Name: elR10090 Date: 04/18/2001


The Hotspot Client VM (1.3.1-rc1-b21 and 1.4.0-beta-b60) for Win32
in compiled mode fails the following test from testbase_nsk:

    nsk/jvmpi/events/mncntentered001

The test passes all other platform and all other modes.

To reproduce the bug run in GammaBase/Bugs/<this bug number>
directory for popframe001:

    sh doit.sh <JAVA_HOME> -Xcomp > log

and to run with java_g:

    sh doit.sh <JAVA_HOME> -g -Xcomp > log

Running on Windows the test produces the following log:

#
# HotSpot Virtual Machine Error, Internal Error
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
#
# Error ID: 53414645504F494E540E435050012D
#
# Problematic Thread: prio=5 tid=0x8a18c60 nid=0x3f4 runnable
#

(safepoint.cpp, 301)

Running with java_g the test produces the following log:

#
# HotSpot Virtual Machine Error, Internal Error
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
#
# Fatal: must be in Java code at this point
#
# Error happened during: thread suspend
#
# Error ID: Z:\hotspot1.3.1\ws\win32\hotspot1.3.1rc1\src\share\vm\runtime\safepoint.cpp, 540
#
# Problematic Thread: prio=5 tid=0x249d38 nid=0x28c runnable
#

This bug also affects the following testbase_nsk tests:

    nsk/jvmpi/events/mncntenter001
    nsk/jvmpi/events/mncntenter002
    nsk/jvmpi/events/mncntexit001

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


Updated by raghunath.verabelli on 03/20/02:

One of our Licensees Sitraka filed an Incident Manager bug report #144059 for the same issue:

Synopsis: monitor contention events crash the JVM with Volano
Bug Description: This has been present since at least 1.3.1, but I thought it was
covered by 4325536 initially. But since that problem has to do with
hprof only as it turns out, and since the problems still occurs in
hopper (b06) I'm reporting it separately.

The following trivial JVMPI client will crash with an Internal
Error (but no additional info) on Windows. The JVM does not exit
but is not consuming noticeable CPU. It doesn't always manage to
get the full error message out.

It used to happen on Solaris as well, but I've not been able to
reproduce that with Hopper (though the problem on Windows makes
me think it may still be there, I just haven't seen it).

Build the client and run it against VolanoMark 2.1.2. For me it
crashes every time on Win2k with Hopper b06 and 1.3.x, and on
Solaris with 1.3.x.

#include <jvmpi.h>

JVMPI_Interface *jvmpi;
static void notify(JVMPI_Event *);

JNIEXPORT jint JNICALL
JVM_OnLoad(JavaVM *jvm, char *options, void *reserved)
{
int res = (*jvm)->GetEnv(jvm, (void **)&jvmpi, JVMPI_VERSION_1);

if (res < 0) {
return JNI_ERR;
}

jvmpi->NotifyEvent = notify;

jvmpi->EnableEvent(JVMPI_EVENT_MONITOR_CONTENDED_ENTER, NULL);
jvmpi->EnableEvent(JVMPI_EVENT_MONITOR_CONTENDED_ENTERED, NULL);
jvmpi->EnableEvent(JVMPI_EVENT_MONITOR_CONTENDED_EXIT, NULL);

return JNI_OK;
}

static void notify(JVMPI_Event * event) {}


Talked to Mandy Chung:
The problem that Sitraka reported looks like a duplicate of:
   4446992 mncntentered001 crashes Client VM in compiled mode

The fix for 4325536 (HPROF: crash when monitor=y option is used)
is to do with hprof only. The fix is not in 1.4.1 b06 yet
and will probably be in b08. The test case that Sitraka gave is
directly related to jvmpi but not hprof.

You can add the bug description from Sitraka to 4446992.

Mandy