1 files changed, 36 insertions, 12 deletions

diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index a905b5519ab6..e2e6719832e1 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -300,7 +300,9 @@ struct cfq_data {
 
         /* List of cfq groups being managed on this device*/
         struct hlist_head cfqg_list;
-        struct rcu_head rcu;
+
+        /* Number of groups which are on blkcg->blkg_list */
+        unsigned int nr_blkcg_linked_grps;
 };
 
 static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
@@ -1063,6 +1065,7 @@ static struct cfq_group * cfq_find_alloc_cfqg(struct cfq_data *cfqd,
                 cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
                                         0);
 
+        cfqd->nr_blkcg_linked_grps++;
         cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
 
         /* Add group on cfqd list */
@@ -3815,15 +3818,11 @@ static void cfq_put_async_queues(struct cfq_data *cfqd)
                 cfq_put_queue(cfqd->async_idle_cfqq);
 }
 
-static void cfq_cfqd_free(struct rcu_head *head)
-{
-        kfree(container_of(head, struct cfq_data, rcu));
-}
-
 static void cfq_exit_queue(struct elevator_queue *e)
 {
         struct cfq_data *cfqd = e->elevator_data;
         struct request_queue *q = cfqd->queue;
+        bool wait = false;
 
         cfq_shutdown_timer_wq(cfqd);
 
@@ -3842,7 +3841,13 @@ static void cfq_exit_queue(struct elevator_queue *e)
 
         cfq_put_async_queues(cfqd);
         cfq_release_cfq_groups(cfqd);
-        cfq_blkiocg_del_blkio_group(&cfqd->root_group.blkg);
+
+        /*
+         * If there are groups which we could not unlink from blkcg list,
+         * wait for a rcu period for them to be freed.
+         */
+        if (cfqd->nr_blkcg_linked_grps)
+                wait = true;
 
         spin_unlock_irq(q->queue_lock);
 
@@ -3852,8 +3857,20 @@ static void cfq_exit_queue(struct elevator_queue *e)
         ida_remove(&cic_index_ida, cfqd->cic_index);
         spin_unlock(&cic_index_lock);
 
-        /* Wait for cfqg->blkg->key accessors to exit their grace periods. */
-        call_rcu(&cfqd->rcu, cfq_cfqd_free);
+        /*
+         * Wait for cfqg->blkg->key accessors to exit their grace periods.
+         * Do this wait only if there are other unlinked groups out
+         * there. This can happen if cgroup deletion path claimed the
+         * responsibility of cleaning up a group before queue cleanup code
+         * get to the group.
+         *
+         * Do not call synchronize_rcu() unconditionally as there are drivers
+         * which create/delete request queue hundreds of times during scan/boot
+         * and synchronize_rcu() can take significant time and slow down boot.
+         */
+        if (wait)
+                synchronize_rcu();
+        kfree(cfqd);
 }
 
 static int cfq_alloc_cic_index(void)
@@ -3909,14 +3926,21 @@ static void *cfq_init_queue(struct request_queue *q)
 
 #ifdef CONFIG_CFQ_GROUP_IOSCHED
         /*
-         * Take a reference to root group which we never drop. This is just
-         * to make sure that cfq_put_cfqg() does not try to kfree root group
+         * Set root group reference to 2. One reference will be dropped when
+         * all groups on cfqd->cfqg_list are being deleted during queue exit.
+         * Other reference will remain there as we don't want to delete this
+         * group as it is statically allocated and gets destroyed when
+         * throtl_data goes away.
          */
-        cfqg->ref = 1;
+        cfqg->ref = 2;
         rcu_read_lock();
         cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg,
                                         (void *)cfqd, 0);
         rcu_read_unlock();
+        cfqd->nr_blkcg_linked_grps++;
+
+        /* Add group on cfqd->cfqg_list */
+        hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
 #endif
         /*
          * Not strictly needed (since RB_ROOT just clears the node and we