Name: elR10090 Date: 04/25/2002


J2SE 1.4.1-b10 crashes with the following asertion:
##
## HotSpot Virtual Machine Error, assertion failure
## Please report this error at
## http://java.sun.com/cgi-bin/bugreport.cgi
##
## Java VM: Java HotSpot(TM) Server VM (1.4.1-beta-b10-debug mixed mode)
##
## assert(map()->req() <= jvms()->endoff(), "no extra garbage on map")
##
## Error ID: /BUILD_AREA/jdk1.4.1/hotspot/src/share/vm/opto/graphKit.cpp, 98
##
## Problematic Thread: prio=1 tid=0x0x80cd060 nid=0x6ee runnable
##
#Current thread is 0x2009
#Dumping core ...

Such crash is observed against Server VM: Linux/x86, Solaris/x86
and Solaris/SPARC/32-bit, and at 64-bit Solaris for SPARC/v9.

To reproduce this crash, use the following testcase:

    javac b4510834.java
    java_g -server b4510834

Here is the testcase sources:

    import java.io.*;

    public class b4510834 {
        private int WAITTIME = 2; // minutes
        private int THREADS = 50;

        public static void main(String args[]) {
            new b4510834().run(args);
        }

        private void run(String args[]) {
            for (int i=0; i<args.length; i++)
                if (args[i].toLowerCase().startsWith("-waittime="))
                    WAITTIME = Integer.parseInt(args[i].substring(10));
                else if (!args[i].startsWith("-"))
                    THREADS = Integer.parseInt(args[i]);
                else
                    throw new IllegalArgumentException(
                        "unknown keyword: " + args[i]);

            buzz = new Buzz [ THREADS ];
            for (int t=0; t<buzz.length; t++) {
                buzz[t] = new Buzz();
                buzz[t].setDaemon(true);
                buzz[t].start();
            };

            Guard guard = new Guard();
            guard.setDaemon(true);
            guard.start();

            new PressBuzz().start();
            try {
                synchronized (buzz) {
                    buzz.wait();
                }
            } catch (InterruptedException exception) {
                error("# run(args[]): " + exception);
            }
        }

        private class PressBuzz extends Thread {
            public void run() {
                synchronized (buzz) {
                    buzz.notify();
                };
                for (int t=0; t<buzz.length; t++)
                    buzz[t].suspend();
            }
        }

        private Buzz buzz[] = null;

        private class Buzz extends Thread {
            public void run() {
                while (true)
                    yield();
            }
        }
    
        private class Guard extends Thread {
            public void run() {
                try {
                    sleep(WAITTIME * 60000L);
                } catch (InterruptedException exception) {
                    error("# Guard.run(): " + exception);
                };
                error("TEST FAILED: time exceeded (" + WAITTIME + " min)");
            }
        }

        private synchronized void error(String message) {
            System.out.println(message);
            Runtime.getRuntime().halt(1);
        }
    }

This test tries to provoke a race between jni->DestroyJavaVM()
(implicitely invoked by finishing the main thread) and thread
suspend efforts - in order to reproduce the bug:
    4510834 JNI DestroyJavaVM/VMThread/sampler thread race hangs VM

The test will appear with the next "r14" release of the testbase_nsk;
its full name would be:
    nsk/regression/b4510834

The crash reported here looks similar (but not identical)
to the crashes reported by the bugs:
    4614060 api.java.lang.ThreadGroup.enumerate0203 fails in solaris platform in -Xmixed
    4426707 runThese crashes in 64Bit mode with assertion at parse1.cpp, 1472
======================================================================