[PATCH] blk-mq: don't handle non-flush requests in blk_insert_flush

[PATCH] blk-mq: don't handle non-flush requests in blk_insert_flush

[Christoph Hellwig]

Return to the normal blk_mq_submit_bio flow if the bio did not end up
actually being a flush because the device didn't support it.  Note that
this is basically impossible to hit without special instrumentation given
that submit_bio_checks already clears these flags usually, so we'd need a
tight race to actually hit this code path.

With this the call to blk_mq_run_hw_queue for the flush requests can be
removed given that the actual flush requests are always issued via the
requeue workqueue which runs the queue unconditionally.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/blk-flush.c | 12 ++++++------
 block/blk-mq.c    | 14 ++++++--------
 block/blk.h       |  2 +-
 3 files changed, 13 insertions(+), 15 deletions(-)

diff --git a/block/blk-flush.c b/block/blk-flush.c
index e9c0b300a1771..ebb11a87c6e30 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -382,7 +382,7 @@ static void mq_flush_data_end_io(struct request *rq, blk_status_t error)
  * @rq is being submitted.  Analyze what needs to be done and put it on the
  * right queue.
  */
-void blk_insert_flush(struct request *rq)
+bool blk_insert_flush(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 	unsigned long fflags = q->queue_flags;	/* may change, cache */
@@ -412,7 +412,7 @@ void blk_insert_flush(struct request *rq)
 	 */
 	if (!policy) {
 		blk_mq_end_request(rq, 0);
-		return;
+		return true;
 	}
 
 	BUG_ON(rq->bio != rq->biotail); /*assumes zero or single bio rq */
@@ -423,10 +423,8 @@ void blk_insert_flush(struct request *rq)
 	 * for normal execution.
 	 */
 	if ((policy & REQ_FSEQ_DATA) &&
-	    !(policy & (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH))) {
-		blk_mq_request_bypass_insert(rq, false, false);
-		return;
-	}
+	    !(policy & (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH)))
+		return false;
 
 	/*
 	 * @rq should go through flush machinery.  Mark it part of flush
@@ -442,6 +440,8 @@ void blk_insert_flush(struct request *rq)
 	spin_lock_irq(&fq->mq_flush_lock);
 	blk_flush_complete_seq(rq, fq, REQ_FSEQ_ACTIONS & ~policy, 0);
 	spin_unlock_irq(&fq->mq_flush_lock);
+
+	return true;
 }
 
 /**
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 9248edd8a7d3b..428e0e0fd5504 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2491,14 +2491,12 @@ void blk_mq_submit_bio(struct bio *bio)
 		return;
 	}
 
-	if (unlikely(is_flush_fua)) {
-		struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
-		/* Bypass scheduler for flush requests */
-		blk_insert_flush(rq);
-		blk_mq_run_hw_queue(hctx, true);
-	} else if (plug && (q->nr_hw_queues == 1 ||
-		   blk_mq_is_shared_tags(rq->mq_hctx->flags) ||
-		   q->mq_ops->commit_rqs || !blk_queue_nonrot(q))) {
+	if (is_flush_fua && blk_insert_flush(rq))
+		return;
+
+	if (plug && (q->nr_hw_queues == 1 ||
+	    blk_mq_is_shared_tags(rq->mq_hctx->flags) ||
+	    q->mq_ops->commit_rqs || !blk_queue_nonrot(q))) {
 		/*
 		 * Use plugging if we have a ->commit_rqs() hook as well, as
 		 * we know the driver uses bd->last in a smart fashion.
diff --git a/block/blk.h b/block/blk.h
index e80350327e6dc..cb98a10a119c8 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -236,7 +236,7 @@ void __blk_account_io_done(struct request *req, u64 now);
  */
 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
 
-void blk_insert_flush(struct request *rq);
+bool blk_insert_flush(struct request *rq);
 
 int elevator_switch_mq(struct request_queue *q,
 			      struct elevator_type *new_e);
-- 
2.30.2

Re: [PATCH] blk-mq: don't handle non-flush requests in blk_insert_flush

[Jens Axboe]

On 10/19/21 6:25 AM, Christoph Hellwig wrote:
> Return to the normal blk_mq_submit_bio flow if the bio did not end up
> actually being a flush because the device didn't support it.  Note that
> this is basically impossible to hit without special instrumentation given
> that submit_bio_checks already clears these flags usually, so we'd need a
> tight race to actually hit this code path.
> 
> With this the call to blk_mq_run_hw_queue for the flush requests can be
> removed given that the actual flush requests are always issued via the
> requeue workqueue which runs the queue unconditionally.

This looks great, thanks.

-- 
Jens Axboe

Re: [PATCH] blk-mq: don't handle non-flush requests in blk_insert_flush

[Jens Axboe]

On Tue, 19 Oct 2021 14:25:53 +0200, Christoph Hellwig wrote:
> Return to the normal blk_mq_submit_bio flow if the bio did not end up
> actually being a flush because the device didn't support it.  Note that
> this is basically impossible to hit without special instrumentation given
> that submit_bio_checks already clears these flags usually, so we'd need a
> tight race to actually hit this code path.
> 
> With this the call to blk_mq_run_hw_queue for the flush requests can be
> removed given that the actual flush requests are always issued via the
> requeue workqueue which runs the queue unconditionally.
> 
> [...]

Applied, thanks!

[1/1] blk-mq: don't handle non-flush requests in blk_insert_flush
      commit: d92ca9d8348fb12c89eac5928bd651c3a485d7b9

Best regards,
-- 
Jens Axboe