dm cache policy smq: allow demotions to happen even during continuous IO
authorJoe Thornber <ejt@redhat.com>
Wed, 3 May 2017 16:01:27 +0000 (12:01 -0400)
committerMike Snitzer <snitzer@redhat.com>
Thu, 4 May 2017 14:27:47 +0000 (10:27 -0400)
dm-cache's smq policy tries hard to do it's work during the idle periods
when there is no IO.  But if there are no idle periods (eg, a long fio
run) we still need to allow some demotions and promotions to occur.

To achieve this, pass @idle=true to queue_promotion()'s
free_target_met() call so that free_target_met() doesn't short-circuit
the possibility of demotion simply because it isn't an idle period.

Fixes: b29d4986d0 ("dm cache: significant rework to leverage dm-bio-prison-v2")
Reported-by: John Harrigan <jharriga@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
drivers/md/dm-cache-policy-smq.c

index e0c40aec5e964eb1428f53ee5e84814d7ba9b75a..d13d9edf8dfe4dc55d88065df4417c8f9ba94942 100644 (file)
@@ -1214,7 +1214,11 @@ static void queue_promotion(struct smq_policy *mq, dm_oblock_t oblock,
                return;
 
        if (allocator_empty(&mq->cache_alloc)) {
-               if (!free_target_met(mq, false))
+               /*
+                * We always claim to be 'idle' to ensure some demotions happen
+                * with continuous loads.
+                */
+               if (!free_target_met(mq, true))
                        queue_demotion(mq);
                return;
        }