1 files changed, 183 insertions, 76 deletions

diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index ab7a9e6a9b1c..f3799432676d 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -185,7 +185,7 @@ struct cfq_group {
 	int nr_cfqq;
 
 	/*
-	 * Per group busy queus average. Useful for workload slice calc. We
+	 * Per group busy queues average. Useful for workload slice calc. We
 	 * create the array for each prio class but at run time it is used
 	 * only for RT and BE class and slot for IDLE class remains unused.
 	 * This is primarily done to avoid confusion and a gcc warning.
@@ -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);
@@ -367,16 +369,16 @@ CFQ_CFQQ_FNS(wait_busy);
 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
 	blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \
 			cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
-			blkg_path(&(cfqq)->cfqg->blkg), ##args);
+			blkg_path(&(cfqq)->cfqg->blkg), ##args)
 
 #define cfq_log_cfqg(cfqd, cfqg, fmt, args...)				\
 	blk_add_trace_msg((cfqd)->queue, "%s " fmt,			\
-				blkg_path(&(cfqg)->blkg), ##args);      \
+				blkg_path(&(cfqg)->blkg), ##args)       \
 
 #else
 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
 	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
-#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0);
+#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0)
 #endif
 #define cfq_log(cfqd, fmt, args...)	\
 	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
@@ -665,15 +667,11 @@ cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2,
 	if (rq2 == NULL)
 		return rq1;
 
-	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
-		return rq1;
-	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
-		return rq2;
-	if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
-		return rq1;
-	else if ((rq2->cmd_flags & REQ_META) &&
-		 !(rq1->cmd_flags & REQ_META))
-		return rq2;
+	if (rq_is_sync(rq1) != rq_is_sync(rq2))
+		return rq_is_sync(rq1) ? rq1 : rq2;
+
+	if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_META)
+		return rq1->cmd_flags & REQ_META ? rq1 : rq2;
 
 	s1 = blk_rq_pos(rq1);
 	s2 = blk_rq_pos(rq2);
@@ -990,9 +988,10 @@ static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
 
 	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
 					st->min_vdisktime);
-	cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u disp=%u charge=%u iops=%u"
-			" sect=%u", used_sl, cfqq->slice_dispatch, charge,
-			iops_mode(cfqd), cfqq->nr_sectors);
+	cfq_log_cfqq(cfqq->cfqd, cfqq,
+		     "sl_used=%u disp=%u charge=%u iops=%u sect=%lu",
+		     used_sl, cfqq->slice_dispatch, charge,
+		     iops_mode(cfqd), cfqq->nr_sectors);
 	cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl,
 					  unaccounted_sl);
 	cfq_blkiocg_set_start_empty_time(&cfqg->blkg);
@@ -1014,28 +1013,47 @@ void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg,
 	cfqg->needs_update = true;
 }
 
-static struct cfq_group * cfq_find_alloc_cfqg(struct cfq_data *cfqd,
-		struct blkio_cgroup *blkcg, int create)
+static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd,
+			struct cfq_group *cfqg, struct blkio_cgroup *blkcg)
 {
-	struct cfq_group *cfqg = NULL;
-	void *key = cfqd;
-	int i, j;
-	struct cfq_rb_root *st;
 	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
 	unsigned int major, minor;
 
-	cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
-	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
+	/*
+	 * Add group onto cgroup list. It might happen that bdi->dev is
+	 * not initialized yet. Initialize this new group without major
+	 * and minor info and this info will be filled in once a new thread
+	 * comes for IO.
+	 */
+	if (bdi->dev) {
 		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
-		cfqg->blkg.dev = MKDEV(major, minor);
-		goto done;
-	}
-	if (cfqg || !create)
-		goto done;
+		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
+					(void *)cfqd, MKDEV(major, minor));
+	} else
+		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 */
+	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
+}
+
+/*
+ * Should be called from sleepable context. No request queue lock as per
+ * cpu stats are allocated dynamically and alloc_percpu needs to be called
+ * from sleepable context.
+ */
+static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd)
+{
+	struct cfq_group *cfqg = NULL;
+	int i, j, ret;
+	struct cfq_rb_root *st;
 
 	cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node);
 	if (!cfqg)
-		goto done;
+		return NULL;
 
 	for_each_cfqg_st(cfqg, i, j, st)
 		*st = CFQ_RB_ROOT;
@@ -1049,43 +1067,94 @@ static struct cfq_group * cfq_find_alloc_cfqg(struct cfq_data *cfqd,
 	 */
 	cfqg->ref = 1;
 
