The Java VM

Architecture & APIs

2003-12087

Contents

Java VM Architecture

Java Runtime Structure

Memory Management
Execution Relatives
Exception and Error Management
Class File Structure
Class Verification
Native Method Support(JNI)

Java APIs

Java Platforms Overview

Java APIs(J2SE)

Java VM Architecture

- Java Runtime Structure

Java VM

Usually referred to Java Runtime(JRE)

Mainly used to execute programs written in Java

Typical runtime system includes:

Execution Engine Virtual(or real hardware ex. ARM)

processor for executing bytecodes
Memory Manager Allocate memory for instances and
arrays and perform garbage collection
Error and Exception Manager Deal with exception

Java VM Architecture

- Java Runtime Structure

Typical runtime system includes(contd):

Native Method Support for calling c/c++ methods

Threads Interface supporting threads and monitors
Class Loader dynamically load Java classes from Java
class files
Security Manager verify that classes are safe and
controlling access to system resources

Java VM Architecture

- Java Runtime structure

Java VM Architecture

- Memory Management

Memory Area

Divided into cells or slots

Slot can usually hold a single data item
All addressing is in terms of the logical memory cells.

Java VM Architecture

- Memory Management

The Method Area

Type Information

Fully qualified name of the type of itself, superclasses,

superinterfaces
Whether or not the type is a class or an interface
Types modifiers(public, abstract, final)

Constant Pool(more detail later)

Set of constants

Field Information

Symbolic references, literals

Name, type, modifiers

Method Information

Name, return & arg. type, modifiers

Bytecodes, exception table, stack frame size
(not native or abstract methods)

Java VM Architecture

- Memory Management

The Method Area

Class Variables

A Reference to Class ClassLoader

A Reference to Class Class

Class information can be accessed through class object

Method Table

Class variables are shared among all instances

Non-finals as part of data for the type that declares them
Finals(constants) as part of data for the type that uses them(get a
copy)

Data structures that speed up access to the raw data

ex) method table can have references to instance methods inherited
from superclass

Method area also can be garbage collected as an unreferenced

instance

Java VM Architecture

- Memory Management

The Heap

The memory for the new object is allocated from a single

heap.
Every application has its own heap
But, All threads share it!
So, careful synchronization of multi-threaded access to
object is needed.
Allocation instruction exists, but freeing instruction doesnt
exists!
Freeing memory occupied by objects that are no longer
referenced is responsible for a garbage collector.
Method area and heap may be on the same heap.

Java VM Architecture

- Memory Management

Object Representation in the heap

Objects can be freely represented in heap.

Two possible solution

- Divides the heap into two parts

- easy for VM to combat heap
fragmentation
- need dereferencing two pointers

(a)

Java VM Architecture

- Memory Management

Object

Representation in the heap

- dereference pointer only once
- make moving objects more
complicated

(b)

Java VM Architecture

- Memory Management

Method

table

Can play an important role in achieving good VM

performance.
May not exist in some implementation that have
extremely low memory requirements.
Method table includes :

Size of methods stack frame

Methods bytecodes
An exception table

Java VM Architecture

- Memory Management

Arrays

in heap

Java VM Architecture

- Memory Management

The Program Counter

Each thread has its own PC.

Can be a native pointer or an offset from the beginning of
methods bytecodes.
If a thread is executing a native method, pc is undefined.

The Stack

Each thread has its own stack area too.

Local variables and operands are thread safe.
Used for local, operand storage
References, not actual objects can exist in stack.
As each method is called, a stack frame is allocated.

Java VM Architecture

- Memory Management

Stack

Stack depth can be estimated at compiletime(will be discussed later)

Locals(include arguments) :

Frame Structure

Instance method has hidden this reference

on its first local slot.
Byte, short, char are converted into int
(due to asymmetry of instruction set)

Frame Data :

Data to support constant pool resolution

Exception table
Normal method return address

Arguments
Locals
Frame data
Operands
Stack Frame
Structure

Java VM Architecture

- Memory Management

Possible

Implementations of the Java Stack

Example code :
public static void addAndPrint() {
double result = addTwoTypes(1, 88.88);
System.out.println(result);
}
public static double addTwoTypes(int i, double d) {
return i + d;
}

Java VM Architecture

- Memory Management

Possible

Implementations of the Java Stack

- Right one uses stack area more efficiently.

- Also saves time because Java VM doesnt need to copy the parameter values.

Java VM Architecture

- Memory Management

Native Method Stack

A native method can access runtime data areas of VM and

also do anything else.
Native method calling is just calling another method within
itself, at the behest of the running Java program.

