[PATCH] __blk_rq_map_user() doesn't need to grab the queue_lock
authorJens Axboe <jens.axboe@oracle.com>
Tue, 19 Dec 2006 07:34:17 +0000 (08:34 +0100)
committerJens Axboe <jens.axboe@oracle.com>
Tue, 19 Dec 2006 07:34:17 +0000 (08:34 +0100)
It was for driver private back_merge_fn hooks, but they don't exist
anymore.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
block/ll_rw_blk.c

index 433797934bd515f309b19fc4c49625a998edd595..5d472e47c125e5d1786ebdc2b107e1abc7928cc9 100644 (file)
@@ -2359,18 +2359,10 @@ static int __blk_rq_map_user(request_queue_t *q, struct request *rq,
         */
        bio_get(bio);
 
-       /*
-        * for most (all? don't know of any) queues we could
-        * skip grabbing the queue lock here. only drivers with
-        * funky private ->back_merge_fn() function could be
-        * problematic.
-        */
-       spin_lock_irq(q->queue_lock);
        if (!rq->bio)
                blk_rq_bio_prep(q, rq, bio);
        else if (!ll_back_merge_fn(q, rq, bio)) {
                ret = -EINVAL;
-               spin_unlock_irq(q->queue_lock);
                goto unmap_bio;
        } else {
                rq->biotail->bi_next = bio;
@@ -2378,7 +2370,6 @@ static int __blk_rq_map_user(request_queue_t *q, struct request *rq,
 
                rq->data_len += bio->bi_size;
        }
-       spin_unlock_irq(q->queue_lock);
 
        return bio->bi_size;