virtio_blk: fix race between start and stop queue
authorMing Lei <tom.leiming@gmail.com>
Fri, 16 May 2014 15:31:21 +0000 (23:31 +0800)
committerJens Axboe <axboe@fb.com>
Tue, 27 May 2014 14:41:10 +0000 (08:41 -0600)
When there isn't enough vring descriptor for adding to vq,
blk-mq will be put as stopped state until some of pending
descriptors are completed & freed.

Unfortunately, the vq's interrupt may come just before
blk-mq's BLK_MQ_S_STOPPED flag is set, so the blk-mq will
still be kept as stopped even though lots of descriptors
are completed and freed in the interrupt handler. The worst
case is that all pending descriptors are freed in the
interrupt handler, and the queue is kept as stopped forever.

This patch fixes the problem by starting/stopping blk-mq
with holding vq_lock.

Cc: Jens Axboe <axboe@kernel.dk>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <axboe@fb.com>
Conflicts:
drivers/block/virtio_blk.c

drivers/block/virtio_blk.c

index 6d8a87f252de563f1251872e5bccbd25e2e64cf7..cb9b1f8326c3c6e1327b2c4505bc94aa143cf215 100644 (file)
@@ -144,11 +144,11 @@ static void virtblk_done(struct virtqueue *vq)
                if (unlikely(virtqueue_is_broken(vq)))
                        break;
        } while (!virtqueue_enable_cb(vq));
-       spin_unlock_irqrestore(&vblk->vq_lock, flags);
 
        /* In case queue is stopped waiting for more buffers. */
        if (req_done)
                blk_mq_start_stopped_hw_queues(vblk->disk->queue);
+       spin_unlock_irqrestore(&vblk->vq_lock, flags);
 }
 
 static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req)
@@ -202,8 +202,8 @@ static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req)
        err = __virtblk_add_req(vblk->vq, vbr, vbr->sg, num);
        if (err) {
                virtqueue_kick(vblk->vq);
-               spin_unlock_irqrestore(&vblk->vq_lock, flags);
                blk_mq_stop_hw_queue(hctx);
+               spin_unlock_irqrestore(&vblk->vq_lock, flags);
                /* Out of mem doesn't actually happen, since we fall back
                 * to direct descriptors */
                if (err == -ENOMEM || err == -ENOSPC)