Java VM Architecture

- Memory Management

Memory

Hierachy

Java VM Architecture

- Execution Relatives

Data

Types

Java VM Architecture

- Execution Relatives

Data Types

Each data types are defined according to the values they

can have.
Every data type except Double and Long needs one
word(slot).
Boolean type
Treated as integer
Boolean arrays are implemented as byte array

Made by newarray 4, but handled by byte array instructions

ReturnAddress
Not visible to programmer
Used internally with subroutine instructions(jsr, ret)
Array Object
Special object support by instruction set
All of array elements have the same type

Java VM Architecture

- Execution Relatives

Instruction Set

Advantages
Stack based ISA - Stack is amenable to platform
independence.
Increase instruction set encoding density

- No instruction fields are needed for specifying registers

Disadvantages
Non-Orthogonal Instruction Set

8-bit opcode can only encode 256 instructions.

some datatypes(short, byte, char) are relegated to second class
status and receive less support in ISA

Hard to Extend

Extending the machine to support 96-bit or 128-bit floats and longs

cannot be done simply.
Use escape or wide opcode to create an extended instruction set.

Java VM Architecture

- Instruction Set

Instruction

Opcode byte + operand(more than zero)

Set Format

Operand can be either of index, immediate data or PCrelative offset.

Wide & escape code can be used to extend instruction
set.

Each of primitive types has its own instructions

that can operate on them.

Array access and type conversion instructions can only

operate on short, byte, and char type.

Java VM Architecture

- Instruction Set

Data-Movement Instructions

There can be different instructions for the same function.

- code density, interpretation performance are related
Pushing Constants onto the Stack
aconst_null, iconst_<n>, ldc(via constant pool),
bipush(direct)
Stack Manipulation
Local Variable relatives
iload (index), iload_<n>
iinc

Java VM Architecture

- Instruction Set

Data-Movement Instructions

Array relatives
newarray, anewarray, multianewarry, <x>aload, <x>astore,
arraylength
Object relatives
new <index(to constant pool)>
(get|put)(static|field)
(checkcast|instanceof) <index(to constant pool)>

Type Conversion
Functional Instructions

Only operates on int, float, double, long.

Operands are converted to standard number representation
before calculation
Convert back to platforms internal representation and be
pushed to stack after calculation.

Java VM Architecture

- Instruction Set

Control Flow Instructions

Designed to expose all control flow paths within a method

Method call

All control flow instructions use fixed, compile-time PC offset.(no

indirection)
Also, jump directly to a method via a fixed index into the constant
pool
This feature enables both complete code discovery and load-time
stack tracking.
invoke(virtual|static|special|interface) <index1,2>
Return PC is saved on a stack(in frame data area), but can not be
accessed directly(only through return)

* Quick instructions
Figure 5.11

Java VM Architecture
- Exceptions and Errors

Exceptions and Errors

Errors caused by either the application behavior and the

limitation of VM
Exception checked or unchecked
checked exception must be encapsulated by try/catch
clause.
unchecked(or runtime) exception caused by dynamic
behavior of program
All exceptions(and errors) must be handled somwhere.
If an exception is not handled by the method that throws the
exception, stack frame is popped until the exception is
handled by some handler.

Java VM Architecture
- Exceptions and Errors

Exception handler is implemented by miniature subroutines.

Use jsr, ret, goto instruction.

athrow throw exception dereferenced by class name on top of the stack

Use exception table to specify an exception handler.

ex)

From

To

Target

Type

12

96

Arithmetic
Exception

Internal data structure for exception table

ExceptionTable {
u2
from_pc;
u2
to_pc;
u2
handler_pc;
u2
catch_type;
}

Java VM Architecture
- Exceptions and Errors

Exception handler example

Example Java code

public class ExceptionTest {

public static void main(String args[]) {

try {
java.io.FileInputStream x
= new java.io.FileInputStream(myfile);
} catch(java.io.FileNotFoundException e) {
System.out.println(Not found);
} finally {
System.out.println(This must be executed);
}

Java VM Architecture

- Class File Structure

Magic Number

0xCAFEBABE(in big-endian order)

Constant Pool

constant - tag(u1) + length(u2) + bytes

Tag specify type of constant

CONST_???

Descriptors

BaseType : B, C, D, F, I, J(long), S, Z(boolean)

ObjectType : L<classname>
ArrayType : [(BaseType|ObjectType|ArrayType)
Method : (<argument_types>)<return_types>
ex) I
=>
int a;
[[J
[Ljava/lang/Object
[[[Z

=>
=>
=>

long[][] a;
java.lang.Object[] a;
boolean[][][] a;

()I
()Ljava/lang/String;
([BII)V

=>
=>
=>

int a();
String a();
void a(byte[], int, int)

Java VM Architecture

- Class File Structure

Access Flags

This, Super Classes, Interface

Specify modifier of class, interface, method, and

field - ACC_???
Specified by indexed constant in constant pool

Field, Method

access_flags(u2) + name_index(u2) +
descriptor_index(u2) + attribute_count(u2) +
attributes_info
name, descriptor are on the constant pool

Java VM Architecture

- Class File Structure

Attribute

Method code, constant value for finals, exception

that a method may throw
Innerclass, LineNumberTable, LocalVariableTable,
Source file name
Code attribute max_stack, max_locals, code,
exception_table

