentium® processors and code written to run on P6 family processors.

13-1
MACHINE-CHECK ARCHITECTURE

13.3. MACHINE-CHECK MSRS

The machine check MSRs in the P6 family processors consist of a set of global control and
status registers and several error-reporting register banks (refer to Figure 13-1). Each error-
reporting bank is associated with a specific hardware unit (or group of hardware units) within
the processor. The RDMSR and WRMSR instructions are used to read and write these
registers.

Global Control Registers Error-Reporting Bank Registers

(One Set for Each Hardware Unit)
63 0 63 0
MCG_CAP Register MCi_CTL Register

63 0 63 0
MCG_STATUS Register MCi_STATUS Register

63 0 63 0
MCG_CTL Register* MCi_ADDR Register

* Not present in the Pentium® Pro 63 0

processor.
MCi_MISC Register

Figure 13-1. Machine-Check MSRs

13.3.1. Machine-Check Global Control MSRs

The machine-check global control registers include the MCG_CAP, MCG_STATUS, and
MCG_CTL MSRs. Refer to Appendix B, Model-Specific Registers for the addresses of these
registers.

13.3.1.1. MCG_CAP MSR

The MCG_CAP MSR is a read-only register that provides information about the machine-check
architecture implementation in the processor (refer to Figure 13-2). It contains the following
field and flag:
Count field, bits 0 through 7
Indicates the number of hardware unit error-reporting banks available in a particular
processor implementation.
MCG_CTL_P (register present) flag, bit 8
Indicates that the MCG_CTL register is present when set, and absent when clear.
Bits 9 through 63 are reserved. The effect of writing to the MCG_CAP register is undefined.
Figure 5-1 shows the bit fields of MCG_CAP.

13-2
 MACHINE-CHECK ARCHITECTURE

63 9 8 7 0

Reserved Count

MCG_CTL_P—MCG_CTL register present

Count—Number of reporting banks

Figure 13-2. MCG_CAP Register

13.3.1.2. MCG_STATUS MSR

The MCG_STATUS MSR describes the current state of the processor after a machine-check
exception has occurred (refer to Figure 13-3). This register contains the following flags:
RIPV (restart IP valid) flag, bit 0
Indicates (when set) that program execution can be restarted reliably at the instruction
pointed to by the instruction pointer pushed on the stack when the machine-check
exception is generated. When clear, the program cannot be reliably restarted at the
pushed instruction pointer.
EIPV (error IP valid) flag, bit 1
Indicates (when set) that the instruction pointed to by the instruction pointer pushed
onto the stack when the machine-check exception is generated is directly associated
with the error. When this flag is cleared, the instruction pointed to may not be associ-
ated with the error.
MCIP (machine check in progress) flag, bit 2
Indicates (when set) that a machine-check exception was generated. Software can set
or clear this flag. The occurrence of a second Machine-Check Event while MCIP is set
will cause the processor to enter a shutdown state.
Bits 3 through 63 in the MCG_STATUS register are reserved.

63 3 2 1 0
M E R
C I I
Reserved I P P
P V V

MCIP—Machine check in progress flag

EIPV—Error IP valid flag
RIPV—Restart IP valid flag

Figure 13-3. MCG_STATUS Register

13-3
MACHINE-CHECK ARCHITECTURE

13.3.1.3. MCG_CTL MSR

The MCG_CTL register is present if the capability flag MCG_CTL_P is set in the MCG_CAP
register. The MCG_CTL register controls the reporting of machine-check exceptions. If present
(MCG_CTL_P flag in the MCG_CAP register is set), writing all 1s to this register enables all
machine-check features and writing all 0s disables all machine-check features. All other values
are undefined and/or implementation specific.

13.3.2. Error-Reporting Register Banks

Each error-reporting register bank can contains an MCi_CTL, MCi_STATUS, MCi_ADDR, and
MCi_MISC MSR. The P6 family processors provide five banks of error-reporting registers. The
first error-reporting register (MC0_CTL) always starts at address 400H. Refer to Table B-1 in
Appendix B, Model-Specific Registers for the addresses of the other error-reporting registers.

13.3.2.1. MCi_CTL MSR

The MCi_CTL MSR controls error reporting for specific errors produced by a particular hard-
ware unit (or group of hardware units). Each of the 64 flags (EEj) represents a potential error.
Setting an EEj flag enables reporting of the associated error and clearing it disables reporting of
the error. Writing the 64-bit value FFFFFFFFFFFFFFFFH to an MCi_CTL register enables
logging of all errors. The processor does not write changes to bits that are not implemented.
Figure 13-4 shows the bit fields of MCi_CTL

NOTE
Operating system or executive software must not modify the contents of the
MC0_CTL register. The MC0_CTL register is internally aliased to the
EBL_CR_POWERON register and as such controls system-specific error
handling features. These features are platform specific. System specific
firmware (the BIOS) is responsible for the appropriate initialization of
MC0_CTL. The P6 family processors only allows the writing of all 1s or all
0s to the MCi_CTL registers.

63 62 61 3 2 1 0
E
E
6
3
E
E
6
2
E
E
6
1
..... E
E
0
2
E E
E E
0 0
1 0

EEj—Error reporting enable flag

(where j is 00 through 63)

Figure 13-4. MCi_CTL Register

13-4
 MACHINE-CHECK ARCHITECTURE

13.3.2.2. MCi_STATUS MSR

The MCi_STATUS MSR contains information related to a machine-check error if its VAL
(valid) flag is set (refer to Figure 13-5). Software is responsible for clearing the MCi_STATUS
register by writing it with all 0s; writing 1s to this register will cause a general-protection excep-
tion to be generated. The flags and fields in this register are as follows:
MCA (machine-check architecture) error code field, bits 0 through 15
Specifies the machine-check architecture-defined error code for the machine-check
error condition detected. The machine-check architecture-defined error codes are
guaranteed to be the same for all Intel Architecture processors that implement the
machine-check architecture. Refer to Section 13.6., “Interpreting the MCA Error
Codes” for information on machine-check error codes.

63 62 6160 59 58 5756 32 31 16 15 0
V P Model-Specific
U E
A O C N C Other Information MCA Error Code
L C Error Code

PCC—Processor context corrupt

ADDRV—MCi_ADDR register valid
MISCV—MCi_MISC register valid
EN—Error enabled
UC—Uncorrected error
OVER—Error overflow
VAL—MCi_STATUS register valid

Figure 13-5. MCi_STATUS Register

Model-specific error code field, bits 16 through 31

Specifies the model-specific error code that uniquely identifies the machine-check
error condition detected. The model-specific error codes may differ among Intel Archi-
tecture processors for the same machine-check error condition.
Other information field, bits 32 through 56
The functions of the bits in this field are implementation specific and are not part of the
machine-check architecture. Software that is intended to be portable among Intel
Architecture processors should not rely on the values in this field.
PCC (processor context corrupt) flag, bit 57
Indicates (when set) that the state of the processor might have been corrupted by the
error condition detected and that reliable restarting of the processor may not be
possible. When clear, this flag indicates that the error did not affect the processor’s
state.
ADDRV (MCi_ADDR register valid) flag, bit 58
Indicates (when set) that the MCi_ADDR register contains the address where the error
occurred (refer to Section 13.3.2.3., “MCi_ADDR MSR”). When clear, this flag indi-
cates that the MCi_ADDR register does not contain the address where the error
occurred. Do not read these registers if they are not implemented in the processor.

13-5
MACHINE-CHECK ARCHITECTURE

MISCV (MCi_MISC register valid) flag, bit 59

Indicates (when set) that the MCi_MISC register contains additional information
regarding the error. When clear, this flag indicates that the MCi_MISC register does
not contain additional information regarding the error. Do not read these registers if
they are not implemented in the processor
EN (error enabled) flag, bit 60
Indicates (when set) that the error was enabled by the associated EEj bit of the
MCi_CTL register.
UC (error uncorrected) flag, bit 61
Indicates (when set) that the processor did not or was not able to correct the error condi-
tion. When clear, this flag indicates that the processor was able to correct the error
condition.
OVER (machine check overflow) flag, bit 62
Indicates (when set) that a machine-check error occurred while the results of a previous
error were still in the error-reporting register bank (that is, the VAL bit was already set
in the MCi_STATUS register). The processor sets the OVER flag and software is
responsible for clearing it. Enabled errors are written over disabled errors, and uncor-
rected errors are written over corrected errors. Uncorrected errors are not written over
previous valid uncorrected errors.
VAL (MCi_STATUS register valid) flag, bit 63
Indicates (when set) that the information within the MCi_STATUS register is valid.
When this flag is set, the processor follows the rules given for the OVER flag in the
MCi_STATUS register when overwriting previously valid entries. The processor sets
the VAL flag and software is responsible for clearing it.

13.3.2.3. MCi_ADDR MSR

The MCi_ADDR MSR contains the address of the code or data memory location that produced
the machine-check error if the ADDRV flag in the MCi_STATUS register is set (refer to Section
13.3.2.2., “MCi_STATUS MSR”). The address returned is either 32-bit offset into a segment,
32-bit linear address, or 36-bit physical address, depending upon the type of error encountered.
Bits 36 through 63 of this register are reserved for future address expansion and are always read
as zeros.

63 36 35 0

Reserved Address

Figure 13-6. Machine-Check Bank Address Register

13-6
 MACHINE-CHECK ARCHITECTURE

13.3.2.4. MCi_MISC MSR

The MCi_MISC MSR contains additional information describing the machine-check error if the
MISCV flag in the MCi_STATUS register is set. This register is not implemented in any of the
error-reporting register banks for the P6 family processors.

13.3.3. Mapping of the Pentium® Processor Machine-Check Errors

to the P6 Family Machine-Check Architecture
The Pentium® processor reports machine-check errors using two registers: P5_MC_TYPE and
P5_MC_ADDR. The P6 family processors map these registers into the MCi_STATUS and
MCi_ADDR registers of the error-reporting register bank that reports on the type of external bus
errors reported in the P5_MC_TYPE and P5_MC_ADDR registers. The information in these
registers can then be accessed in either of two ways:
• By reading the MCi_STATUS and MCi_ADDR registers as part of a generalized machine-
check exception handler written for a P6 family processor.
• By reading the P5_MC_TYPE and P5_MC_ADDR registers with the RDMSR instruction.
The second access capability permits a machine-check exception handler written to run on a
Pentium® processor to be run on a P6 family processor. There is a limitation in that information
returned by the P6 family processor will be encoded differently than it is for the Pentium®
processor. To run the Pentium® processor machine-check exception handler on a P6 family
processor, it must be rewritten to interpret the P5_MC_TYPE register encodings correctly.

13.4. MACHINE-CHECK AVAILABILITY

The machine-check architecture and machine-check exception (#MC) are model-specific
features. Software can execute the CPUID instruction to determine whether a processor imple-
ments these features. Following the execution of the CPUID instruction, the settings of the MCA
flag (bit 14) and MCE flag (bit 7) in the EDX register indicate whether the processor implements
the machine-check architecture and machine-check exception, respectively.

13.5. MACHINE-CHECK INITIALIZATION

To use the processors machine-check architecture, software must initialize the processor to acti-
vate the machine-check exception and the error-reporting mechanism. Example 13-1 gives
pseudocode for performing this initialization. This pseudocode checks for the existence of the
machine-check architecture and exception on the processor, then enables the machine-check
exception and the error-reporting register banks. The pseudocode assumes that the machine-
check exception (#MC) handler has been installed on the system. This initialization procedure
is compatible with the Pentium® and P6 family processors.

13-7
MACHINE-CHECK ARCHITECTURE

Example 13-1. Machine-Check Initialization Pseudocode

EXECUTE the CPUID instruction;
READ bits 7 (MCE) and 14 (MCA) of the EDX register;
IF CPU supports MCE
THEN
IF CPU supports MCA
THEN
IF MCG_CAP.MCG_CTL_P = 1 (* MCG_CTL register is present *)
Set MCG_CTL register to all 1s; (* enables all MCA features *)
FI;
COUNT ← MCG_CAP.Count;
(* determine number of error-reporting banks supported *)
FOR error-reporting banks (1 through COUNT) DO
Set MCi_CTL register to all 1s;
(* enables logging of all errors except for the MC0_CTL register *)
OD
FOR error-reporting banks (0 through COUNT) DO
Set MCi_STATUS register to all 0s; (* clears all errors *)
OD
FI;
Set the MCE flag (bit 6) in CR4 register to enable machine-check exceptions;
FI;
The processor can write valid information (such as an ECC error) into the MCi_STATUS regis-
ters while it is being powered up. As part of the initialization of the MCE exception handler, soft-
ware might examine all the MCi_STATUS registers and log the contents of them, then rewrite
them all to zeros. This procedure is not included in the initialization pseudocode in Example
13-1.

13.6. INTERPRETING THE MCA ERROR CODES

When the processor detects a machine-check error condition, it writes a 16-bit error code in the
MCA Error Code field of one of the MCi_STATUS registers and sets the VAL (valid) flag in
that register. The processor may also write a 16-bit Model-specific Error Code in the
MCi_STATUS register depending on the implementation of the machine-check architecture of
the processor.
The MCA error codes are architecturally defined for Intel Architecture processors; however, the
specific MCi_STATUS register that a code is written into is model specific. To determine the
cause of a machine-check exception, the machine-check exception handler must read the VAL
flag for each MCi_STATUS register, and, if the flag is set, then read the MCA error code field
of the register. It is the encoding of the MCACOD value that determines the type of error being
reported and not the register bank reporting it.
There are two types of MCA error codes: simple error codes and compound error codes.

13-8
 MACHINE-CHECK ARCHITECTURE

13.6.1. Simple Error Codes

Table 13-1 shows the simple error codes. These unique codes indicate global error information.

Table 13-1. Simple Error Codes

Error Code Binary Encoding Meaning
No Error 0000 0000 0000 0000 No error has been reported to this bank of
error-reporting registers.
Unclassified 0000 0000 0000 0001 This error has not been classified into the
MCA error classes.
Microcode ROM Parity 0000 0000 0000 0010 Parity error in internal microcode ROM
Error
External Error 0000 0000 0000 0011 The BINIT# from another processor caused
this processor to enter machine check.
FRC Error 0000 0000 0000 0100 FRC (functional redundancy check)
master/slave error
Internal Unclassified 0000 01xx xxxx xxxx Internal unclassified errors

13.6.2. Compound Error Codes

The compound error codes describe errors related to the TLBs, memory, caches, bus and inter-
connect logic. A set of sub-fields is common to all of the compound error encodings. These sub-
fields describe the type of access, level in the memory hierarchy, and type of request. Table 13-2
shows the general form of the compound error codes. The interpretation column indicates the
name of a compound error. The name is constructed by substituting mnemonics from Tables
13-2 through 13-6 for the sub-field names given within curly braces. For example, the error code
ICACHEL1_RD_ERR is constructed from the form:
{TT}CACHE{LL}_{RRRR}_ERR
where {TT} is replaced by I, {LL} is replaced by L1, and {RRRR} is replaced by RD.
The 2-bit TT sub-field (refer to Table 13-2) indicates the type of transaction (data, instruction,
or generic). It applies to the TLB, cache, and interconnect error conditions. The generic type is
reported when the processor cannot determine the transaction type.

Table 13-2. General Forms of Compound Error Codes

Type Form Interpretation
TLB Errors 0000 0000 0001 TTLL {TT}TLB{LL}_ERR
Memory Hierarchy Errors 0000 0001 RRRR TTLL {TT}CACHE{LL}_{RRRR}_ERR
Bus and Interconnect 0000 1PPT RRRR IILL BUS{LL}_{PP}_{RRRR}_{II}_{T}_ERR
Errors

13-9
MACHINE-CHECK ARCHITECTURE

Table 13-3. Encoding for TT (Transaction Type) Sub-Field

Transaction Type Mnemonic Binary Encoding
Instruction I 00
Data D 01
Generic G 10

The 2-bit LL sub-field (refer to Table 13-4) indicates the level in the memory hierarchy where
the error occurred (level 0, level 1, level 2, or generic). The LL sub-field also applies to the TLB,
cache, and interconnect error conditions. The P6 family processors support two levels in the
cache hierarchy and one level in the TLBs. Again, the generic type is reported when the
processor cannot determine the hierarchy level.

Table 13-4. Level Encoding for LL (Memory Hierarchy Level) Sub-Field

Hierarchy Level Mnemonic Binary Encoding
Level 0 L0 00
Level 1 L1 01
Level 2 L2 10
Generic LG 11

The 4-bit RRRR sub-field (refer to Table 13-5) indicates the type of action associated with the
error. Actions include read and write operations, prefetches, cache evictions, and snoops.
Generic error is returned when the type of error cannot be determined. Generic read and generic
write are returned when the processor cannot determine the type of instruction or data request
that caused the error. Eviction and Snoop requests apply only to the caches. All of the other
requests apply to TLBs, caches and interconnects.

Table 13-5. Encoding of Request (RRRR) Sub-Field

Request Type Mnemonic Binary Encoding
Generic Error ERR 0000
Generic Read RD 0001
Generic Write WR 0010
Data Read DRD 0011
Data Write DWR 0100
Instruction Fetch IRD 0101
Prefetch PREFETCH 0110
Eviction EVICT 0111
Snoop SNOOP 1000

13-10
 MACHINE-CHECK ARCHITECTURE

The bus and interconnect errors are defined with the 2-bit PP (participation), 1-bit T (time-out),
and 2-bit II (memory or I/O) sub-fields, in addition to the LL and RRRR sub-fields (refer to
Table 13-6). The bus error conditions are implementation dependent and related to the type of
bus implemented by the processor. Likewise, the interconnect error conditions are predicated on
a specific implementation-dependent interconnect model that describes the connections
between the different levels of the storage hierarchy. The type of bus is implementation depen-
dent, and as such is not specified in this document. A bus or interconnect transaction consists of
a request involving an address and a response.

Table 13-6. Encodings of PP, T, and II Sub-Fields

Binary
Sub-Field Transaction Mnemonic Encoding
PP (Participation) Local processor originated request SRC 00
Local processor responded to request RES 01
Local processor observed error as third party OBS 10
Generic 11
T (Time-out) Request timed out TIMEOUT 1
Request did not time out NOTIMEOUT 0
II (Memory or I/O) Memory Access M 00
Reserved 01
I/O IO 10
Other transaction 11

13.6.3. Interpreting the Machine-Check Error Codes for External

Bus Errors
Table 13-7 gives additional information for interpreting the MCA error code, model-specific
error code, and other information error code fields for machine-check errors that occur on the
external bus. This information can be used to design a machine-check exception handler for the
processor that offers greater granularity for the external bus errors.

Table 13-7. Encoding of the MCi_STATUS Register for External Bus Errors
Bit
No. Bit Function Bit Description
0-1 MCA Error Undefined.
Code
2-3 MCA Error Bit 2 is set to 1 if the access was a special cycle.
Code Bit 3 is set to 1 if the access was a special cycle OR a I/O cycle.
4-7 MCA Error 00WR; W = 1 for writes, R = 1 for reads.
Code

13-11
MACHINE-CHECK ARCHITECTURE

Table 13-7. Encoding of the MCi_STATUS Register for External Bus Errors (Contd.)
Bit
No. Bit Function Bit Description
8-9 MCA Error Undefined.
Code
10 MCA Error Set to 0 for all EBL errors.
Code Set to 1 for internal watch-dog timer time-out.
For a watch-dog timer time-out, all the MCACOD bits except this bit are set to
0. A watch-dog timer time-out only occurs if the BINIT driver is enabled.
11 MCA Error Set to 1 for EBL errors.
Code Set to 0 for internal watch-dog timer time-out.
12-15 MCA Error Reserved.
Code
16-18 Model- Reserved.
Specific Error
Code
19-24 Model- 000000 for BQ_DCU_READ_TYPE error.
Specific Error 000010 for BQ_IFU_DEMAND_TYPE error.
Code 000011 for BQ_IFU_DEMAND_NC_TYPE error.
000100 for BQ_DCU_RFO_TYPE error.
000101 for BQ_DCU_RFO_LOCK_TYPE error.
000110 for BQ_DCU_ITOM_TYPE error.
001000 for BQ_DCU_WB_TYPE error.
001010 for BQ_DCU_WCEVICT_TYPE error.
001011 for BQ_DCU_WCLINE_TYPE error.
001100 for BQ_DCU_BTM_TYPE error.
001101 for BQ_DCU_INTACK_TYPE error.
001110 for BQ_DCU_INVALL2_TYPE error.
001111 for BQ_DCU_FLUSHL2_TYPE error.
010000 for BQ_DCU_PART_RD_TYPE error.
010010 for BQ_DCU_PART_WR_TYPE error.
010100 for BQ_DCU_SPEC_CYC_TYPE error.
011000 for BQ_DCU_IO_RD_TYPE error.
011001 for BQ_DCU_IO_WR_TYPE error.
011100 for BQ_DCU_LOCK_RD_TYPE error.
011110 for BQ_DCU_SPLOCK_RD_TYPE error.
011101 for BQ_DCU_LOCK_WR_TYPE error.
27-25 Model- 000 for BQ_ERR_HARD_TYPE error.
Specific Error 001 for BQ_ERR_DOUBLE_TYPE error.
Code 010 for BQ_ERR_AERR2_TYPE error.
100 for BQ_ERR_SINGLE_TYPE error.
101 for BQ_ERR_AERR1_TYPE error.
28 Model- 1 if FRC error is active.
Specific Error
Code
29 Model- 1 if BERR is driven.
Specific Error
Code

13-12
 MACHINE-CHECK ARCHITECTURE

Table 13-7. Encoding of the MCi_STATUS Register for External Bus Errors (Contd.)
Bit
No. Bit Function Bit Description
30 Model- 1 if BINIT is driven for this processor.
Specific Error
Code
31 Model- Reserved.
Specific Error
Code
32-34 Other Reserved.
Information
35 Other 1 if BINIT is received from external bus.
Information
BINIT
36 Other This bit is asserted in the MCi_STATUS register if this component has received
Information a parity error on the RS[2:0]# pins for a response transaction. The RS signals
RESPONSE are checked by the RSP# external pin.
PARITY
ERROR
37 Other This bit is asserted in the MCi_STATUS register if this component has received
Information a hard error response on a split transaction (one access that has needed to be
BUS BINIT split across the 64-bit external bus interface into two accesses).
38 Other This bit is asserted in the MCi_STATUS register if this component has
Information experienced a ROB time-out, which indicates that no microinstruction has been
TIMEOUT retired for a predetermined period of time. A ROB time-out occurs when the 15-
BINIT bit ROB time-out counter carries a 1 out of its high order bit.

The timer is cleared when a microinstruction retires, an exception is detected

by the core processor, RESET is asserted, or when a ROB BINIT occurs.

The ROB time-out counter is prescaled by the 8-bit PIC timer which is a divide
by 128 of the bus clock (the bus clock is 1:2, 1:3, 1:4 the core clock). When a
carry out of the 8-bit PIC timer occurs, the ROB counter counts up by one.

While this bit is asserted, it cannot be overwritten by another error.

39-41 Other Reserved.
Information
42 Other This bit is asserted in the MCi_STATUS register if this component has initiated
Information a bus transactions which has received a hard error response. While this bit is
HARD asserted, it cannot be overwritten.
ERROR
43 Other This bit is asserted in the MCi_STATUS register if this component has
Information experienced a failure that causes the IERR pin to be asserted. While this bit is
IERR asserted, it cannot be overwritten.
44 Other This bit is asserted in the MCi_STATUS register if this component has initiated
Information 2 failing bus transactions which have failed due to Address Parity Errors (AERR
AERR asserted). While this bit is asserted, it cannot be overwritten.

13-13
MACHINE-CHECK ARCHITECTURE

Table 13-7. Encoding of the MCi_STATUS Register for External Bus Errors (Contd.)
Bit
No. Bit Function Bit Description
45 Other Uncorrectable ECC error bit is asserted in the MCi_STATUS register for
Information uncorrected ECC errors. While this bit is asserted, the ECC syndrome field will
UECC not be overwritten.
46 Other The correctable ECC error bit is asserted in the MCi_STATUS register for
Information corrected ECC errors.
CECC
47-54 Other The ECC syndrome field in the MCi_STATUS register contains the 8-bit ECC
Information syndrome only if the error was a correctable/uncorrectable ECC error,
SYNDROME and there wasn’t a previous valid ECC error syndrome logged in the
MCi_STATUS register.
A previous valid ECC error in MCi_STATUS is indicated by MCi_STATUS.bit45
(uncorrectable error occurred) being asserted. After processing an ECC error,
machine-check handling software should clear MCi_STATUS.bit45 so that
future ECC error syndromes can be logged.
55-56 Other Reserved.
Information

13.7. GUIDELINES FOR WRITING MACHINE-CHECK SOFTWARE

The machine-check architecture and error logging can be used in two different ways:
• To detect machine errors during normal instruction execution, using the machine-check
exception (#MC).
• To periodically check and log machine errors.
To use the machine-check exception, the operating system or executive software must provide
a machine-check exception handler. This handler can be designed specifically for P6 family
processors or be a portable handler that also handles Pentium® processor machine-check errors.
A special program or utility is required to log machine errors.
Guidelines for writing a machine-check exception handler or a machine-error logging utility are
given in the following sections.

13.7.1. Machine-Check Exception Handler

The machine-check exception (#MC) corresponds to vector 18. To service machine-check
exceptions, a trap gate must be added to the IDT, and the pointer in the trap gate must point to a
machine-check exception handler. Two approaches can be taken to designing the exception
handler:
• The handler can merely log all the machine status and error information, then call a
debugger or shut down the system.
• The handler can analyze the reported error information and, in some cases, attempt to
correct the error and restart the processor.

13-14
 MACHINE-CHECK ARCHITECTURE

Virtually all the machine-check conditions detected with the P6 family processors cannot be
recovered from (they result in abort-type exceptions). The logging of status and error informa-
tion is therefore a baseline implementation. Refer to Section 13.7., “Guidelines for Writing
Machine-Check Software” for more information on logging errors.
For future P6 family processor implementations, where recovery may be possible, the following
things should be considered when writing a machine-check exception handler:
• To determine the nature of the error, the handler must read each of the error-reporting
register banks. The count field in the MCG_CAP register gives number of register banks.
The first register of register bank 0 is at address 400H.
• The VAL (valid) flag in each MCi_STATUS register indicates whether the error
information in the register is valid. If this flag is clear, the registers in that bank do not
contain valid error information and do not need to be checked.
• To write a portable exception handler, only the MCA error code field in the MCi_STATUS
register should be checked. Refer to Section 13.6., “Interpreting the MCA Error Codes” for
information that can be used to write an algorithm to interpret this field.
• The RIPV, PCC, and OVER flags in each MCi_STATUS register indicate whether
recovery from the error is possible. If either of these fields is set, recovery is not possible.
The OVER field indicates that two or more machine-check error occurred. When recovery
is not possible, the handler typically records the error information and signals an abort to
the operating system.
• Corrected errors will have been corrected automatically by the processor. The UC flag in
each MCi_STATUS register indicates whether the processor automatically corrected the
error.
• The RIPV flag in the MCG_STATUS register indicates whether the program can be
restarted at the instruction pointed to by the instruction pointer pushed on the stack when
the exception was generated. If this flag is clear, the processor may still be able to be
restarted (for debugging purposes), but not without loss of program continuity.
• For unrecoverable errors, the EIPV flag in the MCG_STATUS register indicates whether
the instruction pointed to by the instruction pointer pushed on the stack when the exception
was generated is related to the error. If this flag is clear, the pushed instruction may not be
related to the error.
• The MCIP flag in the MCG_STATUS register indicates whether a machine-check
exception was generated. Before returning from the machine-check exception handler,
software should clear this flag so that it can be used reliably by an error logging utility. The
MCIP flag also detects recursion. The machine-check architecture does not support
recursion. When the processor detects machine-check recursion, it enters the shutdown
state.

13-15
MACHINE-CHECK ARCHITECTURE

Example 13-2 gives typical steps carried out by a machine-check exception handler:

Example 13-2. Machine-Check Exception Handler Pseudocode

IF CPU supports MCE
THEN
IF CPU supports MCA
THEN
call errorlogging routine; (* returns restartability *)
FI;
ELSE (* Pentium(R) processor compatible *)
READ P5_MC_ADDR
READ P5_MC_TYPE;
report RESTARTABILITY to console;
FI;
IF error is not restartable
THEN
report RESTARTABILITY to console;
abort system;
FI;
CLEAR MCIP flag in MCG_STATUS;

13.7.2. Pentium® Processor Machine-Check Exception Handling

To make the machine-check exception handler portable to the Pentium® and P6 family proces-
sors, checks can be made (using the CPUID instruction) to determine the processor type. Then
based on the processor type, machine-check exceptions can be handled specifically for Pentium®
or P6 family processors.
When machine-check exceptions are enabled for the Pentium® processor (MCE flag is set in
control register CR0), the machine-check exception handler uses the RDMSR instruction to read
the error type from the P5_MC_TYPE register and the machine check address from the
P5_MC_ADDR register. The handler then normally reports these register values to the system
console before aborting execution (refer to Example 13-2).

13.7.3. Logging Correctable Machine-Check Errors

If a machine-check error is correctable, the processor does not generate a machine-check excep-
tion for it. To detect correctable machine-check errors, a utility program must be written that
reads each of the machine-check error-reporting register banks and logs the results in an
accounting file or data structure. This utility can be implemented in either of the following ways:
• A system daemon that polls the register banks on an infrequent basis, such as hourly or
daily.

13-16
 MACHINE-CHECK ARCHITECTURE

• A user-initiated application that polls the register banks and records the exceptions. Here,
the actual polling service is provided by an operating-system driver or through the system
call interface.
Example 13-3 gives pseudocode for an error logging utility.

Example 13-3. Machine-Check Error Logging Pseudocode

Assume that execution is restartable;
IF the processor supports MCA
THEN
FOR each bank of machine-check registers
DO
READ MCi_STATUS;
IF VAL flag in MCi_STATUS = 1
THEN
IF ADDRV flag in MCi_STATUS = 1
THEN READ MCi_ADDR;
FI;
IF MISCV flag in MCi_STATUS = 1
THEN READ MCi_MISC;
FI;
IF MCIP flag in MCG_STATUS = 1
(* Machine-check exception is in progress *)
AND PCC flag in MCi_STATUS = 1
AND RIPV flag in MCG_STATUS = 0
(* execution is not restartable *)
THEN
RESTARTABILITY = FALSE;
return RESTARTABILITY to calling procedure;
FI;
Save time-stamp counter and processor ID;
Set MCi_STATUS to all 0s;
Execute serializing instruction (i.e., CPUID);
FI;
OD;
FI;

If the processor supports the machine-check architecture, the utility reads through the banks of
error-reporting registers looking for valid register entries, and then saves the values of the
MCi_STATUS, MCi_ADDR, MCi_MISC and MCG_STATUS registers for each bank that is
valid. The routine minimizes processing time by recording the raw data into a system data struc-
ture or file, reducing the overhead associated with polling. User utilities analyze the collected
data in an off-line environment.
When the MCIP flag is set in the MCG_STATUS register, a machine-check exception is in
progress and the machine-check exception handler has called the exception logging routine.
Once the logging process has been completed the exception-handling routine must determine

13-17
MACHINE-CHECK ARCHITECTURE

whether execution can be restarted, which is usually possible when damage has not occurred
(The PCC flag is clear, in the MCi_STATUS register) and when the processor can guarantee that
execution is restartable (the RIPV flag is set in the MCG_STATUS register). If execution cannot
be restarted, the system is not recoverable and the exception-handling routine should signal the
console appropriately before returning the error status to the Operating System kernel for subse-
quent shutdown.
The machine-check architecture allows buffering of exceptions from a given error-reporting
bank although the P6 family processors do not implement this feature. The error logging routine
should provide compatibility with future processors by reading each hardware error-reporting
bank’s MCi_STATUS register and then writing 0s to clear the OVER and VAL flags in this
register. The error logging utility should re-read the MCi_STATUS register for the bank
ensuring that the valid bit is clear. The processor will write the next error into the register bank
and set the VAL flags.
Additional information that should be stored by the exception-logging routine includes the
processor’s time-stamp counter value, which provides a mechanism to indicate the frequency of
exceptions. A multiprocessing operating system stores the identity of the processor node incur-
ring the exception using a unique identifier, such as the processor’s APIC ID (refer to Section
7.5.9., “Interrupt Destination and APIC ID”).
The basic algorithm given in Example 13-3 can be modified to provide more robust recovery
techniques. For example, software has the flexibility to attempt recovery using information
unavailable to the hardware. Specifically, the machine-check exception handler can, after
logging carefully analyze the error-reporting registers when the error-logging routine reports an
error that does not allow execution to be restarted. These recovery techniques can use external
bus related model-specific information provided with the error report to localize the source of
the error within the system and determine the appropriate recovery strategy.

13-18
 14
Code Optimization
 CHAPTER 14
CODE OPTIMIZATION

This chapter describes the more important code optimization techniques for Intel Architecture
processors with and without MMX™ technology, as well as with and without Streaming SIMD
Extensions. The chapter begins with general code-optimization guidelines and continues with a
brief overview of the more important blended techniques for optimizing integer, MMX™ tech-
nology, floating-point, and SIMD floating-point code. A comprehensive discussion of code opti-
mization techniques can be found in the Intel Architecture Optimization Manual, Order Number
242816.

14.1. CODE OPTIMIZATION GUIDELINES

This section contains general guidelines for optimizing applications code, as well as specific
guidelines for optimizing MMX™, floating-point, and SIMD floating-point code. Developers
creating applications that use MMX™ and/or floating-point instructions should apply the first
set of guidelines in addition to the MMX™ and/or floating-point code optimization guidelines.
Developers creating applications that use SIMD floating-point code should apply the first set of
guidelines, as well as the MMX™ and/or floating-point code optimization guidelines, in addi-
tion to the SIMD floating-point code optimization guidelines.

14.1.1. General Code Optimization Guidelines

Use the following guidelines to optimize code to run efficiently across several families of Intel
Architecture processors:
• Use a current generation compiler that produces optimized code to insure that efficient
code is generated from the start of code development.
• Write code that can be optimized by the compiler. For example:
— Minimize the use of global variables, pointers, and complex control flow statements.
— Do not use the “register” modifier.
— Use the “const” modifier.
— Do not defeat the typing system.
— Do not make indirect calls.
— Use minimum sizes for integer and floating-point data types, to enable SIMD paral-
lelism.

14-1
CODE OPTIMIZATION

• Pay attention to the branch prediction algorithm for the target processor. This optimization
is particularly important for P6 family processors. Code that optimizes branch predict-
ability will spend fewer clocks fetching instructions.
• Take advantage of the SIMD capabilities of MMX™ technology and Streaming SIMD
Extensions.
• Avoid partial register stalls.
• Align all data.
• Organize code to minimize instruction cache misses and optimize instruction prefetches.
• Schedule code to maximize pairing on Pentium® processors.
• Avoid prefixed opcodes other than 0FH.
• When possible, load and store data to the same area of memory using the same data sizes
and address alignments; that is, avoid small loads after large stores to the same area of
memory, and avoid large loads after small stores to the same area of memory.
• Use software pipelining.
• Always pair CALL and RET (return) instructions.
• Avoid self-modifying code.
• Do not place data in the code segment.
• Calculate store addresses as soon as possible.
• Avoid instructions that contain 4 or more micro-ops or instructions that are more than 7
bytes long. If possible, use instructions that require 1 micro-op.
• Cleanse partial registers before calling callee-save procedures.

14.1.2. Guidelines for Optimizing MMX™ Code

Use the following guidelines to optimize MMX™ code:
• Do not intermix MMX™ instructions and floating-point instructions.
• Use the opcode reg, mem instruction format whenever possible. This format helps to free
registers and reduce clocks without generating unnecessary loads.
• Put an EMMS instruction at the end of all MMX™ code sections that you know will
transition to floating-point code.
• Optimize data cache bandwidth to MMX™ registers.

14.1.3. Guidelines for Optimizing Floating-Point Code

Use the following guidelines to optimize floating-point code:

14-2
 CODE OPTIMIZATION

• Understand how the compiler handles floating-point code. Look at the assembly dump and
see what transforms are already performed on the program. Study the loop nests in the
application that dominate the execution time.
• Determine why the compiler is not creating the fastest code. For example, look for
dependences that can be resolved by rearranging code
• Look for and correct situations known to cause slow execution of floating-point code, such
as:
— Large memory bandwidth requirements.
— Poor cache locality.
— Long-latency floating-point arithmetic operations.
• Do not use more precision than is necessary. Single precision (32-bits) is faster on some
operations and consumes only half the memory space as double precision (64-bits) or
double extended (80-bits).
• Use a library that provides fast floating-point to integer routines. Many library routines do
more work than is necessary.
• Insure whenever possible that computations stay in range. Out of range numbers cause
very high overhead.
• Schedule code in assembly language using the FXCH instruction. When possible, unroll
loops and pipeline code.
• Perform transformations to improve memory access patterns. Use loop fusion or
compression to keep as much of the computation in the cache as possible.
• Break dependency chains.

14.1.4. Guidelines for Optimizing SIMD Floating-point Code

Generally, it is important to understand and balance port utilization to create efficient SIMD
floating-point code. Use the following guidelines to optimize SIMD floating-point code:
• Balance the limitations of the architecture.
• Schedule instructions to resolve dependencies.
• Schedule utilization of the triple/quadruple rule (port 0, port 1, port 2, 3, and 4).
• Group instructions that utilize the same registers as closely as possible. Take into consider-
ation the resolution of true dependencies.
• Intermix SIMD-fp operations that utilize port 0 and port 1.
• Do not issue consecutive instructions that utilize the same port.
• Use the reciprocal instructions followed by iteration for increased accuracy. These instruc-
tions yield reduced accuracy but execute much faster. If reduced accuracy is acceptable,

14-3
CODE OPTIMIZATION

use them with no iteration. If near full accuracy is needed, use a Newton-Raphson
iteration. If full accuracy is needed, then use divide and square root which provide more
accuracy, but slow down performance.
• Exceptions: mask exceptions to achieve higher performance. Unmasked exceptions may
cause a reduction in the retirement rate.
• Utilize the flush-to-zero mode for higher performance to avoid the penalty of dealing with
denormals and underflows.
• Incorporate the prefetch instruction whenever possible (for details, refer to Chapter 6,
“Optimizing Cache Utilization for Pentium® III processors” ).
• Try to emulate conditional moves by masked compares and logicals instead of using
conditional jumps.
• Utilize MMX™ technology instructions if the computations can be done in SIMD-integer
or for shuffling data or copying data that is not used later in SIMD floating-point computa-
tions.
• If the algorithm requires extended precision, then conversion to SIMD floating-point code
is not advised because the SIMD floating-point instructions are single-precision.

14.2. BRANCH PREDICTION OPTIMIZATION

The P6 family and Pentium® processors provide dynamic branch prediction using the branch
target buffers (BTBs) on the processors. Understanding the flow of branches and improving the
predictability of branches can increase code execution speed significantly.

14.2.1. Branch Prediction Rules

Three elements of dynamic branch prediction are important to understand:
• If the instruction address is not in the BTB, execution is predicted to continue without
branching (fall through).
• Predicted taken branches have a 1 clock delay.
• The BTB stores a four-bit history of branch predictions on Pentium® Pro processors, the
Pentium® II processor family, and the Pentium® III processor.
• The Pentium® II and Pentium® III processor’s BTB pattern matches on the direction of the
last four branches to dynamically predict whether a branch will be taken.
During the process of instruction prefetch, the instruction address of a conditional instruction is
checked with the entries in the BTB. When the address is not in the BTB, execution is predicted
to fall through to the next instruction.
On P6 family processors, branches that do not have a history in the BTB are predicted using a
static prediction algorithm. The static prediction algorithm does the following:

14-4
 CODE OPTIMIZATION

• Predicts unconditional branches to be taken.

• Predicts backward conditional branches to be taken. This rule is suitable for loops.
• Predicts forward conditional branches to be not taken.

14.2.2. Optimizing Branch Predictions in Code

To optimize branch predictions in an application code, apply the following techniques:
• Reduce or eliminate branches (see Section 14.2.3., “Eliminating and Reducing the Number
of Branches”).
• Insure that each CALL instruction has a matching RET instruction. The P6 family of
processors have a return stack buffer that keeps track of the target address of the next RET
instruction. Do not use pops and jumps to return from a CALL instruction; always use the
RET instruction.
• Do not intermingle data with instructions in a code segment. Unconditional jumps, when
not in the BTB, are predicted to be not taken. If data follows a unconditional branch, the
data might be fetched, causing the loss of instruction fetch cycles and valuable instruction-
cache space. When data must be stored in the code segment, move it to the end where it
will not be in the instruction fetch stream.
• Unroll all very short loops. Loops that execute for less than 2 clocks waste loop overhead.
• Write code to follow the static prediction algorithm. The static prediction algorithm
follows the natural flow of program code. Following this algorithm reduces the number of
branch mispredictions.

14.2.3. Eliminating and Reducing the Number of Branches

Eliminating branches improves processor performance by:
• Removing the possibility of branch mispredictions.
• Reducing the number of BTB entries required.
Branches can be eliminated by using the SETcc instruction, or by using the P6 family proces-
sors’ conditional move (CMOVcc or FCMOVcc) instructions.
The following C code example shows conditions that are dependent upon on of the constants A
and B:
/* C Code /*
ebx = (A < B) ? C1 : C2;
This code conditionally compares the values A and B. If the condition is true, EBX is set to C1;
otherwise it is set to C2. The assembly-language equivalent of the C code is shown in the
example below:
; Assembly Code

14-5
CODE OPTIMIZATION

cmp A, B ; condition
jge L30 ; conditional branch
mov ebx, CONST1
jmp L31 ; unconditional branch
L30:
mov ebx, CONST2
L31:
By replacing the JGE instruction as shown in the previous example with a SETcc instruction,
the EBX register is set to either C1 or C2. This code can be optimized to eliminate the branches
as shown in the following code:
xor ebx, ebx ;clear ebx
cmp A, B
setge bl ;When ebx = 0 or 1
;OR the complement condition
dec ebx ;ebx=00...00 or 11...11
and ebx, (CONST2-CONST1) ;ebx=0 or(CONST2-CONST1)
add ebx, min(CONST1,CONST2) ;ebx=CONST1 or CONST2
The optimized code sets register EBX to 0 then compares A and B. If A is greater than or equal
to B then EBX is set to 1. EBX is then decremented and ANDed with the difference of the
constant values. This sets EBX to either 0 or the difference of the values. By adding the
minimum of the two constants the correct value is written to EBX. When CONST1 or CONST2
is equal to zero, the last instruction can be deleted as the correct value already has been written
to EBX.
When ABS(CONST1-CONST2) is 1 of {2,3,5,9}, the following example applies:
xor ebx, ebx
cmp A, B
setge bl ; or the complement condition
lea ebx, [ebx*D+ebx+CONST1-CONST2]
where D stands for ABS(CONST1 − CONST2) − 1.
A second way to remove branches on P6 family processors is to use the new CMOVcc and
FCMOVcc instructions. The following example shows how to use the CMOVcc instruction to
eliminate the branch from a test and branch instruction sequence. If the test sets the equal flag
then the value in register EBX will be moved to register EAX. This branch is data dependent,
and is representative of a unpredictable branch.
test ecx, ecx
jne 1h
mov eax, ebx
1h:
To change the code, the JNE and the MOV instructions are combined into one CMOVcc instruc-
tion, which checks the equal flag. The optimized code is shown below:
test ecx, ecx ; test the flags
cmoveqeax, ebx ; if the equal flag is set, move ebx to eax
1h:

14-6
 CODE OPTIMIZATION

The label 1h: is no longer needed unless it is the target of another branch instruction. These
instructions will generate invalid opcodes when used on previous generation Intel Architecture
processors. Therefore, use the CPUID instruction to check feature bit 15 of the EDX register,
which when set indicates presence of the CMOVcc family of instructions. Do not use the family
and model codes returned by CPUID to test for the presence of specific features.
Additional information on branch optimization can be found in the Intel Architecture Optimiza-
tion Manual.

14.3. REDUCING PARTIAL REGISTER STALLS ON P6 FAMILY

PROCESSORS
On P6 family processors, when a large (32-bit) general-purpose register is read immediately
after a small register (8- or 16-bit) that is contained in the large register has been written, the
read is stalled until the write retires (a minimum of 7 clocks). Consider the example below:
MOV AX, 8
ADD ECX, EAX ; Partial stall occurs on access of
; the EAX register
Here, the first instruction moves the value 8 into the small register AX. The next instruction
accesses the large register EAX. This code sequence results in a partial register stall.
Pentium® and Intel486™ processors do not generate this stall.
Table 14-1 lists the groups of small registers and their corresponding large register for which a
partial register stall can occur. For example, writing to register BL, BH, or BX and subsequently
reading register EBX will result in a stall.

Table 14-1. Small and Large General-Purpose Register Pairs

Small Registers Large Registers
AL AH AX EAX
BL BH BX EBX
CL CH CX ECX
DL DH DX EDX
SP ESP
BP EBP
DI EDI
SI ESI

Because the P6 family processors can execute code out of order, the instructions need not be
immediately adjacent for the stall to occur. The following example also contains a partial stall:
MOV AL, 8
MOV EDX, 0x40
MOV EDI, new_value

14-7
CODE OPTIMIZATION

ADD EDX, EAX ; Partial stall occurs on access of

; the EAX register
In addition, any micro-ops that follow the stalled micro-op will also wait until the clock cycle
after the stalled micro-op continues through the pipe. In general, to avoid stalls, do not read a
large register after writing a small register that is contained in the large register.
Special cases of writing and reading corresponding small and large registers have been imple-
mented in the P6 family processors to simplify the blending of code across processor genera-
tions. The special cases include the XOR and SUB instructions when using EAX, EBX, ECX,
EDX, EBP, ESP, EDI and ESI as shown in the following examples:
xor eax, eax
movb al, mem8
add eax, mem32 ; no partial stall
xor eax, eax
movw ax, mem16
add eax, mem32 ; no partial stall
sub ax, ax
movb al, mem8
add ax, mem16 ; no partial stall
sub eax, eax
movb al, mem8
or ax, mem16 ; no partial stall
xor ah, ah
movb al, mem8
sub ax, mem16 ; no partial stall
In general, when implementing this sequence, always write all zeros to the large register then
write to the lower half of the register.

14-8
 CODE OPTIMIZATION

14.4. ALIGNMENT RULES AND GUIDELINES

The following section gives rules and guidelines for aligning of code and data for optimum code
execution speed.

14.4.1. Alignment Penalties

The following are common penalties for accesses to misaligned data or code:
• On a Pentium® processor, a misaligned access costs 3 clocks.
• On a P6 family processor, a misaligned access that crosses a cache line boundary costs 6 to
9 clocks.
• On a P6 family processor, unaligned accesses that cause a data cache split stall the
processor. A data cache split is a memory access that crosses a 32-byte cache line
boundary.
For best performance, make sure that data structures and arrays greater than 32 bytes, are 32-
byte aligned, and that access patterns to data structures and arrays do not break the alignment
rules.

14.4.2. Code Alignment

The P6 family and Pentium® processors have a cache line size of 32 bytes. Since the prefetch
buffers fetch on 16-byte boundaries, code alignment has a direct impact on prefetch buffer effi-
ciency.
For optimal performance across the Intel Architecture family, it is recommended that:
• A loop entry label should be 16-byte aligned when it is less than 8 bytes away from that
boundary.
• A label that follows a conditional branch should not be aligned.
• A label that follows an unconditional branch or function call should be 16-byte aligned
when it is less than 8 bytes away from that boundary.

14.4.3. Data Alignment

A misaligned access in the data cache or on the bus costs at least 3 extra clocks on the Pentium®
processor. A misaligned access in the data cache, which crosses a cache line boundary, costs 9
to 12 clocks on the P6 family processors. It is recommended that data be aligned on the
following boundaries for optimum code execution on all processors:
• Align 8-bit data on any boundary.
• Align 16-bit data to be contained within an aligned 4-byte word.
• Align 32-bit data on any boundary that is a multiple of 4.
• Align 64-bit data on any boundary that is a multiple of 8.

14-9
CODE OPTIMIZATION

• Align 80-bit data on a 128-bit boundary (that is, any boundary that is a multiple of 16
bytes).
• Align 128-bit SIMD floating-point data on a 128-bit boundary (that is, any boundary that is
a multiple of 16 bytes).

14.4.3.1. ALIGNMENT OF DATA STRUCTURES AND ARRAYS GREATER

THAN 32 BYTES
A 32-byte or greater data structure or array should be aligned such that the beginning of each
structure or array element is aligned on a 32 byte boundary, and such that each structure or array
element does not cross a 32-byte cache line boundary.
Does this general discussion adequately cover the differences between 8, 16, and 32 bit
alignments?

14.4.3.2. ALIGNMENT OF DATA IN MEMORY AND ON THE STACK

On the Pentium® processor, accessing 64-bit variables that are not 8-byte aligned will cost an
extra 3 clocks. On the P6 family processors, accessing a 64-bit variable will cause a data cache
split. Some commercial compilers do not align double precision variables on 8-byte boundaries.
In such cases, the following techniques can be used to force optimum alignment of data:
• Use static variables instead of dynamic (stack) variables.
• Use in-line assembly code that explicitly aligns data.
• In C code, use “malloc” to explicitly allocate variables.
The following sections describe these techniques.
Static Variables
When a compiler allocates stack space for a dynamic variable, it may not align the variable (see
Figure 14-1). However, in most cases, when the compiler allocates space in memory for static
variables, the variables are aligned.

14-10
 CODE OPTIMIZATION

static float a;
float b; b
static float c; b
Stack

a
Memory
c

Figure 14-1. Stack and Memory Layout of Static Variables

Alignment Using Assembly Language

Use in-line assembly code to explicitly align variables. The following example aligns the stack
to 64-bits.
; procedure prologue
push ebp
mov esp, ebp
and ebp, -8
sub esp, 12
; procedure epilogue
add esp, 12
pop ebp
ret
Dynamic Allocation Using MALLOC
When using dynamic allocation, check that the compiler aligns doubleword or quadword values
on 8-byte boundaries. If the compiler does not implement this alignment, then use the following
technique to align doublewords and quadwords for optimum code execution:
1. Allocate memory equal to the size of the array or structure plus 4 bytes.
2. Use “bitwise” and to make sure that the array is aligned, for example:
double a[5];
double *p, *newp;
p = (double*)malloc ((sizeof(double)*5)+4)
newp = (p+4) & (-7)

14-11
CODE OPTIMIZATION

14.5. INSTRUCTION SCHEDULING OVERVIEW

On all Intel Architecture processors, the scheduling of (arrangement of) instructions in the
instruction stream can have a significant affect on the execution speed of the processor. For
example, when executing code on a Pentium ® or later Intel Architecture processor, two 1-clock
instructions that do not have register or data dependencies between them can generally be
executed in parallel (in a single clock) if they are paired—placed adjacent to one another in the
instruction stream. Likewise, a long-latency instruction such as a floating-point instruction can
often be executed in parallel with a sequence of 1-clock integer instructions or shorter latency
floating-point instructions if the instructions are scheduled appropriately in the instruction
stream.
The following sections describe two aspects of scheduling that can provide improved perfor-
mance in Intel Architecture processors: pairing and pipelining. Pairing is generally used to opti-
mize the execution of integer and MMX™ instructions; pipelining is generally used to optimize
the execution of MMX™ and floating-point instructions.

14.5.1. Instruction Pairing Guidelines

The microarchitecture for the Pentium® family of processors (with and without MMX™ tech-
nology) contain two instruction execution pipelines: the U-pipe and the V-pipe. These pipelines
are capable of executing two Intel Architecture instructions in parallel (during the same clock
or clocks) if the two instructions are pairable. Pairable instructions are those instructions that
when they appear adjacent to one another in the instruction stream will normally be executed in
parallel. By ordering a code sequence so that whenever possible pairable instructions occur
sequentially, code can be optimized to take advantage of the Pentium® processor’s two-pipe
microarchitecture.

NOTE
Pairing of instructions improves Pentium® processor performance signifi-
cantly. It does not slow and sometimes improves the performance of P6
family processors.
The following subsections describe the Pentium® processor pairing rules for integer, MMX™,
and, floating-point instructions. The pairing rules are grouped into types, as follows:
• General pairing rules
• Integer instruction pairing rules.
• MMX™ instruction pairing rules.
• Floating-point instruction pairing rules.

14.5.1.1. GENERAL PAIRING RULES

The following are general rules for instruction pairing in code written to run on Pentium®
processors:

14-12
 CODE OPTIMIZATION

• Unpairable instructions are always executed in the U-pipe.

• For paired instructions to execute in parallel, the first instruction of the pair must fall on an
instruction boundary that forces the instruction to be executed in the U-pipe. The following
placements of an instruction in the instruction stream will force an instruction to be
executed in the U-pipe:
— If the first instruction of a pair of pairable instructions is the first instruction in a block
of code, the first instruction will be executed in the U-pipe and the second of the pair
will be executed in the V-pipe, resulting in parallel execution of the two instructions.
— If the first instruction of a pair of pairable instructions follows an unpairable
instruction in the instruction stream, the first of the pairable instructions will be
executed in the U-pipe and the second of the pair in the V-pipe, resulting in parallel
execution.
— After one pair of instructions has been executed in parallel, subsequent pairs will also
be executed in parallel until an unpairable instruction is encountered.
• Parallel execution of paired instructions will not occur if:
— The next two instructions are not pairable instructions.
— The next two instructions have some type of register contention (implicit or explicit).
There are some special exceptions (see “Special Pairs”, in Section 14.5.1.2., “Integer
Pairing Rules”) to this rule where register contention can occur with pairing.
— The instructions are not both in the instruction cache. An exception to this that permits
pairing is if the first instruction is a one byte instruction.
— The processor is operating in single-step mode.
• Instructions that have data dependencies should be separated by at least one other
instruction.
• Pentium® processors without MMX™ technology do not execute a set of paired instruc-
tions if either instruction is longer than 7 bytes; Pentium® processors with MMX™
technology do not execute a set of paired instructions if the first instruction is longer than
11 bytes or the second instruction is longer than 7 bytes. Prefixes are not counted.
• On Pentium® processors without MMX™ technology, prefixed instructions are pairable
only in the U-pipe. On Pentium® processors with MMX™ technology, instructions with
0FH, 66H or 67H prefixes are also pairable in the V-pipe. For this and the previous rule,
stalls at the entrance to the instruction FIFO, on Pentium® processors with MMX™
technology, will prevent pairing.
• Floating-point instructions are not pairable with MMX™ instructions.

14.5.1.2. INTEGER PAIRING RULES

Table 14-2 shows the integer instructions that can be paired. The table is divided into two halves:
one for the U-pipe and one for the V-pipe. Any instruction in the U-pipe list can be paired with
any instruction in the V-pipe list, and vice versa.

14-13
CODE OPTIMIZATION

Table 14-2. Pairable Integer Instructions

Integer Instruction Pairable in U-Pipe Integer Instruction Pairable in V-Pipe
MOV reg, reg ALU reg, imm PUSH reg MOV reg, reg ALU reg, imm PUSH reg
MOV reg, mem ALU mem, imm PUSH imm MOV reg, mem ALU mem, imm PUSH imm
MOV mem, reg ALU eax, imm POP reg MOV mem, reg ALU eax, imm POP reg
MOV reg, imm ALU mem, reg NOP MOV reg, imm ALU mem, reg JMP near
MOV mem, ALU reg, mem SHIFT/ROT by MOV mem, ALU reg, mem Jcc near
imm 1 imm
MOV eax, mem INC/DEC reg SHIFT by imm MOV eax, mem INC./DEC reg 0F Jcc
MOV mem, eax INC/DEC mem TEST reg, r/m MOV m, eax INC/DEC mem CALL near
ALU reg, reg LEA reg, mem TEST acc, imm ALU reg, reg LEA reg, mem NOP
TEST reg, r/m TEST acc, imm

NOTES:
ALU—Arithmetic or logical instruction such as ADD, SUB, or AND. In general, most simple ALU instructions
are pairable.
imm—Immediate.
reg—Register.
mem—Memory location.
r/m—Register or memory location.
acc—Accumulator (EAX or AX register).

General Integer-Instruction Pairability Rules

The following are general rules for pairability of integer instructions. These rules summarize the
pairing of instructions in Table 14-2.
• NP Instructions—The following integer instructions cannot be paired:
— The shift and rotate instructions with a shift count in the CL register.
— Long-arithmetic instructions, such as MUL and DIV.
— Extended instructions, such as RET, ENTER, PUSHA, MOVS, STOS, and LOOPNZ.
— Inter-segment instructions, such as PUSH sreg and CALL far.
• UV Instructions—The following instructions can be paired when issued to the U- or V-
pipes:
— Most 8/32 bit ALU operations, such as ADD, INC, and XOR.
— All 8/32 bit compare instructions, such as CMP and TEST.
— All 8/32 bit stack operations using registers, such as PUSH reg and POP reg.
• PU instructions—The following instructions when issued to the U-pipe can be paired with
a suitable instruction in the V-Pipe. These instructions never execute in the V-pipe.
— Carry and borrow instructions, such as ADC and SBB.

14-14
 CODE OPTIMIZATION

— Prefixed instructions.
— Shift with immediate instructions.
• PV instructions—The following instructions when issued to the V-pipe can be paired with
a suitable instruction in the U-Pipe. The simple control transfer instructions, such as the
CALL near, JMP near, or Jcc instructions, can execute in either the U-pipe or the V-pipe,
but they can be paired with other instructions only when they are in the V-pipe. Since these
instructions change the instruction pointer (EIP), they cannot pair in the U-pipe since the
next instruction may not be adjacent. The PV instructions include both Jcc short and Jcc
near (which have a 0FH prefix) versions of the Jcc instruction.
Unpairability Due to Register Dependencies
Instruction pairing is also affected by instruction operands. The following instruction pairings
will not result in parallel execution because of register contention. Exceptions to these rules are
given in “Special Pairs”, in Section 14.5.1.2., “Integer Pairing Rules”.
• Flow Dependence—The first instruction writes to a register that the second one reads
from, as in the following example:
mov eax, 8
mov [ebp], eax
• Output Dependence—Both instructions write to the same register, as in the following
example.
mov eax, 8
mov eax, [ebp]
This output dependence limitation does not apply to a pair of instructions that write to the
EFLAGS register (for example, two ALU operations that change the condition codes). The
condition code after the paired instructions execute will have the condition from the V-pipe
instruction.
Note that a pair of instructions in which the first reads a register and the second writes to the
same register (anti-dependence) may be paired, as in the following example:
mov eax, ebx
mov ebx, [ebp]
For purposes of determining register contention, a reference to a byte or word register is treated
as a reference to the containing 32-bit register. Therefore, the following instruction pair does not
execute in parallel because of output dependencies on the contents of the EAX register.
mov al, 1
mov ah, 0

14-15
CODE OPTIMIZATION

Special Pairs
Some integer instructions can be paired in spite of the previously described general integer-
instruction rules. These special pairs overcome register dependencies, and most involve implicit
reads/writes to the ESP register or implicit writes to the condition codes:
• Stack Pointer.
push reg/imm ; push reg/imm
push reg/imm ; call
pop reg ; pop reg
• Condition Codes.
cmp ; jcc
add ; jne
Note that the special pairs that consist of PUSH/POP instructions may have only immediate or
register operands, not memory operands.
Restrictions On Pair Execution
Some integer-instruction pairs may be issued simultaneously but will not execute in parallel:
• Data-Cache Conflict—If both instructions access the same data-cache memory bank then
the second request (V-pipe) must wait for the first request to complete. A bank conflict
occurs when bits 2 through 4 of the two physical addresses are the same. A bank conflict
results in a 1-clock penalty on the V-pipe instruction.
• Inter-Pipe Concurrency—Parallel execution of integer instruction pairs preserves memory-
access ordering. A multiclock instruction in the U-pipe will execute alone until its last
memory access.
For example, the following instructions add the contents of the register and the value at the
memory location, then put the result in the register. An add with a memory operand takes 2
clocks to execute. The first clock loads the value from the data cache, and the second clock
performs the addition. Since there is only one memory access in the U-pipe instruction, the add
in the V-pipe can start in the same clock.
add eax, meml
add ebx, mem2 ;1
(add) (add) ; 2 2-cycle
The following instructions add the contents of the register to the memory location and store the
result at the memory location. An add with a memory result takes 3 clocks to execute. The first
clock loads the value, the second performs the addition, and the third stores the result. When
paired, the last clock of the U-pipe instruction overlaps with the first clock of the V-pipe instruc-
tion execution.
add meml, eax ;1
(add) ;2
(add) add mem2, ebx ;3
(add) ;4
(add) ;5

14-16
 CODE OPTIMIZATION

No other instructions may begin execution until the instructions already executing have
completed.
To expose the opportunities for scheduling and pairing, it is better to issue a sequence of simple
instructions rather than a complex instruction that takes the same number of clocks. The simple
instruction sequence can take advantage of more issue slots. The load/store style code genera-
tion requires more registers and increases code size. This impacts Intel486™ processor perfor-
mance, although only as a second order effect. To compensate for the extra registers needed,
extra effort should be put into register allocation and instruction scheduling so that extra regis-
ters are only used when parallelism increases.

14.5.1.3. MMX™ INSTRUCTION PAIRING GUIDELINES

This section specifies guidelines and restrictions for pairing MMX™ instructions with each
other and with integer instructions.
Pairing Two MMX™ Instructions
The following restrictions apply when pairing of two MMX™ instructions:
• Two MMX™ instructions that both use the MMX™ shifter unit (pack, unpack, and shift
instructions) are not pairable because there is only one MMX™ shifter unit. Shift
operations may be issued in either the U-pipe or the V-pipe, but cannot executed in both
pipes in the same clock.
• Two MMX™ instructions that both use the MMX™ multiplier unit (PMULL, PMULH,
PMADD type instructions) are not pairable because there is only one MMX™ multiplier
unit. Multiply operations may be issued in either the U-pipe or the V-pipe, but cannot
executed in both pipes in the same clock.
• MMX™ instructions that access either memory or a general-purpose register can be issued
in the U-pipe only. Do not schedule these instructions to the V-pipe as they will wait and be
issued in the next pair of instructions (and to the U-pipe).
• The MMX™ destination register of the U-pipe instruction should not match the source or
destination register of the V-pipe instruction (dependency check).
• The EMMS instruction is not pairable with other instructions.
• If either the TS flag or the EM flag in control register CR0 is set, MMX™ instructions
cannot be executed in the V-pipe.
Pairing an Integer Instruction in the U-Pipe With an MMX™ Instruction in the V-Pipe
Use the following guidelines for pairing an integer instruction in the U-pipe and an MMX™
instruction in the V-pipe:
• The MMX™ instruction is not the first MMX™ instruction following a floating-point
instruction.
• The V-pipe MMX™ instruction does not access either memory or a general-purpose
register.

