RFR: 8345266: java/util/concurrent/locks/StampedLock/OOMEInStampedLock.java JTREG_TEST_THREAD_FACTORY=Virtual fails with OOME

[David Holmes]

On Tue, 17 Dec 2024 00:44:42 GMT, Patricio Chilano Mateo <pchilanomate at openjdk.org> wrote:

> Please review this small fix. The call to `Continuation::try_preempt()` can fail and return with a pending OOM exception due to insufficient memory while trying to allocate a new stackChunk. On this particular `OOMEInStampedLock.java` test, the OOM is thrown when returning from Object.wait, which is unexpected and causes the test to fail. But this could also happen when a virtual thread returns from monitorenter, which would cause an assert when coming from compiled code (see `SharedRuntime::monitor_enter_helper()`). Since a failing `try_preempt()` call will just prevent the vthread from being unmounted and force it to follow the normal platform thread code there is no need to throw OOM. The fix is to just clear the pending exception if set, and let future bytecodes handle the OOM condition if still present.
> 
> I verified that test `OOMEInStampedLock.java` now passes with the fix. I also added a new more specific test which fails without this fix and passes with it. I also run the change through mach5 tiers1-3.
> 
> Thanks,
> Patricio

Is OOME the only possible exception in this code path?

src/hotspot/share/runtime/continuation.cpp line 164:

> 162:   JVMTI_ONLY(jubm.set_result(res);)
> 163: 
> 164:   if (target->has_pending_exception()) {

Is `target` always the current thread, because we should only ever check and clear exceptions on the current thread. ??

test/jdk/java/lang/Thread/virtual/MonitorOOM.java line 66:

> 64:                         if (testWait) {
> 65:                             lock.wait(5);
> 66:                         }

If you have filled the heap won't you always get the OOME on the monitor enter?

-------------

PR Review: https://git.openjdk.org/jdk/pull/22780#pullrequestreview-2507705115
PR Review Comment: https://git.openjdk.org/jdk/pull/22780#discussion_r1887800294
PR Review Comment: https://git.openjdk.org/jdk/pull/22780#discussion_r1887802027