Quoting Paolo Bonzini (2014-08-27 11:28:23) > Il 27/08/2014 17:53, Michael Roth ha scritto: > >> In case VM does IO while we run a stream job. > >> When stream finishes, the stream coroutine drains all IOs before > >> close the unused image, in bdrv_drain_all() it may find > >> a pending request which is submitted by guest IO coroutine. > >> In order to wait the pending req finish, the subsequent aio_poll() > >> call poll() to wait the req. however, if the req is already done by > >> threadpool and is waiting for the callback, there is no chance to switch > >> back to guest IO coroutine to call the callback and so that the stream > >> coroutine waits in poll() all the time. > > This is the same bug fixed by this patch; > > commit 3c80ca158c96ff902a30883a8933e755988948b1 > Author: Stefan Hajnoczi <stefa...@redhat.com> > Date: Tue Jul 15 16:44:26 2014 +0200 > > thread-pool: avoid deadlock in nested aio_poll() calls > > The thread pool has a race condition if two elements complete before > thread_pool_completion_bh() runs: > > If element A's callback waits for element B using aio_poll() it will > deadlock since pool->completion_bh is not marked scheduled when the > nested aio_poll() runs. > > Fix this by marking the BH scheduled while thread_pool_completion_bh() > is executing. This way any nested aio_poll() loops will enter > thread_pool_completion_bh() and complete the remaining elements. > > Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com>
Thanks. Looks like this missed v2.1.0 so I've gone ahead and queued it for 2.1.1 (along with 5f8127a5811429da85cc4bc273f166aed129fc31). > > Paolo
