If someone else has found active waitqueue and updated bt->wake_index, do not modify it again. This saves an atomic read.
Signed-off-by: Wenbo Wang <mail_weber_w...@163.com> CC: linux-bl...@vger.kernel.org CC: linux-kernel@vger.kernel.org --- block/blk-mq-tag.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c index abdbb47..5ed9111 100644 --- a/block/blk-mq-tag.c +++ b/block/blk-mq-tag.c @@ -356,16 +356,15 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data) static struct bt_wait_state *bt_wake_ptr(struct blk_mq_bitmap_tags *bt) { - int i, wake_index; + int i, wake_index, orig_wake_index; - wake_index = atomic_read(&bt->wake_index); + orig_wake_index = wake_index = atomic_read(&bt->wake_index); for (i = 0; i < BT_WAIT_QUEUES; i++) { struct bt_wait_state *bs = &bt->bs[wake_index]; if (waitqueue_active(&bs->wait)) { - int o = atomic_read(&bt->wake_index); - if (wake_index != o) - atomic_cmpxchg(&bt->wake_index, o, wake_index); + if (wake_index != orig_wake_index) + atomic_cmpxchg(&bt->wake_index, orig_wake_index, wake_index); return bs; } -- 1.8.3.1 --- Not seeing too much benifit from this patch, but it makes the logic here more clear