[Xen-devel] [PATCH] xen-blkfront: fix a deadlock while handling discard response

[Li Dongyang <lidongyang@xxxxxxxxxx>]

When we get EOPNOTSUPP response for a discard request, we will clear
the discard flag on the request queue so we won't attempt to send discard
requests to backend again, and this should be protected under rq->queue_lock.
However, when we setup the request queue, we pass blkif_io_lock to
blk_init_queue so rq->queue_lock is blkif_io_lock indeed,
and this lock is already taken when we are in blkif_interrpt, so remove the
spin_lock/spin_unlock when we clear the discard flag or we will end up
with deadlock here, Thanks
Sorry for the stupid mistake.

Signed-off-by: Li Dongyang <lidongyang@xxxxxxxxxx>
---
 drivers/block/xen-blkfront.c |    7 +++++--
 1 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 86e2c63..d9789f18 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -747,9 +747,12 @@ static irqreturn_t blkif_interrupt(int irq, void *dev_id)
                                           info->gd->disk_name);
                                error = -EOPNOTSUPP;
                                info->feature_discard = 0;
-                               spin_lock(rq->queue_lock);
+                               /*
+                                * rq->queue_lock should be acquired before 
this call,
+                                * it is actually blkif_io_lock, and already 
been taken
+                                * once we are in blkif_interrupt
+                                */
                                queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
-                               spin_unlock(rq->queue_lock);
                        }
                        __blk_end_request_all(req, error);
                        break;
-- 
1.7.6.1


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel