blk-mq: Use request queue-wide tags for tagset-wide sbitmap

The tags used for an IO scheduler are currently per hctx.

As such, when q->nr_hw_queues grows, so does the request queue total IO
scheduler tag depth.

This may cause problems for SCSI MQ HBAs whose total driver depth is
fixed.

Ming and Yanhui report higher CPU usage and lower throughput in scenarios
where the fixed total driver tag depth is appreciably lower than the total
scheduler tag depth:
https://lore.kernel.org/linux-block/440dfcfc-1a2c-bd98-1161-cec4d78c6dfc@huawei.com/T/#mc0d6d4f95275a2743d1c8c3e4dc9ff6c9aa3a76b

In that scenario, since the scheduler tag is got first, much contention
is introduced since a driver tag may not be available after we have got
the sched tag.

Improve this scenario by introducing request queue-wide tags for when
a tagset-wide sbitmap is used. The static sched requests are still
allocated per hctx, as requests are initialised per hctx, as in
blk_mq_init_request(..., hctx_idx, ...) ->
set->ops->init_request(.., hctx_idx, ...).

For simplicity of resizing the request queue sbitmap when updating the
request queue depth, just init at the max possible size, so we don't need
to deal with the possibly with swapping out a new sbitmap for old if
we need to grow.

Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/1620907258-30910-3-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

4 files changed, 72 insertions, 21 deletions

diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index 996a4b2f73aa..045b6878b8c5 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -509,11 +509,9 @@ static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set,
 				   struct blk_mq_hw_ctx *hctx,
 				   unsigned int hctx_idx)
 {
-	unsigned int flags = set->flags & ~BLK_MQ_F_TAG_HCTX_SHARED;
-
 	if (hctx->sched_tags) {
 		blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx);
-		blk_mq_free_rq_map(hctx->sched_tags, flags);
+		blk_mq_free_rq_map(hctx->sched_tags, set->flags);
 		hctx->sched_tags = NULL;
 	}
 }
@@ -523,12 +521,10 @@ static int blk_mq_sched_alloc_tags(struct request_queue *q,
 				   unsigned int hctx_idx)
 {
 	struct blk_mq_tag_set *set = q->tag_set;
-	/* Clear HCTX_SHARED so tags are init'ed */
-	unsigned int flags = set->flags & ~BLK_MQ_F_TAG_HCTX_SHARED;
 	int ret;
 
 	hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests,
-					       set->reserved_tags, flags);
+					       set->reserved_tags, set->flags);
 	if (!hctx->sched_tags)
 		return -ENOMEM;
 
@@ -546,16 +542,50 @@ static void blk_mq_sched_tags_teardown(struct request_queue *q)
 	int i;
 
 	queue_for_each_hw_ctx(q, hctx, i) {
-		/* Clear HCTX_SHARED so tags are freed */
-		unsigned int flags = hctx->flags & ~BLK_MQ_F_TAG_HCTX_SHARED;
-
 		if (hctx->sched_tags) {
-			blk_mq_free_rq_map(hctx->sched_tags, flags);
+			blk_mq_free_rq_map(hctx->sched_tags, hctx->flags);
 			hctx->sched_tags = NULL;
 		}
 	}
 }
 
+static int blk_mq_init_sched_shared_sbitmap(struct request_queue *queue)
+{
+	struct blk_mq_tag_set *set = queue->tag_set;
+	int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(set->flags);
+	struct blk_mq_hw_ctx *hctx;
+	int ret, i;
+
+	/*
+	 * Set initial depth at max so that we don't need to reallocate for
+	 * updating nr_requests.
+	 */
+	ret = blk_mq_init_bitmaps(&queue->sched_bitmap_tags,
+				  &queue->sched_breserved_tags,
+				  MAX_SCHED_RQ, set->reserved_tags,
+				  set->numa_node, alloc_policy);
+	if (ret)
+		return ret;
+
+	queue_for_each_hw_ctx(queue, hctx, i) {
+		hctx->sched_tags->bitmap_tags =
+					&queue->sched_bitmap_tags;
+		hctx->sched_tags->breserved_tags =
+					&queue->sched_breserved_tags;
+	}
+
+	sbitmap_queue_resize(&queue->sched_bitmap_tags,
+			     queue->nr_requests - set->reserved_tags);
+
+	return 0;
+}
+
+static void blk_mq_exit_sched_shared_sbitmap(struct request_queue *queue)
+{
+	sbitmap_queue_free(&queue->sched_bitmap_tags);
+	sbitmap_queue_free(&queue->sched_breserved_tags);
+}
+
 int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
 {
 	struct blk_mq_hw_ctx *hctx;
@@ -580,12 +610,18 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
 	queue_for_each_hw_ctx(q, hctx, i) {
 		ret = blk_mq_sched_alloc_tags(q, hctx, i);
 		if (ret)
-			goto err;
+			goto err_free_tags;
+	}
+
+	if (blk_mq_is_sbitmap_shared(q->tag_set->flags)) {
+		ret = blk_mq_init_sched_shared_sbitmap(q);
+		if (ret)
+			goto err_free_tags;
 	}
 
 	ret = e->ops.init_sched(q, e);
 	if (ret)
-		goto err;
+		goto err_free_sbitmap;
 
 	blk_mq_debugfs_register_sched(q);
 
