[PATCH] scsi: scsi_transport_fc: Change to use per-rport devloss_work_q

[PATCH] scsi: scsi_transport_fc: Change to use per-rport devloss_work_q

[Ewan D. Milne]

Configurations with large numbers of FC rports per host instance are taking a
very long time to complete all devloss work.  Increase potential parallelism
by using a per-rport devloss_work_q for dev_loss_work and fast_io_fail_work.

Signed-off-by: Ewan D. Milne <emilne@redhat.com>
---
 drivers/scsi/scsi_transport_fc.c | 70 ++++++++++++++++++--------------
 include/scsi/scsi_transport_fc.h |  5 +--
 2 files changed, 42 insertions(+), 33 deletions(-)

diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c
index 6b165a3ec6de..82d091d627c0 100644
--- a/drivers/scsi/scsi_transport_fc.c
+++ b/drivers/scsi/scsi_transport_fc.c
@@ -446,13 +446,6 @@ static int fc_host_setup(struct transport_container *tc, struct device *dev,
 		return -ENOMEM;
 
 	fc_host->dev_loss_tmo = fc_dev_loss_tmo;
-	fc_host->devloss_work_q = alloc_workqueue("fc_dl_%d", 0, 0,
-					shost->host_no);
-	if (!fc_host->devloss_work_q) {
-		destroy_workqueue(fc_host->work_q);
-		fc_host->work_q = NULL;
-		return -ENOMEM;
-	}
 
 	fc_bsg_hostadd(shost, fc_host);
 	/* ignore any bsg add error - we just can't do sgio */
@@ -2821,10 +2814,10 @@ fc_flush_work(struct Scsi_Host *shost)
  * 	1 on success / 0 already queued / < 0 for error
  */
 static int
-fc_queue_devloss_work(struct Scsi_Host *shost, struct delayed_work *work,
-				unsigned long delay)
+fc_queue_devloss_work(struct Scsi_Host *shost, struct fc_rport *rport,
+		      struct delayed_work *work, unsigned long delay)
 {
-	if (unlikely(!fc_host_devloss_work_q(shost))) {
+	if (unlikely(!rport->devloss_work_q)) {
 		printk(KERN_ERR
 			"ERROR: FC host '%s' attempted to queue work, "
 			"when no workqueue created.\n", shost->hostt->name);
@@ -2833,7 +2826,7 @@ fc_queue_devloss_work(struct Scsi_Host *shost, struct delayed_work *work,
 		return -EINVAL;
 	}
 
-	return queue_delayed_work(fc_host_devloss_work_q(shost), work, delay);
+	return queue_delayed_work(rport->devloss_work_q, work, delay);
 }
 
 /**
@@ -2841,9 +2834,9 @@ fc_queue_devloss_work(struct Scsi_Host *shost, struct delayed_work *work,
  * @shost:	Pointer to Scsi_Host bound to fc_host.
  */
 static void
-fc_flush_devloss(struct Scsi_Host *shost)
+fc_flush_devloss(struct Scsi_Host *shost, struct fc_rport *rport)
 {
-	if (!fc_host_devloss_work_q(shost)) {
+	if (unlikely(!rport->devloss_work_q)) {
 		printk(KERN_ERR
 			"ERROR: FC host '%s' attempted to flush work, "
 			"when no workqueue created.\n", shost->hostt->name);
@@ -2851,7 +2844,7 @@ fc_flush_devloss(struct Scsi_Host *shost)
 		return;
 	}
 
-	flush_workqueue(fc_host_devloss_work_q(shost));
+	flush_workqueue(rport->devloss_work_q);
 }
 
 
@@ -2913,13 +2906,6 @@ fc_remove_host(struct Scsi_Host *shost)
 		fc_host->work_q = NULL;
 		destroy_workqueue(work_q);
 	}
-
-	/* flush all devloss work items, then kill it  */
-	if (fc_host->devloss_work_q) {
-		work_q = fc_host->devloss_work_q;
-		fc_host->devloss_work_q = NULL;
-		destroy_workqueue(work_q);
-	}
 }
 EXPORT_SYMBOL(fc_remove_host);
 
