CS6456: Graduate

Operating Systems
Brad Campbell – [email protected]
https://www.cs.virginia.edu/~bjc8c/class/cs6456-f19/

Slides modified from CS162 at UCB

1
2
Memory Multiplexing Goals
1. Protection
• Private memory of other processes/OS remains
private
• Even from malicious processes
2. Controlled Overlap
• Want processes and OS to share memory for
communication, cooperation
• Or just to save space (e.g., shared libraries)
3. Uniform View of Memory
• No need to change binary file's addresses
when we want to execute two instances of it
• No need to change a program to let it use
more memory
General Address Translation
Virtual Physical
Addresses Addresses
CPU MMU

• Each process, kernel have a different address

space
• Address Space = All addresses a process can touch
• Two views of memory:
1. View from the CPU – what the program sees (virtual)
2. View from physical memory (physical)
3. MMU is hardware that converts between the two
Segmentation
1
1 4

2
3 2
4
3

user view of physical

memory space memory space
• Manage each region of address space
as an independent unit
• One process uses multiple contiguous
allocations of memory, called segments
Example: Four Segments (16-bit Addresses)
Seg ID # Base Limit
Seg Offset 0 (code) 0x4000 0x0800
15 14 13 0 1 (data) 0x4800 0x1400
Virtual Address Format 2 (shared) 0xF000 0x1000
3 (stack) 0x0000 0x3000
SegID = 0 0x0000
0x0000

SegID = 1 0x4000
0x4000 0x4800
0x5C00
0x8000
Space for
0xC000
Other Apps

0xF000

Virtual Physical
Address Space Address Space
Segmentation
• Internal fragmentation?
• We can have large, unused gaps in virtual address space 0x0000
that don’t correspond to any allocated segment
• But segments must be contiguous in physical memory
0x4000
• Reserve extra space at beginning?
0x4800
• Expansion is difficult
0x5C00
• External fragmentation?
• Yes – may need to rearrange chunks of memory
• Segments are smaller than full process memory image,
hopefully less work
• Sharing? 0xF000
• Yes! Allow virtual addresses in two processes to map to
same segment in physical mem. Virtual
• But can only share at segment granularity Address Space
• One shared library per segment?
• Will we run out of segments?
Swapping
• Problem: What if all segments don’t fit in memory?
• Breaks the illusion of a dedicated machine
• Potential solution: When process isn’t running, move one
or more segments to disk
• Load segment back into main memory before resuming
• An “extreme context switch”
• Need to track whether or not a segment is resident in
memory in processor segment map
• For modern systems, swapping performance is
unacceptable
• Maybe swap some idle memory to disk, but never something
active
How can we do better?
• Allocate at finer granularity than segments
• Allocate memory in fixed-size chunks: pages
• All allocations are full pages
• All holes are multiples of a page size
• Expanding memory footprint: allocate new page
• No need to relocate existing allocations
• No struggle to find free region of proper size
What’s the catch?
• More bookkeeping
• Need to track physical memory location of each page
• Need to track metadata (valid, permissions) per page
• Page size represents a tradeoff
• Larger: Fewer pages to keep track of, more internal
fragmentation
• Smaller: Less wasted space, more pages to track
• Less locality: Sequential access in virtual space
is not necessarily sequential in physical space
• Jumping across page boundaries
Implementing Paging
Virtual
Virtual Address: Page # Offset

PageTablePtr page #0 V,R Physical

Frame # Offset
page #1 V,R
page #2 V,R,W Physical Address
PageTableSize > page #3 V,R,W Check Perm
page #4 N
Access Error page #5 V,R,W Access
Error

Page Table: Process-Specific map for bookkeeping

• Tells us in which physical page frame our virtual page
is stored
• Virtual page number is index into table
• Table itself resides in physical memory
• Trigger fault if page access is not permitted
Address Translation: Overhead
• Segmentation: Segment Table
• One table entry per segment
• Few segments, so typically small
• Potentially store in registers

