RFR: 8369238: Allow virtual thread preemption on some common class initialization paths [v3]

[David Holmes]

On Mon, 20 Oct 2025 19:19:05 GMT, Patricio Chilano Mateo <pchilanomate at openjdk.org> wrote:

>> src/hotspot/share/runtime/objectMonitor.cpp line 315:
>> 
>>> 313: }
>>> 314: 
>>> 315: // Keep object protected during ObjectLocker preemption.
>> 
>> I don't understand why this is necessary.
>
> Once the thread initializing the class finishes the initialization and calls `InstanceKlass::fence_and_clear_init_lock()`, the init_lock oop can be GC’ed, unless there is some other `Handle` protecting it (`ObjectMonitor::_object` is only a `WeakHandle`). But we could still have preempted virtual threads using the ObjectMonitor, so we need to protect the oop. In practice, I don’t see paths in `ObjectMonitor::resume_operation` where we try to get the oop out of the monitor, but I’m not sure if we want to break the invariant where the object needs to be alive while using its ObjectMonitor.

I see - platforms threads have a Handle to the oop whilst they wait, but vthreads have left that code, so we need the OM to maintain the strong reference.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/27802#discussion_r2450563031