@@ -2967,6 +2953,7 @@ fc_rport_final_delete(struct work_struct *work)
 	struct device *dev = &rport->dev;
 	struct Scsi_Host *shost = rport_to_shost(rport);
 	struct fc_internal *i = to_fc_internal(shost->transportt);
+	struct workqueue_struct *work_q;
 	unsigned long flags;
 	int do_callback = 0;
 
@@ -2988,9 +2975,9 @@ fc_rport_final_delete(struct work_struct *work)
 	if (rport->flags & FC_RPORT_DEVLOSS_PENDING) {
 		spin_unlock_irqrestore(shost->host_lock, flags);
 		if (!cancel_delayed_work(&rport->fail_io_work))
-			fc_flush_devloss(shost);
+			fc_flush_devloss(shost, rport);
 		if (!cancel_delayed_work(&rport->dev_loss_work))
-			fc_flush_devloss(shost);
+			fc_flush_devloss(shost, rport);
 		cancel_work_sync(&rport->scan_work);
 		spin_lock_irqsave(shost->host_lock, flags);
 		rport->flags &= ~FC_RPORT_DEVLOSS_PENDING;
@@ -3021,6 +3008,12 @@ fc_rport_final_delete(struct work_struct *work)
 
 	fc_bsg_remove(rport->rqst_q);
 
+	if (rport->devloss_work_q) {
+		work_q = rport->devloss_work_q;
+		rport->devloss_work_q = NULL;
+		destroy_workqueue(work_q);
+	}
+
 	transport_remove_device(dev);
 	device_del(dev);
 	transport_destroy_device(dev);
@@ -3093,6 +3086,22 @@ fc_remote_port_create(struct Scsi_Host *shost, int channel,
 
 	spin_unlock_irqrestore(shost->host_lock, flags);
 
+	rport->devloss_work_q = alloc_workqueue("fc_dl_%d_%d", 0, 0,
+						shost->host_no, rport->number);
+	if (!rport->devloss_work_q) {
+		printk(KERN_ERR "FC Remote Port alloc_workqueue failed\n");
+/*
+ * Note that we have not yet called device_initialize() / get_device()
+ * Cannot reclaim incremented rport->number because we released host_lock
+ */
+		spin_lock_irqsave(shost->host_lock, flags);
+		list_del(&rport->peers);
+		scsi_host_put(shost);			/* for fc_host->rport list */
+		spin_unlock_irqrestore(shost->host_lock, flags);
+		kfree(rport);
+		return NULL;
+	}
+
 	dev = &rport->dev;
 	device_initialize(dev);			/* takes self reference */
 	dev->parent = get_device(&shost->shost_gendev); /* parent reference */
@@ -3255,9 +3264,9 @@ fc_remote_port_add(struct Scsi_Host *shost, int channel,
 				 * be checked and will NOOP the function.
 				 */
 				if (!cancel_delayed_work(&rport->fail_io_work))
-					fc_flush_devloss(shost);
+					fc_flush_devloss(shost, rport);
 				if (!cancel_delayed_work(&rport->dev_loss_work))
-					fc_flush_devloss(shost);
+					fc_flush_devloss(shost, rport);
 
 				spin_lock_irqsave(shost->host_lock, flags);
 
@@ -3451,11 +3460,12 @@ fc_remote_port_delete(struct fc_rport  *rport)
 	/* see if we need to kill io faster than waiting for device loss */
 	if ((rport->fast_io_fail_tmo != -1) &&
 	    (rport->fast_io_fail_tmo < timeout))
-		fc_queue_devloss_work(shost, &rport->fail_io_work,
-					rport->fast_io_fail_tmo * HZ);
+		fc_queue_devloss_work(shost, rport, &rport->fail_io_work,
+				      rport->fast_io_fail_tmo * HZ);
 
 	/* cap the length the devices can be blocked until they are deleted */
-	fc_queue_devloss_work(shost, &rport->dev_loss_work, timeout * HZ);
+	fc_queue_devloss_work(shost, rport, &rport->dev_loss_work,
+			      timeout * HZ);
 }
 EXPORT_SYMBOL(fc_remote_port_delete);
 
