lkml.org 
[lkml]   [2014]   [Aug]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[PATCH] blk-mq: blk_mq_freeze_queue() should allow nesting
While converting to percpu_ref for freezing, add703fda981 ("blk-mq:
use percpu_ref for mq usage count") incorrectly made
blk_mq_freeze_queue() misbehave when freezing is nested due to
percpu_ref_kill() being invoked on an already killed ref.

Fix it by making blk_mq_freeze_queue() kill and kick the queue only
for the outermost freeze attempt. All the nested ones can simply wait
for the ref to reach zero.

While at it, remove unnecessary @wake initialization from
blk_mq_unfreeze_queue().

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Ming Lei <ming.lei@canonical.com>
---
Hello, Ming, Jens.

I think something like this is the correct solution. Freezing should
be nestable after all. Ming, can you please verify whether this fixes
the issue you're seeing?

Thanks.

block/blk-mq.c | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 5189cb1..5978088 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -112,18 +112,22 @@ static void blk_mq_usage_counter_release(struct percpu_ref *ref)
*/
void blk_mq_freeze_queue(struct request_queue *q)
{
+ bool freeze;
+
spin_lock_irq(q->queue_lock);
- q->mq_freeze_depth++;
+ freeze = !q->mq_freeze_depth++;
spin_unlock_irq(q->queue_lock);

- percpu_ref_kill(&q->mq_usage_counter);
- blk_mq_run_queues(q, false);
+ if (freeze) {
+ percpu_ref_kill(&q->mq_usage_counter);
+ blk_mq_run_queues(q, false);
+ }
wait_event(q->mq_freeze_wq, percpu_ref_is_zero(&q->mq_usage_counter));
}

static void blk_mq_unfreeze_queue(struct request_queue *q)
{
- bool wake = false;
+ bool wake;

spin_lock_irq(q->queue_lock);
wake = !--q->mq_freeze_depth;

\
 
 \ /
  Last update: 2014-08-16 14:21    [W:0.049 / U:0.840 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site