14-17
CODE OPTIMIZATION

• The U-pipe integer instruction is a pairable U-pipe integer instruction (see Table 14-2).
Pairing an MMX™ Instruction in the U-Pipe with an Integer Instruction in the V-Pipe
Use the following guidelines for pairing an MMX™ instruction in the U-pipe and an integer
instruction in the V-pipe:
• The U-pipe MMX™ instruction does not access either memory or a general-purpose
register.
• The V-pipe instruction is a pairable integer V-pipe instruction (see Table 14-2).

14.5.2. Pipelining Guidelines

The term pipelining refers to the practice of scheduling instructions in the instruction stream to
reduce processor stalls due to register, data, or data-cache dependencies. The effect of pipelining
on code execution is highly dependent on the family of Intel Architecture processors the code is
intended to run on. Pipelining can greatly increase the performance of code written to run on the
Pentium® family of processors. It is less important for code written to run on the P6 family
processors, because the dynamic execution model that these processors use does a significant
amount of pipelining automatically.
The following subsections describe general pipelining guidelines for MMX™ and floating-
point instructions. These guidelines yield significant improvements in execution speed for code
running on the Pentium® processors and may yield additional improvements in execution speed
on the P6 family processors. Specific pipelining guidelines for the P6 family processors are
given in Section 14.5.3., “Scheduling Rules for P6 Family Processors”

14.5.2.1. MMX™ INSTRUCTION PIPELINING GUIDELINES

All MMX™ instructions can be pipelined on P6 family and Pentium® (with MMX™ tech-
nology) processors, including the multiply instructions. All MMX™ instructions take a single
clock to execute except the MMX™ multiply instructions which take 3 clocks.
Since MMX™ multiply instructions take 3 clocks to execute, the result of a multiply instruction
can be used only by other instructions issued 3 clocks later. For this reason, avoid scheduling a
dependent instruction in the 2 instruction pairs following the multiply.
The store of a register after writing the register must wait for 2 clocks after the update of the
register. Scheduling the store 2 clocks after the update avoids a pipeline stall.

14.5.2.2. FLOATING-POINT PIPELINING GUIDELINES

Many of the floating-point instructions have a latency greater than 1 clock, therefore on
Pentium® processors the next floating-point instruction cannot access the result until the first
operation has finished execution. To hide this latency, instructions should be inserted between
the pair that causes the pipe stall. These instructions can be integer instructions or floating-point
instructions that will not cause a new stall themselves. The number of instructions that should
be inserted depends on the length of the latency. Because of the out-of-order execution capa-

14-18
 CODE OPTIMIZATION

bility of the P6 family processors, stalls will not necessarily occur on an instruction or micro-op
basis. However, if an instruction has a very long latency such as an FDIV, then scheduling can
improve the throughput of the overall application. The following sections list considerations for
floating-point pipelining on Pentium® processors.
Pairing of Floating-Point Instructions
In a Pentium® processor, pairing floating-point instructions with one another (with one excep-
tion) does not result in a performance enhancement because the processor has only one floating-
point unit (FPU). However, some floating-point instructions can be paired with integer instruc-
tions or the FXCH instruction to improve execution times. The following are some general
pairing rules and restrictions for floating-point instructions:
• All floating-point instructions can be executed in the V-pipe and paired with suitable
instructions (generally integer instructions) in the U-pipe.
• The only floating-point instruction that can be executed in the U-pipe is the FXCH
instruction. The FXCH instruction, if executed in the U-pipe can be paired with another
floating-point instruction executing in the V-pipe.
• The floating-point instructions FSCALE, FLDCW, and FST cannot be paired with any
instruction (integer instruction or the FXCH instruction).
Using Integer Instructions to Hide Latencies and Schedule Floating-Point Instructions
When a floating-point instruction depends on the result of the immediately preceding instruc-
tion, and that instruction is also a floating-point instruction, performance can be improved by
placing one or more integer instructions between the two floating-point instructions. This is true
even if the integer instructions perform loop control. The following example restructures a loop
in this manner:
for (i=0; i<Size; i++)
array1 [i] += array2 [i];
; assume eax=Size-1, esi=array1, edi=array2
PENTIUM(R) PROCESSORCLOCKS
LoopEntryPoint:
fld real4 ptr [esi+eax*4] ; 2 - AGI
fadd real4 ptr [edi+eax*4] ;1
fstp real4 ptr [esi+eax*4] ; 5 - waits for fadd
dec eax ;1
jnz LoopEntryPoint
; assume eax=Size-1, esi=array1, edi=array2
jmp LoopEntryPoint
Align 16
TopOfLoop:
fstp real4 ptr [esi+eax*4+4] ; 4 - waits for fadd + AGI
LoopEntryPoint:
fld real4 ptr [esi+eax*4] ;1

14-19
CODE OPTIMIZATION

fadd real4 ptr [edi+eax*4] ;1

dec eax ;1
jnz TopOfLoop
;
fstp real4 ptr [esi+eax*4+4]
By moving the integer instructions between the FADDS and FSTPS instructions, the integer
instructions can be executed while the FADDS instruction is completing in the floating-point unit
and before the FSTPS instruction begins execution. Note that this new loop structure requires a
separate entry point for the first iteration because the loop needs to begin with the FLDS instruc-
tion. Also, there needs to be an additional FSTPS instruction after the conditional jump to finish
the final loop iteration.
Hiding the One-Clock Latency of a Floating-Point Store
A floating-point store must wait an extra clock for its floating-point operand. After an FLD, an
FST must wait 1 clock, as shown in the following example:
fld meml ; 1 fld takes 1 clock
; 2 fst waits, schedule something here
fst mem2 ; 3,4 fst takes 2 clocks
After the common arithmetic operations, FMUL and FADD, which normally have a latency of
3 clocks, FST waits an extra clock for a total of 4 (see following example).
fadd meml ; 1 add takes 3 clocks
; 2 add, schedule something here
; 3 add, schedule something here
; 4 fst waits, schedule something here
fst mem2 ; 5,2 fst takes 2 clocks
Other instructions such as FADDP and FSUBRP also exhibit this type of latency.
In the next example, the store is not dependent on the previous load:
fld meml ;1
fld mem2 ;2
fxch st(l) ;2
fst mem3 ; 3 stores values loaded from meml
Here, a register may be used immediately after it has been loaded (with FLD):
fld mem1 ;l
fadd mem2 ; 2,3,4
Use of a register by a floating-point operation immediately after it has been written by another
FADD, FSUB, or FMUL causes a 2-clock delay. If instructions are inserted between these two,
then latency and a potential stall can be hidden.
Additionally, there are multiclock floating-point instructions (FDIV and FSQRT) that execute in
the floating-point unit pipe (the U-pipe). While executing these instructions in the floating-point
unit pipe, integer instructions can be executed in parallel. Emitting a number of integer instruc-
tions after such an instruction will keep the integer execution units busy (the exact number of
instructions depends on the floating-point instruction’s clock count).

14-20
 CODE OPTIMIZATION

Integer instructions generally overlap with the floating-point operations except when the last
floating-point operation was FXCH. In this case there is a 1 clock delay:
:

U-pipe V-pipe
fadd fxch ;1
; 2 fxch delay
mov eax, 1 inc edx

Integer and Floating-Point Multiply

The integer multiply operations, the MUL and IMUL instructions, are executed by the FPU’s
multiply unit. Therefore, for the Pentium® processor, these instructions cannot be executed in
parallel with a floating-point instruction. This restriction does not apply to the P6 family proces-
sors, because these processors have two internal FPU execution units.
A floating-point multiply instruction (FMUL) delays for 1 clock if the immediately preceding
clock executed an FMUL or an FMUL-FXCH pair. The multiplier can only accept a new pair of
operands every other clock.
Floating-Point Operations with Integer Operands
Floating-point operations that take integer operands (the FIADD or FISUB instruction) should be
avoided. These instructions should be split into two instructions: the FILD instruction and a
floating-point operation. The number of clocks before another instruction can be issued
(throughput) for FIADD is 4, while for FILD and simple floating-point operations it is 1, as
shown in the example below:
.

Complex Instructions Better for Potential Overlap

fiadd [ebp] ; 4 fild [ebp] ; 1
faddp st(l) ; 2

Using the FILD and FADDP instructions in place of FIADD yields 2 free clocks for executing
other instructions.
FSTSW Instruction
The FSTSW instruction that usually appears after a floating-point comparison instruction
(FCOM, FCOMP, FCOMPP) delays for 3 clocks. Other instructions may be inserted after the
comparison instruction to hide this latency. On the P6 family processors the FCMOVcc instruc-
tion can be used instead.

14-21
CODE OPTIMIZATION

Transcendental Instructions
Transcendental instructions execute in the U-pipe and nothing can be overlapped with them, so
an integer instruction following a transcendental instruction will wait until the previous instruc-
tion completes.
Transcendental instructions execute on the Pentium® processor (and later Intel Architecture
processors) much faster than the software emulations of these instructions found in most math
libraries. Therefore, it may be worthwhile in-lining transcendental instructions in place of math
library calls to transcendental functions. Software emulations of transcendental instructions will
execute faster than the equivalent instructions only if accuracy is sacrificed.
FXCH Guidelines
The FXCH instruction costs no extra clocks on the Pentium® processor when all of the following
conditions occur, allowing the instruction to execute in the V-pipe in parallel with another
floating-point instruction executing in the U-pipe:
• A floating-point instruction follows the FXCH instruction.
• A floating-point instruction from the following list immediately precedes the FXCH
instruction: FADD, FSUB, FMUL, FLD, FCOM, FUCOM, FCHS, FTST, FABS, or FDIV.
• An FXCH instruction has already been executed. This is because the instruction boundaries
in the cache are marked the first time the instruction is executed, so pairing only happens
the second time this instruction is executed from the cache.
When the above conditions are true, the instruction is almost “free” and can be used to access
elements in the deeper levels of the floating-point stack instead of storing them and then loading
them again.

14.5.3. Scheduling Rules for P6 Family Processors

The P6 family processors have 3 decoders that translate Intel Architecture macro instructions
into micro operations (micro-ops, also called “uops”). The decoder limitations are as follows:
• The first decoder (decoder 0) can decode instructions up to 7 bytes in length and with up to
4 micro-ops in one clock cycle. The second two decoders (decoders 1 and 2) can decode
instructions that are 1 micro-op instructions, and these instructions will also be decoded in
one clock cycle.
• Three macro instructions in an instruction sequence that fall into this envelope will be
decoded in one clock cycle.
• Macro instructions outside this envelope will be decoded through decoder 0 alone. While
decoder 0 is decoding a long macro instruction, decoders 1 and 2 (second and third
decoders) are quiescent.
Appendix C of the Intel Architecture Optimization Manual lists all Intel macro-instructions and
the decoders on which they can be decoded.

14-22
 CODE OPTIMIZATION

The macro instructions entering the decoder travel through the pipe in order; therefore, if a
macro instruction will not fit in the next available decoder then the instruction must wait until
the next clock to be decoded. It is possible to schedule instructions for the decoder such that the
instructions in the in-order pipeline are less likely to be stalled.
Consider the following examples:
• If the next available decoder for a multimicro-op instruction is not decoder 0, the
multimicro-op instruction will wait for decoder 0 to be available, usually in the next clock,
leaving the other decoders empty during the current clock. Hence, the following two
instructions will take 2 clocks to decode.
add eax, ecx ; 1 uop instruction (decoder 0)
add edx, [ebx] ; 2 uop instruction (stall 1 cycle wait till
; decoder 0 is available)
• During the beginning of the decoding clock, if two consecutive instructions are more than
1 micro-op, decoder 0 will decode one instruction and the next instruction will not be
decoded until the next clock.
add eax, [ebx] ; 2 uop instruction (decoder 0)
mov ecx, [eax] ; 2 uop instruction (stall 1 cycle to wait until
; decoder 0 is available)
add ebx, 8 ; 1 uop instruction (decoder 1)
Instructions of the opcode reg, mem form produce two micro-ops: the load from memory and
the operation micro-op. Scheduling for the decoder template (4-1-1) can improve the decoding
throughput of your application.
In general, the opcode reg, mem forms of instructions are used to reduce register pressure in code
that is not memory bound, and when the data is in the cache. Use simple instructions for
improved speed on the Pentium® and P6 family processors.
The following rules should be observed while using the opcode reg, mem instruction on
Pentium® processors with MMX™ technology:
• Schedule for minimal stalls in the Pentium® processor pipe. Use as many simple instruc-
tions as possible. Generally, 32-bit assembly code that is well optimized for the Pentium®
processor pipeline will execute well on the P6 family processors.
• When scheduling for Pentium® processors, keep in mind the primary stall conditions and
decoder (4-1-1) template on the P6 family processors, as shown in the example below.
pmaddw mm6, [ebx] ; 2 uops instruction (decoder 0)
paddd mm7, mm6 ; 1 uop instruction (decoder 1)
ad ebx, 8 ; 1 uop instruction (decoder 2)

14-23
CODE OPTIMIZATION

14.6. ACCESSING MEMORY

The following subsections describe optimizations that can be obtained when scheduling instruc-
tions that access memory.

14.6.1. Using MMX™ Instructions That Access Memory

An MMX™ instruction may have two register operands (opcode reg, reg) or one register and
one memory operand (opcode reg, mem), where opcode represents the instruction opcode, reg
represents the register, and mem represents memory. The opcode reg, mem instructions are
useful in some cases to reduce register pressure, increase the number of operations per clock,
and reduce code size.
The following discussion assumes that the memory operand is present in the data cache. If it is
not, then the resulting penalty is usually large enough to obviate the scheduling effects discussed
in this section.
In Pentium® processor with MMX™ technology, the opcode reg, mem MMX™ instructions do
not have longer latency than the opcode reg, reg instructions (assuming a cache hit). They do
have more limited pairing opportunities, however. In the Pentium® II and Pentium® III proces-
sors, the opcode reg, mem MMX™ instructions translate into two micro-ops, as opposed to one
micro-op for the opcode reg, reg instructions. Thus, they tend to limit decoding bandwidth and
occupy more resources than the opcode reg, reg instructions.
The recommended usage of the opcode reg, reg instructions depends on whether the MMX™
code is memory-bound (that is, execution speed is limited by memory accesses). As a rule of
thumb, an MMX™ code sequence is considered to be memory-bound if the following inequality
holds:

Instructions NonMMXInstructions
--------------------------------- < MemoryAccesses + -------------------------------------------------------------
2 2

For memory-bound MMX™ code, Intel recommends merging loads whenever the same
memory address is used more than once to reduce memory accesses. For example, the following
code sequence can be speeded up by using a MOVQ instruction in place of the opcode reg,
mem forms of the MMX™ instructions:
OPCODE MM0, [address A]
OPCODE MM1, [address A]
; optimized by use of a MOVQ instruction and opcode reg, mem forms
; of the MMX(TM) instructions
MOVQ MM2, [address A]
OPCODE MM0, MM2
OPCODE MM1, MM2
Another alternative is to incorporate the prefetch instruction introduced in the Pentium® III
processor. Prefetching the data preloads the cache prior to actually needing the data. Proper use
of prefetch can improve performance if the application is not memory bandwidth bound or the

14-24
 CODE OPTIMIZATION

data does not already fit into cache. For more information on proper usage of the prefetch
instruction see the Intel Architecture Optimization Manual order number 245127-001.
For MMX™ code that is not memory-bound, load merging is recommended only if the same
memory address is used more than twice. Where load merging is not possible, usage of the
opcode reg, mem instructions is recommended to minimize instruction count and code size. For
example, the following code sequence can be shortened by removing the MOVQ instruction and
using an opcode reg, mem form of the MMX™ instruction:
MOVQ mm0, [address A]
OPCODE mm1, mm0
; optimized by removing the MOVQ instruction and using an
; opcode reg, mem form of the MMX(TM) instructions
OPCODE mm1, [address A]
In many cases, a MOVQ reg, reg and opcode reg, mem can be replaced by a MOVQ reg, mem and
the opcode reg, reg. This should be done where possible, since it saves one micro-op on the
Pentium® II and Pentium® III processors. The following example is one where the opcode is a
symmetric operation:
MOVQ mm1, mm0 (1 micro-op)
OPCODE mm1, [address A] (2 micro-ops)
One clock can be saved by rewriting the code as follows:
MOVQ mm1, [address A] (1 micro-op)
OPCODE mm1, mm0 (1 micro-op)

14.6.2. Partial Memory Accesses With MMX™ Instructions

The MMX™ registers allow large quantities of data to be moved without stalling the processor.
Instead of loading single array values that are 8-, 16-, or 32-bits long, the values can be loaded
in a single quadword, with the structure or array pointer being incremented accordingly.
Any data that will be manipulated by MMX™ instructions should be loaded using either:
• The MMX™ instruction that loads a 64-bit operand (for example, MOVQ MM0, m64), or
• The register-memory form of any MMX™ instruction that operates on a quadword
memory operand (for example, PMADDW MM0, m64).
All data in MMX™ registers should be stored using the MMX™ instruction that stores a 64-bit
operand (for example, MOVQ m64, MM0).
The goal of these recommendations is twofold. First, the loading and storing of data in MMX™
registers is more efficient using the larger quadword data block sizes. Second, using quadword
data block sizes helps to avoid the mixing of 8-, 16-, or 32-bit load and store operations with 64-
bit MMX™ load and store operations on the same data. This, in turn, prevents situations in
which small loads follow large stores to the same area of memory, or large loads follow small
stores to the same area of memory. The Pentium® II and Pentium® III processors will stall in
these situations.

14-25
CODE OPTIMIZATION

Consider the following examples.

The first example illustrates the effects of a large load after a series of small stores to the same
area of memory (beginning at memory address mem). The large load will stall the processor:
MOV mem, eax ; store dword to address "mem"
MOV mem + 4, ebx ; store dword to address "mem + 4"
:
:
MOVQ mm0, mem ; load qword at address "mem", stalls
The MOVQ instruction in this example must wait for the stores to write memory before it can
access all the data it requires. This stall can also occur with other data types (for example, when
bytes or words are stored and then words or doublewords are read from the same area of
memory). By changing the code sequence as follows, the processor can access the data without
delay:
MOVD mm1, ebx ; build data into a qword first before storing it to memory
MOVD mm2, eax
PSLLQ mm1, 32
POR mm1, mm2
MOVQ mem, mm1 ; store SIMD variable to "mem" as a qword
:
:
MOVQ mm0, mem ; load qword SIMD variable "mem", no stall
The second example illustrates the effect of a series of small loads after a large store to the same
area of memory (beginning at memory address mem). Here, the small loads will stall the
processor:
MOVQ mem, mm0 ; store qword to address "mem"
:
:
MOV bx, mem + 2 ; load word at address "mem + 2" stalls
MOV cx, mem + 4 ; load word at address "mem + 4" stalls
The word loads must wait for the MOVQ instruction to write to memory before they can access
the data they require. This stall can also occur with other data types (for example, when double-
words or words are stored and then words or bytes are read from the same area of memory).
Changing the code sequence as follows allows the processor to access the data without a stall:
MOVQ mem, mm0 ; store qword to address "mem"
:
:
MOVQ mm1, mem ; load qword at address "mem"
MOVD eax, mm1 ; transfer "mem + 2" to ax from
; MMX(TM) register not memory
PSRLQ mm1, 32
SHR eax, 16
MOVD ebx, mm1 ; transfer "mem + 4" to bx from
; MMX register, not memory
AND ebx, 0ffffh

14-26
 CODE OPTIMIZATION

These transformations, in general, increase the number the instructions required to perform the
desired operation. For the Pentium® II and Pentium® III processors, the performance penalty due
to the increased number of instructions is more than offset by the number of clocks saved. For
the Pentium® processor with MMX™ technology, however, the increased number of instruc-
tions can negatively impact performance. For this reason, careful and efficient coding of these
transformations is necessary to minimize any potential negative impact to Pentium® processor
performance.

14.6.3. Write Allocation Effects

P6 family processors have a “write allocate by read-for-ownership” cache, whereas the
Pentium® processor has a “no-write-allocate; write through on write miss” cache.
On P6 family processors, when a write occurs and the write misses the cache, the entire 32-byte
cache line is fetched. On the Pentium® processor, when the same write miss occurs, the write is
simply sent out to memory.
Write allocate is generally advantageous, since sequential stores are merged into burst writes,
and the data remains in the cache for use by later loads. This is why P6 family processors
adopted this write strategy, and why some Pentium® processor system designs implement it for
the L2 cache.
Write allocate can be a disadvantage in code where:
• Just one piece of a cache line is written.
• The entire cache line is not read.
• Strides are larger than the 32-byte cache line.
• Writes to a large number of addresses (greater than 8000).
When a large number of writes occur within an application, and both the stride is longer than the
32-byte cache line and the array is large, every store on a P6 family processor will cause an entire
cache line to be fetched. In addition, this fetch will probably replace one (sometimes two) dirty
cache line(s). The result is that every store causes an additional cache line fetch and slows down
the execution of the program. When many writes occur in a program, the performance decrease
can be significant.
The following Sieve of Erastothenes example program demonstrates these cache effects. In this
example, a large array is stepped through in increasing strides while writing a single value of the
array with zero.

NOTE
This is a very simplistic example used only to demonstrate cache effects.
Many other optimizations are possible in this code.

14-27
CODE OPTIMIZATION

boolean array[max];
for(i=2;i<max;i++) {
array = 1;
}

for(i=2;i<max;i++) {
if( array[i] ) {
for(j=2;j<max;j+=i) {
array[j] = 0; /*here we assign memory to 0 causing
the cache line fetch within the j
loop */
}
}
}
Two optimizations are available for this specific example:
• Optimization 1—In “boolean” in this example there is a “char” array. Here, it may well be
better to make the “boolean” array into an array of bits, thereby reducing the size of the
array, which in turn reduces the number of cache line fetches. The array is packed so that
read-modify-writes are done (since the cache protocol makes every read into a read-
modify-write). Unfortunately, in this example, the vast majority of strides are greater than
256 bits (one cache line of bits), so the performance increase is not significant.
• Optimization 2—Another optimization is to check if the value is already zero before
writing (as shown in the following example), thereby reducing the number of writes to
memory (dirty cache lines)
boolean array[max];
for(i=2;i<max;i++) {
array = 1;
}

for(i=2;i<max;i++) {
if( array[i] ) {
for(j=2;j<max;j+=i) {
if( array[j] != 0 ) { /* check to see if value is
already 0 */
array[j] = 0;
}
}
}
}
The external bus activity is reduced by half because most of the time in the Sieve program the
data is already zero. By checking first, you need only 1 burst bus cycle for the read and you save
the burst bus cycle for every line you do not write. The actual write back of the modified line is
no longer needed, therefore saving the extra cycles.

14-28
 CODE OPTIMIZATION

NOTE
This operation benefits the P6 family processors, but it may not enhance the
performance of Pentium® processors. As such, it should not be considered
generic.

14.7. ADDRESSING MODES AND REGISTER USAGE

On the Pentium® processor, when a register is used as the base component, an additional clock
is used if that register is the destination of the immediately preceding instruction (assuming all
instructions are already in the prefetch queue). For example:
add esi, eax ; esi is destination register
mov eax, [esi] ; esi is base, 1 clock penalty
Since the Pentium® processor has two integer pipelines, a register used as the base or index
component of an effective address calculation (in either pipe) causes an additional clock if that
register is the destination of either instruction from the immediately preceding clock (see Figure
14-2). This effect is known as Address Generation Interlock (AGI). To avoid the AGI, the
instructions should be separated by at least 1 clock by placing other instructions between them.
The MMX™ registers cannot be used as base or index registers, so the AGI does not apply for
MMX™ register destinations.
No penalty occurs in the P6 family processors for the AGI condition.

AGI Penalty
PF

DI PF

D2 DI PF
AGI
E D2 DI

WB E D2

WB E

WB

Figure 14-2. Pipeline Example of AGI Stall

14-29
CODE OPTIMIZATION

Note that some instructions have implicit reads/writes to registers. Instructions that generate
addresses implicitly through ESP (such as PUSH, POP, RET, CALL) also suffer from the AGI
penalty, as shown in the following example:
sub esp, 24
; 1 clock cycle stall
push ebx
mov esp, ebp
; 1 clock cycle stall
pop ebp
The PUSH and POP instructions also implicitly write to the ESP register. These writes, however,
do not cause an AGI when the next instruction addresses through the ESP register. Pentium®
processors “rename” the ESP register from PUSH and POP instructions to avoid the AGI
penalty (see the following example):
push edi ; no stall
mov ebx, [esp]
On Pentium® processors, instructions that include both an immediate and a displacement field
are pairable in the U-pipe. When it is necessary to use constants, it is usually more efficient to
use immediate data instead of loading the constant into a register first. If the same immediate
data is used more than once, however, it is faster to load the constant in a register and then use
the register multiple times, as illustrated in the following example:
mov result, 555 ; 555 is immediate, result is
; displacement
mov word ptr [esp+4], 1 ; 1 is immediate, 4 is displacement
Since MMX™ instructions have 2-byte opcodes (0FH opcode map), any MMX™ instruction
that uses base or index addressing with a 4-byte displacement to access memory will have a
length of 8 bytes. Instructions over 7 bytes can slow macro instruction decoding and should be
avoided where possible. It is often possible to reduce the size of such instructions by adding the
immediate value to the value in the base or index register, thus removing the immediate field.

14.8. INSTRUCTION LENGTH

On Pentium® processors, instructions greater than 7 bytes in length cannot be executed in the V-
pipe. In addition, two instructions cannot be pushed into the instruction FIFO unless both are 7
bytes or less in length. If only one instruction is pushed into the instruction FIFO, pairing will
not occur unless the instruction FIFO already contains at least one instruction. In code where
pairing is very high (as is often the case in MMX™ code) or after a mispredicted branch, the
instruction FIFO may be empty, leading to a loss of pairing whenever the instruction length is
over 7 bytes.
In addition, the P6 family processors can only decode one instruction at a time when an instruc-
tion is longer than 7 bytes.
So, for best performance on all Intel processors, use simple instructions that are less than 8 bytes
in length.

14-30
 CODE OPTIMIZATION

14.9. PREFIXED OPCODES

On the Pentium® processor, an instruction with a prefix is pairable in the U-pipe (PU) if the
instruction (without the prefix) is pairable in both pipes (UV) or in the U-pipe (PU). The prefixes
are issued to the U-pipe and get decoded in 1 clock for each prefix and then the instruction is
issued to the U-pipe and may be paired.
For the P6 family and Pentium® processors, the prefixes that should be avoided for optimum
code execution speeds are:
• Lock.
• Segment override.
• Address size.
• Operand size.
• 2-byte opcode map (0FH) prefix. An exception is the Streaming SIMD Extensions instruc-
tions introduced with the Pentium® III processor. The first byte of these instructions is
0FH. It is not used as a prefix.
• 2-byte opcode map (0FH) prefix.
On Pentium® processors with MMX™ technology, a prefix on an instruction can delay the
parsing and inhibit pairing of instructions.
The following list highlights the effects of instruction prefixes on the Pentium® processor
instruction FIFO:
• There is no penalty on 0FH-prefix instructions.
• An instruction with a 66H or 67H prefix takes 1 clock for prefix detection, another clock
for length calculation, and another clock to enter the instruction FIFO (3 clocks total). It
must be the first instruction to enter the instruction FIFO, and a second instruction can be
pushed with it.
• Instructions with other prefixes (not 0FH, 66H, or 67H) take 1 additional clock to detect
each prefix. These instructions are pushed into the instruction FIFO only as the first
instruction. An instruction with two prefixes will take 3 clocks to be pushed into the
instruction FIFO (2 clocks for the prefixes and 1 clock for the instruction). A second
instruction can be pushed with the first into the instruction FIFO in the same clock.
The impact on performance exists only when the instruction FIFO does not hold at least two
entries. As long as the decoder (D1 stage) has two instructions to decode there is no penalty. The
instruction FIFO will quickly become empty if the instructions are pulled from the instruction
FIFO at the rate of two per clock. So, if the instructions just before the prefixed instruction suffer
from a performance loss (for example, no pairing, stalls due to cache misses, misalignments,
etc.), then the performance penalty of the prefixed instruction may be masked.
On the P6 family processors, instructions longer than 7 bytes in length limit the number of
instructions decoded in each clock. Prefixes add 1 to 2 bytes to the length of an instruction,
possibly limiting the decoder.

14-31
CODE OPTIMIZATION

It is recommended that, whenever possible, prefixed instructions not be used or that they be
scheduled behind instructions which themselves stall the pipe for some other reason.

14.10. INTEGER INSTRUCTION SELECTION AND OPTIMIZATIONS

This section describes both instruction sequences to avoid and sequences to use when generating
optimal assembly code. The information applies to the P6 family processors and the Pentium®
processors with and without MMX™ technology.
• LEA Instruction. The LEA instruction can be used in the following situations to optimize
code execution:
— The LEA instruction may be used sometimes as a three/four operand addition
instruction (for example, LEA ECX, [EAX+EBX+4+a]).
— In many cases, an LEA instruction or a sequence of LEA, ADD, SUB and SHIFT
instructions may be used to replace constant multiply instructions. For the P6 family
processors the constant multiply is faster relative to other instructions than on the
Pentium® processor, therefore the trade off between the two options occurs sooner. It is
recommended that the integer multiply instruction be used in code designed for P6
family processor execution.
— The above technique can also be used to avoid copying a register when both operands
to an ADD instruction are still needed after the ADD, since the LEA instruction need
not overwrite its operands.
The disadvantage of the LEA instruction is that it increases the possibility of an AGI
stall with previous instructions. LEA is useful for shifts of 2, 4, and 8 because on the
Pentium® processor, LEA can execute in either the U- or V-pipe, but the shift can only
execute in the U-pipe. On the P6 family processors, both the LEA and SHIFT instruc-
tions are single micro-op instructions that execute in 1 clock.
• Complex Instructions. For greater execution speed, avoid using complex instructions (for
example, LOOP, ENTER, or LEAVE). Use sequences of simple instructions instead to
accomplish the function of a complex instruction.
• Zero-Extension of Short Integers. On the Pentium® processor, the MOVZX instruction has
a prefix and takes 3 clocks to execute totaling 4 clocks. It is recommended that the
following sequence be used instead of the MOVZX instruction:
xor eax, eax
mov al, mem
If this code occurs within a loop, it may be possible to pull the XOR instruction out of
the loop if the only assignment to EAX is the MOV AL, MEM. This has greater impor-
tance for the Pentium® processor since the MOVZX is not pairable and the new
sequence may be paired with adjacent instructions.
In order to avoid a partial register stall on the P6 family processors, special hardware
has been implemented that allows this code sequence to execute without a stall. Even

14-32
 CODE OPTIMIZATION

so, the MOVZX instruction is a better choice for the P6 family processors than the
alternative sequences.
• PUSH Mem. The PUSH mem instruction takes 4 clocks for the Intel486™ processor. It is
recommended that the following sequence be used in place of a PUSH mem instruction
because it takes only 2 clocks for the Intel486™ processor and increases pairing
opportunity for the Pentium® processor.
mov reg, mem
push reg
• Short Opcodes. Use 1 byte long instructions as much as possible. This will reduce code
size and help increase instruction density in the instruction cache. The most common
example is using the INC and DEC instructions rather than adding or subtracting the
constant 1 with an ADD or SUB instruction. Another common example is using the PUSH
and POP instructions instead of the equivalent sequence.
• 8/16 Bit Operands. With 8-bit operands, try to use the byte opcodes, rather than using 32-
bit operations on sign and zero extended bytes. Prefixes for operand size override apply to
16-bit operands, not to 8-bit operands.
Sign Extension is usually quite expensive. Often, the semantics can be maintained by
zero extending 16-bit operands. Specifically, the C code in the following example does
not need sign extension nor does it need prefixes for operand size overrides.
static short int a, b;
if (a==b) {
...
}
Code for comparing these 16-bit operands might be:

U Pipe V Pipe
xor eax, eax xor ebx, ebx ;1

movw ax, [a] ; 2 (prefix) + 1

movw bx, [b] ; 4 (prefix) + 1

cmp eax, ebx ;6

Of course, this can only be done under certain circumstances, but the circumstances
tend to be quite common. This would not work if the compare was for greater than, less
than, greater than or equal, and so on, or if the values in EAX or EBX were to be used
in another operation where sign extension was required.
The P6 family processors provides special support for the XOR reg, reg instruction
where both operands point to the same register, recognizing that clearing a register does
not depend on the old value of the register. Additionally, special support is provided for
the above specific code sequence to avoid the partial stall.

14-33
CODE OPTIMIZATION

The following straight-forward method may be slower on Pentium® processors.

movsw eax, a ; 1 prefix + 3
movsw ebx, b ;5
cmp ebx, eax ;9
However, the P6 family processors have improved the performance of the MOVZX
instructions to reduce the prevalence of partial stalls. Code written specifically for the
P6 family processors should use the MOVZX instructions.
• Compares. Use the TEST instruction when comparing a value in a register with 0. TEST
essentially ANDs the operands together without writing to a destination register. If a value
is ANDed with itself and the result sets the zero condition flag, the value was zero. TEST
is preferred over an AND instruction because AND writes the result register which may
subsequently cause an AGI or an artificial output dependence on the P6 family processors.
TEST is better than CMP .., 0 because the instruction size is smaller.
Use the TEST instruction when comparing the result of a boolean AND with an imme-
diate constant for equality or inequality if the register is EAX (if (avar & 8) { }).
On the Pentium® processor, the TEST instruction is a 1 clock pairable instruction when
the form is TEST EAX, imm or TEST reg, reg. Other forms of TEST take 2 clocks and
do not pair.
• Address Calculations. Pull address calculations into load and store instructions. Internally,
memory reference instructions can have 4 operands: a relocatable load-time constant, an
immediate constant, a base register, and a scaled index register. (In the segmented model, a
segment register may constitute an additional operand in the linear address calculation.) In
many cases, several integer instructions can be eliminated by fully using the operands of
memory references.
• Clearing a Register. The preferred sequence to move zero to a register is XOR reg, reg.
This sequence saves code space but sets the condition codes. In contexts where the
condition codes must be preserved, use MOV reg, 0.
• Integer Divide. Typically, an integer divide is preceded by a CDQ instruction. (Divide
instructions use EDX: EAX as the dividend and CDQ sets up EDX.) It is better to copy
EAX into EDX, then right shift EDX 31 places to sign extend. On the Pentium® processor,
the copy/shift takes the same number of clocks as CDQ, but the copy/shift scheme allows
two other instructions to execute at the same time. If the value is known to be positive, use
XOR EDX, EDX.
On the P6 family processors, the CDQ instruction is faster, because CDQ is a single
micro-op instruction as opposed to two instructions for the copy/shift sequence.
• Prolog Sequences. Be careful to avoid AGIs in the procedure and function prolog
sequences due to register ESP. Since PUSH can pair with other PUSH instructions, saving
callee-saved registers on entry to functions should use these instructions. If possible, load
parameters before decrementing ESP.

14-34
 CODE OPTIMIZATION

In routines that do not call other routines (leaf routines), use ESP as the base register
to free up EBP. If you are not using the 32-bit flat model, remember that EBP cannot
be used as a general purpose base register because it references the stack segment.
• Avoid Compares with Immediate Zero. Often when a value is compared with zero, the
operation producing the value sets condition codes that can be tested directly by a Jcc
instruction. The most notable exceptions are the MOV and LEA instructions. In these
cases, use the TEST instruction.
• Epilog Sequence. If only 4 bytes were allocated in the stack frame for the current function,
instead of incrementing the stack pointer by 4, use POP instructions to prevent AGIs. For
the Pentium® processor, use two pops for eight bytes.

14-35
CODE OPTIMIZATION

14-36
 15
Debugging and
Performance
Monitoring
 CHAPTER 15
DEBUGGING AND PERFORMANCE MONITORING

The Intel Architecture provides extensive debugging facilities for use in debugging code and
monitoring code execution and processor performance. These facilities are valuable for debug-
ging applications software, system software, and multitasking operating systems.
The debugging support is accessed through the debug registers (DB0 through DB7) and two
model-specific registers (MSRs). The debug registers of the Intel Architecture processors hold
the addresses of memory and I/O locations, called breakpoints. Breakpoints are user-selected
locations in a program, a data-storage area in memory, or specific I/O ports where a programmer
or system designer wishes to halt execution of a program and examine the state of the processor
by invoking debugger software. A debug exception (#DB) is generated when a memory or I/O
access is made to one of these breakpoint addresses. A breakpoint is specified for a particular
form of memory or I/O access, such as a memory read and/or write operation or an I/O read
and/or write operation. The debug registers support both instruction breakpoints and data break-
points. The MSRs (which were introduced into the Intel Architecture in the P6 family proces-
sors) monitor branches, interrupts, and exceptions and record the addresses of the last branch,
interrupt or exception taken and the last branch taken before an interrupt or exception.

15.1. OVERVIEW OF THE DEBUGGING SUPPORT FACILITIES

The following processor facilities support debugging and performance monitoring:
• Debug exception (#DB)—Transfers program control to a debugger procedure or task
when a debug event occurs.
• Breakpoint exception (#BP)—Transfers program control to a debugger procedure or task
when an INT 3 instruction is executed.
• Breakpoint-address registers (DB0 through DB3)—Specifies the addresses of up to 4
breakpoints.
• Debug status register (DB6)—Reports the conditions that were in effect when a debug or
breakpoint exception was generated.
• Debug control register (DB7)—Specifies the forms of memory or I/O access that cause
breakpoints to be generated.
• DebugCtlMSR register—Enables last branch, interrupt, and exception recording; taken
branch traps; the breakpoint reporting pins; and trace messages.
• LastBranchToIP and LastBranchFromIP MSRs—Specifies the source and destination
addresses of the last branch, interrupt, or exception taken. The address saved is the offset in
the code segment of the branch (source) or target (destination) instruction.

15-1
DEBUGGING AND PERFORMANCE MONITORING

• LastExceptionToIP and LastExceptionFromIP MSRs—Specifies the source and

destination addresses of the last branch that was taken prior to an exception or interrupt
being generated. The address saved is the offset in the code segment of the branch (source)
or target (destination) instruction.
• T (trap) flag, TSS—Generates a debug exception (#DB) when an attempt is made to
switch to a task with the T flag set in its TSS.
• RF (resume) flag, EFLAGS register— Suppresses multiple exceptions to the same
instruction.
• TF (trap) flag, EFLAGS register—Generates a debug exception (#DB) after every
execution of an instruction.
• Breakpoint instruction (INT 3)—Generates a breakpoint exception (#BP), which
transfers program control to the debugger procedure or task. This instruction is an
alternative way to set code breakpoints. It is especially useful when more than four
breakpoints are desired, or when breakpoints are being placed in the source code.
These facilities allow a debugger to be called either as a separate task or as a procedure in the
context of the current program or task. The following conditions can be used to invoke the
debugger:
• Task switch to a specific task.
• Execution of the breakpoint instruction.
• Execution of any instruction.
• Execution of an instruction at a specified address.
• Read or write of a byte, word, or doubleword at a specified memory address.
• Write to a byte, word, or doubleword at a specified memory address.
• Input of a byte, word, or doubleword at a specified I/O address.
• Output of a byte, word, or doubleword at a specified I/O address.
• Attempt to change the contents of a debug register.

15.2. DEBUG REGISTERS

The eight debug registers (refer to Figure 15-1) control the debug operation of the processor.
These registers can be written to and read using the move to or from debug register form of the
MOV instruction. A debug register may be the source or destination operand for one of these
instructions. The debug registers are privileged resources; a MOV instruction that accesses these
registers can only be executed in real-address mode, in SMM, or in protected mode at a CPL of
0. An attempt to read or write the debug registers from any other privilege level generates a
general-protection exception (#GP).

15-2
 DEBUGGING AND PERFORMANCE MONITORING

31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
R
LEN R/W LEN R/W LEN R/W LEN R/W s
3 3 2 0 0 G 0 0 1v G L G L G L G L G L DR7
2 1 1 0 0 D E E 3 3 2 2 1 1 0 0
d

31 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0

B B B 0 1 1 1 1 1 1 1 1 B B B B
Reserved (set to 1)
T S D 3 2 1 0
DR6

31 0

Reserved DR5

31 0

Reserved DR4

31 0

Breakpoint 3 Linear Address DR3

31 0

Breakpoint 2 Linear Address DR2

31 0

Breakpoint 1 Linear Address DR1

31 0

Breakpoint 0 Linear Address DR0

Reserved Bits, DO NOT DEFINE

Figure 15-1. Debug Registers

15-3
DEBUGGING AND PERFORMANCE MONITORING

The primary function of the debug registers is to set up and monitor from 1 to 4 breakpoints,
numbered 0 though 3. For each breakpoint, the following information can be specified and
detected with the debug registers:
• The linear address where the breakpoint is to occur.
• The length of the breakpoint location (1, 2, or 4 bytes).
• The operation that must be performed at the address for a debug exception to be generated.
• Whether the breakpoint is enabled.
• Whether the breakpoint condition was present when the debug exception was generated.
The following paragraphs describe the functions of flags and fields in the debug registers.

15.2.1. Debug Address Registers (DR0-DR3)

Each of the four debug-address registers (DR0 through DR3) holds the 32-bit linear address of
a breakpoint (refer to Figure 15-1). Breakpoint comparisons are made before physical address
translation occurs. Each breakpoint condition is specified further by the contents of debug
register DR7.

15.2.2. Debug Registers DR4 and DR5

Debug registers DR4 and DR5 are reserved when debug extensions are enabled (when the DE
flag in control register CR4 is set), and attempts to reference the DR4 and DR5 registers cause
an invalid-opcode exception (#UD) to be generated. When debug extensions are not enabled
(when the DE flag is clear), these registers are aliased to debug registers DR6 and DR7.

15.2.3. Debug Status Register (DR6)

The debug status register (DR6) reports the debug conditions that were sampled at the time the
last debug exception was generated (refer to Figure 15-1). Updates to this register only occur
when an exception is generated. The flags in this register show the following information:
B0 through B3 (breakpoint condition detected) flags (bits 0 through 3)
Indicates (when set) that its associated breakpoint condition was met when a
debug exception was generated. These flags are set if the condition described
for each breakpoint by the LENn, and R/Wn flags in debug control register
DR7 is true. They are set even if the breakpoint is not enabled by the Ln and
Gn flags in register DR7.
BD (debug register access detected) flag (bit 13)
Indicates that the next instruction in the instruction stream will access one of
the debug registers (DR0 through DR7). This flag is enabled when the GD
(general detect) flag in debug control register DR7 is set. Refer to Section
15.2.4., “Debug Control Register (DR7)” for further explanation of the purpose
of this flag.

15-4
 DEBUGGING AND PERFORMANCE MONITORING

BS (single step) flag (bit 14)

Indicates (when set) that the debug exception was triggered by the single-step
execution mode (enabled with the TF flag in the EFLAGS register). The single-
step mode is the highest-priority debug exception. When the BS flag is set, any
of the other debug status bits also may be set.
BT (task switch) flag (bit 15)
Indicates (when set) that the debug exception resulted from a task switch where
the T flag (debug trap flag) in the TSS of the target task was set (refer to Section
6.2.1., “Task-State Segment (TSS)”, in Section 6, “Task Management”, for the
format of a TSS). There is no flag in debug control register DR7 to enable or
disable this exception; the T flag of the TSS is the only enabling flag.
Note that the contents of the DR6 register are never cleared by the processor. To avoid any
confusion in identifying debug exceptions, the debug handler should clear the register before
returning to the interrupted program or task.

15.2.4. Debug Control Register (DR7)

The debug control register (DR7) enables or disables breakpoints and sets breakpoint conditions
(refer to Figure 15-1). The flags and fields in this register control the following things:
L0 through L3 (local breakpoint enable) flags (bits 0, 2, 4, and 6)
Enable (when set) the breakpoint condition for the associated breakpoint for
the current task. When a breakpoint condition is detected and its associated Ln
flag is set, a debug exception is generated. The processor automatically clears
these flags on every task switch to avoid unwanted breakpoint conditions in the
new task.
G0 through G3 (global breakpoint enable) flags (bits 1, 3, 5, and 7)
Enable (when set) the breakpoint condition for the associated breakpoint for all
tasks. When a breakpoint condition is detected and its associated Gn flag is set,
a debug exception is generated. The processor does not clear these flags on a
task switch, allowing a breakpoint to be enabled for all tasks.
LE and GE (local and global exact breakpoint enable) flags (bits 8 and 9)
(Not supported in the P6 family processors.) When set, these flags cause the
processor to detect the exact instruction that caused a data breakpoint condi-
tion. For backward and forward compatibility with other Intel Architecture
processors, Intel recommends that the LE and GE flags be set to 1 if exact
breakpoints are required.
GD (general detect enable) flag (bit 13)
Enables (when set) debug-register protection, which causes a debug exception
to be generated prior to any MOV instruction that accesses a debug register.
When such a condition is detected, the BD flag in debug status register DR6 is
set prior to generating the exception. This condition is provided to support in-
circuit emulators. (When the emulator needs to access the debug registers,
emulator software can set the GD flag to prevent interference from the program

15-5
DEBUGGING AND PERFORMANCE MONITORING

currently executing on the processor.) The processor clears the GD flag upon
entering to the debug exception handler, to allow the handler access to the
debug registers.
R/W0 through R/W3 (read/write) fields (bits 16, 17, 20, 21, 24, 25, 28, and 29)
Specifies the breakpoint condition for the corresponding breakpoint. The DE
(debug extensions) flag in control register CR4 determines how the bits in the
R/Wn fields are interpreted. When the DE flag is set, the processor interprets
these bits as follows:
00—Break on instruction execution only.
01—Break on data writes only.
10—Break on I/O reads or writes.
11—Break on data reads or writes but not instruction fetches.
When the DE flag is clear, the processor interprets the R/Wn bits the same as
for the Intel386™ and Intel486™ processors, which is as follows:
00—Break on instruction execution only.
01—Break on data writes only.
10—Undefined.
11—Break on data reads or writes but not instruction fetches.
LEN0 through LEN3 (Length) fields (bits 18, 19, 22, 23, 26, 27, 30, and 31)
Specify the size of the memory location at the address specified in the corre-
sponding breakpoint address register (DR0 through DR3). These fields are
interpreted as follows:
00—1-byte length
01—2-byte length
10—Undefined
11—4-byte length
If the corresponding RWn field in register DR7 is 00 (instruction execution),
then the LENn field should also be 00. The effect of using any other length is
undefined. Refer to Section 15.2.5., “Breakpoint Field Recognition” for further
information on the use of these fields.

15.2.5. Breakpoint Field Recognition

The breakpoint address registers (debug registers DR0 through DR3) and the LENn fields for
each breakpoint define a range of sequential byte addresses for a data or I/O breakpoint. The
LENn fields permit specification of a 1-, 2-, or 4-byte range beginning at the linear address spec-
ified in the corresponding debug register (DRn). Two-byte ranges must be aligned on word
boundaries and 4-byte ranges must be aligned on doubleword boundaries. I/O breakpoint
addresses are zero extended from 16 to 32 bits for purposes of comparison with the breakpoint
address in the selected debug register. These requirements are enforced by the processor; it uses
the LENn field bits to mask the lower address bits in the debug registers. Unaligned data or I/O
breakpoint addresses do not yield the expected results.

15-6
 DEBUGGING AND PERFORMANCE MONITORING

A data breakpoint for reading or writing data is triggered if any of the bytes participating in an
access is within the range defined by a breakpoint address register and its LENn field. Table 15-1
gives an example setup of the debug registers and the data accesses that would subsequently trap
or not trap on the breakpoints.

Table 15-1. Breakpointing Examples

Debug Register Setup
Debug Register R/Wn Breakpoint Address LENn
DR0 R/W0 = 11 (Read/Write) A0001H LEN0 = 00 (1 byte)
DR1 R/W1 = 01 (Write) A0002H LEN1 = 00 (1 byte)
DR2 R/W2 = 11 (Read/Write) B0002H LEN2 = 01) (2 bytes)
DR3 R/W3 = 01 (Write) C0000H LEN3 = 11 (4 bytes)
Data Accesses
Access Length
Operation Address (In Bytes)
Data operations that trap
- Read or write A0001H 1
- Read or write A0001H 2
- Write A0002H 1
- Write A0002H 2
- Read or write B0001H 4
- Read or write B0002H 1
- Read or write B0002H 2
- Write C0000H 4
- Write C0001H 2
- Write C0003H 1
Data operations that do not trap
- Read or write A0000H 1
- Read A0002H 1
- Read or write A0003H 4
- Read or write B0000H 2
- Read C0000H 2
- Read or write C0004H 4

A data breakpoint for an unaligned operand can be constructed using two breakpoints, where
each breakpoint is byte-aligned, and the two breakpoints together cover the operand. These
breakpoints generate exceptions only for the operand, not for any neighboring bytes.
Instruction breakpoint addresses must have a length specification of 1 byte (the LENn field is
set to 00). The behavior of code breakpoints for other operand sizes is undefined. The processor
recognizes an instruction breakpoint address only when it points to the first byte of an instruc-
tion. If the instruction has any prefixes, the breakpoint address must point to the first prefix.

15.3. DEBUG EXCEPTIONS

The Intel Architecture processors dedicate two interrupt vectors to handling debug exceptions:
vector 1 (debug exception, #DB) and vector 3 (breakpoint exception, #BP). The following

15-7
DEBUGGING AND PERFORMANCE MONITORING

sections describe how these exceptions are generated and typical exception handler operations
for handling these exceptions.

15.3.1. Debug Exception (#DB)—Interrupt Vector 1

The debug-exception handler is usually a debugger program or is part of a larger software
system. The processor generates a debug exception for any of several conditions. The debugger
can check flags in the DR6 and DR7 registers to determine which condition caused the exception
and which other conditions might also apply. Table 15-2 shows the states of these flags
following the generation of each kind of breakpoint condition.

Table 15-2. Debug Exception Conditions

Debug or Breakpoint Condition DR6 Flags Tested DR7 Flags Tested Exception Class
Single-step trap BS = 1 Trap
Instruction breakpoint, at addresses Bn = 1 and R/Wn = 0 Fault
defined by DRn and LENn (GEn or LEn = 1)
Data write breakpoint, at addresses Bn = 1 and R/Wn = 1 Trap
defined by DRn and LENn (GEn or LEn = 1)
I/O read or write breakpoint, at addresses Bn = 1 and R/Wn = 2 Trap
defined by DRn and LENn (GEn or LEn = 1)
Data read or write (but not instruction Bn = 1 and R/Wn = 3 Trap
fetches), at addresses defined by DRn (GEn or LEn = 1)
and LENn
General detect fault, resulting from an BD = 1 Fault
attempt to modify debug registers
(usually in conjunction with in-circuit
emulation)
Task switch BT = 1 Trap

Instruction-breakpoint and general-detect conditions (refer to Section 15.3.1.3., “General-

Detect Exception Condition”) result in faults; other debug-exception conditions result in traps.
The debug exception may report either or both at one time. The following sections describe each
class of debug exception. Refer to Section 5.12., “Exception and Interrupt Reference” in Chapter
5, Interrupt and Exception Handling for additional information about this exception.

15.3.1.1. INSTRUCTION-BREAKPOINT EXCEPTION CONDITION

The processor reports an instruction breakpoint when it attempts to execute an instruction at an
address specified in a breakpoint-address register (DB0 through DR3) that has been set up to
detect instruction execution (R/W flag is set to 0). Upon reporting the instruction breakpoint, the
processor generates a fault-class, debug exception (#DB) before it executes the target instruction

15-8
 DEBUGGING AND PERFORMANCE MONITORING

for the breakpoint. Instruction breakpoints are the highest priority debug exceptions and are
guaranteed to be serviced before any other exceptions that may be detected during the decoding
or execution of an instruction.
Because the debug exception for an instruction breakpoint is generated before the instruction is
executed, if the instruction breakpoint is not removed by the exception handler, the processor
will detect the instruction breakpoint again when the instruction is restarted and generate another
debug exception. To prevent looping on an instruction breakpoint, the Intel Architecture
provides the RF flag (resume flag) in the EFLAGS register (refer to Section 2.3., “System Flags
and Fields in the EFLAGS Register” in Chapter 2, System Architecture Overview). When the RF
flag is set, the processor ignores instruction breakpoints.
All Intel Architecture processors manage the RF flag as follows. The processor sets the RF flag
automatically prior to calling an exception handler for any fault-class exception except a debug
exception that was generated in response to an instruction breakpoint. For debug exceptions
resulting from instruction breakpoints, the processor does not set the RF flag prior to calling the
debug exception handler. The debug exception handler then has the option of disabling the
instruction breakpoint or setting the RF flag in the EFLAGS image on the stack. If the RF flag
in the EFLAGS image is set when the processor returns from the exception handler, it is copied
into the RF flag in the EFLAGS register by the IRETD or task switch instruction that causes the
return. The processor then ignores instruction breakpoints for the duration of the next instruc-
tion. (Note that the POPF, POPFD, and IRET instructions do not transfer the RF image into the
EFLAGS register.) Setting the RF flag does not prevent other types of debug-exception condi-
tions (such as, I/O or data breakpoints) from being detected, nor does it prevent nondebug excep-
tions from being generated. After the instruction is successfully executed, the processor clears
the RF flag in the EFLAGS register, except after an IRETD instruction or after a JMP, CALL,
or INT n instruction that causes a task switch. (Note that the processor also does not set the RF
flag when calling exception or interrupt handlers for trap-class exceptions, for hardware inter-
rupts, or for software-generated interrupts.)
For the Pentium® processor, when an instruction breakpoint coincides with another fault-type
exception (such as a page fault), the processor may generate one spurious debug exception after
the second exception has been handled, even though the debug exception handler set the RF flag
in the EFLAGS image. To prevent this spurious exception with Pentium® processors, all fault-
class exception handlers should set the RF flag in the EFLAGS image.

15.3.1.2. DATA MEMORY AND I/O BREAKPOINT EXCEPTION CONDITIONS

Data memory and I/O breakpoints are reported when the processor attempts to access a memory
or I/O address specified in a breakpoint-address register (DB0 through DR3) that has been set
up to detect data or I/O accesses (R/W flag is set to 1, 2, or 3). The processor generates the excep-
tion after it executes the instruction that made the access, so these breakpoint condition causes
a trap-class exception to be generated.
Because data breakpoints are traps, the original data is overwritten before the trap exception is
generated. If a debugger needs to save the contents of a write breakpoint location, it should save
the original contents before setting the breakpoint. The handler can report the saved value after
the breakpoint is triggered. The address in the debug registers can be used to locate the new
value stored by the instruction that triggered the breakpoint.

15-9
DEBUGGING AND PERFORMANCE MONITORING

The Intel486™ and later Intel Architecture processors ignore the GE and LE flags in DR7. In
the Intel386™ processor, exact data breakpoint matching does not occur unless it is enabled by
setting the LE and/or the GE flags.
The P6 family processors, however, are unable to report data breakpoints exactly for the REP
MOVS and REP STOS instructions until the completion of the iteration after the iteration in
which the breakpoint occurred.
For repeated INS and OUTS instructions that generate an I/O-breakpoint debug exception, the
processor generates the exception after the completion of the first iteration. Repeated INS and
OUTS instructions generate an I/O-breakpoint debug exception after the iteration in which the
memory address breakpoint location is accessed.

15.3.1.3. GENERAL-DETECT EXCEPTION CONDITION

When the GD flag in DR7 is set, the general-detect debug exception occurs when a program
attempts to access any of the debug registers (DR0 through DR7) at the same time they are being
used by another application, such as an emulator or debugger. This additional protection feature
guarantees full control over the debug registers when required. The debug exception handler can
detect this condition by checking the state of the BD flag of the DR6 register. The processor
generates the exception before it executes the MOV instruction that accesses a debug register,
which causes a fault-class exception to be generated.

15.3.1.4. SINGLE-STEP EXCEPTION CONDITION

The processor generates a single-step debug exception if (while an instruction is being executed)
it detects that the TF flag in the EFLAGS register is set. The exception is a trap-class exception,
because the exception is generated after the instruction is executed. (Note that the processor does
not generate this exception after an instruction that sets the TF flag. For example, if the POPF
instruction is used to set the TF flag, a single-step trap does not occur until after the instruction
that follows the POPF instruction.)
The processor clears the TF flag before calling the exception handler. If the TF flag was set in a
TSS at the time of a task switch, the exception occurs after the first instruction is executed in the
new task.
The TF flag normally is not cleared by privilege changes inside a task. The INT n and INTO
instructions, however, do clear this flag. Therefore, software debuggers that single-step code
must recognize and emulate INT n or INTO instructions rather than executing them directly. To
maintain protection, the operating system should check the CPL after any single-step trap to see
if single stepping should continue at the current privilege level.
The interrupt priorities guarantee that, if an external interrupt occurs, single stepping stops.
When both an external interrupt and a single-step interrupt occur together, the single-step inter-
rupt is processed first. This operation clears the TF flag. After saving the return address or
switching tasks, the external interrupt input is examined before the first instruction of the single-
step handler executes. If the external interrupt is still pending, then it is serviced. The external
interrupt handler does not run in single-step mode. To single step an interrupt handler, set a break
point inside the handler and then set the TF flag.

15-10
 DEBUGGING AND PERFORMANCE MONITORING

15.3.1.5. TASK-SWITCH EXCEPTION CONDITION

The processor generates a debug exception after a task switch if the T flag of the new task’s TSS
is set. This exception is generated after program control has passed to the new task, and after the
first instruction of that task is executed. The exception handler can detect this condition by
examining the BT flag of the DR6 register.
Note that, if the debug exception handler is a task, the T bit of its TSS should not be set. Failure
to observe this rule will put the processor in a loop.

15.3.2. Breakpoint Exception (#BP)—Interrupt Vector 3

The breakpoint exception (interrupt 3) is caused by execution of an INT 3 instruction (refer to
Section 5.12., “Exception and Interrupt Reference” in Chapter 5, Interrupt and Exception
Handling). Debuggers use break exceptions in the same way that they use the breakpoint regis-
ters; that is, as a mechanism for suspending program execution to examine registers and memory
locations. With earlier Intel Architecture processors, breakpoint exceptions are used extensively
for setting instruction breakpoints. With the Intel386™ and later Intel Architecture processors,
it is more convenient to set breakpoints with the breakpoint-address registers (DR0 through
DR3). However, the breakpoint exception still is useful for breakpointing debuggers, because
the breakpoint exception can call a separate exception handler. The breakpoint exception is also
useful when it is necessary to set more breakpoints than there are debug registers or when break-
points are being placed in the source code of a program under development.

15.4. LAST BRANCH, INTERRUPT, AND EXCEPTION RECORDING

The P6 family processors provide five MSRs for recording the last branch, interrupt, or excep-
tion taken by the processor: DebugCtlMSR, LastBranchToIP, LastBranchFromIP, LastExcepti-
onToIP, and LastExceptionFromIP. These registers can be used to set breakpoints on branches,
interrupts, and exceptions, and to single-step from one branch to the next.

15.4.1. DebugCtlMSR Register

The DebugCtlMSR register enables last branch, interrupt, and exception recording; taken
branch breakpoints; the breakpoint reporting pins; and trace messages. This register can be
written to using the WRMSR instruction, when operating at privilege level 0 or when in real-
address mode. A protected-mode operating system procedure is required to provide user access
to this register. Figure 15-2 shows the flags in the DebugCtlMSR register. The functions of these
flags are as follows:
LBR (last branch/interrupt/exception) flag (bit 0)
When set, the processor records the source and target addresses for the last
branch and the last exception or interrupt taken by the processor prior to a
debug exception being generated. The processor clears this flag whenever a
debug exception, such as an instruction or data breakpoint or single-step trap
occurs.

15-11
DEBUGGING AND PERFORMANCE MONITORING

31 7 6 5 4 3 2 1 0
P P P P B L
T B B B B T B
R 3 2 1 0 F R

TR—Trace messages enable

PBi—Performance monitoring/breakpoint pins
BTF—Single-step on branches
LBR—Last branch/interrupt/exception

Reserved

Figure 15-2. DebugCtlMSR Register

BTF (single-step on branches) flag (bit 1)

When set, the processor treats the TF flag in the EFLAGS register as a “single-
step on branches” flag rather than a “single-step on instructions” flag. This
mechanism allows single-stepping the processor on taken branches. Software
must set both the BTF and TF flag to enable debug breakpoints on branches;
the processor clears both flags whenever a debug exception occurs.
PBi (performance monitoring/breakpoint pins) flags (bits 2 through 5)
When these flags are set, the performance monitoring/breakpoint pins on the
processor (BP0#, BP1#, BP2#, and BP3#) report breakpoint matches in the
corresponding breakpoint-address registers (DR0 through DR3). The
processor asserts then deasserts the corresponding BPi# pin when a breakpoint
match occurs. When a PBi flag is clear, the performance monitoring/breakpoint
pins report performance events. Processor execution is not affected by
reporting performance events.
TR (trace message enable) flag (bit 6)
When set, trace messages are enabled. Thereafter, when the processor detects
a branch, exception, or interrupt, it sends the “to” and “from” addresses out on
the system bus as part of a branch trace message. A debugging device that is
monitoring the system bus can read these messages and synchronize operations
with branch, exception, and interrupt events. Setting this flag greatly reduces
the performance of the processor. When trace messages are enabled, the values
stored in the LastBranchToIP, LastBranchFromIP, LastExceptionToIP, and
LastExceptionFromIP MSRs are undefined.
Note that the “from” addresses sent out on the system bus may differ from
those stored in the LastBranchFromIP MSRs or LastExceptionFromIP MSRs.
The from address sent out on the bus is always the next instruction in the
instruction stream following a successfully completed instruction. For
example, if a branch completes successfully, the address stored in the Last-
BranchFromIP MSR is the address of the branch instruction, but the address
sent out on the bus in the trace message is the address of the instruction

15-12
 DEBUGGING AND PERFORMANCE MONITORING

following the branch instruction. If the processor faults on the branch, the
address stored in the LastBranchFromIP MSR is again the address of the
branch instruction and that same address is sent out on the bus.

