blk-mq: split out tag initialization, support shared tags

Add a new blk_mq_tag_set structure that gets set up before we initialize
the queue.  A single blk_mq_tag_set structure can be shared by multiple
queues.
Signed-off-by: Christoph Hellwig <hch@lst.de>

Modular export of blk_mq_{alloc,free}_tagset added by me.
Signed-off-by: Jens Axboe <axboe@fb.com>

block/blk-mq-cpumap.c

@@ -80,17 +80,17 @@ int blk_mq_update_queue_map(unsigned int *map, unsigned int nr_queues)
 	return 0;
 }
 
-unsigned int *blk_mq_make_queue_map(struct blk_mq_reg *reg)
+unsigned int *blk_mq_make_queue_map(struct blk_mq_tag_set *set)
 {
 	unsigned int *map;
 
 	/* If cpus are offline, map them to first hctx */
 	map = kzalloc_node(sizeof(*map) * num_possible_cpus(), GFP_KERNEL,
-				reg->numa_node);
+				set->numa_node);
 	if (!map)
 		return NULL;
 
-	if (!blk_mq_update_queue_map(map, reg->nr_hw_queues))
+	if (!blk_mq_update_queue_map(map, set->nr_hw_queues))
 		return map;
 
 	kfree(map);

block/blk-mq-tag.c

 #include <linux/kernel.h>
 #include <linux/module.h>
-#include <linux/percpu_ida.h>
 
 #include <linux/blk-mq.h>
 #include "blk.h"
 #include "blk-mq.h"
 #include "blk-mq-tag.h"
 
-/*
- * Per tagged queue (tag address space) map
- */
-struct blk_mq_tags {
-	unsigned int nr_tags;
-	unsigned int nr_reserved_tags;
-	unsigned int nr_batch_move;
-	unsigned int nr_max_cache;
-
-	struct percpu_ida free_tags;
-	struct percpu_ida reserved_tags;
-};
-
 void blk_mq_wait_for_tags(struct blk_mq_tags *tags)
 {
 	int tag = blk_mq_get_tag(tags, __GFP_WAIT, false);

block/blk-mq-tag.h

 #ifndef INT_BLK_MQ_TAG_H
 #define INT_BLK_MQ_TAG_H
 
-struct blk_mq_tags;
+#include <linux/percpu_ida.h>
+
+/*
+ * Tag address space map.
+ */
+struct blk_mq_tags {
+	unsigned int nr_tags;
+	unsigned int nr_reserved_tags;
+	unsigned int nr_batch_move;
+	unsigned int nr_max_cache;
+
+	struct percpu_ida free_tags;
+	struct percpu_ida reserved_tags;
+
+	struct request **rqs;
+	struct list_head page_list;
+};
+
 
 extern struct blk_mq_tags *blk_mq_init_tags(unsigned int nr_tags, unsigned int reserved_tags, int node);
 extern void blk_mq_free_tags(struct blk_mq_tags *tags);

block/blk-mq.h

 #ifndef INT_BLK_MQ_H
 #define INT_BLK_MQ_H
 
+struct blk_mq_tag_set;
+
 struct blk_mq_ctx {
 	struct {
 		spinlock_t		lock;
@@ -46,8 +48,7 @@ void blk_mq_disable_hotplug(void);
 /*
  * CPU -> queue mappings
  */
-struct blk_mq_reg;
-extern unsigned int *blk_mq_make_queue_map(struct blk_mq_reg *reg);
+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);
 
 void blk_mq_add_timer(struct request *rq);

drivers/block/null_blk.c

@@ -32,6 +32,7 @@ struct nullb {
 	unsigned int index;
 	struct request_queue *q;
 	struct gendisk *disk;
+	struct blk_mq_tag_set tag_set;
 	struct hrtimer timer;
 	unsigned int queue_depth;
 	spinlock_t lock;
@@ -320,10 +321,11 @@ static int null_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
 	return BLK_MQ_RQ_QUEUE_OK;
 }
 
-static struct blk_mq_hw_ctx *null_alloc_hctx(struct blk_mq_reg *reg, unsigned int hctx_index)
+static struct blk_mq_hw_ctx *null_alloc_hctx(struct blk_mq_tag_set *set,
+		unsigned int hctx_index)
 {
-	int b_size = DIV_ROUND_UP(reg->nr_hw_queues, nr_online_nodes);
-	int tip = (reg->nr_hw_queues % nr_online_nodes);
+	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;
 
