linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps
@ 2018-10-30 18:32 Jens Axboe
  2018-10-30 18:32 ` [PATCH 01/16] blk-mq: kill q->mq_map Jens Axboe
                   ` (16 more replies)
  0 siblings, 17 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel

This series adds support for multiple queue maps for blk-mq.
Since blk-mq was introduced, it's only support a single queue
map. This means you can have 1 set of queues, and the mapping
purely depends on what CPU an IO originated from. With this
patch set, drivers can implement mappings that depend on both
CPU and request type - and they can have multiple sets of mappings.

NVMe is used as a proof of concept. It adds support for a separate
write queue set. One way to use this would be to limit the number
of write queues to favor reads, since NVMe does round-robin service
of queues. An easy extension of this would be to add multiple
sets of queues, for prioritized IO.

NVMe also uses this feature to finally make the polling work
efficiently, without triggering interrupts. This both increases
performance (and decreases latency), at a lower system load. At
the same time it's more flexible, as you don't have to worry about
IRQ coalescing and redirection to avoid interrupts disturbing the
workload. This is how polling should have worked from day 1.

This is on top of my mq-conversions branch. It can also be bound
in my mq-maps branch.

Changes since v2:

- Address review comments (mostly code comments)
- Add rq->mq_hctx cache
- Cleanup/improve blk_mq_sched_insert_requests()
- Harden the IRQ affinity sets feature
- Fix assignment of hardware queue type if already mapped
- Improve plug sorting for multiple maps

Changes since v1:

- Ensure irq_calc_affinity_vectors() doesn't return more than 'maxvec'
- Rebase on top of current mq-conversions series


 block/blk-flush.c                     |  11 +-
 block/blk-mq-cpumap.c                 |  19 +--
 block/blk-mq-debugfs.c                |   2 +-
 block/blk-mq-pci.c                    |  10 +-
 block/blk-mq-rdma.c                   |   4 +-
 block/blk-mq-sched.c                  |  12 +-
 block/blk-mq-sched.h                  |   2 +-
 block/blk-mq-sysfs.c                  |  10 ++
 block/blk-mq-tag.c                    |  12 +-
 block/blk-mq-virtio.c                 |   8 +-
 block/blk-mq.c                        | 237 +++++++++++++++++++-----------
 block/blk-mq.h                        |  45 ++++--
 block/blk.h                           |   6 +-
 block/kyber-iosched.c                 |   6 +-
 drivers/block/virtio_blk.c            |   2 +-
 drivers/nvme/host/pci.c               | 263 ++++++++++++++++++++++++++++++----
 drivers/pci/msi.c                     |  14 ++
 drivers/scsi/qla2xxx/qla_os.c         |   5 +-
 drivers/scsi/scsi_lib.c               |   2 +-
 drivers/scsi/smartpqi/smartpqi_init.c |   3 +-
 drivers/scsi/virtio_scsi.c            |   3 +-
 fs/block_dev.c                        |   2 +
 fs/direct-io.c                        |   2 +
 fs/iomap.c                            |   9 +-
 include/linux/blk-mq-pci.h            |   4 +-
 include/linux/blk-mq-virtio.h         |   4 +-
 include/linux/blk-mq.h                |  33 ++++-
 include/linux/blk_types.h             |   4 +-
 include/linux/blkdev.h                |   3 +-
 include/linux/interrupt.h             |   4 +
 kernel/irq/affinity.c                 |  40 ++++--
 31 files changed, 582 insertions(+), 199 deletions(-)

-- 
Jens Axboe



^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 01/16] blk-mq: kill q->mq_map
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:28   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 02/16] blk-mq: abstract out queue map Jens Axboe
                   ` (15 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

It's just a pointer to set->mq_map, use that instead. Move the
assignment a bit earlier, so we always know it's valid.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c         | 13 ++++---------
 block/blk-mq.h         |  4 +++-
 include/linux/blkdev.h |  2 --
 3 files changed, 7 insertions(+), 12 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 21e4147c4810..22d5beaab5a0 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2321,7 +2321,7 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 	 * If the cpu isn't present, the cpu is mapped to first hctx.
 	 */
 	for_each_possible_cpu(i) {
-		hctx_idx = q->mq_map[i];
+		hctx_idx = set->mq_map[i];
 		/* unmapped hw queue can be remapped after CPU topo changed */
 		if (!set->tags[hctx_idx] &&
 		    !__blk_mq_alloc_rq_map(set, hctx_idx)) {
@@ -2331,7 +2331,7 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 			 * case, remap the current ctx to hctx[0] which
 			 * is guaranteed to always have tags allocated
 			 */
-			q->mq_map[i] = 0;
+			set->mq_map[i] = 0;
 		}
 
 		ctx = per_cpu_ptr(q->queue_ctx, i);
@@ -2429,8 +2429,6 @@ static void blk_mq_del_queue_tag_set(struct request_queue *q)
 static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set,
 				     struct request_queue *q)
 {
-	q->tag_set = set;
-
 	mutex_lock(&set->tag_list_lock);
 
 	/*
@@ -2467,8 +2465,6 @@ void blk_mq_release(struct request_queue *q)
 		kobject_put(&hctx->kobj);
 	}
 
-	q->mq_map = NULL;
-
 	kfree(q->queue_hw_ctx);
 
 	/*
@@ -2588,7 +2584,7 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,
 		int node;
 		struct blk_mq_hw_ctx *hctx;
 
-		node = blk_mq_hw_queue_to_node(q->mq_map, i);
+		node = blk_mq_hw_queue_to_node(set->mq_map, i);
 		/*
 		 * If the hw queue has been mapped to another numa node,
 		 * we need to realloc the hctx. If allocation fails, fallback
@@ -2665,8 +2661,6 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 	if (!q->queue_hw_ctx)
 		goto err_percpu;
 
-	q->mq_map = set->mq_map;
-
 	blk_mq_realloc_hw_ctxs(set, q);
 	if (!q->nr_hw_queues)
 		goto err_hctxs;
@@ -2675,6 +2669,7 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 	blk_queue_rq_timeout(q, set->timeout ? set->timeout : 30 * HZ);
 
 	q->nr_queues = nr_cpu_ids;
+	q->tag_set = set;
 
 	q->queue_flags |= QUEUE_FLAG_MQ_DEFAULT;
 
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 9497b47e2526..9536be06d022 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -75,7 +75,9 @@ extern int blk_mq_hw_queue_to_node(unsigned int *map, unsigned int);
 static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
 		int cpu)
 {
-	return q->queue_hw_ctx[q->mq_map[cpu]];
+	struct blk_mq_tag_set *set = q->tag_set;
+
+	return q->queue_hw_ctx[set->mq_map[cpu]];
 }
 
 /*
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index c675e2b5af62..4223ae2d2198 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -412,8 +412,6 @@ struct request_queue {
 
 	const struct blk_mq_ops	*mq_ops;
 
-	unsigned int		*mq_map;
-
 	/* sw queues */
 	struct blk_mq_ctx __percpu	*queue_ctx;
 	unsigned int		nr_queues;
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 02/16] blk-mq: abstract out queue map
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
  2018-10-30 18:32 ` [PATCH 01/16] blk-mq: kill q->mq_map Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:31   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 03/16] blk-mq: provide dummy blk_mq_map_queue_type() helper Jens Axboe
                   ` (14 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

This is in preparation for allowing multiple sets of maps per
queue, if so desired.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq-cpumap.c                 | 10 ++++----
 block/blk-mq-pci.c                    | 10 ++++----
 block/blk-mq-rdma.c                   |  4 ++--
 block/blk-mq-virtio.c                 |  8 +++----
 block/blk-mq.c                        | 34 ++++++++++++++-------------
 block/blk-mq.h                        |  8 +++----
 drivers/block/virtio_blk.c            |  2 +-
 drivers/nvme/host/pci.c               |  2 +-
 drivers/scsi/qla2xxx/qla_os.c         |  5 ++--
 drivers/scsi/scsi_lib.c               |  2 +-
 drivers/scsi/smartpqi/smartpqi_init.c |  3 ++-
 drivers/scsi/virtio_scsi.c            |  3 ++-
 include/linux/blk-mq-pci.h            |  4 ++--
 include/linux/blk-mq-virtio.h         |  4 ++--
 include/linux/blk-mq.h                | 15 +++++++++---
 15 files changed, 64 insertions(+), 50 deletions(-)

diff --git a/block/blk-mq-cpumap.c b/block/blk-mq-cpumap.c
index 3eb169f15842..6e6686c55984 100644
--- a/block/blk-mq-cpumap.c
+++ b/block/blk-mq-cpumap.c
@@ -30,10 +30,10 @@ static int get_first_sibling(unsigned int cpu)
 	return cpu;
 }
 
-int blk_mq_map_queues(struct blk_mq_tag_set *set)
+int blk_mq_map_queues(struct blk_mq_queue_map *qmap)
 {
-	unsigned int *map = set->mq_map;
-	unsigned int nr_queues = set->nr_hw_queues;
+	unsigned int *map = qmap->mq_map;
+	unsigned int nr_queues = qmap->nr_queues;
 	unsigned int cpu, first_sibling;
 
 	for_each_possible_cpu(cpu) {
@@ -62,12 +62,12 @@ EXPORT_SYMBOL_GPL(blk_mq_map_queues);
  * 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 blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int index)
 {
 	int i;
 
 	for_each_possible_cpu(i) {
-		if (index == mq_map[i])
+		if (index == qmap->mq_map[i])
 			return local_memory_node(cpu_to_node(i));
 	}
 
diff --git a/block/blk-mq-pci.c b/block/blk-mq-pci.c
index db644ec624f5..40333d60a850 100644
--- a/block/blk-mq-pci.c
+++ b/block/blk-mq-pci.c
@@ -31,26 +31,26 @@
  * that maps a queue to the CPUs that have irq affinity for the corresponding
  * vector.
  */
-int blk_mq_pci_map_queues(struct blk_mq_tag_set *set, struct pci_dev *pdev,
+int blk_mq_pci_map_queues(struct blk_mq_queue_map *qmap, struct pci_dev *pdev,
 			    int offset)
 {
 	const struct cpumask *mask;
 	unsigned int queue, cpu;
 
-	for (queue = 0; queue < set->nr_hw_queues; queue++) {
+	for (queue = 0; queue < qmap->nr_queues; queue++) {
 		mask = pci_irq_get_affinity(pdev, queue + offset);
 		if (!mask)
 			goto fallback;
 
 		for_each_cpu(cpu, mask)
-			set->mq_map[cpu] = queue;
+			qmap->mq_map[cpu] = queue;
 	}
 
 	return 0;
 
 fallback:
-	WARN_ON_ONCE(set->nr_hw_queues > 1);
-	blk_mq_clear_mq_map(set);
+	WARN_ON_ONCE(qmap->nr_queues > 1);
+	blk_mq_clear_mq_map(qmap);
 	return 0;
 }
 EXPORT_SYMBOL_GPL(blk_mq_pci_map_queues);
diff --git a/block/blk-mq-rdma.c b/block/blk-mq-rdma.c
index 996167f1de18..a71576aff3a5 100644
--- a/block/blk-mq-rdma.c
+++ b/block/blk-mq-rdma.c
@@ -41,12 +41,12 @@ int blk_mq_rdma_map_queues(struct blk_mq_tag_set *set,
 			goto fallback;
 
 		for_each_cpu(cpu, mask)
-			set->mq_map[cpu] = queue;
+			set->map[0].mq_map[cpu] = queue;
 	}
 
 	return 0;
 
 fallback:
-	return blk_mq_map_queues(set);
+	return blk_mq_map_queues(&set->map[0]);
 }
 EXPORT_SYMBOL_GPL(blk_mq_rdma_map_queues);
diff --git a/block/blk-mq-virtio.c b/block/blk-mq-virtio.c
index c3afbca11299..661fbfef480f 100644
--- a/block/blk-mq-virtio.c
+++ b/block/blk-mq-virtio.c
@@ -29,7 +29,7 @@
  * that maps a queue to the CPUs that have irq affinity for the corresponding
  * vector.
  */
-int blk_mq_virtio_map_queues(struct blk_mq_tag_set *set,
+int blk_mq_virtio_map_queues(struct blk_mq_queue_map *qmap,
 		struct virtio_device *vdev, int first_vec)
 {
 	const struct cpumask *mask;
@@ -38,17 +38,17 @@ int blk_mq_virtio_map_queues(struct blk_mq_tag_set *set,
 	if (!vdev->config->get_vq_affinity)
 		goto fallback;
 
-	for (queue = 0; queue < set->nr_hw_queues; queue++) {
+	for (queue = 0; queue < qmap->nr_queues; queue++) {
 		mask = vdev->config->get_vq_affinity(vdev, first_vec + queue);
 		if (!mask)
 			goto fallback;
 
 		for_each_cpu(cpu, mask)
-			set->mq_map[cpu] = queue;
+			qmap->mq_map[cpu] = queue;
 	}
 
 	return 0;
 fallback:
-	return blk_mq_map_queues(set);
+	return blk_mq_map_queues(qmap);
 }
 EXPORT_SYMBOL_GPL(blk_mq_virtio_map_queues);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 22d5beaab5a0..9f149429cfbd 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1974,7 +1974,7 @@ struct blk_mq_tags *blk_mq_alloc_rq_map(struct blk_mq_tag_set *set,
 	struct blk_mq_tags *tags;
 	int node;
 
-	node = blk_mq_hw_queue_to_node(set->mq_map, hctx_idx);
+	node = blk_mq_hw_queue_to_node(&set->map[0], hctx_idx);
 	if (node == NUMA_NO_NODE)
 		node = set->numa_node;
 
@@ -2030,7 +2030,7 @@ int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
 	size_t rq_size, left;
 	int node;
 
-	node = blk_mq_hw_queue_to_node(set->mq_map, hctx_idx);
+	node = blk_mq_hw_queue_to_node(&set->map[0], hctx_idx);
 	if (node == NUMA_NO_NODE)
 		node = set->numa_node;
 
@@ -2321,7 +2321,7 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 	 * If the cpu isn't present, the cpu is mapped to first hctx.
 	 */
 	for_each_possible_cpu(i) {
-		hctx_idx = set->mq_map[i];
+		hctx_idx = set->map[0].mq_map[i];
 		/* unmapped hw queue can be remapped after CPU topo changed */
 		if (!set->tags[hctx_idx] &&
 		    !__blk_mq_alloc_rq_map(set, hctx_idx)) {
@@ -2331,7 +2331,7 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 			 * case, remap the current ctx to hctx[0] which
 			 * is guaranteed to always have tags allocated
 			 */
-			set->mq_map[i] = 0;
+			set->map[0].mq_map[i] = 0;
 		}
 
 		ctx = per_cpu_ptr(q->queue_ctx, i);
@@ -2584,7 +2584,7 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,
 		int node;
 		struct blk_mq_hw_ctx *hctx;
 
-		node = blk_mq_hw_queue_to_node(set->mq_map, i);
+		node = blk_mq_hw_queue_to_node(&set->map[0], i);
 		/*
 		 * If the hw queue has been mapped to another numa node,
 		 * we need to realloc the hctx. If allocation fails, fallback
@@ -2793,18 +2793,18 @@ static int blk_mq_update_queue_map(struct blk_mq_tag_set *set)
 		 * for (queue = 0; queue < set->nr_hw_queues; queue++) {
 		 * 	mask = get_cpu_mask(queue)
 		 * 	for_each_cpu(cpu, mask)
-		 * 		set->mq_map[cpu] = queue;
+		 * 		set->map.mq_map[cpu] = queue;
 		 * }
 		 *
 		 * When we need to remap, the table has to be cleared for
 		 * killing stale mapping since one CPU may not be mapped
 		 * to any hw queue.
 		 */
-		blk_mq_clear_mq_map(set);
+		blk_mq_clear_mq_map(&set->map[0]);
 
 		return set->ops->map_queues(set);
 	} else
-		return blk_mq_map_queues(set);
+		return blk_mq_map_queues(&set->map[0]);
 }
 
 /*
@@ -2859,10 +2859,12 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		return -ENOMEM;
 
 	ret = -ENOMEM;
-	set->mq_map = kcalloc_node(nr_cpu_ids, sizeof(*set->mq_map),
-				   GFP_KERNEL, set->numa_node);
-	if (!set->mq_map)
+	set->map[0].mq_map = kcalloc_node(nr_cpu_ids,
+					  sizeof(*set->map[0].mq_map),
+					  GFP_KERNEL, set->numa_node);
+	if (!set->map[0].mq_map)
 		goto out_free_tags;
+	set->map[0].nr_queues = set->nr_hw_queues;
 
 	ret = blk_mq_update_queue_map(set);
 	if (ret)
@@ -2878,8 +2880,8 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 	return 0;
 
 out_free_mq_map:
-	kfree(set->mq_map);
-	set->mq_map = NULL;
+	kfree(set->map[0].mq_map);
+	set->map[0].mq_map = NULL;
 out_free_tags:
 	kfree(set->tags);
 	set->tags = NULL;
@@ -2894,8 +2896,8 @@ void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
 	for (i = 0; i < nr_cpu_ids; i++)
 		blk_mq_free_map_and_requests(set, i);
 