15.4.2. Last Branch and Last Exception MSRs

The LastBranchToIP and LastBranchFromIP MSRs are 32-bit registers for recording the
instruction pointers for the last branch, interrupt, or exception that the processor took prior to a
debug exception being generated (refer to Figure 15-2). When a branch occurs, the processor
loads the address of the branch instruction into the LastBranchFromIP MSR and loads the target
address for the branch into the LastBranchToIP MSR. When an interrupt or exception occurs
(other than a debug exception), the address of the instruction that was interrupted by the excep-
tion or interrupt is loaded into the LastBranchFromIP MSR and the address of the exception or
interrupt handler that is called is loaded into the LastBranchToIP MSR.
The LastExceptionToIP and LastExceptionFromIP MSRs (also 32-bit registers) record the
instruction pointers for the last branch that the processor took prior to an exception or interrupt
being generated. When an exception or interrupt occurs, the contents of the LastBranchToIP and
LastBranchFromIP MSRs are copied into these registers before the to and from addresses of the
exception or interrupt are recorded in the LastBranchToIP and LastBranchFromIP MSRs.
These registers can be read using the RDMSR instruction.

15.4.3. Monitoring Branches, Exceptions, and Interrupts

When the LBR flag in the DebugCtlMSR register is set, the processor automatically begins
recording branches that it takes, exceptions that are generated (except for debug exceptions), and
interrupts that are serviced. Each time a branch, exception, or interrupt occurs, the processor
records the to and from instruction pointers in the LastBranchToIP and LastBranchFromIP
MSRs. In addition, for interrupts and exceptions, the processor copies the contents of the Last-
BranchToIP and LastBranchFromIP MSRs into the LastExceptionToIP and LastException-
FromIP MSRs prior to recording the to and from addresses of the interrupt or exception.
When the processor generates a debug exception (#DB), it automatically clears the LBR flag
before executing the exception handler, but does not touch the last branch and last exception
MSRs. The addresses for the last branch, interrupt, or exception taken are thus retained in the
LastBranchToIP and LastBranchFromIP MSRs and the addresses of the last branch prior to an
interrupt or exception are retained in the LastExceptionToIP, and LastExceptionFromIP MSRs.
The debugger can use the last branch, interrupt, and/or exception addresses in combination with
code-segment selectors retrieved from the stack to reset breakpoints in the breakpoint-address
registers (DR0 through DR3), allowing a backward trace from the manifestation of a particular
bug toward its source. Because the instruction pointers recorded in the LastBranchToIP, Last-
BranchFromIP, LastExceptionToIP, and LastExceptionFromIP MSRs are offsets into a code
segment, software must determine the segment base address of the code segment associated with

15-13
DEBUGGING AND PERFORMANCE MONITORING

the control transfer to calculate the linear address to be placed in the breakpoint-address regis-
ters. The segment base address can be determined by reading the segment selector for the code
segment from the stack and using it to locate the segment descriptor for the segment in the GDT
or LDT. The segment base address can then be read from the segment descriptor.
Before resuming program execution from a debug-exception handler, the handler should set the
LBR flag again to re-enable last branch and last exception/interrupt recording.

15.4.4. Single-Stepping on Branches, Exceptions, and Interrupts

When the BTF flag in the DebugCtlMSR register and the TF flag in the EFLAGS register are
both set, the processor generates a single-step debug exception the next time it takes a branch,
generates an exception, or services an interrupt. This mechanism allows the debugger to single-
step on control transfers caused by branches, exceptions, or interrupts. This “control-flow single
stepping” helps isolate a bug to a particular block of code before instruction single-stepping
further narrows the search. If the BTF flag is set when the processor generates a debug excep-
tion, the processor clears the flag along with the TF flag. The debugger must reset the BTF flag
before resuming program execution to continue control-flow single stepping.

15.4.5. Initializing Last Branch or Last Exception/Interrupt

Recording
The LastBranchToIP, LastBranchFromIP, LastExceptionToIP, and LastException-FromIP
MSRs are enabled by setting the LBR flag in the DebugCtlMSR register. Control-flow single
stepping is enabled by setting the BTF flag in the DebugCtlMSR register. The processor clears
both the LBR and the BTF flags whenever a debug exception is generated. To re-enable these
mechanisms, the debug-exception handler must thus explicitly set these flags before returning
to the interrupted program.

15.5. TIME-STAMP COUNTER

The Intel Architecture (beginning with the Pentium® processor) defines a time-stamp counter
mechanism that can be used to monitor and identify the relative time of occurrence of processor
events. The time-stamp counter architecture includes an instruction for reading the time-stamp
counter (RDTSC), a feature bit (TCS flag) that can be read with the CPUID instruction, a time-
stamp counter disable bit (TSD flag) in control register CR4, and a model-specific time-stamp
counter.
Following execution of the CPUID instruction, the TSC flag in register EDX (bit 4) indicates
(when set) that the time-stamp counter is present in a particular Intel Architecture processor
implementation. (Refer to “CPUID—CPU Identification” in Chapter 3 of the Intel Architecture
Software Developer’s Manual, Volume 2.)
The time-stamp counter (as implemented in the Pentium® and P6 family processors) is a 64-bit
counter that is set to 0 following the hardware reset of the processor. Following reset, the counter

15-14
 DEBUGGING AND PERFORMANCE MONITORING

is incremented every processor clock cycle, even when the processor is halted by the HLT
instruction or the external STPCLK# pin.
The RDTSC instruction reads the time-stamp counter and is guaranteed to return a monotoni-
cally increasing unique value whenever executed, except for 64-bit counter wraparound. Intel
guarantees, architecturally, that the time-stamp counter frequency and configuration will be such
that it will not wraparound within 10 years after being reset to 0. The period for counter wrap is
several thousands of years in the Pentium® and P6 family processors.
Normally, the RDTSC instruction can be executed by programs and procedures running at any
privilege level and in virtual-8086 mode. The TSD flag in control register CR4 (bit 2) allows
use of this instruction to be restricted to only programs and procedures running at privilege level
0. A secure operating system would set the TSD flag during system initialization to disable user
access to the time-stamp counter. An operating system that disables user access to the time-
stamp counter should emulate the instruction through a user-accessible programming interface.
The RDTSC instruction is not serializing or ordered with other instructions. Thus, it does not
necessarily wait until all previous instructions have been executed before reading the counter.
Similarly, subsequent instructions may begin execution before the RDTSC instruction operation
is performed.
The RDMSR and WRMSR instructions can read and write the time-stamp counter, respectively,
as a model-specific register (TSC). The ability to read and write the time-stamp counter with the
RDMSR and WRMSR instructions is not an architectural feature, and may not be supported by
future Intel Architecture processors. Writing to the time-stamp counter with the WRMSR
instruction resets the count. Only the low order 32-bits of the time-stamp counter can be written
to; the high-order 32 bits are 0 extended (cleared to all 0s).

15.6. PERFORMANCE-MONITORING COUNTERS

The Pentium® processor introduced model-specific performance-monitoring counters to the
Intel Architecture. These counters permit processor performance parameters to be monitored
and measured. The information obtained from these counters can then be used for tuning system
and compiler performance.
In the Intel P6 family of processors, the performance-monitoring counter mechanism was modi-
fied and enhanced to permit a wider variety of events to be monitored and to allow greater
control over the selection of the events to be monitored.
The following sections describe the performance-monitoring counter mechanism in the
Pentium® and P6 family processors.

15.6.1. P6 Family Processor Performance-Monitoring Counters

The P6 family processors provide two 40-bit performance counters, allowing two types of
events to be monitored simultaneously. These counters can either count events or measure dura-
tion. When counting events, a counter is incremented each time a specified event takes place or
a specified number of events takes place. When measuring duration, a counter counts the

15-15
DEBUGGING AND PERFORMANCE MONITORING

number of processor clocks that occur while a specified condition is true. The counters can count
events or measure durations that occur at any privilege level. Table A-1 in Appendix A, Perfor-
mance-Monitoring Events lists the events that can be counted with the P6 family performance
monitoring counters.
The performance-monitoring counters are supported by four MSRs: the performance event
select MSRs (PerfEvtSel0 and PerfEvtSel1) and the performance counter MSRs (PerfCtr0 and
PerfCtr1). These registers can be read from and written to using the RDMSR and WRMSR
instructions, respectively. They can be accessed using these instructions only when operating at
privilege level 0. The PerfCtr0 and PerfCtr1 MSRs can be read from any privilege level using
the RDPMC (read performance-monitoring counters) instruction.

NOTE
The PerfEvtSel0, PerfEvtSel1, PerfCtr0, and PerfCtr1 MSRs and the events
listed in Table A-1 in Appendix A, Performance-Monitoring Events are
model-specific for P6 family processors. They are not guaranteed to be
available in future Intel Architecture processors.

15.6.1.1. PERFEVTSEL0 AND PERFEVTSEL1 MSRS

The PerfEvtSel0 and PerfEvtSel1 MSRs control the operation of the performance-monitoring
counters, with one register used to set up each counter. They specify the events to be counted,
how they should be counted, and the privilege levels at which counting should take place. Figure
15-3 shows the flags and fields in these MSRs.
The functions of the flags and fields in the PerfEvtSel0 and PerfEvtSel1 MSRs are as follows:
Event select field (bits 0 through 7)
Selects the event to be monitored (refer to Table A-1 in Appendix A, Perfor-
mance-Monitoring Events for a list of events and their 8-bit codes).
Unit mask field (bits 8 through 15)
Further qualifies the event selected in the event select field. For example, for
some cache events, the mask is used as a MESI-protocol qualifier of cache
states (refer to Table A-1 in Appendix A, Performance-Monitoring Events).
USR (user mode) flag (bit 16)
Specifies that events are counted only when the processor is operating at priv-
ilege levels 1, 2 or 3. This flag can be used in conjunction with the OS flag.
OS (operating system mode) flag (bit 17)
Specifies that events are counted only when the processor is operating at priv-
ilege level 0. This flag can be used in conjunction with the USR flag.

15-16
 DEBUGGING AND PERFORMANCE MONITORING

31 24 23 22 21 20 19 18 17 16 15 8 7 0
I I U
Counter Mask N E N P E O S Unit Mask Event Select
V N T C S R

INV—Invert counter mask

EN—Enable counters*
INT—APIC interrupt enable
PC—Pin control
E—Edge detect
OS—Operating system mode
USR—User Mode

* Only available in PerfEvtSel0.

Reserved

Figure 15-3. PerfEvtSel0 and PerfEvtSel1 MSRs

E (edge detect) flag (bit 18)

Enables (when set) edge detection of events. The processor counts the number
of deasserted to asserted transitions of any condition that can be expressed by
the other fields. The mechanism is limited in that it does not permit back-to-
back assertions to be distinguished. This mechanism allows software to
measure not only the fraction of time spent in a particular state, but also the
average length of time spent in such a state (for example, the time spent waiting
for an interrupt to be serviced).
PC (pin control) flag (bit 19)
When set, the processor toggles the PMi pins and increments the counter when
performance-monitoring events occur; when clear, the processor toggles the
PMi pins when the counter overflows. The toggling of a pin is defined as asser-
tion of the pin for a single bus clock followed by deassertion
INT (APIC interrupt enable) flag (bit 20)
When set, the processor generates an exception through its local APIC on
counter overflow.
EN (Enable Counters) Flag (bit 22)
This flag is only present in the PerfEvtSel0 MSR. When set, performance
counting is enabled in both performance-monitoring counters; when clear, both
counters are disabled.
INV (invert) flag (bit 23)
Inverts the result of the counter-mask comparison when set, so that both greater
than and less than comparisons can be made.
Counter mask field (bits 24 through 31)
When nonzero, the processor compares this mask to the number of events

15-17
DEBUGGING AND PERFORMANCE MONITORING

counted during a single cycle. If the event count is greater than or equal to this
mask, the counter is incremented by one. Otherwise the counter is not incre-
mented. This mask can be used to count events only if multiple occurrences
happen per clock (for example, two or more instructions retired per clock). If
the counter-mask field is 0, then the counter is incremented each cycle by the
number of events that occurred that cycle.

15.6.1.2. PERFCTR0 AND PERFCTR1 MSRS

The performance-counter MSRs (PerfCtr0 and PerfCtr1) contain the event or duration counts
for the selected events being counted. The RDPMC instruction can be used by programs or
procedures running at any privilege level and in virtual-8086 mode to read these counters. The
PCE flag in control register CR4 (bit 8) allows the use of this instruction to be restricted to only
programs and procedures running at privilege level 0.
The RDPMC instruction is not serializing or ordered with other instructions. Thus, it does not
necessarily wait until all previous instructions have been executed before reading the counter.
Similarly, subsequent instructions may begin execution before the RDPMC instruction opera-
tion is performed.
Only the operating system, executing at privilege level 0, can directly manipulate the perfor-
mance counters, using the RDMSR and WRMSR instructions. A secure operating system would
set the TSD flag during system initialization to disable direct user access to the performance-
monitoring counters, but provide a user-accessible programming interface that emulates the
RDPMC instruction.
The WRMSR instruction cannot arbitrarily write to the performance-monitoring counter MSRs
(PerfCtr0 and PerfCtr1). Instead, the lower-order 32 bits of each MSR may be written with any
value, and the high-order 8 bits are sign-extended according to the value of bit 31. This operation
allows writing both positive and negative values to the performance counters.

15.6.1.3. STARTING AND STOPPING THE PERFORMANCE-MONITORING

COUNTERS
The performance-monitoring counters are started by writing valid setup information in the
PerfEvtSel0 and/or PerfEvtSel1 MSRs and setting the enable counters flag in the PerfEvtSel0
MSR. If the setup is valid, the counters begin counting following the execution of a WRMSR
instruction that sets the enable counter flag. The counters can be stopped by clearing the enable
counters flag or by clearing all the bits in the PerfEvtSel0 and PerfEvtSel1 MSRs. Counter 1
alone can be stopped by clearing the PerfEvtSel1 MSR.

15.6.1.4. EVENT AND TIME-STAMP MONITORING SOFTWARE

To use the performance-monitoring counters and time-stamp counter, the operating system
needs to provide an event-monitoring device driver. This driver should include procedures for
handling the following operations:
• Feature checking.
• Initialize and start counters.

15-18
 DEBUGGING AND PERFORMANCE MONITORING

• Stop counters.
• Read the event counters.
• Read the time-stamp counter.
The event monitor feature determination procedure must determine whether the current
processor supports the performance-monitoring counters and time-stamp counter. This proce-
dure compares the family and model of the processor returned by the CPUID instruction with
those of processors known to support performance monitoring. (The Pentium® and P6 family
processors support performance counters.) The procedure also checks the MSR and TSC flags
returned to register EDX by the CPUID instruction to determine if the MSRs and the RDTSC
instruction are supported.
The initialize and start counters procedure sets the PerfEvtSel0 and/or PerfEvtSel1 MSRs for
the events to be counted and the method used to count them and initializes the counter MSRs
(PerfCtr0 and PerfCtr1) to starting counts. The stop counters procedure stops the performance
counters. (Refer to Section 15.6.1.3., “Starting and Stopping the Performance-Monitoring
Counters” for more information about starting and stopping the counters.)
The read counters procedure reads the values in the PerfCtr0 and PerfCtr1 MSRs, and a read
time-stamp counter procedure reads the time-stamp counter. These procedures would be
provided in lieu of enabling the RDTSC and RDPMC instructions that allow application code
to read the counters.

15.6.2. Monitoring Counter Overflow

The P6 family processors provide the option of generating a local APIC interrupt when a perfor-
mance-monitoring counter overflows. This mechanism is enabled by setting the interrupt enable
flag in either the PerfEvtSel0 or the PerfEvtSel1 MSR. The primary use of this option is for
statistical performance sampling.
To use this option, the operating system should do the following things on the processor for
which performance events are required to be monitored:
• Provide an interrupt vector for handling the counter-overflow interrupt.
• Initialize the APIC PERF local vector entry to enable handling of performance-monitor
counter overflow events.
• Provide an entry in the IDT that points to a stub exception handler that returns without
executing any instructions.
• Provide an event monitor driver that provides the actual interrupt handler and modifies the
reserved IDT entry to point to its interrupt routine.
When interrupted by a counter overflow, the interrupt handler needs to perform the following
actions:
• Save the instruction pointer (EIP register), code-segment selector, TSS segment selector,
counter values and other relevant information at the time of the interrupt.

15-19
DEBUGGING AND PERFORMANCE MONITORING

• Reset the counter to its initial setting and return from the interrupt.
An event monitor application utility or another application program can read the information
collected for analysis of the performance of the profiled application.

15.6.3. Pentium® Processor Performance-Monitoring Counters

The Pentium® processor provides two 40-bit performance counters, which can be used either to
count events or measure duration. The performance-monitoring counters are supported by three
MSRs: the control and event select MSR (CESR) and the performance counter MSRs (CTR0
and CTR1). These registers can be read from and written to using the RDMSR and WRMSR
instructions, respectively. They can be accessed using these instructions only when operating at
privilege level 0. Each counter has an associated external pin (PM0/BP0 and PM1/BP1), which
can be used to indicate the state of the counter to external hardware.

NOTE
The CESR, CTR0, and CTR1 MSRs and the events listed in Table A-1 in
Appendix A, Performance-Monitoring Events are model-specific for the
Pentium® processor.

15.6.3.1. CONTROL AND EVENT SELECT REGISTER (CESR)

The 32-bit control and event select MSR (CESR) is used to control the operation of perfor-
mance-monitoring counters CTR0 and CTR1 and their associated pins (refer to Figure 15-3). To
control each counter, the CESR register contains a 6-bit event select field (ES0 and ES1), a pin
control flag (PC0 and PC1), and a 3-bit counter control field (CC0 and CC1). The functions of
these fields are as follows:
ES0 and ES1 (event select) fields (bits 0 through 5, bits 16 through 21)
Selects (by entering an event code in the field) up to two events to be moni-
tored. Refer to Table A-1 in Appendix A, Performance-Monitoring Events for
a list of available event codes
CC0 and CC1 (counter control) fields (bits 6 through 8, bits 22 through 24)
Controls the operation of the counter. The possible control codes are as
follows:

CCn Meaning
000 Count nothing (counter disabled)
001 Count the selected event while CPL is 0, 1, or 2
010 Count the selected event while CPL is 3
011 Count the selected event regardless of CPL
100 Count nothing (counter disabled)
101 Count clocks (duration) while CPL is 0, 1, or 2
110 Count clocks (duration) while CPL is 3
111 Count clocks (duration) regardless of CPL

15-20
 DEBUGGING AND PERFORMANCE MONITORING

Note that the highest order bit selects between counting events and counting
clocks (duration); the middle bit enables counting when the CPL is 3; and the
low-order bit enables counting when the CPL is 0, 1, or 2.

31 26 25 24 22 21 16 15 10 9 8 6 5 0
P P
C CC1 ES1 C CC0 ESO
1 0

PC1—Pin control 1
CC1—Counter control 1
ES1—Event select 1
PC0—Pin control 0
CC0—Counter control 0
ES0—Event select 0

Reserved

Figure 15-4. CESR MSR (Pentium® Processor Only)

PC0 and PC1 (pin control) flags (bit 9, bits 25)

Selects the function of the external performance-monitoring counter pin
(PM0/BP0 and PM1/BP1). Setting one of these flags to 1 causes the processor
to assert its associated pin when the counter has overflowed; setting the flag to
0 causes the pin to be asserted when the counter has been incremented. These
flags permit the pins to be individually programmed to indicate the overflow or
incremented condition. Note that the external signaling of the event on the pins
will lag the internal event by a few clocks as the signals are latched and buff-
ered.
While a counter need not be stopped to sample its contents, it must be stopped and cleared or
preset before switching to a new event. It is not possible to set one counter separately. If only
one event needs to be changed, the CESR register must be read, the appropriate bits modified,
and all bits must then be written back to CESR. At reset, all bits in the CESR register are cleared.

15.6.3.2. USE OF THE PERFORMANCE-MONITORING PINS

When the performance-monitor pins PM0/BP0 and/or PM1/BP1 are configured to indicate
when the performance-monitor counter has incremented and an “occurrence event” is being
counted, the associated pin is asserted (high) each time the event occurs. When a “duration
event” is being counted the associated PM pin is asserted for the entire duration of the event.
When the performance-monitor pins are configured to indicate when the counter has over-
flowed, the associated PM pin is not asserted until the counter has overflowed.
When the PM0/BP0 and/or PM1/BP1 pins are configured to signal that a counter has incre-
mented, it should be noted that although the counters may increment by 1 or 2 in a single clock,

15-21
DEBUGGING AND PERFORMANCE MONITORING

the pins can only indicate that the event occurred. Moreover, since the internal clock frequency
may be higher than the external clock frequency, a single external clock may correspond to
multiple internal clocks.
A “count up to” function may be provided when the event pin is programmed to signal an over-
flow of the counter. Because the counters are 40 bits, a carry out of bit 39 indicates an overflow.
A counter may be preset to a specific value less then 240 − 1. After the counter has been enabled
and the prescribed number of events has transpired, the counter will overflow. Approximately 5
clocks later, the overflow is indicated externally and appropriate action, such as signaling an
interrupt, may then be taken.
The PM0/BP0 and PM1/BP1 pins also serve to indicate breakpoint matches during in-circuit
emulation, during which time the counter increment or overflow function of these pins is not
available. After RESET, the PM0/BP0 and PM1/BP1 pins are configured for performance moni-
toring, however a hardware debugger may reconfigure these pins to indicate breakpoint
matches.

15.6.3.3. EVENTS COUNTED

The events that the performance-monitoring counters can set to count and record in the CTR0
and CTR1 MSRs are divided into two categories: occurrences and duration. Occurrences events
are counted each time the event takes place. If the PM0/BP0 or PM1/BP1 pins are configured to
indicate when a counter increments, they ar asserted each clock the counter increments. Note
that if an event can happen twice in one clock, the counter increments by 2, however, the pins
are asserted only once.
For duration events, the counter counts the total number of clocks that the condition is true.
When configured to indicate when a counter increments, the PM0/BP0 and/or PM1/BP1 pins
are asserted for the duration of the event.
Table A-2 in Appendix A, Performance-Monitoring Events lists the events that can be counted
with the Pentium® processor performance-monitoring counters.

15-22
 16
8086 Emulation
 8086 EMULATION

CHAPTER 16
8086 EMULATION

Intel Architecture processors (beginning with the Intel386™ processor) provide two ways to
execute new or legacy programs that are assembled and/or compiled to run on an Intel 8086
processor:
• Real-address mode.
• Virtual-8086 mode.
Figure 2-2 in Chapter 2, System Architecture Overview shows the relationship of these operating
modes to protected mode and system management mode (SMM).
When the processor is powered up or reset, it is placed in the real-address mode. This operating
mode almost exactly duplicates the execution environment of the Intel 8086 processor, with
some extensions. Virtually any program assembled and/or compiled to run on an Intel 8086
processor will run on an Intel Architecture processor in this mode.
When running in protected mode, the processor can be switched to virtual-8086 mode to run
8086 programs. This mode also duplicates the execution environment of the Intel 8086
processor, with extensions. In virtual-8086 mode, an 8086 program runs as a separate protected-
mode task. Legacy 8086 programs are thus able to run under an operating system (such as
Microsoft Windows*) that takes advantage of protected mode and to use protected-mode facil-
ities, such as the protected-mode interrupt- and exception-handling facilities. Protected-mode
multitasking permits multiple virtual-8086 mode tasks (with each task running a separate 8086
program) to be run on the processor along with other nonvirtual-8086 mode tasks.
This section describes both the basic real-address mode execution environment and the virtual-
8086-mode execution environment, available on the Intel Architecture processors beginning
with the Intel386™ processor.

16.1. REAL-ADDRESS MODE

The Intel Architecture’s real-address mode runs programs written for the Intel 8086, Intel 8088,
Intel 80186, and Intel 80188 processors, or for the real-address mode of the Intel 286,
Intel386™, Intel486™, Pentium®, Pentium® Pro, Pentium® II, and P6-family processors.
The execution environment of the processor in real-address mode is designed to duplicate the
execution environment of the Intel 8086 processor. To an 8086 program, a processor operating
in real-address mode behaves like a high-speed 8086 processor. The principal features of this
architecture are defined in Chapter 3, Basic Execution Environment, of the Intel Architecture
Software Developer’s Manual, Volume 1. The following is a summary of the core features of the
real-address mode execution environment as would be seen by a program written for the 8086:

16-1
8086 EMULATION

• The processor supports a nominal 1-MByte physical address space (refer to Section
16.1.1., “Address Translation in Real-Address Mode” for specific details). This address
space is divided into segments, each of which can be up to 64 KBytes in length. The base
of a segment is specified with a 16-bit segment selector, which is zero extended to form a
20-bit offset from address 0 in the address space. An operand within a segment is
addressed with a 16-bit offset from the base of the segment. A physical address is thus
formed by adding the offset to the 20-bit segment base (refer to Section 16.1.1., “Address
Translation in Real-Address Mode”).
• All operands in “native 8086 code” are 8-bit or 16-bit values. (Operand size override
prefixes can be used to access 32-bit operands.)
• Eight 16-bit general-purpose registers are provided: AX, BX, CX, DX, SP, BP, SI, and DI.
The extended 32 bit registers (EAX, EBX, ECX, EDX, ESP, EBP, ESI, and EDI) are
accessible to programs that explicitly perform a size override operation.
• Four segment registers are provided: CS, DS, SS, and ES. (The FS and GS registers are
accessible to programs that explicitly access them.) The CS register contains the segment
selector for the code segment; the DS and ES registers contain segment selectors for data
segments; and the SS register contains the segment selector for the stack segment.
• The 8086 16-bit instruction pointer (IP) is mapped to the lower 16-bits of the EIP register.
Note this register is a 32-bit register and unintentional address wrapping may occur.
• The 16-bit FLAGS register contains status and control flags. (This register is mapped to
the 16 least significant bits of the 32-bit EFLAGS register.)
• All of the Intel 8086 instructions are supported (refer to Section 16.1.3., “Instructions
Supported in Real-Address Mode”).
• A single, 16-bit-wide stack is provided for handling procedure calls and invocations of
interrupt and exception handlers. This stack is contained in the stack segment identified
with the SS register. The SP (stack pointer) register contains an offset into the stack
segment. The stack grows down (toward lower segment offsets) from the stack pointer.
The BP (base pointer) register also contains an offset into the stack segment that can be
used as a pointer to a parameter list. When a CALL instruction is executed, the processor
pushes the current instruction pointer (the 16 least-significant bits of the EIP register and,
on far calls, the current value of the CS register) onto the stack. On a return, initiated with
a RET instruction, the processor pops the saved instruction pointer from the stack into the
EIP register (and CS register on far returns). When an implicit call to an interrupt or
exception handler is executed, the processor pushes the EIP, CS, and EFLAGS (low-order
16-bits only) registers onto the stack. On a return from an interrupt or exception handler,
initiated with an IRET instruction, the processor pops the saved instruction pointer and
EFLAGS image from the stack into the EIP, CS, and EFLAGS registers.
• A single interrupt table, called the “interrupt vector table” or “interrupt table,” is provided
for handling interrupts and exceptions (refer to Figure 16-2). The interrupt table (which has
4-byte entries) takes the place of the interrupt descriptor table (IDT, with 8-byte entries)
used when handling protected-mode interrupts and exceptions. Interrupt and exception
vector numbers provide an index to entries in the interrupt table. Each entry provides a
pointer (called a “vector”) to an interrupt- or exception-handling procedure. Refer to

16-2
 8086 EMULATION

Section 16.1.4., “Interrupt and Exception Handling” for more details. It is possible for
software to relocate the IDT by means of the LIDT instruction on Intel Architecture
processors beginning with the Intel386™ processor.
• The floating-point unit (FPU) is active and available to execute FPU instructions in real-
address mode. Programs written to run on the Intel 8087 and Intel 287 math coprocessors
can be run in real-address mode without modification.
The following extensions to the Intel 8086 execution environment are available in the Intel
Architecture’s real-address mode. If backwards compatibility to Intel 286 and Intel 8086 proces-
sors is required, these features should not be used in new programs written to run in real-address
mode.
• Two additional segment registers (FS and GS) are available.
• Many of the integer and system instructions that have been added to P6-family processors
can be executed in real-address mode (refer to Section 16.1.3., “Instructions Supported in
Real-Address Mode”).
• The 32-bit operand prefix can be used in real-address mode programs to execute the 32-bit
forms of instructions. This prefix also allows real-address mode programs to use the
processor’s 32-bit general-purpose registers.
• The 32-bit address prefix can be used in real-address mode programs, allowing 32-bit
offsets.
The following sections describe address formation, registers, available instructions, and inter-
rupt and exception handling in real-address mode. For information on I/O in real-address mode,
refer to Chapter 9, Input/Output, in the Intel Architecture Software Developer’s Manual, Volume
1.

16.1.1. Address Translation in Real-Address Mode

In real-address mode, the processor does not interpret segment selectors as indexes into a
descriptor table; instead, it uses them directly to form linear addresses as the 8086 processor
does. It shifts the segment selector left by 4 bits to form a 20-bit base address (refer to Figure
16-1). The offset into a segment is added to the base address to create a linear address that maps
directly to the physical address space.
When using 8086-style address translation, it is possible to specify addresses larger than 1
MByte. For example, with a segment selector value of FFFFH and an offset of FFFFH, the linear
(and physical) address would be 10FFEFH (1 megabyte plus 64 KBytes). The 8086 processor,
which can form addresses only up to 20 bits long, truncates the high-order bit, thereby “wrap-
ping” this address to FFEFH. When operating in real-address mode, however, the processor does
not truncate such an address and uses it as a physical address. (Note, however, that for Intel
Architecture processors beginning with the Intel486™ processor, the A20M# signal can be used
in real-address mode to mask address line A20, thereby mimicking the 20-bit wrap-around
behavior of the 8086 processor.) Care should be take to ensure that A20M# based address wrap-
ping is handled correctly in multiprocessor based system.

16-3
8086 EMULATION

19 4 3 0

Base 16-bit Segment Selector 0 0 0 0

+ 19 16 15 0

Offset 0 0 0 0 16-bit Effective Address

= 19 0
Linear
20-bit Linear Address
Address

Figure 16-1. Real-Address Mode Address Translation

The Intel Architecture processors beginning with the Intel386™ processor can generate 32-bit
offsets using an address override prefix; however, in real-address mode, the value of a 32-bit
offset may not exceed FFFFH without causing an exception.
For full compatibility with Intel 286 real-address mode, pseudo-protection faults (interrupt 12
or 13) occur if a 32-bit offset is generated outside the range 0 through FFFFH.

16.1.2. Registers Supported in Real-Address Mode

The register set available in real-address mode includes all the registers defined for the 8086
processor plus the new registers introduced inP6-family processors, such as the FS and GS
segment registers, the debug registers, the control registers, and the floating-point unit registers.
The 32-bit operand prefix allows a real-address mode program to use the 32-bit general-purpose
registers (EAX, EBX, ECX, EDX, ESP, EBP, ESI, and EDI).

16.1.3. Instructions Supported in Real-Address Mode

The following instructions make up the core instruction set for the 8086 processor. If backwards
compatibility to the Intel 286 and Intel 8086 processors is required, only these instructions
should be used in a new program written to run in real-address mode.
• Move (MOV) instructions that move operands between general-purpose registers, segment
registers, and between memory and general-purpose registers,
• The exchange (XCHG) instruction.
• Load segment register instructions LDS and LES.
• Arithmetic instructions ADD, ADC, SUB, SBB, MUL, IMUL, DIV, IDIV, INC, DEC,
CMP, and NEG.
• Logical instructions AND, OR, XOR, and NOT.

16-4
 8086 EMULATION

• Decimal instructions DAA, DAS, AAA, AAS, AAM, and AAD.

• Stack instructions PUSH and POP (to general-purpose registers and segment registers).
• Type conversion instructions CWD, CDQ, CBW, and CWDE.
• Shift and rotate instructions SAL, SHL, SHR, SAR, ROL, ROR, RCL, and RCR.
• TEST instruction.
• Control instructions JMP, Jcc, CALL, RET, LOOP, LOOPE, and LOOPNE.
• Interrupt instructions INT n, INTO, and IRET.
• EFLAGS control instructions STC, CLC, CMC, CLD, STD, LAHF, SAHF, PUSHF, and
POPF.
• I/O instructions IN, INS, OUT, and OUTS.
• Load effective address (LEA) instruction, and translate (XLATB) instruction.
• LOCK prefix.
• Repeat prefixes REP, REPE, REPZ, REPNE, and REPNZ.
• Processor halt (HLT) instruction.
• No operation (NOP) instruction.
The following instructions, added to P6-family processors (some in the Intel 286 processor and
the remainder in the Intel386™ processor), can be executed in real-address mode, if backwards
compatibility to the Intel 8086 processor is not required.
• Move (MOV) instructions that operate on the control and debug registers.
• Load segment register instructions LSS, LFS, and LGS.
• Generalized multiply instructions and multiply immediate data.
• Shift and rotate by immediate counts.
• Stack instructions PUSHA, PUSHAD, POPA and POPAD, and PUSH immediate data.
• Move with sign extension instructions MOVSX and MOVZX.
• Long-displacement Jcc instructions.
• Exchange instructions CMPXCHG, CMPXCHG8B, and XADD.
• String instructions MOVS, CMPS, SCAS, LODS, and STOS.
• Bit test and bit scan instructions BT, BTS, BTR, BTC, BSF, and BSR; the byte-set-on
condition instruction SETcc; and the byte swap (BSWAP) instruction.
• Double shift instructions SHLD and SHRD.
• EFLAGS control instructions PUSHF and POPF.
• ENTER and LEAVE control instructions.

16-5
8086 EMULATION

• BOUND instruction.
• CPU identification (CPUID) instruction.
• System instructions CLTS, INVD, WINVD, INVLPG, LGDT, SGDT, LIDT, SIDT,
LMSW, SMSW, RDMSR, WRMSR, RDTSC, and RDPMC.
Execution of any of the other Intel Architecture instructions (not given in the previous two lists)
in real-address mode result in an invalid-opcode exception (#UD) being generated.

16.1.4. Interrupt and Exception Handling

When operating in real-address mode, software must provide interrupt and exception-handling
facilities that are separate from those provided in protected mode. Even during the early stages
of processor initialization when the processor is still in real-address mode, elementary real-
address mode interrupt and exception-handling facilities must be provided to insure reliable
operation of the processor, or the initialization code must insure that no interrupts or exceptions
will occur.
The Intel Architecture processors handle interrupts and exceptions in real-address mode similar
to the way they handle them in protected mode. When a processor receives an interrupt or gener-
ates an exception, it uses the vector number of the interrupt or exception as an index into the
interrupt table. (In protected mode, the interrupt table is called the interrupt descriptor table
(IDT), but in real-address mode, the table is usually called the interrupt vector table, or simply
the interrupt table.) The entry in the interrupt vector table provides a pointer to an interrupt- or
exception-handler procedure. (The pointer consists of a segment selector for a code segment and
a 16-bit offset into the segment.) The processor performs the following actions to make an
implicit call to the selected handler:
1. Pushes the current values of the CS and EIP registers onto the stack. (Only the 16 least-
significant bits of the EIP register are pushed.)
2. Pushes the low-order 16 bits of the EFLAGS register onto the stack.
3. Clears the IF flag in the EFLAGS register to disable interrupts.
4. Clears the TF, RC, and AC flags, in the EFLAGS register.
5. Transfers program control to the location specified in the interrupt vector table.
An IRET instruction at the end of the handler procedure reverses these steps to return program
control to the interrupted program. Exceptions do not return error codes in real-address mode.
The interrupt vector table is an array of 4-byte entries (refer to Figure 16-2). Each entry consists
of a far pointer to a handler procedure, made up of a segment selector and an offset. The
processor scales the interrupt or exception vector by 4 to obtain an offset into the interrupt table.
Following reset, the base of the interrupt vector table is located at physical address 0 and its limit
is set to 3FFH. In the Intel 8086 processor, the base address and limit of the interrupt vector table
cannot be changed. In the P6-family processors, the base address and limit of the interrupt vector
table are contained in the IDTR register and can be changed using the LIDT instruction. (For

16-6
 8086 EMULATION

backward compatibility to Intel 8086 processors, the default base address and limit of the inter-
rupt vector table should not be changed.)

Up to Entry 255

Entry 3
12

Entry 2
8

Entry 1
4

Segment Selector 2
Interrupt Vector 0*
Offset 0
15 0
* Interrupt vector number 0 selects entry 0
(called “interrupt vector 0”) in the interrupt IDTR
vector table. Interrupt vector 0 in turn
points to the start of the interrupt handler
for interrupt 0.

Figure 16-2. Interrupt Vector Table in Real-Address Mode

Table 16-1 shows the interrupt and exception vectors that can be generated in real-address mode
and virtual-8086 mode, and in the Intel 8086 processor. Refer to Chapter 5, Interrupt and Excep-
tion Handling for a description of the exception conditions.

16-7
8086 EMULATION

Table 16-1. Real-Address Mode Exceptions and Interrupts

Vector Real-Address Virtual-8086 Intel 8086
No. Description Mode Mode Processor
0 Divide Error (#DE) Yes Yes Yes
1 Debug Exception (#DB) Yes Yes No
2 NMI Interrupt Yes Yes Yes
3 Breakpoint (#BP) Yes Yes Yes
4 Overflow (#OF) Yes Yes Yes
5 BOUND Range Exceeded (#BR) Yes Yes Reserved
6 Invalid Opcode (#UD) Yes Yes Reserved
7 Device Not Available (#NM) Yes Yes Reserved
8 Double Fault (#DF) Yes Yes Reserved
9 (Intel reserved. Do not use.) Reserved Reserved Reserved
10 Invalid TSS (#TS) Reserved Yes Reserved
11 Segment Not Present (#NP) Reserved Yes Reserved
12 Stack Fault (#SS) Yes Yes Reserved
13 General Protection (#GP)* Yes Yes Reserved
14 Page Fault (#PF) Reserved Yes Reserved
15 (Intel reserved. Do not use.) Reserved Reserved Reserved
16 Floating-Point Error (#MF) Yes Yes Reserved
17 Alignment Check (#AC) Reserved Yes Reserved
18 Machine Check (#MC) Yes Yes Reserved
19 SIMD Floating-Point Numeric Yes Yes Reserved
Error (#XF)
20-31 (Intel reserved. Do not use.) Reserved Reserved Reserved
32-255 User Defined Interrupts Yes Yes Yes

NOTE:
* In the real-address mode, vector 13 is the segment overrun exception. In protected and virtual-8086
modes, this exception covers all general-protection error conditions, including traps to the virtual-8086
monitor from virtual-8086 mode.

16-8
 8086 EMULATION

16.2. VIRTUAL-8086 MODE

Virtual-8086 mode is actually a special type of a task that runs in protected mode. When the
operating-system or executive switches to a virtual-8086-mode task, the processor emulates an
Intel 8086 processor. The execution environment of the processor while in the 8086-emulation
state is the same as is described in Section 16.1., “Real-Address Mode” for real-address mode,
including the extensions. The major difference between the two modes is that in virtual-8086
mode the 8086 emulator uses some protected-mode services (such as the protected-mode inter-
rupt and exception-handling and paging facilities).
As in real-address mode, any new or legacy program that has been assembled and/or compiled
to run on an Intel 8086 processor will run in a virtual-8086-mode task. And several 8086
programs can be run as virtual-8086-mode tasks concurrently with normal protected-mode
tasks, using the processor’s multitasking facilities.

16.2.1. Enabling Virtual-8086 Mode

The processor runs in virtual-8086 mode when the VM (virtual machine) flag in the EFLAGS
register is set. This flag can only be set when the processor switches to a new protected-mode
task or resumes virtual-8086 mode via an IRET instruction.
System software cannot change the state of the VM flag directly in the EFLAGS register (for
example, by using the POPFD instruction). Instead it changes the flag in the image of the
EFLAGS register stored in the TSS or on the stack following a call to an interrupt- or exception-
handler procedure. For example, software sets the VM flag in the EFLAGS image in the TSS
when first creating a virtual-8086 task.
The processor tests the VM flag under three general conditions:
• When loading segment registers, to determine whether to use 8086-style address
translation.
• When decoding instructions, to determine which instructions are not supported in virtual-
8086 mode and which instructions are sensitive to IOPL.
• When checking privileged instructions, on page accesses, or when performing other
permission checks. (Virtual-8086 mode always executes at CPL 3.)

16.2.2. Structure of a Virtual-8086 Task

A virtual-8086-mode task consists of the following items:
• A 32-bit TSS for the task.
• The 8086 program.
• A virtual-8086 monitor.
• 8086 operating-system services.

16-9
8086 EMULATION

The TSS of the new task must be a 32-bit TSS, not a 16-bit TSS, because the 16-bit TSS does
not load the most-significant word of the EFLAGS register, which contains the VM flag. All
TSS’s, stacks, data, and code used to handle exceptions when in virtual-8086 mode must also be
32-bit segments.
The processor enters virtual-8086 mode to run the 8086 program and returns to protected mode
to run the virtual-8086 monitor.
The virtual-8086 monitor is a 32-bit protected-mode code module that runs at a CPL of 0. The
monitor consists of initialization, interrupt- and exception-handling, and I/O emulation proce-
dures that emulate a personal computer or other 8086-based platform. Typically, the monitor is
either part of or closely associated with the protected-mode general-protection (#GP) exception
handler, which also runs at a CPL of 0. As with any protected-mode code module, code-segment
descriptors for the virtual-8086 monitor must exist in the GDT or in the task’s LDT. The virtual-
8086 monitor also may need data-segment descriptors so it can examine the IDT or other parts
of the 8086 program in the first 1 MByte of the address space. The linear addresses above
10FFEFH are available for the monitor, the operating system, and other system software.
The 8086 operating-system services consists of a kernel and/or operating-system procedures
that the 8086 program makes calls to. These services can be implemented in either of the
following two ways:
• They can be included in the 8086 program. This approach is desirable for either of the
following reasons:
— The 8086 program code modifies the 8086 operating-system services.
— There is not sufficient development time to merge the 8086 operating-system services
into main operating system or executive.
• They can be implemented or emulated in the virtual-8086 monitor. This approach is
desirable for any of the following reasons:
— The 8086 operating-system procedures can be more easily coordinated among several
virtual-8086 tasks.
— Memory can be saved by not duplicating 8086 operating-system procedure code for
several virtual-8086 tasks.
— The 8086 operating-system procedures can be easily emulated by calls to the main
operating system or executive.
The approach chosen for implementing the 8086 operating-system services may result in
different virtual-8086-mode tasks using different 8086 operating-system services.

16.2.3. Paging of Virtual-8086 Tasks

Even though a program running in virtual-8086 mode can use only 20-bit linear addresses, the
processor converts these addresses into 32-bit linear addresses before mapping them to the phys-
ical address space. If paging is being used, the 8086 address space for a program running in
virtual-8086 mode can be paged and located in a set of pages in physical address space. If paging

16-10
 8086 EMULATION

is used, it is transparent to the program running in virtual-8086 mode just as it is for any task
running on the processor.
Paging is not necessary for a single virtual-8086-mode task, but paging is useful or necessary in
the following situations:
• When running multiple virtual-8086-mode tasks. Here, paging allows the lower 1 MByte
of the linear address space for each virtual-8086-mode task to be mapped to a different
physical address location.
• When emulating the 8086 address-wraparound that occurs at 1 MByte. When using 8086-
style address translation, it is possible to specify addresses larger than 1 MByte. These
addresses automatically wraparound in the Intel 8086 processor (refer to Section 16.1.1.,
“Address Translation in Real-Address Mode”). If any 8086 programs depend on address
wraparound, the same effect can be achieved in a virtual-8086-mode task by mapping the
linear addresses between 100000H and 110000H and linear addresses between 0 and
10000H to the same physical addresses.
• When sharing the 8086 operating-system services or ROM code that is common to several
8086 programs running as different 8086-mode tasks.
• When redirecting or trapping references to memory-mapped I/O devices.

16.2.4. Protection within a Virtual-8086 Task

Protection is not enforced between the segments of an 8086 program. Either of the following
techniques can be used to protect the system software running in a virtual-8086-mode task from
the 8086 program:
• Reserve the first 1 MByte plus 64 KBytes of each task’s linear address space for the 8086
program. An 8086 processor task cannot generate addresses outside this range.
• Use the U/S flag of page-table entries to protect the virtual-8086 monitor and other system
software in the virtual-8086 mode task space. When the processor is in virtual-8086 mode,
the CPL is 3. Therefore, an 8086 processor program has only user privileges. If the pages
of the virtual-8086 monitor have supervisor privilege, they cannot be accessed by the 8086
program.

16.2.5. Entering Virtual-8086 Mode

Figure 16-3 summarizes the methods of entering and leaving virtual-8086 mode. The processor
switches to virtual-8086 mode in either of the following situations:
• Task switch when the VM flag is set to 1 in the EFLAGS register image stored in the TSS
for the task. Here the task switch can be initiated in either of two ways:
— A CALL or JMP instruction.
— An IRET instruction, where the NT flag in the EFLAGS image is set to 1.
• Return from a protected-mode interrupt or exception handler when the VM flag is set to 1
in the EFLAGS register image on the stack.
16-11
8086 EMULATION

Real Mode
Code
Real-Address
Mode PE=0 or
PE=1 RESET

Task Switch Protected- CALL

Protected Protected- VM=0 Mode Interrupt Virtual-8086
Mode Mode Tasks and Exception Monitor
Handlers RET

Task Switch1
VM=0
VM=1

Interrupt or
Virtual-8086 Exception2
Mode Virtual-8086 #GP Exception3
RESET Mode Tasks
(8086 IRET4
Programs) IRET5

Redirect Interrupt to 8086 Program

Interrupt or Exception Handler6

NOTES:
1. Task switch carried out in either of two ways:
- CALL or JMP where the VM flag in the EFLAGS image is 1.
- IRET where VM is 1 and NT is 1.
2. Hardware interrupt or exception; software interrupt (INT n) when IOPL is 3.
3. General-protection exception caused by software interrupt (INT n), IRET,
POPF, PUSHF, IN, or OUT when IOPL is less than 3.
4. Normal return from protected-mode interrupt or exception handler.
5. A return from the 8086 monitor to redirect an interrupt or exception back
to an interrupt or exception handler in the 8086 program running in virtual-
8086 mode.
6. Internal redirection of a software interrupt (INT n) when VME is 1,
IOPL is <3, and the redirection bit is 1.

Figure 16-3. Entering and Leaving Virtual-8086 Mode

When a task switch is used to enter virtual-8086 mode, the TSS for the virtual-8086-mode task
must be a 32-bit TSS. (If the new TSS is a 16-bit TSS, the upper word of the EFLAGS register
is not in the TSS, causing the processor to clear the VM flag when it loads the EFLAGS register.)
The processor updates the VM flag prior to loading the segment registers from their images in
the new TSS. The new setting of the VM flag determines whether the processor interprets the

16-12
 8086 EMULATION

contents of the segment registers as 8086-style segment selectors or protected-mode segment

selectors. When the VM flag is set, the segment registers are loaded from the TSS, using 8086-
style address translation to form base addresses.
Refer to Section 16.3., “Interrupt and Exception Handling in Virtual-8086 Mode” for informa-
tion on entering virtual-8086 mode on a return from an interrupt or exception handler.

16.2.6. Leaving Virtual-8086 Mode

The processor can leave the virtual-8086 mode only through an interrupt or exception. The
following are situations where an interrupt or exeception will lead to the processor leaving
virtual-8086 mode (refer to Figure 16-3):
• The processor services a hardware interrupt generated to signal the suspension of
execution of the virtual-8086 application. This hardware interrupt may be generated by a
timer or other external mechanism. Upon receiving the hardware interrupt, the processor
enters protected mode and switches to a protected-mode (or another virtual-8086 mode)
task either through a task gate in the protected-mode IDT or through a trap or interrupt gate
that points to a handler that initiates a task switch. A task switch from a virtual-8086 task
to another task loads the EFLAGS register from the TSS of the new task. The value of the
VM flag in the new EFLAGS determines if the new task executes in virtual-8086 mode or
not.
• The processor services an exception caused by code executing the virtual-8086 task or
services a hardware interrupt that “belongs to” the virtual-8086 task. Here, the processor
enters protected mode and services the exception or hardware interrupt through the
protected-mode IDT (normally through an interrupt or trap gate) and the protected-mode
exception- and interrupt-handlers. The processor may handle the exception or interrupt
within the context of the virtual 8086 task and return to virtual-8086 mode on a return from
the handler procedure. The processor may also execute a task switch and handle the
exception or interrupt in the context of another task.
• The processor services a software interrupt generated by code executing in the virtual-
8086 task (such as a software interrupt to call a MS-DOS* operating system routine). The
processor provides several methods of handling these software interrupts, which are
discussed in detail in Section 16.3.3., “Class 3—Software Interrupt Handling in Virtual-
8086 Mode” Most of them involve the processor entering protected mode, often by means
of a general-protection (#GP) exception. In protected mode, the processor can send the
interrupt to the virtual-8086 monitor for handling and/or redirect the interrupt back to the
application program running in virtual-8086 mode task for handling.

Intel Architecture processors that incorporate the virtual mode extension (enabled with the
VME flag in control register CR4) are capable of redirecting software-generated interrupts
back to the program’s interrupt handlers without leaving virtual-8086 mode. Refer to
Section 16.3.3.4., “Method 5: Software Interrupt Handling” for more information on this
mechanism.

16-13
8086 EMULATION

• A hardware reset initiated by asserting the RESET or INIT pin is a special kind of
interrupt. When a RESET or INIT is signaled while the processor is in virtual-8086 mode,
the processor leaves virtual-8086 mode and enters real-address mode.
• Execution of the HLT instruction in virtual-8086 mode will cause a general-protection
(GP#) fault, which the protected-mode handler generally sends to the virtual-8086 monitor.
The virtual-8086 monitor then determines the correct execution sequence after verifying
that it was entered as a result of a HLT execution.
Refer to Section 16.3., “Interrupt and Exception Handling in Virtual-8086 Mode” for informa-
tion on leaving virtual-8086 mode to handle an interrupt or exception generated in virtual-8086
mode.

16.2.7. Sensitive Instructions

When an Intel Architecture processor is running in virtual-8086 mode, the CLI, STI, PUSHF,
POPF, INT n, and IRET instructions are sensitive to IOPL. The IN, INS, OUT, and OUTS
instructions, which are sensitive to IOPL in protected mode, are not sensitive in virtual-8086
mode.
The CPL is always 3 while running in virtual-8086 mode; if the IOPL is less than 3, an attempt
to use the IOPL-sensitive instructions listed above triggers a general-protection exception
(#GP). These instructions are sensitive to IOPL to give the virtual-8086 monitor a chance to
emulate the facilities they affect.

16.2.8. Virtual-8086 Mode I/O

Many 8086 programs written for nonmultitasking systems directly access I/O ports. This prac-
tice may cause problems in a multitasking environment. If more than one program accesses the
same port, they may interfere with each other. Most multitasking systems require application
programs to access I/O ports through the operating system. This results in simplified, centralized
control.
The processor provides I/O protection for creating I/O that is compatible with the environment
and transparent to 8086 programs. Designers may take any of several possible approaches to
protecting I/O ports:
• Protect the I/O address space and generate exceptions for all attempts to perform I/O
directly.
• Let the 8086 program perform I/O directly.
• Generate exceptions on attempts to access specific I/O ports.
• Generate exceptions on attempts to access specific memory-mapped I/O ports.
The method of controlling access to I/O ports depends upon whether they are I/O-port mapped
or memory mapped.

16-14
 8086 EMULATION

16.2.8.1. I/O-PORT-MAPPED I/O

The I/O permission bit map in the TSS can be used to generate exceptions on attempts to access
specific I/O port addresses. The I/O permission bit map of each virtual-8086-mode task deter-
mines which I/O addresses generate exceptions for that task. Because each task may have a
different I/O permission bit map, the addresses that generate exceptions for one task may be
different from the addresses for another task. This differs from protected mode in which, if the
CPL is less than or equal to the IOPL, I/O access is allowed without checking the I/O permission
bit map. Refer to Chapter 9, Input/Output, in the Intel Architecture Software Developer’s
Manual, Volume 1, for more information about the I/O permission bit map.

16.2.8.2. MEMORY-MAPPED I/O

In systems which use memory-mapped I/O, the paging facilities of the processor can be used to
generate exceptions for attempts to access I/O ports. The virtual-8086 monitor may use paging
to control memory-mapped I/O in these ways:
• Map part of the linear address space of each task that needs to perform I/O to the physical
address space where I/O ports are placed. By putting the I/O ports at different addresses (in
different pages), the paging mechanism can enforce isolation between tasks.
• Map part of the linear address space to pages that are not-present. This generates an
exception whenever a task attempts to perform I/O to those pages. System software then
can interpret the I/O operation being attempted.
Software emulation of the I/O space may require too much operating system intervention under
some conditions. In these cases, it may be possible to generate an exception for only the first
attempt to access I/O. The system software then may determine whether a program can be given
exclusive control of I/O temporarily, the protection of the I/O space may be lifted, and the
program allowed to run at full speed.

16.2.8.3. SPECIAL I/O BUFFERS

Buffers of intelligent controllers (for example, a bit-mapped frame buffer) also can be emulated
using page mapping. The linear space for the buffer can be mapped to a different physical space
for each virtual-8086-mode task. The virtual-8086 monitor then can control which virtual buffer
to copy onto the real buffer in the physical address space.

16.3. INTERRUPT AND EXCEPTION HANDLING IN VIRTUAL-8086

MODE
When the processor receives an interrupt or detects an exception condition while in virtual-8086
mode, it invokes an interrupt or exception handler, just as it does in protected or real-address
mode. The interrupt or exception handler that is invoked and the mechanism used to invoke it
depends on the class of interrupt or exception that has been detected or generated and the state
of various system flags and fields.

16-15
8086 EMULATION

In virtual-8086 mode, the interrupts and exceptions are divided into three classes for the
purposes of handling:
• Class 1—All processor-generated exceptions and all hardware interrupts, including the
NMI interrupt and the hardware interrupts sent to the processor’s external interrupt
delivery pins. All class 1 exceptions and interrupts are handled by the protected-mode
exception and interrupt handlers.
• Class 2—Special case for maskable hardware interrupts (Section 5.1.1.2., “Maskable
Hardware Interrupts”, in Chapter 5, Interrupt and Exception Handling) when the virtual
mode extensions are enabled.
• Class 3—All software-generated interrupts, that is interrupts generated with the INT n
instruction1.
The method the processor uses to handle class 2 and 3 interrupts depends on the setting of the
following flags and fields:
• IOPL field (bits 12 and 13 in the EFLAGS register)—Controls how class 3 software
interrupts are handled when the processor is in virtual-8086 mode (refer to Section 2.3.,
“System Flags and Fields in the EFLAGS Register”, in Chapter 2, System Architecture
Overview). This field also controls the enabling of the VIF and VIP flags in the EFLAGS
register when the VME flag is set. The VIF and VIP flags are provided to assist in the
handling of class 2 maskable hardware interrupts.
• VME flag (bit 0 in control register CR4)—Enables the virtual mode extension for the
processor when set (refer to Section 2.5., “Control Registers”, in Chapter 2, System Archi-
tecture Overview).
• Software interrupt redirection bit map (32 bytes in the TSS, refer to Figure
16-5)—Contains 256 flags that indicates how class 3 software interrupts should be handled
when they occur in virtual-8086 mode. A software interrupt can be directed either to the
interrupt and exception handlers in the currently running 8086 program or to the protected-
mode interrupt and exception handlers.
• The virtual interrupt flag (VIF) and virtual interrupt pending flag (VIP) in the EFLAGS
register—Provides virtual interrupt support for the handling of class 2 maskable
hardware interrupts (refer to Section 16.3.2., “Class 2—Maskable Hardware Interrupt
Handling in Virtual-8086 Mode Using the Virtual Interrupt Mechanism”).

NOTE
The VME flag, software interrupt redirection bit map, and VIF and VIP flags
are only available in Intel Architecture processors that support the virtual
mode extensions. These extensions were introduced in the Intel Architecture
with the Pentium® processor.
The following sections describe the actions that processor takes and the possible actions of inter-
rupt and exception handlers for the two classes of interrupts described in the previous para-
graphs. These sections describe three possible types of interrupt and exception handlers:

1. The INT 3 instruction is a special case (refer to the description of the INT n instruction in Chapter 3,
Instruction Set Reference, of the Intel Architecture Software Developer’s Manual, Volume 2).

16-16
 8086 EMULATION

• Protected-mode interrupt and exceptions handlers—These are the handlers that the
processor calls through the protected-mode IDT.
• Virtual-8086 monitor interrupt and exception handlers—These handlers are resident in the
virtual-8086 monitor, and they are commonly accessed through a general-protection
exception (#GP, interrupt 13) that is directed to the protected-mode general-protection
exception handler.
• 8086 program interrupt and exception handlers—These handlers are part of the 8086
program that is running in virtual-8086 mode.
The following sections describe how these handlers are used, depending on the selected class
and method of interrupt and exception handling.

16.3.1. Class 1—Hardware Interrupt and Exception Handling in

Virtual-8086 Mode
In virtual-8086 mode, the Pentium® and P6 family processors handle hardware interrupts and
exceptions in the same manner as they are handled by the Intel486™ and Intel386™ processors.
They invoke the protected-mode interrupt or exception handler that the interrupt or exception
vector points to in the IDT. Here, the IDT entry must contain either a 32-bit trap or interrupt gate
or a task gate. The following sections describe various ways that a virtual-8086 mode interrupt
or exception can be handled after the protected-mode handler has been invoked.
Refer to Section 16.3.2., “Class 2—Maskable Hardware Interrupt Handling in Virtual-8086
Mode Using the Virtual Interrupt Mechanism” for a description of the virtual interrupt mecha-
nism that is available for handling maskable hardware interrupts while in virtual-8086 mode.
When this mechanism is either not available or not enabled, maskable hardware interrupts are
handled in the same manner as exceptions, as described in the following sections.

16.3.1.1. HANDLING AN INTERRUPT OR EXCEPTION THROUGH A

PROTECTED-MODE TRAP OR INTERRUPT GATE
When an interrupt or exception vector points to a 32-bit trap or interrupt gate in the IDT, the gate
must in turn point to a nonconforming, privilege-level 0, code segment. When accessing this
code segment, the processor performs the following steps.
1. Switches to 32-bit protected mode and privilege level 0.
2. Saves the state of the processor on the privilege-level 0 stack. The states of the EIP, CS,
EFLAGS, ESP, SS, ES, DS, FS, and GS registers are saved (refer to Figure 16-4).
3. Clears the segment registers. Saving the DS, ES, FS, and GS registers on the stack and then
clearing the registers lets the interrupt or exception handler safely save and restore these
registers regardless of the type segment selectors they contain (protected-mode or 8086-
style). The interrupt and exception handlers, which may be called in the context of either a
protected-mode task or a virtual-8086-mode task, can use the same code sequences for
saving and restoring the registers for any task. Clearing these registers before execution of
the IRET instruction does not cause a trap in the interrupt handler. Interrupt procedures that

16-17
8086 EMULATION

expect values in the segment registers or that return values in the segment registers must
use the register images saved on the stack for privilege level 0.
4. Clears the VM flag in the EFLAGS register.
5. Begins executing the selected interrupt or exception handler.

Without Error Code With Error Code

ESP from ESP from

Unused Unused
TSS TSS
Old GS Old GS

Old FS Old FS

Old DS Old DS

Old ES Old ES

Old SS Old SS

Old ESP Old ESP

Old EFLAGS Old EFLAGS

Old CS Old CS

Old EIP New ESP Old EIP

Error Code New ESP

Figure 16-4. Privilege Level 0 Stack After Interrupt or Exception in Virtual-8086 Mode

If the trap or interrupt gate references a procedure in a conforming segment or in a segment at a

privilege level other than 0, the processor generates a general-protection exception (#GP). Here,
the error code is the segment selector of the code segment to which a call was attempted.
Interrupt and exception handlers can examine the VM flag on the stack to determine if the inter-
rupted procedure was running in virtual-8086 mode. If so, the interrupt or exception can be
handled in one of three ways:
• The protected-mode interrupt or exception handler that was called can handle the interrupt
or exception.
• The protected-mode interrupt or exception handler can call the virtual-8086 monitor to
handle the interrupt or exception.
• The virtual-8086 monitor (if called) can in turn pass control back to the 8086 program’s
interrupt and exception handler.

16-18
 8086 EMULATION

If the interrupt or exception is handled with a protected-mode handler, the handler can return to
the interrupted program in virtual-8086 mode by executing an IRET instruction. This instruction
loads the EFLAGS and segment registers from the images saved in the privilege level 0 stack
(refer to Figure 16-4). A set VM flag in the EFLAGS image causes the processor to switch back
to virtual-8086 mode. The CPL at the time the IRET instruction is executed must be 0, otherwise
the processor does not change the state of the VM flag.
The virtual-8086 monitor runs at privilege level 0, like the protected-mode interrupt and excep-
tion handlers. It is commonly closely tied to the protected-mode general-protection exception
(#GP, vector 13) handler. If the protected-mode interrupt or exception handler calls the virtual-
8086 monitor to handle the interrupt or exception, the return from the virtual-8086 monitor to
the interrupted virtual-8086 mode program requires two return instructions: a RET instruction
to return to the protected-mode handler and an IRET instruction to return to the interrupted
program.
The virtual-8086 monitor has the option of directing the interrupt and exception back to an inter-
rupt or exception handler that is part of the interrupted 8086 program, as described in Section
16.3.1.2., “Handling an Interrupt or Exception With an 8086 Program Interrupt or Exception
Handler”.