 	/*
@@ -338,7 +340,7 @@ static struct blk_mq_hw_ctx *null_alloc_hctx(struct blk_mq_reg *reg, unsigned in
 
 			tip--;
 			if (!tip)
-				b_size = reg->nr_hw_queues / nr_online_nodes;
+				b_size = set->nr_hw_queues / nr_online_nodes;
 		}
 	}
 
@@ -387,13 +389,17 @@ static struct blk_mq_ops null_mq_ops = {
 	.map_queue      = blk_mq_map_queue,
 	.init_hctx	= null_init_hctx,
 	.complete	= null_softirq_done_fn,
+	.alloc_hctx	= blk_mq_alloc_single_hw_queue,
+	.free_hctx	= blk_mq_free_single_hw_queue,
 };
 
-static struct blk_mq_reg null_mq_reg = {
-	.ops		= &null_mq_ops,
-	.queue_depth	= 64,
-	.cmd_size	= sizeof(struct nullb_cmd),
-	.flags		= BLK_MQ_F_SHOULD_MERGE,
+static struct blk_mq_ops null_mq_ops_pernode = {
+	.queue_rq       = null_queue_rq,
+	.map_queue      = blk_mq_map_queue,
+	.init_hctx	= null_init_hctx,
+	.complete	= null_softirq_done_fn,
+	.alloc_hctx	= null_alloc_hctx,
+	.free_hctx	= null_free_hctx,
 };
 
 static void null_del_dev(struct nullb *nullb)
@@ -402,6 +408,8 @@ static void null_del_dev(struct nullb *nullb)
 
 	del_gendisk(nullb->disk);
 	blk_cleanup_queue(nullb->q);
+	if (queue_mode == NULL_Q_MQ)
+		blk_mq_free_tag_set(&nullb->tag_set);
 	put_disk(nullb->disk);
 	kfree(nullb);
 }
@@ -506,7 +514,7 @@ static int null_add_dev(void)
 
 	nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
 	if (!nullb)
-		return -ENOMEM;
+		goto out;
 
 	spin_lock_init(&nullb->lock);
 
@@ -514,49 +522,47 @@ static int null_add_dev(void)
 		submit_queues = nr_online_nodes;
 
 	if (setup_queues(nullb))
-		goto err;
+		goto out_free_nullb;
 
 	if (queue_mode == NULL_Q_MQ) {
-		null_mq_reg.numa_node = home_node;
-		null_mq_reg.queue_depth = hw_queue_depth;
-		null_mq_reg.nr_hw_queues = submit_queues;
-
-		if (use_per_node_hctx) {
-			null_mq_reg.ops->alloc_hctx = null_alloc_hctx;
-			null_mq_reg.ops->free_hctx = null_free_hctx;
-		} else {
-			null_mq_reg.ops->alloc_hctx = blk_mq_alloc_single_hw_queue;
-			null_mq_reg.ops->free_hctx = blk_mq_free_single_hw_queue;
-		}
-
-		nullb->q = blk_mq_init_queue(&null_mq_reg, nullb);
+		if (use_per_node_hctx)
+			nullb->tag_set.ops = &null_mq_ops_pernode;
+		else
+			nullb->tag_set.ops = &null_mq_ops;
+		nullb->tag_set.nr_hw_queues = submit_queues;
+		nullb->tag_set.queue_depth = hw_queue_depth;
+		nullb->tag_set.numa_node = home_node;
+		nullb->tag_set.cmd_size	= sizeof(struct nullb_cmd);
+		nullb->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
+		nullb->tag_set.driver_data = nullb;
+
+		if (blk_mq_alloc_tag_set(&nullb->tag_set))
+			goto out_cleanup_queues;
+
+		nullb->q = blk_mq_init_queue(&nullb->tag_set);
+		if (!nullb->q)
+			goto out_cleanup_tags;
 	} else if (queue_mode == NULL_Q_BIO) {
 		nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
+		if (!nullb->q)
+			goto out_cleanup_queues;
 		blk_queue_make_request(nullb->q, null_queue_bio);
 		init_driver_queues(nullb);
 	} else {
 		nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
+		if (!nullb->q)
+			goto out_cleanup_queues;
 		blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
-		if (nullb->q)
-			blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
+		blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
 		init_driver_queues(nullb);
 	}
 
