On Mon, 21 Sep 2026 08:03:05 GMT, David Holmes <[email protected]> wrote:
> The `GetThreadSnapshotHandshakeClosure` was added to support virtual threads > in thread dumps. It is very similar to the logic used in the old > safepoint-based thread dump. > `GetThreadSnapshotHandshakeClosure::detect_locks` uses the raw > `javaVFrame::monitors()` method rather than using the `locked_monitors()` > method which already filters out some monitors including those for which > `wait()` has been called. It uses the raw `monitors()` list because it wants > to process eliminated compiled monitors itself, and they are already removed > from `locked_monitors()`. But that means it should be doing its own filtering > of monitors that are being waited-on so they are not reported as locked. This > seems to have been an oversight with the original implementation. > > The logic is somewhat more complicated, particularly in the virtual thread > case, because there are a number of points where we can dump the stack > between `wait0` appearing as the top frame, and the point where we actually > release the monitor. It would be wrong to hide the monitor from the "locked" > section in that case. So we can still report that a thread is waiting-on a > particular object and that it has locked that object, as that is the actual > state of things. But the code no longer reports a released monitor as locked > by the waiting thread. > > Note that `javaVFrame::locked_monitors` is also potentially imprecise in its > reporting. It will report the monitor as locked until > `current_waiting_monitor()` is set, and then elide it. However, as this code > is only used when reporting platform threads this imprecision is not > observable as a platform thread will not respond to a stack dump request in > between the transition to in-VM (for the native call) and the release of the > monitor. > > I also fixed a pre-existing typo whilst in this code. > > Testing > - tiers 1-3 > - tier 5-svc > - updated tests in the PR > > Thanks > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). This pull request has now been integrated. Changeset: d244fcef Author: David Holmes <[email protected]> URL: https://git.openjdk.org/jdk/commit/d244fceff3e48fae405e4da1d055db55344c7c7a Stats: 60 lines in 2 files changed: 33 ins; 0 del; 27 mod 8361926: Thread dump has incorrect locking information when waiting on a monitor Reviewed-by: alanb, sspitsyn ------------- PR: https://git.openjdk.org/jdk/pull/32977
