block, bfq: invoke flush_idle_tree after reparent_active_queues in pd_offline
authorPaolo Valente <paolo.valente@linaro.org>
Sat, 21 Mar 2020 09:45:21 +0000 (10:45 +0100)
committerJens Axboe <axboe@kernel.dk>
Sat, 21 Mar 2020 20:31:03 +0000 (14:31 -0600)
In bfq_pd_offline(), the function bfq_flush_idle_tree() is invoked to
flush the rb tree that contains all idle entities belonging to the pd
(cgroup) being destroyed. In particular, bfq_flush_idle_tree() is
invoked before bfq_reparent_active_queues(). Yet the latter may happen
to add some entities to the idle tree. It happens if, in some of the
calls to bfq_bfqq_move() performed by bfq_reparent_active_queues(),
the queue to move is empty and gets expired.

This commit simply reverses the invocation order between
bfq_flush_idle_tree() and bfq_reparent_active_queues().

Tested-by: cki-project@redhat.com
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/bfq-cgroup.c

index efb89db..68882b9 100644 (file)
@@ -893,13 +893,6 @@ static void bfq_pd_offline(struct blkg_policy_data *pd)
        for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) {
                st = bfqg->sched_data.service_tree + i;
 
-               /*
-                * The idle tree may still contain bfq_queues belonging
-                * to exited task because they never migrated to a different
-                * cgroup from the one being destroyed now.
-                */
-               bfq_flush_idle_tree(st);
-
                /*
                 * It may happen that some queues are still active
                 * (busy) upon group destruction (if the corresponding
@@ -913,6 +906,19 @@ static void bfq_pd_offline(struct blkg_policy_data *pd)
                 * scheduler has taken no reference.
                 */
                bfq_reparent_active_queues(bfqd, bfqg, st, i);
+
+               /*
+                * The idle tree may still contain bfq_queues
+                * belonging to exited task because they never
+                * migrated to a different cgroup from the one being
+                * destroyed now. In addition, even
+                * bfq_reparent_active_queues() may happen to add some
+                * entities to the idle tree. It happens if, in some
+                * of the calls to bfq_bfqq_move() performed by
+                * bfq_reparent_active_queues(), the queue to move is
+                * empty and gets expired.
+                */
+               bfq_flush_idle_tree(st);
        }
 
        __bfq_deactivate_entity(entity, false);