@@ -3514,9 +3524,9 @@ fc_remote_port_rolechg(struct fc_rport  *rport, u32 roles)
 		 * transaction.
 		 */
 		if (!cancel_delayed_work(&rport->fail_io_work))
-			fc_flush_devloss(shost);
+			fc_flush_devloss(shost, rport);
 		if (!cancel_delayed_work(&rport->dev_loss_work))
-			fc_flush_devloss(shost);
+			fc_flush_devloss(shost, rport);
 
 		spin_lock_irqsave(shost->host_lock, flags);
 		rport->flags &= ~(FC_RPORT_FAST_FAIL_TIMEDOUT |
diff --git a/include/scsi/scsi_transport_fc.h b/include/scsi/scsi_transport_fc.h
index d02b55261307..b908aacfef48 100644
--- a/include/scsi/scsi_transport_fc.h
+++ b/include/scsi/scsi_transport_fc.h
@@ -383,6 +383,8 @@ struct fc_rport {	/* aka fc_starget_attrs */
  	struct work_struct stgt_delete_work;
 	struct work_struct rport_delete_work;
 	struct request_queue *rqst_q;	/* bsg support */
+
+	struct workqueue_struct *devloss_work_q;
 } __attribute__((aligned(sizeof(unsigned long))));
 
 /* bit field values for struct fc_rport "flags" field: */
@@ -576,7 +578,6 @@ struct fc_host_attrs {
 
 	/* work queues for rport state manipulation */
 	struct workqueue_struct *work_q;
-	struct workqueue_struct *devloss_work_q;
 
 	/* bsg support */
 	struct request_queue *rqst_q;
@@ -654,8 +655,6 @@ struct fc_host_attrs {
 	(((struct fc_host_attrs *)(x)->shost_data)->npiv_vports_inuse)
 #define fc_host_work_q(x) \
 	(((struct fc_host_attrs *)(x)->shost_data)->work_q)
-#define fc_host_devloss_work_q(x) \
-	(((struct fc_host_attrs *)(x)->shost_data)->devloss_work_q)
 #define fc_host_dev_loss_tmo(x) \
 	(((struct fc_host_attrs *)(x)->shost_data)->dev_loss_tmo)
 #define fc_host_max_ct_payload(x)  \
-- 
2.43.0

Re: [PATCH] scsi: scsi_transport_fc: Change to use per-rport devloss_work_q

[Hannes Reinecke]

On 7/7/25 22:22, Ewan D. Milne wrote:
> Configurations with large numbers of FC rports per host instance are taking a
> very long time to complete all devloss work.  Increase potential parallelism
> by using a per-rport devloss_work_q for dev_loss_work and fast_io_fail_work.
> 
> Signed-off-by: Ewan D. Milne <emilne@redhat.com>
> ---
>   drivers/scsi/scsi_transport_fc.c | 70 ++++++++++++++++++--------------
>   include/scsi/scsi_transport_fc.h |  5 +--
>   2 files changed, 42 insertions(+), 33 deletions(-)
> 

Reviewed-by: Hannes Reinecke <hare@suse.de>

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                  Kernel Storage Architect
hare@suse.de                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich

Re: [PATCH] scsi: scsi_transport_fc: Change to use per-rport devloss_work_q

[Martin K. Petersen]

Ewan,

> Configurations with large numbers of FC rports per host instance are
> taking a very long time to complete all devloss work. Increase
> potential parallelism by using a per-rport devloss_work_q for
> dev_loss_work and fast_io_fail_work.

Applied to 6.17/scsi-staging, thanks!

-- 
Martin K. Petersen

Re: [PATCH] scsi: scsi_transport_fc: Change to use per-rport devloss_work_q

[Martin K. Petersen]

On Mon, 07 Jul 2025 16:22:25 -0400, Ewan D. Milne wrote:

> Configurations with large numbers of FC rports per host instance are taking a
> very long time to complete all devloss work.  Increase potential parallelism
> by using a per-rport devloss_work_q for dev_loss_work and fast_io_fail_work.
> 
> 

Applied to 6.17/scsi-queue, thanks!

[1/1] scsi: scsi_transport_fc: Change to use per-rport devloss_work_q
      https://git.kernel.org/mkp/scsi/c/25236d4844ad

-- 
Martin K. Petersen	Oracle Linux Engineering