• Paging: Page Table

• Allocate physical memory to store table
• Extra memory access per operation
• Once to translate address, second to do actual work
• Tables can get large (1 million entries on 32-bit
machine)
What about Sharing?
Virtual Address Virtual
(Process A): Page # Offset

PageTablePtrA page #0 V,R

page #1 V,R
page #2 V,R,W
page #3 V,R,W
page #4 N
Shared
page #5 V,R,W Page
PageTablePtrB page #0 V,R
page #1 N This physical page
page #2 V,R,W
appears in address
page #3 N
page #4 V,R space of both processes
page #5 V,R,W

Virtual Address Virtual

(Process B): Page # Offset
Paging: Allocation
Page Table
Virtual memory view 11111 11101 Physical memory view
1111 1111 11110 11100
stack 11101 null

1110 0000
11100 null stack 1110 0000
11011 null
11010 null
11001 null
What
1100 0000 11000 null
happens if 10111
10110
null
null
stack grows 10101 null
10100 null
to 1110 0000? 10011 null
heap 10010 10000
1000 0000 10001
10000
01111
01110
heap 0111 000
01111 null
01110 null data
01101 null 0101 000
data 01100 null
0100 0000 01011 01101
01010 01100
01001 01011
01000 01010
00111 null code
00110 null 0001 0000
code 00101 null
0000 0000 00100 null 0000 0000
00011 00101
page #offset 00010 00100
00001 00011
00000 00010
 Summary: Paging
Page Table
Virtual memory view 11111 11101 Physical memory view
1111 1111 11110 11100
stack 11101 10111

1110 0000
11100 10110 stack 1110 0000
11011 null
11010 null
11001 null
1100 0000 11000 null
10111 null stack
10110 null
10101 null
10100 null Allocate
heap
10011
10010
null
10000
new pages
1000 0000 where
10001 01111
heap
10000
01111
01110
null room!0111 000
01110 null data
01101 null 0101 000
data 01100 null
0100 0000 01011 01101
01010 01100
01001 01011
01000 01010
00111 null code
00110 null 0001 0000
code 00101 null
0000 0000 00100 null 0000 0000
00011 00101
page #offset 00010 00100
00001 00011
00000 00010
Paging: Hardware Support
• Special Register to store page table pointer
• Modified in kernel mode only
• Change register values on context switch

• Page Table Lookup

• Two physical memory accesses per virtual
access
• Later: making this faster
• Modern systems do all of this in hardware
Page Table Issues
• What if address space is mostly unused?
• Say we have 4K pages and 2GB of virtual address space
• 512K page table entries
• Megabytes of space for table itself

• How about 64-bit x86 machine? ~256 TB of virt addr space

• 64 billion page table entries
• Gigabytes of space for table itself

• Even though most of this address space is unused by the

process in its lifetime
• Apply tree structure to virtual memory: Multilevel Page Tables
Two-Level Page Tables
Physical Physical
Offset
10 bits 10 bits 12 bits Address:
Page #
Virtual Virtual Virtual
P1 index P2 index Offset
Address:
4KB

PageTablePtr

4 bytes
• A tree of page tables
• Each "node" has a fixed
size
• x86: 1024 4-byte entries
• Still just one register to
change on context switch 4 bytes
Two-Level Page Tables
Physical Physical
Offset
10 bits 10 bits 12 bits Address:
Page #
Virtual Virtual Virtual
P1 index P2 index Offset
Address:
4KB

PageTablePtr

4 bytes

Big gaps in virtual

address space don't need
second level page tables
4 bytes
Two-Level Page Tables
Physical Physical
Offset
10 bits 10 bits 12 bits Address:
Page #
Virtual Virtual Virtual
P1 index P2 index Offset
Address:
4KB

PageTablePtr

4 bytes

But we're also making

two memory accesses
just for translation
4 bytes
Two-Level Paging Example
Virtual memory view Page Table Page Physical memory view
1111 1111 (level 1) Tables
stack (level
11 2)
11101

1111 0000
10 11100 stack 1110 0000
01 10111
00 10110

1100 0000
stack
111 11 null
110 null 10 10000
101 null 01 01111
00 01110
heap 100
1000 0000 011 null
010 heap 0111 000
001 null
000 11 01101
10 01100 data
01 01011 0101 000
data 00 01010
0100 0000
11 00101

page2 #
10 00100
01 00011
code
0001 0000
code 00 00010
0000 0000 0000 0000
page1 #offset
Page Table Discussion
• Good Points:
• Easy memory allocation (fixed size chunks)
• Easy sharing among processes
• No physical memory used for holes in virtual
address sp.

