2007-10-20 12:15:33

by Oleg Nesterov

[permalink] [raw]
Subject: [PATCH 1/2] cfq_exit_queue() should cancel cfq_data->unplug_work

(untested, needs an explicit ack/nack from maintainer)

Spotted by Nick <[email protected]>, perhaps explains the first trace in
http://bugzilla.kernel.org/show_bug.cgi?id=9180.

cfq_exit_queue() should cancel cfqd->unplug_work before freeing cfqd.
blk_sync_queue() seems unneeded, removed.

Q: why cfq_exit_queue() calls cfq_shutdown_timer_wq() twice?

Signed-off-by: Oleg Nesterov <[email protected]>

--- K/block/cfq-iosched.c~1_cfq 2007-08-09 19:59:26.000000000 +0400
+++ K/block/cfq-iosched.c 2007-10-20 15:48:35.000000000 +0400
@@ -2053,7 +2053,7 @@ static void cfq_shutdown_timer_wq(struct
{
del_timer_sync(&cfqd->idle_slice_timer);
del_timer_sync(&cfqd->idle_class_timer);
- blk_sync_queue(cfqd->queue);
+ kblockd_flush_work(&cfqd->unplug_work);
}

static void cfq_put_async_queues(struct cfq_data *cfqd)