tim.bell@Eng 1998-06-03
New description & test case added. Old description may be found
at the bottom.

ThreadTest runs indefinitely if started "java -nojit ThreadTest"

If started "java ThreadTest", a pop-up announces:

    The instruction at "0x01f1d689" referenced memory at
    "0x00000118" The memory could not be "read"

Launching the debugger adds:
    Unhandled exception in java.exe (SYMCJIT.DLL) 0xC0000005: Access Violation


//Begin ThreadTest.java
/*
 * A background thread with certain do/while loop
 * constructs causes the JRE116 JIT VM to blow up.
 *
 * Here is the code (works fine under Solaris and NT with -nojit)
 *
 * --------------------------------------
 *
 * package arbdesk.test;
 */

import java.lang.Thread;

/**
* Test thread operation
*/
public final class ThreadTest implements Runnable, Biteable {
    /** Thread to handle incoming data from the server */
    private Thread receiver = null;

    /** Thread to send "ping" messages */
    private Watchdog keepTalking = null;

    /**
    * Constructor
    */
    public ThreadTest () {
// Start the receiver thread
receiver = new Thread(this);
receiver.start();

// Start the watchdog threads.
// Set their priorities in the order of severity of the situations
// that they are expected to handle.
keepTalking = new Watchdog(this);
keepTalking.setPriority(receiver.getPriority() + 1);
debug("Threads started");
    }

    /**
    * runnable method for the receiver thread
    */
    public void run() {
debug("run()");
while (true) {

try {
Thread.currentThread().sleep(1000);
debug("run() finished sleeping");
}
catch(Exception ex) {
ex.printStackTrace();
}

}
    }

    public void doBite() {
debug("Bitten");
    }

    /**
    * Display a diagnostic message
    */
    private void debug(String msg) {
System.out.println(getClass().getName() + ':' + msg);
    }

    /**
    * Test method
    */
    public static void main(String[] argv) {
ThreadTest t = new ThreadTest ();
    }
}


/**
* Interface to be implemented by classes that need to know when a
* watchdog expires
*/
interface Biteable {
    public abstract void doBite();
}

class Watchdog extends Thread {
    protected int intervalMins;
    protected int timeout;
    protected boolean autoRestart;
    protected Biteable observer;

    public Watchdog (Biteable observer) {
this.observer = observer;
this.timeout = 1;
this.autoRestart = true;
start();
    }

    /**
     * The thread body.
     * Sleep for a minute at a time, decrementing the
     * expiry timeout counter. If the counter reaches
     * zero, call back into the Biteable "doBite"
     * method.
     */
    public void run() {
do {
/*
*** Comment out the loops below to modify the
*** failure behaviour! ***
*/
while (timeout > 0) {
try {
sleep(6 * 1000);// Do nothing for 6 seconds
}
catch(InterruptedException e) {
}
timeout = 0;
}
observer.doBite();
timeout = 1;
} while (autoRestart);
    }
}
// EOF
//End ThreadTest.java

======================================================================

Name: tb29552 Date: 05/18/98


We have a number of JAVA/CORBA servers which we invoke simultaneously
in seperate VM's on a Win NT 4 machine using JDK 1.1.6 with the JIT switched
on. We are also using Visibroker V3.2.

Upon starting up the servers we randomly encounter the following error in
one of the Virtual Machines:

Unhandled Exception in java.exe (SYMCJIT.DLL): 0xC0000005: Access Violation

Although the memory address may be different the disassembly is mostly
the same (it seems to be the same line every time) this is provided below:

01F566F9 push edi
01F566FA push esi
01F566FB and esi,0FEh
01F566FE mov ebx,dword ptr [esi+1Ch]
01F56701 mov esi,dword ptr [esi+2Ch]
01F56704 test esi,esi
01F56706 je 01F567C1
01F5670C push eax
01F5670D mov eax,dword ptr [edi+10h]
01F56710 pop edi

The next line is the problem...

01F56711 mov eax,dword ptr [eax+4]

01F56714 test eax,1Fh
01F56719 je 01F56722
01F5671B mov eax,[01F54C08]
01F56720 jmp 01F56724
01F56722 mov eax,dword ptr [eax]
01F56724 push eax
01F56725 mov eax,dword ptr [ebx+0Ch]
01F56728 test eax,eax
01F5672A je 01F567B6
01F56730 mov edx,dword ptr [eax]
01F56732 mov ecx,dword ptr [eax+4]
01F56735 test byte ptr [esp+4],1
01F5673A je 01F56742

The bytecode we are running on this is mostly compiled to JDK
1.02 using the Visual J++ compiler. The bug does not always
occur and does not effect the same server every time.

Can you shed any light on this problem?
(Review ID: 30168)

More problems even with the new Symantec JIT Version x3.00.050:
At Cadence we can reproduce this bug but the memory location is different.

The instruction at "0x01f42b6f" referenced memory at "0x00000014"
The memory could not be "Read"

nasser.nouri@Corp 1998-07-13