16.3.1.2. HANDLING AN INTERRUPT OR EXCEPTION WITH AN 8086

PROGRAM INTERRUPT OR EXCEPTION HANDLER
Because it was designed to run on an 8086 processor, an 8086 program running in a virtual-
8086-mode task contains an 8086-style interrupt vector table, which starts at linear address 0. If
the virtual-8086 monitor correctly directs an interrupt or exception vector back to the virtual-
8086-mode task it came from, the handlers in the 8086 program can handle the interrupt or
exception. The virtual-8086 monitor must carry out the following steps to send an interrupt or
exception back to the 8086 program:
1. Use the 8086 interrupt vector to locate the appropriate handler procedure in the 8086
program interrupt table.
2. Store the EFLAGS (low-order 16 bits only), CS and EIP values of the 8086 program on the
privilege-level 3 stack. This is the stack that the virtual-8086-mode task is using. (The
8086 handler may use or modify this information.)
3. Change the return link on the privilege-level 0 stack to point to the privilege-level 3
handler procedure.
4. Execute an IRET instruction to pass control to the 8086 program handler.
5. When the IRET instruction from the privilege-level 3 handler triggers a general-protection
exception (#GP) and thus effectively again calls the virtual-8086 monitor, restore the
return link on the privilege-level 0 stack to point to the original, interrupted, privilege-level
3 procedure.
6. Copy the low order 16 bits of the EFLAGS image from the privilege-level 3 stack to the
privilege-level 0 stack (because some 8086 handlers modify these flags to return
information to the code that caused the interrupt).

16-19
8086 EMULATION

7. Execute an IRET instruction to pass control back to the interrupted 8086 program.
Note that if an operating system intends to support all 8086 MS-DOS-based programs, it is
necessary to use the actual 8086 interrupt and exception handlers supplied with the program.
The reason for this is that some programs modify their own interrupt vector table to substitute
(or hook in series) their own specialized interrupt and exception handlers.

16.3.1.3. HANDLING AN INTERRUPT OR EXCEPTION THROUGH A TASK

GATE
When an interrupt or exception vector points to a task gate in the IDT, the processor performs a
task switch to the selected interrupt- or exception-handling task. The following actions are
carried out as part of this task switch:
1. The EFLAGS register with the VM flag set is saved in the current TSS.
2. The link field in the TSS of the called task is loaded with the segment selector of the TSS
for the interrupted virtual-8086-mode task.
3. The EFLAGS register is loaded from the image in the new TSS, which clears the VM flag
and causes the processor to switch to protected mode.
4. The NT flag in the EFLAGS register is set.
5. The processor begins executing the selected interrupt- or exception-handler task.
When an IRET instruction is executed in the handler task and the NT flag in the EFLAGS
register is set, the processors switches from a protected-mode interrupt- or exception-handler
task back to a virtual-8086-mode task. Here, the EFLAGS and segment registers are loaded from
images saved in the TSS for the virtual-8086-mode task. If the VM flag is set in the EFLAGS
image, the processor switches back to virtual-8086 mode on the task switch. The CPL at the time
the IRET instruction is executed must be 0, otherwise the processor does not change the state of
the VM flag.

16.3.2. Class 2—Maskable Hardware Interrupt Handling in Virtual-

8086 Mode Using the Virtual Interrupt Mechanism
Maskable hardware interrupts are those interrupts that are delivered through the INTR# pin or
through an interrupt request to the local APIC (refer to Section 5.1.1.2., “Maskable Hardware
Interrupts”, in Chapter 5, Interrupt and Exception Handling). These interrupts can be inhibited
(masked) from interrupting an executing program or task by clearing the IF flag in the EFLAGS
register.
When the VME flag in control register CR4 is set and the IOPL field in the EFLAGS register is
less than 3, two additional flags are activated in the EFLAGS register:
• VIF (virtual interrupt) flag, bit 19 of the EFLAGS register.
• VIP (virtual interrupt pending) flag, bit 20 of the EFLAGS register.

16-20
 8086 EMULATION

These flags provide the virtual-8086 monitor with more efficient control over handling
maskable hardware interrupts that occur during virtual-8086 mode tasks. They also reduce inter-
rupt-handling overhead, by eliminating the need for all IF related operations (such as PUSHF,
POPF, CLI, and STI instructions) to trap to the virtual-8086 monitor. The purpose and use of
these flags are as follows.

NOTE
The VIF and VIP flags are only available in Intel Architecture processors that
support the virtual mode extensions. These extensions were introduced in the
Intel Architecture with the Pentium® processor. When this mechanism is
either not available or not enabled, maskable hardware interrupts are handled
as class 1 interrupts. Here, if VIF and VIP flags are needed, the virtual-8086
monitor can implement them in software.

Existing 8086 programs commonly set and clear the IF flag in the EFLAGS register to enable
and disable maskable hardware interrupts, respectively; for example, to disable interrupts while
handling another interrupt or an exception. This practice works well in single task environments,
but can cause problems in multitasking and multiple-processor environments, where it is often
desirable to prevent an application program from having direct control over the handling of
hardware interrupts. When using earlier Intel Architecture processors, this problem was often
solved by creating a virtual IF flag in software. The Intel Architecture processors (beginning
with the Pentium® processor) provide hardware support for this virtual IF flag through the VIF
and VIP flags.
The VIF flag is a virtualized version of the IF flag, which an application program running from
within a virtual-8086 task can used to control the handling of maskable hardware interrupts.
When the VIF flag is enabled, the CLI and STI instructions operate on the VIF flag instead of
the IF flag. When an 8086 program executes the CLI instruction, the processor clears the VIF
flag to request that the virtual-8086 monitor inhibit maskable hardware interrupts from inter-
rupting program execution; when it executes the STI instruction, the processor sets the VIF flag
requesting that the virtual-8086 monitor enable maskable hardware interrupts for the 8086
program. But actually the IF flag, managed by the operating system, always controls whether
maskable hardware interrupts are enabled. Also, if under these circumstances an 8086 program
tries to read or change the IF flag using the PUSHF or POPF instructions, the processor will
change the VIF flag instead, leaving IF unchanged.
The VIP flag provides software a means of recording the existence of a deferred (or pending)
maskable hardware interrupt. This flag is read by the processor but never explicitly written by
the processor; it can only be written by software.
If the IF flag is set and the VIF and VIP flags are enabled, and the processor receives a maskable
hardware interrupt (interrupt vector 0 through 255), the processor performs and the interrupt
handler software should perform the following operations:
1. The processor invokes the protected-mode interrupt handler for the interrupt received, as
described in the following steps. These steps are almost identical to those described for

16-21
8086 EMULATION

method 1 interrupt and exception handling in Section 16.3.1.1., “Handling an Interrupt or

Exception Through a Protected-Mode Trap or Interrupt Gate”:
a. Switches to 32-bit protected mode and privilege level 0.
b. Saves the state of the processor on the privilege-level 0 stack. The states of the EIP,
CS, EFLAGS, ESP, SS, ES, DS, FS, and GS registers are saved (refer to Figure 16-4).
In the EFLAGS image on the stack, the IOPL field is set to 3 and the VIF flag is
copied to the IF flag.
c. Clears the segment registers.
d. Clears the VM flag in the EFLAGS register.
e. Begins executing the selected protected-mode interrupt handler.
2. The recommended action of the protected-mode interrupt handler is to read the VM flag
from the EFLAGS image on the stack. If this flag is set, the handler makes a call to the
virtual-8086 monitor.
3. The virtual-8086 monitor should read the VIF flag in the EFLAGS register.
— If the VIF flag is clear, the virtual-8086 monitor sets the VIP flag in the EFLAGS
image on the stack to indicate that there is a deferred interrupt pending and returns to
the protected-mode handler.
— If the VIF flag is set, the virtual-8086 monitor can handle the interrupt if it “belongs”
to the 8086 program running in the interrupted virtual-8086 task; otherwise, it can call
the protected-mode interrupt handler to handle the interrupt.
4. The protected-mode handler executes a return to the program executing in virtual-8086
mode.
5. Upon returning to virtual-8086 mode, the processor continues execution of the 8086
program.
When the 8086 program is ready to receive maskable hardware interrupts, it executes the STI
instruction to set the VIF flag (enabling maskable hardware interrupts). Prior to setting the VIF
flag, the processor automatically checks the VIP flag and does one of the following, depending
on the state of the flag:
• If the VIP flag is clear (indicating no pending interrupts), the processor sets the VIF flag.
• If the VIP flag is set (indicating a pending interrupt), the processor generates a general-
protection exception (#GP).
The recommended action of the protected-mode general-protection exception handler is to then
call the virtual-8086 monitor and let it handle the pending interrupt. After handling the pending
interrupt, the typical action of the virtual-8086 monitor is to clear the VIP flag and set the VIF
flag in the EFLAGS image on the stack, and then execute a return to the virtual-8086 mode. The
next time the processor receives a maskable hardware interrupt, it will then handle it as
described in steps 1 through 5 earlier in this section.
If the processor finds that both the VIF and VIP flags are set at the beginning of an instruction,
it generates a general-protection exception. This action allows the virtual-8086 monitor to

16-22
 8086 EMULATION

handle the pending interrupt for the virtual-8086 mode task for which the VIF flag is enabled.
Note that this situation can only occur immediately following execution of a POPF or IRET
instruction or upon entering a virtual-8086 mode task through a task switch.
Note that the states of the VIF and VIP flags are not modified in real-address mode or during
transitions between real-address and protected modes.

NOTE
The virtual interrupt mechanism described in this section is also available for
use in protected mode, refer to Section 16.4., “Protected-Mode Virtual Inter-
rupts”.

16.3.3. Class 3—Software Interrupt Handling in Virtual-8086 Mode

When the processor receives a software interrupt (an interrupt generated with the INT n instruc-
tion) while in virtual-8086 mode, it can use any of six different methods to handle the interrupt.
The method selected depends on the settings of the VME flag in control register CR4, the IOPL
field in the EFLAGS register, and the software interrupt redirection bit map in the TSS. Table
16-2 lists the six methods of handling software interrupts in virtual-8086 mode and the respec-
tive settings of the VME flag, IOPL field, and the bits in the interrupt redirection bit map for
each method. The table also summarizes the various actions the processor takes for each
method.
The VME flag enables the virtual mode extensions for the Pentium® and P6-family processors.
When this flag is clear, the processor responds to interrupts and exceptions in virtual-8086 mode
in the same manner as an Intel386™ or Intel486™ processor does. When this flag is set, the
virtual mode extension provides the following enhancements to virtual-8086 mode:
• Speeds up the handling of software-generated interrupts in virtual-8086 mode by allowing
the processor to bypass the virtual-8086 monitor and redirect software interrupts back to
the interrupt handlers that are part of the currently running 8086 program.
• Supports virtual interrupts for software written to run on the 8086 processor.
The IOPL value interacts with the VME flag and the bits in the interrupt redirection bit map to
determine how specific software interrupts should be handled.
The software interrupt redirection bit map (refer to Figure 16-5) is a 32-byte field in the TSS.
This map is located directly below the I/O permission bit map in the TSS. Each bit in the inter-
rupt redirection bit map is mapped to an interrupt vector. Bit 0 in the interrupt redirection bit
map (which maps to vector zero in the interrupt table) is located at the I/O base map address in
the TSS minus 32 bytes. When a bit in this bit map is set, it indicates that the associated software
interrupt (interrupt generated with an INT n instruction) should be handled through the
protected-mode IDT and interrupt and exception handlers. When a bit in this bit map is clear,
the processor redirects the associated software interrupt back to the interrupt table in the 8086
program (located at linear address 0 in the program’s address space).

16-23
8086 EMULATION

NOTE
The software interrupt redirection bit map does not affect hardware generated
interrupts and exceptions. Hardware generated interrupts and exceptions are
always handled by the protected-mode interrupt and exception handlers.

Table 16-2. Software Interrupt Handling Methods While in Virtual-8086 Mode

Bit in
Redir.
Method VME IOPL Bitmap* Processor Action
1 0 3 X Interrupt directed to a protected-mode interrupt handler:
- Clears VM and TF flags
- If serviced through interrupt gate, clears IF flag
- Switches to privilege-level 0 stack
- Pushes GS, FS, DS and ES onto privilege-level 0 stack
- Clears GS, FS, DS and ES to 0
- Pushes SS, ESP, EFLAGS, CS and EIP of interrupted task onto
privilege-level 0 stack
- Sets CS and EIP from interrupt gate
2 0 <3 X Interrupt directed to protected-mode general-protection
exception (#GP) handler.
3 1 <3 1 Interrupt directed to a protected-mode general-protection
exception (#GP) handler; VIF and VIP flag support for handling
class 2 maskable hardware interrupts.
4 1 3 1 Interrupt directed to protected-mode interrupt handler: (refer to
method 1 processor action).
5 1 3 0 Interrupt redirected to 8086 program interrupt handler:
- Pushes EFLAGS with NT cleared and IOPL set to 0
- Pushes CS and EIP (lower 16 bits only)
- Clears IF flag
- Clears TF flag
- Loads CS and EIP (lower 16 bits only) from selected entry in
the interrupt vector table of the current virtual-8086 task
6 1 <3 0 Interrupt redirected to 8086 program interrupt handler; VIF and
VIP flag support for handling class 2 maskable hardware
interrupts:
- Pushes EFLAGS with IOPL set to 3 and VIF copied to IF
- Pushes CS and EIP (lower 16 bits only)
- Clears the VIF flag
- Clears TF flag
- Loads CS and EIP (lower 16 bits only) from selected entry in
the interrupt vector table of the current virtual-8086 task

NOTE:
* When set to 0, software interrupt is redirected back to the 8086 program interrupt handler; when set to 1,
interrupt is directed to protected-mode handler.

16-24
 8086 EMULATION

31 24 23 Task-State Segment (TSS) 0

Last byte of bit 1 1 1 1 1 1 1 1
map must be
followed by a byte
with all bits set

I/O Permission Bit Map

Software Interrupt Redirection Bit Map (32 Bytes)

I/O Map Base 64H

I/O base map must
not exceed DFFFH.

Figure 16-5. Software Interrupt Redirection Bit Map in TSS

Redirecting software interrupts back to the 8086 program potentially speeds up interrupt
handling because a switch back and forth between virtual-8086 mode and protected mode is not
required. This latter interrupt-handling technique is particularly useful for 8086 operating
systems (such as MS-DOS) that use the INT n instruction to call operating system procedures.
The CPUID instruction can be used to verify that the virtual mode extension is implemented on
the processor. Bit 1 of the feature flags register (EDX) indicates the availability of the virtual
mode extension (refer to “CPUID—CPU Identification” in Chapter 3 of the Intel Architecture
Software Developer’s Manual, Volume 2).
The following sections describe the six methods (or mechanisms) for handling software inter-
rupts in virtual-8086 mode. Refer to Section 16.3.2., “Class 2—Maskable Hardware Interrupt
Handling in Virtual-8086 Mode Using the Virtual Interrupt Mechanism” for a description of the
use of the VIF and VIP flags in the EFLAGS register for handling maskable hardware interrupts.

16.3.3.1. METHOD 1: SOFTWARE INTERRUPT HANDLING

When the VME flag in control register CR4 is clear and the IOPL field is 3, a Pentium®, or P6-
family processor handles software interrupts in the same manner as they are handled by an
Intel386™ or Intel486™ processor. It executes an implicit call to the interrupt handler in the
protected-mode IDT pointed to by the interrupt vector. Refer to Section 16.3.1., “Class
1—Hardware Interrupt and Exception Handling in Virtual-8086 Mode” for a complete descrip-
tion of this mechanism and its possible uses.

16-25
8086 EMULATION

16.3.3.2. METHODS 2 AND 3: SOFTWARE INTERRUPT HANDLING

When a software interrupt occurs in virtual-8086 mode and the method 2 or 3 conditions are
present, the processor generates a general-protection exception (#GP). Method 2 is enabled
when the VME flag is set to 0 and the IOPL value is less than 3. Here the IOPL value is used to
bypass the protected-mode interrupt handlers and cause any software interrupt that occurs in
virtual-8086 mode to be treated as a protected-mode general-protection exception (#GP). The
general-protection exception handler calls the virtual-8086 monitor, which can then emulate an
8086-program interrupt handler or pass control back to the 8086 program’s handler, as described
in Section 16.3.1.2., “Handling an Interrupt or Exception With an 8086 Program Interrupt or
Exception Handler”.
Method 3 is enabled when the VME flag is set to 1, the IOPL value is less than 3, and the corre-
sponding bit for the software interrupt in the software interrupt redirection bit map is set to 1.
Here, the processor performs the same operation as it does for method 2 software interrupt
handling. If the corresponding bit for the software interrupt in the software interrupt redirection
bit map is set to 0, the interrupt is handled using method 6 (refer to Section 16.3.3.5., “Method
6: Software Interrupt Handling”).

16.3.3.3. METHOD 4: SOFTWARE INTERRUPT HANDLING

Method 4 handling is enabled when the VME flag is set to 1, the IOPL value is 3, and the bit for
the interrupt vector in the redirection bit map is set to 1. Method 4 software interrupt handling
allows method 1 style handling when the virtual mode extension is enabled; that is, the interrupt
is directed to a protected-mode handler (refer to Section 16.3.3.1., “Method 1: Software Inter-
rupt Handling”).

16.3.3.4. METHOD 5: SOFTWARE INTERRUPT HANDLING

Method 5 software interrupt handling provides a streamlined method of redirecting software
interrupts (invoked with the INT n instruction) that occur in virtual 8086 mode back to the 8086
program’s interrupt vector table and its interrupt handlers. Method 5 handling is enabled when
the VME flag is set to 1, the IOPL value is 3, and the bit for the interrupt vector in the redirection
bit map is set to 0. The processor performs the following actions to make an implicit call to the
selected 8086 program interrupt handler:
1. Pushes the low-order 16 bits of the EFLAGS register onto the stack with the NT and IOPL
bits cleared.
2. Pushes the current values of the CS and EIP registers onto the current stack. (Only the 16
least-significant bits of the EIP register are pushed and no stack switch occurs.)
3. Clears the IF flag in the EFLAGS register to disable interrupts.
4. Clears the TF flag, in the EFLAGS register.
5. Locates the 8086 program interrupt vector table at linear address 0 for the 8086-mode task.
6. Loads the CS and EIP registers with values from the interrupt vector table entry pointed to
by the interrupt vector number. Only the 16 low-order bits of the EIP are loaded and the 16

16-26
 8086 EMULATION

high-order bits are set to 0. The interrupt vector table is assumed to be at linear address 0 of
the current virtual-8986 task.
7. Begins executing the selected interrupt handler.
An IRET instruction at the end of the handler procedure reverses these steps to return program
control to the interrupted 8086 program.
Note that with method 5 handling, a mode switch from virtual-8086 mode to protected mode
does not occur. The processor remains in virtual-8086 mode throughout the interrupt-handling
operation.
The method 5 handling actions are virtually identical to the actions the processor takes when
handling software interrupts in real-address mode. The benefit of using method 5 handling to
access the 8086 program handlers is that it avoids the overhead of methods 2 and 3 handling,
which requires first going to the virtual-8086 monitor, then to the 8086 program handler, then
back again to the virtual-8086 monitor, before returning to the interrupted 8086 program (refer
to Section 16.3.1.2., “Handling an Interrupt or Exception With an 8086 Program Interrupt or
Exception Handler”).

NOTE
Methods 1 and 4 handling can handle a software interrupt in a virtual-8086
task with a regular protected-mode handler, but this approach requires all
virtual-8086 tasks to use the same software interrupt handlers, which
generally does not give sufficient latitude to the programs running in the
virtual-8086 tasks, particularly MS-DOS programs.

16.3.3.5. METHOD 6: SOFTWARE INTERRUPT HANDLING

Method 6 handling is enabled when the VME flag is set to 1, the IOPL value is less than 3, and
the bit for the interrupt or exception vector in the redirection bit map is set to 0. With method 6
interrupt handling, software interrupts are handled in the same manner as was described for
method 5 handling (refer to Section 16.3.3.4., “Method 5: Software Interrupt Handling”).
Method 6 differs from method 5 in that with the IOPL value set to less than 3, the VIF and VIP
flags in the EFLAGS register are enabled, providing virtual interrupt support for handling
class 2 maskable hardware interrupts (refer to Section 16.3.2., “Class 2—Maskable Hardware
Interrupt Handling in Virtual-8086 Mode Using the Virtual Interrupt Mechanism”). These flags
provide the virtual-8086 monitor with an efficient means of handling maskable hardware inter-
rupts that occur during a virtual-8086 mode task. Also, because the IOPL value is less than 3
and the VIF flag is enabled, the information pushed on the stack by the processor when invoking
the interrupt handler is slightly different between methods 5 and 6 (refer to Table 16-2).

16.4. PROTECTED-MODE VIRTUAL INTERRUPTS

The Intel Architecture processors (beginning with the Pentium® processor) also support the VIF
and VIP flags in the EFLAGS register in protected mode by setting the PVI (protected-mode

16-27
8086 EMULATION

virtual interrupt) flag in the CR4 register. Setting the PVI flag allows applications running at
privilege level 3 to execute the CLI and STI instructions without causing a general-protection
exception (#GP) or affecting hardware interrupts.
When the PVI flag is set to 1, the CPL is 3, and the IOPL is less than 3, the STI and CLI instruc-
tions set and clear the VIF flag in the EFLAGS register, leaving IF unaffected. In this mode of
operation, an application running in protected mode and at a CPL of 3 can inhibit interrupts in
the same manner as is described in Section 16.3.2., “Class 2—Maskable Hardware Interrupt
Handling in Virtual-8086 Mode Using the Virtual Interrupt Mechanism” for a virtual-8086
mode task. When the application executes the CLI instruction, the processor clears the VIF flag.
If the processor receives a maskable hardware interrupt when the VIF flag is clear, the processor
invokes the protected-mode interrupt handler. This handler checks the state of the VIF flag in
the EFLAGS register. If the VIF flag is clear (indicating that the active task does not want to
have interrupts handled now), the handler sets the VIP flag in the EFLAGS image on the stack
and returns to the privilege-level 3 application, which continues program execution. When the
application executes a STI instruction to set the VIF flag, the processor automatically invokes
the general-protection exception handler, which can then handle the pending interrupt. After
handing the pending interrupt, the handler typically sets the VIF flag and clears the VIP flag in
the EFLAGS image on the stack and executes a return to the application program. The next time
the processor receives a maskable hardware interrupt, the processor will handle it in the normal
manner for interrupts received while the processor is operating at a CPL of 3.
As with the virtual mode extension (enabled with the VME flag in the CR4 register), the
protected-mode virtual interrupt extension only affects maskable hardware interrupts (interrupt
vectors 32 through 255). NMI interrupts and exceptions are handled in the normal manner.
When protected-mode virtual interrupts are disabled (that is, when the PVI flag in control
register CR4 is set to 0, the CPL is less than 3, or the IOPL value is 3), then the CLI and STI
instructions execute in a manner compatible with the Intel486™ processor. That is, if the CPL
is greater (less privileged) than the I/O privilege level (IOPL), a general-protection exception
occurs. If the IOPL value is 3, CLI and STI clear or set the IF flag, respectively.
PUSHF, POPF, and IRET are executed like in the Intel486™ processor, regardless of whether
protected-mode virtual interrupts are enabled.
It is only possible to enter virtual-8086 mode through a task switch or the execution of an IRET
instruction, and it is only possible to leave virtual-8086 mode by faulting to a protected-mode
interrupt handler (typically the general-protection exception handler, which in turn calls the
virtual 8086-mode monitor). In both cases, the EFLAGS register is saved and restored. This is
not true, however, in protected mode when the PVI flag is set and the processor is not in virtual-
8086 mode. Here, it is possible to call a procedure at a different privilege level, in which case
the EFLAGS register is not saved or modified. However, the states of VIF and VIP flags are
never examined by the processor when the CPL is not 3.

16-28
 17
Mixing 16-Bit and
32-Bit Code
 MIXING 16-BIT AND 32-BIT CODE

CHAPTER 17
MIXING 16-BIT AND 32-BIT CODE

Program modules written to run on Intel Architecture processors can be either 16-bit modules
or 32-bit modules. Table 17-1 shows the characteristic of 16-bit and 32-bit modules.

Table 17-1. Characteristics of 16-Bit and 32-Bit Program Modules

Characteristic 16-Bit Program Modules 32-Bit Program Modules
Segment Size 0 to 64 KBytes 0 to 4 GBytes
Operand Sizes 8 bits and 16 bits 8 bits and 32 bits
Pointer Offset Size (Address Size) 16 bits 32 bits
Stack Pointer Size 16 Bits 32 Bits
Control Transfers Allowed to Code 16 Bits 32 Bits
Segments of This Size

The Intel Architecture processors function most efficiently when executing 32-bit program
modules. They can, however, also execute 16-bit program modules, in any of the following
ways:
• In real-address mode.
• In virtual-8086 mode.
• System management mode (SMM).
• As a protected-mode task, when the code, data, and stack segments for the task are all
configured as a 16-bit segments.
• By integrating 16-bit and 32-bit segments into a single protected-mode task.
• By integrating 16-bit operations into 32-bit code segments.
Real-address mode, virtual-8086 mode, and SMM are native 16-bit modes. A legacy program
assembled and/or compiled to run on an Intel 8086 or Intel 286 processor should run in real-
address mode or virtual-8086 mode without modification. Sixteen-bit program modules can also
be written to run in real-address mode for handling system initialization or to run in SMM for
handling system management functions. Refer to Chapter 16, 8086 Emulation for detailed infor-
mation on real-address mode and virtual-8086 mode; refer to Chapter 12, System Management
Mode (SMM) for information on SMM.
This chapter describes how to integrate 16-bit program modules with 32-bit program modules
when operating in protected mode and how to mix 16-bit and 32-bit code within 32-bit code
segments.

17-1
MIXING 16-BIT AND 32-BIT CODE

17.1. DEFINING 16-BIT AND 32-BIT PROGRAM MODULES

The following Intel Architecture mechanisms are used to distinguish between and support 16-
bit and 32-bit segments and operations:
• The D (default operand and address size) flag in code-segment descriptors.
• The B (default stack size) flag in stack-segment descriptors.
• 16-bit and 32-bit call gates, interrupt gates, and trap gates.
• Operand-size and address-size instruction prefixes.
• 16-bit and 32-bit general-purpose registers.
The D flag in a code-segment descriptor determines the default operand-size and address-size
for the instructions of a code segment. (In real-address mode and virtual-8086 mode, which do
not use segment descriptors, the default is 16 bits.) A code segment with its D flag set is a 32-bit
segment; a code segment with its D flag clear is a 16-bit segment.
The B flag in the stack-segment descriptor specifies the size of stack pointer (the 32-bit ESP
register or the 16-bit SP register) used by the processor for implicit stack references. The B flag
for all data descriptors also controls upper address range for expand down segments.
When transferring program control to another code segment through a call gate, interrupt gate,
or trap gate, the operand size used during the transfer is determined by the type of gate used (16-
bit or 32-bit), (not by the D-flag or prefix of the transfer instruction). The gate type determines
how return information is saved on the stack (or stacks).
For most efficient and trouble-free operation of the processor, 32-bit programs or tasks should
have the D flag in the code-segment descriptor and the B flag in the stack-segment descriptor
set, and 16-bit programs or tasks should have these flags clear. Program control transfers from
16-bit segments to 32-bit segments (and vice versa) are handled most efficiently through call,
interrupt, or trap gates.
Instruction prefixes can be used to override the default operand size and address size of a code
segment. These prefixes can be used in real-address mode as well as in protected mode and
virtual-8086 mode. An operand-size or address-size prefix only changes the size for the duration
of the instruction.

17.2. MIXING 16-BIT AND 32-BIT OPERATIONS WITHIN A CODE

SEGMENT
The following two instruction prefixes allow mixing of 32-bit and 16-bit operations within one
segment:
• The operand-size prefix (66H)
• The address-size prefix (67H)

These prefixes reverse the default size selected by the D flag in the code-segment descriptor. For
example, the processor can interpret the (MOV mem, reg) instruction in any of four ways:

17-2
 MIXING 16-BIT AND 32-BIT CODE

• In a 32-bit code segment:

— Moves 32 bits from a 32-bit register to memory using a 32-bit effective address.
— If preceded by an operand-size prefix, moves 16 bits from a 16-bit register to memory
using a 32-bit effective address.
— If preceded by an address-size prefix, moves 32 bits from a 32-bit register to memory
using a 16-bit effective address.
— If preceded by both an address-size prefix and an operand-size prefix, moves 16 bits
from a 16-bit register to memory using a 16-bit effective address.
• In a 16-bit code segment:
— Moves 16 bits from a 16-bit register to memory using a 16-bit effective address.
— If preceded by an operand-size prefix, moves 32 bits from a 32-bit register to memory
using a 16-bit effective address.
— If preceded by an address-size prefix, moves 16 bits from a 16-bit register to memory
using a 32-bit effective address.
— If preceded by both an address-size prefix and an operand-size prefix, moves 32 bits
from a 32-bit register to memory using a 32-bit effective address.
The previous examples show that any instruction can generate any combination of operand size
and address size regardless of whether the instruction is in a 16- or 32-bit segment. The choice
of the 16- or 32-bit default for a code segment is normally based on the following criteria:
• Performance—Always use 32-bit code segments when possible. They run much faster
than 16-bit code segments on P6 family processors, and somewhat faster on earlier Intel
Architecture processors.
• The operating system the code segment will be running on—If the operating system is a
16-bit operating system, it may not support 32-bit program modules.
• Mode of operation—If the code segment is being designed to run in real-address mode,
virtual-8086 mode, or SMM, it must be a 16-bit code segment.
• Backward compatibility to earlier Intel Architecture processors—If a code segment
must be able to run on an Intel 8086 or Intel 286 processor, it must be a 16-bit code
segment.

17.3. SHARING DATA AMONG MIXED-SIZE CODE SEGMENTS

Data segments can be accessed from both 16-bit and 32-bit code segments. When a data segment
that is larger than 64 KBytes is to be shared among 16- and 32-bit code segments, the data that
is to be accessed from the 16-bit code segments must be located within the first 64 KBytes of
the data segment. The reason for this is that 16-bit pointers by definition can only point to the
first 64 KBytes of a segment.

17-3
MIXING 16-BIT AND 32-BIT CODE

A stack that spans less than 64 KBytes can be shared by both 16- and 32-bit code segments. This
class of stacks includes:
• Stacks in expand-up segments with the G (granularity) and B (big) flags in the stack-
segment descriptor clear.
• Stacks in expand-down segments with the G and B flags clear.
• Stacks in expand-up segments with the G flag set and the B flag clear and where the stack
is contained completely within the lower 64 KBytes. (Offsets greater than FFFFH can be
used for data, other than the stack, which is not shared.)
Refer to Section 3.4.3., “Segment Descriptors” in Chapter 3, Protected-Mode Memory Manage-
ment for a description of the G and B flags and the expand-down stack type.
The B flag cannot, in general, be used to change the size of stack used by a 16-bit code segment.
This flag controls the size of the stack pointer only for implicit stack references such as those
caused by interrupts, exceptions, and the PUSH, POP, CALL, and RET instructions. It does not
control explicit stack references, such as accesses to parameters or local variables. A 16-bit code
segment can use a 32-bit stack only if the code is modified so that all explicit references to the
stack are preceded by the 32-bit address-size prefix, causing those references to use 32-bit
addressing and explicit writes to the stack pointer are preceded by a 32-bit operand-size prefix.
In 32-bit, expand-down segments, all offsets may be greater than 64 KBytes; therefore, 16-bit
code cannot use this kind of stack segment unless the code segment is modified to use 32-bit
addressing.

17.4. TRANSFERRING CONTROL AMONG MIXED-SIZE CODE

SEGMENTS
There are three ways for a procedure in a 16-bit code segment to safely make a call to a 32-bit
code segment:
• Make the call through a 32-bit call gate.
• Make a 16-bit call to a 32-bit interface procedure. The interface procedure then makes a
32-bit call to the intended destination.
• Modify the 16-bit procedure, inserting an operand-size prefix before the call, to change it
to a 32-bit call.
Likewise, there are three ways for procedure in a 32-bit code segment to safely make a call to a
16-bit code segment:
• Make the call through a 16-bit call gate. Here, the EIP value at the CALL instruction
cannot exceed FFFFH.
• Make a 32-bit call to a 16-bit interface procedure. The interface procedure then makes a
16-bit call to the intended destination.
• Modify the 32-bit procedure, inserting an operand-size prefix before the call, changing it to
a 16-bit call. Be certain that the return offset does not exceed FFFFH.

17-4
 MIXING 16-BIT AND 32-BIT CODE

These methods of transferring program control overcome the following architectural limitations
imposed on calls between 16-bit and 32-bit code segments:
• Pointers from 16-bit code segments (which by default can only be 16-bits) cannot be used
to address data or code located beyond FFFFH in a 32-bit segment.
• The operand-size attributes for a CALL and its companion RETURN instruction must be
the same to maintain stack coherency. This is also true for implicit calls to interrupt and
exception handlers and their companion IRET instructions.
• A 32-bit parameters (particularly a pointer parameter) greater than FFFFH cannot be
squeezed into a 16-bit parameter location on a stack.
• The size of the stack pointer (SP or ESP) changes when switching between 16-bit and
32-bit code segments.
These limitations are discussed in greater detail in the following sections.

17.4.1. Code-Segment Pointer Size

For control-transfer instructions that use a pointer to identify the next instruction (that is, those
that do not use gates), the operand-size attribute determines the size of the offset portion of the
pointer. The implications of this rule are as follows:
• A JMP, CALL, or RET instruction from a 32-bit segment to a 16-bit segment is always
possible using a 32-bit operand size, providing the 32-bit pointer does not exceed FFFFH.
• A JMP, CALL, or RET instruction from a 16-bit segment to a 32-bit segment cannot
address a destination greater than FFFFH, unless the instruction is given an operand-size
prefix.
Refer to Section 17.4.5., “Writing Interface Procedures” for an interface procedure that can
transfer program control from 16-bit segments to destinations in 32-bit segments beyond
FFFFH.

17.4.2. Stack Management for Control Transfer

Because the stack is managed differently for 16-bit procedure calls than for 32-bit calls, the
operand-size attribute of the RET instruction must match that of the CALL instruction (refer to
Figure 17-1). On a 16-bit call, the processor pushes the contents of the 16-bit IP register and (for
calls between privilege levels) the 16-bit SP register. The matching RET instruction must also
use a 16-bit operand size to pop these 16-bit values from the stack into the 16-bit registers.
A 32-bit CALL instruction pushes the contents of the 32-bit EIP register and (for inter-privilege-
level calls) the 32-bit ESP register. Here, the matching RET instruction must use a 32-bit
operand size to pop these 32-bit values from the stack into the 32-bit registers. If the two parts
of a CALL/RET instruction pair do not have matching operand sizes, the stack will not be
managed correctly and the values of the instruction pointer and stack pointer will not be restored
to correct values.

17-5
MIXING 16-BIT AND 32-BIT CODE

Without Privilege Transition

After 16-bit Call After 32-bit Call

31 0 31 0

PARM 2 PARM 1 PARM 2

Stack
Growth
CS IP SP PARM 1

CS

EIP ESP

With Privilege Transition

After 16-bit Call After 32-bit Call

31 0 31 0

SS SP SS

PARM 2 PARM 1 ESP

Stack
Growth
CS IP SP PARM 2

PARM 1

CS

EIP ESP

Undefined

Figure 17-1. Stack after Far 16- and 32-Bit Calls

While executing 32-bit code, if a call is made to a 16-bit code segment which is at the same or
a more privileged level (that is, the DPL of the called code segment is less than or equal to the
CPL of the calling code segment) through a 16-bit call gate, then the upper 16-bits of the ESP
register may be unreliable upon returning to the 32-bit code segment (that is, after executing a
RET in the 16-bit code segment).
When the CALL instruction and its matching RET instruction are in code segments that have D
flags with the same values (that is, both are 32-bit code segments or both are 16-bit code
segments), the default settings may be used. When the CALL instruction and its matching RET
instruction are in segments which have different D-flag settings, an operand-size prefix must be
used.

17-6
 MIXING 16-BIT AND 32-BIT CODE

17.4.2.1. CONTROLLING THE OPERAND-SIZE ATTRIBUTE FOR A CALL

Three things can determine the operand-size of a call:
• The D flag in the segment descriptor for the calling code segment.
• An operand-size instruction prefix.
• The type of call gate (16-bit or 32-bit), if a call is made through a call gate.
When a call is made with a pointer (rather than a call gate), the D flag for the calling code
segment determines the operand-size for the CALL instruction. This operand-size attribute can
be overridden by prepending an operand-size prefix to the CALL instruction. So, for example,
if the D flag for a code segment is set for 16 bits and the operand-size prefix is used with a CALL
instruction, the processor will cause the information stored on the stack to be stored in 32-bit
format. If the call is to a 32-bit code segment, the instructions in that code segment will be able
to read the stack coherently. Also, a RET instruction from the 32-bit code segment without an
operand-size prefix will maintain stack coherency with the 16-bit code segment being returned
to.
When a CALL instruction references a call-gate descriptor, the type of call is determined by the
type of call gate (16-bit or 32-bit). The offset to the destination in the code segment being called
is taken from the gate descriptor; therefore, if a 32-bit call gate is used, a procedure in a 16-bit
code segment can call a procedure located more than 64 Kbytes from the base of a 32-bit code
segment, because a 32-bit call gate uses a 32-bit offset.
Note that regardless of the operand size of the call and how it is determined, the size of the stack
pointer used (SP or ESP) is always controlled by the B flag in the stack-segment descriptor
currently in use (that is, when B is clear, SP is used, and when B is set, ESP is used).
An unmodified 16-bit code segment that has run successfully on an 8086 processor or in
real-mode on a P6-family processor will have its D flag clear and will not use operand-size over-
ride prefixes. As a result, all CALL instructions in this code segment will use the 16-bit operand-
size attribute. Procedures in these code segments can be modified to safely call procedures to
32-bit code segments in either of two ways:
• Relink the CALL instruction to point to 32-bit call gates (refer to Section 17.4.2.2.,
“Passing Parameters With a Gate”).
• Add a 32-bit operand-size prefix to each CALL instruction.

17.4.2.2. PASSING PARAMETERS WITH A GATE

When referencing 32-bit gates with 16-bit procedures, it is important to consider the number of
parameters passed in each procedure call. The count field of the gate descriptor specifies the size
of the parameter string to copy from the current stack to the stack of a more privileged (numer-
ically lower privilege level) procedure. The count field of a 16-bit gate specifies the number of
16-bit words to be copied, whereas the count field of a 32-bit gate specifies the number of 32-bit
doublewords to be copied. The count field for a 32-bit gate must thus be half the size of the
number of words being placed on the stack by a 16-bit procedure. Also, the 16-bit procedure
must use an even number of words as parameters.

17-7
MIXING 16-BIT AND 32-BIT CODE

17.4.3. Interrupt Control Transfers

A program-control transfer caused by an exception or interrupt is always carried out through an
interrupt or trap gate (located in the IDT). Here, the type of the gate (16-bit or 32-bit) determines
the operand-size attribute used in the implicit call to the exception or interrupt handler procedure
in another code segment.
A 32-bit interrupt or trap gate provides a safe interface to a 32-bit exception or interrupt handler
when the exception or interrupt occurs in either a 32-bit or a 16-bit code segment. It is sometimes
impractical, however, to place exception or interrupt handlers in 16-bit code segments, because
only 16-bit return addresses are saved on the stack. If an exception or interrupt occurs in a 32-bit
code segment when the EIP was greater than FFFFH, the 16-bit handler procedure cannot
provide the correct return address.

17.4.4. Parameter Translation

When segment offsets or pointers (which contain segment offsets) are passed as parameters
between 16-bit and 32-bit procedures, some translation is required. If a 32-bit procedure passes
a pointer to data located beyond 64 KBytes to a 16-bit procedure, the 16-bit procedure cannot
use it. Except for this limitation, interface code can perform any format conversion between
32-bit and 16-bit pointers that may be needed.
Parameters passed by value between 32-bit and 16-bit code also may require translation between
32-bit and 16-bit formats. The form of the translation is application-dependent.

17.4.5. Writing Interface Procedures

Placing interface code between 32-bit and 16-bit procedures can be the solution to the following
interface problems:
• Allowing procedures in 16-bit code segments to call procedures with offsets greater than
FFFFH in 32-bit code segments.
• Matching operand-size attributes between companion CALL and RET instructions.
• Translating parameters (data), including managing parameter strings with a variable count
or an odd number of 16-bit words.
• The possible invalidation of the upper bits of the ESP register.
The interface procedure is simplified where these rules are followed.
1. The interface procedure must reside in a 32-bit code segment (the D flag for the code-
segment descriptor is set).
2. All procedures that may be called by 16-bit procedures must have offsets not greater than
FFFFH.
3. All return addresses saved by 16-bit procedures must have offsets not greater than FFFFH.

17-8
 MIXING 16-BIT AND 32-BIT CODE

The interface procedure becomes more complex if any of these rules are violated. For example,
if a 16-bit procedure calls a 32-bit procedure with an entry point beyond FFFFH, the interface
procedure will need to provide the offset to the entry point. The mapping between 16- and 32-bit
addresses is only performed automatically when a call gate is used, because the gate descriptor
for a call gate contains a 32-bit address. When a call gate is not used, the interface code must
provide the 32-bit address.
The structure of the interface procedure depends on the types of calls it is going to support, as
follows:
• Calls from 16-bit procedures to 32-bit procedures. Calls to the interface procedure from
a 16-bit code segment are made with 16-bit CALL instructions (by default, because the D
flag for the calling code-segment descriptor is clear), and 16-bit operand-size prefixes are
used with RET instructions to return from the interface procedure to the calling procedure.
Calls from the interface procedure to 32-bit procedures are performed with 32-bit CALL
instructions (by default, because the D flag for the interface procedure’s code segment is
set), and returns from the called procedures to the interface procedure are performed with
32-bit RET instructions (also by default).
• Calls from 32-bit procedures to 16-bit procedures. Calls to the interface procedure from
a 32-bit code segment are made with 32-bit CALL instructions (by default), and returns to
the calling procedure from the interface procedure are made with 32-bit RET instructions
(also by default). Calls from the interface procedure to 16-bit procedures require the CALL
instructions to have the operand-size prefixes, and returns from the called procedures to the
interface procedure are performed with 16-bit RET instructions (by default).

17-9
MIXING 16-BIT AND 32-BIT CODE

17-10
 18
Intel Architecture
Compatibility
 CHAPTER 18
INTEL ARCHITECTURE COMPATIBILITY

All Intel Architecture processors are binary compatible. Compatibility means that, within
certain limited constraints, programs that execute on previous generations of Intel Architecture
processors will produce identical results when executed on later Intel Architecture processors.
The compatibility constraints and any implementation differences between the Intel Architec-
ture processors are described in this chapter.
Each new Intel Architecture processor has enhanced the software visible architecture from that
found in earlier Intel Architecture processors. Those enhancements have been defined with
consideration for compatibility with previous and future processors. This chapter also summa-
rizes the compatibility considerations for those extensions.

18.1. INTEL ARCHITECTURE FAMILIES AND CATEGORIES

Intel Architecture processors are referred to in several different ways in this chapter, depending
on the type of compatibility information being related, as described in the following:
• Intel Architecture Processors—All the Intel processors based on the Intel Architecture,
which include the 8086/88, Intel 286, Intel386™, Intel486™, Pentium®, and P6 family
processors.
• 32-bit Processors—All the Intel Architecture processors that use a 32-bit architecture,
which include the Intel386™, Intel486™, Pentium®, and P6 family processors.
• 16-bit Processors—All the Intel Architecture processors that use a 16-bit architecture,
which include the 8086/88 and Intel 286 processors.
• P6 Family Processors—All the Intel Architecture processors that are based on the P6
family micro-architecture, which include the Pentium® Pro, Pentium® II, Pentium® III and
future P6 family processors.

18.2. RESERVED BITS

Throughout this manual, certain bits are marked as reserved in many register and memory layout
descriptions. When bits are marked as undefined or reserved, it is essential for compatibility
with future processors that software treat these bits as having a future, though unknown effect.
Software should follow these guidelines in dealing with reserved bits:

• Do not depend on the states of any reserved bits when testing the values of registers or
memory locations that contain such bits. Mask out the reserved bits before testing.
• Do not depend on the states of any reserved bits when storing them to memory or to a
register.

18-1
INTEL ARCHITECTURE COMPATIBILITY

• Do not depend on the ability to retain information written into any reserved bits.
• When loading a register, always load the reserved bits with the values indicated in the
documentation, if any, or reload them with values previously read from the same register.
Software written for existing Intel Architecture processor that handles reserved bits correctly
will port to future Intel Architecture processors without generating protection exceptions.

18.3. ENABLING NEW FUNCTIONS AND MODES

Most of the new control functions defined for the P6 family and Pentium® processors are
enabled by new mode flags in the control registers (primarily register CR4). This register is
undefined for Intel Architecture processors earlier than the Pentium ® processor. Attempting to
access this register with an Intel486™ or earlier Intel Architecture processor results in an
invalid-opcode exception (#UD). Consequently, programs that execute correctly on the
Intel486™ or earlier Intel Architecture processor cannot erroneously enable these functions.
Attempting to set a reserved bit in register CR4 to a value other than its original value results in
a general-protection exception (#GP). So, programs that execute on the P6 family and Pentium®
processors cannot erroneously enable functions that may be implemented in future Intel Archi-
tecture processors.
The P6 family and Pentium® processors do not check for attempts to set reserved bits in model-
specific registers. It is the obligation of the software writer to enforce this discipline. These
reserved bits may be used in future Intel processors.

18.4. DETECTING THE PRESENCE OF NEW FEATURES

THROUGH SOFTWARE
Software can check for the presence of new architectural features and extensions in either of two
ways:
• Test for the presence of the feature or extension — Software can test for the presence of
new flags in the EFLAGS register and control registers. If these flags are reserved
(meaning not present in the processor executing the test), an exception is generated.
Likewise, software can attempt to execute a new instruction, which results in an invalid-
opcode exception (#UD) being generated if it is not supported.
• Execute the CPUID instruction — The CPUID instruction (added to the Intel Architecture
in the Pentium® processor) indicates the presence of new features directly.
Refer to Chapter 10, Processor Identification and Feature Determination, in the Intel Architec-
ture Software Developer’s Manual, Volume 1, for detailed information on detecting new
processor features and extensions.

18-2
 INTEL ARCHITECTURE COMPATIBILITY

18.5. MMX™ TECHNOLOGY

The Pentium® processor with MMX™ technology introduced the MMX™ technology and a set
of MMX™ instructions to the Intel Architecture. The MMX™ instructions are summarized in
Chapter 6, Instruction Set Summary, in the Intel Architecture Software Developer’s Manual,
Volume 1 and are described in detail in Chapter 3 in the Intel Architecture Software Developer’s
Manual, Volume 2. The MMX™ technology and MMX™ instructions are also included in the
Pentium® II and Pentium® III processors.

18.6. STREAMING SIMD EXTENSIONS

The Pentium® III processor introduced the Streaming SIMD Extensions. This is a set of new
instructions added to enhance perfomance of several classes of applications. The Streaming
SIMD Extensions are summarized in Chapter 6, Instruction Set Summary, in the Intel Architec-
ture Software Developer’s Manual, Volume 1 and are described in detail in Chapter 3 in the Intel
Architecture Software Developer’s Manual, Volume 2. Several of these new instructions operate
in the same register space as the MMX™ instructions. When using these instructions, the rules
that apply to MMX™ technology programming apply to this subset of the new instructions as
well.

18.7. NEW INSTRUCTIONS IN THE PENTIUM® AND LATER INTEL

ARCHITECTURE PROCESSORS
Table 18-1 identifies the instructions introduced into the Intel Architecture in the Pentium® and
later Intel Architecture processors.

Table 18-1. New Instructions in the Pentium® and Later Intel Architecture Processors
Instruction CPUID Identification Bits Introduced In
Streaming SIMD Extensions EDX, Bit 25 Pentium® III processor
SYSENTER/SYSEXIT(fast system call) EDX, Bit 11 Pentium® II processor
FXSAVE/FXRSTOR(fast save/restore) EDX, Bit 24 Pentium® II processor
CMOVcc (conditional move) EDX, Bit 15 Pentium® Pro processor
FCMOVcc (floating-point conditional move) EDX, Bits 0 and 15
FCOMI (floating-point compare and set EDX, Bits 0 and 15
EFLAGS)
RDPMC (read performance monitoring EAX, Bits 8-11, set to 6H;
counters) refer to Note 1
UD2 (undefined) EAX, Bits 8-11, set to 6H

18-3
INTEL ARCHITECTURE COMPATIBILITY

Table 18-1. New Instructions in the Pentium® and Later Intel Architecture Processors
Instruction CPUID Identification Bits Introduced In
CMPXCHG8B (compare and exchange 8 EDX, Bit 8 Pentium® processor
bytes)
CPUID (CPU identification) None; refer to Note 2
RDTSC (read time-stamp counter) EDX, Bit 4
RDMSR (read model-specific register) EDX, Bit 5
WRMSR (write model-specific register) EDX, Bit 5
MMX™ Instructions EDX, Bit 23

NOTES:
1. The RDPMC instruction was introduced in the P6 family of processors and added to later model Pentium®
processors. This instruction is model specific in nature and not architectural.
2. The CPUID instruction is available in all Pentium® and P6 family processors and in later models of the
Intel486™ processors. The ability to set and clear the ID flag (bit 21) in the EFLAGS register indicates the
availability of the CPUID instruction.

18-4
 INTEL ARCHITECTURE COMPATIBILITY

18.7.1. Instructions Added Prior to the Pentium® Processor

The following instructions were added in the Intel486™ processor:
• BSWAP (byte swap) instruction.
• XADD (exchange and add) instruction.
• CMPXCHG (compare and exchange) instruction.
• ΙNVD (invalidate cache) instruction.
• WBINVD (write-back and invalidate cache) instruction.
• INVLPG (invalidate TLB entry) instruction.

The following instructions were added in the Intel386™ processor:

• LSS, LFS, and LGS (load SS, FS, and GS registers).
• Long-displacement conditional jumps.
• Single-bit instructions.
• Bit scan instructions.
• Double-shift instructions.
• Byte set on condition instruction.
• Move with sign/zero extension.
• Generalized multiply instruction.
• MOV to and from control registers.
• MOV to and from test registers (now obsolete).
• MOV to and from debug registers.

• RSM (resume from SMM). This instruction was introduced in the Intel386™ SL and
Intel486™ SL processors.
The following instructions were added in the Intel 387 math coprocessor:
• FPREM1.
• FUCOM, FUCOMP, and FUCOMPP.

18.8. OBSOLETE INSTRUCTIONS

The MOV to and from test registers instructions were removed from the Pentium® and future
Intel Architecture processors. Execution of these instructions generates an invalid-opcode
exception (#UD).

18-5
INTEL ARCHITECTURE COMPATIBILITY

18.9. UNDEFINED OPCODES

All new instructions defined for Intel Architecture processors use binary encodings that were
reserved on earlier-generation processors. Attempting to execute a reserved opcode always
results in an invalid-opcode (#UD) exception being generated. Consequently, programs that
execute correctly on earlier-generation processors cannot erroneously execute these instructions
and thereby produce unexpected results when executed on later Intel Architecture processors.

18.10.NEW FLAGS IN THE EFLAGS REGISTER

The section titled “EFLAGS Register” in Chapter 3 of the Intel Architecture Software Devel-
oper’s Manual, Volume 1, shows the configuration of flags in the EFLAGS register for the P6
family processors. No new flags have been added to this register in the P6 family processors.
The flags added to this register in the Pentium® and Intel486™ processors are described in the
following sections.
The following flags were added to the EFLAGS register in the Pentium® processor:
• VIF (virtual interrupt flag), bit 19.
• VIP (virtual interrupt pending), bit 20.
• ID (identification flag), bit 21.

The AC flag (bit 18) was added to the EFLAGS register in the Intel486™ processor.

18.10.1. Using EFLAGS Flags to Distinguish Between 32-Bit Intel

Architecture Processors
The following bits in the EFLAGS register that can be used to differentiate between the 32-bit
Intel Architecture processors:

• Bit 18 (the AC flag) can be used to distinguish an Intel386™ processor from the P6 family,
Pentium®, and Intel486™ processors. Since it is not implemented on the Intel386™
processor, it will always be clear.
• Bit 21 (the ID flag) indicates whether an application can execute the CPUID instruction.
The ability to set and clear this bit indicates that the processor is a P6 family or Pentium®
processor. The CPUID instruction can then be used to determine which processor.
• Bits 19 (the VIF flag) and 20 (the VIP flag) will always be zero on processors that do not
support virtual mode extensions, which includes all 32-bit processors prior to the Pentium®
processor.
Refer to Chapter 10, Processor Identification and Feature Determination, in the Intel Architec-
ture Software Developer’s Manual, Volume 1, for more information on identifying processors.

18-6
 INTEL ARCHITECTURE COMPATIBILITY

18.11.STACK OPERATIONS
This section identifies the differences in stack implementation between the various Intel Archi-
tecture processors.

18.11.1. PUSH SP
The P6 family, Pentium®, Intel486™, Intel386™, and Intel 286 processors push a different value
on the stack for a PUSH SP instruction than the 8086 processor. The 32-bit processors push the
value of the SP register before it is decremented as part of the push operation; the 8086 processor
pushes the value of the SP register after it is decremented. If the value pushed is important,
replace PUSH SP instructions with the following three instructions:
PUSH BP
MOV BP, SP
XCHG BP, [BP]

This code functions as the 8086 processor PUSH SP instruction on the P6 family, Pentium®,
Intel486™, Intel386™, and Intel 286 processors.

18.11.2. EFLAGS Pushed on the Stack

The setting of the stored values of bits 12 through 15 (which includes the IOPL field and the NT
flag) in the EFLAGS register by the PUSHF instruction, by interrupts, and by exceptions is
different with the 32-bit Intel Architecture processors than with the 8086 and Intel 286 proces-
sors. The differences are as follows:
• 8086 processor—bits 12 through 15 are always set.
• Intel 286 processor—bits 12 through 15 are always cleared in real-address mode.
• 32-bit processors in real-address mode—bit 15 (reserved) is always cleared, and bits 12
through 14 have the last value loaded into them.

18.12.FPU
This section addresses the issues that must be faced when porting floating-point software
designed to run on earlier Intel Architecture processors and math coprocessors to a Pentium® or
P6 family processor with integrated FPU. To software, a P6 family processor looks very much
like a Pentium® processor. Floating-point software which runs on a Pentium® or Intel486™ DX
processor, or on an Intel486™ SX processor/Intel 487 SX math coprocessor system or an
Intel386™ processor/Intel 387 math coprocessor system, will run with at most minor modifica-
tions on a P6 family processor. To port code directly from an Intel 286 processor/Intel 287
math coprocessor system or an Intel 8086 processor/8087 math coprocessor system to the
Pentium® and P6 family processors, certain additional issues must be addressed.

18-7
INTEL ARCHITECTURE COMPATIBILITY

In the following sections, the term “32-bit Intel Architecture FPUs” refers to the P6 family,
Pentium®, and Intel486™ DX processors, and to the Intel 487 SX and Intel 387 math coproces-
sors; the term “16-bit Intel Architecture math coprocessors” refers to the Intel 287 and 8087
math coprocessors.

18.12.1. Control Register CR0 Flags

The ET, NE, and MP flags in control register CR0 control the interface between the integer unit
of an Intel Architecture processor and either its internal FPU or an external math coprocessor.
The effect of these flags in the various Intel Architecture processors are described in the
following paragraphs.
The ET (extension type) flag (bit 4 of the CR0 register) is used in the Intel386™ processor to
indicate whether the math coprocessor in the system is an Intel 287 math coprocessor (flag is
clear) or an Intel 387 DX math coprocessor (flag is set). This bit is hardwired to 1 in the P6
family, Pentium®, and Intel486™ processors.
The NE (Numeric Exception) flag (bit 5 of the CR0 register) is used in the P6 family, Pentium®,
and Intel486™ processors to determine whether unmasked floating-point exceptions are
reported internally through interrupt vector 16 (flag is set) or externally through an external
interrupt (flag is clear). On a hardware reset, the NE flag is initialized to 0, so software using the
automatic internal error-reporting mechanism must set this flag to 1. This flag is nonexistent on
the Intel386™ processor.
As on the Intel 286 and Intel386™ processors, the MP (monitor coprocessor) flag (bit 1 of
register CR0) determines whether the WAIT/FWAIT instructions or waiting-type floating-point
instructions trap when the context of the FPU is different from that of the currently-executing
task. If the MP and TS flag are set, then a WAIT/FWAIT instruction and waiting instructions
will cause a device-not-available exception (interrupt vector 7). The MP flag is used on the Intel
286 and Intel386™ processors to support the use of a WAIT/FWAIT instruction to wait on a
device other than a math coprocessor. The device reports its status through the BUSY# pin.
Since the P6 family, Pentium®, and Intel486™ processors do not have such a pin, the MP flag
has no relevant use and should be set to 1 for normal operation.

18.12.2. FPU Status Word

This section identifies differences to the FPU status word for the different Intel Architecture
processors and math coprocessors, the reason for the differences, and their impact on software.

18.12.2.1. CONDITION CODE FLAGS (C0 THROUGH C3)

The following information pertains to differences in the use of the condition code flags (C0
through C3) located in bits 8, 9, 10, and 14 of the FPU status word.
After execution of an FINIT instruction or a hardware reset on a 32-bit Intel Architecture FPU,
the condition code flags are set to 0. The same operations on a 16-bit Intel Architecture math

18-8
 INTEL ARCHITECTURE COMPATIBILITY

coprocessor leave these flags intact (they contain their prior value). This difference in operation
has no impact on software and provides a consistent state after reset.
Transcendental instruction results in the core range of the P6 family and Pentium® processors
may differ from the Intel486™ DX processor and Intel 487 SX math coprocessor by 2 to 3 units
in the last place (ulps)—(refer to “Transcendental Instruction Accuracy” in Chapter 7 of the Intel
Architecture Software Developer’s Manual, Volume 1). As a result, the value saved in the C1 flag
may also differ.
After an incomplete FPREM/FPREM1 instruction, the C0, C1, and C3 flags are set to 0 on the
32-bit Intel Architecture FPUs. After the same operation on a 16-bit Intel Architecture math
coprocessor, these flags are left intact.
On the 32-bit Intel Architecture FPUs, the C2 flag serves as an incomplete flag for the FTAN
instruction. On the 16-bit Intel Architecture math coprocessors, the C2 flag is undefined for the
FPTAN instruction. This difference has no impact on software, because Intel 287 or 8087
programs do not check C2 after an FPTAN instruction. The use of this flag on later processors
allows fast checking of operand range.

18.12.2.2. STACK FAULT FLAG

When unmasked stack overflow or underflow occurs on a 32-bit Intel Architecture FPU, the IE
flag (bit 0) and the SF flag (bit 6) of the FPU status word are set to indicate a stack fault and
condition code flag C1 is set or cleared to indicate overflow or underflow, respectively. When
unmasked stack overflow or underflow occurs on a 16-bit Intel Architecture math coprocessor,
only the IE flag is set. Bit 6 is reserved on these processors. The addition of the SF flag on a 32-
bit Intel Architecture FPU has no impact on software. Existing exception handlers need not
change, but may be upgraded to take advantage of the additional information.

18.12.3. FPU Control Word

Only affine closure is supported for infinity control on a 32-bit Intel Architecture FPU. The
infinity control flag (bit 12 of the FPU control word) remains programmable on these proces-
sors, but has no effect. This change was made to conform to IEEE Standard 754. On a 16-bit
Intel Architecture math coprocessor, both affine and projective closures are supported, as deter-
mined by the setting of bit 12. After a hardware reset, the default value of bit 12 is projective.
Software that requires projective infinity arithmetic may give different results.

18.12.4. FPU Tag Word

When loading the tag word of a 32-bit Intel Architecture FPU, using an FLDENV, FRSTOR, or
FXRSTOR (Pentium® III processor only) instruction, the processor examines the incoming tag
and classifies the location only as empty or nonempty. Thus, tag values of 00, 01, and 10 are
interpreted by the processor to indicate a nonempty location. The tag value of 11 is interpreted
by the processor to indicate an empty location. Subsequent operations on a nonempty register
always examine the value in the register, not the value in its tag. The FSTENV, FSAVE, and

18-9
INTEL ARCHITECTURE COMPATIBILITY

FXSAVE (Pentium® III processor only) instructions examine the nonempty registers and put the
correct values in the tags before storing the tag word.
The corresponding tag for a 16-bit Intel Architecture math coprocessor is checked before each
register access to determine the class of operand in the register; the tag is updated after every
change to a register so that the tag always reflects the most recent status of the register. Software
can load a tag with a value that disagrees with the contents of a register (for example, the register
contains a valid value, but the tag says special). Here, the 16-bit Intel Architecture math copro-
cessors honor the tag and do not examine the register.
Software written to run on a 16-bit Intel Architecture math coprocessor may not operate
correctly on a 16-bit Intel Architecture FPU, if it uses FLDENV, FRSTOR, or FXRSTOR
(Pentium® III processor only) to change tags to values (other than to empty) that are different
from actual register contents.
The encoding in the tag word for the 32-bit Intel Architecture FPUs for unsupported data
formats (including pseudo-zero and unnormal) is special (10B), to comply with the IEEE Stan-
dard 754. The encoding in the 16-bit Intel Architecture math coprocessors for pseudo-zero and
unnormal is valid (00B) and the encoding for other unsupported data formats is special (10B).
Code that recognizes the pseudo-zero or unnormal format as valid must therefore be changed if
it is ported to a 32-bit Intel Architecture FPU.

