scsi: core: Avoid calling synchronize_rcu() for each device in scsi_host_block()

scsi_host_block() calls scsi_internal_device_block() for each scsi_device and
scsi_internal_device_block() calls blk_mq_quiesce_queue() for each LUN.

Since synchronize_rcu() is called from blk_mq_quiesce_queue(), this can cause
substantial slowdowns on systems with many LUNs.

Use scsi_internal_device_block_nowait() to implement scsi_host_block() so it
is sufficient to run synchronize_rcu() once. This is safe since SCSI does not
set the BLK_MQ_F_BLOCKING flag.

[mkp: commit desc and comment tweaks]

Link: https://lore.kernel.org/r/[email protected]
Cc: Steffen Maier <[email protected]>
Cc: Bart Van Assche <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Dexuan Cui <[email protected]>
Cc: Hannes Reinecke <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Reviewed-by: Hannes Reinecke <[email protected]>
Signed-off-by: Ming Lei <[email protected]>
Signed-off-by: Martin K. Petersen <[email protected]>

1 files changed, 17 insertions, 1 deletions

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 47835c4b4ee0..ff0449fbe3a0 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -2841,11 +2841,27 @@ scsi_host_block(struct Scsi_Host *shost)
 	struct scsi_device *sdev;
 	int ret = 0;
 
+	/*
+	 * Call scsi_internal_device_block_nowait so we can avoid
+	 * calling synchronize_rcu() for each LUN.
+	 */
 	shost_for_each_device(sdev, shost) {
-		ret = scsi_internal_device_block(sdev);
+		mutex_lock(&sdev->state_mutex);
+		ret = scsi_internal_device_block_nowait(sdev);
+		mutex_unlock(&sdev->state_mutex);
 		if (ret)
 			break;
 	}
+
+	/*
+	 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
+	 * calling synchronize_rcu() once is enough.
+	 */
+	WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
+
+	if (!ret)
+		synchronize_rcu();
+
 	return ret;
 }
 EXPORT_SYMBOL_GPL(scsi_host_block);