From 4acb83417cadfdcbe64215f9d0ddcf3132af808e Mon Sep 17 00:00:00 2001 From: Keith Busch Date: Fri, 9 Sep 2022 11:40:22 -0700 Subject: sbitmap: fix batched wait_cnt accounting Batched completions can clear multiple bits, but we're only decrementing the wait_cnt by one each time. This can cause waiters to never be woken, stalling IO. Use the batched count instead. Link: https://bugzilla.kernel.org/show_bug.cgi?id=215679 Signed-off-by: Keith Busch Link: https://lore.kernel.org/r/20220909184022.1709476-1-kbusch@fb.com Signed-off-by: Jens Axboe --- lib/sbitmap.c | 37 +++++++++++++++++++++++-------------- 1 file changed, 23 insertions(+), 14 deletions(-) (limited to 'lib') diff --git a/lib/sbitmap.c b/lib/sbitmap.c index cbfd2e677d87..624fa7f118d1 100644 --- a/lib/sbitmap.c +++ b/lib/sbitmap.c @@ -599,24 +599,31 @@ static struct sbq_wait_state *sbq_wake_ptr(struct sbitmap_queue *sbq) return NULL; } -static bool __sbq_wake_up(struct sbitmap_queue *sbq) +static bool __sbq_wake_up(struct sbitmap_queue *sbq, int *nr) { struct sbq_wait_state *ws; unsigned int wake_batch; - int wait_cnt; + int wait_cnt, cur, sub; bool ret; + if (*nr <= 0) + return false; + ws = sbq_wake_ptr(sbq); if (!ws) return false; - wait_cnt = atomic_dec_return(&ws->wait_cnt); - /* - * For concurrent callers of this, callers should call this function - * again to wakeup a new batch on a different 'ws'. - */ - if (wait_cnt < 0) - return true; + cur = atomic_read(&ws->wait_cnt); + do { + /* + * For concurrent callers of this, callers should call this + * function again to wakeup a new batch on a different 'ws'. + */ + if (cur == 0) + return true; + sub = min(*nr, cur); + wait_cnt = cur - sub; + } while (!atomic_try_cmpxchg(&ws->wait_cnt, &cur, wait_cnt)); /* * If we decremented queue without waiters, retry to avoid lost @@ -625,6 +632,8 @@ static bool __sbq_wake_up(struct sbitmap_queue *sbq) if (wait_cnt > 0) return !waitqueue_active(&ws->wait); + *nr -= sub; + /* * When wait_cnt == 0, we have to be particularly careful as we are * responsible to reset wait_cnt regardless whether we've actually @@ -660,12 +669,12 @@ static bool __sbq_wake_up(struct sbitmap_queue *sbq) sbq_index_atomic_inc(&sbq->wake_index); atomic_set(&ws->wait_cnt, wake_batch); - return ret; + return ret || *nr; } -void sbitmap_queue_wake_up(struct sbitmap_queue *sbq) +void sbitmap_queue_wake_up(struct sbitmap_queue *sbq, int nr) { - while (__sbq_wake_up(sbq)) + while (__sbq_wake_up(sbq, &nr)) ; } EXPORT_SYMBOL_GPL(sbitmap_queue_wake_up); @@ -705,7 +714,7 @@ void sbitmap_queue_clear_batch(struct sbitmap_queue *sbq, int offset, atomic_long_andnot(mask, (atomic_long_t *) addr); smp_mb__after_atomic(); - sbitmap_queue_wake_up(sbq); + sbitmap_queue_wake_up(sbq, nr_tags); sbitmap_update_cpu_hint(&sbq->sb, raw_smp_processor_id(), tags[nr_tags - 1] - offset); } @@ -733,7 +742,7 @@ void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr, * waiter. See the comment on waitqueue_active(). */ smp_mb__after_atomic(); - sbitmap_queue_wake_up(sbq); + sbitmap_queue_wake_up(sbq, 1); sbitmap_update_cpu_hint(&sbq->sb, cpu, nr); } EXPORT_SYMBOL_GPL(sbitmap_queue_clear); -- cgit v1.2.3