Hello,

On Wed, Dec 13, 2017 at 11:30:48AM +0800, jianchao.wang wrote:

quoted

+	} else {
+		srcu_idx = srcu_read_lock(hctx->queue_rq_srcu);
+		if (!blk_mark_rq_complete(rq))
+			__blk_mq_complete_request(rq);
+		srcu_read_unlock(hctx->queue_rq_srcu, srcu_idx);

The __blk_mq_complete_request() could be executed in irq context. There should not be any 
sleeping operations in it. If just synchronize with the timeout path to ensure the aborted_gstate
to be seen, only rcu is needed here ,as well as the blk_mq_timeout_work.

Sure, but it's just a lot cleaner to use the same to protect both
issue and completion; otherwise, whoever who wants to synchronize
against them have to do awkward double rcu locking.

Thanks.

-- 
tejun