RFR (XS) 8199012: JVM crash during LevelDB JNI access
david.holmes at oracle.com
Thu Mar 8 21:39:03 UTC 2018
On 9/03/2018 12:31 AM, Daniel D. Daugherty wrote:
> On 3/6/18 6:31 PM, David Holmes wrote:
>> webrev: http://cr.openjdk.java.net/~dholmes/8199012/webrev/
> No comments.
> Thumbs up!
> Since you're in that function would you mind fixing the indenting
> on these two lines:
> L4223: HOTSPOT_JNI_DETACHCURRENTTHREAD_RETURN(JNI_OK);
> L4231: HOTSPOT_JNI_DETACHCURRENTTHREAD_RETURN((uint32_t) JNI_ERR);
> Both need two more spaces... no webrev needed if you decide
> to fix this.
Will do! Also found a typo "middel" that I fixed.
>> bug: https://bugs.openjdk.java.net/browse/JDK-8199012
>> JNI DetachCurrentThread is defined as a no-op if called by a thread
>> that is not attached. But the check for that was occurring after we
>> called VM_Exit::block_if_vm_exited(). If the thread was not in fact
>> attached then we have a NULL current thread and that can cause in
>> crash in VM_Exit::block_if_vm_exited() if we actually need to block -
>> which explains the scenario reported in the bug.
>> Simple, obvious, fix is to do the "not attached" check first.
>> - Mach5 CI equivalent testing (tier1/2)
>> - internal JNI tests
>> I marked this noreg-hard as a regression test is difficult to set up
>> due to the need to coordinate with a VM shutdown.
More information about the hotspot-runtime-dev