On Sun, 2013-05-19 at 18:32 -0400, Rik van Riel wrote: > > > > One fix would be a loop in do_smart_update(): > > - first check the global queue > > - then the per-semaphore queues > > - if one of the per-semaphore queues made progress: check the global > > queue again > > - if the global queue made progress: check the per semaphore queues again > > ... > > Would that be as simple as making do_smart_update() loop back to > the top if update_queue on a single semaphore's queue returns > a non-zero value (something was changed), and there are complex > operations pending?
I've been looking at the code for a while and this approach seems quite reasonable. I'd still like Manfred's feedback though. I ran pgbench and your semop-multi program, nothing suspicious. > ---8<--- > > Subject: ipc,sem: move restart loop to do_smart_update > > A complex operation may be sleeping on a semaphore to become > a certain value. A sleeping simple operation may turn the > semaphore into that value. > > Having the restart loop inside update_queue means we may be > missing the complex operation (which lives on a different > queue), and result in a semaphore lockup. > > The lockup can be avoided by moving the restart loop into > do_smart_update, so the list of pending complex operations > will also be checked if required. > > Signed-off-by: Rik van Riel <r...@redhat.com> > Reported-by: Manfred Spraul <manf...@colorfullife.com> Acked-by: Davidlohr Bueso <davidlohr.bu...@hp.com> -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