+	ret = blkio_alloc_blkg_stats(&cfqg->blkg);
+	if (ret) {
+		kfree(cfqg);
+		return NULL;
+	}
+
+	return cfqg;
+}
+
+static struct cfq_group *
+cfq_find_cfqg(struct cfq_data *cfqd, struct blkio_cgroup *blkcg)
+{
+	struct cfq_group *cfqg = NULL;
+	void *key = cfqd;
+	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
+	unsigned int major, minor;
+
 	/*
-	 * Add group onto cgroup list. It might happen that bdi->dev is
-	 * not initialized yet. Initialize this new group without major
-	 * and minor info and this info will be filled in once a new thread
-	 * comes for IO. See code above.
+	 * This is the common case when there are no blkio cgroups.
+	 * Avoid lookup in this case
 	 */
-	if (bdi->dev) {
-		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
-		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
-					MKDEV(major, minor));
-	} else
-		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
-					0);
-
-	cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
+	if (blkcg == &blkio_root_cgroup)
+		cfqg = &cfqd->root_group;
+	else
+		cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
 
-	/* Add group on cfqd list */
-	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
+	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
+		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
+		cfqg->blkg.dev = MKDEV(major, minor);
+	}
 
-done:
 	return cfqg;
 }
 
 /*
- * Search for the cfq group current task belongs to. If create = 1, then also
- * create the cfq group if it does not exist. request_queue lock must be held.
+ * Search for the cfq group current task belongs to. request_queue lock must
+ * be held.
  */
-static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
+static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd)
 {
 	struct blkio_cgroup *blkcg;
-	struct cfq_group *cfqg = NULL;
+	struct cfq_group *cfqg = NULL, *__cfqg = NULL;
+	struct request_queue *q = cfqd->queue;
+
+	rcu_read_lock();
+	blkcg = task_blkio_cgroup(current);
+	cfqg = cfq_find_cfqg(cfqd, blkcg);
+	if (cfqg) {
+		rcu_read_unlock();
+		return cfqg;
+	}
+
+	/*
+	 * Need to allocate a group. Allocation of group also needs allocation
+	 * of per cpu stats which in-turn takes a mutex() and can block. Hence
+	 * we need to drop rcu lock and queue_lock before we call alloc.
+	 *
+	 * Not taking any queue reference here and assuming that queue is
+	 * around by the time we return. CFQ queue allocation code does
+	 * the same. It might be racy though.
+	 */
+
+	rcu_read_unlock();
+	spin_unlock_irq(q->queue_lock);
+
+	cfqg = cfq_alloc_cfqg(cfqd);
+
+	spin_lock_irq(q->queue_lock);
 
 	rcu_read_lock();
 	blkcg = task_blkio_cgroup(current);
-	cfqg = cfq_find_alloc_cfqg(cfqd, blkcg, create);
-	if (!cfqg && create)
+
+	/*
+	 * If some other thread already allocated the group while we were
+	 * not holding queue lock, free up the group
+	 */
+	__cfqg = cfq_find_cfqg(cfqd, blkcg);
+
+	if (__cfqg) {
+		kfree(cfqg);
+		rcu_read_unlock();
+		return __cfqg;
+	}
+
+	if (!cfqg)
 		cfqg = &cfqd->root_group;
+
+	cfq_init_add_cfqg_lists(cfqd, cfqg, blkcg);
 	rcu_read_unlock();
 	return cfqg;
 }
@@ -1118,6 +1187,7 @@ static void cfq_put_cfqg(struct cfq_group *cfqg)
 		return;
 	for_each_cfqg_st(cfqg, i, j, st)
 		BUG_ON(!RB_EMPTY_ROOT(&st->rb));
+	free_percpu(cfqg->blkg.stats_cpu);
 	kfree(cfqg);
 }
 
