On Mon, 7 Sep 2026 13:14:41 GMT, Jorn Vernee <[email protected]> wrote:

>> See the JBS issue for the extended problem description.
>> 
>> Native threads that were attached to the JVM through an FFM upcall are 
>> automatically detached from the VM when they join/terminate. However, if a 
>> native thread tries to terminate after the VM has exited without going 
>> through `DestroyJavaVm` (e.g. as a result of calling `System.exit`), they 
>> will block in `VM_Exit::block_if_vm_exited` inside `DetachCurrentThread`. 
>> This may happen for instance when they try to join after the JVM has exited 
>> in an `atexit` handler.
>> 
>> This patch adds a check before trying to detach the thread to see if the VM 
>> has exited and bails out if it has. This does not prevent issues as a result 
>> of a race between the VM exiting and the thread joining, but it does prevent 
>> issues in the more comment case where a thread simply outlives the VM.
>> 
>> ---------
>> - [X] I confirm that I make this contribution in accordance with the 
>> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai).
>
> Jorn Vernee has updated the pull request incrementally with one additional 
> commit since the last revision:
> 
>   Indentation
>   
>   Co-authored-by: David Holmes <[email protected]>

test/jdk/java/foreign/detachafterexit/TestDetachAfterExit.java line 88:

> 86: 
> 87:                 System.out.println("[main] Waiting for callback...");
> 88:                 while (!flag.get()) {

`cb` sets `flag` in the callback, but the native worker may still be executing 
I think. Can the `main` thread then close the arena (freeing the upcall stub) 
and invoke `exit(0)` before the worker returns to native code? If so, we could 
be exposed to use-after-free and/or leave the worker in an unknown state. Is 
there a more robust way of handshaking or can we use a timed delay here?

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

PR Review Comment: https://git.openjdk.org/jdk/pull/32686#discussion_r3956623643

Reply via email to