JVM Crash in GCTaskThread with Java 7u21

Question

We have a server based Java application that has been running for the last 6 months with few problems. The server uses 32-bit Java 7u21 configured with minimum and maximum heap of 1433m. Recently, in the space of 1 week, we have experienced 2 separate JVM crashes. For both crashes, the crash log shows an EXCEPTION ACCESS VIOLATION in VM code and the current thread as the GCTaskThread.
We understand this crash to be happening during Garbage Collection. The application threads are all blocked and the VM has reached a safepoint. The young generation is being garbage collected and at the moment of the crash, both survivor spaces are showing occupancy - which seems very wrong to us!
Heap at time of crash:
par new generation total 235968K, used 233398K [0x047a0000, 0x147a0000, 0x147a0000)
eden space 209792K, 99% used [0x047a0000, 0x11474e78, 0x11480000)
from space 26176K, 90% used [0x12e10000, 0x14528bb8, 0x147a0000)
to space 26176K, 100% used [0x11480000, 0x12e10000, 0x12e10000)
concurrent-mark-sweep generation total 1206272K, used 917730K [0x147a0000, 0x5e1a0000, 0x5e1a0000)
concurrent-mark-sweep perm gen total 131072K, used 78075K [0x5e1a0000, 0x661a0000, 0x661a0000)
We have been unable to reproduce the problem on demand, or identify any unusual application activity at the time of the crashes. We are running Java 7u21 which is not the latest release. We could upgrade to the latest version of Java 7, however without knowledge and understanding of the causes of the crash, we cannot be confident that the upgrade will resolve the matter.
We would appreciate any insights anyone can offer into determining the cause of the JVM crash or suggestions on ways that we can confirm that we have resolved the problem if we upgrade to use the latest version of Java 7.

Answer 1

Looks like JNI is being used.  Which is the most like cause.