dm bufio: when done scanning return from __scan immediately
authorMikulas Patocka <mpatocka@redhat.com>
Wed, 1 Oct 2014 17:29:48 +0000 (13:29 -0400)
committerMike Snitzer <snitzer@redhat.com>
Mon, 6 Oct 2014 00:03:37 +0000 (20:03 -0400)
When __scan frees the required number of buffer entries that the
shrinker requested (nr_to_scan becomes zero) it must return.  Before
this fix the __scan code exited only the inner loop and continued in the
outer loop -- which could result in reduced performance due to extra
buffers being freed (e.g. unnecessarily evicted thinp metadata needing
to be synchronously re-read into bufio's cache).

Also, move dm_bufio_cond_resched to __scan's inner loop, so that
iterating the bufio client's lru lists doesn't result in scheduling
latency.

Reported-by: Joe Thornber <thornber@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 3.2+
drivers/md/dm-bufio.c

index eab64bcaab66654f12b97cc60dd46cee40115ccb..9ea5b6041eb2080e1d36b9320c8305669ffc6afd 100644 (file)
@@ -1473,9 +1473,9 @@ static long __scan(struct dm_bufio_client *c, unsigned long nr_to_scan,
                list_for_each_entry_safe_reverse(b, tmp, &c->lru[l], lru_list) {
                        freed += __cleanup_old_buffer(b, gfp_mask, 0);
                        if (!--nr_to_scan)
-                               break;
+                               return freed;
+                       dm_bufio_cond_resched();
                }
-               dm_bufio_cond_resched();
        }
        return freed;
 }