-
Notifications
You must be signed in to change notification settings - Fork 721
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Spin during VirtualThread MountBegin and UnmountBegin #18439
Conversation
@fengxue-IS Requesting your review. fyi @tajila |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
code LGTM, only question is about moving carrier suspend to unmountBegin
means that the threadObject would still point to vthread. Would this affect any other JVMTI API?
In the above approach, the halt flag is set too late. As soon as the
continuation swaps the J9VMThread context, the thread begins execution
and is capable of triggering JVMTI events.
My understanding is that J9_PRIVATE_FLAGS_VIRTUAL_THREAD_HIDDEN_FRAMES
flag set during mount/unmountBegin
should prevent JVMTI events from being dispatched. are there cases which have not been covered?
i don't believe so. carrier thread will stay suspended. the virtual will look as if it is still running. if someone suspends the virtual thread, it will halt while re-acquiring VM access.
in #16654, we didn't disable all JVMTI events with |
jenkins test sanity.functional,sanity.openjdk,extended.openjdk aix,zlinux jdk21 |
Currently, the halt flag is set in VirtualThread MountEnd if a virtual thread is suspended via JVMTI, and in VirtualThread UnmountEnd if a carrier thread is suspended via JVMTI. In the above approach, the halt flag is set too late. As soon as the continuation swaps the J9VMThread context, the thread begins execution and is capable of triggering JVMTI events. To avoid the above issue, the above steps are moved into VirtualThread MountBegin and UnmountBegin. This prevents the continuation to swap the J9VMThread context. Currently, the halt flag is set without invoking exitVThreadTransitionCritical. This prevents JVMTI to resume the halted thread and cause a hang. The new approach spins, invokes exitVThreadTransitionCritical and releases VM access to allow JVMTI to resume the suspended thread. The better approach will be to fail mount if the thread is suspended and retry later. Currently, his approach cannot be implemented because VirtualThread.java does not support this approach. Related: eclipse-openj9#17865 Related: eclipse-openj9#17869 Related: eclipse-openj9#18370 Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
Old PR build: https://openj9-jenkins.osuosl.org/job/PullRequest-OpenJ9/4736. It should still be good since the last update only added the TODO comments. |
Currently, the halt flag is set in VirtualThread MountEnd if a virtual
thread is suspended via JVMTI, and in VirtualThread UnmountEnd if a
carrier thread is suspended via JVMTI.
In the above approach, the halt flag is set too late. As soon as the
continuation swaps the J9VMThread context, the thread begins execution
and is capable of triggering JVMTI events.
To avoid the above issue, the above steps are moved into VirtualThread
MountBegin and UnmountBegin. This prevents the continuation to swap
the J9VMThread context.
Currently, the halt flag is set without invoking
exitVThreadTransitionCritical. This prevents JVMTI to resume the halted
thread and cause a hang. The new approach spins, invokes
exitVThreadTransitionCritical and releases VM access to allow JVMTI to
resume the suspended thread.
The better approach will be to fail mount if the thread is suspended
and retry later. Currently, his approach cannot be implemented because
VirtualThread.java does not support this approach.
Related: #17865
Related: #17869
Related: #18370