blk-mq: pass in suggested NUMA node to ->alloc_hctx()

Drivers currently have to figure this out on their own, and they
are missing information to do it properly. The ones that did
attempt to do it, do it wrong.

So just pass in the suggested node directly to the alloc
function.

Signed-off-by: Jens Axboe <axboe@fb.com>

block/blk-mq-cpumap.c

@@ -96,3 +96,19 @@ unsigned int *blk_mq_make_queue_map(struct blk_mq_tag_set *set)
 	kfree(map);
 	return NULL;
 }
+
+/*
+ * We have no quick way of doing reverse lookups. This is only used at
+ * queue init time, so runtime isn't important.
+ */
+int blk_mq_hw_queue_to_node(unsigned int *mq_map, unsigned int index)
+{
+	int i;
+
+	for_each_possible_cpu(i) {
+		if (index == mq_map[i])
+			return cpu_to_node(i);
+	}
+
+	return NUMA_NO_NODE;
+}

block/blk-mq.c

@@ -1297,10 +1297,10 @@ struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q, const int cpu)
 EXPORT_SYMBOL(blk_mq_map_queue);
 
 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *set,
-						   unsigned int hctx_index)
+						   unsigned int hctx_index,
+						   int node)
 {
-	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL,
-				set->numa_node);
+	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL, node);
 }
 EXPORT_SYMBOL(blk_mq_alloc_single_hw_queue);
 
@@ -1752,6 +1752,7 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	struct blk_mq_hw_ctx **hctxs;
 	struct blk_mq_ctx *ctx;
 	struct request_queue *q;
+	unsigned int *map;
 	int i;
 
 	ctx = alloc_percpu(struct blk_mq_ctx);
@@ -1764,8 +1765,14 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	if (!hctxs)
 		goto err_percpu;
 
+	map = blk_mq_make_queue_map(set);
+	if (!map)
+		goto err_map;
+
 	for (i = 0; i < set->nr_hw_queues; i++) {
-		hctxs[i] = set->ops->alloc_hctx(set, i);
+		int node = blk_mq_hw_queue_to_node(map, i);
+
+		hctxs[i] = set->ops->alloc_hctx(set, i, node);
 		if (!hctxs[i])
 			goto err_hctxs;
 
@@ -1773,7 +1780,7 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 			goto err_hctxs;
 
 		atomic_set(&hctxs[i]->nr_active, 0);
-		hctxs[i]->numa_node = NUMA_NO_NODE;
+		hctxs[i]->numa_node = node;
 		hctxs[i]->queue_num = i;
 	}
 
@@ -1784,15 +1791,12 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	if (percpu_counter_init(&q->mq_usage_counter, 0))
 		goto err_map;
 
-	q->mq_map = blk_mq_make_queue_map(set);
-	if (!q->mq_map)
-		goto err_map;
-
 	setup_timer(&q->timeout, blk_mq_rq_timer, (unsigned long) q);
 	blk_queue_rq_timeout(q, 30000);
 
 	q->nr_queues = nr_cpu_ids;
 	q->nr_hw_queues = set->nr_hw_queues;
+	q->mq_map = map;
 
 	q->queue_ctx = ctx;
 	q->queue_hw_ctx = hctxs;
@@ -1844,16 +1848,16 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 err_flush_rq:
 	kfree(q->flush_rq);
 err_hw:
-	kfree(q->mq_map);
-err_map:
 	blk_cleanup_queue(q);
 err_hctxs:
+	kfree(map);
 	for (i = 0; i < set->nr_hw_queues; i++) {
 		if (!hctxs[i])
 			break;
 		free_cpumask_var(hctxs[i]->cpumask);
 		set->ops->free_hctx(hctxs[i], i);
 	}
+err_map:
 	kfree(hctxs);
 err_percpu:
 	free_percpu(ctx);

block/blk-mq.h

@@ -52,6 +52,7 @@ void blk_mq_disable_hotplug(void);
  */
 extern unsigned int *blk_mq_make_queue_map(struct blk_mq_tag_set *set);
 extern int blk_mq_update_queue_map(unsigned int *map, unsigned int nr_queues);
+extern int blk_mq_hw_queue_to_node(unsigned int *map, unsigned int);
 
 /*
  * Basic implementation of sparser bitmap, allowing the user to spread

drivers/block/null_blk.c

@@ -322,39 +322,10 @@ static int null_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
 }
 
 static struct blk_mq_hw_ctx *null_alloc_hctx(struct blk_mq_tag_set *set,
-		unsigned int hctx_index)
+					     unsigned int hctx_index,
+					     int node)
 {
-	int b_size = DIV_ROUND_UP(set->nr_hw_queues, nr_online_nodes);
-	int tip = (set->nr_hw_queues % nr_online_nodes);
-	int node = 0, i, n;
-
-	/*
-	 * Split submit queues evenly wrt to the number of nodes. If uneven,
-	 * fill the first buckets with one extra, until the rest is filled with
-	 * no extra.
-	 */
-	for (i = 0, n = 1; i < hctx_index; i++, n++) {
-		if (n % b_size == 0) {
-			n = 0;
-			node++;
-
-			tip--;
-			if (!tip)
-				b_size = set->nr_hw_queues / nr_online_nodes;
-		}
-	}
-
-	/*
-	 * A node might not be online, therefore map the relative node id to the
-	 * real node id.
-	 */
-	for_each_online_node(n) {
-		if (!node)
-			break;
-		node--;
-	}
-
-	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL, n);
+	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL, node);
 }
 
 static void null_free_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_index)

include/linux/blk-mq.h

@@ -80,7 +80,7 @@ struct blk_mq_tag_set {
 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, struct request *);
 typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
 typedef struct blk_mq_hw_ctx *(alloc_hctx_fn)(struct blk_mq_tag_set *,
-		unsigned int);
+		unsigned int, int);
 typedef void (free_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
@@ -165,7 +165,7 @@ struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw, g
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 
 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
-struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int);
+struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
 void blk_mq_free_single_hw_queue(struct blk_mq_hw_ctx *, unsigned int);
 
 void blk_mq_end_io(struct request *rq, int error);