Re: Review Request (XS) 8030027: nsk/jvmti/scenarios/hotswap/HS101/hs101t006 Crashed the vm on Linux-amd64: SIGSEGV in JavaThread::last_java_vframe(RegisterMap*)+0xfa

[serguei.spit...@oracle.com]

On 12/19/13 2:19 AM, David Holmes wrote:

Hi Serguei,

On 19/12/2013 6:01 PM, serguei.spit...@oracle.com wrote:

Please, review the fix for:
   https://bugs.openjdk.java.net/browse/JDK-8030027


Open webrev:

http://cr.openjdk.java.net/~sspitsyn/webrevs/2013/hotspot/8030027-JVMTI-HS101.1/

Summary:
   The bug is that the JavaThread::last_java_vframe (called from the
VMOp_GetCurrentLocation)
   crashes on the target thread stack that is non-walkable because the
thread is exiting.
   The fix is to check a condition !_thread->is_exiting() in the
VMOp_GetCurrentLocation.
   The fix is for JDK 9.

Looks good to me. I wonder how many more of these termination races are still lurking.

Thanks, David!
There are some more complicated shutdown races in the JDI + JDWP-agent code.
I'm still thinking how to fix them better.

Thanks,
Serguei

David

Testing:

The originally failing test nsk/jvmti/scenarios/hotswap/HS101/hs101t006

   In progress: nsk.jvmti, nsk.jdi, nsk.jdwp


Thanks,
Serguei