blk-mq: move cpuhp callback registering out of q->sysfs_lock

Registering and unregistering cpuhp callback requires global cpu hotplug lock,
which is used everywhere. Meantime q->sysfs_lock is used in block layer
almost everywhere.

It is easy to trigger lockdep warning[1] by connecting the two locks.

Fix the warning by moving blk-mq's cpuhp callback registering out of
q->sysfs_lock. Add one dedicated global lock for covering registering &
unregistering hctx's cpuhp, and it is safe to do so because hctx is
guaranteed to be live if our request_queue is live.

[1] https://lore.kernel.org/lkml/Z04pz3AlvI4o0Mr8@agluck-desk3/

Cc: Reinette Chatre <[e-mail]>
Cc: Fenghua Yu <[e-mail]>
Cc: Peter Newman <[e-mail]>
Cc: Babu Moger <[e-mail]>
Reported-by: Luck Tony <[e-mail]>
Signed-off-by: Ming Lei <[e-mail]>
Tested-by: Tony Luck <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Jens Axboe <[e-mail]>

block/blk-mq.c

···
 
 static DEFINE_PER_CPU(struct llist_head, blk_cpu_done);
 static DEFINE_PER_CPU(call_single_data_t, blk_cpu_csd);
+static DEFINE_MUTEX(blk_mq_cpuhp_lock);
 
 static void blk_mq_insert_request(struct request *rq, blk_insert_t flags);
 static void blk_mq_request_bypass_insert(struct request *rq,
···
 	return 0;
 }
 
-static void blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx)
+static void __blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx)
 {
-	if (!(hctx->flags & BLK_MQ_F_STACKING))
+	lockdep_assert_held(&blk_mq_cpuhp_lock);
+
+	if (!(hctx->flags & BLK_MQ_F_STACKING) &&
+	    !hlist_unhashed(&hctx->cpuhp_online)) {
 		cpuhp_state_remove_instance_nocalls(CPUHP_AP_BLK_MQ_ONLINE,
 						    &hctx->cpuhp_online);
-	cpuhp_state_remove_instance_nocalls(CPUHP_BLK_MQ_DEAD,
-					    &hctx->cpuhp_dead);
+		INIT_HLIST_NODE(&hctx->cpuhp_online);
+	}
+
+	if (!hlist_unhashed(&hctx->cpuhp_dead)) {
+		cpuhp_state_remove_instance_nocalls(CPUHP_BLK_MQ_DEAD,
+						    &hctx->cpuhp_dead);
+		INIT_HLIST_NODE(&hctx->cpuhp_dead);
+	}
+}
+
+static void blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx)
+{
+	mutex_lock(&blk_mq_cpuhp_lock);
+	__blk_mq_remove_cpuhp(hctx);
+	mutex_unlock(&blk_mq_cpuhp_lock);
+}
+
+static void __blk_mq_add_cpuhp(struct blk_mq_hw_ctx *hctx)
+{
+	lockdep_assert_held(&blk_mq_cpuhp_lock);
+
+	if (!(hctx->flags & BLK_MQ_F_STACKING) &&
+	    hlist_unhashed(&hctx->cpuhp_online))
+		cpuhp_state_add_instance_nocalls(CPUHP_AP_BLK_MQ_ONLINE,
+				&hctx->cpuhp_online);
+
+	if (hlist_unhashed(&hctx->cpuhp_dead))
+		cpuhp_state_add_instance_nocalls(CPUHP_BLK_MQ_DEAD,
+				&hctx->cpuhp_dead);
+}
+
+static void __blk_mq_remove_cpuhp_list(struct list_head *head)
+{
+	struct blk_mq_hw_ctx *hctx;
+
+	lockdep_assert_held(&blk_mq_cpuhp_lock);
+
+	list_for_each_entry(hctx, head, hctx_list)
+		__blk_mq_remove_cpuhp(hctx);
+}
+
+/*
+ * Unregister cpuhp callbacks from exited hw queues
+ *
+ * Safe to call if this `request_queue` is live
+ */
+static void blk_mq_remove_hw_queues_cpuhp(struct request_queue *q)
+{
+	LIST_HEAD(hctx_list);
+
+	spin_lock(&q->unused_hctx_lock);
+	list_splice_init(&q->unused_hctx_list, &hctx_list);
+	spin_unlock(&q->unused_hctx_lock);
+
+	mutex_lock(&blk_mq_cpuhp_lock);
+	__blk_mq_remove_cpuhp_list(&hctx_list);
+	mutex_unlock(&blk_mq_cpuhp_lock);
+
+	spin_lock(&q->unused_hctx_lock);
+	list_splice(&hctx_list, &q->unused_hctx_list);
+	spin_unlock(&q->unused_hctx_lock);
+}
+
+/*
+ * Register cpuhp callbacks from all hw queues
+ *
+ * Safe to call if this `request_queue` is live
+ */
+static void blk_mq_add_hw_queues_cpuhp(struct request_queue *q)
+{
+	struct blk_mq_hw_ctx *hctx;
+	unsigned long i;
+
+	mutex_lock(&blk_mq_cpuhp_lock);
+	queue_for_each_hw_ctx(q, hctx, i)
+		__blk_mq_add_cpuhp(hctx);
+	mutex_unlock(&blk_mq_cpuhp_lock);
 }
 
 /*
···
 	if (set->ops->exit_hctx)
 		set->ops->exit_hctx(hctx, hctx_idx);
 
-	blk_mq_remove_cpuhp(hctx);
-
 	xa_erase(&q->hctx_table, hctx_idx);
 
 	spin_lock(&q->unused_hctx_lock);
···
 	queue_for_each_hw_ctx(q, hctx, i) {
 		if (i == nr_queue)
 			break;
+		blk_mq_remove_cpuhp(hctx);
 		blk_mq_exit_hctx(q, set, hctx, i);
 	}
 }
···
 
 	if (xa_insert(&q->hctx_table, hctx_idx, hctx, GFP_KERNEL))
 		goto exit_flush_rq;
-
-	if (!(hctx->flags & BLK_MQ_F_STACKING))
-		cpuhp_state_add_instance_nocalls(CPUHP_AP_BLK_MQ_ONLINE,
-				&hctx->cpuhp_online);
-	cpuhp_state_add_instance_nocalls(CPUHP_BLK_MQ_DEAD, &hctx->cpuhp_dead);
 
 	return 0;
 
···
 	INIT_DELAYED_WORK(&hctx->run_work, blk_mq_run_work_fn);
 	spin_lock_init(&hctx->lock);
 	INIT_LIST_HEAD(&hctx->dispatch);
+	INIT_HLIST_NODE(&hctx->cpuhp_dead);
+	INIT_HLIST_NODE(&hctx->cpuhp_online);
 	hctx->queue = q;
 	hctx->flags = set->flags & ~BLK_MQ_F_TAG_QUEUE_SHARED;
 
···
 	xa_for_each_start(&q->hctx_table, j, hctx, j)
 		blk_mq_exit_hctx(q, set, hctx, j);
 	mutex_unlock(&q->sysfs_lock);
+
+	/* unregister cpuhp callbacks for exited hctxs */
+	blk_mq_remove_hw_queues_cpuhp(q);
+
+	/* register cpuhp for new initialized hctxs */
+	blk_mq_add_hw_queues_cpuhp(q);
 }
 
 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,