@@ -605,7 +641,10 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
 
 	return 0;
 
-err:
+err_free_sbitmap:
+	if (blk_mq_is_sbitmap_shared(q->tag_set->flags))
+		blk_mq_exit_sched_shared_sbitmap(q);
+err_free_tags:
 	blk_mq_sched_free_requests(q);
 	blk_mq_sched_tags_teardown(q);
 	q->elevator = NULL;
@@ -643,5 +682,7 @@ void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e)
 	if (e->type->ops.exit_sched)
 		e->type->ops.exit_sched(e);
 	blk_mq_sched_tags_teardown(q);
+	if (blk_mq_is_sbitmap_shared(q->tag_set->flags))
+		blk_mq_exit_sched_shared_sbitmap(q);
 	q->elevator = NULL;
 }
diff --git a/block/blk-mq-sched.h b/block/blk-mq-sched.h
index 5b18ab915c65..aff037cfd8e7 100644
--- a/block/blk-mq-sched.h
+++ b/block/blk-mq-sched.h
@@ -5,6 +5,8 @@
 #include "blk-mq.h"
 #include "blk-mq-tag.h"
 
+#define MAX_SCHED_RQ (16 * BLKDEV_MAX_RQ)
+
 void blk_mq_sched_assign_ioc(struct request *rq);
 
 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index f597d40de10b..86f87346232a 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -13,6 +13,7 @@
 #include <linux/delay.h>
 #include "blk.h"
 #include "blk-mq.h"
+#include "blk-mq-sched.h"
 #include "blk-mq-tag.h"
 
 /*
@@ -590,8 +591,6 @@ int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx,
 	 */
 	if (tdepth > tags->nr_tags) {
 		struct blk_mq_tag_set *set = hctx->queue->tag_set;
-		/* Only sched tags can grow, so clear HCTX_SHARED flag  */
-		unsigned int flags = set->flags & ~BLK_MQ_F_TAG_HCTX_SHARED;
 		struct blk_mq_tags *new;
 		bool ret;
 
@@ -602,21 +601,21 @@ int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx,
 		 * We need some sort of upper limit, set it high enough that
 		 * no valid use cases should require more.
 		 */
-		if (tdepth > 16 * BLKDEV_MAX_RQ)
+		if (tdepth > MAX_SCHED_RQ)
 			return -EINVAL;
 
 		new = blk_mq_alloc_rq_map(set, hctx->queue_num, tdepth,
-				tags->nr_reserved_tags, flags);
+				tags->nr_reserved_tags, set->flags);
 		if (!new)
 			return -ENOMEM;
 		ret = blk_mq_alloc_rqs(set, new, hctx->queue_num, tdepth);
 		if (ret) {
-			blk_mq_free_rq_map(new, flags);
+			blk_mq_free_rq_map(new, set->flags);
 			return -ENOMEM;
 		}
 
 		blk_mq_free_rqs(set, *tagsptr, hctx->queue_num);
-		blk_mq_free_rq_map(*tagsptr, flags);
+		blk_mq_free_rq_map(*tagsptr, set->flags);
 		*tagsptr = new;
 	} else {
 		/*
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 001e196bdebd..f11d4018ce2e 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -3640,15 +3640,24 @@ int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr)
 		} else {
 			ret = blk_mq_tag_update_depth(hctx, &hctx->sched_tags,
 							nr, true);
+			if (blk_mq_is_sbitmap_shared(set->flags)) {
+				hctx->sched_tags->bitmap_tags =
+					&q->sched_bitmap_tags;
+				hctx->sched_tags->breserved_tags =
+					&q->sched_breserved_tags;
+			}
 		}
 		if (ret)
 			break;
 		if (q->elevator && q->elevator->type->ops.depth_updated)
 			q->elevator->type->ops.depth_updated(hctx);
 	}
-
-	if (!ret)
+	if (!ret) {
 		q->nr_requests = nr;
+		if (q->elevator && blk_mq_is_sbitmap_shared(set->flags))
+			sbitmap_queue_resize(&q->sched_bitmap_tags,
+					     nr - set->reserved_tags);
+	}
 
 	blk_mq_unquiesce_queue(q);
 	blk_mq_unfreeze_queue(q);