-	kfree(set->mq_map);
-	set->mq_map = NULL;
+	kfree(set->map[0].mq_map);
+	set->map[0].mq_map = NULL;
 
 	kfree(set->tags);
 	set->tags = NULL;
@@ -3056,7 +3058,7 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 			pr_warn("Increasing nr_hw_queues to %d fails, fallback to %d\n",
 					nr_hw_queues, prev_nr_hw_queues);
 			set->nr_hw_queues = prev_nr_hw_queues;
-			blk_mq_map_queues(set);
+			blk_mq_map_queues(&set->map[0]);
 			goto fallback;
 		}
 		blk_mq_map_swqueue(q);
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 9536be06d022..889f0069dd80 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -70,14 +70,14 @@ void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
 /*
  * CPU -> queue mappings
  */
-extern int blk_mq_hw_queue_to_node(unsigned int *map, unsigned int);
+extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
 
 static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
 		int cpu)
 {
 	struct blk_mq_tag_set *set = q->tag_set;
 
-	return q->queue_hw_ctx[set->mq_map[cpu]];
+	return q->queue_hw_ctx[set->map[0].mq_map[cpu]];
 }
 
 /*
@@ -206,12 +206,12 @@ static inline void blk_mq_put_driver_tag(struct request *rq)
 	__blk_mq_put_driver_tag(hctx, rq);
 }
 
-static inline void blk_mq_clear_mq_map(struct blk_mq_tag_set *set)
+static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap)
 {
 	int cpu;
 
 	for_each_possible_cpu(cpu)
-		set->mq_map[cpu] = 0;
+		qmap->mq_map[cpu] = 0;
 }
 
 #endif
diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 086c6bb12baa..6e869d05f91e 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -624,7 +624,7 @@ static int virtblk_map_queues(struct blk_mq_tag_set *set)
 {
 	struct virtio_blk *vblk = set->driver_data;
 
-	return blk_mq_virtio_map_queues(set, vblk->vdev, 0);
+	return blk_mq_virtio_map_queues(&set->map[0], vblk->vdev, 0);
 }
 
 #ifdef CONFIG_VIRTIO_BLK_SCSI
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index f30031945ee4..e5d783cb6937 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -435,7 +435,7 @@ static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
 {
 	struct nvme_dev *dev = set->driver_data;
 
-	return blk_mq_pci_map_queues(set, to_pci_dev(dev->dev),
+	return blk_mq_pci_map_queues(&set->map[0], to_pci_dev(dev->dev),
 			dev->num_vecs > 1 ? 1 /* admin queue */ : 0);
 }
 
diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
index 3e2665c66bc4..ca9ac124f218 100644
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -6934,11 +6934,12 @@ static int qla2xxx_map_queues(struct Scsi_Host *shost)
 {
 	int rc;
 	scsi_qla_host_t *vha = (scsi_qla_host_t *)shost->hostdata;
+	struct blk_mq_queue_map *qmap = &shost->tag_set.map[0];
 
 	if (USER_CTRL_IRQ(vha->hw))
-		rc = blk_mq_map_queues(&shost->tag_set);
+		rc = blk_mq_map_queues(qmap);
 	else
-		rc = blk_mq_pci_map_queues(&shost->tag_set, vha->hw->pdev, 0);
+		rc = blk_mq_pci_map_queues(qmap, vha->hw->pdev, 0);
 	return rc;
 }
 
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 651be30ba96a..ed81b8e74cfe 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1812,7 +1812,7 @@ static int scsi_map_queues(struct blk_mq_tag_set *set)
 
 	if (shost->hostt->map_queues)
 		return shost->hostt->map_queues(shost);
-	return blk_mq_map_queues(set);
+	return blk_mq_map_queues(&set->map[0]);
 }
 
 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
diff --git a/drivers/scsi/smartpqi/smartpqi_init.c b/drivers/scsi/smartpqi/smartpqi_init.c
index a25a07a0b7f0..bac084260d80 100644
--- a/drivers/scsi/smartpqi/smartpqi_init.c
+++ b/drivers/scsi/smartpqi/smartpqi_init.c
@@ -5319,7 +5319,8 @@ static int pqi_map_queues(struct Scsi_Host *shost)
 {
 	struct pqi_ctrl_info *ctrl_info = shost_to_hba(shost);
 
-	return blk_mq_pci_map_queues(&shost->tag_set, ctrl_info->pci_dev, 0);
+	return blk_mq_pci_map_queues(&shost->tag_set.map[0],
+					ctrl_info->pci_dev, 0);
 }
 
 static int pqi_getpciinfo_ioctl(struct pqi_ctrl_info *ctrl_info,
diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
index 1c72db94270e..c3c95b314286 100644
--- a/drivers/scsi/virtio_scsi.c
+++ b/drivers/scsi/virtio_scsi.c
@@ -719,8 +719,9 @@ static void virtscsi_target_destroy(struct scsi_target *starget)
 static int virtscsi_map_queues(struct Scsi_Host *shost)
 {
 	struct virtio_scsi *vscsi = shost_priv(shost);
+	struct blk_mq_queue_map *qmap = &shost->tag_set.map[0];
 
-	return blk_mq_virtio_map_queues(&shost->tag_set, vscsi->vdev, 2);
+	return blk_mq_virtio_map_queues(qmap, vscsi->vdev, 2);
 }
 
 /*
diff --git a/include/linux/blk-mq-pci.h b/include/linux/blk-mq-pci.h
index 9f4c17f0d2d8..0b1f45c62623 100644
--- a/include/linux/blk-mq-pci.h
+++ b/include/linux/blk-mq-pci.h
@@ -2,10 +2,10 @@
 #ifndef _LINUX_BLK_MQ_PCI_H
 #define _LINUX_BLK_MQ_PCI_H
 
-struct blk_mq_tag_set;
+struct blk_mq_queue_map;
 struct pci_dev;
 
-int blk_mq_pci_map_queues(struct blk_mq_tag_set *set, struct pci_dev *pdev,
+int blk_mq_pci_map_queues(struct blk_mq_queue_map *qmap, struct pci_dev *pdev,
 			  int offset);
 
 #endif /* _LINUX_BLK_MQ_PCI_H */
diff --git a/include/linux/blk-mq-virtio.h b/include/linux/blk-mq-virtio.h
index 69b4da262c45..687ae287e1dc 100644
--- a/include/linux/blk-mq-virtio.h
+++ b/include/linux/blk-mq-virtio.h
@@ -2,10 +2,10 @@
 #ifndef _LINUX_BLK_MQ_VIRTIO_H
 #define _LINUX_BLK_MQ_VIRTIO_H
 
-struct blk_mq_tag_set;
+struct blk_mq_queue_map;
 struct virtio_device;
 
-int blk_mq_virtio_map_queues(struct blk_mq_tag_set *set,
+int blk_mq_virtio_map_queues(struct blk_mq_queue_map *qmap,
 		struct virtio_device *vdev, int first_vec);
 
 #endif /* _LINUX_BLK_MQ_VIRTIO_H */
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 5c8418ebbfd6..da88e539601b 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -74,10 +74,19 @@ struct blk_mq_hw_ctx {
 	struct srcu_struct	srcu[0];
 };
 
+struct blk_mq_queue_map {
+	unsigned int *mq_map;
+	unsigned int nr_queues;
+};
+
+enum {
+	HCTX_MAX_TYPES = 1,
+};
+
 struct blk_mq_tag_set {
-	unsigned int		*mq_map;
+	struct blk_mq_queue_map	map[HCTX_MAX_TYPES];
 	const struct blk_mq_ops	*ops;
-	unsigned int		nr_hw_queues;
+	unsigned int		nr_hw_queues;	/* nr hw queues across maps */
 	unsigned int		queue_depth;	/* max hw supported */
 	unsigned int		reserved_tags;
 	unsigned int		cmd_size;	/* per-request extra data */
@@ -294,7 +303,7 @@ void blk_mq_freeze_queue_wait(struct request_queue *q);
 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
 				     unsigned long timeout);
 
-int blk_mq_map_queues(struct blk_mq_tag_set *set);
+int blk_mq_map_queues(struct blk_mq_queue_map *qmap);
 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 
 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 03/16] blk-mq: provide dummy blk_mq_map_queue_type() helper
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
  2018-10-30 18:32 ` [PATCH 01/16] blk-mq: kill q->mq_map Jens Axboe
  2018-10-30 18:32 ` [PATCH 02/16] blk-mq: abstract out queue map Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:32   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping Jens Axboe
                   ` (13 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

Doesn't do anything right now, but it's needed as a prep patch
to get the interfaces right.

While in there, correct the blk_mq_map_queue() CPU type to an unsigned
int.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.h | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/block/blk-mq.h b/block/blk-mq.h
index 889f0069dd80..d9facfb9ca51 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -73,13 +73,20 @@ void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
 extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
 
 static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
-		int cpu)
+						     unsigned int cpu)
 {
 	struct blk_mq_tag_set *set = q->tag_set;
 
 	return q->queue_hw_ctx[set->map[0].mq_map[cpu]];
 }
 
+static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
+							  unsigned int hctx_type,
+							  unsigned int cpu)
+{
+	return blk_mq_map_queue(q, cpu);
+}
+
 /*
  * sysfs helpers
  */
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (2 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 03/16] blk-mq: provide dummy blk_mq_map_queue_type() helper Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:37   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues Jens Axboe
                   ` (12 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

Prep patch for being able to place request based not just on
CPU location, but also on the type of request.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-flush.c      |  7 +++---
 block/blk-mq-debugfs.c |  4 +++-
 block/blk-mq-sched.c   | 16 ++++++++++----
 block/blk-mq-tag.c     |  5 +++--
 block/blk-mq.c         | 50 +++++++++++++++++++++++-------------------
 block/blk-mq.h         |  6 +++--
 block/blk.h            |  6 ++---
 7 files changed, 57 insertions(+), 37 deletions(-)

diff --git a/block/blk-flush.c b/block/blk-flush.c
index 9baa9a119447..7922dba81497 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -219,7 +219,7 @@ static void flush_end_io(struct request *flush_rq, blk_status_t error)
 
 	/* release the tag's ownership to the req cloned from */
 	spin_lock_irqsave(&fq->mq_flush_lock, flags);
-	hctx = blk_mq_map_queue(q, flush_rq->mq_ctx->cpu);
+	hctx = blk_mq_map_queue(q, flush_rq->cmd_flags, flush_rq->mq_ctx->cpu);
 	if (!q->elevator) {
 		blk_mq_tag_set_rq(hctx, flush_rq->tag, fq->orig_rq);
 		flush_rq->tag = -1;
@@ -307,7 +307,8 @@ static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq,
 	if (!q->elevator) {
 		fq->orig_rq = first_rq;
 		flush_rq->tag = first_rq->tag;
-		hctx = blk_mq_map_queue(q, first_rq->mq_ctx->cpu);
+		hctx = blk_mq_map_queue(q, first_rq->cmd_flags,
+					first_rq->mq_ctx->cpu);
 		blk_mq_tag_set_rq(hctx, first_rq->tag, flush_rq);
 	} else {
 		flush_rq->internal_tag = first_rq->internal_tag;
@@ -330,7 +331,7 @@ static void mq_flush_data_end_io(struct request *rq, blk_status_t error)
 	unsigned long flags;
 	struct blk_flush_queue *fq = blk_get_flush_queue(q, ctx);
 
-	hctx = blk_mq_map_queue(q, ctx->cpu);
+	hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
 
 	if (q->elevator) {
 		WARN_ON(rq->tag < 0);
diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index 9ed43a7c70b5..fac70c81b7de 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -427,8 +427,10 @@ struct show_busy_params {
 static void hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
 {
 	const struct show_busy_params *params = data;
+	struct blk_mq_hw_ctx *hctx;
 
-	if (blk_mq_map_queue(rq->q, rq->mq_ctx->cpu) == params->hctx)
+	hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu);
+	if (hctx == params->hctx)
 		__blk_mq_debugfs_rq_show(params->m,
 					 list_entry_rq(&rq->queuelist));
 }
diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index 29bfe8017a2d..8125e9393ec2 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -311,7 +311,7 @@ bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio)
 {
 	struct elevator_queue *e = q->elevator;
 	struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, bio->bi_opf, ctx->cpu);
 	bool ret = false;
 
 	if (e && e->type->ops.mq.bio_merge) {
@@ -367,7 +367,9 @@ void blk_mq_sched_insert_request(struct request *rq, bool at_head,
 	struct request_queue *q = rq->q;
 	struct elevator_queue *e = q->elevator;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx;
+
+	hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
 
 	/* flush rq in flush machinery need to be dispatched directly */
 	if (!(rq->rq_flags & RQF_FLUSH_SEQ) && op_is_flush(rq->cmd_flags)) {
@@ -400,9 +402,15 @@ void blk_mq_sched_insert_requests(struct request_queue *q,
 				  struct blk_mq_ctx *ctx,
 				  struct list_head *list, bool run_queue_async)
 {
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
-	struct elevator_queue *e = hctx->queue->elevator;
+	struct blk_mq_hw_ctx *hctx;
+	struct elevator_queue *e;
+	struct request *rq;
+
+	/* For list inserts, requests better be on the same hw queue */
+	rq = list_first_entry(list, struct request, queuelist);
+	hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
 
+	e = hctx->queue->elevator;
 	if (e && e->type->ops.mq.insert_requests)
 		e->type->ops.mq.insert_requests(hctx, list, false);
 	else {
diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index 4254e74c1446..478a959357f5 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -168,7 +168,8 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data)
 		io_schedule();
 
 		data->ctx = blk_mq_get_ctx(data->q);
-		data->hctx = blk_mq_map_queue(data->q, data->ctx->cpu);
+		data->hctx = blk_mq_map_queue(data->q, data->cmd_flags,
+						data->ctx->cpu);
 		tags = blk_mq_tags_from_data(data);
 		if (data->flags & BLK_MQ_REQ_RESERVED)
 			bt = &tags->breserved_tags;
@@ -530,7 +531,7 @@ u32 blk_mq_unique_tag(struct request *rq)
 	struct blk_mq_hw_ctx *hctx;
 	int hwq = 0;
 
-	hctx = blk_mq_map_queue(q, rq->mq_ctx->cpu);
+	hctx = blk_mq_map_queue(q, rq->cmd_flags, rq->mq_ctx->cpu);
 	hwq = hctx->queue_num;
 
 	return (hwq << BLK_MQ_UNIQUE_TAG_BITS) |
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 9f149429cfbd..e3febb5691c4 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -332,8 +332,8 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
 }
 
 static struct request *blk_mq_get_request(struct request_queue *q,
-		struct bio *bio, unsigned int op,
-		struct blk_mq_alloc_data *data)
+					  struct bio *bio,
+					  struct blk_mq_alloc_data *data)
 {
 	struct elevator_queue *e = q->elevator;
 	struct request *rq;
@@ -347,8 +347,9 @@ static struct request *blk_mq_get_request(struct request_queue *q,
 		put_ctx_on_error = true;
 	}
 	if (likely(!data->hctx))
-		data->hctx = blk_mq_map_queue(q, data->ctx->cpu);
-	if (op & REQ_NOWAIT)
+		data->hctx = blk_mq_map_queue(q, data->cmd_flags,
+						data->ctx->cpu);
+	if (data->cmd_flags & REQ_NOWAIT)
 		data->flags |= BLK_MQ_REQ_NOWAIT;
 
 	if (e) {
@@ -359,9 +360,10 @@ static struct request *blk_mq_get_request(struct request_queue *q,
 		 * dispatch list. Don't include reserved tags in the
 		 * limiting, as it isn't useful.
 		 */
-		if (!op_is_flush(op) && e->type->ops.mq.limit_depth &&
+		if (!op_is_flush(data->cmd_flags) &&
+		    e->type->ops.mq.limit_depth &&
 		    !(data->flags & BLK_MQ_REQ_RESERVED))
-			e->type->ops.mq.limit_depth(op, data);
+			e->type->ops.mq.limit_depth(data->cmd_flags, data);
 	} else {
 		blk_mq_tag_busy(data->hctx);
 	}
@@ -376,8 +378,8 @@ static struct request *blk_mq_get_request(struct request_queue *q,
 		return NULL;
 	}
 
-	rq = blk_mq_rq_ctx_init(data, tag, op);
-	if (!op_is_flush(op)) {
+	rq = blk_mq_rq_ctx_init(data, tag, data->cmd_flags);
+	if (!op_is_flush(data->cmd_flags)) {
 		rq->elv.icq = NULL;
 		if (e && e->type->ops.mq.prepare_request) {
 			if (e->type->icq_cache && rq_ioc(bio))
@@ -394,7 +396,7 @@ static struct request *blk_mq_get_request(struct request_queue *q,
 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
 		blk_mq_req_flags_t flags)
 {
-	struct blk_mq_alloc_data alloc_data = { .flags = flags };
+	struct blk_mq_alloc_data alloc_data = { .flags = flags, .cmd_flags = op };
 	struct request *rq;
 	int ret;
 
@@ -402,7 +404,7 @@ struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
 	if (ret)
 		return ERR_PTR(ret);
 
