On Fri, 2018-07-27 at 10:46 -0600, Keith Busch wrote: > On Fri, Jul 27, 2018 at 09:20:42AM -0700, Bart Van Assche wrote: > > + ret = req->q->mq_ops->timeout(req, reserved); > > + /* > > + * BLK_EH_DONT_RESET_TIMER means that the block driver either > > + * completed the request or still owns the request and will > > + * continue processing the timeout asynchronously. In the > > + * latter case, if blk_mq_complete_request() was called while > > + * the timeout handler was in progress, ignore that call. > > + */ > > + if (ret == BLK_EH_DONT_RESET_TIMER) > > + return; > > This is how completions get lost.
The new approach for handling completions that occur while the .timeout() callback in progress is as follows: * blk_mq_complete_request() executes the following code: if (blk_mq_change_rq_state(rq, MQ_RQ_TIMED_OUT, MQ_RQ_COMPLETE)) return; * blk_mq_rq_timed_out() executes the following code: if (blk_mq_rq_state(req) == MQ_RQ_COMPLETE) { __blk_mq_complete_request(req); return; } As one can see __blk_mq_complete_request() gets called if this race occurs. Bart.