@@ -1176,7 +1246,7 @@ void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg)
 }
 
 #else /* GROUP_IOSCHED */
-static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
+static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd)
 {
 	return &cfqd->root_group;
 }
@@ -1210,7 +1280,6 @@ static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 	struct cfq_rb_root *service_tree;
 	int left;
 	int new_cfqq = 1;
-	int group_changed = 0;
 
 	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
 						cfqq_type(cfqq));
@@ -1281,7 +1350,7 @@ static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 	rb_link_node(&cfqq->rb_node, parent, p);
 	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
 	service_tree->count++;
-	if ((add_front || !new_cfqq) && !group_changed)
+	if (add_front || !new_cfqq)
 		return;
 	cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
 }
@@ -1955,8 +2024,8 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 	 */
 	if (sample_valid(cic->ttime_samples) &&
 	    (cfqq->slice_end - jiffies < cic->ttime_mean)) {
-		cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%d",
-				cic->ttime_mean);
+		cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu",
+			     cic->ttime_mean);
 		return;
 	}
 
@@ -2029,7 +2098,7 @@ cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 
 	WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
 
-	return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
+	return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
 }
 
 /*
@@ -2704,8 +2773,11 @@ static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
 	smp_wmb();
 	cic->key = cfqd_dead_key(cfqd);
 
-	if (ioc->ioc_data == cic)
+	if (rcu_dereference(ioc->ioc_data) == cic) {
+		spin_lock(&ioc->lock);
 		rcu_assign_pointer(ioc->ioc_data, NULL);
+		spin_unlock(&ioc->lock);
+	}
 
 	if (cic->cfqq[BLK_RW_ASYNC]) {
 		cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]);
@@ -2911,7 +2983,7 @@ cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync,
 	struct cfq_group *cfqg;
 
 retry:
-	cfqg = cfq_get_cfqg(cfqd, 1);
+	cfqg = cfq_get_cfqg(cfqd);
 	cic = cfq_cic_lookup(cfqd, ioc);
 	/* cic always exists here */
 	cfqq = cic_to_cfqq(cic, is_sync);
@@ -3718,9 +3790,6 @@ new_queue:
 	return 0;
 
 queue_fail:
-	if (cic)
-		put_io_context(cic->ioc);
-
 	cfq_schedule_dispatch(cfqd);
 	spin_unlock_irqrestore(q->queue_lock, flags);
 	cfq_log(cfqd, "set_request fail");
@@ -3815,15 +3884,11 @@ 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 +3907,13 @@ 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 +3923,25 @@ 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();
+
+#ifdef CONFIG_CFQ_GROUP_IOSCHED
+	/* Free up per cpu stats for root group */
+	free_percpu(cfqd->root_group.blkg.stats_cpu);
+#endif
+	kfree(cfqd);
 }
 
 static int cfq_alloc_cic_index(void)
@@ -3886,8 +3974,12 @@ static void *cfq_init_queue(struct request_queue *q)
 		return NULL;
 
 	cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
-	if (!cfqd)
+	if (!cfqd) {
+		spin_lock(&cic_index_lock);
+		ida_remove(&cic_index_ida, i);
+		spin_unlock(&cic_index_lock);
 		return NULL;
+	}
 
 	/*
 	 * Don't need take queue_lock in the routine, since we are
@@ -3909,14 +4001,29 @@ 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;
+
+	if (blkio_alloc_blkg_stats(&cfqg->blkg)) {
+		kfree(cfqg);
+		kfree(cfqd);
+		return NULL;
+	}
+
 	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