-	rq = blk_mq_get_request(q, NULL, op, &alloc_data);
+	rq = blk_mq_get_request(q, NULL, &alloc_data);
 	blk_queue_exit(q);
 
 	if (!rq)
@@ -420,7 +422,7 @@ EXPORT_SYMBOL(blk_mq_alloc_request);
 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
 	unsigned int op, blk_mq_req_flags_t flags, unsigned int hctx_idx)
 {
-	struct blk_mq_alloc_data alloc_data = { .flags = flags };
+	struct blk_mq_alloc_data alloc_data = { .flags = flags, .cmd_flags = op };
 	struct request *rq;
 	unsigned int cpu;
 	int ret;
@@ -453,7 +455,7 @@ struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
 	cpu = cpumask_first_and(alloc_data.hctx->cpumask, cpu_online_mask);
 	alloc_data.ctx = __blk_mq_get_ctx(q, cpu);
 
-	rq = blk_mq_get_request(q, NULL, op, &alloc_data);
+	rq = blk_mq_get_request(q, NULL, &alloc_data);
 	blk_queue_exit(q);
 
 	if (!rq)
@@ -467,7 +469,7 @@ static void __blk_mq_free_request(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
 	const int sched_tag = rq->internal_tag;
 
 	blk_pm_mark_last_busy(rq);
@@ -484,7 +486,7 @@ void blk_mq_free_request(struct request *rq)
 	struct request_queue *q = rq->q;
 	struct elevator_queue *e = q->elevator;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
 
 	if (rq->rq_flags & RQF_ELVPRIV) {
 		if (e && e->type->ops.mq.finish_request)
@@ -976,8 +978,9 @@ bool blk_mq_get_driver_tag(struct request *rq)
 {
 	struct blk_mq_alloc_data data = {
 		.q = rq->q,
-		.hctx = blk_mq_map_queue(rq->q, rq->mq_ctx->cpu),
+		.hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu),
 		.flags = BLK_MQ_REQ_NOWAIT,
+		.cmd_flags = rq->cmd_flags,
 	};
 	bool shared;
 
@@ -1141,7 +1144,7 @@ bool blk_mq_dispatch_rq_list(struct request_queue *q, struct list_head *list,
 
 		rq = list_first_entry(list, struct request, queuelist);
 
-		hctx = blk_mq_map_queue(rq->q, rq->mq_ctx->cpu);
+		hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu);
 		if (!got_budget && !blk_mq_get_dispatch_budget(hctx))
 			break;
 
@@ -1572,7 +1575,8 @@ void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
 void blk_mq_request_bypass_insert(struct request *rq, bool run_queue)
 {
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, rq->cmd_flags,
+							ctx->cpu);
 
 	spin_lock(&hctx->lock);
 	list_add_tail(&rq->queuelist, &hctx->dispatch);
@@ -1782,7 +1786,8 @@ blk_status_t blk_mq_request_issue_directly(struct request *rq)
 	int srcu_idx;
 	blk_qc_t unused_cookie;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, rq->cmd_flags,
+							ctx->cpu);
 
 	hctx_lock(hctx, &srcu_idx);
 	ret = __blk_mq_try_issue_directly(hctx, rq, &unused_cookie, true);
@@ -1816,7 +1821,7 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
 {
 	const int is_sync = op_is_sync(bio->bi_opf);
 	const int is_flush_fua = op_is_flush(bio->bi_opf);
-	struct blk_mq_alloc_data data = { .flags = 0 };
+	struct blk_mq_alloc_data data = { .flags = 0, .cmd_flags = bio->bi_opf };
 	struct request *rq;
 	unsigned int request_count = 0;
 	struct blk_plug *plug;
@@ -1839,7 +1844,7 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
 
 	rq_qos_throttle(q, bio, NULL);
 
-	rq = blk_mq_get_request(q, bio, bio->bi_opf, &data);
+	rq = blk_mq_get_request(q, bio, &data);
 	if (unlikely(!rq)) {
 		rq_qos_cleanup(q, bio);
 		if (bio->bi_opf & REQ_NOWAIT)
@@ -1908,6 +1913,7 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
 
 		if (same_queue_rq) {
 			data.hctx = blk_mq_map_queue(q,
+					same_queue_rq->cmd_flags,
 					same_queue_rq->mq_ctx->cpu);
 			blk_mq_try_issue_directly(data.hctx, same_queue_rq,
 					&cookie);
@@ -2262,7 +2268,7 @@ static void blk_mq_init_cpu_queues(struct request_queue *q,
 		 * Set local node, IFF we have more than one hw queue. If
 		 * not, we remain on the home node of the device
 		 */
-		hctx = blk_mq_map_queue(q, i);
+		hctx = blk_mq_map_queue_type(q, 0, i);
 		if (nr_hw_queues > 1 && hctx->numa_node == NUMA_NO_NODE)
 			hctx->numa_node = local_memory_node(cpu_to_node(i));
 	}
@@ -2335,7 +2341,7 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 		}
 
 		ctx = per_cpu_ptr(q->queue_ctx, i);
-		hctx = blk_mq_map_queue(q, i);
+		hctx = blk_mq_map_queue_type(q, 0, i);
 
 		cpumask_set_cpu(i, hctx->cpumask);
 		ctx->index_hw = hctx->nr_ctx;
diff --git a/block/blk-mq.h b/block/blk-mq.h
index d9facfb9ca51..6a8f8b60d8ba 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -73,6 +73,7 @@ void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
 extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
 
 static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
+						     unsigned int flags,
 						     unsigned int cpu)
 {
 	struct blk_mq_tag_set *set = q->tag_set;
@@ -84,7 +85,7 @@ static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *
 							  unsigned int hctx_type,
 							  unsigned int cpu)
 {
-	return blk_mq_map_queue(q, cpu);
+	return blk_mq_map_queue(q, hctx_type, cpu);
 }
 
 /*
@@ -135,6 +136,7 @@ struct blk_mq_alloc_data {
 	struct request_queue *q;
 	blk_mq_req_flags_t flags;
 	unsigned int shallow_depth;
+	unsigned int cmd_flags;
 
 	/* input & output parameter */
 	struct blk_mq_ctx *ctx;
@@ -209,7 +211,7 @@ static inline void blk_mq_put_driver_tag(struct request *rq)
 	if (rq->tag == -1 || rq->internal_tag == -1)
 		return;
 
-	hctx = blk_mq_map_queue(rq->q, rq->mq_ctx->cpu);
+	hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu);
 	__blk_mq_put_driver_tag(hctx, rq);
 }
 
diff --git a/block/blk.h b/block/blk.h
index 2bf1cfeeb9c0..78ae94886acf 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -104,10 +104,10 @@ static inline void queue_flag_clear(unsigned int flag, struct request_queue *q)
 	__clear_bit(flag, &q->queue_flags);
 }
 
-static inline struct blk_flush_queue *blk_get_flush_queue(
-		struct request_queue *q, struct blk_mq_ctx *ctx)
+static inline struct blk_flush_queue *
+blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
 {
-	return blk_mq_map_queue(q, ctx->cpu)->fq;
+	return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx->cpu)->fq;
 }
 
 static inline void __blk_get_queue(struct request_queue *q)
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (3 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:49   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory Jens Axboe
                   ` (11 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

The mapping used to be dependent on just the CPU location, but
now it's a tuple of (type, cpu) instead. This is a prep patch
for allowing a single software queue to map to multiple hardware
queues. No functional changes in this patch.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq-sched.c   |  2 +-
 block/blk-mq.c         | 22 ++++++++++++++++------
 block/blk-mq.h         |  2 +-
 block/kyber-iosched.c  |  6 +++---
 include/linux/blk-mq.h |  3 ++-
 5 files changed, 23 insertions(+), 12 deletions(-)

diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index 8125e9393ec2..d232ecf3290c 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -110,7 +110,7 @@ static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
 static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx,
 					  struct blk_mq_ctx *ctx)
 {
-	unsigned idx = ctx->index_hw;
+	unsigned short idx = ctx->index_hw[hctx->type];
 
 	if (++idx == hctx->nr_ctx)
 		idx = 0;
diff --git a/block/blk-mq.c b/block/blk-mq.c
index e3febb5691c4..34afbad0ebf6 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -75,14 +75,18 @@ static bool blk_mq_hctx_has_pending(struct blk_mq_hw_ctx *hctx)
 static void blk_mq_hctx_mark_pending(struct blk_mq_hw_ctx *hctx,
 				     struct blk_mq_ctx *ctx)
 {
-	if (!sbitmap_test_bit(&hctx->ctx_map, ctx->index_hw))
-		sbitmap_set_bit(&hctx->ctx_map, ctx->index_hw);
+	const int bit = ctx->index_hw[hctx->type];
+
+	if (!sbitmap_test_bit(&hctx->ctx_map, bit))
+		sbitmap_set_bit(&hctx->ctx_map, bit);
 }
 
 static void blk_mq_hctx_clear_pending(struct blk_mq_hw_ctx *hctx,
 				      struct blk_mq_ctx *ctx)
 {
-	sbitmap_clear_bit(&hctx->ctx_map, ctx->index_hw);
+	const int bit = ctx->index_hw[hctx->type];
+
+	sbitmap_clear_bit(&hctx->ctx_map, bit);
 }
 
 struct mq_inflight {
@@ -954,7 +958,7 @@ static bool dispatch_rq_from_ctx(struct sbitmap *sb, unsigned int bitnr,
 struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx,
 					struct blk_mq_ctx *start)
 {
-	unsigned off = start ? start->index_hw : 0;
+	unsigned off = start ? start->index_hw[hctx->type] : 0;
 	struct dispatch_rq_data data = {
 		.hctx = hctx,
 		.rq   = NULL,
@@ -2342,10 +2346,16 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 
 		ctx = per_cpu_ptr(q->queue_ctx, i);
 		hctx = blk_mq_map_queue_type(q, 0, i);
-
+		hctx->type = 0;
 		cpumask_set_cpu(i, hctx->cpumask);
-		ctx->index_hw = hctx->nr_ctx;
+		ctx->index_hw[hctx->type] = hctx->nr_ctx;
 		hctx->ctxs[hctx->nr_ctx++] = ctx;
+
+		/*
+		 * If the nr_ctx type overflows, we have exceeded the
+		 * amount of sw queues we can support.
+		 */
+		BUG_ON(!hctx->nr_ctx);
 	}
 
 	mutex_unlock(&q->sysfs_lock);
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 6a8f8b60d8ba..1821f448f7c4 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -17,7 +17,7 @@ struct blk_mq_ctx {
 	}  ____cacheline_aligned_in_smp;
 
 	unsigned int		cpu;
-	unsigned int		index_hw;
+	unsigned short		index_hw[HCTX_MAX_TYPES];
 
 	/* incremented at dispatch time */
 	unsigned long		rq_dispatched[2];
diff --git a/block/kyber-iosched.c b/block/kyber-iosched.c
index 728757a34fa0..b824a639d5d4 100644
--- a/block/kyber-iosched.c
+++ b/block/kyber-iosched.c
@@ -576,7 +576,7 @@ static bool kyber_bio_merge(struct blk_mq_hw_ctx *hctx, struct bio *bio)
 {
 	struct kyber_hctx_data *khd = hctx->sched_data;
 	struct blk_mq_ctx *ctx = blk_mq_get_ctx(hctx->queue);
-	struct kyber_ctx_queue *kcq = &khd->kcqs[ctx->index_hw];
+	struct kyber_ctx_queue *kcq = &khd->kcqs[ctx->index_hw[hctx->type]];
 	unsigned int sched_domain = kyber_sched_domain(bio->bi_opf);
 	struct list_head *rq_list = &kcq->rq_list[sched_domain];
 	bool merged;
@@ -602,7 +602,7 @@ static void kyber_insert_requests(struct blk_mq_hw_ctx *hctx,
 
 	list_for_each_entry_safe(rq, next, rq_list, queuelist) {
 		unsigned int sched_domain = kyber_sched_domain(rq->cmd_flags);
-		struct kyber_ctx_queue *kcq = &khd->kcqs[rq->mq_ctx->index_hw];
+		struct kyber_ctx_queue *kcq = &khd->kcqs[rq->mq_ctx->index_hw[hctx->type]];
 		struct list_head *head = &kcq->rq_list[sched_domain];
 
 		spin_lock(&kcq->lock);
@@ -611,7 +611,7 @@ static void kyber_insert_requests(struct blk_mq_hw_ctx *hctx,
 		else
 			list_move_tail(&rq->queuelist, head);
 		sbitmap_set_bit(&khd->kcq_map[sched_domain],
-				rq->mq_ctx->index_hw);
+				rq->mq_ctx->index_hw[hctx->type]);
 		blk_mq_sched_request_inserted(rq);
 		spin_unlock(&kcq->lock);
 	}
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index da88e539601b..466b9202b69c 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -37,7 +37,8 @@ struct blk_mq_hw_ctx {
 	struct blk_mq_ctx	*dispatch_from;
 	unsigned int		dispatch_busy;
 
-	unsigned int		nr_ctx;
+	unsigned short		type;
+	unsigned short		nr_ctx;
 	struct blk_mq_ctx	**ctxs;
 
 	spinlock_t		dispatch_wait_lock;
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (4 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:53   ` Sagi Grimberg
  2018-11-01 21:59   ` Omar Sandoval
  2018-10-30 18:32 ` [PATCH 07/16] blk-mq: support multiple hctx maps Jens Axboe
                   ` (10 subsequent siblings)
  16 siblings, 2 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

It can be useful for a user to verify what type a given hardware
queue is, expose this information in sysfs.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq-sysfs.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/block/blk-mq-sysfs.c b/block/blk-mq-sysfs.c
index aafb44224c89..2d737f9e7ba7 100644
--- a/block/blk-mq-sysfs.c
+++ b/block/blk-mq-sysfs.c
@@ -161,6 +161,11 @@ static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
 	return ret;
 }
 
+static ssize_t blk_mq_hw_sysfs_type_show(struct blk_mq_hw_ctx *hctx, char *page)
+{
+	return sprintf(page, "%u\n", hctx->type);
+}
+
 static struct attribute *default_ctx_attrs[] = {
 	NULL,
 };
@@ -177,11 +182,16 @@ static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
 	.attr = {.name = "cpu_list", .mode = 0444 },
 	.show = blk_mq_hw_sysfs_cpus_show,
 };
+static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_type = {
+	.attr = {.name = "type", .mode = 0444 },
+	.show = blk_mq_hw_sysfs_type_show,
+};
 
 static struct attribute *default_hw_ctx_attrs[] = {
 	&blk_mq_hw_sysfs_nr_tags.attr,
 	&blk_mq_hw_sysfs_nr_reserved_tags.attr,
 	&blk_mq_hw_sysfs_cpus.attr,
+	&blk_mq_hw_sysfs_type.attr,
 	NULL,
 };
 
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 07/16] blk-mq: support multiple hctx maps
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (5 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  0:59   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 08/16] blk-mq: separate number of hardware queues from nr_cpu_ids Jens Axboe
                   ` (9 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

Add support for the tag set carrying multiple queue maps, and
for the driver to inform blk-mq how many it wishes to support
through setting set->nr_maps.

