blk-throtl: optimize IOPS throttle for large IO scenarios

After patch 54efd50 (block: make generic_make_request handle
arbitrarily sized bios), the IO through io-throttle may be larger,
and these IOs may be further split into more small IOs. However,
IOPS throttle does not seem to be aware of this change, which
makes the calculation of IOPS of large IOs incomplete, resulting
in disk-side IOPS that does not meet expectations. Maybe we should
fix this problem.

We can reproduce it by set max_sectors_kb of disk to 128, set
blkio.write_iops_throttle to 100, run a dd instance inside blkio
and use iostat to watch IOPS:

dd if=/dev/zero of=/dev/sdb bs=1M count=1000 oflag=direct

As a result, without this change the average IOPS is 1995, with
this change the IOPS is 98.

Signed-off-by: Chunguang Xu <brookxu@tencent.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/65869aaad05475797d63b4c3fed4f529febe3c26.1627876014.git.brookxu@tencent.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/block/blk-merge.c b/block/blk-merge.c
index f8707ff7e2fcca88b95490be4c01a99484db8164..eeba8422ae8231b047d2279b448bd3673b25bb10 100644 (file)
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -348,6 +348,8 @@ void __blk_queue_split(struct bio **bio, unsigned int *nr_segs)
                trace_block_split(split, (*bio)->bi_iter.bi_sector);
                submit_bio_noacct(*bio);
                *bio = split;
+
+               blk_throtl_charge_bio_split(*bio);
        }
 }
 

diff --git a/block/blk-throttle.c b/block/blk-throttle.c
index b1b22d863bdf8955831122334caa37ee825f5e37..55c49015e5333d98fa9b7d8f0059881788b500b8 100644 (file)
--- a/block/blk-throttle.c
+++ b/block/blk-throttle.c
@@ -178,6 +178,9 @@ struct throtl_grp {
        unsigned int bad_bio_cnt; /* bios exceeding latency threshold */
        unsigned long bio_cnt_reset_time;
 
+       atomic_t io_split_cnt[2];
+       atomic_t last_io_split_cnt[2];
+
        struct blkg_rwstat stat_bytes;
        struct blkg_rwstat stat_ios;
 };
@@ -777,6 +780,8 @@ static inline void throtl_start_new_slice_with_credit(struct throtl_grp *tg,
        tg->bytes_disp[rw] = 0;
        tg->io_disp[rw] = 0;
 
+       atomic_set(&tg->io_split_cnt[rw], 0);
+
        /*
         * Previous slice has expired. We must have trimmed it after last
         * bio dispatch. That means since start of last slice, we never used
@@ -799,6 +804,9 @@ static inline void throtl_start_new_slice(struct throtl_grp *tg, bool rw)
        tg->io_disp[rw] = 0;
        tg->slice_start[rw] = jiffies;
        tg->slice_end[rw] = jiffies + tg->td->throtl_slice;
+
+       atomic_set(&tg->io_split_cnt[rw], 0);
+
        throtl_log(&tg->service_queue,
                   "[%c] new slice start=%lu end=%lu jiffies=%lu",
                   rw == READ ? 'R' : 'W', tg->slice_start[rw],
@@ -1031,6 +1039,9 @@ static bool tg_may_dispatch(struct throtl_grp *tg, struct bio *bio,
                                jiffies + tg->td->throtl_slice);
        }
 
+       if (iops_limit != UINT_MAX)
+               tg->io_disp[rw] += atomic_xchg(&tg->io_split_cnt[rw], 0);
+
        if (tg_with_in_bps_limit(tg, bio, bps_limit, &bps_wait) &&
            tg_with_in_iops_limit(tg, bio, iops_limit, &iops_wait)) {
                if (wait)
@@ -2052,12 +2063,14 @@ static void throtl_downgrade_check(struct throtl_grp *tg)
        }
 
        if (tg->iops[READ][LIMIT_LOW]) {
+               tg->last_io_disp[READ] += atomic_xchg(&tg->last_io_split_cnt[READ], 0);
                iops = tg->last_io_disp[READ] * HZ / elapsed_time;
                if (iops >= tg->iops[READ][LIMIT_LOW])
                        tg->last_low_overflow_time[READ] = now;
        }
 
        if (tg->iops[WRITE][LIMIT_LOW]) {
+               tg->last_io_disp[WRITE] += atomic_xchg(&tg->last_io_split_cnt[WRITE], 0);
                iops = tg->last_io_disp[WRITE] * HZ / elapsed_time;
                if (iops >= tg->iops[WRITE][LIMIT_LOW])
                        tg->last_low_overflow_time[WRITE] = now;
@@ -2176,6 +2189,25 @@ static inline void throtl_update_latency_buckets(struct throtl_data *td)
 }
 #endif
 
+void blk_throtl_charge_bio_split(struct bio *bio)
+{
+       struct blkcg_gq *blkg = bio->bi_blkg;
+       struct throtl_grp *parent = blkg_to_tg(blkg);
+       struct throtl_service_queue *parent_sq;
+       bool rw = bio_data_dir(bio);
+
+       do {
+               if (!parent->has_rules[rw])
+                       break;
+
+               atomic_inc(&parent->io_split_cnt[rw]);
+               atomic_inc(&parent->last_io_split_cnt[rw]);
+
+               parent_sq = parent->service_queue.parent_sq;
+               parent = sq_to_tg(parent_sq);
+       } while (parent);
+}
+
 bool blk_throtl_bio(struct bio *bio)
 {
        struct request_queue *q = bio->bi_bdev->bd_disk->queue;

diff --git a/block/blk.h b/block/blk.h
index db6f82bbb6836164a656dad599012361e4c013fe..148bdcd3aa08493ce170211029f0032e635d5711 100644 (file)
--- a/block/blk.h
+++ b/block/blk.h
@@ -293,11 +293,13 @@ int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
 extern int blk_throtl_init(struct request_queue *q);
 extern void blk_throtl_exit(struct request_queue *q);
 extern void blk_throtl_register_queue(struct request_queue *q);
+extern void blk_throtl_charge_bio_split(struct bio *bio);
 bool blk_throtl_bio(struct bio *bio);
 #else /* CONFIG_BLK_DEV_THROTTLING */
 static inline int blk_throtl_init(struct request_queue *q) { return 0; }
 static inline void blk_throtl_exit(struct request_queue *q) { }
 static inline void blk_throtl_register_queue(struct request_queue *q) { }
+static inline void blk_throtl_charge_bio_split(struct bio *bio) { }
 static inline bool blk_throtl_bio(struct bio *bio) { return false; }
 #endif /* CONFIG_BLK_DEV_THROTTLING */
 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW