blk-mq: bitmap tag: fix and cleanup
cover blk-mq: bitmap tag: fix and cleanup
Hi Jens and guys,
I just take a look at the new bitmap based tag allocation patches in -next tree, and play it for a while.
Some of them are fixes, and some of them are cleanup, please review.
Thanks,
--
Ming Lei
PATCH 1/5 blk-mq: bitmap tag: use clear_bit_unlock in bt_clear_tag()
The unlock memory barrier need to order access to req in free path and clearing tag bit, otherwise either request free path may see a allocated request, or initialized request in allocate path might be modified by the ongoing free path.
--- block/blk-mq-tag.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
block/blk-mq-tag.c +5 -1
@@ -235,7 +235,11 @@ static void bt_clear_tag(struct blk_mq_bitmap_tags *bt, unsigned int tag)const int index =TAG_TO_INDEX (tag);struct bt_wait_state *bs; -clear_bit (TAG_TO_BIT (tag), &bt->map[index].word); +/* +* The unlock memory barrier need to order access to req in free +* path and clearing tag bit +*/ +clear_bit_unlock (TAG_TO_BIT (tag), &bt->map[index].word); bs =bt_wake_ptr (bt);if (bs &&atomic_dec_and_test (&bs->wait_cnt)) {
Signature · 1 line
1.7.9.5
PATCH 2/5 blk-mq: bitmap tag: remove barrier in bt_clear_tag()
In reply to Ming LeiThe barrier isn't necessary because both atomic_dec_and_test() and wake_up() implicate one barrier.
--- block/blk-mq-tag.c | 1 - 1 file changed, 1 deletion(-)
block/blk-mq-tag.c +0 -1
@@ -243,7 +243,6 @@ static void bt_clear_tag(struct blk_mq_bitmap_tags *bt, unsigned int tag) bs =bt_wake_ptr (bt);if (bs &&atomic_dec_and_test (&bs->wait_cnt)) { -smp_mb__after_clear_bit ();atomic_set (&bs->wait_cnt,BT_WAIT_BATCH );bt_index_inc (&bt->wake_index);wake_up (&bs->wait);
Signature · 1 line
1.7.9.5
PATCH 3/5 blk-mq: bitmap tag: select random tag betweet 0 and (depth - 1)
In reply to Ming LeiThe selected tag should be selected at random between 0 and (depth - 1) with probability 1/depth, instead between 0 and (depth - 2) with probability 1/(depth - 1).
--- block/blk-mq-tag.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
block/blk-mq-tag.c +1 -1
@@ -427,7 +427,7 @@ void blk_mq_tag_init_last_tag(struct blk_mq_tags *tags, unsigned int *tag) {unsigned int depth = tags->nr_tags - tags->nr_reserved_tags; - *tag =prandom_u32 () % (depth -1 ); + *tag =prandom_u32 () % depth; }ssize_t blk_mq_tag_sysfs_show (struct blk_mq_tags *tags,char *page)
Signature · 1 line
1.7.9.5
PATCH 4/5 blk-mq: bitmap tag: cleanup blk_mq_init_tags
In reply to Ming LeiBoth nr_cache and nr_tags arn't needed for bitmap tag anymore.
--- block/blk-mq-tag.c | 4 ---- 1 file changed, 4 deletions(-)
block/blk-mq-tag.c +0 -4
@@ -394,7 +394,6 @@ enomem:struct blk_mq_tags *blk_mq_init_tags (unsigned int total_tags,unsigned int reserved_tags,int node) { -unsigned int nr_tags, nr_cache;struct blk_mq_tags *tags;if (total_tags >BLK_MQ_TAG_MAX ) {
@@ -406,9 +405,6 @@ struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags,if (!tags)return NULL ; - nr_tags = total_tags - reserved_tags; - nr_cache = nr_tags /num_online_cpus (); - tags->nr_tags = total_tags; tags->nr_reserved_tags = reserved_tags;
Signature · 1 line
1.7.9.5
PATCH 5/5 blk-mq: bitmap tag: fix wait batch for allocation
In reply to Ming LeiOne interesting point of bitmap tag allocation is that it may wait for at least BT_WAIT_BATCH times tag free for a blocked allocation. Obviously, it may hang allocation if the depth is smaller than BT_WAIT_BATCH.
This patch simply sets the wait count as 1 if depth is smaller than BT_WAIT_BATCH to avoid the problem.
Maybe better idea is that it should be set as one ratio of depth(1/8 or others), but it may need more tests for verification.
--- block/blk-mq-tag.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-)
block/blk-mq-tag.c +12 -2
@@ -135,6 +135,16 @@ static struct bt_wait_state *bt_wait_ptr(struct blk_mq_bitmap_tags *bt,return bs; } +static void bs_reset_wait_cnt (struct blk_mq_bitmap_tags *bt, +struct bt_wait_state *bs) +{ +int cnt; + + cnt = bt->depth <BT_WAIT_BATCH ?1 :BT_WAIT_BATCH ; + +atomic_set (&bs->wait_cnt, cnt); +} +static int bt_get (struct blk_mq_bitmap_tags *bt,struct blk_mq_hw_ctx *hctx,unsigned int *last_tag,gfp_t gfp) {
@@ -160,7 +170,7 @@ static int bt_get(struct blk_mq_bitmap_tags *bt, struct blk_mq_hw_ctx *hctx,break ;if (was_empty) -atomic_set (&bs->wait_cnt,BT_WAIT_BATCH ); +bs_reset_wait_cnt (bt, bs);io_schedule (); }while (1 );
@@ -243,7 +253,7 @@ static void bt_clear_tag(struct blk_mq_bitmap_tags *bt, unsigned int tag) bs =bt_wake_ptr (bt);if (bs &&atomic_dec_and_test (&bs->wait_cnt)) { -atomic_set (&bs->wait_cnt,BT_WAIT_BATCH ); +bs_reset_wait_cnt (bt, bs);bt_index_inc (&bt->wake_index);wake_up (&bs->wait); }
Signature · 1 line
1.7.9.5
Quoted 7 lines · Ming Lei
Hi Jens and guys,
I just take a look at the new bitmap based tag allocation patches in -next tree, and play it for a while.
Some of them are fixes, and some of them are cleanup, please review.
Thanks, applied 1-4, #5 I think you are looking at an older version, this one I already fixed yesterday.
Signature · 1 line
Jens Axboe
Quoted 11 lines · Jens Axboe
Quoted 7 lines · Ming Lei
Hi Jens and guys,
I just take a look at the new bitmap based tag allocation patches in -next tree, and play it for a while.
Some of them are fixes, and some of them are cleanup, please review.
Thanks, applied 1-4, #5 I think you are looking at an older version, this one I already fixed yesterday.
Thanks, it should be because I pulled from -next instead of your tree directly.
Thanks,
Signature · 1 line
Ming Lei