drbd: performance - don't lose unplug events

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>

1 files changed, 6 insertions, 1 deletions

diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index 3678d3d66c6c..d09aac4a84ec 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -505,7 +505,7 @@ void __req_mod(struct drbd_request *req, enum drbd_req_event what,
                  * corresponding hlist_del is in _req_may_be_done() */
                 hlist_add_head(&req->colision, ar_hash_slot(mdev, req->sector));
 
-                set_bit(UNPLUG_REMOTE, &mdev->flags); /* why? */
+                set_bit(UNPLUG_REMOTE, &mdev->flags);
 
                 D_ASSERT(req->rq_state & RQ_NET_PENDING);
                 req->rq_state |= RQ_NET_QUEUED;
@@ -536,6 +536,11 @@ void __req_mod(struct drbd_request *req, enum drbd_req_event what,
                  *
                  * Add req to the (now) current epoch (barrier). */
 
+                /* otherwise we may lose an unplug, which may cause some remote
+                 * io-scheduler timeout to expire, increasing maximum latency,
+                 * hurting performance. */
+                set_bit(UNPLUG_REMOTE, &mdev->flags);
+
                 /* see drbd_make_request_common,
                  * just after it grabs the req_lock */
                 D_ASSERT(test_bit(CREATE_BARRIER, &mdev->flags) == 0);