This adds an mq_ops helper for drivers that support more than 1
map, mq_ops->flags_to_type(). The function takes request/bio flags
and CPU, and returns a queue map index for that. We then use the
type information in blk_mq_map_queue() to index the map set.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c         | 92 ++++++++++++++++++++++++++++--------------
 block/blk-mq.h         | 33 +++++++++++----
 include/linux/blk-mq.h | 14 +++++++
 3 files changed, 100 insertions(+), 39 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 34afbad0ebf6..9d6e2f6f8ee9 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2257,7 +2257,8 @@ static int blk_mq_init_hctx(struct request_queue *q,
 static void blk_mq_init_cpu_queues(struct request_queue *q,
 				   unsigned int nr_hw_queues)
 {
-	unsigned int i;
+	struct blk_mq_tag_set *set = q->tag_set;
+	unsigned int i, j;
 
 	for_each_possible_cpu(i) {
 		struct blk_mq_ctx *__ctx = per_cpu_ptr(q->queue_ctx, i);
@@ -2272,9 +2273,11 @@ static void blk_mq_init_cpu_queues(struct request_queue *q,
 		 * Set local node, IFF we have more than one hw queue. If
 		 * not, we remain on the home node of the device
 		 */
-		hctx = blk_mq_map_queue_type(q, 0, i);
-		if (nr_hw_queues > 1 && hctx->numa_node == NUMA_NO_NODE)
-			hctx->numa_node = local_memory_node(cpu_to_node(i));
+		for (j = 0; j < set->nr_maps; j++) {
+			hctx = blk_mq_map_queue_type(q, j, i);
+			if (nr_hw_queues > 1 && hctx->numa_node == NUMA_NO_NODE)
+				hctx->numa_node = local_memory_node(cpu_to_node(i));
+		}
 	}
 }
 
@@ -2309,7 +2312,7 @@ static void blk_mq_free_map_and_requests(struct blk_mq_tag_set *set,
 
 static void blk_mq_map_swqueue(struct request_queue *q)
 {
-	unsigned int i, hctx_idx;
+	unsigned int i, j, hctx_idx;
 	struct blk_mq_hw_ctx *hctx;
 	struct blk_mq_ctx *ctx;
 	struct blk_mq_tag_set *set = q->tag_set;
@@ -2345,17 +2348,28 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 		}
 
 		ctx = per_cpu_ptr(q->queue_ctx, i);
-		hctx = blk_mq_map_queue_type(q, 0, i);
-		hctx->type = 0;
-		cpumask_set_cpu(i, hctx->cpumask);
-		ctx->index_hw[hctx->type] = hctx->nr_ctx;
-		hctx->ctxs[hctx->nr_ctx++] = ctx;
+		for (j = 0; j < set->nr_maps; j++) {
+			hctx = blk_mq_map_queue_type(q, j, i);
 
-		/*
-		 * If the nr_ctx type overflows, we have exceeded the
-		 * amount of sw queues we can support.
-		 */
-		BUG_ON(!hctx->nr_ctx);
+			/*
+			 * If the CPU is already set in the mask, then we've
+			 * mapped this one already. This can happen if
+			 * devices share queues across queue maps.
+			 */
+			if (cpumask_test_cpu(i, hctx->cpumask))
+				continue;
+
+			cpumask_set_cpu(i, hctx->cpumask);
+			hctx->type = j;
+			ctx->index_hw[hctx->type] = hctx->nr_ctx;
+			hctx->ctxs[hctx->nr_ctx++] = ctx;
+
+			/*
+			 * If the nr_ctx type overflows, we have exceeded the
+			 * amount of sw queues we can support.
+			 */
+			BUG_ON(!hctx->nr_ctx);
+		}
 	}
 
 	mutex_unlock(&q->sysfs_lock);
@@ -2523,6 +2537,7 @@ struct request_queue *blk_mq_init_sq_queue(struct blk_mq_tag_set *set,
 	memset(set, 0, sizeof(*set));
 	set->ops = ops;
 	set->nr_hw_queues = 1;
+	set->nr_maps = 1;
 	set->queue_depth = queue_depth;
 	set->numa_node = NUMA_NO_NODE;
 	set->flags = set_flags;
@@ -2802,6 +2817,8 @@ static int blk_mq_alloc_rq_maps(struct blk_mq_tag_set *set)
 static int blk_mq_update_queue_map(struct blk_mq_tag_set *set)
 {
 	if (set->ops->map_queues) {
+		int i;
+
 		/*
 		 * transport .map_queues is usually done in the following
 		 * way:
@@ -2809,18 +2826,21 @@ static int blk_mq_update_queue_map(struct blk_mq_tag_set *set)
 		 * for (queue = 0; queue < set->nr_hw_queues; queue++) {
 		 * 	mask = get_cpu_mask(queue)
 		 * 	for_each_cpu(cpu, mask)
-		 * 		set->map.mq_map[cpu] = queue;
+		 * 		set->map[x].mq_map[cpu] = queue;
 		 * }
 		 *
 		 * When we need to remap, the table has to be cleared for
 		 * killing stale mapping since one CPU may not be mapped
 		 * to any hw queue.
 		 */
-		blk_mq_clear_mq_map(&set->map[0]);
+		for (i = 0; i < set->nr_maps; i++)
+			blk_mq_clear_mq_map(&set->map[i]);
 
 		return set->ops->map_queues(set);
-	} else
+	} else {
+		BUG_ON(set->nr_maps > 1);
 		return blk_mq_map_queues(&set->map[0]);
+	}
 }
 
 /*
@@ -2831,7 +2851,7 @@ static int blk_mq_update_queue_map(struct blk_mq_tag_set *set)
  */
 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 {
-	int ret;
+	int i, ret;
 
 	BUILD_BUG_ON(BLK_MQ_MAX_DEPTH > 1 << BLK_MQ_UNIQUE_TAG_BITS);
 
@@ -2854,6 +2874,11 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		set->queue_depth = BLK_MQ_MAX_DEPTH;
 	}
 
+	if (!set->nr_maps)
+		set->nr_maps = 1;
+	else if (set->nr_maps > HCTX_MAX_TYPES)
+		return -EINVAL;
+
 	/*
 	 * If a crashdump is active, then we are potentially in a very
 	 * memory constrained environment. Limit us to 1 queue and
@@ -2875,12 +2900,14 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		return -ENOMEM;
 
 	ret = -ENOMEM;
-	set->map[0].mq_map = kcalloc_node(nr_cpu_ids,
-					  sizeof(*set->map[0].mq_map),
-					  GFP_KERNEL, set->numa_node);
-	if (!set->map[0].mq_map)
-		goto out_free_tags;
-	set->map[0].nr_queues = set->nr_hw_queues;
+	for (i = 0; i < set->nr_maps; i++) {
+		set->map[i].mq_map = kcalloc_node(nr_cpu_ids,
+						  sizeof(struct blk_mq_queue_map),
+						  GFP_KERNEL, set->numa_node);
+		if (!set->map[i].mq_map)
+			goto out_free_mq_map;
+		set->map[i].nr_queues = set->nr_hw_queues;
+	}
 
 	ret = blk_mq_update_queue_map(set);
 	if (ret)
@@ -2896,9 +2923,10 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 	return 0;
 
 out_free_mq_map:
-	kfree(set->map[0].mq_map);
-	set->map[0].mq_map = NULL;
-out_free_tags:
+	for (i = 0; i < set->nr_maps; i++) {
+		kfree(set->map[i].mq_map);
+		set->map[i].mq_map = NULL;
+	}
 	kfree(set->tags);
 	set->tags = NULL;
 	return ret;
@@ -2907,13 +2935,15 @@ EXPORT_SYMBOL(blk_mq_alloc_tag_set);
 
 void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
 {
-	int i;
+	int i, j;
 
 	for (i = 0; i < nr_cpu_ids; i++)
 		blk_mq_free_map_and_requests(set, i);
 
-	kfree(set->map[0].mq_map);
-	set->map[0].mq_map = NULL;
+	for (j = 0; j < set->nr_maps; j++) {
+		kfree(set->map[j].mq_map);
+		set->map[j].mq_map = NULL;
+	}
 
 	kfree(set->tags);
 	set->tags = NULL;
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 1821f448f7c4..8329017badc8 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -72,20 +72,37 @@ void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
  */
 extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
 
-static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
-						     unsigned int flags,
-						     unsigned int cpu)
+/*
+ * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue
+ * @q: request queue
+ * @hctx_type: the hctx type index
+ * @cpu: CPU
+ */
+static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
+							  unsigned int hctx_type,
+							  unsigned int cpu)
 {
 	struct blk_mq_tag_set *set = q->tag_set;
 
-	return q->queue_hw_ctx[set->map[0].mq_map[cpu]];
+	return q->queue_hw_ctx[set->map[hctx_type].mq_map[cpu]];
 }
 
-static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
-							  unsigned int hctx_type,
-							  unsigned int cpu)
+/*
+ * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue
+ * @q: request queue
+ * @flags: request command flags
+ * @cpu: CPU
+ */
+static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
+						     unsigned int flags,
+						     unsigned int cpu)
 {
-	return blk_mq_map_queue(q, hctx_type, cpu);
+	int hctx_type = 0;
+
+	if (q->mq_ops->flags_to_type)
+		hctx_type = q->mq_ops->flags_to_type(q, flags);
+
+	return blk_mq_map_queue_type(q, hctx_type, cpu);
 }
 
 /*
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 466b9202b69c..26768c8f5af5 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -85,7 +85,14 @@ enum {
 };
 
 struct blk_mq_tag_set {
+	/*
+	 * map[] holds ctx -> hctx mappings, one map exists for each type
+	 * that the driver wishes to support. There are no restrictions
+	 * on maps being of the same size, and it's perfectly legal to
+	 * share maps between types.
+	 */
 	struct blk_mq_queue_map	map[HCTX_MAX_TYPES];
+	unsigned int		nr_maps;	/* nr entries in map[] */
 	const struct blk_mq_ops	*ops;
 	unsigned int		nr_hw_queues;	/* nr hw queues across maps */
 	unsigned int		queue_depth;	/* max hw supported */
@@ -109,6 +116,8 @@ struct blk_mq_queue_data {
 
 typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *,
 		const struct blk_mq_queue_data *);
+/* takes rq->cmd_flags as input, returns a hardware type index */
+typedef int (flags_to_type_fn)(struct request_queue *, unsigned int);
 typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *);
 typedef void (put_budget_fn)(struct blk_mq_hw_ctx *);
 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
@@ -133,6 +142,11 @@ struct blk_mq_ops {
 	 */
 	queue_rq_fn		*queue_rq;
 
+	/*
+	 * Return a queue map type for the given request/bio flags
+	 */
+	flags_to_type_fn	*flags_to_type;
+
 	/*
 	 * Reserve budget before queue request, once .queue_rq is
 	 * run, it is driver's responsibility to release the
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 08/16] blk-mq: separate number of hardware queues from nr_cpu_ids
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (6 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 07/16] blk-mq: support multiple hctx maps Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:00   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 09/16] blk-mq: cache request hardware queue mapping Jens Axboe
                   ` (8 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

With multiple maps, nr_cpu_ids is no longer the maximum number of
hardware queues we support on a given devices. The initializer of
the tag_set can have set ->nr_hw_queues larger than the available
number of CPUs, since we can exceed that with multiple queue maps.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c | 28 +++++++++++++++++++++-------
 1 file changed, 21 insertions(+), 7 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 9d6e2f6f8ee9..1ca48cf3bbc7 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2668,6 +2668,19 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,
 	mutex_unlock(&q->sysfs_lock);
 }
 
+/*
+ * Maximum number of hardware queues we support. For single sets, we'll never
+ * have more than the CPUs (software queues). For multiple sets, the tag_set
+ * user may have set ->nr_hw_queues larger.
+ */
+static unsigned int nr_hw_queues(struct blk_mq_tag_set *set)
+{
+	if (set->nr_maps == 1)
+		return nr_cpu_ids;
+
+	return max(set->nr_hw_queues, nr_cpu_ids);
+}
+
 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 						  struct request_queue *q)
 {
@@ -2687,7 +2700,8 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 	/* init q->mq_kobj and sw queues' kobjects */
 	blk_mq_sysfs_init(q);
 
-	q->queue_hw_ctx = kcalloc_node(nr_cpu_ids, sizeof(*(q->queue_hw_ctx)),
+	q->nr_queues = nr_hw_queues(set);
+	q->queue_hw_ctx = kcalloc_node(q->nr_queues, sizeof(*(q->queue_hw_ctx)),
 						GFP_KERNEL, set->numa_node);
 	if (!q->queue_hw_ctx)
 		goto err_percpu;
@@ -2699,7 +2713,6 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 	INIT_WORK(&q->timeout_work, blk_mq_timeout_work);
 	blk_queue_rq_timeout(q, set->timeout ? set->timeout : 30 * HZ);
 
-	q->nr_queues = nr_cpu_ids;
 	q->tag_set = set;
 
 	q->queue_flags |= QUEUE_FLAG_MQ_DEFAULT;
@@ -2889,12 +2902,13 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		set->queue_depth = min(64U, set->queue_depth);
 	}
 	/*
-	 * There is no use for more h/w queues than cpus.
+	 * There is no use for more h/w queues than cpus if we just have
+	 * a single map
 	 */
-	if (set->nr_hw_queues > nr_cpu_ids)
+	if (set->nr_maps == 1 && set->nr_hw_queues > nr_cpu_ids)
 		set->nr_hw_queues = nr_cpu_ids;
 
-	set->tags = kcalloc_node(nr_cpu_ids, sizeof(struct blk_mq_tags *),
+	set->tags = kcalloc_node(nr_hw_queues(set), sizeof(struct blk_mq_tags *),
 				 GFP_KERNEL, set->numa_node);
 	if (!set->tags)
 		return -ENOMEM;
@@ -2937,7 +2951,7 @@ void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
 {
 	int i, j;
 
-	for (i = 0; i < nr_cpu_ids; i++)
+	for (i = 0; i < nr_hw_queues(set); i++)
 		blk_mq_free_map_and_requests(set, i);
 
 	for (j = 0; j < set->nr_maps; j++) {
@@ -3069,7 +3083,7 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 
 	lockdep_assert_held(&set->tag_list_lock);
 
-	if (nr_hw_queues > nr_cpu_ids)
+	if (set->nr_maps == 1 && nr_hw_queues > nr_cpu_ids)
 		nr_hw_queues = nr_cpu_ids;
 	if (nr_hw_queues < 1 || nr_hw_queues == set->nr_hw_queues)
 		return;
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 09/16] blk-mq: cache request hardware queue mapping
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (7 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 08/16] blk-mq: separate number of hardware queues from nr_cpu_ids Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:01   ` Sagi Grimberg
  2018-11-01  9:27   ` Hannes Reinecke
  2018-10-30 18:32 ` [PATCH 10/16] blk-mq: cleanup and improve list insertion Jens Axboe
                   ` (7 subsequent siblings)
  16 siblings, 2 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

We call blk_mq_map_queue() a lot, at least two times for each
request per IO, sometimes more. Since we now have an indirect
call as well in that function. cache the mapping so we don't
have to re-call blk_mq_map_queue() for the same request
multiple times.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-flush.c      | 12 ++++--------
 block/blk-mq-debugfs.c |  4 +---
 block/blk-mq-sched.c   |  6 ++----
 block/blk-mq-tag.c     |  9 +--------
 block/blk-mq.c         | 22 +++++++++-------------
 block/blk-mq.h         |  5 +----
 include/linux/blkdev.h |  1 +
 7 files changed, 19 insertions(+), 40 deletions(-)

diff --git a/block/blk-flush.c b/block/blk-flush.c
index 7922dba81497..2ff590b31a9d 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -219,7 +219,7 @@ static void flush_end_io(struct request *flush_rq, blk_status_t error)
 
 	/* release the tag's ownership to the req cloned from */
 	spin_lock_irqsave(&fq->mq_flush_lock, flags);
-	hctx = blk_mq_map_queue(q, flush_rq->cmd_flags, flush_rq->mq_ctx->cpu);
+	hctx = flush_rq->mq_hctx;
 	if (!q->elevator) {
 		blk_mq_tag_set_rq(hctx, flush_rq->tag, fq->orig_rq);
 		flush_rq->tag = -1;
@@ -268,7 +268,6 @@ static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq,
 	struct request *first_rq =
 		list_first_entry(pending, struct request, flush.list);
 	struct request *flush_rq = fq->flush_rq;
-	struct blk_mq_hw_ctx *hctx;
 
 	/* C1 described at the top of this file */
 	if (fq->flush_pending_idx != fq->flush_running_idx || list_empty(pending))
@@ -303,13 +302,12 @@ static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq,
 	 * just for cheating put/get driver tag.
 	 */
 	flush_rq->mq_ctx = first_rq->mq_ctx;
+	flush_rq->mq_hctx = first_rq->mq_hctx;
 
 	if (!q->elevator) {
 		fq->orig_rq = first_rq;
 		flush_rq->tag = first_rq->tag;
-		hctx = blk_mq_map_queue(q, first_rq->cmd_flags,
-					first_rq->mq_ctx->cpu);
-		blk_mq_tag_set_rq(hctx, first_rq->tag, flush_rq);
+		blk_mq_tag_set_rq(flush_rq->mq_hctx, first_rq->tag, flush_rq);
 	} else {
 		flush_rq->internal_tag = first_rq->internal_tag;
 	}
@@ -326,13 +324,11 @@ static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq,
 static void mq_flush_data_end_io(struct request *rq, blk_status_t error)
 {
 	struct request_queue *q = rq->q;
-	struct blk_mq_hw_ctx *hctx;
+	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
 	unsigned long flags;
 	struct blk_flush_queue *fq = blk_get_flush_queue(q, ctx);
 
-	hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
-
 	if (q->elevator) {
 		WARN_ON(rq->tag < 0);
 		blk_mq_put_driver_tag_hctx(hctx, rq);
diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index fac70c81b7de..cde19be36135 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -427,10 +427,8 @@ struct show_busy_params {
 static void hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
 {
 	const struct show_busy_params *params = data;
-	struct blk_mq_hw_ctx *hctx;
 
-	hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu);
-	if (hctx == params->hctx)
+	if (rq->mq_hctx == params->hctx)
 		__blk_mq_debugfs_rq_show(params->m,
 					 list_entry_rq(&rq->queuelist));
 }
diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index d232ecf3290c..8bc1f37acca2 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -367,9 +367,7 @@ void blk_mq_sched_insert_request(struct request *rq, bool at_head,
 	struct request_queue *q = rq->q;
 	struct elevator_queue *e = q->elevator;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx;
-
-	hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 
 	/* flush rq in flush machinery need to be dispatched directly */
 	if (!(rq->rq_flags & RQF_FLUSH_SEQ) && op_is_flush(rq->cmd_flags)) {
@@ -408,7 +406,7 @@ void blk_mq_sched_insert_requests(struct request_queue *q,
 
