On Fri, Aug 10, 2018 at 10:59:37AM +0800, jianchao.wang wrote:

Hi Omar

On 08/10/2018 04:26 AM, Omar Sandoval wrote:

quoted

@@ -524,6 +524,9 @@ inline void __blk_mq_end_request(struct request *rq, blk_status_t error)
 		blk_stat_add(rq, now);
 	}
 
+	if (rq->internal_tag != -1)
+		blk_mq_sched_completed_request(rq, now);
+

Is it OK to move the io scheduler completed callback into __blk_mq_end_request ?

It's ok in the sense that only Kyber uses it.

There is a relatively long distance between the __blk_mq_complete_request and __blk_mq_end_request,
especially the driver's mq_ops.complete and the bio_endio.

That's a fair point. We could maybe consolidate all of the I/O time
measurements to __blk_mq_complete_request() and have a separate callback
for the total time measurement in __blk_mq_end_request(), but I'm not
sure it's worth adding another hook for that.