• Bad Points:
• Overhead: At least 1 pointer per 4K page
• Page tables need to be contiguous (tricky
allocation)
• Clever design (x86) can limit them to one page
• Several memory accesses per translation
• Later: making this faster
Page Tables for Huge Address Spaces
• Want full 64 bits of address space?
• How many levels of indirection needed?

2 bits 10 bits 10 bits 10 bits 10 bits 10 bits 12 bits

Virtual P2 Virtual P3 Virtual P4 Virtual P5 Virtual P6
Offset
Index Index Index Index Index

• Overhead: Extra space for tables in phys. mem.

• And additional memory accesses per
translation
• 64-bit x86 does not do this
• Virtual Addresses are 48 bits long
• 4-level page table
Alternative: Inverted Page Table
• Until Now: Size of page table proportional to size of
virtual address space
• New Structure: Size of page table proportional to
size of physical address space
• Good if most of address space not in use
• Particularly 64-bit address space
• Bad: Complexity Virtual
Offset
Page #
• How do we deal with
hash collisions?
Physical
Hash Page # Offset
Table
Combining Segments & Pages

• One approach: Each segment has its own page

table
• Fragmentation/allocation advantages of pages
Virtual Virtual Virtual
Seg # Page # Offset
Address:
page #0 V,R
page #1 V,R Physical
Base0 Limit0 V Page # Offset
Base1 Limit1 V page #2 V,R,W
Base2 Limit2 V Physical Address
page #3 V,R,W
Base3 Limit3 N page #4 N
Base4 Limit4 V page #5 V,R,W Check Permissions
Base5 Limit5 N
Base6 Limit6 N
Access
Base7 Limit7 V
> Access
Error Error
Sharing a Segment
Virtual Virtual
Process A: Seg # Page # Offset page #0 V,R
page #1 V,R
page #2 V,R,W
Base0 Limit0 V page #3 V,R,W
Base1 Limit1 V
page #4 N
Base2 Limit2 V
Base3 Limit3 N page #5 V,R,W
Base4 Limit4 V Shared Segment
Base5 Limit5 N
Base6 Limit6 N Base0 Limit0 V
Base7 Limit7 V Base1 Limit1 V
Base2 Limit2 V
Base3 Limit3 N
Base4 Limit4 V
Base5 Limit5 N
Base6 Limit6 N
Base7 Limit7 V

Virtual Virtual
Process B: Seg # Page # Offset
Combining Segments & Pages

• Good:
• Just need one page table per segment, even if
shared
• Easy memory allocation (no external fragmentation)
• Easy sharing of segments, just change segment
table entry
• Can still share at page granularity if necessary

• Bad
• Overhead: 1 pointer per page
• At least two extra lookups per memory access
 Address Translation Comparison
Advantages Disadvantages
Simple Fast context External fragmentation
Segmentation switching: Segment
mapping maintained
by CPU
Paging (single- No external Large table size ~ virtual
level page) fragmentation, fast memory
easy allocation Internal fragmentation

Paged Table size ~ # of Multiple memory references

segmentation pages in virtual per page access
memory, fast easy
Two-level pages
allocation

Inverted Table Table size ~ # of Hash function more

pages in physical complex
memory No cache locality of page
Real World: 32-bit x86
• Has both segmentation and paging
• Segmentation different from what we've
described
• Segment identified by instruction, not address

• Note: x86 actually offers multiple modes of

memory operation, we'll talk about common
one
 Real Example: Intel x86 (Old Days)
80386 Special Registers

