Name: js5519 Date: 02/19/99


Access violations are raised by javai.dll for JDK1.1.6 and
JDK1.1.7B on NT. The application is a large multi-threaded
server that is 100% Pure Java.

The violations occur with the JIT enabled and disabled.

When the debugger is started, the violation is c000..005.
The offending instructions include:
0x1003e3cd movzx ecx, word ptr [eax+3Ch]
     ecx = 0, eax = 29797
0x10037c16 mov exc, dword ptr[eax+4]
      [eax+4] = b0071938

The instruction pointer has also been 0x0.

When using java_g with JDK1.1.7B I get the following message:

* panic src\share\java\runtime\executeJava.c, line 234
assertion failure



Access violations occur in javai.dll when the client or server
begin excecuting. Both the client and server are 100% pure
Java and are multithreaded. Both the client and server run
on a 2-cpu Intel NT Server 4.0.

The violations occur primarily on the client after the client
has started several threads. Each of the threads executes the
same code, and the threads are started in a loop.

The workaround is to add a call to sleep after each thread
is started, so that the threads will not initially execute
the code concurrently. When the threads other than the
first thread execute, most of the code that the will execute
has already been executed by the first thread.

This code...

Thread[] threads = ...;
for (i = 0; i < threads.length; i++) {
  threads[i].start();
}

Became...

Thread[] threads = ...;
for (i = 0; i < threads.length; i++) {
  threads[i].start();
  try { Thread.currentThread().sleep(500); }
  catch(InterruptedException ie) { }
}


My assumption is that there is a race condition that is
exposed when several threads concurrently excecute code
for the first time. The addition of the sleep prevents this.

The access violations occur with JDKs 1.1.6, 1.1.7B
with the JIT enabled and disabled. The violations occur
with JDK1.2 with the original JIT enabled.

JDK 1.1.6 used the updated JIT

JDK1.1.6 JIT -> adding the sleep call stopped the faults
JDK1.1.6 no JIT -> adding the sleep call stopped the faults
JDK1.1.7B JIT -> adding the sleep call did not stop the faults
                 (I am unabled to prevent all threads from
                  starting concurrently)
JDK1.1.7B no JIT -> adding the sleep call stopped the faults
JDK1.2 original JIT -> adding the sleep call did not stop the faults
JDK1.2 early access JIT -> the sleep call was not required
JDK1.2 early acess no JIT -> the sleep call was not required
(Review ID: 54220)
======================================================================