 	/* For list inserts, requests better be on the same hw queue */
 	rq = list_first_entry(list, struct request, queuelist);
-	hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
+	hctx = rq->mq_hctx;
 
 	e = hctx->queue->elevator;
 	if (e && e->type->ops.mq.insert_requests)
diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index 478a959357f5..fb836d818b80 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -527,14 +527,7 @@ int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx,
  */
 u32 blk_mq_unique_tag(struct request *rq)
 {
-	struct request_queue *q = rq->q;
-	struct blk_mq_hw_ctx *hctx;
-	int hwq = 0;
-
-	hctx = blk_mq_map_queue(q, rq->cmd_flags, rq->mq_ctx->cpu);
-	hwq = hctx->queue_num;
-
-	return (hwq << BLK_MQ_UNIQUE_TAG_BITS) |
+	return (rq->mq_hctx->queue_num << BLK_MQ_UNIQUE_TAG_BITS) |
 		(rq->tag & BLK_MQ_UNIQUE_TAG_MASK);
 }
 EXPORT_SYMBOL(blk_mq_unique_tag);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 1ca48cf3bbc7..b86d725958d3 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -300,6 +300,7 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
 	/* csd/requeue_work/fifo_time is initialized before use */
 	rq->q = data->q;
 	rq->mq_ctx = data->ctx;
+	rq->mq_hctx = data->hctx;
 	rq->rq_flags = rq_flags;
 	rq->cpu = -1;
 	rq->cmd_flags = op;
@@ -473,10 +474,11 @@ static void __blk_mq_free_request(struct request *rq)
 {
 	struct request_queue *q = rq->q;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 	const int sched_tag = rq->internal_tag;
 
 	blk_pm_mark_last_busy(rq);
+	rq->mq_hctx = NULL;
 	if (rq->tag != -1)
 		blk_mq_put_tag(hctx, hctx->tags, ctx, rq->tag);
 	if (sched_tag != -1)
@@ -490,7 +492,7 @@ void blk_mq_free_request(struct request *rq)
 	struct request_queue *q = rq->q;
 	struct elevator_queue *e = q->elevator;
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, rq->cmd_flags, ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 
 	if (rq->rq_flags & RQF_ELVPRIV) {
 		if (e && e->type->ops.mq.finish_request)
@@ -982,7 +984,7 @@ bool blk_mq_get_driver_tag(struct request *rq)
 {
 	struct blk_mq_alloc_data data = {
 		.q = rq->q,
-		.hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu),
+		.hctx = rq->mq_hctx,
 		.flags = BLK_MQ_REQ_NOWAIT,
 		.cmd_flags = rq->cmd_flags,
 	};
@@ -1148,7 +1150,7 @@ bool blk_mq_dispatch_rq_list(struct request_queue *q, struct list_head *list,
 
 		rq = list_first_entry(list, struct request, queuelist);
 
-		hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu);
+		hctx = rq->mq_hctx;
 		if (!got_budget && !blk_mq_get_dispatch_budget(hctx))
 			break;
 
@@ -1578,9 +1580,7 @@ void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
  */
 void blk_mq_request_bypass_insert(struct request *rq, bool run_queue)
 {
-	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, rq->cmd_flags,
-							ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 
 	spin_lock(&hctx->lock);
 	list_add_tail(&rq->queuelist, &hctx->dispatch);
@@ -1789,9 +1789,7 @@ blk_status_t blk_mq_request_issue_directly(struct request *rq)
 	blk_status_t ret;
 	int srcu_idx;
 	blk_qc_t unused_cookie;
-	struct blk_mq_ctx *ctx = rq->mq_ctx;
-	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, rq->cmd_flags,
-							ctx->cpu);
+	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 
 	hctx_lock(hctx, &srcu_idx);
 	ret = __blk_mq_try_issue_directly(hctx, rq, &unused_cookie, true);
@@ -1916,9 +1914,7 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
 		blk_mq_put_ctx(data.ctx);
 
 		if (same_queue_rq) {
-			data.hctx = blk_mq_map_queue(q,
-					same_queue_rq->cmd_flags,
-					same_queue_rq->mq_ctx->cpu);
+			data.hctx = same_queue_rq->mq_hctx;
 			blk_mq_try_issue_directly(data.hctx, same_queue_rq,
 					&cookie);
 		}
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 8329017badc8..74cb2f524824 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -223,13 +223,10 @@ static inline void blk_mq_put_driver_tag_hctx(struct blk_mq_hw_ctx *hctx,
 
 static inline void blk_mq_put_driver_tag(struct request *rq)
 {
-	struct blk_mq_hw_ctx *hctx;
-
 	if (rq->tag == -1 || rq->internal_tag == -1)
 		return;
 
-	hctx = blk_mq_map_queue(rq->q, rq->cmd_flags, rq->mq_ctx->cpu);
-	__blk_mq_put_driver_tag(hctx, rq);
+	__blk_mq_put_driver_tag(rq->mq_hctx, rq);
 }
 
 static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap)
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 4223ae2d2198..7b351210ebcd 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -129,6 +129,7 @@ enum mq_rq_state {
 struct request {
 	struct request_queue *q;
 	struct blk_mq_ctx *mq_ctx;
+	struct blk_mq_hw_ctx *mq_hctx;
 
 	int cpu;
 	unsigned int cmd_flags;		/* op and common flags */
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 10/16] blk-mq: cleanup and improve list insertion
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (8 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 09/16] blk-mq: cache request hardware queue mapping Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:03   ` Sagi Grimberg
  2018-11-01  9:28   ` Hannes Reinecke
  2018-10-30 18:32 ` [PATCH 11/16] blk-mq: improve plug list sorting Jens Axboe
                   ` (6 subsequent siblings)
  16 siblings, 2 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

It's somewhat strange to have a list insertion function that
relies on the fact that the caller has mapped things correctly.
Pass in the hardware queue directly for insertion, which makes
for a much cleaner interface and implementation.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq-sched.c |  8 +-------
 block/blk-mq-sched.h |  2 +-
 block/blk-mq.c       | 25 ++++++++++++++-----------
 3 files changed, 16 insertions(+), 19 deletions(-)

diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
index 8bc1f37acca2..6e7375246e2f 100644
--- a/block/blk-mq-sched.c
+++ b/block/blk-mq-sched.c
@@ -396,17 +396,11 @@ void blk_mq_sched_insert_request(struct request *rq, bool at_head,
 		blk_mq_run_hw_queue(hctx, async);
 }
 
-void blk_mq_sched_insert_requests(struct request_queue *q,
+void blk_mq_sched_insert_requests(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx,
 				  struct list_head *list, bool run_queue_async)
 {
-	struct blk_mq_hw_ctx *hctx;
 	struct elevator_queue *e;
-	struct request *rq;
-
-	/* For list inserts, requests better be on the same hw queue */
-	rq = list_first_entry(list, struct request, queuelist);
-	hctx = rq->mq_hctx;
 
 	e = hctx->queue->elevator;
 	if (e && e->type->ops.mq.insert_requests)
diff --git a/block/blk-mq-sched.h b/block/blk-mq-sched.h
index 8a9544203173..ffd7b5989d63 100644
--- a/block/blk-mq-sched.h
+++ b/block/blk-mq-sched.h
@@ -19,7 +19,7 @@ void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx);
 
 void blk_mq_sched_insert_request(struct request *rq, bool at_head,
 				 bool run_queue, bool async);
-void blk_mq_sched_insert_requests(struct request_queue *q,
+void blk_mq_sched_insert_requests(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx,
 				  struct list_head *list, bool run_queue_async);
 
diff --git a/block/blk-mq.c b/block/blk-mq.c
index b86d725958d3..51b8166959b9 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1623,11 +1623,12 @@ static int plug_ctx_cmp(void *priv, struct list_head *a, struct list_head *b)
 
 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
 {
+	struct blk_mq_hw_ctx *this_hctx;
 	struct blk_mq_ctx *this_ctx;
 	struct request_queue *this_q;
 	struct request *rq;
 	LIST_HEAD(list);
-	LIST_HEAD(ctx_list);
+	LIST_HEAD(rq_list);
 	unsigned int depth;
 
 	list_splice_init(&plug->mq_list, &list);
@@ -1635,6 +1636,7 @@ void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
 	list_sort(NULL, &list, plug_ctx_cmp);
 
 	this_q = NULL;
+	this_hctx = NULL;
 	this_ctx = NULL;
 	depth = 0;
 
@@ -1642,30 +1644,31 @@ void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
 		rq = list_entry_rq(list.next);
 		list_del_init(&rq->queuelist);
 		BUG_ON(!rq->q);
-		if (rq->mq_ctx != this_ctx) {
-			if (this_ctx) {
+		if (rq->mq_hctx != this_hctx || rq->mq_ctx != this_ctx) {
+			if (this_hctx) {
 				trace_block_unplug(this_q, depth, !from_schedule);
-				blk_mq_sched_insert_requests(this_q, this_ctx,
-								&ctx_list,
+				blk_mq_sched_insert_requests(this_hctx, this_ctx,
+								&rq_list,
 								from_schedule);
 			}
 
-			this_ctx = rq->mq_ctx;
 			this_q = rq->q;
+			this_ctx = rq->mq_ctx;
+			this_hctx = rq->mq_hctx;
 			depth = 0;
 		}
 
 		depth++;
-		list_add_tail(&rq->queuelist, &ctx_list);
+		list_add_tail(&rq->queuelist, &rq_list);
 	}
 
 	/*
-	 * If 'this_ctx' is set, we know we have entries to complete
-	 * on 'ctx_list'. Do those.
+	 * If 'this_hctx' is set, we know we have entries to complete
+	 * on 'rq_list'. Do those.
 	 */
-	if (this_ctx) {
+	if (this_hctx) {
 		trace_block_unplug(this_q, depth, !from_schedule);
-		blk_mq_sched_insert_requests(this_q, this_ctx, &ctx_list,
+		blk_mq_sched_insert_requests(this_hctx, this_ctx, &rq_list,
 						from_schedule);
 	}
 }
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 11/16] blk-mq: improve plug list sorting
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (9 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 10/16] blk-mq: cleanup and improve list insertion Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:04   ` Sagi Grimberg
  2018-11-01  9:30   ` Hannes Reinecke
  2018-10-30 18:32 ` [PATCH 12/16] blk-mq: initial support for multiple queue maps Jens Axboe
                   ` (5 subsequent siblings)
  16 siblings, 2 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

Currently we only look at the software queue, but with support
for multiple maps, we should also look at the hardware queue.
This is important since we'll flush out the request list if
either the software queue or hardware queue don't match.

This sorts by software queue first, then hardware queue if
that differs. Finally we sort by request location like before.
This minimizes the flush points per plug list.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 51b8166959b9..5a34c9374dc7 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1611,14 +1611,21 @@ void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx,
 	spin_unlock(&ctx->lock);
 }
 
-static int plug_ctx_cmp(void *priv, struct list_head *a, struct list_head *b)
+static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
 {
 	struct request *rqa = container_of(a, struct request, queuelist);
 	struct request *rqb = container_of(b, struct request, queuelist);
 
-	return !(rqa->mq_ctx < rqb->mq_ctx ||
-		 (rqa->mq_ctx == rqb->mq_ctx &&
-		  blk_rq_pos(rqa) < blk_rq_pos(rqb)));
+	if (rqa->mq_ctx < rqb->mq_ctx)
+		return -1;
+	else if (rqa->mq_ctx > rqb->mq_ctx)
+		return 1;
+	else if (rqa->mq_hctx < rqb->mq_hctx)
+		return -1;
+	else if (rqa->mq_hctx > rqb->mq_hctx)
+		return 1;
+
+	return blk_rq_pos(rqa) > blk_rq_pos(rqb);
 }
 
 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
@@ -1633,7 +1640,7 @@ void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
 
 	list_splice_init(&plug->mq_list, &list);
 
-	list_sort(NULL, &list, plug_ctx_cmp);
+	list_sort(NULL, &list, plug_rq_cmp);
 
 	this_q = NULL;
 	this_hctx = NULL;
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 12/16] blk-mq: initial support for multiple queue maps
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (10 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 11/16] blk-mq: improve plug list sorting Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:14   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs Jens Axboe
                   ` (4 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

Add a queue offset to the tag map. This enables users to map
iteratively, for each queue map type they support.

Bump maximum number of supported maps to 2, we're now fully
able to support more than 1 map.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq-cpumap.c  | 9 +++++----
 block/blk-mq-pci.c     | 2 +-
 block/blk-mq-virtio.c  | 2 +-
 include/linux/blk-mq.h | 3 ++-
 4 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/block/blk-mq-cpumap.c b/block/blk-mq-cpumap.c
index 6e6686c55984..03a534820271 100644
--- a/block/blk-mq-cpumap.c
+++ b/block/blk-mq-cpumap.c
@@ -14,9 +14,10 @@
 #include "blk.h"
 #include "blk-mq.h"
 
-static int cpu_to_queue_index(unsigned int nr_queues, const int cpu)
+static int cpu_to_queue_index(struct blk_mq_queue_map *qmap,
+			      unsigned int nr_queues, const int cpu)
 {
-	return cpu % nr_queues;
+	return qmap->queue_offset + (cpu % nr_queues);
 }
 
 static int get_first_sibling(unsigned int cpu)
@@ -44,11 +45,11 @@ int blk_mq_map_queues(struct blk_mq_queue_map *qmap)
 		 * performace optimizations.
 		 */
 		if (cpu < nr_queues) {
-			map[cpu] = cpu_to_queue_index(nr_queues, cpu);
+			map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu);
 		} else {
 			first_sibling = get_first_sibling(cpu);
 			if (first_sibling == cpu)
-				map[cpu] = cpu_to_queue_index(nr_queues, cpu);
+				map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu);
 			else
 				map[cpu] = map[first_sibling];
 		}
diff --git a/block/blk-mq-pci.c b/block/blk-mq-pci.c
index 40333d60a850..1dce18553984 100644
--- a/block/blk-mq-pci.c
+++ b/block/blk-mq-pci.c
@@ -43,7 +43,7 @@ int blk_mq_pci_map_queues(struct blk_mq_queue_map *qmap, struct pci_dev *pdev,
 			goto fallback;
 
 		for_each_cpu(cpu, mask)
-			qmap->mq_map[cpu] = queue;
+			qmap->mq_map[cpu] = qmap->queue_offset + queue;
 	}
 
 	return 0;
diff --git a/block/blk-mq-virtio.c b/block/blk-mq-virtio.c
index 661fbfef480f..370827163835 100644
--- a/block/blk-mq-virtio.c
+++ b/block/blk-mq-virtio.c
@@ -44,7 +44,7 @@ int blk_mq_virtio_map_queues(struct blk_mq_queue_map *qmap,
 			goto fallback;
 
 		for_each_cpu(cpu, mask)
-			qmap->mq_map[cpu] = queue;
+			qmap->mq_map[cpu] = qmap->queue_offset + queue;
 	}
 
 	return 0;
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 26768c8f5af5..8e80d5043079 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -78,10 +78,11 @@ struct blk_mq_hw_ctx {
 struct blk_mq_queue_map {
 	unsigned int *mq_map;
 	unsigned int nr_queues;
+	unsigned int queue_offset;
 };
 
 enum {
-	HCTX_MAX_TYPES = 1,
+	HCTX_MAX_TYPES = 2,
 };
 
 struct blk_mq_tag_set {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (11 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 12/16] blk-mq: initial support for multiple queue maps Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:17   ` Sagi Grimberg
  2018-11-02 14:37   ` Ming Lei
  2018-10-30 18:32 ` [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes Jens Axboe
                   ` (3 subsequent siblings)
  16 siblings, 2 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe, Thomas Gleixner

A driver may have a need to allocate multiple sets of MSI/MSI-X
interrupts, and have them appropriately affinitized. Add support for
defining a number of sets in the irq_affinity structure, of varying
sizes, and get each set affinitized correctly across the machine.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 drivers/pci/msi.c         | 14 ++++++++++++++
 include/linux/interrupt.h |  4 ++++
 kernel/irq/affinity.c     | 40 ++++++++++++++++++++++++++++++---------
 3 files changed, 49 insertions(+), 9 deletions(-)

diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
index af24ed50a245..e6c6e10b9ceb 100644
--- a/drivers/pci/msi.c
+++ b/drivers/pci/msi.c
@@ -1036,6 +1036,13 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
 	if (maxvec < minvec)
 		return -ERANGE;
 
+	/*
+	 * If the caller is passing in sets, we can't support a range of
+	 * vectors. The caller needs to handle that.
+	 */
+	if (affd->nr_sets && minvec != maxvec)
+		return -EINVAL;
+
 	if (WARN_ON_ONCE(dev->msi_enabled))
 		return -EINVAL;
 
@@ -1087,6 +1094,13 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
 	if (maxvec < minvec)
 		return -ERANGE;
 
+	/*
+	 * If the caller is passing in sets, we can't support a range of
+	 * supported vectors. The caller needs to handle that.
+	 */
+	if (affd->nr_sets && minvec != maxvec)
+		return -EINVAL;
+
 	if (WARN_ON_ONCE(dev->msix_enabled))
 		return -EINVAL;
 
diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
index 1d6711c28271..ca397ff40836 100644
--- a/include/linux/interrupt.h
+++ b/include/linux/interrupt.h
@@ -247,10 +247,14 @@ struct irq_affinity_notify {
  *			the MSI(-X) vector space
  * @post_vectors:	Don't apply affinity to @post_vectors at end of
  *			the MSI(-X) vector space
+ * @nr_sets:		Length of passed in *sets array
+ * @sets:		Number of affinitized sets
  */
 struct irq_affinity {
 	int	pre_vectors;
 	int	post_vectors;
+	int	nr_sets;
+	int	*sets;
 };
 