Today: Register points to

segment table in physical
memory
Intel x86
• Six segments: cs (code), ds (data), ss
(stack),
es, fs, gs (extras)
• Instructions identify segment to use
• mov [es:bx], ax
• Some instructions have default segments, e.g.
push and pop always refer to ss (stack)
• Underused in modern operating systems
• In 64-bit x86, only fs and gs support
enforcement of base and bound
Real Page Table Entries
• So far, we've said they contain a physical
frame #
• Or the frame # of a page table lower on the
"tree"
• But we want some extra information
• Present/valid bit: indicate unallocated
regions of physical address space
• Protection bits, e.g. to set certain regions
of memory to read-only for sharing
32-bit x86 Page Table Entry
• 10/10/12 Split of Virtual Address
• Top-level page tables called directories

PCD
PWT
Page Frame Number Free
0 L D A U W P
(Physical Page Number) (OS)
31-12 11-9 8 7 6 5 4 3 2 1 0

• PFN: Physical Page number of page or next page table

• P: Present Bit (Is page mapping valid?)
• W: Is this page writable?
• U: Can we access this page while in user mode?
• A: Set by hardware when page is first accessed
• D: Page marked dirty if it is ever modified
• PWT: Write-through cache behavior
• PCD: Disable caching for this page
Paging Tricks
• What does it mean if a page table entry doesn't
have the valid (present) bit set?
• Region of address space is invalid or
• Page is not loaded and ready yet

• When program accesses an invalid PTE, OS gets

an exception (a page fault or protection fault)
• Options
• Terminate program (access was actually invalid)
• Get page ready and restart instruction
Example Paging Tricks
1. Demand Paging: Swapping for pages
• Keep only active pages in memory
• Remember: not common on modern systems,
except perhaps when first loading program
• Response: Load in page from disk, retry operation
2. Copy on Write (remember fork?)
• Temporarily mark pages as read-only
• Allocate new pages when OS receives protection
fault
3. Zero-Fill on Demand
• Slow to overwrite new pages with all zeros
• Page starts as invalid, zero it out when it's
actually used
Recall: Dual-Mode Operation
• Process cannot modify its own page tables
• Otherwise, it could access all physical memory
• Even access or modify kernel
• Hardware distinguishes between user mode
and kernel mode with special CPU register
• Protects Page Table Pointer
• Kernel has to ensure contents of page tables
themselves are not pointed to by any mapping
• Remember: Page table can mark some frames
as off limits for user-mode processes
Synchronous Exceptions
• System calls are on example of a synchronous
exception (a "trap" into the kernel)
• Other exceptions: page fault, access fault, bus
error
• rerun the triggering instruction
• After fixing something
• Relies on precise exceptions
Faulting

Faulting
User
Inst.

Inst.
Load
OS Page
Precise Exceptions

• Definition: Machine's state is as if the program

executed up to the offending instruction
• Hardware has to complete previous
instructions
• Remember pipelining
• May need to revert side effects if instruction was
partially executed

• OS relies on hardware to enforce this property

Caching

• Cache: Repository for copies that can be

accessed more quickly than the originals

• Goal: Improve performance

• We'll see lots of applications!

• Memory
• Address translations
• File contents, file name to number mappings
• DNS records (name to IP addr mappings)
Why Bother with Caching?
Growing latency gap between CPU and Memory

1000 CPU
µProc
“Moore’s Law” 60%/yr.
(2X/1.5yr)
Performance

100 Processor-Memory
Performance Gap:
(grows 50% / year)
10
“Less’ Law?” DRAM
DRAM
9%/yr.
1 (2X/10 yrs)
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
Time
Average Access Time
• Average Access Time =
(Hit Rate x Hit Time) + (Miss Rate x Miss
Time)

• Intuition: Caching is good when most

accesses are for a small portion of the
possible items
• "Most Accesses": High Hit Rate
• "Small Portion": Hits can be faster than
misses
Why Does Caching Help? Locality!
Probability
of reference