Class file limitation

u2 - constant pool entries, field count, method count,

bytecode length(per method), local variables, operand
stack, exception table length
u1 - array dimensions, arg. to a method

Java VM Architecture

- Class File Structure

ClassStruct_java.txt

ClassStruct.txt

Java VM Architecture

- Class Verification

Class Verification
Class binaries are sometimes unsafe and may crash VM.
Must take all control path and prove that the program is safe in
each case.
Halting Problem
Studied by Alan Turing and Kurt Godel
In general case, it is not possible to take a description of a
program and decide whether or not the program will complete, let
alone whether is behaves well or not.
Operand Stack Tracking
For each alternative way in a method for reaching an instruction X,
the stack state and the local variable state must be equivalent.
Figure 5.10

(b) stack size is different. (c) operand types are different

Java VM Architecture

- Class Verification

Operand Stack Tracking

ex) iconst_4
istore_1
Loop:
aconst_null
iinc 1 1
iload_1

Operand stack is not equivalent at Loop.

Stack tracking can be done in static-time

Because control flows are determined in static-time.
Execution engine doesnt need to perform runtime checks for
following items :

Stack limits
Types of arguments to JVM instructions
Accesses or assignments to local variables

Java VM Architecture

- Class Verification

Passing verification
Structure

Check that the class file structure is met.

Check also the contents of bytecode

ex) this_class field must be the index of a CONSTANT_Class

Magic field must be the value 0xCAFEBABE

Check that all byte code offsets are within method boundary.
Type of constant and constant referencing instruction must be the same.

Environment

Other classes that one class depends, and the methods and fields of
those
Type conflict and access conflict
Doesnt immediately check if the referenced class really exist.

The constant pool can also contain references to classes that havent been
loaded yet.

JVM verifier tries to delay the checks until they are necessary.

Speed up the initial loading time for a class

Java VM Architecture

- Class Verification

Passing verification

Environment
ex) invokenonvirtual myclass/funmethod()LFunClass

-> putfield myclass/myfield LFunClass;

-> putfield myclass/myfield LAnotherClass;

Content
Each instruction should bee invoked with the correct types
for its operands and stack values.

The maximum stack length must not be exceeded.

Use pop2, pop to retrieve long value from the stack

Stack size is specified in the Code attribute

Dont use the stack in complex ways.

Only push items onto the stack just before they are needed.

Java VM Architecture

- Class Verification

Working of bytecode verifier

Trace all the static control flow paths and simulate a stack
symbolically.
Steps :
When instruction is first encountered, stores stack and local
var. state in table

separate entry is maintained in the table for every instruction in

bytecode

Then check that Instruction is begin run with the correct

types
Emulate the instructions effect on the stack and local var.
When a branch instruction is met, look at all the possible
destinations

If a destination has not been seen previously, verifier recursively

examines.
Else, verifier compares the current state with recorded state

Java VM Architecture

- Class Verification

Working of bytecode verifier

State comparison
Two states are identical move forward
Two states are incompatible verifier complains!
Two state are compatible merges two states

int
float
DataInputStream

int
float
BufferedInputStream

int
float
Vector

int
float
int

merge

incompatible

int
float
FilterInputStream

Java VM Architecture

- Native Method Support

Java

Java side and native side can interoperate each

other by JNI.
Native side method invocation from Java side

Native Interface(JNI)

Use native keywords for modifier of function

Generate header file for native function by javah
ex) JNITest_java.txt ->JNITest_header.txt

Java side method invocation from Native side

Create JVM and call method through API

ex) CreateJVM.txt

Java APIs

- Java Platform Overview

J2SE(Standard

Edition)

API for developing general user app. or client app.

J2EE(Enterprise

API for developing large enterprise software

infrastructure
ex) EJB, servlet, JSP, JMS, etc.

J2ME(Micro

Edition)

Edition)

Light-weight platform for embedded system

Java APIs

- J2SE APIs

Serialization
RMI is used for communicating between objects in different VM.
Parameters or return values must be converted to
implementation-independent form in RMI.
Serialization may be used for object to be saved in persistent
storage.
In order to serialize an object, it must implements the Serializable
interface.
Reflection
Determine class information at run time
Classes in java.lang.reflect package

Array, Constructor, Field, Method, Modifier, etc.

Object.getClass() -> Class.get(Fields|Methods| )

ex) Method can be called by invoke method of Method class.

Java APIs

- J2SE APIs

Thread

Multithreading support is provided by java.lang.Thread

class(and Runnable interface)
Libraries can communicate with the underlying OS.
Thread execute run() method during its lifetime.

Synchronization through monitor

Suppported by instruction
monitorenter and monitorexit
Locks are associated with each object and each
class(through Class object).
Class Object declares five methods that enable
programmers to access the Java Virtual Machines support
for the coordination aspect of synchronization.
notify, notifyAll, wait

Java APIs

Synchronization Example