 #if defined(CONFIG_SMP)
diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c
index f4f29b9d90ee..2046a0f0f0f1 100644
--- a/kernel/irq/affinity.c
+++ b/kernel/irq/affinity.c
@@ -180,6 +180,7 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
 	int curvec, usedvecs;
 	cpumask_var_t nmsk, npresmsk, *node_to_cpumask;
 	struct cpumask *masks = NULL;
+	int i, nr_sets;
 
 	/*
 	 * If there aren't any vectors left after applying the pre/post
@@ -210,10 +211,23 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
 	get_online_cpus();
 	build_node_to_cpumask(node_to_cpumask);
 
-	/* Spread on present CPUs starting from affd->pre_vectors */
-	usedvecs = irq_build_affinity_masks(affd, curvec, affvecs,
-					    node_to_cpumask, cpu_present_mask,
-					    nmsk, masks);
+	/*
+	 * Spread on present CPUs starting from affd->pre_vectors. If we
+	 * have multiple sets, build each sets affinity mask separately.
+	 */
+	nr_sets = affd->nr_sets;
+	if (!nr_sets)
+		nr_sets = 1;
+
+	for (i = 0, usedvecs = 0; i < nr_sets; i++) {
+		int this_vecs = affd->sets ? affd->sets[i] : affvecs;
+		int nr;
+
+		nr = irq_build_affinity_masks(affd, curvec, this_vecs,
+					      node_to_cpumask, cpu_present_mask,
+					      nmsk, masks + usedvecs);
+		usedvecs += nr;
+	}
 
 	/*
 	 * Spread on non present CPUs starting from the next vector to be
@@ -258,13 +272,21 @@ int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity
 {
 	int resv = affd->pre_vectors + affd->post_vectors;
 	int vecs = maxvec - resv;
-	int ret;
+	int set_vecs;
 
 	if (resv > minvec)
 		return 0;
 
-	get_online_cpus();
-	ret = min_t(int, cpumask_weight(cpu_possible_mask), vecs) + resv;
-	put_online_cpus();
-	return ret;
+	if (affd->nr_sets) {
+		int i;
+
+		for (i = 0, set_vecs = 0;  i < affd->nr_sets; i++)
+			set_vecs += affd->sets[i];
+	} else {
+		get_online_cpus();
+		set_vecs = cpumask_weight(cpu_possible_mask);
+		put_online_cpus();
+	}
+
+	return resv + min(set_vecs, vecs);
 }
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (12 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:57   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 15/16] block: add REQ_HIPRI and inherit it from IOCB_HIPRI Jens Axboe
                   ` (2 subsequent siblings)
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

NVMe does round-robin between queues by default, which means that
sharing a queue map for both reads and writes can be problematic
in terms of read servicing. It's much easier to flood the queue
with writes and reduce the read servicing.

Implement two queue maps, one for reads and one for writes. The
write queue count is configurable through the 'write_queues'
parameter.

By default, we retain the previous behavior of having a single
queue set, shared between reads and writes. Setting 'write_queues'
to a non-zero value will create two queue sets, one for reads and
one for writes, the latter using the configurable number of
queues (hardware queue counts permitting).

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 drivers/nvme/host/pci.c | 174 +++++++++++++++++++++++++++++++++++++---
 1 file changed, 162 insertions(+), 12 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index e5d783cb6937..17170686105f 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -74,11 +74,29 @@ static int io_queue_depth = 1024;
 module_param_cb(io_queue_depth, &io_queue_depth_ops, &io_queue_depth, 0644);
 MODULE_PARM_DESC(io_queue_depth, "set io queue depth, should >= 2");
 
+static int queue_count_set(const char *val, const struct kernel_param *kp);
+static const struct kernel_param_ops queue_count_ops = {
+	.set = queue_count_set,
+	.get = param_get_int,
+};
+
+static int write_queues;
+module_param_cb(write_queues, &queue_count_ops, &write_queues, 0644);
+MODULE_PARM_DESC(write_queues,
+	"Number of queues to use for writes. If not set, reads and writes "
+	"will share a queue set.");
+
 struct nvme_dev;
 struct nvme_queue;
 
 static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
 
+enum {
+	NVMEQ_TYPE_READ,
+	NVMEQ_TYPE_WRITE,
+	NVMEQ_TYPE_NR,
+};
+
 /*
  * Represents an NVM Express device.  Each nvme_dev is a PCI function.
  */
@@ -92,6 +110,7 @@ struct nvme_dev {
 	struct dma_pool *prp_small_pool;
 	unsigned online_queues;
 	unsigned max_qid;
+	unsigned io_queues[NVMEQ_TYPE_NR];
 	unsigned int num_vecs;
 	int q_depth;
 	u32 db_stride;
@@ -134,6 +153,17 @@ static int io_queue_depth_set(const char *val, const struct kernel_param *kp)
 	return param_set_int(val, kp);
 }
 
+static int queue_count_set(const char *val, const struct kernel_param *kp)
+{
+	int n = 0, ret;
+
+	ret = kstrtoint(val, 10, &n);
+	if (n > num_possible_cpus())
+		n = num_possible_cpus();
+
+	return param_set_int(val, kp);
+}
+
 static inline unsigned int sq_idx(unsigned int qid, u32 stride)
 {
 	return qid * 2 * stride;
@@ -218,9 +248,20 @@ static inline void _nvme_check_size(void)
 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
 }
 
+static unsigned int max_io_queues(void)
+{
+	return num_possible_cpus() + write_queues;
+}
+
+static unsigned int max_queue_count(void)
+{
+	/* IO queues + admin queue */
+	return 1 + max_io_queues();
+}
+
 static inline unsigned int nvme_dbbuf_size(u32 stride)
 {
-	return ((num_possible_cpus() + 1) * 8 * stride);
+	return (max_queue_count() * 8 * stride);
 }
 
 static int nvme_dbbuf_dma_alloc(struct nvme_dev *dev)
@@ -431,12 +472,41 @@ static int nvme_init_request(struct blk_mq_tag_set *set, struct request *req,
 	return 0;
 }
 
+static int queue_irq_offset(struct nvme_dev *dev)
+{
+	/* if we have more than 1 vec, admin queue offsets us 1 */
+	if (dev->num_vecs > 1)
+		return 1;
+
+	return 0;
+}
+
 static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
 {
 	struct nvme_dev *dev = set->driver_data;
+	int i, qoff, offset;
+
+	offset = queue_irq_offset(dev);
+	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
+		struct blk_mq_queue_map *map = &set->map[i];
 
-	return blk_mq_pci_map_queues(&set->map[0], to_pci_dev(dev->dev),
-			dev->num_vecs > 1 ? 1 /* admin queue */ : 0);
+		map->nr_queues = dev->io_queues[i];
+		if (!map->nr_queues) {
+			BUG_ON(i == NVMEQ_TYPE_READ);
+
+			/* shared set, resuse read set parameters */
+			map->nr_queues = dev->io_queues[NVMEQ_TYPE_READ];
+			qoff = 0;
+			offset = queue_irq_offset(dev);
+		}
+
+		map->queue_offset = qoff;
+		blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset);
+		qoff += map->nr_queues;
+		offset += map->nr_queues;
+	}
+
+	return 0;
 }
 
 /**
@@ -849,6 +919,14 @@ static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
 	return ret;
 }
 
+static int nvme_flags_to_type(struct request_queue *q, unsigned int flags)
+{
+	if ((flags & REQ_OP_MASK) == REQ_OP_READ)
+		return NVMEQ_TYPE_READ;
+
+	return NVMEQ_TYPE_WRITE;
+}
+
 static void nvme_pci_complete_rq(struct request *req)
 {
 	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
@@ -1476,6 +1554,7 @@ static const struct blk_mq_ops nvme_mq_admin_ops = {
 
 static const struct blk_mq_ops nvme_mq_ops = {
 	.queue_rq	= nvme_queue_rq,
+	.flags_to_type	= nvme_flags_to_type,
 	.complete	= nvme_pci_complete_rq,
 	.init_hctx	= nvme_init_hctx,
 	.init_request	= nvme_init_request,
@@ -1888,18 +1967,53 @@ static int nvme_setup_host_mem(struct nvme_dev *dev)
 	return ret;
 }
 
+static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int nr_io_queues)
+{
+	unsigned int this_w_queues = write_queues;
+
+	/*
+	 * Setup read/write queue split
+	 */
+	if (nr_io_queues == 1) {
+		dev->io_queues[NVMEQ_TYPE_READ] = 1;
+		dev->io_queues[NVMEQ_TYPE_WRITE] = 0;
+		return;
+	}
+
+	/*
+	 * If 'write_queues' is set, ensure it leaves room for at least
+	 * one read queue
+	 */
+	if (this_w_queues >= nr_io_queues)
+		this_w_queues = nr_io_queues - 1;
+
+	/*
+	 * If 'write_queues' is set to zero, reads and writes will share
+	 * a queue set.
+	 */
+	if (!this_w_queues) {
+		dev->io_queues[NVMEQ_TYPE_WRITE] = 0;
+		dev->io_queues[NVMEQ_TYPE_READ] = nr_io_queues;
+	} else {
+		dev->io_queues[NVMEQ_TYPE_WRITE] = this_w_queues;
+		dev->io_queues[NVMEQ_TYPE_READ] = nr_io_queues - this_w_queues;
+	}
+}
+
 static int nvme_setup_io_queues(struct nvme_dev *dev)
 {
 	struct nvme_queue *adminq = &dev->queues[0];
 	struct pci_dev *pdev = to_pci_dev(dev->dev);
 	int result, nr_io_queues;
 	unsigned long size;
-
+	int irq_sets[2];
 	struct irq_affinity affd = {
-		.pre_vectors = 1
+		.pre_vectors = 1,
+		.nr_sets = ARRAY_SIZE(irq_sets),
+		.sets = irq_sets,
 	};
 
-	nr_io_queues = num_possible_cpus();
+	nr_io_queues = max_io_queues();
 	result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
 	if (result < 0)
 		return result;
@@ -1934,13 +2048,48 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	 * setting up the full range we need.
 	 */
 	pci_free_irq_vectors(pdev);
-	result = pci_alloc_irq_vectors_affinity(pdev, 1, nr_io_queues + 1,
-			PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
-	if (result <= 0)
-		return -EIO;
+
+	/*
+	 * For irq sets, we have to ask for minvec == maxvec. This passes
+	 * any reduction back to us, so we can adjust our queue counts and
+	 * IRQ vector needs.
+	 */
+	do {
+		nvme_calc_io_queues(dev, nr_io_queues);
+		irq_sets[0] = dev->io_queues[NVMEQ_TYPE_READ];
+		irq_sets[1] = dev->io_queues[NVMEQ_TYPE_WRITE];
+		if (!irq_sets[1])
+			affd.nr_sets = 1;
+
+		/*
+		 * Need IRQs for read+write queues, and one for the admin queue
+		 */
+		nr_io_queues = irq_sets[0] + irq_sets[1] + 1;
+
+		result = pci_alloc_irq_vectors_affinity(pdev, nr_io_queues,
+				nr_io_queues,
+				PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
+
+		/*
+		 * Need to reduce our vec counts
+		 */
+		if (result == -ENOSPC) {
+			nr_io_queues--;
+			if (!nr_io_queues)
+				return result;
+			continue;
+		} else if (result <= 0)
+			return -EIO;
+		break;
+	} while (1);
+
 	dev->num_vecs = result;
 	dev->max_qid = max(result - 1, 1);
 
+	dev_info(dev->ctrl.device, "%d/%d/%d read/write queues\n",
+					dev->io_queues[NVMEQ_TYPE_READ],
+					dev->io_queues[NVMEQ_TYPE_WRITE]);
+
 	/*
 	 * Should investigate if there's a performance win from allocating
 	 * more queues than interrupt vectors; it might allow the submission
@@ -2042,6 +2191,7 @@ static int nvme_dev_add(struct nvme_dev *dev)
 	if (!dev->ctrl.tagset) {
 		dev->tagset.ops = &nvme_mq_ops;
 		dev->tagset.nr_hw_queues = dev->online_queues - 1;
+		dev->tagset.nr_maps = NVMEQ_TYPE_NR;
 		dev->tagset.timeout = NVME_IO_TIMEOUT;
 		dev->tagset.numa_node = dev_to_node(dev->dev);
 		dev->tagset.queue_depth =
@@ -2489,8 +2639,8 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (!dev)
 		return -ENOMEM;
 
-	dev->queues = kcalloc_node(num_possible_cpus() + 1,
-			sizeof(struct nvme_queue), GFP_KERNEL, node);
+	dev->queues = kcalloc_node(max_queue_count(), sizeof(struct nvme_queue),
+					GFP_KERNEL, node);
 	if (!dev->queues)
 		goto free;
 
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 15/16] block: add REQ_HIPRI and inherit it from IOCB_HIPRI
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (13 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-31  1:58   ` Sagi Grimberg
  2018-10-30 18:32 ` [PATCH 16/16] nvme: add separate poll queue map Jens Axboe
  2018-10-30 18:35 ` [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Keith Busch
  16 siblings, 1 reply; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

We use IOCB_HIPRI to poll for IO in the caller instead of scheduling.
This information is not available for (or after) IO submission. The
driver may make different queue choices based on the type of IO, so
make the fact that we will poll for this IO known to the lower layers
as well.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/block_dev.c            | 2 ++
 fs/direct-io.c            | 2 ++
 fs/iomap.c                | 9 ++++++++-
 include/linux/blk_types.h | 4 +++-
 4 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/fs/block_dev.c b/fs/block_dev.c
index 38b8ce05cbc7..8bb8090c57a7 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -232,6 +232,8 @@ __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
 		bio.bi_opf = dio_bio_write_op(iocb);
 		task_io_account_write(ret);
 	}
+	if (iocb->ki_flags & IOCB_HIPRI)
+		bio.bi_opf |= REQ_HIPRI;
 
 	qc = submit_bio(&bio);
 	for (;;) {
diff --git a/fs/direct-io.c b/fs/direct-io.c
index 093fb54cd316..ffb46b7aa5f7 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -1265,6 +1265,8 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	} else {
 		dio->op = REQ_OP_READ;
 	}
+	if (iocb->ki_flags & IOCB_HIPRI)
+		dio->op_flags |= REQ_HIPRI;
 
 	/*
 	 * For AIO O_(D)SYNC writes we need to defer completions to a workqueue
diff --git a/fs/iomap.c b/fs/iomap.c
index ec15cf2ec696..50ad8c8d1dcb 100644
--- a/fs/iomap.c
+++ b/fs/iomap.c
@@ -1554,6 +1554,7 @@ iomap_dio_zero(struct iomap_dio *dio, struct iomap *iomap, loff_t pos,
 		unsigned len)
 {
 	struct page *page = ZERO_PAGE(0);
+	int flags = REQ_SYNC | REQ_IDLE;
 	struct bio *bio;
 
 	bio = bio_alloc(GFP_KERNEL, 1);
@@ -1562,9 +1563,12 @@ iomap_dio_zero(struct iomap_dio *dio, struct iomap *iomap, loff_t pos,
 	bio->bi_private = dio;
 	bio->bi_end_io = iomap_dio_bio_end_io;
 
+	if (dio->iocb->ki_flags & IOCB_HIPRI)
+		flags |= REQ_HIPRI;
+
 	get_page(page);
 	__bio_add_page(bio, page, len, 0);
-	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC | REQ_IDLE);
+	bio_set_op_attrs(bio, REQ_OP_WRITE, flags);
 
 	atomic_inc(&dio->ref);
 	return submit_bio(bio);
@@ -1663,6 +1667,9 @@ iomap_dio_bio_actor(struct inode *inode, loff_t pos, loff_t length,
 				bio_set_pages_dirty(bio);
 		}
 
+		if (dio->iocb->ki_flags & IOCB_HIPRI)
+			bio->bi_opf |= REQ_HIPRI;
+
 		iov_iter_advance(dio->submit.iter, n);
 
 		dio->size += n;
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 093a818c5b68..d6c2558d6b73 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -322,6 +322,8 @@ enum req_flag_bits {
 	/* command specific flags for REQ_OP_WRITE_ZEROES: */
 	__REQ_NOUNMAP,		/* do not free blocks when zeroing */
 
+	__REQ_HIPRI,
+
 	/* for driver use */
 	__REQ_DRV,
 	__REQ_SWAP,		/* swapping request. */
@@ -342,8 +344,8 @@ enum req_flag_bits {
 #define REQ_RAHEAD		(1ULL << __REQ_RAHEAD)
 #define REQ_BACKGROUND		(1ULL << __REQ_BACKGROUND)
 #define REQ_NOWAIT		(1ULL << __REQ_NOWAIT)
-
 #define REQ_NOUNMAP		(1ULL << __REQ_NOUNMAP)
+#define REQ_HIPRI		(1ULL << __REQ_HIPRI)
 
 #define REQ_DRV			(1ULL << __REQ_DRV)
 #define REQ_SWAP		(1ULL << __REQ_SWAP)
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* [PATCH 16/16] nvme: add separate poll queue map
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (14 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 15/16] block: add REQ_HIPRI and inherit it from IOCB_HIPRI Jens Axboe
@ 2018-10-30 18:32 ` Jens Axboe
  2018-10-30 18:35 ` [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Keith Busch
  16 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-30 18:32 UTC (permalink / raw)
  To: linux-block, linux-nvme, linux-kernel; +Cc: Jens Axboe

Adds support for defining a variable number of poll queues, currently
configurable with the 'poll_queues' module parameter. Defaults to
a single poll queue.