18.12.5. Data Types

This section discusses the differences of data types for the various Intel Architecture FPUs and
math coprocessors.

18.12.5.1. NaNs
The 32-bit Intel Architecture FPUs distinguish between signaling NaNs (SNaNs) and quiet
NaNs (QNaNs). These FPUs only generate QNaNs and normally do not generate an exception
upon encountering a QNaN. An invalid operation exception (#I) is generated only upon encoun-
tering a SNaN, except for the FCOM, FIST, and FBSTP instructions, which also generates an
invalid operation exceptions for a QNaNs. This behavior matches the IEEE Standard 754.
The 16-bit Intel Architecture math coprocessors only generate one kind of NaN (the equivalent
of a QNaN), but the raise an invalid operation exception upon encountering any kind of NaN.
When porting software written to run on a 16-bit Intel Architecture math coprocessor to a 32-bit
Intel Architecture FPU, uninitialized memory locations that contain QNaNs should be changed
to SNaNs to cause the FPU or math coprocessor to fault when uninitialized memory locations
are referenced.

18.12.5.2. PSEUDO-ZERO, PSEUDO-NaN, PSEUDO-INFINITY, AND

UNNORMAL FORMATS
The 32-bit Intel Architecture FPUs neither generate nor support the pseudo-zero, pseudo-NaN,
pseudo-infinity, and unnormal formats. Whenever they encounter them in an arithmetic opera-

18-10
 INTEL ARCHITECTURE COMPATIBILITY

tion, they raise an invalid operation exception. The 16-bit Intel Architecture math coprocessors
define and support special handling for these formats. Support for these formats was dropped to
conform with the IEEE Standard 754.
This change should not impact software ported from 16-bit Intel Architecture math coprocessors
to 32-bit Intel Architecture FPUs. The 32-bit Intel Architecture FPUs do not generate these
formats, and therefore will not encounter them unless software explicitly loads them in the data
registers. The only affect may be in how software handles the tags in the tag word (refer to
Section 18.12.4., “FPU Tag Word”).

18.12.6. Floating-Point Exceptions

This section identifies the implementation differences in exception handling for floating-point
instructions in the various Intel Architecture FPUs and math coprocessors.

18.12.6.1. DENORMAL OPERAND EXCEPTION (#D)

When the denormal operand exception is masked, the 32-bit Intel Architecture FPUs automati-
cally normalize denormalized numbers when possible; whereas, the 16-bit Intel Architecture
math coprocessors return a denormal result. A program written to run on a 16-bit Intel Architec-
ture math coprocessor that uses the denormal exception solely to normalize denormalized
operands is redundant when run on the 32-bit Intel Architecture FPUs. If such a program is run
on 32-bit Intel Architecture FPUs, performance can be improved by masking the denormal
exception. Floating-point programs run faster when the FPU performs normalization of denor-
malized operands.
The denormal operand exception is not raised for transcendental instructions and the FXTRACT
instruction on the 16-bit Intel Architecture math coprocessors. This exception is raised for these
instructions on the 32-bit Intel Architecture FPUs. The exception handlers ported to these latter
processors need to be changed only if the handlers gives special treatment to different opcodes.

18.12.6.2. NUMERIC OVERFLOW EXCEPTION (#O)

On the 32-bit Intel Architecture FPUs, when the numeric overflow exception is masked and the
rounding mode is set to chop (toward 0), the result is the largest positive or smallest negative
number. The 16-bit Intel Architecture math coprocessors do not signal the overflow exception
when the masked response is not ∞; that is, they signal overflow only when the rounding control
is not set to round to 0. If rounding is set to chop (toward 0), the result is positive or negative ∞.
Under the most common rounding modes, this difference has no impact on existing software.
If rounding is toward 0 (chop), a program on a 32-bit Intel Architecture FPU produces, under
overflow conditions, a result that is different in the least significant bit of the significand,
compared to the result on a 16-bit Intel Architecture math coprocessor. The reason for this differ-
ence is IEEE Standard 754 compatibility.
When the overflow exception is not masked, the precision exception is flagged on the 32-bit
Intel Architecture FPUs. When the result is stored in the stack, the significand is rounded
according to the precision control (PC) field of the FPU control word or according to the opcode.

18-11
INTEL ARCHITECTURE COMPATIBILITY

On the 16-bit Intel Architecture math coprocessors, the precision exception is not flagged and
the significand is not rounded. The impact on existing software is that if the result is stored on
the stack, a program running on a 32-bit Intel Architecture FPU produces a different result under
overflow conditions than on a 16-bit Intel Architecture math coprocessor. The difference is
apparent only to the exception handler. This difference is for IEEE Standard 754 compatibility.

18.12.6.3. NUMERIC UNDERFLOW EXCEPTION (#U)

When the underflow exception is masked on the 32-bit Intel Architecture FPUs, the underflow
exception is signaled when both the result is tiny and denormalization results in a loss of accu-
racy. When the underflow exception is unmasked and the instruction is supposed to store the
result on the stack, the significand is rounded to the appropriate precision (according to the PC
flag in the FPU control word, for those instructions controlled by PC, otherwise to extended
precision), after adjusting the exponent.
When the underflow exception is masked on the 16-bit Intel Architecture math coprocessors and
rounding is toward 0, the underflow exception flag is raised on a tiny result, regardless of loss
of accuracy. When the underflow exception is not masked and the destination is the stack, the
significand is not rounded, but instead is left as is.
When the underflow exception is masked, this difference has no impact on existing software.
The underflow exception occurs less often when rounding is toward 0.
When the underflow exception not masked. A program running on a 32-bit Intel Architecture
FPU produces a different result during underflow conditions than on a 16-bit Intel Architecture
math coprocessor if the result is stored on the stack. The difference is only in the least significant
bit of the significand and is apparent only to the exception handler.

18.12.6.4. EXCEPTION PRECEDENCE

There is no difference in the precedence of the denormal operand exception on the 32-bit Intel
Architecture FPUs, whether it be masked or not. When the denormal operand exception is not
masked on the 16-bit Intel Architecture math coprocessors, it takes precedence over all
other exceptions. This difference causes no impact on existing software, but some unneeded
normalization of denormalized operands is prevented on the Intel486™ processor and Intel 387
math coprocessor.

18.12.6.5. CS AND EIP FOR FPU EXCEPTIONS

On the Intel 32-bit Intel Architecture FPUs, the values from the CS and EIP registers saved for
floating-point exceptions point to any prefixes that come before the floating-point instruction.
On the 8087 math coprocessor, the saved CS and IP registers points to the floating-point
instruction.

18.12.6.6. FPU ERROR SIGNALS

The floating-point error signals to the P6 family, Pentium®, and Intel486™ processors do not
pass through an interrupt controller; an INT# signal from an Intel 387, Intel 287 or 8087 math

18-12
 INTEL ARCHITECTURE COMPATIBILITY

coprocessors does. If an 8086 processor uses another exception for the 8087 interrupt, both
exception vectors should call the floating-point-error exception handler. Some instructions in a
floating-point-error exception handler may need to be deleted if they use the interrupt controller.
The P6 family, Pentium®, and Intel486™ processors have signals that, with the addition of
external logic, support reporting for emulation of the interrupt mechanism used in many
personal computers.
On the P6 family, Pentium®, and Intel486™ processors, an undefined floating-point opcode will
cause an invalid-opcode exception (#UD, interrupt vector 6). Undefined floating-point opcodes,
like legal floating-point opcodes, cause a device not available exception (#NM, interrupt vector
7) when either the TS or EM flag in control register CR0 is set. The P6 family, Pentium®, and
Intel486™ processors do not check for floating-point error conditions on encountering an unde-
fined floating-point opcode.

18.12.6.7. ASSERTION OF THE FERR# PIN

When using the MS-DOS compatibility mode for handing floating-point exceptions, the FERR#
pin must be connected to an input to an external interrupt controller. An external interrupt is then
generated when the FERR# output drives the input to the interrupt controller and the interrupt
controller in turn drives the INTR pin on the processor. For the P6 family and Intel386™ proces-
sors, an unmasked floating-point exception always causes the FERR# pin to be asserted upon
completion of the instruction that caused the exception. For the Pentium® and Intel486™ proces-
sors, an unmasked floating-point exception may cause the FERR# pin to be asserted either at the
end of the instruction causing the exception or immediately before execution of the next
floating-point instruction. (Note that the next floating-point instruction would not be executed
until the pending unmasked exception has been handled.) Refer to Appendix D in the Intel
Architecture Software Developer’s Manual, Volume 1, for a complete description of the required
mechanism for handling floating-point exceptions using the MS-DOS compatibility mode.

18.12.6.8. INVALID OPERATION EXCEPTION ON DENORMALS

An invalid operation exception is not generated on the 32-bit Intel Architecture FPUs upon
encountering a denormal value when executing a FSQRT, FDIV, or FPREM instruction or upon
conversion to BCD or to integer. The operation proceeds by first normalizing the value. On the
16-bit Intel Architecture math coprocessors, upon encountering this situation, the invalid
operation exception is generated. This difference has no impact on existing software. Software
running on the 32-bit Intel Architecture FPUs continues to execute in cases where the 16-bit
Intel Architecture math coprocessors trap. The reason for this change was to eliminate an excep-
tion from being raised.

18.12.6.9. ALIGNMENT CHECK EXCEPTIONS (#AC)

If alignment checking is enabled, a misaligned data operand on the P6 family, Pentium®, and
Intel486™ processors causes an alignment check exception (#AC) when a program or proce-
dure is running at privilege-level 3, except for the stack portion of the
FSAVE/FNSAVE/FXSAVE and FRSTOR/FXRSTOR instructions.

18-13
INTEL ARCHITECTURE COMPATIBILITY

18.12.6.10. SEGMENT NOT PRESENT EXCEPTION DURING FLDENV

On the Intel486™ processor, when a segment not present exception (#NP) occurs in the middle
of an FLDENV instruction, it can happen that part of the environment is loaded and part not. In
such cases, the FPU control word is left with a value of 007FH. The P6 family and Pentium®
processors ensure the internal state is correct at all times by attempting to read the first and last
bytes of the environment before updating the internal state.

18.12.6.11. DEVICE NOT AVAILABLE EXCEPTION (#NM)

The device-not-available exception (#NM, interrupt 7) will occur in the P6 family, Pentium®,
and Intel486™ processors as described in Section 2.5., “Control Registers” in Chapter 2, System
Architecture Overview, and Section 5.12., “Exception and Interrupt Reference” in Chapter 5,
Interrupt and Exception Handling.

18.12.6.12. COPROCESSOR SEGMENT OVERRUN EXCEPTION

The coprocessor segment overrun exception (interrupt 9) does not occur in the P6 family,
Pentium®, and Intel486™ processors. In situations where the Intel 387 math coprocessor would
cause an interrupt 9, the P6 family, Pentium®, and Intel486™ processors simply abort the
instruction. To avoid undetected segment overruns, it is recommended that the floating-point
save area be placed in the same page as the TSS. This placement will prevent the FPU environ-
ment from being lost if a page fault occurs during the execution of an FLDENV, FRSTOR, or
FXRSTOR instructions while the operating system is performing a task switch.

18.12.6.13. GENERAL PROTECTION EXCEPTION (#GP)

A general-protection exception (#GP, interrupt 13) occurs if the starting address of a floating-
point operand falls outside a segment’s size. An exception handler should be included to report
these programming errors.

18.12.6.14. FLOATING-POINT ERROR EXCEPTION (#MF)

In real mode and protected mode (not including virtual-8086 mode), interrupt vector 16 must
point to the floating-point exception handler. In virtual 8086 mode, the virtual-8086 monitor can
be programmed to accommodate a different location of the interrupt vector for floating-point
exceptions.

18.12.7. Changes to Floating-Point Instructions

This section identifies the differences in floating-point instructions for the various Intel FPU and
math coprocessor architectures, the reason for the differences, and their impact on software.

18-14
 INTEL ARCHITECTURE COMPATIBILITY

18.12.7.1. FDIV, FPREM, AND FSQRT INSTRUCTIONS

The 32-bit Intel Architecture FPUs support operations on denormalized operands and, when
detected, an underflow exception can occur, for compatibility with the IEEE Standard 754. The
16-bit Intel Architecture math coprocessors do not operate on denormalized operands or return
underflow results. Instead, they generate an invalid operation exception when they detect an
underflow condition. An existing underflow exception handler will require change only if it
gives different treatment to different opcodes. Also, it is possible that fewer invalid operation
exceptions will occur.

18.12.7.2. FSCALE INSTRUCTION

With the 32-bit Intel Architecture FPUs, the range of the scaling operand is not restricted. If (0
< | ST(1) < 1), the scaling factor is 0; therefore, ST(0) remains unchanged. If the rounded result
is not exact or if there was a loss of accuracy (masked underflow), the precision exception is
signaled. With the 16-bit Intel Architecture math coprocessors, the range of the scaling operand
is restricted. If (0 < | ST(1) | < 1), the result is undefined and no exception is signaled. The
impact of this difference on exiting software is that different results are delivered on the 32-bit
and 16-bit FPUs and math coprocessors when (0 < | ST(1) | < 1).

18.12.7.3. FPREM1 INSTRUCTION

The 32-bit Intel Architecture FPUs compute a partial remainder according to the IEEE Standard
754. This instruction does not exist on the 16-bit Intel Architecture math coprocessors. The
availability of the FPREM1 instruction has is no impact on existing software.

18.12.7.4. FPREM INSTRUCTION

On the 32-bit Intel Architecture FPUs, the condition code flags C0, C3, C1 in the status word
correctly reflect the three low-order bits of the quotient following execution of the FPREM
instruction. On the 16-bit Intel Architecture math coprocessors, the quotient bits are incorrect
when performing a reduction of (64N + M) when (N ≥ 1) and M is 1 or 2. This difference does
not affect existing software; software that works around the bug should not be affected.

18.12.7.5. FUCOM, FUCOMP, AND FUCOMPP INSTRUCTIONS

When executing the FUCOM, FUCOMP, and FUCOMPP instructions, the 32-bit Intel Archi-
tecture FPUs perform unordered compare according to IEEE Standard 754. These instructions
do not exist on the 16-bit Intel Architecture math coprocessors. The availability of these new
instructions has no impact on existing software.

18.12.7.6. FPTAN INSTRUCTION

On the 32-bit Intel Architecture FPUs, the range of the operand for the FPTAN instruction is
much less restricted (| ST(0) | < 263) than on earlier math coprocessors. The instruction reduces
the operand internally using an internal π/4 constant that is more accurate. The range of the

18-15
INTEL ARCHITECTURE COMPATIBILITY

operand is restricted to (| ST(0) | < π/4) on the 16-bit Intel Architecture math coprocessors; the
operand must be reduced to this range using FPREM. This change has no impact on existing
software.

18.12.7.7. STACK OVERFLOW

On the 32-bit Intel Architecture FPUs, if an FPU stack overflow occurs when the invalid oper-
ation exception is masked, the FPU returns the real, integer, or BCD-integer indefinite value to
the destination operand, depending on the instruction being executed. On the 16-bit Intel Archi-
tecture math coprocessors, the original operand remains unchanged following a stack overflow,
but it is loaded into register ST(1). This difference has no impact on existing software.

18.12.7.8. FSIN, FCOS, AND FSINCOS INSTRUCTIONS

On the 32-bit Intel Architecture FPUs, these instructions perform three common trigonometric
functions. These instructions do not exist on the 16-bit Intel Architecture math coprocessors.
The availability of these instructions has no impact on existing software, but using them
provides a performance upgrade.

18.12.7.9. FPATAN INSTRUCTION

On the 32-bit Intel Architecture FPUs, the range of operands for the FPATAN instruction is unre-
stricted. On the 16-bit Intel Architecture math coprocessors, the absolute value of the operand
in register ST(0) must be smaller than the absolute value of the operand in register ST(1). This
difference has impact on existing software.

18.12.7.10. F2XM1 INSTRUCTION

The 32-bit Intel Architecture FPUs support a wider range of operands (–1 < ST (0) < + 1) for
the F2XM1 instruction. The supported operand range for the 16-bit Intel Architecture math
coprocessors is (0 ≤ ST(0) ≤ 0.5). This difference has no impact on existing software.

18.12.7.11. FLD INSTRUCTION

On the 32-bit Intel Architecture FPUs, when using the FLD instruction to load an extended-real
value, a denormal operand exception is not generated because the instruction is not arithmetic.
The 16-bit Intel Architecture math coprocessors do report a denormal operand exception in this
situation. This difference does not affect existing software.
On the 32-bit Intel Architecture FPUs, loading a denormal value that is in single- or double-real
format causes the value to be converted to extended-real format. Loading a denormal value on
the 16-bit Intel Architecture math coprocessors causes the value to be converted to an unnormal.
If the next instruction is FXTRACT or FXAM, the 32-bit Intel Architecture FPUs will give a
different result than the 16-bit Intel Architecture math coprocessors. This change was made for
IEEE Standard 754 compatibility.

18-16
 INTEL ARCHITECTURE COMPATIBILITY

On the 32-bit Intel Architecture FPUs, loading an SNaN that is in single- or double-real format
causes the FPU to generate an invalid operation exception. The 16-bit Intel Architecture math
coprocessors do not raise an exception when loading a signaling NaN. The invalid operation
exception handler for 16-bit math coprocessor software needs to be updated to handle this condi-
tion when porting software to 32-bit FPUs. This change was made for IEEE Standard 754
compatibility.

18.12.7.12. FXTRACT INSTRUCTION

On the 32-bit Intel Architecture FPUs, if the operand is 0 for the FXTRACT instruction, the
divide-by-zero exception is reported and –∞ is delivered to register ST(1). If the operand is +∞,
no exception is reported. If the operand is 0 on the 16-bit Intel Architecture math coprocessors,
0 is delivered to register ST(1) and no exception is reported. If the operand is +∞, the invalid
operation exception is reported. These differences have no impact on existing software. Soft-
ware usually bypasses 0 and ∞. This change is due to the IEEE 754 recommendation to fully
support the “logb” function.

18.12.7.13. LOAD CONSTANT INSTRUCTIONS

On 32-bit Intel Architecture FPUs, rounding control is in effect for the load constant instruc-
tions. Rounding control is not in effect for the 16-bit Intel Architecture math coprocessors.
Results for the FLDPI, FLDLN2, FLDLG2, and FLDL2E instructions are the same as for the
16-bit Intel Architecture math coprocessors when rounding control is set to round to nearest or
round to +∞. They are the same for the FLDL2T instruction when rounding control is set to
round to nearest, round to –∞, or round to zero. Results are different from the 16-bit Intel Archi-
tecture math coprocessors in the least significant bit of the mantissa if rounding control is set to
round to –∞ or round to 0 for the FLDPI, FLDLN2, FLDLG2, and FLDL2E instructions; they
are different for the FLDL2T instruction if round to +∞ is specified. These changes were imple-
mented for compatibility with IEEE 754 recommendations.

18.12.7.14. FSETPM INSTRUCTION

With the 32-bit Intel Architecture FPUs, the FSETPM instruction is treated as NOP (no opera-
tion). This instruction informs the Intel 287 math coprocessor that the processor is in protected
mode. This change has no impact on existing software. The 32-bit Intel Architecture FPUs
handle all addressing and exception-pointer information, whether in protected mode or not.

18.12.7.15. FXAM INSTRUCTION

With the 32-bit Intel Architecture FPUs, if the FPU encounters an empty register when
executing the FXAM instruction, it will generate combinations of C0 through C3 equal to 1101
or 1111. The 16-bit Intel Architecture math coprocessors may generate these combinations,
among others. This difference has no impact on existing software; it provides a performance
upgrade to provide repeatable results.

18-17
INTEL ARCHITECTURE COMPATIBILITY

18.12.7.16. FSAVE AND FSTENV INSTRUCTIONS

With the 32-bit Intel Architecture FPUs, the address of a memory operand pointer stored by
FSAVE or FSTENV is undefined if the previous floating-point instruction did not refer to
memory

18.12.8. Transcendental Instructions

The floating-point results of the P6 family and Pentium® processors for transcendental instruc-
tions in the core range may differ from the Intel486™ processors by about 2 or 3 ulps (refer to
“Transcendental Instruction Accuracy” in Chapter 7 of the Intel Architecture Software Devel-
oper’s Manual, Volume 1). Condition code flag C1 of the status word may differ as a result. The
exact threshold for underflow and overflow will vary by a few ulps. The P6 family and Pentium ®
processors’ results will have a worst case error of less than 1 ulp when rounding to the nearest-
even and less than 1.5 ulps when rounding in other modes. The transcendental instructions are
guaranteed to be monotonic, with respect to the input operands, throughout the domain
supported by the instruction.
Transcendental instructions may generate different results in the round-up flag (C1) on the
32-bit Intel Architecture FPUs. The round-up flag is undefined for these instructions on the
16-bit Intel Architecture math coprocessors. This difference has no impact on existing software.

18.12.9. Obsolete Instructions

The 8087 math coprocessor instructions FENI and FDISI and the Intel 287 math coprocessor
instruction FSETPM are treated as integer NOP instructions in the 32-bit Intel Architecture
FPUs. If these opcodes are detected in the instruction stream, no specific operation is performed
and no internal states are affected.

18.12.10.WAIT/FWAIT Prefix Differences

On the Intel486™ processor, when a WAIT/FWAIT instruction precedes a floating-point
instruction (one which itself automatically synchronizes with the previous floating-point
instruction), the WAIT/FWAIT instruction is treated as a no-op. Pending floating-point excep-
tions from a previous floating-point instruction are processed not on the WAIT/FWAIT instruc-
tion but on the floating-point instruction following the WAIT/FWAIT instruction. In such a case,
the report of a floating-point exception may appear one instruction later on the Intel486™
processor than on a P6 family or Pentium® FPU, or on Intel 387 math coprocessor.

18.12.11.Operands Split Across Segments and/or Pages

On the P6 family, Pentium®, and Intel486™ processor FPUs, when the first half of an operand
to be written is inside a page or segment and the second half is outside, a memory fault can cause
the first half to be stored but not the second half. In this situation, the Intel 387 math coprocessor
stores nothing.

18-18
 INTEL ARCHITECTURE COMPATIBILITY

18.12.12.FPU Instruction Synchronization

On the 32-bit Intel Architecture FPUs, all floating-point instructions are automatically synchro-
nized; that is, the processor automatically waits until the previous floating-point instruction has
completed before completing the next floating-point instruction. No explicit WAIT/FWAIT
instructions are required to assure this synchronization. For the 8087 math coprocessors, explicit
waits are required before each floating-point instruction to ensure synchronization. Although
8087 programs having explicit WAIT instructions execute perfectly on the 32-bit Intel Architec-
ture processors without reassembly, these WAIT instructions are unnecessary.

18.13. SERIALIZING INSTRUCTIONS

Certain instructions have been defined to serialize instruction execution to ensure that modifi-
cations to flags, registers and memory are completed before the next instruction is executed (or
in P6 family processor terminology “committed to machine state”). Because the P6 family
processors use branch-prediction and out-of-order execution techniques to improve perfor-
mance, instruction execution is not generally serialized until the results of an executed instruc-
tion are committed to machine state (refer to Chapter 2, Introduction to the Intel Architecture,
in the Intel Architecture Software Developer’s Manual, Volume 1). As a result, at places in a
program or task where it is critical to have execution completed for all previous instructions
before executing the next instruction (for example, at a branch, at the end of a procedure, or in
multiprocessor dependent code), it is useful to add a serializing instruction. Refer to Section 7.4.,
“Serializing Instructions” in Chapter 7, Multiple-Processor Management for more information
on serializing instructions.

18.14. FPU AND MATH COPROCESSOR INITIALIZATION

Table 8-1 in Chapter 8, Processor Management and Initialization shows the states of the FPUs
in the P6 family, Pentium®, Intel486™ processors and of the Intel 387 math coprocessor and
Intel 287 coprocessor following a power-up, reset, or INIT, or following the execution of an
FINIT/FNINIT instruction. The following is some additional compatibility information
concerning the initialization of Intel Architecture FPUs and math coprocessors.

18.14.1. Intel 387 and Intel 287 Math Coprocessor Initialization

Following an Intel386™ processor reset, the processor identifies its coprocessor type (Intel 287
or Intel 387 DX math coprocessor) by sampling its ERROR# input some time after the falling
edge of RESET# signal and before execution of the first floating-point instruction. The Intel 287
coprocessor keeps its ERROR# output in inactive state after hardware reset; the Intel 387 copro-
cessor keeps its ERROR# output in active state after hardware reset.
Upon hardware reset or execution of the FINIT/FNINIT instruction, the Intel 387 math copro-
cessor signals an error condition. The P6 family, Pentium®, and Intel486™ processors, like the
Intel 287 coprocessor, do not.

18-19
INTEL ARCHITECTURE COMPATIBILITY

18.14.2. Intel486™ SX Processor and Intel 487 SX Math

Coprocessor Initialization
When initializing an Intel486™ SX processor and an Intel 487 SX math coprocessor, the initial-
ization routine should check the presence of the math coprocessor and should set the FPU related
flags (EM, MP, and NE) in control register CR0 accordingly (refer to Section 2.5., “Control
Registers” in Chapter 2, System Architecture Overview for a complete description of these
flags). Table 18-1 gives the recommended settings for these flags when the math coprocessor is
present. The FSTCW instruction will give a value of FFFFH for the Intel486™ SX micropro-
cessor and 037FH for the Intel 487 SX math coprocessor.

Table 18-1. Recommended Values of the FP Related Bits for Intel486™ SX

Microprocessor/Intel 487 SX Math Coprocessor System
CR0 Flags Intel486™ SX Processor Only Intel 487 SX Math Coprocessor Present
EM 1 0
MP 0 1
NE 1 0, for MS-DOS* systems
1, for user-defined exception handler

The EM and MP flags in register CR0 are interpreted as shown in Table 18-2.

Table 18-2. EM and MP Flag Interpretation

EM MP Interpretation
0 0 Floating-point instructions are passed to FPU; WAIT/FWAIT and
other waiting-type instructions ignore TS.
0 1 Floating-point instructions are passed to FPU; WAIT/FWAIT and
other waiting-type instructions test TS.
1 0 Floating-point instructions trap to emulator; WAIT/FWAIT and
other waiting-type instructions ignore TS.
1 1 Floating-point instructions trap to emulator; WAIT/FWAIT and
other waiting-type instructions test TS.

Following is an example code sequence to initialize the system and check for the presence of
Intel486™ SX processor/Intel 487 SX math coprocessor.
fninit
fstcw mem_loc
mov ax, mem_loc
cmp ax, 037fh
jz Intel487_SX_Math_CoProcessor_present;ax=037fh
jmp Intel486_SX_microprocessor_present;ax=ffffh

If the Intel 487 SX math coprocessor is not present, the following code can be run to set the CR0
register for the Intel486™ SX processor.

18-20
 INTEL ARCHITECTURE COMPATIBILITY

mov eax, cr0

and eax, fffffffdh ;make MP=0
or eax, 0024h ;make EM=1, NE=1
mov cr0, eax

This initialization will cause any floating-point instruction to generate a device not available
exception (#NH), interrupt 7. The software emulation will then take control to execute these
instructions. This code is not required if an Intel 487 SX math coprocessor is present in the
system. In that case, the typical initialization routine for the Intel486™ SX microprocessor will
be adequate.
Also, when designing an Intel486™ SX processor based system with an Intel 487 SX math
coprocessor, timing loops should be independent of clock speed and clocks per instruction. One
way to attain this is to implement these loops in hardware and not in software (for example,
BIOS).

18.15. CONTROL REGISTERS

The following sections identify the new control registers and control register flags and fields that
were introduced to the 32-bit Intel Architecture in various processor families. Refer to Figure
2-5 in Chapter 2, System Architecture Overview for the location of these flags and fields in the
control registers.
The Pentium® III processor introduced one new control flag in control register CR4:
• OSXMMEXCPT (bit 10)—The OS will set this bit if it supports unmasked SIMD floating-
point exceptions.
The Pentium® II processor introduced one new control flag in control register CR4:
• OSFXSR (bit 9)—The OS supports saving and restoring the Pentium® III processor state
during context switches.

The Pentium® Pro processor introduced three new control flags in control register CR4:
• PAE (bit 5)—Physical address extension. Enables paging mechanism to reference 36-bit
physical addresses when set; restricts physical addresses to 32 bits when clear (refer to
Section 18.16.1.1., “Physical Memory Addressing Extension” in Chapter 18, Intel Archi-
tecture Compatibility).
• PGE (bit 7)—Page global enable. Inhibits flushing of frequently-used or shared pages on
task switches (refer to Section 18.16.1.2., “Global Pages” in Chapter 18, Intel Architecture
Compatibility).
• PCE (bit 8)—Performance-monitoring counter enable. Enables execution of the RDPMC
instruction at any protection level.

The content of CR4 is 0H following a hardware reset.

18-21
INTEL ARCHITECTURE COMPATIBILITY

Control register CR4 was introduced in the Pentium® processor. This register contains flags that
enable certain new extensions provided in the Pentium® processor:

• VME—Virtual-8086 mode extensions. Enables support for a virtual interrupt flag in

virtual-8086 mode (refer to Section 16.3., “Interrupt and Exception Handling in Virtual-
8086 Mode” in Chapter 16, 8086 Emulation).
• PVI—Protected-mode virtual interrupts. Enables support for a virtual interrupt flag in
protected mode (refer to Section 16.4., “Protected-Mode Virtual Interrupts” in Chapter 16,
8086 Emulation).
• TSD—Time-stamp disable. Restricts the execution of the RDTSC instruction to
procedures running at privileged level 0.
• DE—Debugging extensions. Causes an undefined opcode (#UD) exception to be
generated when debug registers DR4 and DR5 are references for improved performance
(refer to Section 15.2.2., “Debug Registers DR4 and DR5” in Chapter 15, Debugging and
Performance Monitoring).
• PSE—Page size extensions. Enables 4-MByte pages when set (refer to Section 3.6.1.,
“Paging Options” in Chapter 3, Protected-Mode Memory Management).
• MCE—Machine-check enable. Enables the machine-check exception, allowing exception
handling for certain hardware error conditions (refer to Chapter 13, Machine-Check Archi-
tecture).

The Intel486™ processor introduced five new flags in control register CR0:
• NE—Numeric error. Enables the normal mechanism for reporting floating-point numeric
errors.
• WP—Write protect. Write-protects user-level pages against supervisor-mode accesses.
• AM—Alignment mask. Controls whether alignment checking is performed. Operates in
conjunction with the AC (Alignment Check) flag.
• NW—Not write-through. Enables write-throughs and cache invalidation cycles when clear
and disables invalidation cycles and write-throughs that hit in the cache when set.
• CD—Cache disable. Enables the internal cache when clear and disables the cache when
set.

The Intel486™ processor introduced two new flags in control register CR3:
• PCD—Page-level cache disable. The state of this flag is driven on the PCD# pin during
bus cycles that are not paged, such as interrupt acknowledge cycles, when paging is
enabled. The PCD# pin is used to control caching in an external cache on a cycle-by-cycle
basis.
• PWT—Page-level write-through. The state of this flag is driven on the PWT# pin during
bus cycles that are not paged, such as interrupt acknowledge cycles, when paging is
enabled. The PWT# pin is used to control write through in an external cache on a cycle-by-
cycle basis.

18-22
 INTEL ARCHITECTURE COMPATIBILITY

18.16. MEMORY MANAGEMENT FACILITIES

The following sections describe the new memory management facilities available in the various
Intel Architecture processors and some compatibility differences.

18.16.1. New Memory Management Control Flags

The Pentium® Pro processor introduced three new memory management features: physical
memory addressing extension, the global bit in page-table entries, and general support for larger
page sizes. These features are only available when operating in protected mode.

18.16.1.1. PHYSICAL MEMORY ADDRESSING EXTENSION

The new PAE (physical address extension) flag in control register CR4, bit 5, enables 4 addi-
tional address lines on the processor, allowing 36-bit physical addresses. This option can only
be used when paging is enabled, using a new page-table mechanism provided to support the
larger physical address range (refer to Section 3.8., “Physical Address Extension” in Chapter 3,
Protected-Mode Memory Management).

18.16.1.2. GLOBAL PAGES

The new PGE (page global enable) flag in control register CR4, bit 7, provides a mechanism for
preventing frequently used pages from being flushed from the translation lookaside buffer
(TLB). When this flag is set, frequently used pages (such as pages containing kernel procedures
or common data tables) can be marked global by setting the global flag in a page-directory or
page-table entry. On a task switch or a write to control register CR3 (which normally causes the
TLBs to be flushed), the entries in the TLB marked global are not flushed. Marking pages global
in this manner prevents unnecessary reloading of the TLB due to TLB misses on frequently used
pages. Refer to Section 3.7., “Translation Lookaside Buffers (TLBs)” in Chapter 3, Protected-
Mode Memory Management for a detailed description of this mechanism.

18.16.1.3. LARGER PAGE SIZES

The P6 family processors support large page sizes. This facility is enabled with the PSE (page
size extension) flag in control register CR4, bit 4. When this flag is set, the processor supports
either 4-KByte or 4-MByte page sizes when normal paging is used and 4-KByte and 2-MByte
page sizes when the physical address extension is used. Refer to Section 3.6.1., “Paging
Options” in Chapter 3, Protected-Mode Memory Management for more information about large
page sizes.

18.16.2. CD and NW Cache Control Flags

The CD and NW flags in control register CR0 were introduced in the Intel486™ processor. In
the P6 family and Pentium® processors, these flags are used to implement a writeback strategy
for the data cache; in the Intel486™ processor, they implement a write-through strategy. Refer

18-23
INTEL ARCHITECTURE COMPATIBILITY

to Table 9-4, in Chapter 9, Memory Cache Control for a comparison of these bits on the P6
family, Pentium®, and Intel486™ processors. For complete information on caching, refer to
Chapter 9, Memory Cache Control.

18.16.3. Descriptor Types and Contents

Operating-system code that manages space in descriptor tables often contains an invalid value
in the access-rights field of descriptor-table entries to identify unused entries. Access rights
values of 80H and 00H remain invalid for the P6 family, Pentium®, Intel486™, Intel386™, and
Intel 286 processors. Other values that were invalid on the Intel 286 processor may be valid on
the 32-bit processors because uses for these bits have been defined.

18.16.4. Changes in Segment Descriptor Loads

On the Intel386™ processor, loading a segment descriptor always causes a locked read and write
to set the accessed bit of the descriptor. On the P6 family, Pentium®, and Intel486™ processors,
the locked read and write occur only if the bit is not already set.

18.17. DEBUG FACILITIES

The P6 family and Pentium® processors include extensions to the Intel486™ processor debug-
ging support for breakpoints. To use the new breakpoint features, it is necessary to set the DE
flag in control register CR4.

18.17.1. Differences in Debug Register DR6

It is not possible to write a 1 to reserved bit 12 in debug status register DR6 on the P6 family
and Pentium® processors; however, it is possible to write a 1 in this bit on the Intel486™
processor. Refer to Table 8-1 in Chapter 8, Processor Management and Initialization for the
different setting of this register following a power-up or hardware reset.

18.17.2. Differences in Debug Register DR7

The P6 family and Pentium® processors determines the type of breakpoint access by the R/W0
through R/W3 fields in debug control register DR7 as follows:
00 Break on instruction execution only.
01 Break on data writes only.
10 Undefined if the DE flag in control register CR4 is cleared; break on I/O reads or writes
but not instruction fetches if the DE flag in control register CR4 is set.
11 Break on data reads or writes but not instruction fetches.

18-24
 INTEL ARCHITECTURE COMPATIBILITY

On the P6 family and Pentium® processors, reserved bits 11, 12, 14 and 15 are hard-wired to 0.
On the Intel486™ processor, however, bit 12 can be set. Refer to Table 8-1 in Chapter 8,
Processor Management and Initialization for the different settings of this register following a
power-up or hardware reset.

18.17.3. Debug Registers DR4 and DR5

Although the DR4 and DR5 registers are documented as reserved, previous generations of
processors aliased references to these registers to debug registers DR6 and DR7, respectively.
When debug extensions are not enabled (the DE flag in control register CR4 is cleared), the P6
family and Pentium® processors remain compatible with existing software by allowing these
aliased references. When debug extensions are enabled (the DE flag is set), attempts to reference
registers DR4 or DR5 will result in an invalid-opcode exception (#UD).

18.17.4. Recognition of Breakpoints

For the Pentium® processor, it is recommended that debuggers execute the LGDT instruction
before returning to the program being debugged to ensure that breakpoints are detected. This
operation does not need to be performed on the P6 family, Intel486™, or Intel386™ processors.

18.18. TEST REGISTERS

The implementation of test registers on the Intel486™ processor used for testing the cache and
TLB has been redesigned using MSRs on the P6 family and Pentium® processors. (Note that
MSRs used for this function are different on the P6 family and Pentium® processors.) The MOV
to and from test register instructions generate invalid-opcode exceptions (#UD) on the P6 family
processors.

18.19. Exceptions and/or Exception Conditions

This section describes the new exceptions and exception conditions added to the 32-bit Intel
Architecture processors and implementation differences in existing exception handling. Refer to
Chapter 5, Interrupt and Exception Handling for a detailed description of the Intel Architecture
exceptions.
The Pentium® III processor introduced new state with the SIMD floating-point registers.
Computations involving data in these registers can produce exceptions. A new control/status
register is used to determine which exception or exceptions have occurred. When an exception
associated with the SIMD floating-point registers occurs, an interrupt is generated.
• Streaming SIMD Extensions exception (#XF, interrupt 19)—New exceptions associated
with the SIMD floating-point registers and resulting computations.

18-25
INTEL ARCHITECTURE COMPATIBILITY

No new exceptions were added to the Pentium® II and Pentium® Pro processors. The set of avail-
able exceptions is the same as for the Pentium® processor. However, the following exception
condition was added to the Intel Architecture with the Pentium® Pro processor:
• Machine-check exception (#MC, interrupt 18)—New exception conditions. Many
exception conditions have been added to the machine-check exception and a new archi-
tecture has been added for handling and reporting on hardware errors. Refer to Chapter 13,
Machine-Check Architecture for a detailed description of the new conditions.

The following exceptions and/or exception conditions were added to the Intel Architecture with
the Pentium® processor:
• Machine-check exception (#MC, interrupt 18)—New exception. This exception reports
parity and other hardware errors. It is a model-specific exception and may not be
implemented or implemented differently in future processors. The MCE flag in control
register CR4 enables the machine-check exception. When this bit is clear (which it is at
reset), the processor inhibits generation of the machine-check exception.
• General-protection exception (#GP, interrupt 13)—New exception condition added. An
attempt to write a 1 to a reserved bit position of a special register causes a general-
protection exception to be generated.
• Page-fault exception (#PF, interrupt 14)—New exception condition added. When a 1 is
detected in any of the reserved bit positions of a page-table entry, page-directory entry, or
page-directory pointer during address translation, a page-fault exception is generated.

The following exception was added to the Intel486™ processor:

• Alignment-check exception (#AC, interrupt 17)—New exception. Reports unaligned
memory references when alignment checking is being performed.

The following exceptions and/or exception conditions were added to the Intel386™ processor:
• Divide-error exception (#DE, interrupt 0)
— Change in exception handling. Divide-error exceptions on the Intel386™ processors
always leave the saved CS:IP value pointing to the instruction that failed. On the 8086
processor, the CS:IP value points to the next instruction.
— Change in exception handling. The Intel386™ processors can generate the largest
negative number as a quotient for the IDIV instruction (80H and 8000H). The 8086
processor generates a divide-error exception instead.
• Invalid-opcode exception (#UD, interrupt 6)—New exception condition added. Improper
use of the LOCK instruction prefix can generate an invalid-opcode exception.
• Page-fault exception (#PF, interrupt 14)—New exception condition added. If paging is
enabled in a 16-bit program, a page-fault exception can be generated as follows. Paging
can be used in a system with 16-bit tasks if all tasks use the same page directory. Because
there is no place in a 16-bit TSS to store the PDBR register, switching to a 16-bit task does
not change the value of the PDBR register. Tasks ported from the Intel 286 processor
should be given 32-bit TSSs so they can make full use of paging.

18-26
 INTEL ARCHITECTURE COMPATIBILITY

• General-protection exception (#GP, interrupt 13)—New exception condition added. The

Intel386™ processor sets a limit of 15 bytes on instruction length. The only way to violate
this limit is by putting redundant prefixes before an instruction. A general-protection
exception is generated if the limit on instruction length is violated. The 8086 processor has
no instruction length limit.

18.19.1. Machine-Check Architecture

The Pentium® Pro processor introduced a new architecture to the Intel Architecture for handling
and reporting on machine-check exceptions. This machine-check architecture (described in
detail in Chapter 13, Machine-Check Architecture) greatly expands the ability of the processor
to report on internal hardware errors.

18.19.2. Priority OF Exceptions

The priority of exceptions are broken down into several major categories:
1. Traps on the previous instruction
2. External interrupts
3. Faults on fetching the next instruction
4. Faults in decoding the next instruction
5. Faults on executing an instruction
There are no changes in the priority of these major categories between the different processors,
however, exceptions within these categories are implementation dependent and may change
from processor to processor.

18.20. INTERRUPTS
The following differences in handling interrupts are found among the Intel Architecture
processors.

18.20.1. Interrupt Propagation Delay

External hardware interrupts may be recognized on different instruction boundaries on the P6
family, Pentium®, Intel486™, and Intel386™ processors, due to the superscaler designs of the
P6 family and Pentium® processors. Therefore, the EIP pushed onto the stack when servicing an
interrupt may be different for the P6 family, Pentium®, Intel486™, and Intel386™ processors.

18-27
INTEL ARCHITECTURE COMPATIBILITY

18.20.2. NMI Interrupts

After an NMI interrupt is recognized by the P6 family, Pentium®, Intel486™, Intel386™, and
Intel 286 processors, the NMI interrupt is masked until the first IRET instruction is executed,
unlike the 8086 processor.

18.20.3. IDT Limit

The LIDT instruction can be used to set a limit on the size of the IDT. A double-fault exception
(#DF) is generated if an interrupt or exception attempts to read a vector beyond the limit. Shut-
down then occurs on the 32-bit Intel Architecture processors if the double-fault handler vector
is beyond the limit. (The 8086 processor does not have a shutdown mode nor a limit.)

18.21. TASK SWITCHING AND TSS

This section identifies the implementation differences of task switching, additions to the TSS
and the handling of TSSs and TSS segment selectors.

18.21.1. P6 Family and Pentium® Processor TSS

When the virtual mode extensions are enabled (by setting the VME flag in control register CR4),
the TSS in the P6 family and Pentium® processors contain an interrupt redirection bit map,
which is used in virtual-8086 mode to redirect interrupts back to an 8086 program.

18.21.2. TSS Selector Writes

During task state saves, the Intel486™ processor writes 2-byte segment selectors into a 32-bit
TSS, leaving the upper 16 bits undefined. For performance reasons, the P6 family and Pentium®
processors write 4-byte segment selectors into the TSS, with the upper 2 bytes being 0. For
compatibility reasons, code should not depend on the value of the upper 16 bits of the selector
in the TSS.

18.21.3. Order of Reads/Writes to the TSS

The order of reads and writes into the TSS is processor dependent. The P6 family and Pentium®
processors may generate different page-fault addresses in control register CR2 in the same TSS
area than the Intel486™ and Intel386™ processors, if a TSS crosses a page boundary (which is
not recommended).

18-28
 INTEL ARCHITECTURE COMPATIBILITY

18.21.4. Using A 16-Bit TSS with 32-Bit Constructs

Task switches using 16-bit TSSs should be used only for pure 16-bit code. Any new code written
using 32-bit constructs (operands, addressing, or the upper word of the EFLAGS register)
should use only 32-bit TSSs. This is due to the fact that the 32-bit processors do not save the
upper 16 bits of EFLAGS to a 16-bit TSS. A task switch back to a 16-bit task that was executing
in virtual mode will never re-enable the virtual mode, as this flag was not saved in the upper half
of the EFLAGS value in the TSS. Therefore, it is strongly recommended that any code using
32-bit constructs use a 32-bit TSS to ensure correct behavior in a multitasking environment.

18.21.5. Differences in I/O Map Base Addresses

The Intel486™ processor considers the TSS segment to be a 16-bit segment and wraps around
the 64K boundary. Any I/O accesses check for permission to access this I/O address at the I/O
base address plus the I/O offset. If the I/O map base address exceeds the specified limit of
0DFFFH, an I/O access will wrap around and obtain the permission for the I/O address at an
incorrect location within the TSS. A TSS limit violation does not occur in this situation on the
Intel486™ processor. However, the P6 family and Pentium® processors consider the TSS to be
a 32-bit segment and a limit violation occurs when the I/O base address plus the I/O offset is
greater than the TSS limit. By following the recommended specification for the I/O base address
to be less than 0DFFFH, the Intel486™ processor will not wrap around and access incorrect
locations within the TSS for I/O port validation and the P6 family and Pentium® processors will
not experience general-protection exceptions (#GP). Figure 18-1 demonstrates the different
areas accessed by the Intel486™ and the P6 family and Pentium® processors.

18-29
INTEL ARCHITECTURE COMPATIBILITY

Intel486™ Processor P6 family and Pentium® Processors

FFFFH + 10H = Outside Segment
for I/O Validation

FFFFH FFFFH

I/O Map I/O Map

Base Addres FFFFH Base Addres FFFFH

FFFFH + 10H = FH
for I/O Validation

0H 0H

I/O access at port 10H checks I/O access at port 10H checks
bitmap at I/O map base address bitmap at I/O address FFFFH + 10H,
FFFFH + 10H = offset 10H. which exceeds segment limit.
Offset FH from beginning of Wrap around does not occur,
TSS segment results because general-protection exception (#GP)
wraparound occurs. occurs.

Figure 18-1. I/O Map Base Address Differences

18.22. CACHE MANAGEMENT

The P6 family processors include two levels of internal caches: L1 (level 1) and L2 (level 2).
The L1 cache is divided into an instruction cache and a data cache; the L2 cache is a general-
purpose cache. Refer to Section 9.1., “Internal Caches, TLBs, and Buffers”, in Chapter 9,
Memory Cache Control, for a description of these caches. (Note that although the Pentium® II
processor L2 cache is physically located on a separate chip in the cassette, it is considered an
internal cache.)
The Pentium® processor includes separate level 1 instruction and data caches. The data cache
supports a writeback (or alternatively write-through, on a line by line basis) policy for memory
updates. Refer to the Pentium® Processor Data Book for more information about the organiza-
tion and operation of the Pentium® processor caches.
The Intel486™ processor includes a single level 1 cache for both instructions and data.
The meaning of the CD and NW flags in control register CR0 have been redefined for the P6
family and Pentium® processors. For these processors, the recommended value (00B) enables
writeback for the data cache of the Pentium® processor and for the L1 data cache and L2 cache
of the P6 family processors. In the Intel486™ processor, setting these flags to (00B) enables
write-through for the cache.
External system hardware can force the Pentium® processor to disable caching or to use the
write-through cache policy should that be required. Refer to the Pentium® Processor Data Book

18-30
 INTEL ARCHITECTURE COMPATIBILITY

for more information about hardware control of the Pentium® processor caches. In the P6 family
processors, the MTRRs can be used to override the CD and NW flags (refer to Table 9-6, in
Chapter 9, Memory Cache Control).
The P6 family and Pentium® processors support page-level cache management in the same
manner as the Intel486™ processor by using the PCD and PWT flags in control register CR3,
the page-directory entries, and the page-table entries. The Intel486™ processor, however, is not
affected by the state of the PWT flag since the internal cache of the Intel486™ processor is a
write-through cache.

18.22.1. Self-Modifying Code with Cache Enabled

On the Intel486™ processor, a write to an instruction in the cache will modify it in both the
cache and memory. If the instruction was prefetched before the write, however, the old version
of the instruction could be the one executed. To prevent this problem, it is necessary to flush the
instruction prefetch unit of the Intel486™ processor by coding a jump instruction immediately
after any write that modifies an instruction. The P6 family and Pentium® processors, however,
check whether a write may modify an instruction that has been prefetched for execution. This
check is based on the linear address of the instruction. If the linear address of an instruction is
found to be present in the prefetch queue, the P6 family and Pentium® processors flush the
prefetch queue, eliminating the need to code a jump instruction after any writes that modify an
instruction.
Because the linear address of the write is checked against the linear address of the instructions
that have been prefetched, special care must be taken for self-modifying code to work correctly
when the physical addresses of the instruction and the written data are the same, but the linear
addresses differ. In such cases, it is necessary to execute a serializing operation to flush the
prefetch queue after the write and before executing the modified instruction. Refer to Section
7.4., “Serializing Instructions” in Chapter 7, Multiple-Processor Management for more infor-
mation on serializing instructions.

NOTE
The check on linear addresses described above is not in practice a concern for
compatibility. Applications that include self-modifying code use the same
linear address for modifying and fetching the instruction. System software,
such as a debugger, that might possibly modify an instruction using a
different linear address than that used to fetch the instruction must execute a
serializing operation, such as IRET, before the modified instruction is
executed.

18.23. PAGING
This section identifies enhancements made to the paging mechanism and implementation differ-
ences in the paging mechanism for various Intel Architecture processors.

18-31
INTEL ARCHITECTURE COMPATIBILITY

18.23.1. Large Pages

The Pentium® processor extended the memory management/paging facilities of the Intel Archi-
tecture to allow large (4Mbytes) pages sizes (refer to Section 3.6.1., “Paging Options” in
Chapter 3, Protected-Mode Memory Management). The initial P6 family processor (the
Pentium® Pro processor) added a 2MByte page size to the Intel Architecture in conjunction with
the physical address extension (PAE) feature (refer to Section 3.8., “Physical Address Exten-
sion” in Chapter 3, Protected-Mode Memory Management).
The availability of large pages on any Intel Architecture processor can be determined via feature
bit 3 (PSE) of register EDX after the CPUID instruction has been execution with an argument
of 1. Intel processors that do not support the CPUID instruction do not support page size
enhancements. (Refer to “CPUID—CPU Identification” in Chapter 3, Instruction Set Reference,
of the Intel Architecture Software Developer’s Manual, Volume 2, and AP-485, Intel Processor
Identification and the CPUID Instruction, for more information on the CPUID instruction.)

18.23.2. PCD and PWT Flags

The PCD and PWT flags were introduced to the Intel Architecture in the Intel486™ processor
to control the caching of pages:

• PCD (page-level cache disable) flag—Controls caching on a page-by-page basis.

• PWT (page-level write-through) flag—Controls the write-through/writeback caching
policy on a page-by-page basis. Since the internal cache of the Intel486™ processor is a
write-through cache, it is not affected by the state of the PWT flag.

18.23.3. Enabling and Disabling Paging

Paging is enabled and disabled by loading a value into control register CR0 that modifies the PG
flag. For backward and forward compatibility with all Intel Architecture processors, Intel
recommends that the following operations be performed when enabling or disabling paging:

1. Execute a MOV CR0, REG instruction to either set (enable paging) or clear (disable
paging) the PG flag.
2. Execute a near JMP instruction.
The sequence bounded by the MOV and JMP instructions should be identity mapped (that is,
the instructions should reside on a page whose linear and physical addresses are identical).
For the P6 family processors, the MOV CR0, REG instruction is serializing, so the jump oper-
ation is not required. However, for backwards compatibility, the JMP instruction should still be
included.

18-32
 INTEL ARCHITECTURE COMPATIBILITY

18.24. STACK OPERATIONS

This section identifies the differences in the stack mechanism for the various Intel Architecture
processors.

18.24.1. Selector Pushes and Pops

When pushing a segment selector onto the stack, the Intel486™ processor writes 2 bytes onto
4-byte stacks and decrements ESP by 4. The P6 family and Pentium® processors write 4 bytes,
with the upper 2 bytes being zeros.
When popping a segment selector from the stack, the Intel486™ processor reads only 2 bytes.
The P6 family and Pentium® processors read 4 bytes and discard the upper 2 bytes. This opera-
tion may have an effect if the ESP is close to the stack-segment limit. On the P6 family and
Pentium® processors, stack location at ESP plus 4 may be above the stack limit, in which case a
stack fault exception (#SS) will be generated. On the Intel486™ processor, stack location at ESP
plus 2 may be less than the stack limit and no exception is generated.
For a POP-to-memory instruction that meets the following conditions:
• The stack segment size is 16-bit
• Any 32-bit addressing form with the SIB byte specifying ESP as the base register
• The initial stack pointer is FFFCh (32-bit operand) or FFFEh (16-bit operand) and will
wrap around to 0h as a result of the POP operation
the result of the memory write is specific to the processor-family. For example, in Pentium® II
and Pentium® Pro processors, the result of the memory write is SS:0h plus any scaled index and
displacement. In Pentium® and Pentium® Pro processors, the result of the memory write may be
either a stack fault (real mode or protected mode with stack segment size of 64Kbyte), or write
to SS:10000h plus any scaled index and displacement (protected mode and stack segment size
exceeds 64Kbyte).

18.24.2. Error Code Pushes

The Intel486™ processor implements the error code pushed on the stack as a 16-bit value. When
pushed onto a 32-bit stack, the Intel486™ processor only pushes 2 bytes and updates ESP by 4.
The P6 family and Pentium® processors’ error code is a full 32 bits with the upper 16 bits set to
zero. The P6 family and Pentium® processors, therefore, push 4 bytes and update ESP by 4. Any
code that relies on the state of the upper 16 bits may produce inconsistent results.

18.24.3. Fault Handling Effects on the Stack

During the handling of certain instructions, such as CALL and PUSHA, faults may occur in
different sequences for the different processors. For example, during far calls, the Intel486™
processor pushes the old CS and EIP before a possible branch fault is resolved. A branch fault

18-33
INTEL ARCHITECTURE COMPATIBILITY

is a fault from a branch instruction occurring from a segment limit or access rights violation. If
a branch fault is taken, the Intel486™ and P6 family processors will have corrupted memory
below the stack pointer. However, the ESP register is backed up to make the instruction restart-
able. The P6 family processors issue the branch before the pushes. Therefore, if a branch fault
does occur, these processors do not corrupt memory below the stack pointer. This implementa-
tion difference, however, does not constitute a compatibility problem, as only values at or above
the stack pointer are considered to be valid.

18.24.4. Interlevel RET/IRET From a 16-Bit Interrupt or Call Gate

If a call or interrupt is made from a 32-bit stack environment through a 16-bit gate, only 16 bits
of the old ESP can be pushed onto the stack. On the subsequent RET/IRET, the 16-bit ESP is
popped but the full 32-bit ESP is updated since control is being resumed in a 32-bit stack envi-
ronment. The Intel486™ processor writes the SS selector into the upper 16 bits of ESP. The P6
family and Pentium® processors write zeros into the upper 16 bits.

18.25. MIXING 16- AND 32-BIT SEGMENTS

The features of the 16-bit Intel 286 processor are an object-code compatible subset of those of
the 32-bit Intel Architecture processors. The D (default operation size) flag in segment descrip-
tors indicates whether the processor treats a code or data segment as a 16-bit or 32-bit segment;
the B(default stack size) flag in segment descriptors indicates whether the processor treats a
stack segment as a 16-bit or 32-bit segment.
The segment descriptors used by the Intel 286 processor are supported by the 32-bit Intel Archi-
tecture processors if the Intel-reserved word (highest word) of the descriptor is clear. On the
32-bit Intel Architecture processors, this word includes the upper bits of the base address and
the segment limit.
The segment descriptors for data segments, code segments, local descriptor tables (there are no
descriptors for global descriptor tables), and task gates are the same for the 16- and 32-bit
processors. Other 16-bit descriptors (TSS segment, call gate, interrupt gate, and trap gate) are
supported by the 32-bit processors. The 32-bit processors also have descriptors for TSS
segments, call gates, interrupt gates, and trap gates that support the 32-bit architecture. Both
kinds of descriptors can be used in the same system.
For those segment descriptors common to both 16- and 32-bit processors, clear bits in the
reserved word cause the 32-bit processors to interpret these descriptors exactly as an Intel 286
processor does, that is:

• Base Address—The upper 8 bits of the 32-bit base address are clear, which limits base
addresses to 24 bits.
• Limit—The upper 4 bits of the limit field are clear, restricting the value of the limit field to
64 Kbytes.
• Granularity bit—The G (granularity) flag is clear, indicating the value of the 16-bit limit is
interpreted in units of 1 byte.

18-34
 INTEL ARCHITECTURE COMPATIBILITY

• Big bit—In a data-segment descriptor, the B flag is clear in the segment descriptor used by
the 32-bit processors, indicating the segment is no larger than 64 Kbytes.
• Default bit—In a code-segment descriptor, the D flag is clear, indicating 16-bit addressing
and operands are the default. In a stack-segment descriptor, the D flag is clear, indicating
use of the SP register (instead of the ESP register) and a 64-Kbyte maximum segment
limit.
For information on mixing 16- and 32-bit code in applications, refer to Chapter 17, Mixing 16-
Bit and 32-Bit Code.

18.26. SEGMENT AND ADDRESS WRAPAROUND

This section discusses differences in segment and address wraparound between the P6 family,
Pentium®, Intel486™, Intel386™, Intel 286, and 8086 processors.

18.26.1. Segment Wraparound

On the 8086 processor, an attempt to access a memory operand that crosses offset 65,535 or
0FFFFH or offset 0 (for example, moving a word to offset 65,535 or pushing a word when the
stack pointer is set to 1) causes the offset to wrap around modulo 65,536 or 010000H. With the
Intel 286 processor, any base and offset combination that addresses beyond 16 MBytes wraps
around to the 1 MByte of the address space. The P6 family, Pentium®, Intel486™, and
Intel386™ processors in real-address mode generate an exception in these cases:
• A general-protection exception (#GP) if the segment is a data segment (that is, if the CS,
DS, ES, FS, or GS register is being used to address the segment).
• A stack-fault exception (#SS) if the segment is a stack segment (that is, if the SS register is
being used).
An exception to this behavior occurs when a stack access is data aligned, and the stack pointer
is pointing to the last aligned piece of data that size at the top of the stack (ESP is FFFFFFFCH).
When this data is popped, no segment limit violation occurs and the stack pointer will wrap
around to 0.
The address space of the P6 family, Pentium®, and Intel486™ processors may wraparound at 1
MByte in real-address mode. An external A20M# pin forces wraparound if enabled. On Intel
8086 processors, it is possible to specify addresses greater than 1 MByte. For example, with a
selector value FFFFH and an offset of FFFFH, the effective address would be 10FFEFH (1
MByte plus 65519 bytes). The 8086 processor, which can form addresses up to 20 bits long,
truncates the uppermost bit, which “wraps” this address to FFEFH. However, the P6 family,
Pentium®, and Intel486™ processors do not truncate this bit if A20M# is not enabled.
If a stack operation wraps around the address limit, shutdown occurs. (The 8086 processor does
not have a shutdown mode nor a limit.)

18-35
INTEL ARCHITECTURE COMPATIBILITY

18.27. WRITE BUFFERS AND MEMORY ORDERING

The Pentium® Pro and Pentium® II processors provide a write buffer for temporary storage of
writes (stores) to memory (refer to Section 9.11., “Write Buffer”, in Chapter 9, Memory Cache
Control). The Pentium® III processor has 4 write buffers. Writes stored in the write buffer(s) are
always written to memory in program order, with the exception of “fast string” store operations
(refer to Section 7.2.3., “Out of Order Stores From String Operations in P6 Family Processors”
in Chapter 7, Multiple-Processor Management).
The Pentium® processor has two write buffers, one corresponding to each of the pipelines.
Writes in these buffers are always written to memory in the order they were generated by the
processor core.
It should be noted that only memory writes are buffered and I/O writes are not. The P6 family,
Pentium®, and Intel486™ processors do not synchronize the completion of memory writes on
the bus and instruction execution after a write. An I/O, locked, or serializing instruction needs
to be executed to synchronize writes with the next instruction (refer to Section 7.4., “Serializing
Instructions” in Chapter 7, Multiple-Processor Management).
The P6 family processors use processor ordering to maintain consistency in the order that data
is read (loaded) and written (stored) in a program and the order the processor actually carries out
the reads and writes. With this type of ordering, reads can be carried out speculatively and in any
order, reads can pass buffered writes, and writes to memory are always carried out in program
order. (Refer to Section 7.2., “Memory Ordering” in Chapter 7, Multiple-Processor Manage-
ment for more information about processor ordering.) The Pentium ® III processor introduced a
new instruction to serialize writes and make them globally visible. Memory ordering issues can
arise between a producer and a consumer of data. The SFENCE instruction provides a perfor-
mance-efficient way of ensuring ordering between routines that produce weakly-ordered results
and routines that consume this data.
No re-ordering of reads occurs on the Pentium® processor, except under the condition noted in
Section 7.2.1., “Memory Ordering in the Pentium® and Intel486™ Processors” in Chapter 7,
Multiple-Processor Management, and in the following paragraph describing the Intel486™
processor. Specifically, the write buffers are flushed before the IN instruction is executed. No
reads (as a result of cache miss) are reordered around previously generated writes sitting in the
write buffers. The implication of this is that the write buffers will be flushed or emptied before
a subsequent bus cycle is run on the external bus.
On both the Intel486™ and Pentium® processors, under certain conditions, a memory read will
go onto the external bus before the pending memory writes in the buffer even though the writes
occurred earlier in the program execution. A memory read will only be reordered in front of all
writes pending in the buffers if all writes pending in the buffers are cache hits and the read is a
cache miss. Under these conditions, the Intel486™ and Pentium® processors will not read from
an external memory location that needs to be updated by one of the pending writes.
During a locked bus cycle, the Intel486™ processor will always access external memory, it will
never look for the location in the on-chip cache. All data pending in the Intel486™ processor's
write buffers will be written to memory before a locked cycle is allowed to proceed to the
external bus. Thus, the locked bus cycle can be used for eliminating the possibility of reordering
read cycles on the Intel486™ processor. The Pentium® processor does check its cache on a read-

18-36
 INTEL ARCHITECTURE COMPATIBILITY

modify-write access and, if the cache line has been modified, writes the contents back to
memory before locking the bus. The P6 family processors write to their cache on a read-modify-
write operation (if the access does not split across a cache line) and does not write back to system
memory. If the access does split across a cache line, it locks the bus and accesses system
memory.
I/O reads are never reordered in front of buffered memory writes on an Intel Architecture
processor. This ensures an update of all memory locations before reading the status from an I/O
device.

