This was discovered in a round-about way with AWT shutdown code that uses JVM RawMonitors. The RawMonitor code checks for VM exit and calls VM_Exit::wait_if_vm_exited, which checks if the current thread is the shutdown-thread else it blocks it by locking the Threads_lock. However, if the current thread is not attached (which RawMonitors allow) then we can't use the Threads_lock (current thread is null). In such a situation we should not attempt to block the unattached thread.

Credit to [~ihse] for identifying the problem.