From: Hannes Reinecke <hare@xxxxxxx> scsi_target_block() calls scsi_stop_queue() for each scsi_device and scsi_stop_queue() calls blk_mq_wait_quiesce_done() for each LUN. As blk_mq_wait_quiesce_done() comes down to synchronize_rcu() for SCSI queues, this can cause substantial delay for scsi_target_block() on a target with a lot of logical units (we measured more than 100s delay for blocking a FC rport with 2048 LUNs). Simplify scsi_stop_queue(), which is only called in this code path, to never wait for the quiescing to finish. Rather call synchronize_rcu() from scsi_target_block() after iterating over all devices. Also, move the call to scsi_stop_queue() in scsi_internal_device_block() out of the code section where the state_mutex is held. This patch uses the same basic idea as commit f983622ae605 ("scsi: core: Avoid calling synchronize_rcu() for each device in scsi_host_block()"). Signed-off-by: Hannes Reinecke <hare@xxxxxxx> Signed-off-by: Martin Wilck <mwilck@xxxxxxxx> --- drivers/scsi/scsi_lib.c | 29 +++++++++++++++-------------- 1 file changed, 15 insertions(+), 14 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index b7c569a42aa4..ed5bc9e1dbed 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -2725,24 +2725,18 @@ void scsi_start_queue(struct scsi_device *sdev) blk_mq_unquiesce_queue(sdev->request_queue); } -static void scsi_stop_queue(struct scsi_device *sdev, bool nowait) +static void scsi_stop_queue(struct scsi_device *sdev) { /* * The atomic variable of ->queue_stopped covers that * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue. * * However, we still need to wait until quiesce is done - * in case that queue has been stopped. + * in case that queue has been stopped. This is done in + * scsi_target_block() for all devices of the target. */ - if (!cmpxchg(&sdev->queue_stopped, 0, 1)) { - if (nowait) - blk_mq_quiesce_queue_nowait(sdev->request_queue); - else - blk_mq_quiesce_queue(sdev->request_queue); - } else { - if (!nowait) - blk_mq_wait_quiesce_done(sdev->request_queue->tag_set); - } + if (!cmpxchg(&sdev->queue_stopped, 0, 1)) + blk_mq_quiesce_queue_nowait(sdev->request_queue); } /** @@ -2769,7 +2763,7 @@ int scsi_internal_device_block_nowait(struct scsi_device *sdev) * request queue. */ if (!ret) - scsi_stop_queue(sdev, true); + scsi_stop_queue(sdev); return ret; } EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); @@ -2795,9 +2789,9 @@ static int scsi_internal_device_block(struct scsi_device *sdev) mutex_lock(&sdev->state_mutex); err = __scsi_internal_device_block_nowait(sdev); - if (err == 0) - scsi_stop_queue(sdev, false); mutex_unlock(&sdev->state_mutex); + if (err == 0) + scsi_stop_queue(sdev); return err; } @@ -2910,6 +2904,13 @@ scsi_target_block(struct device *dev) device_block); else device_for_each_child(dev, NULL, target_block); + + /* + * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag, + * so blk_mq_wait_quiesce_done() comes down to just synchronize_rcu(). + * Just calling it once is enough. + */ + synchronize_rcu(); } EXPORT_SYMBOL_GPL(scsi_target_block); -- 2.40.1