0 2n - 1
Address Space

• Temporal Locality (Locality in Time):

• Keep recently accessed data items closer to
processor
• Spatial Locality (Locality in Space):
• Move contiguous blocks to Main
the upper levels
To Processor Cache Memory

Blk X
From Processor Blk Y
Memory Hierarchy
• Take advantage of the principle of
locality to:
• Present as much memory as in the cheapest
technology
• Provide access
Processor at speed offered by the fastest
technology
Control Tertiary
Secondary
Second Main Storage Storage
(Tape)
Cache
On-Chip
Registers

Level Memory (Disk)

Datapath
Cache (DRAM)
(SRAM)

Speed (ns): 1s 10s-100s 100s 10,000,000s 10,000,000,000s

(10s sec)
Size (bytes): 100s Ks-Ms Ms-Gs (10s ms)
Gs-Ts Ts-Ps
Types of Cache Miss

1. Compulsory ("cold start"): First access to a

block
• Insignificant in any long-lived process
2. Capacity: Not enough space in cache
3. Conflict: Memory locations map to same
cache location
4. Coherence (invalidation): Memory updated
externally
• e.g. multi-core system, or on I/O
Where Can a Block Go?
• Example: Block 12 placed in 8
32-Block Address Space:
block cache

Block 1111111111222222222233
no. 01234567890123456789012345678901

Direct Set associative: Fully associative:

mapped: block 12 can go block 12 can go
block 12 can go anywhere in set 0 anywhere
only into block 4 (12 mod 4)
Block (120 mod
1 2 3 48)5 6 7 Block 01234567 Block 01234567
no. no. no.

Set Set Set Set

0 1 2 3
Which block should be replaced on a miss?

• Easy for Direct Mapped: Only one possibility

• Set Associative or Fully Associative:
• Random
• LRU (Least Recently Used)

• Miss rates for a workload:

2-way 4-way 8-way
Size LRU Random LRU Random LRU
Random
16 KB 5.2% 5.7% 4.7% 5.3% 4.4% 5.0%
64 KB 1.9% 2.0% 1.5% 1.7% 1.4% 1.5%
256 KB1.15% 1.17% 1.13% 1.13% 1.12% 1.12%
What Happens When We Write to a Cached
Address?
• Write Through: Update both cache block and
corresponding location in main memory
• Simple to implement, but now we wait for writes?

• Write Back: Update only the cached copy at

time of write. Update main memory when
cache block is removed later on.
• Repeated writes not all sent to RAM
• A read may trigger a write (evicting a cache block)
• More complex (need modified "dirty" bit)
Caching Applied to Address Translation

Virtual TLB Physical

Address Cached? Address
CPU Yes Physical
No Memory
a ve lt
S su
Re
Translate
(MMU)

Data Read or Write

(untranslated)
Caching Address Translations
• Locality in page accesses?

• Yes: Spatial locality, just like CPU cache

• TLB: Cache of page table entries

TLB and Context Switches
• Do nothing upon context switch?
• New process could use old process's address
space!

• Option 1: Invalidate ("flush") entire TLB

• Simple, but very poor performance

• Option 2: TLB entries store a process ID

• Called tagged TLB
• Requires additional hardware
TLB and Page Table Changes
• Think about what happens we we use fork
• OS marks all pages in address space as
read-only
• After parent returns from fork, it tries to
write to its stack
• Triggers a protection fault. OS makes a
copy of the stack page, updates page table.
• Restarts instruction.
TLB and Page Table Changes

What if TLB had cached the old read/write page

table entry for the stack page?

• OS marks all pages in address space as

read-only
• After parent returns from fork, it tries to
write to its stack
• Triggers a protection fault. OS makes a copy
of the stack page, updates page table.
• Restarts instruction.
How do we invalidate TLB entries?

• Hardware could keep track of where page

table for each entry is and monitor that
memory for updates…
• Very complicated!
• Especially for multi-level page tables and
tagged TLBs

• Instead: The OS must invalidate TLB

entries
• So TLB is not entirely transparent to OS