[PATCH V2 2/6] blk-mq: support concurrent blk_mq_quiesce_queue()

[Ming Lei]

Turns out that blk_mq_freeze_queue() isn't stronger[1] than
blk_mq_quiesce_queue() because dispatch may still be in-progress after
queue is frozen, and in several cases, such as switching io scheduler,
and updating hw queues, we still need to quiesce queue as a supplement
of freezing queue.

As we need to extend uses of blk_mq_quiesce_queue(), it is inevitable
for us to need support cucurrent quiesce, especailly we can't let
unquiesce happen when there is quiesce happenning from other contexts.

This patch introduces q->mq_quiesce_depth to deal concurrent quiesce,
and we only unquiesce queue when it is the last one from all contexts.

[1] https://marc.info/?l=linux-block&m=150993988115872&w=2

Suggested-by: Bart Van Assche <bart.vanassche at wdc.com>
Signed-off-by: Ming Lei <ming.lei at redhat.com>
---
 block/blk-mq.c           | 21 ++++++++++++++++-----
 drivers/nvme/host/core.c |  4 ++--
 include/linux/blk-mq.h   |  7 ++++++-
 include/linux/blkdev.h   |  2 ++
 4 files changed, 26 insertions(+), 8 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 11097477eeab..5d69c8075339 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -196,7 +196,8 @@ void blk_mq_quiesce_queue_nowait(struct request_queue *q)
 	unsigned long flags;
 
 	spin_lock_irqsave(q->queue_lock, flags);
-	queue_flag_set(QUEUE_FLAG_QUIESCED, q);
+	if (!q->quiesce_depth++)
+		queue_flag_set(QUEUE_FLAG_QUIESCED, q);
 	spin_unlock_irqrestore(q->queue_lock, flags);
 }
 EXPORT_SYMBOL_GPL(blk_mq_quiesce_queue_nowait);
@@ -232,22 +233,32 @@ EXPORT_SYMBOL_GPL(blk_mq_quiesce_queue);
 /*
  * blk_mq_unquiesce_queue() - counterpart of blk_mq_quiesce_queue()
  * @q: request queue.
+ * @force: force to unquiesce if set
  *
  * This function recovers queue into the state before quiescing
  * which is done by blk_mq_quiesce_queue.
+ *
+ * Note: @force should be passed as true only when it is done before
+ * cleanup queue, for other cases, please don't use this way.
  */
-void blk_mq_unquiesce_queue(struct request_queue *q)
+void __blk_mq_unquiesce_queue(struct request_queue *q, bool force)
 {
 	unsigned long flags;
+	int depth;
 
 	spin_lock_irqsave(q->queue_lock, flags);
-	queue_flag_clear(QUEUE_FLAG_QUIESCED, q);
+	if (q->quiesce_depth > 0)
+		q->quiesce_depth--;
+	depth = q->quiesce_depth;
+	if (!depth || force)
+		queue_flag_clear(QUEUE_FLAG_QUIESCED, q);
 	spin_unlock_irqrestore(q->queue_lock, flags);
 
 	/* dispatch requests which are inserted during quiescing */
-	blk_mq_run_hw_queues(q, true);
+	if (!depth || force)
+		blk_mq_run_hw_queues(q, true);
 }
-EXPORT_SYMBOL_GPL(blk_mq_unquiesce_queue);
+EXPORT_SYMBOL_GPL(__blk_mq_unquiesce_queue);
 
 void blk_mq_wake_waiters(struct request_queue *q)
 {
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index f837d666cbd4..1ab1168cd46a 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -3371,7 +3371,7 @@ void nvme_kill_queues(struct nvme_ctrl *ctrl)
 
 	/* Forcibly unquiesce queues to avoid blocking dispatch */
 	if (ctrl->admin_q)
-		blk_mq_unquiesce_queue(ctrl->admin_q);
+		__blk_mq_unquiesce_queue(ctrl->admin_q, true);
 
 	list_for_each_entry(ns, &ctrl->namespaces, list) {
 		/*
@@ -3384,7 +3384,7 @@ void nvme_kill_queues(struct nvme_ctrl *ctrl)
 		blk_set_queue_dying(ns->queue);
 
 		/* Forcibly unquiesce queues to avoid blocking dispatch */
-		blk_mq_unquiesce_queue(ns->queue);
+		__blk_mq_unquiesce_queue(ns->queue, true);
 	}
 	mutex_unlock(&ctrl->namespaces_mutex);
 }
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 95c9a5c862e2..8a01822dc09e 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -265,7 +265,7 @@ void blk_mq_start_hw_queues(struct request_queue *q);
 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
 void blk_mq_quiesce_queue(struct request_queue *q);
-void blk_mq_unquiesce_queue(struct request_queue *q);
+void __blk_mq_unquiesce_queue(struct request_queue *q, bool force);
 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
 bool blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
@@ -286,6 +286,11 @@ void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 
 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
 
+static inline void blk_mq_unquiesce_queue(struct request_queue *q)
+{
+	__blk_mq_unquiesce_queue(q, false);
+}
+
 /*
  * Driver command data is immediately after the request. So subtract request
  * size to get back to the original request, add request size to get the PDU.
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 8089ca17db9a..ee3731f417c0 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -570,6 +570,8 @@ struct request_queue {
 	int			bypass_depth;
 	atomic_t		mq_freeze_depth;
 
+	int			quiesce_depth;
+
 #if defined(CONFIG_BLK_DEV_BSG)
 	bsg_job_fn		*bsg_job_fn;
 	struct bsg_class_device bsg_dev;
-- 
2.9.5