Name: clC74495 Date: 06/18/98


My configuration is the Java JRE 1.1.6 on Windows NT 4.0 Server
using LiveSoftware JRun Servlet runner for integration into
Microsoft Internet Information Server (IIS) 3.0. The system
is a 300MHz Dual Pentium II PC server.

There seems to be a problem in the function jni_AttachCurrentThread()
in the JNI.C source file where the PrivateInfo member of the
Java Thread object (java_thr) seems to be being written to a
zero value by some other thread of execution. This ultimately
leads to an Access Violation in the sysThreadSetPriority() function
which cannot be passed a zero value for the thread id.

The three important lines of code in jni_AttachCurrentThread()
are the following.

1: THREAD(java_thr)->PrivateInfo = (long)sys_thr;
2: THREAD(java_thr)->priority = NormalPriority;
3: threadSetPriority(java_thr, NormalPriority);

The java_thr->obj->PrivateInfo member is set to the value
sys_thr, but when threadSetPriority() macro is called
only two statements later, the debugger reports that
java_thr->obj->PrivateInfo is now zero and no longer equals
sys_thr, which it was just set to. Because a zero value
is passed to threadSetPriority() an Access violation occurs
within the function sysThreadSetPriority() which is called
by the macro.

I have checked the assembly code around these statements
and it seems compeletely valid.

This bug seems to be extremely hard to reproduce and occurs
after anywhere between 100,000 to 500,000 calls to the
jni_AttachCurrentThread() function -- about 1 to 2 hours into
a heavy load placed on the IIS web server. This function is
called whenever IIS spawns a request to be serviced by the
JRun servlet engine and our servlet. If needed, I can go into
more detail regarding what type of load needs to be placed
on the IIS server to reproduce this problem.

My best guess after analyzing the problem is that some other
thread of execution within the Java VM manipulates the
Java thread object which results in setting it's PrivateInfo
member to zero.

This bug has the potential of delaying the shipment of an
important new graphics product for Macromedia. Therefore,
it is important that we have help from Sun to track down
and resolve whatever problem is occuring.

Mike Thompson
###@###.###
(Review ID: 33865)
======================================================================