I found this while investigating JDK-8319897. It seems like jmm_GetThreadInfo calls Threads::owning_thread_from_object, which calls LockStack::contains, on running threads. This is a race and Undefined Behavior. I've heard some explanation that has been deemed to be safe after investigating the race and thinking about what could actually happens. That is not a safe way to handle races, certainly not when we are changing the implementation of the LockStack.

This needs to be fixed somehow. If others don't agree then we should at least add a warning in the code + document here why we think it is OK to have this race.