block: Limit number of items taken from the I/O scheduler in one go

Flushes bypass the I/O scheduler and get added to hctx->dispatch
in blk_mq_sched_bypass_insert.  This can happen while a kworker is running
hctx->run_work work item and is past the point in
blk_mq_sched_dispatch_requests where hctx->dispatch is checked.

The blk_mq_do_dispatch_sched call is not guaranteed to end in bounded time,
because the I/O scheduler can feed an arbitrary number of commands.

Since we have only one hctx->run_work, the commands waiting in
hctx->dispatch will wait an arbitrary length of time for run_work to be
rerun.

A similar phenomenon exists with dispatches from the software queue.

The solution is to poll hctx->dispatch in blk_mq_do_dispatch_sched and
blk_mq_do_dispatch_ctx and return from the run_work handler and let it
rerun.
Signed-off-by: Salman Qazi <sqazi@google.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/blk-mq-sched.c

@@ -86,12 +86,16 @@ void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx)
  * Only SCSI implements .get_budget and .put_budget, and SCSI restarts
  * its queue by itself in its completion handler, so we don't need to
  * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE.
+ *
+ * Returns -EAGAIN if hctx->dispatch was found non-empty and run_work has to
+ * be run again.  This is necessary to avoid starving flushes.
  */
-static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
+static int blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
 {
 	struct request_queue *q = hctx->queue;
 	struct elevator_queue *e = q->elevator;
 	LIST_HEAD(rq_list);
+	int ret = 0;
 
 	do {
 		struct request *rq;
@@ -99,6 +103,11 @@ static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
 		if (e->type->ops.has_work && !e->type->ops.has_work(hctx))
 			break;
 
+		if (!list_empty_careful(&hctx->dispatch)) {
+			ret = -EAGAIN;
+			break;
+		}
+
 		if (!blk_mq_get_dispatch_budget(hctx))
 			break;
 
@@ -123,6 +132,8 @@ static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
 		 */
 		list_add(&rq->queuelist, &rq_list);
 	} while (blk_mq_dispatch_rq_list(q, &rq_list, true));
+
+	return ret;
 }
 
 static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx,
@@ -140,16 +151,25 @@ static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx,
  * Only SCSI implements .get_budget and .put_budget, and SCSI restarts
  * its queue by itself in its completion handler, so we don't need to
  * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE.
+ *
+ * Returns -EAGAIN if hctx->dispatch was found non-empty and run_work has to
+ * to be run again.  This is necessary to avoid starving flushes.
  */
-static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx)
+static int blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx)
 {
 	struct request_queue *q = hctx->queue;
 	LIST_HEAD(rq_list);
 	struct blk_mq_ctx *ctx = READ_ONCE(hctx->dispatch_from);
+	int ret = 0;
 
 	do {
 		struct request *rq;
 
+		if (!list_empty_careful(&hctx->dispatch)) {
+			ret = -EAGAIN;
+			break;
+		}
+
 		if (!sbitmap_any_bit_set(&hctx->ctx_map))
 			break;
 
@@ -183,21 +203,17 @@ static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx)
 	} while (blk_mq_dispatch_rq_list(q, &rq_list, true));
 
 	WRITE_ONCE(hctx->dispatch_from, ctx);
+	return ret;
 }
 
-void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
+int __blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
 {
 	struct request_queue *q = hctx->queue;
 	struct elevator_queue *e = q->elevator;
 	const bool has_sched_dispatch = e && e->type->ops.dispatch_request;
+	int ret = 0;
 	LIST_HEAD(rq_list);
 
-	/* RCU or SRCU read lock is needed before checking quiesced flag */
-	if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q)))
-		return;
-
-	hctx->run++;
-
 	/*
 	 * If we have previous entries on our dispatch list, grab them first for
 	 * more fair dispatch.
@@ -226,19 +242,41 @@ void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
 		blk_mq_sched_mark_restart_hctx(hctx);
 		if (blk_mq_dispatch_rq_list(q, &rq_list, false)) {
 			if (has_sched_dispatch)
-				blk_mq_do_dispatch_sched(hctx);
+				ret = blk_mq_do_dispatch_sched(hctx);
 			else
-				blk_mq_do_dispatch_ctx(hctx);
+				ret = blk_mq_do_dispatch_ctx(hctx);
 		}
 	} else if (has_sched_dispatch) {
-		blk_mq_do_dispatch_sched(hctx);
+		ret = blk_mq_do_dispatch_sched(hctx);
 	} else if (hctx->dispatch_busy) {
 		/* dequeue request one by one from sw queue if queue is busy */
-		blk_mq_do_dispatch_ctx(hctx);
+		ret = blk_mq_do_dispatch_ctx(hctx);
 	} else {
 		blk_mq_flush_busy_ctxs(hctx, &rq_list);
 		blk_mq_dispatch_rq_list(q, &rq_list, false);
 	}
+
+	return ret;
+}
+
+void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
+{
+	struct request_queue *q = hctx->queue;
+
+	/* RCU or SRCU read lock is needed before checking quiesced flag */
+	if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q)))
+		return;
+
+	hctx->run++;
+
+	/*
+	 * A return of -EAGAIN is an indication that hctx->dispatch is not
+	 * empty and we must run again in order to avoid starving flushes.
+	 */
+	if (__blk_mq_sched_dispatch_requests(hctx) == -EAGAIN) {
+		if (__blk_mq_sched_dispatch_requests(hctx) == -EAGAIN)
+			blk_mq_run_hw_queue(hctx, true);
+	}
 }
 
 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,