And now we finally have poll support without triggering interrupts!

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 drivers/nvme/host/pci.c | 97 +++++++++++++++++++++++++++++++++--------
 include/linux/blk-mq.h  |  2 +-
 2 files changed, 81 insertions(+), 18 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 17170686105f..305d8d3826d7 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -86,6 +86,10 @@ MODULE_PARM_DESC(write_queues,
 	"Number of queues to use for writes. If not set, reads and writes "
 	"will share a queue set.");
 
+static int poll_queues = 1;
+module_param_cb(poll_queues, &queue_count_ops, &poll_queues, 0644);
+MODULE_PARM_DESC(poll_queues, "Number of queues to use for polled IO.");
+
 struct nvme_dev;
 struct nvme_queue;
 
@@ -94,6 +98,7 @@ static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
 enum {
 	NVMEQ_TYPE_READ,
 	NVMEQ_TYPE_WRITE,
+	NVMEQ_TYPE_POLL,
 	NVMEQ_TYPE_NR,
 };
 
@@ -202,6 +207,7 @@ struct nvme_queue {
 	u16 last_cq_head;
 	u16 qid;
 	u8 cq_phase;
+	u8 polled;
 	u32 *dbbuf_sq_db;
 	u32 *dbbuf_cq_db;
 	u32 *dbbuf_sq_ei;
@@ -250,7 +256,7 @@ static inline void _nvme_check_size(void)
 
 static unsigned int max_io_queues(void)
 {
-	return num_possible_cpus() + write_queues;
+	return num_possible_cpus() + write_queues + poll_queues;
 }
 
 static unsigned int max_queue_count(void)
@@ -500,8 +506,15 @@ static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
 			offset = queue_irq_offset(dev);
 		}
 
+		/*
+		 * The poll queue(s) doesn't have an IRQ (and hence IRQ
+		 * affinity), so use the regular blk-mq cpu mapping
+		 */
 		map->queue_offset = qoff;
-		blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset);
+		if (i != NVMEQ_TYPE_POLL)
+			blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset);
+		else
+			blk_mq_map_queues(map);
 		qoff += map->nr_queues;
 		offset += map->nr_queues;
 	}
@@ -892,7 +905,7 @@ static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
 	 * We should not need to do this, but we're still using this to
 	 * ensure we can drain requests on a dying queue.
 	 */
-	if (unlikely(nvmeq->cq_vector < 0))
+	if (unlikely(nvmeq->cq_vector < 0 && !nvmeq->polled))
 		return BLK_STS_IOERR;
 
 	ret = nvme_setup_cmd(ns, req, &cmnd);
@@ -921,6 +934,8 @@ static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
 
 static int nvme_flags_to_type(struct request_queue *q, unsigned int flags)
 {
+	if (flags & REQ_HIPRI)
+		return NVMEQ_TYPE_POLL;
 	if ((flags & REQ_OP_MASK) == REQ_OP_READ)
 		return NVMEQ_TYPE_READ;
 
@@ -1094,7 +1109,10 @@ static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
 		struct nvme_queue *nvmeq, s16 vector)
 {
 	struct nvme_command c;
-	int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;
+	int flags = NVME_QUEUE_PHYS_CONTIG;
+
+	if (vector != -1)
+		flags |= NVME_CQ_IRQ_ENABLED;
 
 	/*
 	 * Note: we (ab)use the fact that the prp fields survive if no data
@@ -1106,7 +1124,10 @@ static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
 	c.create_cq.cqid = cpu_to_le16(qid);
 	c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
 	c.create_cq.cq_flags = cpu_to_le16(flags);
-	c.create_cq.irq_vector = cpu_to_le16(vector);
+	if (vector != -1)
+		c.create_cq.irq_vector = cpu_to_le16(vector);
+	else
+		c.create_cq.irq_vector = 0;
 
 	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
 }
@@ -1348,13 +1369,14 @@ static int nvme_suspend_queue(struct nvme_queue *nvmeq)
 	int vector;
 
 	spin_lock_irq(&nvmeq->cq_lock);
-	if (nvmeq->cq_vector == -1) {
+	if (nvmeq->cq_vector == -1 && !nvmeq->polled) {
 		spin_unlock_irq(&nvmeq->cq_lock);
 		return 1;
 	}
 	vector = nvmeq->cq_vector;
 	nvmeq->dev->online_queues--;
 	nvmeq->cq_vector = -1;
+	nvmeq->polled = false;
 	spin_unlock_irq(&nvmeq->cq_lock);
 
 	/*
@@ -1366,7 +1388,8 @@ static int nvme_suspend_queue(struct nvme_queue *nvmeq)
 	if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
 		blk_mq_quiesce_queue(nvmeq->dev->ctrl.admin_q);
 
-	pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq);
+	if (vector != -1)
+		pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq);
 
 	return 0;
 }
@@ -1500,7 +1523,7 @@ static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
 	spin_unlock_irq(&nvmeq->cq_lock);
 }
 
-static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
+static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled)
 {
 	struct nvme_dev *dev = nvmeq->dev;
 	int result;
@@ -1510,7 +1533,11 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
 	 * A queue's vector matches the queue identifier unless the controller
 	 * has only one vector available.
 	 */
-	vector = dev->num_vecs == 1 ? 0 : qid;
+	if (!polled)
+		vector = dev->num_vecs == 1 ? 0 : qid;
+	else
+		vector = -1;
+
 	result = adapter_alloc_cq(dev, qid, nvmeq, vector);
 	if (result)
 		return result;
@@ -1527,15 +1554,20 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
 	 * xxx' warning if the create CQ/SQ command times out.
 	 */
 	nvmeq->cq_vector = vector;
+	nvmeq->polled = polled;
 	nvme_init_queue(nvmeq, qid);
-	result = queue_request_irq(nvmeq);
-	if (result < 0)
-		goto release_sq;
+
+	if (vector != -1) {
+		result = queue_request_irq(nvmeq);
+		if (result < 0)
+			goto release_sq;
+	}
 
 	return result;
 
 release_sq:
 	nvmeq->cq_vector = -1;
+	nvmeq->polled = false;
 	dev->online_queues--;
 	adapter_delete_sq(dev, qid);
 release_cq:
@@ -1686,7 +1718,7 @@ static int nvme_pci_configure_admin_queue(struct nvme_dev *dev)
 
 static int nvme_create_io_queues(struct nvme_dev *dev)
 {
-	unsigned i, max;
+	unsigned i, max, rw_queues;
 	int ret = 0;
 
 	for (i = dev->ctrl.queue_count; i <= dev->max_qid; i++) {
@@ -1697,8 +1729,17 @@ static int nvme_create_io_queues(struct nvme_dev *dev)
 	}
 
 	max = min(dev->max_qid, dev->ctrl.queue_count - 1);
+	if (max != 1 && dev->io_queues[NVMEQ_TYPE_POLL]) {
+		rw_queues = dev->io_queues[NVMEQ_TYPE_READ] +
+				dev->io_queues[NVMEQ_TYPE_WRITE];
+	} else {
+		rw_queues = max;
+	}
+
 	for (i = dev->online_queues; i <= max; i++) {
-		ret = nvme_create_queue(&dev->queues[i], i);
+		bool polled = i > rw_queues;
+
+		ret = nvme_create_queue(&dev->queues[i], i, polled);
 		if (ret)
 			break;
 	}
@@ -1970,6 +2011,7 @@ static int nvme_setup_host_mem(struct nvme_dev *dev)
 static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int nr_io_queues)
 {
 	unsigned int this_w_queues = write_queues;
+	unsigned int this_p_queues = poll_queues;
 
 	/*
 	 * Setup read/write queue split
@@ -1977,9 +2019,28 @@ static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int nr_io_queues)
 	if (nr_io_queues == 1) {
 		dev->io_queues[NVMEQ_TYPE_READ] = 1;
 		dev->io_queues[NVMEQ_TYPE_WRITE] = 0;
+		dev->io_queues[NVMEQ_TYPE_POLL] = 0;
 		return;
 	}
 
+	/*
+	 * Configure number of poll queues, if set
+	 */
+	if (this_p_queues) {
+		/*
+		 * We need at least one queue left. With just one queue, we'll
+		 * have a single shared read/write set.
+		 */
+		if (this_p_queues >= nr_io_queues) {
+			this_w_queues = 0;
+			this_p_queues = nr_io_queues - 1;
+		}
+
+		dev->io_queues[NVMEQ_TYPE_POLL] = this_p_queues;
+		nr_io_queues -= this_p_queues;
+	} else
+		dev->io_queues[NVMEQ_TYPE_POLL] = 0;
+
 	/*
 	 * If 'write_queues' is set, ensure it leaves room for at least
 	 * one read queue
@@ -2084,11 +2145,13 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	} while (1);
 
 	dev->num_vecs = result;
-	dev->max_qid = max(result - 1, 1);
+	result = max(result - 1, 1);
+	dev->max_qid = result + dev->io_queues[NVMEQ_TYPE_POLL];
 
-	dev_info(dev->ctrl.device, "%d/%d/%d read/write queues\n",
+	dev_info(dev->ctrl.device, "%d/%d/%d read/write/poll queues\n",
 					dev->io_queues[NVMEQ_TYPE_READ],
-					dev->io_queues[NVMEQ_TYPE_WRITE]);
+					dev->io_queues[NVMEQ_TYPE_WRITE],
+					dev->io_queues[NVMEQ_TYPE_POLL]);
 
 	/*
 	 * Should investigate if there's a performance win from allocating
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 8e80d5043079..b31f6f016621 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -82,7 +82,7 @@ struct blk_mq_queue_map {
 };
 
 enum {
-	HCTX_MAX_TYPES = 2,
+	HCTX_MAX_TYPES = 3,
 };
 
 struct blk_mq_tag_set {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps
  2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
                   ` (15 preceding siblings ...)
  2018-10-30 18:32 ` [PATCH 16/16] nvme: add separate poll queue map Jens Axboe
@ 2018-10-30 18:35 ` Keith Busch
  16 siblings, 0 replies; 48+ messages in thread
From: Keith Busch @ 2018-10-30 18:35 UTC (permalink / raw)
  To: Jens Axboe; +Cc: linux-block, linux-nvme, linux-kernel

On Tue, Oct 30, 2018 at 12:32:36PM -0600, Jens Axboe wrote:
> This series adds support for multiple queue maps for blk-mq.
> Since blk-mq was introduced, it's only support a single queue
> map. This means you can have 1 set of queues, and the mapping
> purely depends on what CPU an IO originated from. With this
> patch set, drivers can implement mappings that depend on both
> CPU and request type - and they can have multiple sets of mappings.
> 
> NVMe is used as a proof of concept. It adds support for a separate
> write queue set. One way to use this would be to limit the number
> of write queues to favor reads, since NVMe does round-robin service
> of queues. An easy extension of this would be to add multiple
> sets of queues, for prioritized IO.
> 
> NVMe also uses this feature to finally make the polling work
> efficiently, without triggering interrupts. This both increases
> performance (and decreases latency), at a lower system load. At
> the same time it's more flexible, as you don't have to worry about
> IRQ coalescing and redirection to avoid interrupts disturbing the
> workload. This is how polling should have worked from day 1.
> 
> This is on top of my mq-conversions branch. It can also be bound
> in my mq-maps branch.

Looks good to me. For the series:

Reviewed-by: Keith Busch <keith.busch@intel.com>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 01/16] blk-mq: kill q->mq_map
  2018-10-30 18:32 ` [PATCH 01/16] blk-mq: kill q->mq_map Jens Axboe
@ 2018-10-31  0:28   ` Sagi Grimberg
  0 siblings, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:28 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 02/16] blk-mq: abstract out queue map
  2018-10-30 18:32 ` [PATCH 02/16] blk-mq: abstract out queue map Jens Axboe
@ 2018-10-31  0:31   ` Sagi Grimberg
  2018-10-31 14:17     ` Jens Axboe
  0 siblings, 1 reply; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:31 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel


> @@ -41,12 +41,12 @@ int blk_mq_rdma_map_queues(struct blk_mq_tag_set *set,
>   			goto fallback;
>   
>   		for_each_cpu(cpu, mask)
> -			set->mq_map[cpu] = queue;
> +			set->map[0].mq_map[cpu] = queue;

Any reason not to change the signature like the other mappers?

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 03/16] blk-mq: provide dummy blk_mq_map_queue_type() helper
  2018-10-30 18:32 ` [PATCH 03/16] blk-mq: provide dummy blk_mq_map_queue_type() helper Jens Axboe
@ 2018-10-31  0:32   ` Sagi Grimberg
  0 siblings, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:32 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping
  2018-10-30 18:32 ` [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping Jens Axboe
@ 2018-10-31  0:37   ` Sagi Grimberg
  2018-10-31 14:18     ` Jens Axboe
  0 siblings, 1 reply; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:37 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel


>   static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
> +						     unsigned int flags,
>   						     unsigned int cpu)
>   {
>   	struct blk_mq_tag_set *set = q->tag_set;
> @@ -84,7 +85,7 @@ static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *
>   							  unsigned int hctx_type,
>   							  unsigned int cpu)
>   {
> -	return blk_mq_map_queue(q, cpu);
> +	return blk_mq_map_queue(q, hctx_type, cpu);
>   }

Haven't read through the rest of the patch set, but it looks a bit
strange that a type enumeration is being passed on as flags field.

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues
  2018-10-30 18:32 ` [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues Jens Axboe
@ 2018-10-31  0:49   ` Sagi Grimberg
  2018-10-31 14:19     ` Jens Axboe
  0 siblings, 1 reply; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:49 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel



> @@ -2342,10 +2346,16 @@ static void blk_mq_map_swqueue(struct request_queue *q)
>   
>   		ctx = per_cpu_ptr(q->queue_ctx, i);
>   		hctx = blk_mq_map_queue_type(q, 0, i);
> -
> +		hctx->type = 0;
>   		cpumask_set_cpu(i, hctx->cpumask);
> -		ctx->index_hw = hctx->nr_ctx;
> +		ctx->index_hw[hctx->type] = hctx->nr_ctx;
>   		hctx->ctxs[hctx->nr_ctx++] = ctx;
> +
> +		/*
> +		 * If the nr_ctx type overflows, we have exceeded the
> +		 * amount of sw queues we can support.
> +		 */
> +		BUG_ON(!hctx->nr_ctx);

Was this added because nr_ctx became short? What was the motivation
for that? save space? structure alignment?

Perhaps this should be mentioned in the change log (as it contradicts
that no functional changes exist in this patch). Or, if this can happen
even when 32bit nr_ctx, it should be a separate small patch?

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory
  2018-10-30 18:32 ` [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory Jens Axboe
@ 2018-10-31  0:53   ` Sagi Grimberg
  2018-10-31 14:21     ` Jens Axboe
  2018-11-01 21:59   ` Omar Sandoval
  1 sibling, 1 reply; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:53 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel


> It can be useful for a user to verify what type a given hardware
> queue is, expose this information in sysfs.

I assume the user is expected to understand the meaning of the
enumeration it sees in accessing this field correct?

Would be nice to output some meaningful string but I'm not
yet sure how to do that, but would be nice...

Other than that,
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 07/16] blk-mq: support multiple hctx maps
  2018-10-30 18:32 ` [PATCH 07/16] blk-mq: support multiple hctx maps Jens Axboe
@ 2018-10-31  0:59   ` Sagi Grimberg
  2018-10-31 14:23     ` Jens Axboe
  0 siblings, 1 reply; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  0:59 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel


> diff --git a/block/blk-mq.h b/block/blk-mq.h
> index 1821f448f7c4..8329017badc8 100644
> --- a/block/blk-mq.h
> +++ b/block/blk-mq.h
> @@ -72,20 +72,37 @@ void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
>    */
>   extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
>   
> -static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
> -						     unsigned int flags,
> -						     unsigned int cpu)
> +/*
> + * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue
> + * @q: request queue
> + * @hctx_type: the hctx type index
> + * @cpu: CPU
> + */
> +static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
> +							  unsigned int hctx_type,
> +							  unsigned int cpu)
>   {
>   	struct blk_mq_tag_set *set = q->tag_set;
>   
> -	return q->queue_hw_ctx[set->map[0].mq_map[cpu]];
> +	return q->queue_hw_ctx[set->map[hctx_type].mq_map[cpu]];
>   }
>   
> -static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
> -							  unsigned int hctx_type,
> -							  unsigned int cpu)
> +/*
> + * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue
> + * @q: request queue
> + * @flags: request command flags
> + * @cpu: CPU
> + */
> +static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
> +						     unsigned int flags,
> +						     unsigned int cpu)
>   {
> -	return blk_mq_map_queue(q, hctx_type, cpu);
> +	int hctx_type = 0;
> +
> +	if (q->mq_ops->flags_to_type)
> +		hctx_type = q->mq_ops->flags_to_type(q, flags);

nitpick: I think that rq_flags_to_type is a bit more descriptive here.
just a personal opinion.