18.28. BUS LOCKING

The Intel 286 processor performs the bus locking differently than the Intel P6 family, Pentium®,
Intel486™, and Intel386™ processors. Programs that use forms of memory locking specific to
the Intel 286 processor may not run properly when run on later processors.
A locked instruction is guaranteed to lock only the area of memory defined by the destination
operand, but may lock a larger memory area. For example, typical 8086 and Intel 286 configu-
rations lock the entire physical memory space. Programmers should not depend on this.
On the Intel 286 processor, the LOCK prefix is sensitive to IOPL. If the CPL is greater than the
IOPL, a general-protection exception (#GP) is generated. On the Intel386™ DX, Intel486™,
and Pentium®, and P6 family processors, no check against IOPL is performed.
The Pentium® processor automatically asserts the LOCK# signal when acknowledging external
interrupts. After signaling an interrupt request, an external interrupt controller may use the data
bus to send the interrupt vector to the processor. After receiving the interrupt request signal, the
processor asserts LOCK# to insure that no other data appears on the data bus until the interrupt
vector is received. This bus locking does not occur on the P6 family processors.

18.29. BUS HOLD

Unlike the 8086 and Intel 286 processors, but like the Intel386™ and Intel486™ processors, the
P6 family and Pentium® processors respond to requests for control of the bus from other poten-
tial bus masters, such as DMA controllers, between transfers of parts of an unaligned operand,
such as two words which form a doubleword. Unlike the Intel386™ processor, the P6 family,
Pentium® and Intel486™ processors respond to bus hold during reset initialization.

18.30. TWO WAYS TO RUN INTEL 286 PROCESSOR TASKS

When porting 16-bit programs to run on 32-bit Intel Architecture processors, there are two
approaches to consider:
• Porting an entire 16-bit software system to a 32-bit processor, complete with the old
operating system, loader, and system builder. Here, all tasks will have 16-bit TSSs. The 32-
bit processor is being used as if it were a faster version of the 16-bit processor.

18-37
INTEL ARCHITECTURE COMPATIBILITY

• Porting selected 16-bit applications to run in a 32-bit processor environment with a 32-bit
operating system, loader, and system builder. Here, the TSSs used to represent 286 tasks
should be changed to 32-bit TSSs. It is possible to mix 16 and 32-bit TSSs, but the benefits
are small and the problems are great. All tasks in a 32-bit software system should have 32-
bit TSSs. It is not necessary to change the 16-bit object modules themselves; TSSs are
usually constructed by the operating system, by the loader, or by the system builder. Refer
to Chapter 17, Mixing 16-Bit and 32-Bit Code for more detailed information about mixing
16-bit and 32-bit code.
Because the 32-bit processors use the contents of the reserved word of 16-bit segment descrip-
tors, 16-bit programs that place values in this word may not run correctly on the 32-bit
processors.

18.31. MODEL-SPECIFIC EXTENSIONS TO THE INTEL

ARCHITECTURE
Certain extensions to the Intel Architecture are specific to a processor or family of Intel Archi-
tecture processors and may not be implemented or implemented in the same way in future
processors. The following sections describe these model-specific extensions. The CPUID
instruction indicates the availability of some of the model-specific features.

18.31.1. Model-Specific Registers

The Pentium® processor introduced a set of model-specific registers (MSRs) for use in control-
ling hardware functions and performance monitoring. To access these MSRs, two new instruc-
tions were added to the Intel Architecture: read MSR (RDMSR) and write MSR (WRMSR). The
MSRs in the Pentium® processor are not guaranteed to be duplicated or provided in the next
generation Intel Architecture processors.
The P6 family processors greatly increased the number of MSRs available to software. Refer to
Appendix B, Model-Specific Registers for a complete list of the available MSRs. The new regis-
ters control the debug extensions, the performance counters, the machine-check exception capa-
bility, the machine-check architecture, and the MTRRs. These registers are accessible using the
RDMSR and WRMSR instructions. Specific information on some of these new MSRs is
provided in the following sections. As with the Pentium® processor MSR, the P6 family
processor MSRs are not guaranteed to be duplicated or provided in the next generation Intel
Architecture processors.

18.31.2. RDMSR and WRMSR Instructions

The RDMSR (read model-specific register) and WRMSR (write model-specific register)
instructions recognize a much larger number of model-specific registers in the P6 family proces-
sors. (Refer to “RDMSR—Read from Model Specific Register” and “WRMSR—Write to
Model Specific Register” in Chapter 3 of the Intel Architecture Software Developer’s Manual,
Volume 2, for more information about these instructions.)

18-38
 INTEL ARCHITECTURE COMPATIBILITY

18.31.3. Memory Type Range Registers

Memory type range registers (MTRRs) are a new feature introduced into the Intel Architecture
in the Pentium® Pro processor. MTRRs allow the processor to optimize memory operations for
different types of memory, such as RAM, ROM, frame buffer memory, and memory-mapped
I/O.
MTRRs are MSRs that contain an internal map of how physical address ranges are mapped to
various types of memory. The processor uses this internal memory map to determine the cache-
ability of various physical memory locations and the optimal method of accessing memory loca-
tions. For example, if a memory location is specified in an MTRR as write-through memory, the
processor handles accesses to this location as follows. It reads data from that location in lines
and caches the read data or maps all writes to that location to the bus and updates the cache to
maintain cache coherency. In mapping the physical address space with MTRRs, the processor
recognizes five types of memory: uncacheable (UC), uncacheable, speculatable, write-
combining (USWC), write-through (WT), write-protected (WP), and writeback (WB).
Earlier Intel Architecture processors (such as the Intel486™ and Pentium® processors) used the
KEN# (cache enable) pin and external logic to maintain an external memory map and signal
cacheable accesses to the processor. The MTRR mechanism simplifies hardware designs by
eliminating the KEN# pin and the external logic required to drive it.
Refer to Chapter 8, Processor Management and Initialization and Appendix B, Model-Specific
Registers for more information on the MTRRs.

18.31.4. Machine-Check Exception and Architecture

The Pentium® processor introduced a new exception called the machine-check exception (#MC,
interrupt 18). This exception is used to detect hardware-related errors, such as a parity error on
a read cycle.
The P6 family processors extend the types of errors that can be detected and that generate a
machine-check exception. It also provides a new machine-check architecture for recording
information about a machine-check error and provides extended recovery capability.
The machine-check architecture provides several banks of reporting registers for recording
machine-check errors. Each bank of registers is associated with a specific hardware unit in the
processor. The primary focus of the machine checks is on bus and interconnect operations;
however, checks are also made of translation lookaside buffer (TLB) and cache operations.
The machine-check architecture can correct some errors automatically and allow for reliable
restart of instruction execution. It also collects sufficient information for software to use in
correcting other machine errors not corrected by hardware.
Refer to Chapter 13, Machine-Check Architecture for more information on the machine-check
exception and the machine-check architecture.

18-39
INTEL ARCHITECTURE COMPATIBILITY

18.31.5. Performance-Monitoring Counters

The P6 family and Pentium® processors provide two performance-monitoring counters for use
in monitoring internal hardware operations. These counters are event counters that can be
programmed to count a variety of different types of events, such as the number of instructions
decoded, number of interrupts received, or number of cache loads. Appendix A, Performance-
Monitoring Events lists all the events that can be counted (Table A-1 for the P6 family proces-
sors and Table A-2 for the Pentium® processors). The counters are set up, started, and stopped
using two MSRs and the RDMSR and WRMSR instructions. For the P6 family processors, the
current count for a particular counter can be read using the new RDPMC instruction.
The performance-monitoring counters are useful for debugging programs, optimizing code,
diagnosing system failures, or refining hardware designs. Refer to Chapter 15, Debugging and
Performance Monitoring for more information on these counters.

18-40
 A
Performance-
Monitoring Events
 PERFORMANCE-MONITORING EVENTS

APPENDIX A
PERFORMANCE-MONITORING EVENTS

This appendix contains list of the performance-monitoring events that can be monitored with the
Intel Architecture processors. In the Intel Architecture processors, the ability to monitor perfor-
mance events and the events that can be monitored are model specific. Section A.1., “P6 Family
Processor Performance-Monitoring Events” lists and describes the events that can be monitored
with the P6 family of processors. Section A.2., “Pentium® Processor Performance-Monitoring
Events” lists and describes the events that can be monitored with Pentium® processors.

A.1. P6 FAMILY PROCESSOR PERFORMANCE-MONITORING

EVENTS
Table A-1 lists the events that can be counted with the performance-monitoring counters and
read with the RDPMC instruction for the P6 family of processors. The unit column gives the
microarchitecture or bus unit that produces the event; the event number column gives the hexa-
decimal number identifying the event; the mnemonic event name column gives the name of the
event; the unit mask column gives the unit mask required (if any); the description column de-
scribes the event; and the comments column gives additional information about the event.
These performance-monitoring events are intended to be used as guides for performance tuning.
The counter values reported are not guaranteed to be absolutely accurate and should be used as
a relative guide for tuning. Known discrepancies are documented where applicable.
Some performance events are model specific. Those added in later generations of the P6 family
processors are listed in this table. Performance events are not architecturally guaranteed in fu-
ture versions of the P6 family processors. All performance event encodings not listed in Table
A-1 are reserved and their use will result in undefined counter results.
Refer to the end of the table for notes related to certain entries in the table.

A-1
PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
Data Cache 43H DATA_MEM_REFS 00H All loads from any memory type.
Unit (DCU) All stores to any memory type.
Each part of a split is counted
separately. The internal logic
counts not only memory loads
and stores, but also internal
retries.

Note: 80-bit floating-point

accesses are double counted,
since they are decomposed into
a 16-bit exponent load and a 64-
bit mantissa load. Memory
accesses are only counted
when they are actually
performed (such as a load that
gets squashed because a
previous cache miss is
outstanding to the same
address, and which finally gets
performed, is only counted
once).

Does not include I/O accesses,

or other nonmemory accesses.
45H DCU_LINES_IN 00H Total lines allocated in the DCU.
46H DCU_M_LINES_IN 00H Number of M state lines
allocated in the DCU.
47H DCU_M_LINES_OUT 00H Number of M state lines evicted
from the DCU. This includes
evictions via snoop HITM,
intervention or replacement.
48H DCU_MISS_ 00H Weighted number of cycles An access that also
OUTSTANDING while a DCU miss is misses the L2 is short-
outstanding, incremented by the changed by 2 cycles
number of outstanding cache (i.e., if counts N cycles,
misses at any particular time. should be N+2 cycles).

Cacheable read requests only Subsequent loads to the

are considered. same cache line will not
result in any additional
Uncacheable requests are counts.
excluded.
Count value not precise,
Read-for-ownerships are but still useful.
counted, as well as line fills,
invalidates, and stores.
Instruction 80H IFU_IFETCH 00H Number of instruction fetches,
Fetch Unit both cacheable and
(IFU) noncacheable, including UC
fetches.
81H IFU_IFETCH_MISS 00H Number of instruction fetch
misses.

All instruction fetches that do not

hit the IFU (i.e., that produce
memory requests).

Includes UC accesses.
85H ITLB_MISS 00H Number of ITLB misses.

A-2
 PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
86H IFU_MEM_STALL 00H Number of cycles instruction
fetch is stalled, for any reason.

Includes IFU cache misses,

ITLB misses, ITLB faults, and
other minor stalls.
87H ILD_STALL 00H Number of cycles that the
instruction length decoder is
stalled.
L2 Cache1 28H L2_IFETCH MESI Number of L2 instruction
0FH fetches.

This event indicates that a

normal instruction fetch was
received by the L2.

The count includes only L2

cacheable instruction fetches; it
does not include UC instruction
fetches.

It does not include ITLB miss

accesses.
29H L2_LD MESI Number of L2 data loads.
0FH
This event indicates that a
normal, unlocked, load memory
access was received by the L2.

It includes only L2 cacheable

memory accesses; it does not
include I/O accesses, other
nonmemory accesses, or
memory accesses such as
UC/WT memory accesses.

It does include L2 cacheable

TLB miss memory accesses.
2AH L2_ST MESI Number of L2 data stores.
0FH
This event indicates that a
normal, unlocked, store memory
access was received by the L2.

Specifically, it indicates that the

DCU sent a read-for-ownership
request to the L2.

It also includes Invalid to

Modified requests sent by the
DCU to the L2.

It includes only L2 cacheable

memory accesses; it does not
include I/O accesses, other
nonmemory accesses, or
memory accesses such as
UC/WT memory accesses.

It includes TLB miss memory

accesses.
24H L2_LINES_IN 00H Number of lines allocated in the
L2.

A-3
PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
26H L2_LINES_OUT 00H Number of lines removed from
the L2 for any reason.
25H L2_M_LINES_INM 00H Number of modified lines
allocated in the L2.
27H L2_M_LINES_OUTM 00H Number of modified lines
removed from the L2 for any
reason.
2EH L2_RQSTS MESI Total number of L2 requests.
0FH
21H L2_ADS 00H Number of L2 address strobes.
22H L2_DBUS_BUSY 00H Number of cycles during which
the L2 cache data bus was busy.
23H L2_DBUS_BUSY_RD 00H Number of cycles during which
the data bus was busy
transferring read data from L2 to
the processor.
External Bus 62H BUS_DRDY_ 00H Number of clocks during which Unit Mask = 00H counts
Logic (EBL)2 CLOCKS (Self) DRDY# is asserted. bus clocks when the
20H processor is driving
(Any) Utilization of the external system DRDY#.
data bus during data transfers.
Unit Mask = 20H counts
in processor clocks
when any agent is
driving DRDY#.
63H BUS_LOCK_ 00H Number of clocks during which Always counts in
CLOCKS (Self) LOCK# is asserted on the processor clocks.
20H external system bus.3
(Any)
60H BUS_REQ_ 00H Number of bus requests Counts only DCU full-
OUTSTANDING (Self) outstanding. line cacheable reads, not
RFOs, writes, instruction
This counter is incremented by fetches, or anything else.
the number of cacheable read Counts “waiting for bus
bus requests outstanding in any to complete” (last data
given cycle. chunk received).
65H BUS_TRAN_BRD 00H Number of burst read
(Self) transactions.
20H
(Any)
66H BUS_TRAN_RFO 00H Number of completed read for
(Self) ownership transactions.
20H
(Any)
67H BUS_TRANS_WB 00H Number of completed write back
(Self) transactions.
20H
(Any)
68H BUS_TRAN_ 00H Number of completed instruction
IFETCH (Self) fetch transactions.
20H
(Any)

A-4
 PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
69H BUS_TRAN_INVAL 00H Number of completed invalidate
(Self) transactions.
20H
(Any)
6AH BUS_TRAN_PWR 00H Number of completed partial
(Self) write transactions.
20H
(Any)
6BH BUS_TRANS_P 00H Number of completed partial
(Self) transactions.
20H
(Any)
6CH BUS_TRANS_IO 00H Number of completed I/O
(Self) transactions.
20H
(Any)
6DH BUS_TRAN_DEF 00H Number of completed deferred
(Self) transactions.
20H
(Any)
6EH BUS_TRAN_BURST 00H Number of completed burst
(Self) transactions.
20H
(Any)
70H BUS_TRAN_ANY 00H Number of all completed bus
(Self) transactions.
20H
(Any) Address bus utilization can be
calculated knowing the minimum
address bus occupancy.

Includes special cycles, etc.

6FH BUS_TRAN_MEM 00H Number of completed memory
(Self) transactions.
20H
(Any)
64H BUS_DATA_RCV 00H Number of bus clock cycles
(Self) during which this processor is
receiving data.
61H BUS_BNR_DRV 00H Number of bus clock cycles
(Self) during which this processor is
driving the BNR# pin.

A-5
PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
7AH BUS_HIT_DRV 00H Number of bus clock cycles Includes cycles due to
(Self) during which this processor is snoop stalls.
driving the HIT# pin.
The event counts
correctly, but the BPMi
pins function as follows
based on the setting of
the PC bits (bit 19 in the
PerfEvtSel0 and
PerfEvtSel1 registers):

If the core-clock-to- bus-

clock ratio is 2:1 or 3:1,
and a PC bit is set, the
BPMipins will be
asserted for a single
clock when the counters
overflow.

If the PC bit is clear, the

processor toggles the
BPMipins when the
counter overflows.

If the clock ratio is not

2:1 or 3:1, the BPMi
pins will not function for
these performance-
monitoring counter
events.
7BH BUS_HITM_DRV 00H Number of bus clock cycles Includes cycles due to
(Self) during which this processor is snoop stalls.
driving the HITM# pin.
The event counts
correctly, but the BPM i
pins function as follows
based on the setting of
the PC bits (bit 19 in the
PerfEvtSel0 and
PerfEvtSel1 registers):

If the core-clock-to- bus-

clock ratio is 2:1 or 3:1,
and a PC bit is set, the
BPMipins will be
asserted for a single
clock when the counters
overflow.

If the PC bit is clear, the

processor toggles the
BPMipins when the
counter overflows.

If the clock ratio is not

2:1 or 3:1, the BPMi
pins will not function for
these performance-
monitoring counter
events.
7EH BUS_SNOOP_STALL 00H Number of clock cycles during
(Self) which the bus is snoop stalled.

A-6
 PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
Floating- C1H FLOPS 00H Number of computational Counter 0 only.
Point Unit floating-point operations retired.

Excludes floating-point
computational operations that
cause traps or assists.

Includes floating-point
computational operations
executed by the assist handler.

Includes internal sub-operations

for complex floating-point
instructions like
transcendentals.

Excludes floating-point loads

and stores.
10H FP_COMP_OPS_ 00H Number of computational Counter 0 only.
EXE floating-point operations
executed.

The number of FADD, FSUB,

FCOM, FMULs, integer MULs
and IMULs, FDIVs, FPREMs,
FSQRTS, integer DIVs, and
IDIVs.

Note not the number of cycles,

but the number of operations.

This event does not distinguish

an FADD used in the middle of a
transcendental flow from a
separate FADD instruction.
11H FP_ASSIST 00H Number of floating-point Counter 1 only.
exception cases handled by
microcode. This event includes
counts due to
speculative execution.
12H MUL 00H Number of multiplies. Counter 1 only.

Note: Includes integer as well as

FP multiplies and is speculative.
13H DIV 00H Number of divides. Counter 1 only.

Note: Includes integer as well as

FP divides and is speculative.
14H CYCLES_DIV_BUSY 00H Number of cycles during which Counter 0 only.
the divider is busy, and cannot
accept new divides.

Note: Includes integer and FP

divides, FPREM, FPSQRT, etc.,
and is speculative.

A-7
PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
Memory 03H LD_BLOCKS 00H Number of store buffer blocks.
Ordering
Includes counts caused by
preceding stores whose
addresses are unknown,
preceding stores whose
addresses are known but whose
data is unknown, and preceding
stores that conflicts with the load
but which incompletely overlap
the load.
04H SB_DRAINS 00H Number of store buffer drain
cycles.

Incremented every cycle the

store buffer is draining.

Draining is caused by serializing

operations like CPUID,
synchronizing operations like
XCHG, interrupt
acknowledgment, as well as
other conditions (such as cache
flushing).
05H MISALIGN_ 00H Number of misaligned data It should be noted that
MEM_REF memory references. MISALIGN_MEM_REF
is only an approximation
Incremented by 1 every cycle, to the true number of
during which either the proc load misaligned memory
or store pipeline dispatches a references.
misaligned uop.
The value returned is
Counting is performed if it is the roughly proportional to
first or second half, or if it is the number of
blocked, squashed, or missed. misaligned memory
accesses, i.e., the size
Note: In this context, misaligned of the problem.
means crossing a 64-bit
boundary.
07H EMON_KNI_PREF_ Number of Streaming SIMD Counters 0 and 1.
DISPATCHED extensions prefetch/weakly- Pentium® III processor
ordered instructions dispatched only.
(speculative prefetches are
included in counting)
0: prefetch NTA
1: prefetch T1
2: prefetch T2
00H 3: weakly ordered stores
01H
02H
03H
4BH EMON_KNI_PREF_ Number of prefetch/weakly- Counters 0 and 1.
MISS ordered instructions that miss all Pentium® III processor
caches. only.
0: prefetch NTA
00H 1: prefetch T1
01H 2: prefetch T2
02H 3: weakly ordered stores
03H

A-8
 PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
Instruction C0H INST_RETIRED OOH Number of instructions retired. A hardware interrupt
Decoding received during/after the
and last iteration of the REP
Retirement STOS flow causes the
counter to undercount by
1 instruction.
C2H UOPS_RETIRED 00H Number of UOPs retired.
D0H INST_DECODED 00H Number of instructions decoded.
D8H EMON_KNI_INST_ Number of Streaming SIMD Counters 0 and 1.
RETIRED extensions retired Pentium® III processor
0: packed & scalar only.
00H 1: scalar
01H
D9H EMON_KNI_COMP_ Number of Streaming SIMD Counters 0 and 1.
INST_RET extensions computation Pentium® III processor
instructions retired. only.
0: packed and scalar
00H 1: scalar
01H
Interrupts C8H HW_INT_RX 00H Number of hardware interrupts
received.
C6H CYCLES_INT_ 00H Number of processor cycles for
MASKED which interrupts are disabled.
C7H CYCLES_INT_ 00H Number of processor cycles for
PENDING_ which interrupts are disabled
AND_MASKED and interrupts are pending.
Branches C4H BR_INST_RETIRED 00H Number of branch instructions
retired.
C5H BR_MISS_PRED_ 00H Number of mispredicted
RETIRED branches retired.
C9H BR_TAKEN_ 00H Number of taken branches
RETIRED retired.
CAH BR_MISS_PRED_ 00H Number of taken mispredictions
TAKEN_RET branches retired.
E0H BR_INST_DECODED 00H Number of branch instructions
decoded.
E2H BTB_MISSES 00H Number of branches for which
the BTB did not produce a
prediction.
E4H BR_BOGUS 00H Number of bogus branches.
E6H BACLEARS 00H Number of times BACLEAR is
asserted.

This is the number of times that

a static branch prediction was
made, in which the branch
decoder decided to make a
branch prediction because the
BTB did not.

A-9
PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
Stalls A2H RESOURCE_STALLS 00H Incremented by 1 during every
cycle for which there is a
resource related stall.

Includes register renaming

buffer entries, memory buffer
entries.

Does not include stalls due to

bus queue full, too many cache
misses, etc.

In addition to resource related

stalls, this event counts some
other events.

Includes stalls arising during

branch misprediction recovery,
such as if retirement of the
mispredicted branch is delayed
and stalls arising while store
buffer is draining from
synchronizing operations.
D2H PARTIAL_RAT_ 00H Number of cycles or events for
STALLS partial stalls.

Note: Includes flag partial stalls.

Segment 06H SEGMENT_REG_ 00H Number of segment register
Register LOADS loads.
Loads
Clocks 79H CPU_CLK_ 00H Number of cycles during which
UNHALTED the processor is not halted.
MMX™ Unit B0H MMX_INSTR_EXEC 00H Number of MMX™ Instructions Available in Intel®
Executed. Celeron, Pentium® II
and Pentium® II Xeon
processors only.

Does not account for

MOVQ and MOVD
stores from register to
memory.
B1H MMX_SAT_ 00H Number of MMX™ Saturating Available in Pentium® II
INSTR_EXEC Instructions Executed. & Pentium® III
processors only.

B2H MMX_UOPS_EXEC 0FH Number of MMX™ UOPS Available in Pentium® II

Executed. & Pentium® III
processors only.
B3H MMX_INSTR_ 01H MMX™ packed multiply Available in Pentium® II
TYPE_EXEC instructions executed. & Pentium® III
02H MMX™ packed shift instructions processors only.
executed.
04H MMX™ pack operation
instructions executed.
08H MMX™ unpack operation
instructions executed.
MMX™ packed logical
10H instructions executed.
MMX™ packed arithmetic
20H instructions executed.

A-10
 PERFORMANCE-MONITORING EVENTS

Table A-1. Events That Can Be Counted with the P6 Family Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event Unit
Unit Num. Name Mask Description Comments
CCH FP_MMX_TRANS 00H Transitions from MMX™ Available in Pentium® II
instruction to floating-point & Pentium® III
instructions. processors only.
01H Transitions from floating-point
instructions to MMX ™
instructions.
CDH MMX_ASSIST 00H Number of MMX™ Assists (that Available in Pentium® II
is, the number of EMMS & Pentium® III
instructions executed). processors only.
CEH MMX_INSTR_RET 00H Number of MMX™ Instructions Available in Pentium® II
Retired. processor only.
Segment D4H SEG_RENAME_ Number of Segment Register Available in Pentium® II
Register STALLS Renaming Stalls: & Pentium® III
Renaming Segment register ES processors only.
01H Segment register DS
02H Segment register FS
04H Segment register FS
08H Segment registers ES + DS +
0FH FS + GS
D5H SEG_REG_ Number of Segment Register Available in Pentium® II
RENAMES Renames: & Pentium® III
01H Segment register ES processors only.
02H Segment register DS
04H Segment register FS
08H Segment register FS
0FH Segment registers ES + DS +
FS + GS
D6H RET_SEG_ 00H Number of segment register Available in Pentium® II
RENAMES rename events retired. & Pentium® III
processors only.

NOTES:
1. Several L2 cache events, where noted, can be further qualified using the Unit Mask (UMSK) field in the
PerfEvtSel0 and PerfEvtSel1 registers. The lower 4 bits of the Unit Mask field are used in conjunction
with L2 events to indicate the cache state or cache states involved. The P6 family processors identify
cache states using the “MESI” protocol and consequently each bit in the Unit Mask field represents one of
the four states: UMSK[3] = M (8H) state, UMSK[2] = E (4H) state, UMSK[1] = S (2H) state, and UMSK[0]
= I (1H) state. UMSK[3:0] = MESI” (FH) should be used to collect data for all states; UMSK = 0H, for the
applicable events, will result in nothing being counted.
2. All of the external bus logic (EBL) events, except where noted, can be further qualified using the Unit
Mask (UMSK) field in the PerfEvtSel0 and PerfEvtSel1 registers. Bit 5 of the UMSK field is used in con-
junction with the EBL events to indicate whether the processor should count transactions that are self-
generated (UMSK[5] = 0) or transactions that result from any processor on the bus (UMSK[5] = 1).
3. L2 cache locks, so it is possible to have a zero count.

A-11
PERFORMANCE-MONITORING EVENTS

A.2. PENTIUM® PROCESSOR PERFORMANCE-MONITORING

EVENTS
Table A-2 lists the events that can be counted with the performance-monitoring counters for the
Pentium® processor. The Event Number column gives the hexadecimal code that identifies the
event and that is entered in the ES0 or ES1 (event select) fields of the CESR MSR. The Mne-
monic Event Name column gives the name of the event, and the Description and Comments col-
umns give detailed descriptions of the events. Most events can be counted with either counter 0
or counter 1; however, some events can only be counted with only counter 0 or only counter 1
(as noted).

NOTE
The events in the table that are shaded are implemented only in the Pentium®
processor with MMX technology.

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters
Event Mnemonic Event
Num. Name Description Comments
00H DATA_READ Number of memory data Split cycle reads are counted
reads (internal data cache individually. Data Memory Reads that
hit and miss combined). are part of TLB miss processing are not
included. These events may occur at a
maximum of two per clock. I/O is not
included.
01H DATA_WRITE Number of memory data Split cycle writes are counted
writes (internal data cache individually. These events may occur at
hit and miss combined), a maximum of two per clock. I/O is not
I/O is not included. included.
0H2 DATA_TLB_MISS Number of misses to the
data cache translation
look-aside buffer.
03H DATA_READ_MISS Number of memory read Additional reads to the same cache line
accesses that miss the after the first BRDY# of the burst line fill
internal data cache is returned but before the final (fourth)
whether or not the access BRDY# has been returned, will not
is cacheable or cause the counter to be incremented
noncacheable. additional times. Data accesses that
are part of TLB miss processing are not
included. Accesses directed to I/O
space are not included.
04H DATA WRITE MISS Number of memory write Data accesses that are part of TLB
accesses that miss the miss processing are not included.
internal data cache Accesses directed to I/O space are not
whether or not the access included.
is cacheable or
noncacheable.

A-12
 PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
05H WRITE_HIT_TO_ Number of write hits to These are the writes that may be held
M-_OR_E- exclusive or modified lines up if EWBE# is inactive. These events
STATE_LINES in the data cache. may occur a maximum of two per clock.
06H DATA_CACHE_ Number of dirty lines (all) Replacements and internal and external
LINES_ that are written back, snoops can all cause writeback and are
WRITTEN_BACK regardless of the cause. counted.
07H EXTERNAL_ Number of accepted Assertions of EADS# outside of the
SNOOPS external snoops whether sampling interval are not counted, and
they hit in the code cache no internal snoops are counted.
or data cache or neither.
08H EXTERNAL_DATA_ Number of external Snoop hits to a valid line in either the
CACHE_SNOOP_ snoops to the data cache. data cache, the data line fill buffer, or
HITS one of the write back buffers are all
counted as hits.
09H MEMORY Number of data memory These accesses are not necessarily run
ACCESSES IN reads or writes that are in parallel due to cache misses, bank
BOTH PIPES paired in both pipes of the conflicts, etc.
pipeline.
0AH BANK CONFLICTS Number of actual bank
conflicts.
0BH MISALIGNED DATA Number of memory or I/O A 2- or 4-byte access is misaligned
MEMORY OR I/O reads or writes that are when it crosses a 4-byte boundary; an
REFERENCES misaligned. 8-byte access is misaligned when it
crosses an 8-byte boundary. Ten byte
accesses are treated as two separate
accesses of 8 and 2 bytes each.
0CH CODE READ Number of instruction Individual 8-byte noncacheable
reads whether the read is instruction reads are counted.
cacheable or
noncacheable.
0DH CODE TLB MISS Number of instruction Individual 8-byte noncacheable
reads that miss the code instruction reads are counted.
TLB whether the read is
cacheable or
noncacheable.
0EH CODE CACHE MISS Number of instruction Individual 8-byte noncacheable
reads that miss the instruction reads are counted.
internal code cache
whether the read is
cacheable or
noncacheable.

A-13
PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
0FH ANY SEGMENT Number of writes into any Segment loads are caused by explicit
REGISTER LOADED segment register in real or segment register load instructions, far
protected mode including control transfers, and task switches. Far
the LDTR, GDTR, IDTR, control transfers and task switches
and TR. causing a privilege level change will
signal this event twice. Note that
interrupts and exceptions may initiate a
far control transfer.
10H Reserved
11H Reserved
12H Branches Number of taken and not Also counted as taken branches are
taken branches, including serializing instructions, VERR and
conditional branches, VERW instructions, some segment
jumps, calls, returns, descriptor loads, hardware interrupts
software interrupts, and (including FLUSH#), and programmatic
interrupt returns. exceptions that invoke a trap or fault
handler. The pipe is not necessarily
flushed. The number of branches
actually executed is measured, not the
number of predicted branches.
13H BTB_HITS Number of BTB hits that Hits are counted only for those
occur. instructions that are actually executed.
14H TAKEN_BRANCH_ Number of taken This event type is a logical OR of taken
OR_BTB_HIT branches or BTB hits that branches and BTB hits. It represents an
occur. event that may cause a hit in the BTB.
Specifically, it is either a candidate for a
space in the BTB or it is already in the
BTB.
15H PIPELINE FLUSHES Number of pipeline The counter will not be incremented for
flushes that occur. serializing instructions (serializing
Pipeline flushes are instructions cause the prefetch queue
caused by BTB misses on to be flushed but will not trigger the
taken branches, Pipeline Flushed event counter) and
mispredictions, software interrupts (software interrupts
exceptions, interrupts, do not flush the pipeline).
and some segment
descriptor loads.

A-14
 PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
16H INSTRUCTIONS_ Number of instructions Invocations of a fault handler are
EXECUTED executed (up to two per considered instructions. All hardware
clock). and software interrupts and exceptions
will also cause the count to be
incremented. Repeat prefixed string
instructions will only increment this
counter once despite the fact that the
repeat loop executes the same
instruction multiple times until the loop
criteria is satisfied. This applies to all
the Repeat string instruction prefixes
(i.e., REP, REPE, REPZ, REPNE, and
REPNZ). This counter will also only
increment once per each HLT
instruction executed regardless of how
many cycles the processor remains in
the HALT state.
17H INSTRUCTIONS_ Number of instructions This event is the same as the 16H
EXECUTED_ V PIPE executed in the V_pipe. It event except it only counts the number
indicates the number of of instructions actually executed in the
instructions that were V-pipe.
paired.
18H BUS_CYCLE_ Number of clocks while a The count includes HLDA, AHOLD, and
DURATION bus cycle is in progress. BOFF# clocks.
This event measures bus
use.
19H WRITE_BUFFER_ Number of clocks while Full write buffers stall data memory
FULL_STALL_ the pipeline is stalled due read misses, data memory write
DURATION to full write buffers. misses, and data memory write hits to
S-state lines. Stalls on I/O accesses are
not included.
1AH WAITING_FOR_ Number of clocks while Data TLB Miss processing is also
DATA_MEMORY_ the pipeline is stalled included in the count. The pipeline stalls
READ_STALL_ while waiting for data while a data memory read is in progress
DURATION memory reads. including attempts to read that are not
bypassed while a line is being filled.
1BH STALL ON WRITE Number of stalls on writes
TO AN E- OR M- to E- or M-state lines
STATE LINE
1CH LOCKED BUS Number of locked bus Only the read portion of the locked
CYCLE cycles that occur as the read-modify-write is counted. Split
result of the LOCK prefix locked cycles (SCYC active) count as
or LOCK instruction, two separate accesses. Cycles
page-table updates, and restarted due to BOFF# are not re-
descriptor table updates. counted.

A-15
PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
1DH I/O READ OR Number of bus cycles Misaligned I/O accesses will generate
WRITE CYCLE directed to I/O space. two bus cycles. Bus cycles restarted
due to BOFF# are not re-counted.
1EH NONCACHEABLE_ Number of noncacheable Cycles restarted due to BOFF# are not
MEMORY_READS instruction or data re-counted.
memory read bus cycles.
Count includes read
cycles caused by TLB
misses, but does not
include read cycles to I/O
space.
1FH PIPELINE_AGI_ Number of address An AGI occurs when the instruction in
STALLS generation interlock (AGI) the execute stage of either of U- or V-
stalls. An AGI occurring in pipelines is writing to either the index or
both the U- and V- base address register of an instruction
pipelines in the same in the D2 (address generation) stage of
clock signals this event either the U- or V- pipelines.
twice.
20H Reserved
21H Reserved
22H FLOPS Number of floating-point Number of floating-point adds,
operations that occur. subtracts, multiplies, divides,
remainders, and square roots are
counted. The transcendental
instructions consist of multiple adds and
multiplies and will signal this event
multiple times. Instructions generating
the divide-by-zero, negative square
root, special operand, or stack
exceptions will not be counted.
Instructions generating all other
floating-point exceptions will be
counted. The integer multiply
instructions and other instructions
which use the FPU will be counted.

A-16
 PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
23H BREAKPOINT Number of matches on The counters is incremented regardless
MATCH ON DR0 register DR0 breakpoint. if the breakpoints are enabled or not.
REGISTER However, if breakpoints are not
enabled, code breakpoint matches will
not be checked for instructions
executed in the V-pipe and will not
cause this counter to be incremented.
(They are checked on instruction
executed in the U-pipe only when
breakpoints are not enabled.) These
events correspond to the signals driven
on the BP[3:0] pins. Refer to Chapter
15, Debugging and Performance
Monitoring, for more information.
24H BREAKPOINT Number of matches on Refer to comment for 23H event.
MATCH ON DR1 register DR1 breakpoint.
REGISTER
25H BREAKPOINT Number of matches on Refer to comment for 23H event.
MATCH ON DR2 register DR2 breakpoint.
REGISTER
26H BREAKPOINT Number of matches on Refer to comment for 23H event.
MATCH ON DR3 register DR3 breakpoint.
REGISTER
27H HARDWARE Number of taken INTR
INTERRUPTS and NMI interrupts.
28H DATA_READ_OR_ Number of memory data Split cycle reads and writes are counted
WRITE reads and/or writes individually. Data Memory Reads that
(internal data cache hit are part of TLB miss processing are not
and miss combined). included. These events may occur at a
maximum of two per clock. I/O is not
included.
29H DATA_READ_MISS Number of memory read Additional reads to the same cache line
OR_WRITE MISS and/or write accesses that after the first BRDY# of the burst line fill
miss the internal data is returned but before the final (fourth)
cache whether or not the BRDY# has been returned, will not
access is cacheable or cause the counter to be incremented
noncacheable. additional times. Data accesses that
are part of TLB miss processing are not
included. Accesses directed to I/O
space are not included.

A-17
PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
2AH BUS_OWNERSHIP_ The time from LRM bus The ratio of the 2AH events counted on
LATENCY (Counter ownership request to bus counter 0 and counter 1 is the average
0) ownership granted (that stall time due to bus ownership conflict.
is, the time from the
earlier of a PBREQ (0),
PHITM# or HITM#
assertion to a PBGNT
assertion).
2AH BUS OWNERSHIP The number of buss The ratio of the 2AH events counted on
TRANSFERS ownership transfers (that counter 0 and counter 1 is the average
(Counter 1) is, the number of PBREQ stall time due to bus ownership conflict.
(0) assertions.
2BH MMX_ Number of MMX™
INSTRUCTIONS_ instructions executed in
EXECUTED_ the U-pipe.
U-PIPE (Counter 0)
2BH MMX_ Number of MMX™
INSTRUCTIONS_ instructions executed in
EXECUTED_ the V-pipe.
V-PIPE (Counter 1)
2CH CACHE_M- Number of times a If the average memory latencies of the
STATE_LINE_ processor identified a hit system are known, this event enables
SHARING to a modified line due to a the user to count the Write Backs on
(Counter 0) memory access in the PHITM(O) penalty and the Latency on
other processor (PHITM Hit Modified(I) penalty.
(O)).
2CH CACHE_LINE_ Number of shared data
SHARING lines in the L1 cache
(Counter 1) (PHIT (O)).
2DH EMMS_ Number of EMMS
INSTRUCTIONS_ instructions executed.
EXECUTED
(Counter 0)
2DH TRANSITIONS_ Number of transitions This event counts the first floating-point
BETWEEN_MMX_ between MMX™ and instruction following an MMX™
AND_FP_ floating-point instructions instruction or first MMX™ instruction
INSTRUCTIONS or vice versa. An even following a floating-point instruction.
(Counter 1) count indicates the The count may be used to estimate the
processor is in MMX™ penalty in transitions between floating-
state. an odd count point state and MMX™ state.
indicates it is in FP state.
2DH BUS_UTILIZATION_ Number of clocks the bus
DUE_TO_ is busy due to the
PROCESSOR_ processor’s own activity,
ACTIVITY i.e., the bus activity that is
(Counter 0) caused by the processor.

A-18
 PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
2EH WRITES_TO_ Number of write accesses The count includes write cycles caused
NONCACHEABLE_ to noncacheable memory. by TLB misses and I/O write cycles.
MEMORY Cycles restarted due to BOFF# are not
(Counter 1) re-counted.
2FH SATURATING_ Number of saturating
MMX_ MMX™ instructions
INSTRUCTIONS_ executed, independently
EXECUTED of whether they actually
(Counter 0) saturated.
2FH SATURATIONS_ Number of MMX™ If an MMX™ instruction operating on 4
PERFORMED instructions that used doublewords saturated in three out of
(Counter 1) saturating arithmetic and the four results, the counter will be
that at least one of its incremented by one only.
results actually saturated.
30H NUMBER_OF_ Number of cycles the This event will enable the user to
CYCLES_NOT_IN_ processor is not idle due calculate “net CPI”. Note that during the
HALT_STATE to HLT instruction. time that the processor is executing the
(Counter 0) HLT instruction, the Time-Stamp
Counter is not disabled. Since this
event is controlled by the Counter
Controls CC0, CC1 it can be used to
calculate the CPI at CPL=3, which the
TSC cannot provide.
30H DATA_CACHE_ Number of clocks the
TLB_MISS_ pipeline is stalled due to a
STALL_DURATION data cache translation
(Counter 1) look-aside buffer (TLB)
miss.
31H MMX_ Number of MMX™
INSTRUCTION_ instruction data reads.
DATA_READS
(Counter 0)
31H MMX_ Number of MMX™
INSTRUCTION_ instruction data read
DATA_READ_ misses.
MISSES
(Counter 1)
32H FLOATING_POINT_ Number of clocks while
STALLS_DURATION pipe is stalled due to a
(Counter 0) floating-point freeze.
32H TAKEN_BRANCHES Number of taken
(Counter 1) branches.
33H D1_STARVATION_ Number of times D1 stage The D1 stage can issue 0, 1, or 2
AND_FIFO_IS_ cannot issue ANY instructions per clock if those are
EMPTY instructions since the available in an instructions FIFO buffer.
(Counter 0) FIFO buffer is empty.

A-19
PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
33H D1_STARVATION_ Number of times the D1 The D1 stage can issue 0, 1, or 2
AND_ONLY_ONE_ stage issues just a single instructions per clock if those are
INSTRUCTION_IN_ instruction since the FIFO available in an instructions FIFO buffer.
FIFO buffer had just one When combined with the previously
(Counter 1) instruction ready. defined events, Instruction Executed
(16H) and Instruction Executed in the V-
pipe (17H), this event enables the user
to calculate the numbers of time pairing
rules prevented issuing of two
instructions.
34H MMX_ Number of data writes
INSTRUCTION_ caused by MMX™
DATA_WRITES instructions.
(Counter 0)
34H MMX_ Number of data write
INSTRUCTION_ misses caused by MMX™
DATA_WRITE_ instructions.
MISSES
(Counter 1)
35H PIPELINE_ Number of pipeline The count includes any pipeline flush
FLUSHES_DUE_ flushes due to wrong due to a branch that the pipeline did not
TO_WRONG_ branch predictions follow correctly. It includes cases where
BRANCH_ resolved in either the E- a branch was not in the BTB, cases
PREDICTIONS stage or the WB-stage. where a branch was in the BTB but was
(Counter 0) mispredicted, and cases where a
branch was correctly predicted but to
the wrong address. Branches are
resolved in either the Execute stage (E-
stage) or the Writeback stage (WB-
stage). In the later case, the
misprediction penalty is larger by one
clock. The difference between the 35H
event count in counter 0 and counter 1
is the number of E-stage resolved
branches.
35H PIPELINE_ Number of pipeline Refer to note for event 35H (Counter 0).
FLUSHES_DUE_ flushes due to wrong
TO_WRONG_ branch predictions
BRANCH_ resolved in the WB-stage.
PREDICTIONS_
RESOLVED_IN_
WB-STAGE (Counter
1)
36H MISALIGNED_ Number of misaligned
DATA_MEMORY_ data memory references
REFERENCE_ON_ when executing MMX™
MMX_ instructions.
INSTRUCTIONS
(Counter 0)

A-20
 PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
36H PIPELINE_ Number clocks during
ISTALL_FOR_MMX_ pipeline stalls caused by
INSTRUCTION_ waits form MMX™
DATA_MEMORY_ instruction data memory
READS reads.
(Counter 1)
37H MISPREDICTED_ Number of returns The count is the difference between the
OR_ predicted incorrectly or total number of executed returns and
UNPREDICTED_ not predicted at all. the number of returns that were
RETURNS correctly predicted. Only RET
(Counter 1) instructions are counted (for example,
IRET instructions are not counted).
37H PREDICTED_ Number of predicted Only RET instructions are counted (for
RETURNS returns (whether they are example, IRET instructions are not
(Counter 1) predicted correctly and counted).
incorrectly.
38H MMX_MULTIPLY_ Number of clocks the pipe The counter will not be incremented if
UNIT_INTERLOCK is stalled since the there is another cause for a stall. For
(Counter 0) destination of previous each occurrence of a multiply interlock
MMX™ multiply this event will be counted twice (if the
instruction is not ready stalled instruction comes on the next
yet. clock after the multiply) or by one (if the
stalled instruction comes two clocks
after the multiply).
38H MOVD/MOVQ_ Number of clocks a
STORE_STALL_ MOVD/MOVQ instruction
DUE_TO_ store is stalled in D2 stage
PREVIOUS_MMX_ due to a previous MMX™
OPERATION operation with a
(Counter 1) destination to be used in
the store instruction.
39H RETURNS Number or returns Only RET instructions are counted;
(Counter 0) executed. IRET instructions are not counted. Any
exception taken on a RET instruction
and any interrupt recognized by the
processor on the instruction boundary
prior to the execution of the RET
instruction will also cause this counter
to be incremented.
39H Reserved
3AH BTB_FALSE_ Number of false entries in False entries are causes for
ENTRIES the Branch Target Buffer. misprediction other than a wrong
(Counter 0) prediction.

A-21
PERFORMANCE-MONITORING EVENTS

Table A-2. Events That Can Be Counted with the Pentium® Processor Performance-
Monitoring Counters (Contd.)
Event Mnemonic Event
Num. Name Description Comments
3AH BTB_MISS_ Number of times the BTB
PREDICTION_ON_ predicted a not-taken
NOT-TAKEN_ branch as taken.
BRANCH
(Counter 1)
3BH FULL_WRITE_ Number of clocks while
BUFFER_STALL_ the pipeline is stalled due
DURATION_ to full write buffers while
WHILE_ executing MMX™
EXECUTING_MMX_ instructions.
INSTRUCTIONS
(Counter 0)
3BH STALL_ON_MMX_ Number of clocks during
INSTRUCTION_ stalls on MMX™
WRITE_TO E-_OR_ instructions writing to E-
M-STATE_LINE or M-state lines.
(Counter 1)

A-22
 B
Model-Specific
Registers
 APPENDIX B
MODEL-SPECIFIC REGISTERS

Table B-1 lists the model-specific registers (MSRs) that can be read with the RDMSR and writ-
ten with the WRMSR instructions. Register addresses are given in both hexadecimal and deci-
mal; the register name is the mnemonic register name; the bit description describes individual
bits in registers.

NOTE
The registers with addresses 0H, 1H, 10H, 11H, 12H, and 13H in Table B-1
are available only in the Pentium® processor. Code code that accesses
registers 0H, 1H, and 10H will run on a P6 family processor without
generating exceptions; however, code that accesses registers 11H, 12H, and
13H will generate exceptions on a P6 family processor. The MSRs in this
table that are shaded are available only in the Pentium® II and later processors
in the P6 family.

Table B-1. Model-Specific Registers (MSRs)

Register Address
Hex Dec Register Name Bit Description
0H 0 P5_MC_ADDR
(Pentium® Processor Only)
1H 1 P5_MC_TYPE
(Pentium® Processor Only)
10H 16 TSC
11H 17 CESR
(Pentium® Processor Only)
12H 18 CTR0
(Pentium® Processor Only)
13H 19 CTR1
(Pentium® Processor Only)
1BH 27 APICBASE
7:0 Reserved
8 Boot Strap Processor indicator Bit. BSP= 1
10:9 Reserved
11 APIC Global Enable Bit - Permanent til reset
Enabled = 1, Disabled = 0
31:12 APIC Base Address

B-1
MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
63:32 Reserved
2AH 42 EBL_CR_POWERON
0 Reserved1
1 Data Error Checking Enable
1 = Enabled
0 = Disabled
Read/Write
2 Response Error Checking Enable
FRCERR Observation Enable
1 = Enabled
0 = Disabled
Read/Write
3 AERR# Drive Enable
1 = Enabled
0 = Disabled
Read/Write
4 BERR# Enable for initiator bus requests
1 = Enabled
0 = Disabled
Read/Write
5 Reserved
6 BERR# Driver Enable for initiator internal errors
1 = Enabled
0 = Disabled
Read/Write
7 BINIT# Driver Enable
1 = Enabled
0 = Disabled
Read/Write
8 Output Tri-state Enabled
1 = Enabled
0 = Disabled
Read
9 Execute BIST
1 = Enabled
0 = Disabled
Read
10 AERR# Observation Enabled
1 = Enabled
0 = Disabled
Read
11 Reserved

B-2
 MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
12 BINIT# Observation Enabled
1 = Enabled
0 = Disabled
Read
13 In Order Queue Depth
1=1
0=8
Read
14 1Mbyte Power on Reset Vector
1 = 1Mbyte
0 = 4Gbytes
Read Only
15 FRC Mode Enable
1 = Enabled
0 = Disabled
Read Only
17:16 APIC Cluster ID
Read
19:18 Reserved
21: 20 Symmetric Arbitration ID
Read
24:22 Clock Frequency Ratio
Read
25 Reserved
26 Low Power Mode Enable
Read/Write
63:27 Reserved1
33H 51 TEST_CTL Test Control Register
29:0 Reserved
30 Streaming Buffer Disable
31 Disable LOCK# assertion for split locked access
79H 121 BIOS_UPDT_TRIG BIOS Update Trigger Register
88 136 BBL_CR_D0[63:0] Chunk 0 data register D[63:0]: used to write to and
read from the L2
89 137 BBL_CR_D1[63:0] Chunk 1 data register D[63:0]: used to write to and
read from the L2
8A 138 BBL_CR_D2[63:0] Chunk 2 data register D[63:0]: used to write to and
read from the L2

B-3
MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
8BH 139 BIOS_SIGN/BBL_CR_D3[ BIOS Update Signature Register or Chunk 3 data
63:0] register D[63:0]: used to write to and read from the L2
depending on the usage model
C1H 193 PERFCTR0
C2H 194 PERFCTR1
FEH 254 MTRRcap
116 278 BBL_CR_ADDR [63:0] Address register: used to send specified address (A31-
A3) to L2 during cache initialization accesses.
BBL_CR_ADDR [63:32] Reserved,
BBL_CR_ADDR [31:3] Address bits [35:3]
BBL_CR_ADDR [2:0] Reserved Set to 0.
118 280 BBL_CR_DECC[63:0] Data ECC register D[7:0]: used to write ECC and read
ECC to/from L2
119 281 BBL_CR_CTL Control register: used to program L2 commands to be
issued via cache configuration accesses mechanism.
Also receives L2 lookup response

BL_CR_CTL[63:22] Reserved
BBL_CR_CTL[21] Processor number2
Disable = 1
Enable = 0
BBL_CR_CTL[20:19] Reserved
BBL_CR_CTL[18] User supplied ECC
BBL_CR_CTL[17] Reserved
BBL_CR_CTL[16] L2 Hit
BBL_CR_CTL[15:14] Reserved
BBL_CR_CTL[13:12] State from L2
Modified - 11,Exclusive - 10, Shared - 01, Invalid - 00
BBL_CR_CTL[11:10] Way from L2
Way 0 - 00, Way 1 - 01, Way 2 - 10, Way 3 - 11
BBL_CR_CTL[9:8] Way to L2
BBL_CR_CTL[7] Reserved
BBL_CR_CTL[6:5] State to L2
BBL_CR_CTL[4:0] L2 Command
01100 Data Read w/ LRU update (RLU)
01110 Tag Read w/ Data Read (TRR)
01111 Tag Inquire (TI)
00010 L2 Control Register Read (CR)
00011 L2 Control Register Write (CW)
010 + MESI encode Tag Write w/ Data Read (TWR)
111 + MESI encode Tag Write w/ Data Write (TWW)
100 + MESI encode Tag Write (TW)
11A 282 BBL_CR_TRIG Trigger register: used to initiate a cache configuration
accesses access, Write only with Data=0.
11B 283 BBL_CR_BUSY Busy register: indicates when a cache configuration
accesses L2 command is in progress. D[0] = 1 = BUSY

B-4
 MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
11E 286 BBL_CR_CTL3 Control register 3: used to configure the L2 Cache

BBL_CR_CTL3[63:26] Reserved
BBL_CR_CTL3[25] Cache bus fraction (read only)
BBL_CR_CTL3[24] Reserved
BBL_CR_CTL3[23] L2 Hardware Disable (read only)
BBL_CR_CTL3[22:20] L2 Physical Address Range support
111 64Gbytes
110 32Gbytes
101 16Gbytes
100 8Gbytes
011 4Gbytes
010 2Gbytes
001 1Gbytes
000 512Mbytes
BBL_CR_CTL3[19] Reserved
BBL_CR_CTL3[18] Cache State error checking enable (read/write)
BBL_CR_CTL3[17:13 Cache size per bank (read/write)
00001 256Kbytes
00010 512Kbytes
00100 1Mbyte
01000 2Mbyte
10000 4Mbytes
BBL_CR_CTL3[12:11] Number of L2 banks (read only)
BBL_CR_CTL3[10:9] L2 Associativity (read only)
00 Direct Mapped
01 2 Way
10 4 Way
11 Reserved
BBL_CR_CTL3[8] L2 Enabled (read/write)
BBL_CR_CTL3[7] CRTN Parity Check Enable (read/write)
BBL_CR_CTL3[6] Address Parity Check Enable (read/write)
BBL_CR_CTL3[5] ECC Check Enable (read/write)
BBL_CR_CTL3[4:1] L2 Cache Latency (read/write)
BBL_CR_CTL3[0] L2 Configured (read/write)
179H 377 MCG_CAP
17AH 378 MCG_STATUS
17BH 379 MCG_CTL
186H 390 EVNTSEL0
7:0 Event Select
(Refer to Performance Counter section for a list of
event encodings)
15:8 UMASK:
Unit Mask Register Set to 0 to enable all count options
16 USER:
Controls the counting of events at Privilege levels of 1,
2, and 3

B-5
MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
17 OS:
Controls the counting of events at Privilege level of 0
18 E:
Occurrence/Duration Mode Select
1 = Occurrence
0 = Duration
19 PC:
Enabled the signaling of performance counter overflow
via BP0 pin
20 INT:
Enables the signaling of counter overflow via input to
APIC
1 = Enable
0 = Disable
22 ENABLE:
Enables the counting of performance events in both
counters
1 = Enable
0 = Disable
23 INV:
Inverts the result of the CMASK condition
1 = Inverted
0 = Non-Inverted
31:24 CMASK:
Counter Mask
187H 391 EVNTSEL1
7:0 Event Select
(Refer to Performance Counter section for a list of
event encodings)
15:8 UMASK:
Unit Mask Register Set to Zero to enable all count
options
16 USER:
Controls the counting of events at Privilege levels of 1,
2, and 3
17 OS:
Controls the counting of events at Privilege level of 0
18 E:
Occurrence/Duration Mode Select
1 = Occurrence
0 = Duration
19 PC:
Enabled the signaling of performance counter overflow
via BP0 pin.

B-6
 MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
20 INT:
Enables the signaling of counter overflow via input to
APIC
1 = Enable
0 = Disable
23 INV:
Inverts the result of the CMASK condition
1 = Inverted
0 = Non-Inverted
31:24 CMASK:
Counter Mask
1D9H 473 DEBUGCTLMSR
0 Enable/Disable Last Branch Records
1 Branch Trap Flag
2 Performance Monitoring/Break Point Pins
3 Performance Monitoring/Break Point Pins
4 Performance Monitoring/Break Point Pins
5 Performance Monitoring/Break Point Pins
6 Enable/Disable Execution Trace Messages
13:7 Reserved
14 Enable/Disable Execution Trace Messages
15 Enable/Disable Execution Trace Messages
1DBH 475 LASTBRANCHFROMIP
1DCH 476 LASTBRANCHTOIP
1DDH 477 LASTINTFROMIP
1DEH 478 LASTINTTOIP
1E0H 480 ROB_CR_BKUPTMPDR6
1:0 Reserved
2 Fast String Enable bit. Default is enabled
200H 512 MTRRphysBase0
201H 513 MTRRphysMask0
202H 514 MTRRphysBase1
203H 515 MTRRphysMask1
204H 516 MTRRphysBase2
205H 517 MTRRphysMask2

B-7
MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
206H 518 MTRRphysBase3
207H 519 MTRRphysMask3
208H 520 MTRRphysBase4
209H 521 MTRRphysMask4
20AH 522 MTRRphysBase5
20BH 523 MTRRphysMask5
20CH 524 MTRRphysBase6
20DH 525 MTRRphysMask6
20EH 526 MTRRphysBase7
20FH 527 MTRRphysMask7
250H 592 MTRRfix64K_00000
258H 600 MTRRfix16K_80000
259H 601 MTRRfix16K_A0000
268H 616 MTRRfix4K_C0000
269H 617 MTRRfix4K_C8000
26AH 618 MTRRfix4K_D0000
26BH 619 MTRRfix4K_D8000
26CH 620 MTRRfix4K_E0000
26DH 621 MTRRfix4K_E8000
26EH 622 MTRRfix4K_F0000
26FH 623 MTRRfix4K_F8000
2FFH 767 MTRRdefType
2:0 Default memory type
10 Fixed MTRR enable
11 MTRR Enable
400H 1024 MC0_CTL
401H 1025 MC0_STATUS
63 MC_STATUS_V
62 MC_STATUS_O
61 MC_STATUS_UC
60 MC_STATUS_EN
59 MC_STATUS_MISCV

B-8
 MODEL-SPECIFIC REGISTERS

Table B-1. Model-Specific Registers (MSRs) (Contd.)

Register Address
Hex Dec Register Name Bit Description
58 MC_STATUS_ADDRV
57 MC_STATUS_DAM
31:16 MC_STATUS_MCACOD
15:0 MC_STATUS_MSCOD
402H 1026 MC0_ADDR
403H 1027 MC0_MISC Defined in MCA architecture but not implemented in
the P6 family processors
404H 1028 MC1_CTL
405H 1029 MC1_STATUS Bit definitions same as MC0_STATUS
406H 1030 MC1_ADDR
407H 1031 MC1_MISC Defined in MCA architecture but not implemented in
the P6 family processors
408H 1032 MC2_CTL
409H 1033 MC2_STATUS Bit definitions same as MC0_STATUS
40AH 1034 MC2_ADDR
40BH 1035 MC2_MISC Defined in MCA architecture but not implemented in
the P6 family processors
40CH 1036 MC4_CTL
40DH 1037 MC4_STATUS Bit definitions same as MC0_STATUS
40EH 1038 MC4_ADDR Defined in MCA architecture but not implemented in P6
Family processors
40FH 1039 MC4_MISC Defined in MCA architecture but not implemented in
the P6 family processors
410H 1040 MC3_CTL
411H 1041 MC3_STATUS Bit definitions same as MC0_STATUS
412H 1042 MC3_ADDR
413H 1043 MC3_MISC Defined in MCA architecture but not implemented in
the P6 family processors

NOTES:
1. Bit 0 of this register has been redefined several times, and is no longer used in Pentium® Pro processors.
2. The processor number feature may be disabled by setting bit 21 of the BBL_CR_CTL MSR (model-spe-
cific register address 119h) to “1”. Once set, bit 21 of the BBL_CR_CTL may not be cleared. This bit is
write-once. The processor number feature will be disabled until the processor is reset.
3. The Pentium® III processor will prevent FSB frequency overclocking with a new shutdown mechanism. If
the FSB frequency selected is greater than the internal FSB frequency the processor will shutdown. If the
FSB selected is less than the internal FSB frequency the BIOS may choose to use bit 11 to implement its
own shutdown policy.

B-9
 C
Dual-Processor
Bootup Sequence
Example
(Specific to Pentium® Processors)
 APPENDIX C
DUAL-PROCESSOR (DP) BOOTUP SEQUENCE
EXAMPLE (SPECIFIC TO PENTIUM®
PROCESSORS)

The following example shows the DP protocol for booting two Pentium® processors (a primary
processor and a secondary processor) in a DP system and initializing their APICs. For dual-pro-
cessor systems based on Pentium® processors, the APIC ID of the primary processor is always 0.
The following constants and data definitions are used in the accompanying code examples. They
are based on the addresses of the APIC registers as defined in Table 7-1 in Chapter 7.
ICR_LOW EQU 0FEE00300H
ICR_HI EQU 0FEE00310H
SVR EQU 0FEE000F0H
APIC_ID EQU 0FEE00020H
LVT3 EQU 0FEE00370H
APIC_ENABLED EQU 100H
BOOT_ID DW ?
UPGRD_ID DW ?

C.1. PRIMARY PROCESSOR’S SEQUENCE OF EVENTS

1. The primary processor boots at the Intel Architecture address and executes until it is ready
to activate the secondary processor.
2. Initialization software should execute the CPUID instruction to determine if the primary
processor is a “GenuineIntel.” The values of EAX and EDX should be saved into a config-
uration RAM space for use later.
If the type field (in the EAX register following CPUID instruction execution) is 01B in bits
13 and 14, respectively, the processor is a future Pentium® OverDrive® processor and the
Pentium® processor (735/90, 815/100, 1000,120, 1110/133) has been put to sleep. This
means the system is a uniprocessor system and normal AT system configuration can
continue. Go to step 14 to configure the APIC.
If the type field is 00B, the processor is the primary processor and detection of the
secondary processor is required. Continue with steps 3 through 13.

C-1
DUAL-PROCESSOR (DP) BOOTUP SEQUENCE EXAMPLE (SPECIFIC

3. The following operation can be used to detect the secondary processor:

Set a timer before sending the start-up IPI to the secondary processor. In the secondary
processor’s initialization routine, it should write a value into memory indicating its
presence. The primary processor can then use the timer expiration to check if something
has been written into memory. If the timer expires and nothing has been written into
memory, the secondary processor is not present or some error has occurred.
4. Load start-up code for the secondary processor to execute into a 4-KByte page in the lower
1 MByte of memory.
5. Switch to protected mode (to access APIC address space above 1 MByte).
6. Determine the Pentium® processor’s APIC ID from the local APIC ID register (default is
0):
MOV ESI, APIC_ID; address of local APIC ID register
MOV EAX, [ESI]
AND EAX, 0F000000H; zero out all other bits except APIC ID
MOV BOOT_ID, EAX; save in memory

Save the ID in the configuration RAM (optional).

