blk-mq: mark request queue as mq asap
authorMing Lei <tom.leiming@gmail.com>
Fri, 12 Feb 2016 07:27:00 +0000 (15:27 +0800)
committerJens Axboe <axboe@fb.com>
Sun, 14 Feb 2016 22:35:14 +0000 (15:35 -0700)
commit66841672161efb9e3be4a1dbd9755020bb1d86b7
tree878b571be06833c5b407d7eafe2e336e4acad7e9
parentd5df731ab804e0d917f44099bfeb88a5f1488a3d
blk-mq: mark request queue as mq asap

Currently q->mq_ops is used widely to decide if the queue
is mq or not, so we should set the 'flag' asap so that both
block core and drivers can get the correct mq info.

For example, commit 868f2f0b720(blk-mq: dynamic h/w context count)
moves the hctx's initialization before setting q->mq_ops in
blk_mq_init_allocated_queue(), then cause blk_alloc_flush_queue()
to think the queue is non-mq and don't allocate command size
for the per-hctx flush rq.

This patches should fix the problem reported by Sasha.

Cc: Keith Busch <keith.busch@intel.com>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Fixes: 868f2f0b720 ("blk-mq: dynamic h/w context count")
Signed-off-by: Jens Axboe <axboe@fb.com>
block/blk-mq.c