[PATCH] blk-throtl: avoid double charge of bio IOPS due to split

[PATCH] blk-throtl: avoid double charge of bio IOPS due to split

[brookxu]

From: Chunguang Xu <brookxu@tencent.com>

After commit 900e08075202("block: move queue enter logic into
blk_mq_submit_bio()"), submit_bio_checks() moved to __submit_bio_fops()
and blk_mq_submit_bio(). The IOs go through blk_mq_submit_bio()
may be splited before entering blk-throtl, so we need to check
whether the BIO is throttled, and only update the io_split_cnt
for the THROTTLED bio to avoid double charge.

Signed-off-by: Chunguang Xu <brookxu@tencent.com>
---
 block/blk-throttle.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/block/blk-throttle.c b/block/blk-throttle.c
index 39bb6e6..2b12fc7 100644
--- a/block/blk-throttle.c
+++ b/block/blk-throttle.c
@@ -2049,6 +2049,9 @@ void blk_throtl_charge_bio_split(struct bio *bio)
 	struct throtl_service_queue *parent_sq;
 	bool rw = bio_data_dir(bio);
 
+	if (!bio_flagged(bio, BIO_THROTTLED))
+		return;
+
 	do {
 		if (!parent->has_rules[rw])
 			break;
-- 
1.8.3.1

Re: [PATCH] blk-throtl: avoid double charge of bio IOPS due to split

[Ming Lei]

Hi Chunguang,

On Mon, Jan 10, 2022 at 04:29:23PM +0800, brookxu wrote:
> From: Chunguang Xu <brookxu@tencent.com>
> 
> After commit 900e08075202("block: move queue enter logic into
> blk_mq_submit_bio()"), submit_bio_checks() moved to __submit_bio_fops()
> and blk_mq_submit_bio(). The IOs go through blk_mq_submit_bio()
> may be splited before entering blk-throtl, so we need to check
> whether the BIO is throttled, and only update the io_split_cnt
> for the THROTTLED bio to avoid double charge.

Today Jens has merged the patch[1] of "block: don't protect
submit_bio_checks by q_usage_counter", so the behavior is switched
back to previous behavior: running bio check before calling
blk_mq_submit_bio() or ->submit_bio().

Then looks your patch isn't needn't any more.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git/commit/?h=for-5.17/block&id=9d497e2941c30a060ba62d5485b3bc9d91ffb09e


Thanks,
Ming