7. Determine APIC ID of the secondary processor and save it in the configuration RAM
(optional).
MOV EAX, BOOT_ID
XOR EAX, 100000H; toggle lower bit of ID field (bit 24)
MOV SECOND_ID, EAX

8. Convert the base address of the 4-KByte page for the secondary processor’s bootup code
into 8-bit vector. The 8-bit vector defines the address of a 4-KByte page in the real-address
mode address space (1-MByte space). For example, a vector of 0BDH specifies a start-up
memory address of 000BD000H.
Use steps 9 and 10 to use the LVT APIC error handling entry to deal with unsuccessful
delivery of the start-up IPI.
9. Enable the local APIC by writing to spurious vector register (SVR). This is required to do
APIC error handling via the local vector table.
MOV ESI, SVR ; address of SVR
MOV EAX, [ESI]
OR EAX, APIC_ENABLED; set bit 8 to enable (0 on reset)
MOV [ESI], EAX

C-2
 DUAL-PROCESSOR (DP) BOOTUP SEQUENCE EXAMPLE (SPECIFIC

10. Program LVT3 (APIC error interrupt vector) of the local vector table with an 8-bit vector
for handling APIC errors.
MOV ESI, LVT3
MOV EAX, [ESI]
AND EAX, FFFFFF00H; clear out previous vector
OR EAX, 000000xxH; xx is the 8-bit vector for APIC error
; handling.
MOV [ESI], EAX

11. Write APIC ICRH with address of the secondary processor’s APIC.

MOV ESI,
ICR_HI ; address of ICR high dword
MOV EAX,
[ESI] ; get high word of ICR
AND EAX,
0F0FFFFFFH; zero out ID Bits
OR EAX,
SECOND_ID; write ID into appropriate bits - don’t
; affect reserved bits
MOV [ESI], SECOND_ID; write upgrade ID to destination field

12. Set the timer with an appropriate value (~100 milliseconds).

13. Write APIC ICRL to send a start-up IPI message to the secondary processor via the APIC.
MOV ESI,
ICR_LOW; write address of ICR low dword
MOV EAX,
[ESI] ; get low dword of ICR
AND EAX,
0FFF0F800H; zero out delivery mode and vector fields
OR EAX,
000006xxH; 6 selects delivery mode 110 (StartUp IPI)
; xx should be vector of 4kb page as
; computed in Step 8.
MOV [ESI], EAX

14. Configure the APIC as appropriate.

C-3
C.2. SECONDARY PROCESSOR’S SEQUENCE OF EVENTS
FOLLOWING RECEIPT OF START-UP IPI
If the secondary processor’s APIC is to be used for symmetric multiprocessing, the secondary
processor must undertake the following steps:
1. Switch to protected mode to access the APIC addresses.
2. Initialize its local APIC by writing to bit 8 of the SVR register and programming its LVT3
for error handling.
3. Configure the APIC as appropriate.
4. Enable interrupts.
5. (Optional.) Execute the CPUID instruction and write the results into the configuration
RAM.
6. Do either of the following:
— Execute a HALT instruction and wait for an IPI from the operating system.
— Continue execution.
 D
Multiple-Processor
(MP) Bootup
Sequence Example
(Specific to P6 Family Processors)
 APPENDIX D
MULTIPLE-PROCESSOR (MP) BOOTUP
SEQUENCE EXAMPLE (SPECIFIC TO P6 FAMILY
PROCESSORS)

The following example illustrates the use of the MP protocol to boot two P6 family processors
in a multiple-processor (MP) system and initialize their APICs. The primary processor (the pro-
cessor that won the “race for the flag”) is called the boot strap processor (BSP) and the second-
ary processor is called the application processor (AP).
The following constants and data definitions are used in the accompanying code examples. They
are based on the addresses of the APIC registers as defined in Table 7-1 in Chapter 7.
ICR_LOW EQU 0FEE00300H
ICR_HI EQU 0FEE00310H
SVR EQU 0FEE000F0H
APIC_ID EQU 0FEE00020H
LVT3 EQU 0FEE00370H
APIC_ENABLED EQU 100H
BOOT_ID DW ?
SECOND_ID DW ?

D.1. BSP’S SEQUENCE OF EVENTS

1. The BSP boots at the Intel Architecture address and executes until it is ready to activate the
AP.
2. Initialization software should execute the CPUID instruction to determine if the BSP is a
“GenuineIntel.” The values of EAX and EDX should be saved into a configuration RAM
space for use later.
3. The following operation can be used to detect the AP:
Set a timer before sending the start-up IPI to the AP. In the AP’s initialization routine, it
should write a value into memory indicating its presence. The BSP can then use the timer
expiration to check if something has been written into memory. If the timer expires and
nothing has been written into memory, the AP is not present or some error has occurred.
4. Load start-up code for the AP to execute into a 4-KByte page in the lower 1 MByte of
memory.

D-1
MULTIPLE-PROCESSOR (MP) BOOTUP SEQUENCE EXAMPLE

5. Switch to protected mode (to access APIC address space above 1 MByte) or change the
APIC base to less than 1 MByte and insure it is mapped to an uncached (UC) memory
type.
6. Determine the BSP’s APIC ID from the local APIC ID register (default is 0):
MOV ESI, APIC_ID; address of local APIC ID register
MOV EAX, [ESI]
AND EAX, 0F000000H; zero out all other bits except APIC ID
MOV BOOT_ID, EAX; save in memory

Save the ID in the configuration RAM (optional).

7. Determine APIC ID of the AP and save it in the configuration RAM (optional).
MOV EAX, BOOT_ID
XOR EAX, 100000H; toggle lower bit of ID field (bit 24)
MOV SECOND_ID, EAX

8. Convert the base address of the 4-KByte page for the AP’s bootup code into 8-bit vector.
The 8-bit vector defines the address of a 4-KByte page in the real-address mode address
space (1-MByte space). For example, a vector of 0BDH specifies a start-up memory
address of 000BD000H.
Use steps 9 and 10 to use the LVT APIC error handling entry to deal with unsuccessful
delivery of the start-up IPI.
9. Enable the local APIC by writing to spurious vector register (SVR). This is required to do
APIC error handling via the local vector table.
MOV ESI, SVR ; address of SVR
MOV EAX, [ESI]
OR EAX, APIC_ENABLED; set bit 8 to enable (0 on reset)
MOV [ESI], EAX

10. Program LVT3 (APIC error interrupt vector) of the local vector table with an 8-bit vector
for handling APIC errors.
MOV ESI, LVT3
MOV EAX, [ESI]
AND EAX, FFFFFF00H; clear out previous vector
OR EAX, 000000xxH; xx is the 8-bit vector for APIC error
; handling.
MOV [ESI], EAX

11. Write APIC ICRH with address of the AP’s APIC.

MOV ESI,
ICR_HI ; address of ICR high dword
MOV EAX,
[ESI] ; get high word of ICR
AND EAX,
0F0FFFFFFH; zero out ID Bits
OR EAX,
SECOND_ID; write ID into appropriate bits - don’t
; affect reserved bits
MOV [ESI], SECOND_ID; write upgrade ID to destination field

D-2
 MULTIPLE-PROCESSOR (MP) BOOTUP SEQUENCE EXAMPLE

12. Initialize the memory location into which the AP will write to signal it’s presence.
13. Set the timer with an appropriate value (~100 milliseconds).
14. Write APIC ICRL to send a start-up IPI message to the AP via the APIC.
MOV ESI,
ICR_LOW; write address of ICR low dword
MOV EAX,
[ESI] ; get low dword of ICR
AND EAX,
0FFF0F800H; zero out delivery mode and vector fields
OR EAX,
000006xxH; 6 selects delivery mode 110 (StartUp IPI)
; xx should be vector of 4kb page as
; computed in Step 8.
MOV [ESI], EAX

15. Wait for the timer interrupt or an AP signal appearing in memory.

16. If necessary, reconfigure the APIC and continue with the remaining system diagnostics as
appropriate.

D.2. AP’S SEQUENCE OF EVENTS FOLLOWING RECEIPT OF

START-UP IPI
If the AP’s APIC is to be used for symmetric multiprocessing, the AP must undertake the fol-
lowing steps:
1. Switch to protected mode to access the APIC addresses.
2. Initialize its local APIC by writing to bit 8 of the SVR register and programming its LVT3
for error handling.
3. Configure the APIC as appropriate.
4. Enable interrupts.
5. (Optional) Execute the CPUID instruction and write the results into the configuration
RAM.
6. Write into the memory location that is being used to signal to the BSP that the AP is
executing.
7. Do either of the following:
— Continue execution (that is, self-configuration, MP Specification Configuration table
completion).
— Execute a HLT instruction and wait for an IPI from the operating system.

D-3
 E
Programming the
LINT0 and LINT1
Inputs
 APPENDIX E
PROGRAMMING THE LINT0 AND LINT1 INPUTS

The following procedure describes how to program the LINT0 and LINT1 local APIC pins on
a processor after multiple processors have been booted and initialized (as described in Appendix
C and Appendix D). In this example, LINT0 is programmed to be the ExtINT pin and LINT1 is
programmed to be the NMI pin.

E.1. CONSTANTS
The following constants are defined:
LVT1 EQU 0FEE00350H
LVT2 EQU 0FEE00360H
LVT3 EQU 0FEE00370H
SVR EQU 0FEE000F0H

E.2. LINT[0:1] PINS PROGRAMMING PROCEDURE

Use the following to program the LINT[1:0] pins:
1. Mask 8259 interrupts.
2. Enable APIC via SVR (spurious vector register) if not already enabled.
MOV ESI, SVR ; address of SVR
MOV EAX, [ESI]
OR EAX, APIC_ENABLED; set bit 8 to enable (0 on reset)
MOV [ESI], EAX

3. Program LVT1 as an ExtINT which delivers the signal to the INTR signal of all processors
cores listed in the destination as an interrupt that originated in an externally connected
interrupt controller.
MOV ESI,
LVT1
MOV EAX,
[ESI]
AND EAX,
0FFFE58FFH; mask off bits 8-10, 12, 14 and 16
OR EAX,
700H ; Bit 16=0 for not masked, Bit 15=0 for edge
; triggered, Bit 13=0 for high active input
; polarity, Bits 8-10 are 111b for ExtINT
MOV [ESI], EAX ; Write to LVT1

E-1
PROGRAMMING THE LINT0 AND LINT1 INPUTS

4. Program LVT2 as NMI, which delivers the signal on the NMI signal of all processor cores
listed in the destination.
MOV ESI,
LVT2
MOV EAX,
[ESI]
AND EAX,
0FFFE58FFH; mask off bits 8-10 and 15
OR EAX,
000000400H; Bit 16=0 for not masked, Bit 15=0 edge
; triggered, Bit 13=0 for high active input
; polarity, Bits 8-10 are 100b for NMI
MOV [ESI], EAX ; Write to LVT2
;Unmask 8259 interrupts and allow NMI.

E-2
 INDEX

Numerics bus message format . . . . . . . . . . . . . . . . . 7-37

16-bit code, mixing with 32-bit code. . . . . . . . .17-1 description of. . . . . . . . . . . . . . . . . . . . . . . 7-13
32-bit code, mixing with 16-bit code. . . . . . . . .17-1 diagram of . . . . . . . . . . . . . . . . . . . . . . . . . 7-14
8086 EOI message format . . . . . . . . . . . . . . . . . 7-37
emulation, support for . . . . . . . . . . . . . . . . .16-1 nonfocused lowest priority message . . . . . 7-38
processor, exceptions and interrupts . . . . .16-8 short message format . . . . . . . . . . . . . . . . 7-37
8086/8088 processor . . . . . . . . . . . . . . . . . . . .18-7 SMI message . . . . . . . . . . . . . . . . . . . . . . 12-2
8087 math coprocessor . . . . . . . . . . . . . . . . . .18-7 status cycles . . . . . . . . . . . . . . . . . . . . . . . 7-40
82489DX, software visible differences between the structure of . . . . . . . . . . . . . . . . . . . . . . . . 7-14
local APIC on a Pentium Pro processor APIC (see also I/O APIC or Loal APIC)
and the 82489DX . . . . . . . . . . . . . . .7-44 APIC_BASE_MSR . . . . . . . . . . . . . . . . . . . . . 7-19
APR (arbitration priority register), local APIC . 7-32
Arbitration
A APIC bus . . . . . . . . . . . . . . . . . . . . . . . . . . 7-36
A (accessed) flag, page-table entry . . . . . . . . .3-27 priority, local APIC. . . . . . . . . . . . . . . . . . . 7-22
A20M# signal . . . . . . . . . . . . . . . . . . . . 16-3, 18-35 ARPL instruction. . . . . . . . . . . . . . . . . . .2-20, 4-30
Aborts Atomic operations
description of . . . . . . . . . . . . . . . . . . . . . . . .5-5 automatic bus locking . . . . . . . . . . . . . . . . . 7-3
restarting a program or task after . . . . . . . . .5-7 effects of a locked operation on internal
AC (alignment check) flag, processor caches. . . . . . . . . . . . . . . . . . 7-6
EFLAGS register . . . . . . 2-9, 5-50, 18-6 guaranteed, description of. . . . . . . . . . . . . . 7-2
Access rights overview of . . . . . . . . . . . . . . . . . . . . . .7-2, 7-3
checking . . . . . . . . . . . . . . . . . . . . . . . . . . .2-20 software-controlled bus locking. . . . . . . . . . 7-4
checking caller privileges . . . . . . . . . . . . . .4-28 Auto HALT restart
description of . . . . . . . . . . . . . . . . . . . . . . .4-26 field, SMM . . . . . . . . . . . . . . . . . . . . . . . . 12-13
invalid values . . . . . . . . . . . . . . . . . . . . . .18-24 SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-13
ADC instruction . . . . . . . . . . . . . . . . . . . . . . . . .7-4 Automatic bus locking. . . . . . . . . . . . . . . . . . . . 7-3
ADD instruction . . . . . . . . . . . . . . . . . . . . . . . . .7-4
Address
size prefix . . . . . . . . . . . . . . . . . . . . . . . . . .17-2
B
space, of task . . . . . . . . . . . . . . . . . . . . . . .6-17 B (busy) flag, TSS descriptor . 6-7, 6-12, 6-16, 7-3
Address translation B (default stack size) flag, segment descriptor . . .
2-MByte pages . . . . . . . . . . . . . . . . . . . . . .3-32 17-2, 18-34
4-KByte pages . . . . . . . . . . . . . . . . . 3-20, 3-30 B0-B3 (breakpoint condition detected) flags,
4-MByte pages . . . . . . . . . . . . . . . . . . . . . .3-21 DR6 register . . . . . . . . . . . . . . . . . . 15-4
in real-address mode . . . . . . . . . . . . . . . . .16-3 Backlink (see Previous task link)
logical to linear . . . . . . . . . . . . . . . . . . . . . . .3-7 Base address fields, segment descriptor . . . . 3-11
overview . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-6 BD (debug register access detected) flag,
Addressing, segments . . . . . . . . . . . . . . . . . . . .1-7 DR6 register . . . . . . . . . . . . .15-4, 15-10
Advanced programmable interrupt controller Binary numbers . . . . . . . . . . . . . . . . . . . . . . . . 1-7
(see APIC, I/O APIC, or Loal APIC) BINIT# signal . . . . . . . . . . . . . . . . . . . . . . . . . 2-22
Alignment Bit order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6
alignment check exception . . . . . . . . . . . . .5-50 BOUND instruction . . . . . . . . . . . . . . . . . .5-3, 5-27
checking . . . . . . . . . . . . . . . . . . . . . . . . . . .4-30 BOUND range exceeded exception (#BR) . . . 5-27
exception . . . . . . . . . . . . . . . . . . . . . . . . .18-13 BP0#, BP1#, BP2#, and BP3# pins . . . . . . . 15-12
Alignment check exception (#AC) . . . 5-50, 18-13, Breakpoint exception (#BP) 5-3, 5-25, 15-1, 15-11
18-26 Breakpoints
AM (alignment mask) flag, breakpoint exception (#BP). . . . . . . . . . . . 15-1
CR0 control register . . . . . . . 2-14, 18-22 data breakpoint . . . . . . . . . . . . . . . . . . . . . 15-7
AND instruction . . . . . . . . . . . . . . . . . . . . . . . . .7-4 data breakpoint exception conditions . . . . 15-9
APIC Base field, APIC_BASE_MSR . . . . . . . .7-19 description of. . . . . . . . . . . . . . . . . . . . . . . 15-1
APIC bus DR0-DR3 debug registers. . . . . . . . . . . . . 15-4
arbitration mechanism and protocol . . . . . .7-36 example. . . . . . . . . . . . . . . . . . . . . . . . . . . 15-7
bus arbitration . . . . . . . . . . . . . . . . . . . . . . .7-15

INDEX-1
INDEX

exception . . . . . . . . . . . . . . . . . . . . . . . . . .5-25 multiple-processor considerations. . . . . . . 9-31

field recognition. . . . . . . . . . . . . . . . . . . . . .15-6 NW flag, CR0 control register . . . . .9-12, 18-23
general-detect exception condition . . . . . .15-10 operating modes . . . . . . . . . . . . . . . . . . . . 9-11
instruction breakpoint . . . . . . . . . . . . . . . . .15-7 overview of . . . . . . . . . . . . . . . . . . . . . . . . . 9-1
instruction breakpoint exception condition .15-8 PCD flag, CR3 control register . . . . . . . . . 9-12
I/O breakpoint exception conditions . . . . . .15-9 PCD flag, page-directory entries . . . 9-12, 9-13,
LEN0 - LEN3 (Length) fields, DR7 register.15-6 9-32
R/W0-R/W3 (read/write) fields, PCD flag, page-table entries . . 9-12, 9-13, 9-32
DR7 register . . . . . . . . . . . . . . . . . . . . .15-6 precedence of controls . . . . . . . . . . . . . . . 9-13
single-step exception condition. . . . . . . . .15-10 preventing caching . . . . . . . . . . . . . . . . . . 9-14
task-switch exception condition . . . . . . . .15-11 protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-9
BS (single step) flag, DR6 register. . . . . . . . . .15-5 PWT flag, CR3 control register . . . . . . . . . 9-12
BSP (bootstrap processor) flag, PWT flag, page-directory entries. . . .9-12, 9-32
APIC_BASE_MSR . . . . . . . . . . . . . .7-19 PWT flag, page-table entries. . . . . . .9-12, 9-32
BSWAP instruction. . . . . . . . . . . . . . . . . . . . . .18-5 remapping memory types . . . . . . . . . . . . . 9-27
BT (task switch) flag, DR6 register. . . . 15-5, 15-11 setting up memory ranges with MTRRs . . 9-21
BTC instruction. . . . . . . . . . . . . . . . . . . . . . . . . .7-4 variable-range MTRRs . . . . . . . . . . . . . . . 9-23
BTF (single-step on branches) flag, Caches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6
DebugCtlMSR register . . . . 15-12, 15-14 cache hit . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-5
BTR instruction. . . . . . . . . . . . . . . . . . . . . . . . . .7-4 cache line . . . . . . . . . . . . . . . . . . . . . . . . . . 9-4
BTS instruction. . . . . . . . . . . . . . . . . . . . . . . . . .7-4 cache line fill . . . . . . . . . . . . . . . . . . . . . . . . 9-5
Built-in self-test (BIST) cache write hit . . . . . . . . . . . . . . . . . . . . . . . 9-5
description of . . . . . . . . . . . . . . . . . . . . . . . .8-1 description of. . . . . . . . . . . . . . . . . . . . . . . . 9-1
performing. . . . . . . . . . . . . . . . . . . . . . . . . . .8-2 effects of a locked operation on internal
Bus processor caches. . . . . . . . . . . . . . . . . . 7-6
arbitration, APIC bus. . . . . . . . . . . . . . . . . .7-15 enabling. . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-8
errors, detected with machine-check management, instructions . . . . . . . . . . . . . 2-21
architecture . . . . . . . . . . . . . . . . . . . . .13-11 Caching
hold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-37 cache control protocol . . . . . . . . . . . . . . . . . 9-9
locking. . . . . . . . . . . . . . . . . . . . . . . . 7-3, 18-37 cache line . . . . . . . . . . . . . . . . . . . . . . . . . . 9-4
Byte order. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-6 cache mechanisms in Intel Architecture
processors . . . . . . . . . . . . . . . . . . . . . 18-30
caching terminology . . . . . . . . . . . . . . . . . . 9-4
C choosing a memory type . . . . . . . . . . . . . . . 9-8
C (conforming) flag, segment descriptor . . . . .4-13 flushing TLBs . . . . . . . . . . . . . . . . . . . . . . 9-17
C1 flag, FPU status word . . . . . . . . . . . 18-9 , 18-18 implicit caching . . . . . . . . . . . . . . . . . . . . . 9-16
C2 flag, FPU status word . . . . . . . . . . . . . . . . .18-9 internal caches . . . . . . . . . . . . . . . . . . . . . . 9-1
Cache control . . . . . . . . . . . . . . . . . . . . . . . . . .9-18 L1 (level 1) cache . . . . . . . . . . . . . . . . . . . . 9-2
cache management instructions . . . . . . . . .9-15 L2 (level 2) cache . . . . . . . . . . . . . . . . . . . . 9-2
cache mechanisms in Intel Architecture methods of caching available . . . . . . . . . . . 9-5
processors. . . . . . . . . . . . . . . . . . . . . .18-30 MTRRs, description of. . . . . . . . . . . . . . . . 9-18
caching terminology . . . . . . . . . . . . . . . . . . .9-4 operating modes . . . . . . . . . . . . . . . . . . . . 9-11
CD flag, CR0 control register . . . . . . 9-9, 18-23 overview of . . . . . . . . . . . . . . . . . . . . . . . . . 9-1
choosing a memory type. . . . . . . . . . . . . . . .9-8 self-modifying code, effect on . . . . .9-15, 18-31
fixed-range MTRRs. . . . . . . . . . . . . . . . . . .9-22 snooping . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-5
flags and fields . . . . . . . . . . . . . . . . . . . . . . .9-9 TLBs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-4
flushing TLBs . . . . . . . . . . . . . . . . . . . . . . .9-17 UC (uncacheable) memory type . . . . . . . . . 9-5
G (global) flag, page-directory entries . . . 9-12, WB (write back) memory type . . . . . . . . . . . 9-6
9-17 WC (write combining) memory type . . . . . . 9-6
G (global) flag, page-table entries . . 9-12, 9-17 WP (write protected) memory type . . . . . . . 9-7
internal caches . . . . . . . . . . . . . . . . . . . . . . .9-1 write buffer . . . . . . . . . . . . . . . . . . . . .9-4, 9-17
MemTypeGet() function . . . . . . . . . . . . . . .9-28 write-back caching . . . . . . . . . . . . . . . . . . . 9-5
MemTypeSet() function . . . . . . . . . . . . . . .9-29 WT (write through) memory type. . . . . . . . . 9-6
MESI protocol . . . . . . . . . . . . . . . . . . . . 9-4, 9-9 Call gates
methods of caching available . . . . . . . . . . . .9-5 16-bit, interlevel return from . . . . . . . . . . 18-34
MTRR initialization . . . . . . . . . . . . . . . . . . .9-27 accessing a code segment through . . . . . 4-17
MTRR precedences . . . . . . . . . . . . . . . . . .9-26 description of. . . . . . . . . . . . . . . . . . . . . . . 4-16
MTRRs, description of . . . . . . . . . . . . . . . .9-18 for 16-bit and 32-bit code modules . . . . . . 17-2

INDEX-2
 INDEX

introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-3 guidelines, floating-point code. . . . . . . . . . 14-2

mechanism . . . . . . . . . . . . . . . . . . . . . . . . .4-18 guidelines, general . . . . . . . . . . . . . . . . . . 14-1
privilege level checking rules . . . . . . . . . . .4-19 guidelines, MMX code. . . . . . . . . . . .14-2, 14-3
CALL instruction. 3-9, 4-12, 4-13, 4-17, 4-23, 6-3, instruction length . . . . . . . . . . . . . . . . . . . 14-30
6-10, 6-12, 17-7 instruction pairing, general integer-instruction
Caller access privileges, checking . . . . . . . . . .4-28 pairability rules. . . . . . . . . . . . . . . . . . 14-14
Calls instruction pairing, general rules . . . . . . . 14-12
between 16- and 32-bit code segments . . .17-4 instruction pairing, guidelines . . . . . . . . . 14-12
controlling the operand-size attribute instruction pairing, integer pairing rules . 14-13
for a call. . . . . . . . . . . . . . . . . . . . . . . . .17-7 instruction pairing, MMX instruction pairing
returning from . . . . . . . . . . . . . . . . . . . . . . .4-23 guidelines. . . . . . . . . . . . . . . . . . . . . . 14-17
CC0 and CC1 (counter control) fields, CESR MSR instruction pairing, pairing MMX and integer
(Pentium processor). . . . . . . . . . . .15-20 instructions. . . . . . . . . . . . . . . .14-17, 14-18
CD (cache disable) flag, CR0 control register 2-13, instruction pairing, pairing two MMX
8-8, 9-9, 9-11, 9-13, 9-14, 9-31, 9-32, instructions. . . . . . . . . . . . . . . . . . . . . 14-17
18-22, 18-23, 18-30 instruction pairing, restrictions on pair execution
CESR (control and event select) MSR (Pentium 14-16
processor) . . . . . . . . . . . . . . . . . . .15-20 instruction pairing, special pairs . . . . . . . 14-16
CLI instruction . . . . . . . . . . . . . . . . . . . . . . . . . .5-9 instruction pairing, unpairability due to
CLTS instruction. . . . . . . . . . . . . . . . . . . 2-20, 4-25 register dependencies . . . . . . . . . . . . 14-15
Cluster model, local APIC . . . . . . . . . . . . . . . .7-21 instruction scheduling, overview . . . . . . . 14-12
CMOVcc instructions . . . . . . . . . . . . . . . . . . . .18-3 integer divide. . . . . . . . . . . . . . . . . . . . . . 14-34
CMPXCHG instruction . . . . . . . . . . . . . . . 7-4, 18-5 integer instruction selection and
CMPXCHG8B instruction . . . . . . . . . . . . . 7-4, 18-4 optimizations . . . . . . . . . . . . . . . . . . . 14-32
Code modules LEA instruction . . . . . . . . . . . . . . . . . . . . 14-32
16 bit vs. 32 bit . . . . . . . . . . . . . . . . . . . . . .17-2 partial register stalls, reducing . . . . . . . . . 14-7
mixing 16-bit and 32-bit code . . . . . . . . . . .17-1 pipelining, floating-point instructions . . . . 14-18
sharing data among mixed-size code pipelining, floating-point operations with integer
segments. . . . . . . . . . . . . . . . . . . . . . . .17-3 operands . . . . . . . . . . . . . . . . . . . . . . 14-21
transferring control among mixed-size code pipelining, FSTSW instruction . . . . . . . . . 14-21
segments. . . . . . . . . . . . . . . . . . . . . . . .17-4 pipelining, FXCH guidelines . . . . . . . . . . 14-22
Code optimization pipelining, guidelines. . . . . . . . . . . . . . . . 14-18
8/16 bit operands . . . . . . . . . . . . . . . . . . .14-33 pipelining, hiding the one-clock latency of a
accessing memory . . . . . . . . . . . . . . . . . .14-24 floating-point store . . . . . . . . . . . . . . . 14-20
accessing memory, using MMX pipelining, integer and floating-point
instructions . . . . . . . . . . . . . . . 14-24, 14-25 multiply. . . . . . . . . . . . . . . . . . . . . . . . 14-21
accessing memory, write allocation pipelining, MMX instructions . . . . . . . . . . 14-18
effects . . . . . . . . . . . . . . . . . . . . . . . . .14-27 pipelining, pairing of floating-point
address calculations . . . . . . . . . . . . . . . . .14-34 instructions. . . . . . . . . . . . . . . . . . . . . 14-19
addressing modes and register usage . . .14-29 pipelining, transcendental instructions . . 14-22
alignment, code . . . . . . . . . . . . . . . . . . . . .14-9 pipelining, using integer instructions to hide
alignment, data . . . . . . . . . . . . . . . . . . . . . .14-9 latencies and schedule floating-point
alignment, data structures and arrays . . .14-10 instructions. . . . . . . . . . . . . . . . . . . . . 14-19
alignment, dynamic allocation using prefixed opcodes. . . . . . . . . . . . . . . . . . . 14-31
malloc . . . . . . . . . . . . . . . . . . . . . . . . .14-11 prolog sequences . . . . . . . . . . . . . . . . . . 14-34
alignment, memory and stack. . . . . . . . . .14-10 PUSH mem instruction . . . . . . . . . . . . . . 14-33
alignment, of static variables . . . . . . . . . .14-10 scheduling, rules for Pentium II and Pentium Pro
alignment, penalties . . . . . . . . . . . . . . . . . .14-9 processors . . . . . . . . . . . . . . . . . . . . . 14-22
alignment, rules and guidelines . . . . . . . . .14-9 short opcodes . . . . . . . . . . . . . . . . . . . . . 14-33
alignment, using in-line assembly code . .14-11 zero-extension of short integers . . . . . . . 14-32
branch prediction, eliminating and reducing Code optimizations
number of branches . . . . . . . . . . . . . . .14-5 compares . . . . . . . . . . . . . . . . . . . . . . . . 14-34
branch prediction, optimization . . . . . 14-4, 14-5 Code segments
branch prediction, rules . . . . . . . . . . . . . . .14-4 accessing data in . . . . . . . . . . . . . . . . . . . 4-12
clearing a register . . . . . . . . . . . . . . . . . . .14-34 accessing through a call gate . . . . . . . . . . 4-17
compares with immediate zero . . . . . . . . .14-35 description of. . . . . . . . . . . . . . . . . . . . . . . 3-13
complex instructions . . . . . . . . . . . . . . . . .14-32 descriptor format . . . . . . . . . . . . . . . . . . . . . 4-3
epilog sequence . . . . . . . . . . . . . . . . . . . .14-35 descriptor layout . . . . . . . . . . . . . . . . . . . . . 4-3

INDEX-3
INDEX

direct calls or jumps to . . . . . . . . . . . . . . . .4-13 CS segment selector, CPL field . . . . . . . . . . . . 4-3

executable (defined) . . . . . . . . . . . . . . . . . .3-12 CTR0 and CTR1 (performance counters) MSRs
pointer size . . . . . . . . . . . . . . . . . . . . . . . . .17-5 (Pentium processor) . . . . . .15-20, 15-22
privilege level checking when transferring Current privilege level (see CPL)
program control between code Current-count register, local APIC . . . . . . . . . 7-44
segments. . . . . . . . . . . . . . . . . . . . . . . .4-12
Compatibility
Intel Architecture. . . . . . . . . . . . . . . . . . . . .18-1 D
software . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-6 D (default operation size) flag,
Condition code flags, FPU status word segment descriptor. . . . . . . .17-2, 18-34
compatibility information . . . . . . . . . . . . . . .18-8 D (dirty) flag, page-table entry . . . . . . . . . . . . 3-27
Conforming code segments Data breakpoint exception conditions. . . . . . . 15-9
accessing . . . . . . . . . . . . . . . . . . . . . . . . . .4-15 Data segments
C (conforming) flag . . . . . . . . . . . . . . . . . . .4-13 description of. . . . . . . . . . . . . . . . . . . . . . . 3-13
description of . . . . . . . . . . . . . . . . . . . . . . .3-14 descriptor layout . . . . . . . . . . . . . . . . . . . . . 4-3
Context, task (see Task state) expand-down type. . . . . . . . . . . . . . . . . . . 3-12
Control registers privilege level checking when accessing. . . 4-9
CR0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-12 DB0-DB3 breakpoint-address registers . . . . . 15-1
CR1 (reserved) . . . . . . . . . . . . . . . . . . . . . .2-12 DB6 debug status register . . . . . . . . . . . . . . . 15-1
CR2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-12 DB7 debug control register. . . . . . . . . . . . . . . 15-1
CR3 (PDBR) . . . . . . . . . . . . . . . . . . . . 2-5, 2-12 DE (debugging extensions) flag, CR4 control
CR4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-12 register . . . . . 2-17, 18-22, 18-24, 18-25
description of . . . . . . . . . . . . . . . . . . . . . . .2-12 DE (denormal operand exception) flag, FPU
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 status word . . . . . . . . . . . . .11-17, 11-19
qualification of flags with CPUID Debug exception (#DB) 5-9, 5-23, 6-6, 15-1, 15-8,
instruction . . . . . . . . . . . . . . . . . . . . . . .2-18 15-13
Coprocessor segment overrun exception . . . 5-34, Debug registers
18-14 description of. . . . . . . . . . . . . . . . . . . . . . . 15-2
Counter mask field, PerfEvtSel0 and PerfEvtSel1 introduction to . . . . . . . . . . . . . . . . . . . . . . . 2-5
MSRs (P6 family processors) . . . .15-17 loading. . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-21
CPL DebugCtlMSR register . . . . . . . . . . . . .15-1, 15-11
description of . . . . . . . . . . . . . . . . . . . . . . . .4-8 Debugging facilities
field, CS segment selector . . . . . . . . . . . . . .4-3 debug registers . . . . . . . . . . . . . . . . . . . . . 15-2
CPUID instruction. . 2-18, 7-12, 9-20, 13-7, 15-14, exceptions . . . . . . . . . . . . . . . . . . . . . . . . . 15-7
15-19, 18-2, 18-4, 18-38 last branch, interrupt, and exception
CR0 control register . . . . . . . . . . . . . . . . . . . . .18-8 recording . . . . . . . . . . . . . . . . . . . . . . 15-11
description of . . . . . . . . . . . . . . . . . . . . . . .2-12 masking debug exceptions . . . . . . . . . . . . . 5-9
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 overview of . . . . . . . . . . . . . . . . . . . . . . . . 15-1
state following processor reset . . . . . . . . . . .8-2 performance-monitoring counters . . . . . . 15-15
CR1 control register (reserved) . . . . . . . . . . . .2-12 time-stamp counter . . . . . . . . . . . . . . . . . 15-14
CR2 control register DEC instruction. . . . . . . . . . . . . . . . . . . . . . . . . 7-4
description of . . . . . . . . . . . . . . . . . . . . . . .2-12 Denormal operand exception (#D) . . .11-19, 18-11
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 Denormalized operand . . . . . . . . . . . . . . . . . 18-15
CR3 control register (PDBR) Device-not-available exception (#NM) . . 5-30, 8-8,
associated with a task. . . . . . . . . . . . . . 6-1, 6-3 18-13, 18-14
description of . . . . . . . . . . . . . . . . . . 2-12, 3-23 DFR (destination format register), local APIC 7-21
in TSS . . . . . . . . . . . . . . . . . . . . . . . . . 6-6, 6-17 DIV instruction . . . . . . . . . . . . . . . . . . . . . . . . 5-22
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 Divide configuration register, local APIC . . . . 7-43
loading during initialization . . . . . . . . . . . . .8-13 Divide-error exception (#DE) . . . . . . . .5-22, 18-26
memory management. . . . . . . . . . . . . . . . . .2-5 Division-by-zero exception (#Z) . . . . . . . . . . 11-18
CR4 control register . . . . . . . . . . . . . . . . . . . . .18-2 Double-fault exception (#DF) . . . . . . . .5-32, 18-28
description of . . . . . . . . . . . . . . . . . . . . . . .2-12 DPL (descriptor privilege level) field, segment
inclusion in Intel Architecture . . . . . . . . . .18-21 descriptor . . . . . . . . . . . . . 3-12, 4-2, 4-8
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 DR0-DR3 breakpoint-address registers . . . . . 15-4,
CS register . . . . . . . . . . . . . . . . . . . . . . . . . . .18-12 15-12, 15-13
saving on call to exception or interrupt DR4-DR5 debug registers . . . . . . . . . .15-4, 18-25
handler . . . . . . . . . . . . . . . . . . . . . . . . .5-15 DR6 debug status register . . . . . . . . . . . . . . . 15-4
state following initialization . . . . . . . . . . . . . .8-6

INDEX-4
 INDEX

B0-B3 (breakpoint condition detected) Event select field, PerfEvtSel0 and PerfEvtSel1
flags. . . . . . . . . . . . . . . . . . . . . . . . . . . .15-4 MSRs (P6 family processors) . . . . 15-16
BD (debug register access detected) flag. .15-4 Exception handler
BS (single step) flag . . . . . . . . . . . . . . . . . .15-5 calling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
BT (task switch) flag . . . . . . . . . . . . . . . . . .15-5 defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
debug exception (#DB) . . . . . . . . . . . . . . . .5-23 flag usage by handler procedure. . . . . . . . 5-18
reserved bits . . . . . . . . . . . . . . . . . . . . . . .18-24 machine-check exceptions (#MC). . . . . . 13-14
DR7 debug control register . . . . . . . . . . . . . . .15-5 procedures . . . . . . . . . . . . . . . . . . . . . . . . 5-15
G0-G3 (global breakpoint enable) flags . . .15-5 protection of handler procedures . . . . . . . 5-17
GD (general detect enable) flag . . . . . . . . .15-5 task . . . . . . . . . . . . . . . . . . . . . . . . . . .5-18, 6-3
GE (global exact breakpoint enable) flag . .15-5 Exception priority, FPU exceptions. . .11-13, 18-12
L0-L3 (local breakpoint enable) flags . . . . .15-5 Exceptions
LE local exact breakpoint enable) flag . . . .15-5 alignment check . . . . . . . . . . . . . . . . . . . 18-13
LEN0-LEN3 (Length) fields. . . . . . . . . . . . .15-6 classifications . . . . . . . . . . . . . . . . . . . . . . . 5-4
R/W0-R/W3 (read/write) fields . . . . 15-6, 18-24 conditions checked during a task switch . . 6-13
D/B (default operation size/default stack pointer coprocessor segment overrun. . . . . . . . . 18-14
size and/or upper bound) flag, segment description of. . . . . . . . . . . . . . . . . . . . .2-4, 5-1
descriptor . . . . . . . . . . . . . . . . . 3-12, 4-5 device not available. . . . . . . . . . . . . . . . . 18-14
double fault . . . . . . . . . . . . . . . . . . . . . . . . 5-32
error code . . . . . . . . . . . . . . . . . . . . . . . . . 5-20
E floating-point error . . . . . . . . . . . . . . . . . . 18-14
E (edge detect) flag, PerfEvtSel0 and PerfEvtSel1 general protection . . . . . . . . . . . . . . . . . . 18-14
MSRs (P6 family processors) . . . .15-17 handler mechanism. . . . . . . . . . . . . . . . . . 5-15
E (enable/disable APIC) flag, handler procedures . . . . . . . . . . . . . . . . . . 5-15
APIC_BASE_MSR . . . . . . . . . . . . . .7-19 handling. . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
E (expansion direction) flag, segment handling in real-address mode . . . . . . . . . 16-6
descriptor . . . . . . . . . . . . . . . . . . 4-2, 4-5 handling in SMM . . . . . . . . . . . . . . . . . . . 12-10
E (MTRRs enabled) flag, MTRRdefType handling in virtual-8086 mode . . . . . . . . . 16-15
register . . . . . . . . . . . . . . . . . . 7-19, 9-22 handling through a task gate in virtual-8086
EFLAGS register mode . . . . . . . . . . . . . . . . . . . . . . . . . 16-20
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 handling through a trap or interrupt gate in
new flags. . . . . . . . . . . . . . . . . . . . . . . . . . .18-6 virtual-8086 mode . . . . . . . . . . . . . . . 16-17
saved in TSS . . . . . . . . . . . . . . . . . . . . . . . .6-4 IDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-11
saving on call to exception or interrupt initializing for protected-mode operation . . 8-12
handler . . . . . . . . . . . . . . . . . . . . . . . . .5-15 invalid opcode . . . . . . . . . . . . . . . . . . . . . . 18-6
using flags to distinguish between 32-bit Intel masking debug exceptions . . . . . . . . . . . . . 5-9
Architecture processors. . . . . . . . . . . . .18-6 masking when switching stack segments . 5-10
EIP register . . . . . . . . . . . . . . . . . . . . . . . . . .18-12 notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-8
saved in TSS . . . . . . . . . . . . . . . . . . . . . . . .6-4 overview of . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
saving on call to exception or interrupt priorities among simultaneous exceptions and
handler . . . . . . . . . . . . . . . . . . . . . . . . .5-15 interrupts . . . . . . . . . . . . . . . . . . . . . . . 5-10
state following initialization . . . . . . . . . . . . . .8-6 priority of . . . . . . . . . . . . . . . . . . . . . . . . . 18-27
EM (emulation) flag, CR0 control register . . . 2-15, reference information on all exceptions . . 5-21
5-30, 8-6, 8-8 restarting a task or program . . . . . . . . . . . . 5-7
EOI (end-of-interrupt register), local APIC . . . .7-33 segment not present . . . . . . . . . . . . . . . . 18-14
Error code sources of . . . . . . . . . . . . . . . . . . . . . . . . . . 5-3
exception, description of . . . . . . . . . . . . . . .5-20 summary of . . . . . . . . . . . . . . . . . . . . . . . . . 5-6
pushing on stack. . . . . . . . . . . . . . . . . . . .18-33 vectors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
Error signals . . . . . . . . . . . . . . . . . . . . 18-12, 18-13 Executable code segment, size . . . . . . . . . . . 3-12
ERROR# input . . . . . . . . . . . . . . . . . . . . . . . .18-19 Expand-down data segment type . . . . . . . . . . 3-12
ERROR# output . . . . . . . . . . . . . . . . . . . . . . .18-19 External bus errors, detected with machine-check
ES0 and ES1 (event select) fields, CESR MSR architecture. . . . . . . . . . . . . . . . . . 13-11
(Pentium processor). . . . . . .15-20, A-12
ESP register, saving on call to exception or interrupt
handler . . . . . . . . . . . . . . . . . . . . . . .5-15 F
ESR (error status register), local APIC . . . . . .7-42 F2XM1 instruction. . . . . . . . . . . . . . . . . . . . . 18-16
ET (extension type) flag, CR0 control register .2-14 Fast string operations . . . . . . . . . . . . . . . . . . . . 7-9
ET (extension type) flag, CR0 register . . . . . . .18-8

INDEX-5
INDEX

Faults instruction synchronization . . . . . . . . . . . 18-19

description of . . . . . . . . . . . . . . . . . . . . . . . .5-4 setting up for software emulation of FPU
restarting a program or task after . . . . . . . . .5-7 functions . . . . . . . . . . . . . . . . . . . . . . . . 8-8
FCMOVcc instructions . . . . . . . . . . . . . . . . . . .18-3 using in SMM . . . . . . . . . . . . . . . . . . . . . 12-11
FCOMI instruction . . . . . . . . . . . . . . . . . . . . . .18-3 FPU control word
FCOMIP instruction . . . . . . . . . . . . . . . . . . . . .18-3 compatibility, Intel Architecture processors 18-9
FCOS instruction . . . . . . . . . . . . . . . . . . . . . .18-16 RC field . . . . . . . . . . . . . . . . . . . . . . .11-3, 11-4
FDISI instruction (obsolete) . . . . . . . . . . . . . .18-18 FPU status word
FDIV instruction . . . . . . . . . . . . . . . . . 18-13, 18-15 condition code flags . . . . . . . . . . . . . . . . . 18-8
FE (fixed MTRRs enabled) flag, MTRRdefType OE flag . . . . . . . . . . . . . . . . . . . . . . . . . . 11-19
register . . . . . . . . . . . . . . . . . . . . . . .9-22 FPU tag word . . . . . . . . . . . . . . . . . . . . . . . . . 18-9
Feature determination, of processor . . . . . . . .18-2 FRSTOR instruction . . . . . . . . . . . . . .18-13, 18-14
Feature information, processor . . . . . . . . . . . .18-2 FSAVE/FNSAVE instructions . . . . . . .18-13, 18-18
FENI instruction (obsolete). . . . . . . . . . . . . . .18-18 FSCALE instruction . . . . . . . . . . . . . . . . . . . 18-15
FINIT/FNINIT instructions . . . . . . . . . . 18-8, 18-19 FSIN instruction . . . . . . . . . . . . . . . . . . . . . . 18-16
FIX (fixed range registers supported) flag, FSINCOS instruction . . . . . . . . . . . . . . . . . . 18-16
MTRRcap register . . . . . . . . . . . . . .9-21 FSQRT instruction . . . . . . . . . . . . . . .18-13, 18-15
Fixed-range MTRRs FSTENV/FNSTENV instructions . . . . . . . . . 18-18
description of . . . . . . . . . . . . . . . . . . . . . . .9-22 FTAN instruction. . . . . . . . . . . . . . . . . . . . . . . 18-9
mapping to physical memory . . . . . . . . . . .9-23 FUCOM instruction . . . . . . . . . . . . . . . . . . . . 18-15
Flat model, local APIC . . . . . . . . . . . . . . . . . . .7-21 FUCOMI instruction . . . . . . . . . . . . . . . . . . . . 18-3
Flat segmentation model . . . . . . . . . . . . . . 3-3, 3-4 FUCOMIP instruction . . . . . . . . . . . . . . . . . . . 18-3
FLD instruction . . . . . . . . . . . . . . . . . . . . . . . .18-16 FUCOMP instruction. . . . . . . . . . . . . . . . . . . 18-15
FLDENV instruction . . . . . . . . . . . . . . . . . . . .18-14 FUCOMPP instruction . . . . . . . . . . . . . . . . . 18-15
FLDL2E instruction. . . . . . . . . . . . . . . . . . . . .18-17 FWAIT instruction . . . . . . . . . . . . . . . . . . . . . . 5-30
FLDL2T instruction. . . . . . . . . . . . . . . . . . . . .18-17 FXAM instruction . . . . . . . . . . . . . . . .18-16, 18-17
FLDLG2 instruction . . . . . . . . . . . . . . . . . . . .18-17 FXTRACT instruction . . . . . . . 18-11, 18-16, 18-17
FLDLN2 instruction . . . . . . . . . . . . . . . . . . . .18-17
FLDPI instruction . . . . . . . . . . . . . . . . . . . . . .18-17
Floating-point error exception (#MF) . . 5-48, 5-53, G
18-14 G (global) flag
Floating-point exceptions page-directory entries . . . . . . . . . . . .9-12, 9-17
denormal operand exception . . . . 11-19, 18-11 page-table entries . . . . . . . . . . . . . . .9-12, 9-17
division-by-zero. . . . . . . . . . . . . . . . . . . . .11-18 page-table entry . . . . . . . . . . . . . . . . . . . . 3-27
exception conditions . . . . . . . . . . . . . . . . .11-16 G (granularity) flag, segment descriptor 3-10, 3-12,
exception priority. . . . . . . . . . . . . . . . . . . .11-13 4-2, 4-5
inexact result (precision). . . . . . . . . . . . . .11-21 G0-G3 (global breakpoint enable) flags,
invalid arithmetic operand. . . . . . . . . . . . .11-17 DR7 register . . . . . . . . . . . . . . . . . . 15-5
invalid operation . . . . . . . . . . . . . . . . . . . .18-17 Gate descriptors
numeric overflow. . . . . . . . . . . . . . 11-19, 18-11 call gates . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16
numeric underflow . . . . . . . . . . . . 11-20, 18-12 description of. . . . . . . . . . . . . . . . . . . . . . . 4-16
saved CS and EIP values . . . . . . . . . . . . .18-12 Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
software handling . . . . . . . . . . . . . . . . . . .11-15 GD (general detect enable) flag,
stack underflow. . . . . . . . . . . . . . . . . . . . .11-17 DR7 register . . . . . . . . . . . . .15-5, 15-10
FLUSH# pin . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-2 GDT
Focus processor, local APIC . . . . . . . . . . . . . .7-22 description of. . . . . . . . . . . . . . . . . . . .2-3, 3-17
FPATAN instruction . . . . . . . . . . . . . . . . . . . .18-16 index into with index field of segment
FPREM instruction . . . . . . . . . . 18-9, 18-13, 18-15 selector . . . . . . . . . . . . . . . . . . . . . . . . . 3-7
FPREM1 instruction . . . . . . . . . . . . . . . 18-9, 18-15 initializing. . . . . . . . . . . . . . . . . . . . . . . . . . 8-12
FPTAN instruction . . . . . . . . . . . . . . . . 18-9, 18-15 pointers to exception and interrupt
FPU handlers . . . . . . . . . . . . . . . . . . . . . . . . 5-15
compatibility with Intel Architecture FPUs and segment descriptors in . . . . . . . . . . . . . . . . 3-9
math coprocessors . . . . . . . . . . . . . . . .18-7 selecting with TI (table indicator) flag of segment
configuring the FPU environment . . . . . . . . .8-6 selector . . . . . . . . . . . . . . . . . . . . . . . . . 3-8
device-not-available exception . . . . . . . . . .5-30 task switching . . . . . . . . . . . . . . . . . . . . . . 6-10
error signals . . . . . . . . . . . . . . . . . 18-12, 18-13 task-gate descriptor. . . . . . . . . . . . . . . . . . . 6-8
floating-point error exception . . . . . . . . . . .5-48 TSS descriptors. . . . . . . . . . . . . . . . . . . . . . 6-6
initialization . . . . . . . . . . . . . . . . . . . . . . . . . .8-6

INDEX-6
 INDEX

use in address translation. . . . . . . . . . . . . . .3-7 IE (invalid operation exception) flag, FPU

GDTR register status word . . . . . . . . . . . . . . . . . . . 18-9
description of . . . . . . . . . . . . . . 2-3, 2-10, 3-17 IEEE 754 and 854 standards for floating-point
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 arithmetic . . . . . . . . . . . . . . .18-9, 18-10
limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-5 IF (interrupt enable) flag, EFLAGS register . . . 2-8,
loading during initialization . . . . . . . . . . . . .8-12 5-8, 5-15, 5-18, 12-10, 16-6, 16-26
storing . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-18 IN instruction. . . . . . . . . . . . . . . . . . . . .7-10, 18-36
GE (global exact breakpoint enable) flag, INC instruction . . . . . . . . . . . . . . . . . . . . . . . . . 7-4
DR7 register. . . . . . . . . . . . . 15-5, 15-10 Index field, segment selector . . . . . . . . . . . . . . 3-7
General-detect exception condition . . . . . . . .15-10 Inexact Result (Precision) Exception . . . . . . 11-21
General-protection exception (#GP) 3-14, 4-7, 4-8, Inexact result (precision) exception (#P) . . . 11-21
4-14, 4-15, 5-17, 5-41, 6-7, 15-2, 18-14, Inexact result, FPU . . . . . . . . . . . . . . . . . . . . . 11-4
18-26, 18-27, 18-35, 18-37 INIT interrupt. . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
General-purpose registers Initial-count register, local APIC . . . . . . . . . . . 7-44
saved in TSS . . . . . . . . . . . . . . . . . . . . . . . .6-4 Initialization
Global descriptor table register (see GDTR) built-in self-test (BIST). . . . . . . . . . . . . .8-1, 8-2
Global descriptor table (see GDT) CS register state following . . . . . . . . . . . . . 8-6
dual-processor (DP) bootup sequence for
Pentium processors . . . . . . . . . . . . . . . . C-1
H EIP register state following . . . . . . . . . . . . . 8-6
HALT state . . . . . . . . . . . . . . . . . . . . . . . . . . .12-13 example. . . . . . . . . . . . . . . . . . . . . . . . . . . 8-16
relationship to SMI interrupt . . . . . . . . . . . .12-3 first instruction executed . . . . . . . . . . . . . . . 8-6
Hardware reset FPU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-6
description of . . . . . . . . . . . . . . . . . . . . . . . .8-1 hardware reset . . . . . . . . . . . . . . . . . . . . . . 8-1
processor state after reset . . . . . . . . . . . . . .8-2 IDT, protected mode . . . . . . . . . . . . . . . . . 8-12
state of MTRRs following . . . . . . . . . . . . . .9-18 IDT, real-address mode . . . . . . . . . . . . . . 8-10
value of SMBASE following . . . . . . . . . . . .12-4 Intel486 SX processor and Intel 487 SX math
Hexadecimal numbers . . . . . . . . . . . . . . . . . . . .1-7 coprocessor . . . . . . . . . . . . . . . . . . . 18-20
HITM# line . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-5 local APIC . . . . . . . . . . . . . . . . . . . . . . . . . 7-35
HLT instruction . . . 2-22, 4-25, 5-33, 12-13, 12-14, location of software-initialization code. . . . . 8-6
15-15 model and stepping information . . . . . . . . . 8-5
multiple-processor (MP) bootup sequence for
I P6 family processors . . . . . . . . . . . . . . . D-1
multitasking environment . . . . . . . . . . . . . 8-13
ID (identification) flag, EFLAGS register 2-10, 18-6
overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
IDIV instruction. . . . . . . . . . . . . . . . . . . 5-22, 18-26
paging . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-12
IDT
processor state after reset . . . . . . . . . . . . . 8-2
calling interrupt- and exception-handlers
protected mode . . . . . . . . . . . . . . . . . . . . . 8-11
from . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-15
real-address mode . . . . . . . . . . . . . . . . . . 8-10
changing base and limit in real-address
RESET# pin . . . . . . . . . . . . . . . . . . . . . . . . 8-1
mode . . . . . . . . . . . . . . . . . . . . . . . . . . .16-6
setting up exception- and interrupt-handling
description of . . . . . . . . . . . . . . . . . . . . . . .5-11
facilities . . . . . . . . . . . . . . . . . . . . . . . . 8-12
handling NMI interrrupts during
INIT# pin . . . . . . . . . . . . . . . . . . . . . . . . . . .5-2, 8-2
initialization . . . . . . . . . . . . . . . . . . . . . .8-11 INIT# signal . . . . . . . . . . . . . . . . . . . . . . . . . . 2-22
initializing, for protected-mode operation . .8-12 INS instruction . . . . . . . . . . . . . . . . . . . . . . . 15-10
initializing, for real-address mode Instruction operands . . . . . . . . . . . . . . . . . . . . . 1-7
operation . . . . . . . . . . . . . . . . . . . . . . . .8-10 Instruction set
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-4 new instructions . . . . . . . . . . . . . . . . . . . . 18-3
limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-28 obsolete instructions . . . . . . . . . . . . . . . . . 18-5
structure in real-address mode . . . . . . . . . .16-7 Instruction-breakpoint exception condition . . . 15-8
task switching . . . . . . . . . . . . . . . . . . . . . . .6-10 Instructions
task-gate descriptor . . . . . . . . . . . . . . . . . . .6-8 privileged. . . . . . . . . . . . . . . . . . . . . . . . . . 4-25
types of descriptors allowed . . . . . . . . . . . .5-13 serializing . . . . . . . . . . . . . . . . . . . . . . . . 18-19
use in real-address mode . . . . . . . . . . . . . .16-6 supported in real-address mode . . . . . . . . 16-4
IDTR register system. . . . . . . . . . . . . . . . . . . . . . . . .2-6, 2-18
description of . . . . . . . . . . . . . . . . . . 2-11, 5-13 INT 3 instruction . . . . . . . . . . . . . . . . . . .5-25, 15-2
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-4
INT instruction . . . . . . . . . . . . . . . . . . . . . . . . 4-12
limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-5
INT n instruction . . . . . . . . . . . . . . . . . 3-9, 5-1, 5-3
loading in real-address mode . . . . . . . . . . .16-6
storing . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-18

INDEX-7
INT (APIC interrupt enable) flag, PerfEvtSel0 and handling through a trap or interrupt gate in
PerfEvtSel1 MSRs (P6 family processors) virtual-8086 mode . . . . . . . . . . . . . . . 16-17
15-17 IDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-11
INT3 instruction . . . . . . . . . . . . . . . . . . . . . 3-9, 5-3 IDTR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-11
Intel 287 math coprocessor . . . . . . . . . . . . . . .18-7 initializing for protected-mode operation . . 8-12
Intel 387 math coprocessor system . . . . . . . . .18-7 interrupt descriptor table register (see IDTR)
Intel 487 SX math coprocessor . . . . . . 18-7, 18-20 interrupt descriptor table (see IDT)
Intel 8086 processor. . . . . . . . . . . . . . . . . . . . .18-7 local APIC . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
Intel Architecture local APIC sources . . . . . . . . . . . . . . . . . . 7-15
compatibility . . . . . . . . . . . . . . . . . . . . . . . .18-1 maskable hardware interrupts. . . . . . .2-8, 7-23
processors . . . . . . . . . . . . . . . . . . . . . . . . .18-1 masking maskable hardware interrupts . . . 5-8
Intel286 processor . . . . . . . . . . . . . . . . . . . . . .18-7 masking when switching stack segments . 5-10
Intel386 DX processor . . . . . . . . . . . . . . . . . . .18-7 overview of . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
Intel486 DX processor . . . . . . . . . . . . . . . . . . .18-7 priorities among simultaneous exceptions and
Intel486 SX processor . . . . . . . . . . . . . 18-7, 18-20 interrupts . . . . . . . . . . . . . . . . . . . . . . . 5-10
Interprivilege level calls propagation delay . . . . . . . . . . . . . . . . . . 18-27
call mechanism . . . . . . . . . . . . . . . . . . . . . .4-17 restarting a task or program . . . . . . . . . . . . 5-7
stack switching . . . . . . . . . . . . . . . . . . . . . .4-21 software. . . . . . . . . . . . . . . . . . . . . . . . . . . 5-55
Interrupt command register (ICR), local APIC .7-25 summary of . . . . . . . . . . . . . . . . . . . . . . . . . 5-6
Interrupt gates user defined . . . . . . . . . . . . . . . . . . . .5-4, 5-55
16-bit, interlevel return from . . . . . . . . . . .18-34 valid APIC interrupts . . . . . . . . . . . . . . . . . 7-15
clearing IF flag . . . . . . . . . . . . . . . . . . 5-9, 5-18 vectors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
difference between interrupt and trap gates . . INTn instruction . . . . . . . . . . . . . . . . . . . . . . 15-10
5-18 INTO instruction . . . . . . . . . . 3-9, 5-3, 5-26, 15-10
for 16-bit and 32-bit code modules . . . . . . .17-2 INTR# pin . . . . . . . . . . . . . . . . . . . . . . . . . .5-2, 5-8
handling a virtual-8086 mode interrupt or Invalid arithmetic operand exception (#IA), FPU
exception through . . . . . . . . . . . . . . . .16-17 description of. . . . . . . . . . . . . . . . . . . . . . 11-17
in IDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-13 Invalid opcode exception (#UD) . 5-28, 12-3, 15-4,
introduction to . . . . . . . . . . . . . . . . . . . . 2-3, 2-4 18-6, 18-13
layout of . . . . . . . . . . . . . . . . . . . . . . . . . . .5-13 Invalid operation exception. . . . . . . . . . . . . . 11-17
Interrupt handler Invalid operation exception, FPU . . . .18-13, 18-17
calling . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-15 Invalid TSS exception (#TS). . . . . . . . . . .5-35, 6-7
defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-1 Invalid-opcode exception (#UD) . . . . .18-25, 18-26
flag usage by handler procedure . . . . . . . .5-18 INVD instruction . . . . 2-21, 4-25, 7-12, 9-15, 18-5
procedures . . . . . . . . . . . . . . . . . . . . . . . . .5-15 INVLPG instruction . . . . . . . 2-21, 4-25, 7-12, 18-5
protection of handler procedures . . . . . . . .5-17 IOPL (I/O privilege level) field, EFLAGS register
task . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-18, 6-3 description of. . . . . . . . . . . . . . . . . . . . . . . . 2-8
Interrupt redirection bit map field (in TSS) . . .16-16 restoring on return from exception or interrupt h
Interrupts andler. . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
acceptance, local APIC. . . . . . . . . . . . . . . .7-30 sensitive instructions in virtual-8086
APIC priority levels . . . . . . . . . . . . . . . . . . .7-15 mode . . . . . . . . . . . . . . . . . . . . . . . . . 16-14
automatic bus locking when IRET instruction . . 3-9, 5-8, 5-9, 5-15, 5-18, 6-10,
acknowledging. . . . . . . . . . . . . . . . . . .18-37 6-12, 7-12, 16-6, 16-27
control transfers between 16- and 32-bit code IRETD instruction . . . . . . . . . . . . . . . . . . . . . . 7-12
modules. . . . . . . . . . . . . . . . . . . . . . . . .17-8 IRR (interrupt request register), local APIC . . 7-30
description of . . . . . . . . . . . . . . . . . . . . 2-4, 5-1 ISR (in-service register), local APIC . . . . . . . . 7-30
distribution mechanism, local APIC . . . . . .7-22 I/O
enabling and disabling . . . . . . . . . . . . . . . . .5-8 breakpoint exception conditions . . . . . . . . 15-9
handler mechanism . . . . . . . . . . . . . . . . . .5-15 in virtual-8086 mode . . . . . . . . . . . . . . . . 16-14
handler procedures. . . . . . . . . . . . . . . . . . .5-15 instruction restart flag, SMM revision indentifier
handling . . . . . . . . . . . . . . . . . . . . . . . . . . .5-15 field . . . . . . . . . . . . . . . . . . . . .12-15, 12-16
handling in real-address mode . . . . . . . . . .16-6 instructions, restarting following an SMI
handling in SMM . . . . . . . . . . . . . . . . . . . .12-10 interrupt . . . . . . . . . . . . . . . . . . . . . . . 12-15
handling in virtual-8086 mode. . . . . . . . . .16-15 I/O permission bit map, TSS . . . . . . . . . . . . 6-6
handling multiple NMIs . . . . . . . . . . . . . . . . .5-8 map base address field, TSS . . . . . . . . . . . 6-6
handling through a task gate in virtual-8086 I/O APIC
mode . . . . . . . . . . . . . . . . . . . . . . . . . .16-20 bus arbitration . . . . . . . . . . . . . . . . . . . . . . 7-15
description of. . . . . . . . . . . . . . . . . . . . . . . 7-13
 INDEX

external interrupts . . . . . . . . . . . . . . . . . . . . .5-2 task-gate descriptor. . . . . . . . . . . . . . . . . . . 6-8

interrupt sources . . . . . . . . . . . . . . . . . . . . .7-15 use in address translation . . . . . . . . . . . . . . 3-7
relationship of local APIC to I/O APIC . . . .7-14 LDTR register
valid interrupts . . . . . . . . . . . . . . . . . . . . . .7-15 description of. . . . . . . . . . . . . . . . . . .2-11, 3-18
introduction to . . . . . . . . . . . . . . . . . . . .2-3, 2-5
limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
J storing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-18
JMP instruction. . 3-9, 4-12, 4-13, 4-17, 6-3, 6-10, LE (local exact breakpoint enable) flag,
6-12 DR7 register . . . . . . . . . . . . .15-5 , 15-10
LEN0-LEN3 (Length) fields, DR7 register . . . 15-6
K LES instruction . . . . . . . . . . . . . . . . 3-9, 4-10, 5-28
LFS instruction . . . . . . . . . . . . . . . . . . . . .3-9, 4-10
KEN# pin . . . . . . . . . . . . . . . . . . . . . . . . . . . .18-39
LGDT instruction. . . 2-20, 4-25, 7-12, 8-12, 18-25
LGS instruction . . . . . . . . . . . . . . . . . . . . .3-9, 4-10
L LIDT instruction2-20, 4-25, 5-13, 7-12, 8-10, 16-6,
L0-L3 (local breakpoint enable) flags, 18-28
DR7 register. . . . . . . . . . . . . . . . . . .15-5 Limit checking
L1 (level 1) cache description of. . . . . . . . . . . . . . . . . . . . . . . . 4-5
description of . . . . . . . . . . . . . . . . . . . . . . . .9-2 pointer offsets are within limits . . . . . . . . . 4-28
disabling . . . . . . . 9-4, 9-5, 9-8, 9-9, 9-15, 9-19 Limit field, segment descriptor . . . . . . . . . .4-2, 4-5
introduction of . . . . . . . . . . . . . . . . . . . . . .18-30 Linear address
MESI cache protocol. . . . . . . . . . . . . . . . . . .9-9 description of. . . . . . . . . . . . . . . . . . . . . . . . 3-6
L2 (level 2) cache introduction to . . . . . . . . . . . . . . . . . . . . . . . 2-5
description of . . . . . . . . . . . . . . . . . . . . . . . .9-2 Linear address space . . . . . . . . . . . . . . . . . . . . 3-6
disabling . . . . . . . 9-4, 9-5, 9-8, 9-9, 9-15, 9-19 defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
introduction of . . . . . . . . . . . . . . . . . . . . . .18-30 of task . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-17
MESI cache protocol. . . . . . . . . . . . . . . . . . .9-9 Link (to previous task) field, TSS . . . . . . . . . . 5-19
LAR instruction. . . . . . . . . . . . . . . . . . . . 2-20, 4-26 Linking tasks
Larger page sizes mechanism . . . . . . . . . . . . . . . . . . . . . . . . 6-14
introduction of . . . . . . . . . . . . . . . . . . . . . .18-32 modifying task linkages . . . . . . . . . . . . . . . 6-16
support for. . . . . . . . . . . . . . . . . . . . . . . . .18-23 LINT pins
Last branch, interrupt, and exception recording function of . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
description of . . . . . . . . . . . . . . . . . . . . . .15-11 programming . . . . . . . . . . . . . . . . . . . . . . . . E-1
initialization . . . . . . . . . . . . . . . . . . . . . . . .15-14 LLDT instruction . . . . . . . . . . . . . . 2-20, 4-25, 7-12
LastBranchFromIP MSR . . . . . 15-1, 15-13, 15-14 LMSW instruction . . . . . . . . . . . . . . . . . .2-20, 4-25
LastBranchToIP MSR . . . . . . . 15-1, 15-13, 15-14 Local APIC
LastExceptionFromIP MSR . . . 15-2, 15-13, 15-14 APIC_BASE_MSR . . . . . . . . . . . . . . . . . . 7-19
LastExceptionToIP MSR . . . . . 15-2, 15-13, 15-14 APR (arbitration priority register). . . . . . . . 7-32
LBR (last branch/interrupt/exception) flag, arbitration priority . . . . . . . . . . . . . . . . . . . 7-22
DebugCtlMSR register . . . 15-11, 15-13, block diagram . . . . . . . . . . . . . . . . . . . . . . 7-16
15-14 bus arbitration . . . . . . . . . . . . . . . . . . . . . . 7-15
LDR (logical destination register), local APIC .7-20 cluster model. . . . . . . . . . . . . . . . . . . . . . . 7-21
LDS instruction. . . . . . . . . . . . . . . . . . . . . 3-9, 4-10 current-count register . . . . . . . . . . . . . . . . 7-44
LDT description of. . . . . . . . . . . . . . . . . . . . . . . 7-13
associated with a task. . . . . . . . . . . . . . . . . .6-3 DFR (destination format register) . . . . . . . 7-21
description of . . . . . . . . . . . . . . . . . . . . . . .3-18 divide configuration register . . . . . . . . . . . 7-43
index into with index field of segment enabling or disabling . . . . . . . . . . . . . . . . . 7-19
selector . . . . . . . . . . . . . . . . . . . . . . . . . .3-7 EOI (end-of-interrupt register) . . . . . . . . . . 7-33
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-3 ESR (error status register) . . . . . . . . . . . . 7-42
pointer to in TSS . . . . . . . . . . . . . . . . . . . . . .6-5 external interrupts . . . . . . . . . . . . . . . . . . . . 5-2
pointers to exception and interrupt flat model. . . . . . . . . . . . . . . . . . . . . . . . . . 7-21
handlers. . . . . . . . . . . . . . . . . . . . . . . . .5-15 focus processor. . . . . . . . . . . . . . . . . . . . . 7-22
segment descriptors in . . . . . . . . . . . . . . . . .3-9 ID. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-20
segment selector field, TSS . . . . . . . . . . . .6-17 identifying BSP . . . . . . . . . . . . . . . . . . . . . 7-19
selecting with TI (table indicator) flag of segment indicating performance-monitoring counter
selector . . . . . . . . . . . . . . . . . . . . . . . . . .3-8 overflow . . . . . . . . . . . . . . . . . . . . . . . 15-19
setting up during initialization . . . . . . . . . . .8-12 initial-count register . . . . . . . . . . . . . . . . . . 7-44
task switching . . . . . . . . . . . . . . . . . . . . . . .6-10 initialization . . . . . . . . . . . . . . . . . . . . . . . . 7-35