-	if (!nullb->q)
-		goto queue_fail;
-
 	nullb->q->queuedata = nullb;
 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
 
 	disk = nullb->disk = alloc_disk_node(1, home_node);
-	if (!disk) {
-queue_fail:
-		blk_cleanup_queue(nullb->q);
-		cleanup_queues(nullb);
-err:
-		kfree(nullb);
-		return -ENOMEM;
-	}
+	if (!disk)
+		goto out_cleanup_blk_queue;
 
 	mutex_lock(&lock);
 	list_add_tail(&nullb->list, &nullb_list);
@@ -579,6 +585,18 @@ static int null_add_dev(void)
 	sprintf(disk->disk_name, "nullb%d", nullb->index);
 	add_disk(disk);
 	return 0;
+
+out_cleanup_blk_queue:
+	blk_cleanup_queue(nullb->q);
+out_cleanup_tags:
+	if (queue_mode == NULL_Q_MQ)
+		blk_mq_free_tag_set(&nullb->tag_set);
+out_cleanup_queues:
+	cleanup_queues(nullb);
+out_free_nullb:
+	kfree(nullb);
+out:
+	return -ENOMEM;
 }
 
 static int __init null_init(void)

drivers/block/virtio_blk.c

@@ -30,6 +30,9 @@ struct virtio_blk
 	/* The disk structure for the kernel. */
 	struct gendisk *disk;
 
+	/* Block layer tags. */
+	struct blk_mq_tag_set tag_set;
+
 	/* Process context for config space updates */
 	struct work_struct config_work;
 
@@ -480,8 +483,9 @@ static const struct device_attribute dev_attr_cache_type_rw =
 	__ATTR(cache_type, S_IRUGO|S_IWUSR,
 	       virtblk_cache_type_show, virtblk_cache_type_store);
 
-static int virtblk_init_request(void *data, struct blk_mq_hw_ctx *hctx,
-		struct request *rq, unsigned int nr)
+static int virtblk_init_request(void *data, struct request *rq,
+		unsigned int hctx_idx, unsigned int request_idx,
+		unsigned int numa_node)
 {
 	struct virtio_blk *vblk = data;
 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
@@ -495,18 +499,12 @@ static struct blk_mq_ops virtio_mq_ops = {
 	.map_queue	= blk_mq_map_queue,
 	.alloc_hctx	= blk_mq_alloc_single_hw_queue,
 	.free_hctx	= blk_mq_free_single_hw_queue,
-	.init_request	= virtblk_init_request,
 	.complete	= virtblk_request_done,
+	.init_request	= virtblk_init_request,
 };
 
-static struct blk_mq_reg virtio_mq_reg = {
-	.ops		= &virtio_mq_ops,
-	.nr_hw_queues	= 1,
-	.queue_depth	= 0, /* Set in virtblk_probe */
-	.numa_node	= NUMA_NO_NODE,
-	.flags		= BLK_MQ_F_SHOULD_MERGE,
-};
-module_param_named(queue_depth, virtio_mq_reg.queue_depth, uint, 0444);
+static unsigned int virtblk_queue_depth;
+module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 
 static int virtblk_probe(struct virtio_device *vdev)
 {
@@ -562,20 +560,32 @@ static int virtblk_probe(struct virtio_device *vdev)
 	}
 
 	/* Default queue sizing is to fill the ring. */
-	if (!virtio_mq_reg.queue_depth) {
-		virtio_mq_reg.queue_depth = vblk->vq->num_free;
+	if (!virtblk_queue_depth) {
+		virtblk_queue_depth = vblk->vq->num_free;
 		/* ... but without indirect descs, we use 2 descs per req */
 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
-			virtio_mq_reg.queue_depth /= 2;
+			virtblk_queue_depth /= 2;
 	}
-	virtio_mq_reg.cmd_size =
+
+	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
+	vblk->tag_set.ops = &virtio_mq_ops;
+	vblk->tag_set.nr_hw_queues = 1;
+	vblk->tag_set.queue_depth = virtblk_queue_depth;
+	vblk->tag_set.numa_node = NUMA_NO_NODE;
+	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
+	vblk->tag_set.cmd_size =
 		sizeof(struct virtblk_req) +
 		sizeof(struct scatterlist) * sg_elems;
+	vblk->tag_set.driver_data = vblk;
 
