scsi: core: avoid host-wide host_busy counter for scsi_mq

It isn't necessary to check the host depth in scsi_queue_rq() any more
since it has been respected by blk-mq before calling scsi_queue_rq() via
getting driver tag.

Lots of LUNs may attach to same host and per-host IOPS may reach millions,
so we should avoid expensive atomic operations on the host-wide counter in
the IO path.

This patch implements scsi_host_busy() via blk_mq_tagset_busy_iter() for
reading the count of busy IOs for scsi_mq.

It is observed that IOPS is increased by 15% in IO test on scsi_debug (32
LUNs, 32 submit queues, 1024 can_queue, libaio/dio) in a dual-socket
system.

[mkp: clarified commit message]

Cc: Omar Sandoval <[email protected]>,
Cc: "Martin K. Petersen" <[email protected]>,
Cc: James Bottomley <[email protected]>,
Cc: Christoph Hellwig <[email protected]>,
Cc: Don Brace <[email protected]>
Cc: Kashyap Desai <[email protected]>
Cc: Mike Snitzer <[email protected]>
Cc: Hannes Reinecke <[email protected]>
Cc: Laurence Oberman <[email protected]>
Cc: Bart Van Assche <[email protected]>
Signed-off-by: Ming Lei <[email protected]>
Reviewed-by: Bart Van Assche <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

Author: Ming Lei

drivers/scsi/hosts.c

@@ -563,13 +563,35 @@ struct Scsi_Host *scsi_host_get(struct Scsi_Host *shost)
 }
 EXPORT_SYMBOL(scsi_host_get);
 
+struct scsi_host_mq_in_flight {
+	int cnt;
+};
+
+static void scsi_host_check_in_flight(struct request *rq, void *data,
+		bool reserved)
+{
+	struct scsi_host_mq_in_flight *in_flight = data;
+
+	if (blk_mq_request_started(rq))
+		in_flight->cnt++;
+}
+
 /**
  * scsi_host_busy - Return the host busy counter
  * @shost:	Pointer to Scsi_Host to inc.
  **/
 int scsi_host_busy(struct Scsi_Host *shost)
 {
-	return atomic_read(&shost->host_busy);
+	struct scsi_host_mq_in_flight in_flight = {
+		.cnt = 0,
+	};
+
+	if (!shost->use_blk_mq)
+		return atomic_read(&shost->host_busy);
+
+	blk_mq_tagset_busy_iter(&shost->tag_set, scsi_host_check_in_flight,
+			&in_flight);
+	return in_flight.cnt;
 }
 EXPORT_SYMBOL(scsi_host_busy);
 

drivers/scsi/scsi_lib.c

@@ -345,7 +345,8 @@ static void scsi_dec_host_busy(struct Scsi_Host *shost)
 	unsigned long flags;
 
 	rcu_read_lock();
-	atomic_dec(&shost->host_busy);
+	if (!shost->use_blk_mq)
+		atomic_dec(&shost->host_busy);
 	if (unlikely(scsi_host_in_recovery(shost))) {
 		spin_lock_irqsave(shost->host_lock, flags);
 		if (shost->host_failed || shost->host_eh_scheduled)
@@ -444,7 +445,12 @@ static inline bool scsi_target_is_busy(struct scsi_target *starget)
 
 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
 {
-	if (shost->can_queue > 0 &&
+	/*
+	 * blk-mq can handle host queue busy efficiently via host-wide driver
+	 * tag allocation
+	 */
+
+	if (!shost->use_blk_mq && shost->can_queue > 0 &&
 	    atomic_read(&shost->host_busy) >= shost->can_queue)
 		return true;
 	if (atomic_read(&shost->host_blocked) > 0)
@@ -1600,9 +1606,12 @@ static inline int scsi_host_queue_ready(struct request_queue *q,
 	if (scsi_host_in_recovery(shost))
 		return 0;
 
-	busy = atomic_inc_return(&shost->host_busy) - 1;
+	if (!shost->use_blk_mq)
+		busy = atomic_inc_return(&shost->host_busy) - 1;
+	else
+		busy = 0;
 	if (atomic_read(&shost->host_blocked) > 0) {
-		if (busy)
+		if (busy || scsi_host_busy(shost))
 			goto starved;
 
 		/*
@@ -1616,7 +1625,7 @@ static inline int scsi_host_queue_ready(struct request_queue *q,
 				     "unblocking host at zero depth\n"));
 	}
 
-	if (shost->can_queue > 0 && busy >= shost->can_queue)
+	if (!shost->use_blk_mq && shost->can_queue > 0 && busy >= shost->can_queue)
 		goto starved;
 	if (shost->host_self_blocked)
 		goto starved;
@@ -1702,7 +1711,9 @@ static void scsi_kill_request(struct request *req, struct request_queue *q)
 	 * with the locks as normal issue path does.
 	 */
 	atomic_inc(&sdev->device_busy);
-	atomic_inc(&shost->host_busy);
+
+	if (!shost->use_blk_mq)
+		atomic_inc(&shost->host_busy);
 	if (starget->can_queue > 0)
 		atomic_inc(&starget->target_busy);