INDEX-9
INDEX

interrupt acceptance . . . . . . . . . . . . . . . . . .7-30 M

interrupt acceptance decision flow chart. . .7-30 Machine-check architecture
interrupt command register (ICR) . . . . . . . .7-25 availability of machine-check architecture and
interrupt destination . . . . . . . . . . . . . . . . . .7-20 exception . . . . . . . . . . . . . . . . . . . . . . . 13-7
interrupt distribution mechanism. . . . . . . . .7-22 compatibility with Pentium processor
interrupt sources . . . . . . . . . . . . . . . . . . . . .7-15 implementation . . . . . . . . . . . . . . . . . . 13-1
IRR (interrupt request register) . . . . . . . . . .7-30 error codes, compound . . . . . . . . . . . . . . . 13-9
ISR (in-service register) . . . . . . . . . . . . . . .7-30 error codes, interpreting . . . . . . . . . . . . . . 13-8
LDR (logical destination register) . . . . . . . .7-20 error codes, simple . . . . . . . . . . . . . . . . . . 13-9
local vector table (LVT). . . . . . . . . . . . . . . .7-23 error-reporting MSRs . . . . . . . . . . . . . . . . 13-4
logical destination mode . . . . . . . . . . . . . . .7-20 first introduced. . . . . . . . . . . . . . . . . . . . . 18-27
LVT (local-APIC version register) . . . . . . . .7-36 global MSRs . . . . . . . . . . . . . . . . . . . . . . . 13-2
MDA (message destination address) . . . . .7-20 guidelines for writing machine-check
new features incorporated in the Pentium Pro software . . . . . . . . . . . . . . . . . . . . . . . 13-14
processor. . . . . . . . . . . . . . . . . . . . . . . .7-45 initialization of . . . . . . . . . . . . . . . . . . . . . . 13-7
physical destination mode . . . . . . . . . . . . .7-20 introduction of in Intel Architecture
PPR (processor priority register) . . . . . . . .7-32 processors . . . . . . . . . . . . . . . . . . . . . 18-39
register address map . . . . . . . . . . . . . . . . .7-18 logging correctable machine-check errors 13-16
relationship of local APIC to I/O APIC . . . .7-14 machine-check error codes, external bus
relocating base address . . . . . . . . . . . . . . .7-19 errors . . . . . . . . . . . . . . . . . . . . . . . . . 13-11
serial bus . . . . . . . . . . . . . . . . . . . . . . . . . . .5-2 machine-check exception handler. . . . . . 13-14
SMI interrupt . . . . . . . . . . . . . . . . . . . . . . . .12-2 MCG_CAP MSR . . . . . . . . . . . . . . . . . . . . 13-2
software visible differences between the local MCG_CTL MSR . . . . . . . . . . . . . . . . . . . . 13-4
APIC on a Pentium Pro processor and the MCi_ADDR MSRs. . . . . . . . . . . . . . . . . . . 13-6
82489DX . . . . . . . . . . . . . . . . . . . . . . . .7-44 MCi_CTL MSRs . . . . . . . . . . . . . . . . . . . . 13-4
spurious interrupt . . . . . . . . . . . . . . . . . . . .7-33 MCi_MISC MSRs . . . . . . . . . . . . . . . . . . . 13-7
state after a software (INIT) reset . . . . . . . .7-35 MCi_STATUS MSRs. . . . . . . . . . . . . . . . . 13-5
state after INIT-deassert message . . . . . . .7-35 MSRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13-2
state after power-up reset. . . . . . . . . . . . . .7-35 overview . . . . . . . . . . . . . . . . . . . . . . . . . . 13-1
state of . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-33 P5_MC_ADDR MSR . . . . . . . . . . . . . . . . . 13-7
SVR (spurious-interrupt vector register) . . .7-34 P5_MC_TYPE MSR . . . . . . . . . . . . . . . . . 13-7
timer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-43 Pentium processor machine-check exception
TMR (trigger mode register) . . . . . . . . . . . .7-30 handling . . . . . . . . . . . . . . . . . . . . . . . 13-16
TPR (task priority register) . . . . . . . . . . . . .7-31 Pentium processor style error reporting . . 13-7
valid interrupts . . . . . . . . . . . . . . . . . . . . . .7-15 Machine-check exception (#MC) 5-52, 13-1, 13-7,
Local APIC version register . . . . . . . . . . . . . . .7-36 13-14, 18-26, 18-39
Local descriptor table register (see LDTR) Maskable hardware interrupts
Local descriptor table (see LDT) delivered with local APIC . . . . . . . . . . . . . 7-23
Local vector table (LVT), local APIC . . . . . . . .7-23 description of. . . . . . . . . . . . . . . . . . . . . . . . 5-2
LOCK prefix . 2-22, 5-28, 7-2, 7-3, 7-4, 7-9, 18-37 handling with virtual interrupt mechanism 16-20
Locked (atomic) operations masking. . . . . . . . . . . . . . . . . . . . . . . . .2-8, 5-8
automatic bus locking . . . . . . . . . . . . . . . . . .7-3 Masked responses
bus locking . . . . . . . . . . . . . . . . . . . . . . . . . .7-3 to denormal operand exception. . . . . . . . 11-19
effects of a locked operation on internal to FPU stack overflow or underflow
processor caches . . . . . . . . . . . . . . . . . .7-6 exception . . . . . . . . . . . . . . . . . . . . . . 11-17
loading a segment descriptor . . . . . . . . . .18-24 to inexact result (precision) exception. . . 11-21
on Intel Architecture processors . . . . . . . .18-37 to numeric overflow exception. . . . . . . . . 11-20
overview of . . . . . . . . . . . . . . . . . . . . . . . . . .7-2 MCA (machine-check architecture) flag, CPUID
software-controlled bus locking . . . . . . . . . .7-4 instruction . . . . . . . . . . . . . . . . . . . . 13-7
LOCK# signal . . . . . . . . . . . 2-22, 7-2, 7-3, 7-4, 7-6 MCE (machine-check enable) flag, CR4 control
Logical address space, of task. . . . . . . . . . . . .6-18 register . . . . . . . . . . . . . . . . .2-17, 18-22
Logical address, description of. . . . . . . . . . . . . .3-6 MCE (machine-check exception) flag, CPUID
Logical destination mode, local APIC. . . . . . . .7-20 instruction . . . . . . . . . . . . . . . . . . . . 13-7
LSL instruction . . . . . . . . . . . . . . . . . . . . 2-20, 4-28 MCG_CAP MSR. . . . . . . . . . . . . . . . . .13-2, 13-15
LSS instruction . . . . . . . . . . . . . . . . . . . . . 3-9, 4-10 MCG_CTL MSR . . . . . . . . . . . . . . . . . . . . . . . 13-4
LTR instruction . . . . . . . 2-20, 4-25, 6-8, 7-12, 8-13 MCG_STATUS MSR . . . . . . . . . . . . .13-15, 13-17
LVT (local vector table), local APIC . . . . . . . . .7-23 MCi_ADDR MSRs . . . . . . . . . . . . . . . . . . . . 13-17
MCi_CTL MSRs . . . . . . . . . . . . . . . . . . . . . . . 13-4

INDEX-10
 INDEX

MCi_MISC MSRs . . . . . . . . . . . . . . . . . 13-7, 13-17 MTRR flag, EDX feature information register . 9-20
MCi_STATUS MSRs . . . . . . . . 13-5, 13-15, 13-17 MTRRcap register . . . . . . . . . . . . . . . . . . . . . 9-20
MDA (message destination address), local MTRRdefType register . . . . . . . . . . . . . . . . . . 9-21
APIC. . . . . . . . . . . . . . . . . . . . . . . . .7-20 MTRRfix16K_80000 and MTRRfix16K_A0000
Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-1 (fixed range) MTRRs . . . . . . . . . . . 9-23
Memory management MTRRfix4K_C0000. and MTRRfix4K_F8000 (fixed
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 range) MTRRs . . . . . . . . . . . . . . . . 9-23
overview . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-1 MTRRfix64K_00000 (fixed range) MTRR. . . . 9-22
paging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-1 MTRRphysBasen (variable range) MTRRs . . 9-23
segmentation . . . . . . . . . . . . . . . . . . . . . . . .3-1 MTRRphysMaskn (variable range) MTRRs . . 9-23
Memory ordering MTRRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-9
in Intel Architecture processors . . . . . . . .18-36 address mapping for fixed-range MTRRs . 9-23
overview . . . . . . . . . . . . . . . . . . . . . . . . . . . .7-6 cache control. . . . . . . . . . . . . . . . . . . . . . . 9-12
processor ordering . . . . . . . . . . . . . . . . . . . .7-6 description of. . . . . . . . . . . . . . . . . . . .8-9, 9-18
snooping mechanism . . . . . . . . . . . . . . . . . .7-8 enabling caching . . . . . . . . . . . . . . . . . . . . . 8-8
write forwarding . . . . . . . . . . . . . . . . . . . . . .7-8 example of base and mask calculations . . 9-25
write ordering . . . . . . . . . . . . . . . . . . . . . . . .7-6 feature identification . . . . . . . . . . . . . . . . . 9-20
Memory type range registers (see MTRRs) fixed-range registers . . . . . . . . . . . . . . . . . 9-22
Memory types initialization of . . . . . . . . . . . . . . . . . . . . . . 9-27
caching methods, defined. . . . . . . . . . . . . . .9-5 introduction of in Intel Architecture
choosing . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-8 processors . . . . . . . . . . . . . . . . . . . . 18-39
MTRR types . . . . . . . . . . . . . . . . . . . . . . . .9-19 large page size considerations . . . . . . . . . 9-32
UC (uncacheable). . . . . . . . . . . . . . . . . . . . .9-5 mapping physical memory with . . . . . . . . . 9-20
WB (write back) . . . . . . . . . . . . . . . . . . . . . .9-6 memory types and their properties . . . . . . 9-19
WC (write combining) . . . . . . . . . . . . . . . . . .9-6 MemTypeGet() function . . . . . . . . . . . . . . 9-28
WP (write protected) . . . . . . . . . . . . . . . . . . .9-7 MemTypeSet() function. . . . . . . . . . . . . . . 9-29
WT (write through) . . . . . . . . . . . . . . . . . . . .9-6 MTRRcap register . . . . . . . . . . . . . . . . . . . 9-20
MemTypeGet() function . . . . . . . . . . . . . . . . . .9-28 MTRRdefType register . . . . . . . . . . . . . . . 9-21
MemTypeSet() function . . . . . . . . . . . . . . . . . .9-29 multiple-processor considerations. . . . . . . 9-31
MESI cache protocol precedence of cache controls . . . . . . . . . . 9-13
described . . . . . . . . . . . . . . . . . . . . . . . 9-4, 9-9 precedences . . . . . . . . . . . . . . . . . . . . . . . 9-26
Mixing 16-bit and 32-bit code programming interface . . . . . . . . . . . . . . . 9-28
on Intel Architecture processors . . . . . . . .18-34 remapping memory types . . . . . . . . . . . . . 9-27
overview . . . . . . . . . . . . . . . . . . . . . . . . . . .17-1 setting memory ranges . . . . . . . . . . . . . . . 9-21
MMX instructions state of following a hardware reset . . . . . . 9-18
pairing guidelines . . . . . . . . . . . . . . . . . . .14-17 variable-range registers . . . . . . . . . . . . . . 9-23
Mode switching Multiple-processor initialization
between real-address and protected mode 8-13 MP protocol . . . . . . . . . . . . . . . . . . . .7-45, 7-46
example . . . . . . . . . . . . . . . . . . . . . . . . . . .8-16 procedure . . . . . . . . . . . . . . . . . . . . . . . . . 7-48
to SMM . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-2 Multiple-processor management
Model and stepping information, following bus locking . . . . . . . . . . . . . . . . . . . . . . . . . 7-3
processor initialization or reset . . . . .8-5 guaranteed atomic operations. . . . . . . . . . . 7-2
Model-specific registers (see MSRs) interprocessor and self-interrupts . . . . . . . 7-25
MOV instruction . . . . . . . . . . . . . . . . . . . . 3-9, 4-10 local APIC . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
MOV (control registers) instructions. . . 2-20, 4-25, memory ordering . . . . . . . . . . . . . . . . . . . . . 7-6
7-12, 8-14 MP protocol . . . . . . . . . . . . . . . . . . . .7-45, 7-46
MOV (debug registers) instructions . . . 2-21, 4-25, overview of . . . . . . . . . . . . . . . . . . . . . . . . . 7-1
7-12, 15-10 SMM considerations . . . . . . . . . . . . . . . . 12-17
MP (monitor coprocessor) flag, CR0 control register Multiple-processor system
2-16, 5-30, 8-6, 8-8 MP protocol . . . . . . . . . . . . . . . . . . . .7-45, 7-46
MP (monitor coprocessor) flag, CR0 register. .18-8 relationship of local and I/O APICs . . . . . . 7-14
MSRs Multisegment model . . . . . . . . . . . . . . . . . . . . . 3-5
description of . . . . . . . . . . . . . . . . . . . . . . . .8-8 Multitasking
introduction of in Intel Architecture processors initialization for . . . . . . . . . . . . . . . . . . . . . 8-13
18-38 linking tasks. . . . . . . . . . . . . . . . . . . . . . . . 6-14
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 mechanism, description of . . . . . . . . . . . . . 6-3
machine-check architecture . . . . . . . . . . . .13-2 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1
reading and writing . . . . . . . . . . . . . . . . . . .2-23 setting up TSS. . . . . . . . . . . . . . . . . . . . . . 8-13

INDEX-11
INDEX

setting up TSS descriptor . . . . . . . . . . . . . .8-13 Operands

operand-size prefix . . . . . . . . . . . . . . . . . . 17-2
OR instruction. . . . . . . . . . . . . . . . . . . . . . . . . . 7-4
N OS (operating system mode) flag, PerfEvtSel0 and
NaN PerfEvtSel1 MSRs (P6 family
compatibility, Intel Architecture processors . . . processors). . . . . . . . . . . . . . . . . . 15-16
18-10 OUT instruction. . . . . . . . . . . . . . . . . . . . . . . . 7-10
NE (numeric error) flag, CR0 control register. 2-14, OUTS instruction . . . . . . . . . . . . . . . . . . . . . 15-10
5-48, 8-6, 8-8, 18-22 Overflow exception (#OF). . . . . . . . . . . . . . . . 5-26
NE (numeric error) flag, CR0 register . . . . . . .18-8 Overflow, FPU stack. . . . . . . . . . . . . . . . . . . 11-17
NEG instruction . . . . . . . . . . . . . . . . . . . . . . . . .7-4
NMI interrupt . . . . . . . . . . . . . . . . . . . . . 2-22, 7-13
description of . . . . . . . . . . . . . . . . . . . . . . . .5-2 P
handling during initialization . . . . . . . . . . . .8-10 P (present) flag
handling in SMM . . . . . . . . . . . . . . . . . . . .12-10 page-directory entry . . . . . . . . . . . . . . . . . 5-44
handling multiple NMIs . . . . . . . . . . . . . . . . .5-8 page-table entry . . . . . . . . . . . . . . . .3-25, 5-44
masking . . . . . . . . . . . . . . . . . . . . . . . . . .18-28 P (segment-present) flag, segment descriptor 3-12
receiving when processor is shutdown . . . .5-33 P5_MC_ADDR MSR . . . . . . . . . . . . . .13-7, 13-16
reference information . . . . . . . . . . . . . . . . .5-24 P5_MC_TYPE MSR . . . . . . . . . . . . . . .13-7, 13-16
vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-4 P6 family processors
NMI# pin. . . . . . . . . . . . . . . . . . . . . . . . . . 5-2, 5-24 description of. . . . . . . . . . . . . . . . . . . . . . . . 1-1
Nonconforming code segments list of events counted with
accessing . . . . . . . . . . . . . . . . . . . . . . . . . .4-14 performance-monitoring counters . . . . . A-1
C (conforming) flag . . . . . . . . . . . . . . . . . . .4-13 PAE (physical address extension) flag, CR4 control
description of . . . . . . . . . . . . . . . . . . . . . . .3-14 register . 2-17, 3-19, 3-29, 18-21, 18-23
Nonmaskable interrupt (see NMI) Page base address field, page-table entry . . . 3-25
NOT instruction . . . . . . . . . . . . . . . . . . . . . . . . .7-4 Page directory
Notation base address. . . . . . . . . . . . . . . . . . . . . . . 3-23
bit and byte order . . . . . . . . . . . . . . . . . . . . .1-6 base address (PDBR) . . . . . . . . . . . . . . . . . 6-6
exceptions. . . . . . . . . . . . . . . . . . . . . . . . . . .1-8 description of. . . . . . . . . . . . . . . . . . . . . . . 3-20
hexadecimal and binary numbers. . . . . . . . .1-7 introduction to . . . . . . . . . . . . . . . . . . . . . . . 2-5
instruction operands . . . . . . . . . . . . . . . . . . .1-7 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
reserved bits . . . . . . . . . . . . . . . . . . . . . . . . .1-6 setting up during initialization . . . . . . . . . . 8-13
segmented addressing . . . . . . . . . . . . . . . . .1-7 Page frame (see Page)
Notational conventions. . . . . . . . . . . . . . . . . . . .1-5 Page tables
NT (nested task) flag, EFLAGS register. 2-9, 6-10, description of. . . . . . . . . . . . . . . . . . . . . . . 3-20
6-12, 6-14 introduction to . . . . . . . . . . . . . . . . . . . . . . . 2-5
Null segment selector, checking for . . . . . . . . . .4-7 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
Numeric overflow exception (#O). . . . 11-19, 18-11 setting up during initialization . . . . . . . . . . 8-13
Numeric underflow exception (#U). . . 11-20, 18-12 Page-directory entries
NV(invert)flag,PerfEvtSel0MSR(P6familyprocessors) automatic bus locking while updating . . . . . 7-4
15-17 caching in TLBs. . . . . . . . . . . . . . . . . . . . . . 9-4
NW (not writethrough) flag, CR0 control page-table base address field . . . . . . . . . . 3-25
register . . . . . . . . . 2-13, 8-8, 9-11, 9-12, R/W (read/write) flag . . . . . . . . . . 4-2 , 4-3, 4-32
9-14, 9-31, 9-32 structure of . . . . . . . . . . . . . . . . . . . . . . . . 3-23
NW (not write-through) flag, CR0 control U/S (user/supervisor) flag . . . . . . 4-2, 4-3, 4-31
register . . . . . . . . . . 18-22, 18-23, 18-30 Page-directory-pointer (PDPTR) table . . . . . . 3-30
Page-fault exception (#PF). . . . . 3-18, 5-44, 18-26
Pages
O descripiton of. . . . . . . . . . . . . . . . . . . . . . . 3-20
Obsolete instructions . . . . . . . . . . . . . . 18-5, 18-18 disabling protection of . . . . . . . . . . . . . . . . . 4-2
OE (numeric overflow exception) flag, FPU status enabling protection of . . . . . . . . . . . . . . . . . 4-2
word . . . . . . . . . . . . . . . . . . 11-18, 11-19 introduction to . . . . . . . . . . . . . . . . . . . . . . . 2-5
OF flag, EFLAGS register . . . . . . . . . . . . . . . .5-26 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2
Opcodes PG flag, CR0 control register . . . . . . . . . . . 4-2
undefined . . . . . . . . . . . . . . . . . . . . . . . . . .18-6 Pages, split . . . . . . . . . . . . . . . . . . . . . . . . . . 18-18
Operand Page-table base address field, page-directory
instruction . . . . . . . . . . . . . . . . . . . . . . . . . . .1-7 entry . . . . . . . . . . . . . . . . . . . . . . . . 3-25

INDEX-12
 INDEX

Page-table entries PerfCtr0 MSR and PerfCtr1 MSRs (P6 family

automatic bus locking while updating . . . . . .7-4 processors). . . . . . . . . . . . . . . . . . 15-18
caching in TLBs . . . . . . . . . . . . . . . . . . . . . .9-4 PerfEvtSel0 and PerfEvtSel1 MSRs (P6 family
effect of implicit caching on. . . . . . . . . . . . .9-16 processors). . . . . . . . . . . . . . . . . . 15-16
page base address field . . . . . . . . . . . . . . .3-25 Performance-monitoring counters
R/W (read/write) flag. . . . . . . . . . 4-2, 4-3, 4-32 description of. . . . . . . . . . . . . . . . . . . . . . 15-15
structure of . . . . . . . . . . . . . . . . . . . . . . . . .3-23 events that can be counted (P6 family
U/S (user/supervisor) flag . . . . . . 4-2, 4-3, 4-31 processors) . . . . . . . . . . . . . . . . . . . . . . A-1
Paging events that can be counted (Pentium
combining segment and page-level processors) . . . . . . . . . . . . . . . 15-22, A-12
protection. . . . . . . . . . . . . . . . . . . . . . . .4-33 introduction of in Intel Architecture
combining with segmentation . . . . . . . . . . . .3-6 processors . . . . . . . . . . . . . . . . . . . . . 18-40
defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-1 monitoring counter overflow (P6 family
initializing . . . . . . . . . . . . . . . . . . . . . . . . . .8-12 processors) . . . . . . . . . . . . . . . . . . . . 15-19
introduction to . . . . . . . . . . . . . . . . . . . . . . . .2-5 overflow, monitoring (P6 family
large page size MTRR considerations . . . .9-32 processors) . . . . . . . . . . . . . . . . . . . . 15-19
linear address translation (4-KByte pages).3-20 overview of . . . . . . . . . . . . . . . . . . . . . . . . . 2-6
linear address translation (4-MByte pages) 3-21 P6 family processors. . . . . . . . . . . . . . . . 15-15
mapping segments to pages. . . . . . . . . . . .3-39 Pentium II processor . . . . . . . . . . . . . . . . 15-15
mixing 4-KByte and 4-MByte pages . . . . . .3-22 Pentium Pro processor . . . . . . . . . . . . . . 15-15
page boundaries regarding TSS. . . . . . . . . .6-6 Pentium processor . . . . . . . . . . . . . . . . . 15-20
page-fault exception . . . . . . . . . . . . . . . . . .5-44 reading . . . . . . . . . . . . . . . . . . . . . .2-22, 15-18
page-level protection . . . . . . . . . . . . . 4-2, 4-30 setting up (P6 family processors) . . . . . . 15-16
page-level protection flags . . . . . . . . . . . . .4-31 software drivers for . . . . . . . . . . . . . . . . . 15-18
virtual-8086 tasks . . . . . . . . . . . . . . . . . . .16-10 starting and stopping. . . . . . . . . . . . . . . . 15-18
Parameter Performance-monitoring events
passing, between 16- and 32-bit call gates 17-7 list of events . . . . . . . . . . . . . . . . . . . . . . . . A-1
translation, between 16- and 32-bit code PG (paging) flag, CR0 control register . 2-13, 3-19,
segments. . . . . . . . . . . . . . . . . . . . . . . .17-8 3-26, 4-2, 8-13, 8-14, 12-8, 18-32
PBi (performance monitoring/breakpoint pins) flags, PGE (page global enable) flag, CR4 control
DebugCtlMSR register . . . . . . . . . .15-12 register . . . . . . 2-17, 3-27, 18-21, 18-23
PC (pin control) flag, PerfEvtSel0 and PerfEvtSel1 PhysBase field, MTRRphysBasen register. . . 9-24
MSRs (P6 family processors) . . . .15-17 Physical address extension
PC0 and PC1 (pin control) fields, CESR MSR access full extended physical address
(Pentium processor). . . . . . . . . . . .15-21 space . . . . . . . . . . . . . . . . . . . . . . . . . . 3-32
PCD (page-level cache disable) flag description of. . . . . . . . . . . . . . . . . . . . . . . 3-29
CR3 control register . 2-16, 9-12, 18-22, 18-31 page-directory entries . . . . . . . . . . . . . . . . 3-33
page-directory entries . . . 8-8, 9-12, 9-13, 9-32 page-table entries . . . . . . . . . . . . . . . . . . . 3-33
page-table entries . 3-26, 8-8, 9-12, 9-13, 9-32, Physical address space
18-32 defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
PCE (performance-monitoring counter enable) flag, description of. . . . . . . . . . . . . . . . . . . . . . . . 3-6
CR4 control register . . 2-18, 4-25, 18-21 mapped to a task. . . . . . . . . . . . . . . . . . . . 6-17
PCE (performance-monitoring counter enable) flag, Physical addressing . . . . . . . . . . . . . . . . . . . . . 2-5
CR4 control register (P6 family Physical destination mode, local APIC . . . . . . 7-20
processors) . . . . . . . . . . . . . . . . . .15-18 Physical memory
PDBR (see CR3 control register) mapping of with fixed-range MTRRs . . . . . 9-23
PE (inexact result exception) flag, FPU status mapping of with variable-range MTRRs . . 9-23
word . . . . . . . . . . . . . . . . . . . 11-4, 11-21 PhysMask, MTRRphysMaskn register . . . . . . 9-24
PE (protection enable) flag, CR0 control PM0/BP0 and PM1/BP1 (performance-monitor)
register . . . . . 2-16, 4-2, 8-13, 8-14, 12-8 pins (Pentium processor) . 15-20, 15-21,
Pentium Pro processor. . . . . . . . . . . . . . . . . . . .1-1 15-22
Pentium processors . . . . . . . . . . . . . . . . . . . . .18-7 Pointers
list of events counted with code-segment pointer size . . . . . . . . . . . . 17-5
performance-monitoring counters . . . . A-12 limit checking. . . . . . . . . . . . . . . . . . . . . . . 4-28
performance-monitoring counters. . . . . . .15-20 validation . . . . . . . . . . . . . . . . . . . . . . . . . . 4-25
PerfCtr0 and PerfCtr1 MSRs (P6 family POP instruction. . . . . . . . . . . . . . . . . . . . . . . . . 3-9
processors) . . . . . . . . . . . . . . . . . .15-16 POPF instruction . . . . . . . . . . . . . . . . . .5-9, 15-10
PPR (processor priority register), local APIC . 7-32

INDEX-13
INDEX

Previous task link field, TSS. . . . . . 6-4, 6-14, 6-16 CR3 control register . 2-16, 9-12, 18-22, 18-31
Priority levels, APIC interrupts . . . . . . . . . . . . .7-15 page-directory entries . . . . . . . . 8-8, 9-12, 9-32
Privilege levels page-table entries . . . . . 8-8, 9-12, 9-32, 18-32
checking when accessing data segments . .4-9 page-table entry . . . . . . . . . . . . . . . . . . . . 3-26
checking, for call gates . . . . . . . . . . . . . . . .4-17
checking, when transferring program control
between code segments . . . . . . . . . . . .4-12 Q
description of . . . . . . . . . . . . . . . . . . . . . . . .4-8 QNaN
protection rings . . . . . . . . . . . . . . . . . . . . . . .4-9 compatibility, Intel Architecture
Privileged instructions . . . . . . . . . . . . . . . . . . .4-25 processors . . . . . . . . . . . . . . . . . . . . . 18-10
Processor identification
earlier Intel architecture processors . . . . . .9-33 R
Processor management
RC (rounding control) field, FPU control
initialization . . . . . . . . . . . . . . . . . . . . . . . . . .8-1
word . . . . . . . . . . . . . . . . . . . .11-3, 11-4
local APIC . . . . . . . . . . . . . . . . . . . . . . . . . .7-13
RDMSR instruction2-23, 4-25, 9-20, 15-13, 15-15,
overview of . . . . . . . . . . . . . . . . . . . . . . . . . .7-1
15-16, 15-18, 15-20, 18-4, 18-38
snooping mechanism . . . . . . . . . . . . . . . . . .7-8
RDPMC instruction2-22, 4-25, 15-16, 15-18, 18-3,
processor number . . . . . . . . . . . . . . . . . . B-4, B-9
18-21, 18-40
Processor ordering, description of . . . . . . . . . . .7-7
RDTSC instruction . . . . . . 2-22, 4-25, 15-15, 18-4
Protected mode
Read/write
IDT initialization . . . . . . . . . . . . . . . . . . . . .8-12
protection, page level . . . . . . . . . . . . . . . . 4-32
initialization for . . . . . . . . . . . . . . . . . . . . . .8-11
rights, checking . . . . . . . . . . . . . . . . . . . . . 4-27
mixing 16-bit and 32-bit code modules . . . .17-2
Real-address mode
mode switching . . . . . . . . . . . . . . . . . . . . . .8-13
8086 emulation . . . . . . . . . . . . . . . . . . . . . 16-1
PE flag, CR0 register . . . . . . . . . . . . . . . . . .4-2
address translation in . . . . . . . . . . . . . . . . 16-3
switching to . . . . . . . . . . . . . . . . . . . . . 4-2, 8-14
system data structures required during description of. . . . . . . . . . . . . . . . . . . . . . . 16-1
initialization . . . . . . . . . . . . . . . . . 8-11, 8-12 exceptions and interrupts . . . . . . . . . . . . . 16-8
Protection IDT initialization. . . . . . . . . . . . . . . . . . . . . 8-10
combining segment and page-level IDT, changing base and limit of. . . . . . . . . 16-6
protection. . . . . . . . . . . . . . . . . . . . . . . .4-33 IDT, structure of . . . . . . . . . . . . . . . . . . . . 16-7
disabling . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-2 IDT, use of. . . . . . . . . . . . . . . . . . . . . . . . . 16-6
enabling . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-2 initialization . . . . . . . . . . . . . . . . . . . . . . . . 8-10
flags used for page-level protection . . . . . . .4-2 instructions supported . . . . . . . . . . . . . . . . 16-4
flags used for segment-level protection . . . .4-2 interrupt and exception handling . . . . . . . . 16-6
of exception- and interrupt-handler procedures mode switching . . . . . . . . . . . . . . . . . . . . . 8-13
5-17 native 16-bit mode. . . . . . . . . . . . . . . . . . . 17-1
overview of . . . . . . . . . . . . . . . . . . . . . . . . . .4-1 overview of . . . . . . . . . . . . . . . . . . . . . . . . 16-1
page level . . . . . . . . . . . . . . . . . . . . . . 4-2, 4-32 registers supported . . . . . . . . . . . . . . . . . . 16-4
page level, overriding . . . . . . . . . . . . . . . . .4-32 switching to . . . . . . . . . . . . . . . . . . . . . . . . 8-15
page level, overview . . . . . . . . . . . . . . . . . .4-30 Related literature . . . . . . . . . . . . . . . . . . . . . . . 1-9
page-level protection flags . . . . . . . . . . . . .4-31 Requested privilege level (see RPL)
read/write, page level . . . . . . . . . . . . . . . . .4-32 Reserved bits . . . . . . . . . . . . . . . . . . . . . .1-6, 18-1
RESET# pin . . . . . . . . . . . . . . . . . . . . . .5-2, 18-19
segment level . . . . . . . . . . . . . . . . . . . . . . . .4-2
RESET# signal . . . . . . . . . . . . . . . . . . . . . . . . 2-22
user/supervisor type . . . . . . . . . . . . . . . . . .4-31
Reset, hardware
Protection rings . . . . . . . . . . . . . . . . . . . . . . . . .4-9
receiving when processor is shutdown . . . 5-33
PS (page size) flag, page-table entry. . . . . . . .3-27
Restarting program or task, following an exception
PSE (page size extension) flag, CR4 control
or interrupt . . . . . . . . . . . . . . . . . . . . 5-7
register . . . 2-17, 3-19, 3-21, 3-22, 9-17,
18-22, 18-23 Restricting addressable domain . . . . . . . . . . . 4-31
Pseudo-infinity . . . . . . . . . . . . . . . . . . . . . . . .18-10 RET instruction . . . . . . . . . . 4-12, 4-13, 4-23, 17-7
Returning
Pseudo-NaN. . . . . . . . . . . . . . . . . . . . . . . . . .18-10
from a called procedure . . . . . . . . . . . . . . 4-23
Pseudo-zero. . . . . . . . . . . . . . . . . . . . . . . . . .18-10
from an interrupt or exception handler . . . 5-15
PUSH instruction . . . . . . . . . . . . . . . . . . . . . . .18-7
RF (resume) flag, EFLAGS register . 2-9, 5-9, 15-2
PUSHF instruction . . . . . . . . . . . . . . . . . . 5-9, 18-7
Rounding
PVI (protected-mode virtual interrupts) flag, CR4
control, RC field of FPU control word . . . . 11-3
control register . . . . . . . . . . . 2-17 , 18-22
PWT (page-level write-through) flag modes, FPU . . . . . . . . . . . . . . . . . . .11-3, 11-4

INDEX-14
 INDEX

results, FPU . . . . . . . . . . . . . . . . . . . . . . . .11-5 basic flat model . . . . . . . . . . . . . . . . . . . . . . 3-3

RPL code type. . . . . . . . . . . . . . . . . . . . . . . . . . 3-13
description of . . . . . . . . . . . . . . . . . . . . 3-8, 4-9 combining segment and page-level
field, segment selector . . . . . . . . . . . . . . . . .4-2 protection . . . . . . . . . . . . . . . . . . . . . . . 4-33
RSM instruction . . . . 2-22, 7-12, 12-1, 12-2, 12-3, combining with paging. . . . . . . . . . . . . . . . . 3-6
12-11, 12-16, 18-5 data type . . . . . . . . . . . . . . . . . . . . . . . . . . 3-13
R/S# pin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-2 defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
R/W (read/write) flag disabling protection of . . . . . . . . . . . . . . . . . 4-2
page-directory entry . . . . . . . . . . 4-2, 4-3, 4-32 enabling protection of . . . . . . . . . . . . . . . . . 4-2
page-table entry . . . . . . . . 3-26, 4-2, 4-3, 4-32 mapping to pages . . . . . . . . . . . . . . . . . . . 3-39
R/W0-R/W3 (read/write) fields, DR7 register . 15-6, multisegment usage model . . . . . . . . . . . . . 3-5
18-24 protected flat model. . . . . . . . . . . . . . . . . . . 3-4
segment-level protection . . . . . . . . . . . . . . . 4-2
segment-not-present exception. . . . . . . . . 5-37
S system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
S (descriptor type) flag, segment descriptor . 3-11, types, checking access rights . . . . . . . . . . 4-26
3-13, 4-2, 4-6 typing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
SBB instruction. . . . . . . . . . . . . . . . . . . . . . . . . .7-4 using . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-3
Segment descriptors wraparound . . . . . . . . . . . . . . . . . . . . . . . 18-35
access rights. . . . . . . . . . . . . . . . . . . . . . . .4-26 Self-interrupts, local APIC . . . . . . . . . . . . . . . 7-25
access rights, invalid values . . . . . . . . . . .18-24 Self-modifying code, effect on caches . . . . . . 9-15
automatic bus locking while updating . . . . . .7-3 Serializing instructions . . . . . . . . . . . . .7-11 , 18-19
base address fields. . . . . . . . . . . . . . . . . . .3-11 SF (stack fault) flag, FPU status word . . . . . . 18-9
code type . . . . . . . . . . . . . . . . . . . . . . . . . . .4-3 SGDT instruction . . . . . . . . . . . . . . . . . .2-20, 3-18
data type . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-3 Shutdown
description of . . . . . . . . . . . . . . . . . . . . 2-3, 3-9 resulting from double fault. . . . . . . . . . . . . 5-33
DPL (descriptor privilege level) field . . 3-12, 4-2 resulting from out of IDT limit condition. . . 5-33
D/B (default operation size/default stack pointer SIDT instruction . . . . . . . . . . . . . . 2-20, 3-18, 5-13
size and/or upper bound) flag . . . . 3-12, 4-5 Single-stepping
E (expansion direction) flag . . . . . . . . . 4-2, 4-5 breakpoint exception condition . . . . . . . . 15-10
G (granularity) flag . . . . . . . . . . . 3-12, 4-2, 4-5 on branches . . . . . . . . . . . . . . . . . . . . . . 15-14
limit field . . . . . . . . . . . . . . . . . . . . . . . . 4-2, 4-5 on exceptions . . . . . . . . . . . . . . . . . . . . . 15-14
loading . . . . . . . . . . . . . . . . . . . . . . . . . . .18-24 on interrupts . . . . . . . . . . . . . . . . . . . . . . 15-14
P (segment-present) flag . . . . . . . . . . . . . .3-12 TF (trap) flag, EFLAGS register . . . . . . . 15-10
S (descriptor type) flag . . . 3-11, 3-13, 4-2, 4-6 SLDT instruction . . . . . . . . . . . . . . . . . . . . . . . 2-20
segment limit field . . . . . . . . . . . . . . . . . . . .3-10 SLTR instruction . . . . . . . . . . . . . . . . . . . . . . . 3-18
system type. . . . . . . . . . . . . . . . . . . . . . . . . .4-3 SMBASE
tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-16 default value . . . . . . . . . . . . . . . . . . . . . . . 12-4
TSS descriptor . . . . . . . . . . . . . . . . . . . . . . .6-6 relocation of. . . . . . . . . . . . . . . . . . . . . . . 12-14
type field . . . . . . . . . . . . . . 3-11, 3-13, 4-2, 4-6 SMI handler
type field, encoding. . . . . . . . . . . . . . 3-14, 3-15 description of. . . . . . . . . . . . . . . . . . . . . . . 12-1
when P (segment-present) flag is clear . . .3-13 execution environment for. . . . . . . . . . . . . 12-8
Segment limit exiting from . . . . . . . . . . . . . . . . . . . . . . . . 12-3
checking . . . . . . . . . . . . . . . . . . . . . . . . . . .2-20 location in SMRAM . . . . . . . . . . . . . . . . . . 12-4
field, segment descriptor. . . . . . . . . . . . . . .3-10 SMI interrupt . . . . . . . . . . . . . . . . . . . . . .2-22, 7-13
Segment not present exception (#NP) . . . . . . .3-12 description of. . . . . . . . . . . . . . . . . . .12-1, 12-2
Segment registers priority . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
description of . . . . . . . . . . . . . . . . . . . . . . . .3-8 switching to SMM . . . . . . . . . . . . . . . . . . . 12-2
saved in TSS . . . . . . . . . . . . . . . . . . . . . . . .6-4 SMI# pin . . . . . . . . . . . . . . . . . . . . 5-2, 12-2, 12-15
Segment selectors SMM
description of . . . . . . . . . . . . . . . . . . . . . . . .3-7 auto halt restart . . . . . . . . . . . . . . . . . . . . 12-13
index field . . . . . . . . . . . . . . . . . . . . . . . . . . .3-7 executing the HLT instruction in . . . . . . . 12-14
null . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4-7 exiting from . . . . . . . . . . . . . . . . . . . . . . . . 12-3
RPL field . . . . . . . . . . . . . . . . . . . . . . . . 3-8, 4-2 handling exceptions and interrupts . . . . . 12-10
TI (table indicator) flag . . . . . . . . . . . . . . . . .3-8 I/O instruction restart. . . . . . . . . . . . . . . . 12-15
Segmented addressing . . . . . . . . . . . . . . . . . . .1-7 native 16-bit mode. . . . . . . . . . . . . . . . . . . 17-1
Segment-not-present exception (#NP). . . . . . .5-37 overview of . . . . . . . . . . . . . . . . . . . . . . . . 12-1
Segments revision identifier . . . . . . . . . . . . . . . . . . . 12-12

INDEX-15
INDEX

revision identifier field . . . . . . . . . . . . . . . .12-12 description of. . . . . . . . . . . . . . . . . . . . . . . 4-31

switching to . . . . . . . . . . . . . . . . . . . . . . . . .12-2 U/S (user/supervisor) flag . . . . . . . . . . . . . 4-31
switching to from other operating modes . .12-2 SVR (spurious-interrupt vector register), local
using FPU in . . . . . . . . . . . . . . . . . . . . . . .12-11 APIC . . . . . . . . . . . . . . . . . . . . . . . . 7-34
SMRAM System
caching . . . . . . . . . . . . . . . . . . . . . . . . . . . .12-7 architecture . . . . . . . . . . . . . . . . . . . . . . . . . 2-1
description of . . . . . . . . . . . . . . . . . . . . . . .12-1 instructions . . . . . . . . . . . . . . . . . . . . .2-6, 2-18
state save map . . . . . . . . . . . . . . . . . . . . . .12-5 registers, introduction to . . . . . . . . . . . . . . . 2-5
structure of . . . . . . . . . . . . . . . . . . . . . . . . .12-4 segment descriptor, layout of . . . . . . . . . . . 4-3
SMSW instruction. . . . . . . . . . . . . . . . . . . . . . .2-20 System-management mode (see SMM)
SNaN
compatibility, Intel Architecture
processors. . . . . . . . . . . . . . . . 18-10, 18-17 T
Snooping mechanism. . . . . . . . . . . . . . . . . 7-8, 9-5 T (debug trap) flag, TSS . . . . . . . . . . . . . .6-6, 15-2
Software interrupts . . . . . . . . . . . . . . . . . . . . . . .5-3 Task gates
Software-controlled bus locking . . . . . . . . . . . . .7-4 descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . 6-8
Split pages . . . . . . . . . . . . . . . . . . . . . . . . . . .18-18 executing a task . . . . . . . . . . . . . . . . . . . . . 6-3
Spurious interrupt, local APIC . . . . . . . . . . . . .7-33 handling a virtual-8086 mode interrupt or
SS register, saving on call to exception or interrupt exception through . . . . . . . . . . . . . . . 16-20
handler . . . . . . . . . . . . . . . . . . . . . . .5-15 in IDT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-13
Stack fault exception (#SS) . . . . . . . . . . . . . . .5-39 introduction to . . . . . . . . . . . . . . . . . . . .2-3, 2-4
Stack fault, FPU . . . . . . . . . . . . . . . . . . 18-9, 18-16 layout of. . . . . . . . . . . . . . . . . . . . . . . . . . . 5-13
Stack overflow exception, FPU . . . . . . . . . . .11-17 referencing of TSS descriptor . . . . . . . . . . 5-19
Stack pointers Task management . . . . . . . . . . . . . . . . . . . . . . 6-1
privilege level 0, 1, and 2 stacks. . . . . . . . . .6-6 data structures . . . . . . . . . . . . . . . . . . . . . . 6-4
size of . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-12 mechanism, description of . . . . . . . . . . . . . 6-3
Stack segments Task register. . . . . . . . . . . . . . . . . . . . . . . . . . 3-18
privilege level checks when loading the SS description of. . . . . . . . . . . . . . . . 2-11, 6-1, 6-8
register . . . . . . . . . . . . . . . . . . . . . . . . .4-12 initializing. . . . . . . . . . . . . . . . . . . . . . . . . . 8-13
size of stack pointer . . . . . . . . . . . . . . . . . .3-12 introduction to . . . . . . . . . . . . . . . . . . . . . . . 2-5
Stack switching Task switching
inter-privilege level calls . . . . . . . . . . . . . . .4-21 description of. . . . . . . . . . . . . . . . . . . . . . . . 6-3
masking exceptions and interrupts when exception condition . . . . . . . . . . . . . . . . . 15-11
switching stacks . . . . . . . . . . . . . . . . . .5-10 operation . . . . . . . . . . . . . . . . . . . . . . . . . . 6-10
on call to exception or interrupt handler . . .5-15 preventing recursive task switching . . . . . 6-16
Stack underflow exception, FPU . . . . . . . . . .11-17 T (debug trap) flag. . . . . . . . . . . . . . . . . . . . 6-6
Stack-fault exception (#SS) . . . . . . . . . . . . . .18-35 Tasks
Stacks address space. . . . . . . . . . . . . . . . . . . . . . 6-17
error code pushes. . . . . . . . . . . . . . . . . . .18-33 description of. . . . . . . . . . . . . . . . . . . . . . . . 6-1
faults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-39 exception-handler task . . . . . . . . . . . . . . . 5-15
for privilege levels 0, 1, and 2 . . . . . . . . . . .4-21 executing. . . . . . . . . . . . . . . . . . . . . . . . . . . 6-3
interlevel RET/IRET from a 16-bit interrupt or Intel 286 processor tasks . . . . . . . . . . . . 18-37
call gate . . . . . . . . . . . . . . . . . . . . . . . .18-34 interrupt-handler task . . . . . . . . . . . . . . . . 5-15
managment of control transfers for 16- and interrupts and exceptions . . . . . . . . . . . . . 5-18
32-bit procedure calls . . . . . . . . . . . . . .17-5 linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-14
operation on pushes and pops . . . . . . . . .18-33 logical address space . . . . . . . . . . . . . . . . 6-18
pointers to in TSS . . . . . . . . . . . . . . . . . . . . .6-6 management . . . . . . . . . . . . . . . . . . . . . . . . 6-1
stack switching . . . . . . . . . . . . . . . . . . . . . .4-21 mapping to linear and physical address
usage on call to exception or interrupt spaces . . . . . . . . . . . . . . . . . . . . . . . . . 6-17
handler . . . . . . . . . . . . . . . . . . . . . . . .18-33 restart following an exception or interrupt . . 5-7
Stepping information, following processor state (context) . . . . . . . . . . . . . . . . . . . .6-2, 6-3
initialization or reset . . . . . . . . . . . . . .8-5 structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1
STI instruction . . . . . . . . . . . . . . . . . . . . . . . . . .5-9 switching . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-3
STPCLK# pin . . . . . . . . . . . . . . . . . . . . . 5-2, 15-15 task management data structures. . . . . . . . 6-4
STR instruction. . . . . . . . . . . . . . . . . . . . . 3-18, 6-8 Task-state segment (see TSS)
STRT instruction . . . . . . . . . . . . . . . . . . . . . . .2-20 Test registers . . . . . . . . . . . . . . . . . . . . . . . . 18-25
SUB instruction . . . . . . . . . . . . . . . . . . . . . . . . .7-4 TF (trap) flag, EFLAGS register . 2-8, 5-18, 12-10,
Supervisor mode 15-2, 15-10, 15-12, 15-14, 16-6, 16-26

INDEX-16
 INDEX

TI (table indicator) flag, segment selector . . . . .3-8 task register. . . . . . . . . . . . . . . . . . . . . . . . . 6-8

Timer, local APIC . . . . . . . . . . . . . . . . . . . . . . .7-43 using 16-bit TSSs in a 32-bit environment 18-29
Time-stamp counter virtual-mode extensions . . . . . . . . . . . . . 18-28
description of . . . . . . . . . . . . . . . . . . . . . .15-14 TSS descriptor
reading . . . . . . . . . . . . . . . . . . . . . . . . . . . .2-22 B (busy) flag . . . . . . . . . . . . . . . . . . . . . . . . 6-7
software drivers for . . . . . . . . . . . . . . . . . .15-18 initialization for multitasking . . . . . . . . . . . 8-13
TLBs structure of . . . . . . . . . . . . . . . . . . . . . . . . . 6-6
description of . . . . . . . . . . . . . . . 3-19, 9-1, 9-4 TSS segment selector
flushing . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-17 field, task-gate descriptor . . . . . . . . . . . . . . 6-8
invalidating (flushing) . . . . . . . . . . . . . . . . .2-21 writes. . . . . . . . . . . . . . . . . . . . . . . . . . . . 18-28
relationship to PGE flag . . . . . . . . . 3-27, 18-23 Type
relationship to PSE flag . . . . . . . . . . 3-22, 9-17 checking . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
TMR (Trigger Mode Register), local APIC . . . .7-30 field, MTRRdefType register . . . . . . . . . . . 9-21
TPR (task priority register), local APIC . . . . . .7-31 field, MTRRphysBasen register . . . . . . . . 9-24
TR (trace message enable) flag, DebugCtlMSR field, segment descriptor .3-11, 3-13, 3-15, 4-2,
register . . . . . . . . . . . . . . . . . . . . . .15-12 4-6
Transcendental instruction accuracy . . 18-9, 18-18 of segment . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
Translation lookaside buffer (see TLB)
Trap gates
difference between interrupt and trap U
gates . . . . . . . . . . . . . . . . . . . . . . . . . . .5-18 UD2 instruction . . . . . . . . . . . . . . . . . . . .5-28, 18-3
for 16-bit and 32-bit code modules . . . . . . .17-2 UE (numeric overflow exception) flag, FPU status
handling a virtual-8086 mode interrupt or word . . . . . . . . . . . . . . . . . . . . . . . 11-21
exception through . . . . . . . . . . . . . . . .16-17 Uncached (UC) memory type
in IDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5-13 description of. . . . . . . . . . . . . . . . . . . . . . . . 9-5
introduction to . . . . . . . . . . . . . . . . . . . . 2-3, 2-4 effect on memory ordering . . . . . . . . . . . . 7-10
layout of . . . . . . . . . . . . . . . . . . . . . . . . . . .5-13 use of . . . . . . . . . . . . . . . . . . . . . . . . . .8-9, 9-8
Traps Undefined
description of . . . . . . . . . . . . . . . . . . . . . . . .5-5 opcodes. . . . . . . . . . . . . . . . . . . . . . . . . . . 18-6
restarting a program or task after . . . . . . . . .5-7 Underflow, FPU stack. . . . . . . . . . . . . . . . . . 11-17
TS (task switched) flag, CR0 control Unit mask field, PerfEvtSel0 and PerfEvtSel1 MSRs
register . . . . . . . . . . . . . 2-14, 5-30, 6-12 (P6 family processors) . . . . . . . . . 15-17
TSD (time-stamp counter disable) flag, CR4 control Un-normal number . . . . . . . . . . . . . . . . . . . . 18-10
register .2-17, 4-25, 15-15, 15-18, 18-22 User mode
TSS description of. . . . . . . . . . . . . . . . . . . . . . . 4-31
16-bit TSS, structure of. . . . . . . . . . . . . . . .6-19 U/S (user/supervisor) flag . . . . . . . . . . . . . 4-31
32-bit TSS, structure of. . . . . . . . . . . . . . . . .6-4 User-defined interrupts . . . . . . . . . . . . . . .5-4, 5-55
CR3 control register (PDBR) . . . . . . . 6-6, 6-17 USR (user mode) flag, PerfEvtSel0 and
description of . . . . . . . . . . . . 2-3, 2-4, 6-1, 6-4 PerfEvtSel1 MSRs (P6 family
EFLAGS register. . . . . . . . . . . . . . . . . . . . . .6-4 processors). . . . . . . . . . . . . . . . . . 15-16
EIP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6-4 U/S (user/supervisor) flag
executing a task . . . . . . . . . . . . . . . . . . . . . .6-3 page-directory entry . . . . . . . . . . 4-2, 4-3, 4-31
floating-point save area . . . . . . . . . . . . . .18-14 page-table entries . . . . . . . . . . . . . . . . . . 16-11
general-purpose registers. . . . . . . . . . . . . . .6-4 page-table entry . . . . . . . . 3-26, 4-2, 4-3, 4-31
initialization for multitasking . . . . . . . . . . . .8-13
invalid TSS exception . . . . . . . . . . . . . . . . .5-35 V
I/O map base address field. . . . . . . . 6-6, 18-29
V (valid) flag, MTRRphysMaskn register . . . . 9-24
I/O permission bit map . . . . . . . . . . . . . . . . .6-6
Variable-range MTRRs, description of . . . . . . 9-23
LDT segment selector field . . . . . . . . . 6-5, 6-17
link field. . . . . . . . . . . . . . . . . . . . . . . . . . . .5-19 VCNT (variable range registers count) field,
order of reads/writes to . . . . . . . . . . . . . . .18-28 MTRRcap register . . . . . . . . . . . . . 9-20
page-directory base address (PDBR). . . . .3-23 Vectors
pointed to by task-gate descriptor. . . . . . . . .6-8 exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
previous task link field. . . . . . . . 6-4, 6-14, 6-16 interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
privilege-level 0, 1, and 2 stacks. . . . . . . . .4-21 reserved . . . . . . . . . . . . . . . . . . . . . . . . . . 7-15
referenced by task gate . . . . . . . . . . . . . . .5-19 VERR instruction . . . . . . . . . . . . . . . . . .2-20, 4-27
VERW instruction . . . . . . . . . . . . . . . . . .2-20, 4-27
segment registers . . . . . . . . . . . . . . . . . . . . .6-4
VIF flag, EFLAGS register . . . . . . . . . . . . . . . 18-6
T (debug trap) flag . . . . . . . . . . . . . . . . . . . .6-6

INDEX-17
INDEX

VIF (virtual interrupt) flag, EFLAGS register . .2-10 WRMSR instruction 2-22, 2-23, 4-25, 7-12, 15-11,
VIP (virtual interrupt pending) flag, EFLAGS 15-15, 15-16, 15-18, 15-20, 18-4, 18-38
register . . . . . . . . . . . . . . . . . . 2-10, 18-6 WT (write through) memory type . . . . . . . .9-6, 9-8
Virtual memory . . . . . . . . . . . . . . . . . . . . . . 2-5, 3-1
Virtual-8086 mode
8086 emulation . . . . . . . . . . . . . . . . . . . . . .16-1 X
description of . . . . . . . . . . . . . . . . . . . . . . .16-9 XADD instruction . . . . . . . . . . . . . . . . . . .7-4, 18-5
emulating 8086 operating system calls. . .16-25 XCHG instruction . . . . . . . . . . . . . . . 7-3, 7-4, 7-10
enabling . . . . . . . . . . . . . . . . . . . . . . . . . . .16-9 XOR instruction . . . . . . . . . . . . . . . . . . . . . . . . 7-4
entering. . . . . . . . . . . . . . . . . . . . . . . . . . .16-11
exception and interrupt handling, Z
overview . . . . . . . . . . . . . . . . . . . . . . .16-15
ZF flag, EFLAGS register . . . . . . . . . . . . . . . . 4-27
exceptions and interrupts, handling through a
task gate . . . . . . . . . . . . . . . . . . . . . . .16-19
exceptions and interrupts, handling through a
trap or interrupt gate . . . . . . . . . . . . . .16-17
handling exceptions and interrupts through a
task gate . . . . . . . . . . . . . . . . . . . . . . .16-20
IOPL sensitive instructions . . . . . . . . . . . .16-14
I/O-port-mapped I/O . . . . . . . . . . . . . . . . .16-15
leaving . . . . . . . . . . . . . . . . . . . . . . . . . . .16-13
memory mapped I/O . . . . . . . . . . . . . . . . .16-15
native 16-bit mode . . . . . . . . . . . . . . . . . . .17-1
overview of . . . . . . . . . . . . . . . . . . . . . . . . .16-1
paging of virtual-8086 tasks . . . . . . . . . . .16-10
protection within a virtual-8086 task . . . . .16-11
special I/O buffers. . . . . . . . . . . . . . . . . . .16-15
structure of a virtual-8086 task . . . . . . . . . .16-9
virtual I/O . . . . . . . . . . . . . . . . . . . . . . . . .16-14
Virtual-8086 tasks
paging of . . . . . . . . . . . . . . . . . . . . . . . . . .16-10
protection within . . . . . . . . . . . . . . . . . . . .16-11
structure of . . . . . . . . . . . . . . . . . . . . . . . . .16-9
VM (virtual-8086 mode) flag, EFLAGS register .2-9
VME (virtual-8086 mode extensions) flag, CR4
control register . . . . . . . . . . . 2-17, 18-22

W
WAIT instruction. . . . . . . . . . . . . . . . . . . . . . . .5-30
WAIT/FWAIT instructions. . . . . 18-8, 18-18, 18-19
WB (write back) memory type . . . . . . . . . . 9-6, 9-8
WBINVD instruction . . 2-21, 4-25, 7-12, 9-15, 18-5
WC (write combining)
flag, MTRRcap register. . . . . . . . . . . . . . . .9-21
memory type . . . . . . . . . . . . . . . . . . . . . 9-6, 9-8
WP (write protected) memory type. . . . . . . . . . .9-7
WP (write protect) flag, CR0 control register . 2-14,
4-32, 18-22
Write
forwarding . . . . . . . . . . . . . . . . . . . . . . . . . . .7-8
hit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9-5
Write back (WB) memory type . . . . . . . . . . . . .7-10
Write buffer
description of . . . . . . . . . . . . . . . . . . . . . . . .9-4
in Intel Architecture processors . . . . . . . .18-36
operation of. . . . . . . . . . . . . . . . . . . . . . . . .9-17
Write-back caching. . . . . . . . . . . . . . . . . . . . . . .9-5

INDEX-18