-	q = vblk->disk->queue = blk_mq_init_queue(&virtio_mq_reg, vblk);
+	err = blk_mq_alloc_tag_set(&vblk->tag_set);
+	if (err)
+		goto out_put_disk;
+
+	q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
 	if (!q) {
 		err = -ENOMEM;
-		goto out_put_disk;
+		goto out_free_tags;
 	}
 
 	q->queuedata = vblk;
@@ -678,6 +688,8 @@ static int virtblk_probe(struct virtio_device *vdev)
 out_del_disk:
 	del_gendisk(vblk->disk);
 	blk_cleanup_queue(vblk->disk->queue);
+out_free_tags:
+	blk_mq_free_tag_set(&vblk->tag_set);
 out_put_disk:
 	put_disk(vblk->disk);
 out_free_vq:
@@ -704,6 +716,8 @@ static void virtblk_remove(struct virtio_device *vdev)
 	del_gendisk(vblk->disk);
 	blk_cleanup_queue(vblk->disk->queue);
 
+	blk_mq_free_tag_set(&vblk->tag_set);
+
 	/* Stop all the virtqueues. */
 	vdev->config->reset(vdev);
 

include/linux/blk-mq.h

@@ -33,8 +33,6 @@ struct blk_mq_hw_ctx {
 	unsigned int 		nr_ctx_map;
 	unsigned long		*ctx_map;
 
-	struct request		**rqs;
-	struct list_head	page_list;
 	struct blk_mq_tags	*tags;
 
 	unsigned long		queued;
@@ -42,7 +40,6 @@ struct blk_mq_hw_ctx {
 #define BLK_MQ_MAX_DISPATCH_ORDER	10
 	unsigned long		dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
 
-	unsigned int		queue_depth;
 	unsigned int		numa_node;
 	unsigned int		cmd_size;	/* per-request extra data */
 
@@ -50,7 +47,7 @@ struct blk_mq_hw_ctx {
 	struct kobject		kobj;
 };
 
-struct blk_mq_reg {
+struct blk_mq_tag_set {
 	struct blk_mq_ops	*ops;
 	unsigned int		nr_hw_queues;
 	unsigned int		queue_depth;
@@ -59,18 +56,22 @@ struct blk_mq_reg {
 	int			numa_node;
 	unsigned int		timeout;
 	unsigned int		flags;		/* BLK_MQ_F_* */
+	void			*driver_data;
+
+	struct blk_mq_tags	**tags;
 };
 
 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_reg *,unsigned int);
+typedef struct blk_mq_hw_ctx *(alloc_hctx_fn)(struct blk_mq_tag_set *,
+		unsigned 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);
-typedef int (init_request_fn)(void *, struct blk_mq_hw_ctx *,
-		struct request *, unsigned int);
-typedef void (exit_request_fn)(void *, struct blk_mq_hw_ctx *,
-		struct request *, unsigned int);
+typedef int (init_request_fn)(void *, struct request *, unsigned int,
+		unsigned int, unsigned int);
+typedef void (exit_request_fn)(void *, struct request *, unsigned int,
+		unsigned int);
 
 struct blk_mq_ops {
 	/*
@@ -127,10 +128,13 @@ enum {
 	BLK_MQ_MAX_DEPTH	= 2048,
 };
 
-struct request_queue *blk_mq_init_queue(struct blk_mq_reg *, void *);
+struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
 int blk_mq_register_disk(struct gendisk *);
 void blk_mq_unregister_disk(struct gendisk *);
 
+int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
+void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
+
 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
 
 void blk_mq_insert_request(struct request *, bool, bool, bool);
@@ -139,10 +143,10 @@ void blk_mq_free_request(struct request *rq);
 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
 struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp);
 struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw, gfp_t gfp);
-struct request *blk_mq_rq_from_tag(struct request_queue *q, unsigned int tag);
+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_reg *, unsigned int);
+struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int);
 void blk_mq_free_single_hw_queue(struct blk_mq_hw_ctx *, unsigned int);
 
 bool blk_mq_end_io_partial(struct request *rq, int error,
@@ -173,12 +177,6 @@ static inline void *blk_mq_rq_to_pdu(struct request *rq)
 	return (void *) rq + sizeof(*rq);
 }
 
-static inline struct request *blk_mq_tag_to_rq(struct blk_mq_hw_ctx *hctx,
-					       unsigned int tag)
-{
-	return hctx->rqs[tag];
-}
-
 #define queue_for_each_hw_ctx(q, hctx, i)				\
 	for ((i) = 0; (i) < (q)->nr_hw_queues &&			\
 	     ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)