blk-mq: Fix wrong wakeup batch configuration which will cause hang

Commit 0b3a3c0a74497 ("blk-mq: fix tag_get wait task can't be
awakened") will recalculate wake_batch when incrementing or decrementing
active_queues to avoid wake_batch > hctx_max_depth. At the same time, in
order to not affect performance as much as possible, the minimum wakeup
batch is set to 4. But when the QD is small (such as QD=1), if inc or dec
active_queues increases wakeup batch, that can lead to a hang:

Fix this problem with the following strategies:
QD          :  >= 32 | < 32
---------------------------------
wakeup batch:  8~4   | 3~1

Fixes: 0b3a3c0a74497 ("blk-mq: fix tag_get wait task can't be awakened")
Link: https://lore.kernel.org/linux-block/78cafe94-a787-e006-8851-69906f0c2128@huawei.com/T/#t
Reported-by: Alex Xu (Hello71) <alex_y_xu@yahoo.ca>
Signed-off-by: Laibin Qiu <qiulaibin@huawei.com>
Tested-by: Alex Xu (Hello71) <alex_y_xu@yahoo.ca>
Link: https://lore.kernel.org/r/20220127100047.1763746-1-qiulaibin@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/lib/sbitmap.c b/lib/sbitmap.c
index 6220fa67fb7e6cd2c383efe28b4b9cced6d8e70a..09d293c30fd2d38be702b6a0fa88a35a219abfed 100644 (file)
--- a/lib/sbitmap.c
+++ b/lib/sbitmap.c
@@ -488,9 +488,13 @@ void sbitmap_queue_recalculate_wake_batch(struct sbitmap_queue *sbq,
                                            unsigned int users)
 {
        unsigned int wake_batch;
+       unsigned int min_batch;
+       unsigned int depth = (sbq->sb.depth + users - 1) / users;
 
-       wake_batch = clamp_val((sbq->sb.depth + users - 1) /
-                       users, 4, SBQ_WAKE_BATCH);
+       min_batch = sbq->sb.depth >= (4 * SBQ_WAIT_QUEUES) ? 4 : 1;
+
+       wake_batch = clamp_val(depth / SBQ_WAIT_QUEUES,
+                       min_batch, SBQ_WAKE_BATCH);
        __sbitmap_queue_update_wake_batch(sbq, wake_batch);
 }
 EXPORT_SYMBOL_GPL(sbitmap_queue_recalculate_wake_batch);