Support the new changes to the VirtualThread states

[babsingh]

RUNNABLE=2 state was removed, and all other states are re-numbered
due to the removal of the RUNNABLE state.

Two new states are added: UNPARKED=9 and YIELDED=11.

Currently, VirtualThread.yield incorrectly consumes the parking
permit. The parking permit should only be consumed when continuing
from VirtualThread.park. The two new states are added to address
this incorrect behaviour.

Depends on ibmruntimes/openj9-openjdk-jdk21#103

[babsingh]

This is to support the fix for https://bugs.openjdk.org/browse/JDK-8321270. The fix is currently in JDK22 and JDK23 (JDK-next). The fix will eventually be ported to JDK21 as per https://bugs.openjdk.org/browse/JDK-8321270.

Commit title -- 8321270: Virtual Thread.yield consumes parking permit

This PR doesn't have ifdefs for JDK22+. So, this PR shouldn't be merged until the fix is ported to JDK21.

ibmruntimes/openj9-openjdk-jdk21#103 backports the fix to JDK21. To avoid any side-effects from JDK-8321270's fix, the backport can wait until the 0.42 and 0.43 targets are released.

If JDK-8321270's fix is ported by the RI to JDK21, then we can just merge this PR and close ibmruntimes/openj9-openjdk-jdk21#103.

fyi @pshipton @tajila

[hangshao0]

LGTM.

[hangshao0]

Jenkins test sanity.functional,sanity.openjdk aix jdk21 depends ibmruntimes/openj9-openjdk-jdk21#103

[hangshao0]

To avoid any side-effects from JDK-8321270's fix, the backport can wait until the 0.42 and 0.43 targets are released.

Does this really matter ? 0.42 and 0.43 have already branched.

[pshipton]

Does this really matter ? 0.42 and 0.43 have already branched.

It does make it easier to merge 21.0.2 to 0.43, and any other fixes in the extensions for 0.42/0.43 (and we do need one). I won't merge it now.

I've added it to the 0.43 milestone just in case JDK-8321270 shows up in 21.0.2.

[babsingh]

We should show caution with these changes. Before merging, we should also run the extended.openjdk test suite, which has the JVMTI tests, at least on two platforms (aix and alinux). This is to make sure that these changes have no side-effects (such as timeouts/hangs).

[hangshao0]

Jenkins test extended.openjdk aix jdk21 depends ibmruntimes/openj9-openjdk-jdk21#103

[hangshao0]

Jenkins test extended.openjdk alinux jdk21 depends ibmruntimes/openj9-openjdk-jdk21#103

[babsingh]

In the extended.openjdk builds, only known and unrelated failures are seen:

• JDK21 serviceability/jvmti/vthread/FollowReferences/VThreadStackRefTest.java ERROR class VThreadStackRefTest$VThreadMountedJNIReferenced: ref count = 1 (expected 1), thread id = 0 (expected 27) #18426
• OpenJDK serviceability/jvmti/vthread/SuspendResume2/SuspendResume2 no-vmcontinuations timeout #18675 (might be related to [Loom] Implement JVM_VirtualThreadDisableSuspend #18671 and https://bugs.openjdk.org/browse/JDK-8311218)
• JDK18 AIX java/nio/channels/FileChannel/Transfer2GPlus.java - RuntimeException: Transfer too small: 1043881984 #14504
• jdk_security3_0_FAILED RuntimeException: No 64-bit serial number, SunCertPathBuilderException: unable to find valid certification path to requested target and others #18602
• openjdk zip or inflate/deflate failures on RH9 (z15+) or AIX 7.3 (p10) #14948

[hangshao0]

I've added it to the 0.43 milestone just in case JDK-8321270 shows up in 21.0.2.

Does 21.0.2 have the JDK-8321270 fix ?

[hangshao0]

Does 21.0.2 have the JDK-8321270 fix ?

I see this is moved to Release 0.44, so I guess the answer is no.

[babsingh]

I see this is moved to Release 0.44, so I guess the answer is no.

Yes, @pshipton plans to merge it in a few days after monitoring the new extension repo changes.

[hangshao0]

OK, I'll let Peter merge this.