[PATCH 3/4] blk-mq: use hw tag for scheduling if hw tag space is big enough

[Ming Lei]

When tag space of one device is big enough, we use hw tag
directly for I/O scheduling.

Now the decision is made if hw queue depth is not less than
q->nr_requests and the tag set isn't shared.

Signed-off-by: Ming Lei <ming....@redhat.com>
---
 block/blk-mq-sched.c |  8 ++++++++
 block/blk-mq-sched.h | 15 +++++++++++++++
 block/blk-mq.c       | 18 +++++++++++++++++-
 3 files changed, 40 insertions(+), 1 deletion(-)

diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index 45a675f07b8b..4681e27c127e 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -507,6 +507,7 @@ int blk_mq_init_sched(struct request_queue *q, struct 
elevator_type *e)
        struct elevator_queue *eq;
        unsigned int i;
        int ret;
+       bool auto_hw_tag;
 
        if (!e) {
                q->elevator = NULL;
@@ -519,7 +520,14 @@ int blk_mq_init_sched(struct request_queue *q, struct 
elevator_type *e)
         */
        q->nr_requests = 2 * BLKDEV_MAX_RQ;
 
+       auto_hw_tag = blk_mq_sched_may_use_hw_tag(q);
+
        queue_for_each_hw_ctx(q, hctx, i) {
+               if (auto_hw_tag)
+                       hctx->flags |= BLK_MQ_F_SCHED_USE_HW_TAG;
+               else
+                       hctx->flags &= ~BLK_MQ_F_SCHED_USE_HW_TAG;
+
                ret = blk_mq_sched_alloc_tags(q, hctx, i);
                if (ret)
                        goto err;
diff --git a/block/blk-mq-sched.h b/block/blk-mq-sched.h
index edafb5383b7b..22a19c118044 100644
--- a/block/blk-mq-sched.h
+++ b/block/blk-mq-sched.h
@@ -129,4 +129,19 @@ static inline bool blk_mq_sched_needs_restart(struct 
blk_mq_hw_ctx *hctx)
        return test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
 }
 
+/*
+ * If this queue has enough hardware tags and doesn't share tags with
+ * other queues, just use hw tag directly for scheduling.
+ */
+static inline bool blk_mq_sched_may_use_hw_tag(struct request_queue *q)
+{
+       if (q->tag_set->flags & BLK_MQ_F_TAG_SHARED)
+               return false;
+
+       if (blk_mq_get_queue_depth(q) < q->nr_requests)
+               return false;
+
+       return true;
+}
+
 #endif
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 04761fb76ab4..b0bd1fb4b0f8 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2649,6 +2649,19 @@ int blk_mq_get_queue_depth(struct request_queue *q)
        return tags->bitmap_tags.sb.depth;
 }
 
+static void blk_mq_update_sched_flag(struct request_queue *q)
+{
+       struct blk_mq_hw_ctx *hctx;
+       int i;
+
+       if (!blk_mq_sched_may_use_hw_tag(q))
+               queue_for_each_hw_ctx(q, hctx, i)
+                       hctx->flags &= ~BLK_MQ_F_SCHED_USE_HW_TAG;
+       else
+               queue_for_each_hw_ctx(q, hctx, i)
+                       hctx->flags |= BLK_MQ_F_SCHED_USE_HW_TAG;
+}
+
 int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr)
 {
        struct blk_mq_tag_set *set = q->tag_set;
@@ -2681,9 +2694,12 @@ int blk_mq_update_nr_requests(struct request_queue *q, 
unsigned int nr)
                        break;
        }
 
-       if (!ret)
+       if (!ret) {
                q->nr_requests = nr;
 
+               blk_mq_update_sched_flag(q);
+       }
+
        blk_mq_unfreeze_queue(q);
        blk_mq_start_stopped_hw_queues(q, true);
 
-- 
2.9.3