Other than that,
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 08/16] blk-mq: separate number of hardware queues from nr_cpu_ids
  2018-10-30 18:32 ` [PATCH 08/16] blk-mq: separate number of hardware queues from nr_cpu_ids Jens Axboe
@ 2018-10-31  1:00   ` Sagi Grimberg
  0 siblings, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:00 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 09/16] blk-mq: cache request hardware queue mapping
  2018-10-30 18:32 ` [PATCH 09/16] blk-mq: cache request hardware queue mapping Jens Axboe
@ 2018-10-31  1:01   ` Sagi Grimberg
  2018-11-01  9:27   ` Hannes Reinecke
  1 sibling, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:01 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 10/16] blk-mq: cleanup and improve list insertion
  2018-10-30 18:32 ` [PATCH 10/16] blk-mq: cleanup and improve list insertion Jens Axboe
@ 2018-10-31  1:03   ` Sagi Grimberg
  2018-11-01  9:28   ` Hannes Reinecke
  1 sibling, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:03 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 11/16] blk-mq: improve plug list sorting
  2018-10-30 18:32 ` [PATCH 11/16] blk-mq: improve plug list sorting Jens Axboe
@ 2018-10-31  1:04   ` Sagi Grimberg
  2018-11-01  9:30   ` Hannes Reinecke
  1 sibling, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:04 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 12/16] blk-mq: initial support for multiple queue maps
  2018-10-30 18:32 ` [PATCH 12/16] blk-mq: initial support for multiple queue maps Jens Axboe
@ 2018-10-31  1:14   ` Sagi Grimberg
  0 siblings, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:14 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs
  2018-10-30 18:32 ` [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs Jens Axboe
@ 2018-10-31  1:17   ` Sagi Grimberg
  2018-11-02 14:37   ` Ming Lei
  1 sibling, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:17 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel; +Cc: Thomas Gleixner

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes
  2018-10-30 18:32 ` [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes Jens Axboe
@ 2018-10-31  1:57   ` Sagi Grimberg
  2018-10-31 14:32     ` Jens Axboe
  0 siblings, 1 reply; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:57 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel


> +static int queue_irq_offset(struct nvme_dev *dev)
> +{
> +	/* if we have more than 1 vec, admin queue offsets us 1 */

offsets us by 1?

> +	if (dev->num_vecs > 1)
> +		return 1;
> +
> +	return 0;
> +}
> +

...

> @@ -1934,13 +2048,48 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
>   	 * setting up the full range we need.
>   	 */
>   	pci_free_irq_vectors(pdev);
> -	result = pci_alloc_irq_vectors_affinity(pdev, 1, nr_io_queues + 1,
> -			PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
> -	if (result <= 0)
> -		return -EIO;
> +
> +	/*
> +	 * For irq sets, we have to ask for minvec == maxvec. This passes
> +	 * any reduction back to us, so we can adjust our queue counts and
> +	 * IRQ vector needs.
> +	 */
> +	do {
> +		nvme_calc_io_queues(dev, nr_io_queues);
> +		irq_sets[0] = dev->io_queues[NVMEQ_TYPE_READ];
> +		irq_sets[1] = dev->io_queues[NVMEQ_TYPE_WRITE];
> +		if (!irq_sets[1])
> +			affd.nr_sets = 1;
> +
> +		/*
> +		 * Need IRQs for read+write queues, and one for the admin queue
> +		 */
> +		nr_io_queues = irq_sets[0] + irq_sets[1] + 1;
> +
> +		result = pci_alloc_irq_vectors_affinity(pdev, nr_io_queues,
> +				nr_io_queues,
> +				PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
> +
> +		/*
> +		 * Need to reduce our vec counts
> +		 */
> +		if (result == -ENOSPC) {
> +			nr_io_queues--;
> +			if (!nr_io_queues)
> +				return result;
> +			continue;
> +		} else if (result <= 0)
> +			return -EIO;
> +		break;
> +	} while (1);
> +
>   	dev->num_vecs = result;
>   	dev->max_qid = max(result - 1, 1);
>   
> +	dev_info(dev->ctrl.device, "%d/%d/%d read/write queues\n",
> +					dev->io_queues[NVMEQ_TYPE_READ],
> +					dev->io_queues[NVMEQ_TYPE_WRITE]);
> +

Perhaps it would be better if we move this code into a function.

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 15/16] block: add REQ_HIPRI and inherit it from IOCB_HIPRI
  2018-10-30 18:32 ` [PATCH 15/16] block: add REQ_HIPRI and inherit it from IOCB_HIPRI Jens Axboe
@ 2018-10-31  1:58   ` Sagi Grimberg
  0 siblings, 0 replies; 48+ messages in thread
From: Sagi Grimberg @ 2018-10-31  1:58 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 02/16] blk-mq: abstract out queue map
  2018-10-31  0:31   ` Sagi Grimberg
@ 2018-10-31 14:17     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-31 14:17 UTC (permalink / raw)
  To: Sagi Grimberg, linux-block, linux-nvme, linux-kernel

On 10/30/18 6:31 PM, Sagi Grimberg wrote:
> 
>> @@ -41,12 +41,12 @@ int blk_mq_rdma_map_queues(struct blk_mq_tag_set *set,
>>   			goto fallback;
>>   
>>   		for_each_cpu(cpu, mask)
>> -			set->mq_map[cpu] = queue;
>> +			set->map[0].mq_map[cpu] = queue;
> 
> Any reason not to change the signature like the other mappers?

No particular reason, outside of not wanting to touch more than
I had to.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping
  2018-10-31  0:37   ` Sagi Grimberg
@ 2018-10-31 14:18     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-31 14:18 UTC (permalink / raw)
  To: Sagi Grimberg, linux-block, linux-nvme, linux-kernel

On 10/30/18 6:37 PM, Sagi Grimberg wrote:
> 
>>   static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
>> +						     unsigned int flags,
>>   						     unsigned int cpu)
>>   {
>>   	struct blk_mq_tag_set *set = q->tag_set;
>> @@ -84,7 +85,7 @@ static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *
>>   							  unsigned int hctx_type,
>>   							  unsigned int cpu)
>>   {
>> -	return blk_mq_map_queue(q, cpu);
>> +	return blk_mq_map_queue(q, hctx_type, cpu);
>>   }
> 
> Haven't read through the rest of the patch set, but it looks a bit
> strange that a type enumeration is being passed on as flags field.

It's impossible to get it 100% right at this point, which is why
it looks a bit odd. If you read on, it'll make sense.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues
  2018-10-31  0:49   ` Sagi Grimberg
@ 2018-10-31 14:19     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-31 14:19 UTC (permalink / raw)
  To: Sagi Grimberg, linux-block, linux-nvme, linux-kernel

On 10/30/18 6:49 PM, Sagi Grimberg wrote:
> 
> 
>> @@ -2342,10 +2346,16 @@ static void blk_mq_map_swqueue(struct request_queue *q)
>>   
>>   		ctx = per_cpu_ptr(q->queue_ctx, i);
>>   		hctx = blk_mq_map_queue_type(q, 0, i);
>> -
>> +		hctx->type = 0;
>>   		cpumask_set_cpu(i, hctx->cpumask);
>> -		ctx->index_hw = hctx->nr_ctx;
>> +		ctx->index_hw[hctx->type] = hctx->nr_ctx;
>>   		hctx->ctxs[hctx->nr_ctx++] = ctx;
>> +
>> +		/*
>> +		 * If the nr_ctx type overflows, we have exceeded the
>> +		 * amount of sw queues we can support.
>> +		 */
>> +		BUG_ON(!hctx->nr_ctx);
> 
> Was this added because nr_ctx became short? What was the motivation
> for that? save space? structure alignment?

Right, I made it smaller to not take up more space after this change.
And since it is now smaller, adding a check for wrap seemed like
a good idea.

> Perhaps this should be mentioned in the change log (as it contradicts
> that no functional changes exist in this patch). Or, if this can happen
> even when 32bit nr_ctx, it should be a separate small patch?

Could happen for any type, but more likely now of course. I'll
mention this in the changelog.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory
  2018-10-31  0:53   ` Sagi Grimberg
@ 2018-10-31 14:21     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-31 14:21 UTC (permalink / raw)
  To: Sagi Grimberg, linux-block, linux-nvme, linux-kernel

On 10/30/18 6:53 PM, Sagi Grimberg wrote:
> 
>> It can be useful for a user to verify what type a given hardware
>> queue is, expose this information in sysfs.
> 
> I assume the user is expected to understand the meaning of the
> enumeration it sees in accessing this field correct?
> 
> Would be nice to output some meaningful string but I'm not
> yet sure how to do that, but would be nice...

The type is driver private, there's no specific meaning associated
with the type. We could make it output a string provided that
the driver had a way to query it. For now I think the type is
enough, it'll allow you to see which ones are different.

> Other than that,
> Reviewed-by: Sagi Grimberg <sagi@grimberg.me>

Thanks

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 07/16] blk-mq: support multiple hctx maps
  2018-10-31  0:59   ` Sagi Grimberg
@ 2018-10-31 14:23     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-31 14:23 UTC (permalink / raw)
  To: Sagi Grimberg, linux-block, linux-nvme, linux-kernel

On 10/30/18 6:59 PM, Sagi Grimberg wrote:
>> +static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
>> +						     unsigned int flags,
>> +						     unsigned int cpu)
>>   {
>> -	return blk_mq_map_queue(q, hctx_type, cpu);
>> +	int hctx_type = 0;
>> +
>> +	if (q->mq_ops->flags_to_type)
>> +		hctx_type = q->mq_ops->flags_to_type(q, flags);
> 
> nitpick: I think that rq_flags_to_type is a bit more descriptive here.
> just a personal opinion.

Sure, I can make that change.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes
  2018-10-31  1:57   ` Sagi Grimberg
@ 2018-10-31 14:32     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-10-31 14:32 UTC (permalink / raw)
  To: Sagi Grimberg, linux-block, linux-nvme, linux-kernel

On 10/30/18 7:57 PM, Sagi Grimberg wrote:
> 
>> +static int queue_irq_offset(struct nvme_dev *dev)
>> +{
>> +	/* if we have more than 1 vec, admin queue offsets us 1 */
> 
> offsets us by 1?

Fixed

>> @@ -1934,13 +2048,48 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
>>   	 * setting up the full range we need.
>>   	 */
>>   	pci_free_irq_vectors(pdev);
>> -	result = pci_alloc_irq_vectors_affinity(pdev, 1, nr_io_queues + 1,
>> -			PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
>> -	if (result <= 0)
>> -		return -EIO;
>> +
>> +	/*
>> +	 * For irq sets, we have to ask for minvec == maxvec. This passes
>> +	 * any reduction back to us, so we can adjust our queue counts and
>> +	 * IRQ vector needs.
>> +	 */
>> +	do {
>> +		nvme_calc_io_queues(dev, nr_io_queues);
>> +		irq_sets[0] = dev->io_queues[NVMEQ_TYPE_READ];
>> +		irq_sets[1] = dev->io_queues[NVMEQ_TYPE_WRITE];
>> +		if (!irq_sets[1])
>> +			affd.nr_sets = 1;
>> +
>> +		/*
>> +		 * Need IRQs for read+write queues, and one for the admin queue
>> +		 */
>> +		nr_io_queues = irq_sets[0] + irq_sets[1] + 1;
>> +
>> +		result = pci_alloc_irq_vectors_affinity(pdev, nr_io_queues,
>> +				nr_io_queues,
>> +				PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
>> +
>> +		/*
>> +		 * Need to reduce our vec counts
>> +		 */
>> +		if (result == -ENOSPC) {
>> +			nr_io_queues--;
>> +			if (!nr_io_queues)
>> +				return result;
>> +			continue;
>> +		} else if (result <= 0)
>> +			return -EIO;
>> +		break;
>> +	} while (1);
>> +
>>   	dev->num_vecs = result;
>>   	dev->max_qid = max(result - 1, 1);
>>   
>> +	dev_info(dev->ctrl.device, "%d/%d/%d read/write queues\n",
>> +					dev->io_queues[NVMEQ_TYPE_READ],
>> +					dev->io_queues[NVMEQ_TYPE_WRITE]);
>> +
> 
> Perhaps it would be better if we move this code into a function.

Agree, I've done that now.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 09/16] blk-mq: cache request hardware queue mapping
  2018-10-30 18:32 ` [PATCH 09/16] blk-mq: cache request hardware queue mapping Jens Axboe
  2018-10-31  1:01   ` Sagi Grimberg
@ 2018-11-01  9:27   ` Hannes Reinecke
  2018-11-01 12:22     ` Jens Axboe
  1 sibling, 1 reply; 48+ messages in thread
From: Hannes Reinecke @ 2018-11-01  9:27 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

On 10/30/18 7:32 PM, Jens Axboe wrote:
> We call blk_mq_map_queue() a lot, at least two times for each
> request per IO, sometimes more. Since we now have an indirect
> call as well in that function. cache the mapping so we don't
> have to re-call blk_mq_map_queue() for the same request
> multiple times.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>   block/blk-flush.c      | 12 ++++--------
>   block/blk-mq-debugfs.c |  4 +---
>   block/blk-mq-sched.c   |  6 ++----
>   block/blk-mq-tag.c     |  9 +--------
>   block/blk-mq.c         | 22 +++++++++-------------
>   block/blk-mq.h         |  5 +----
>   include/linux/blkdev.h |  1 +
>   7 files changed, 19 insertions(+), 40 deletions(-)
> 
Hehe.
I had been wondering about this since the original mq implementation, 
but was told not to worry :-)

Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes



^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 10/16] blk-mq: cleanup and improve list insertion
  2018-10-30 18:32 ` [PATCH 10/16] blk-mq: cleanup and improve list insertion Jens Axboe
  2018-10-31  1:03   ` Sagi Grimberg
@ 2018-11-01  9:28   ` Hannes Reinecke
  1 sibling, 0 replies; 48+ messages in thread
From: Hannes Reinecke @ 2018-11-01  9:28 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

On 10/30/18 7:32 PM, Jens Axboe wrote:
> It's somewhat strange to have a list insertion function that
> relies on the fact that the caller has mapped things correctly.
> Pass in the hardware queue directly for insertion, which makes
> for a much cleaner interface and implementation.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>   block/blk-mq-sched.c |  8 +-------
>   block/blk-mq-sched.h |  2 +-
>   block/blk-mq.c       | 25 ++++++++++++++-----------
>   3 files changed, 16 insertions(+), 19 deletions(-)
> 
Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 11/16] blk-mq: improve plug list sorting
  2018-10-30 18:32 ` [PATCH 11/16] blk-mq: improve plug list sorting Jens Axboe
  2018-10-31  1:04   ` Sagi Grimberg
@ 2018-11-01  9:30   ` Hannes Reinecke
  1 sibling, 0 replies; 48+ messages in thread
From: Hannes Reinecke @ 2018-11-01  9:30 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-nvme, linux-kernel

On 10/30/18 7:32 PM, Jens Axboe wrote:
> Currently we only look at the software queue, but with support
> for multiple maps, we should also look at the hardware queue.
> This is important since we'll flush out the request list if
> either the software queue or hardware queue don't match.
> 
> This sorts by software queue first, then hardware queue if
> that differs. Finally we sort by request location like before.
> This minimizes the flush points per plug list.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>   block/blk-mq.c | 17 ++++++++++++-----
>   1 file changed, 12 insertions(+), 5 deletions(-)
> 
Reviewed-by: Hannes Reinecke <hare@suse.com>

Cheers,

Hannes

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 09/16] blk-mq: cache request hardware queue mapping
  2018-11-01  9:27   ` Hannes Reinecke
@ 2018-11-01 12:22     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-11-01 12:22 UTC (permalink / raw)
  To: Hannes Reinecke, linux-block, linux-nvme, linux-kernel

On 11/1/18 3:27 AM, Hannes Reinecke wrote:
> On 10/30/18 7:32 PM, Jens Axboe wrote:
>> We call blk_mq_map_queue() a lot, at least two times for each
>> request per IO, sometimes more. Since we now have an indirect
>> call as well in that function. cache the mapping so we don't
>> have to re-call blk_mq_map_queue() for the same request
>> multiple times.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>   block/blk-flush.c      | 12 ++++--------
>>   block/blk-mq-debugfs.c |  4 +---
>>   block/blk-mq-sched.c   |  6 ++----
>>   block/blk-mq-tag.c     |  9 +--------
>>   block/blk-mq.c         | 22 +++++++++-------------
>>   block/blk-mq.h         |  5 +----
>>   include/linux/blkdev.h |  1 +
>>   7 files changed, 19 insertions(+), 40 deletions(-)
>>
> Hehe.
> I had been wondering about this since the original mq implementation, 
> but was told not to worry :-)

To be fair, it was just an array index. Now it's got an indirect
call, so it's more expensive.
 > Reviewed-by: Hannes Reinecke <hare@suse.com>

Thanks!

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory
  2018-10-30 18:32 ` [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory Jens Axboe
  2018-10-31  0:53   ` Sagi Grimberg
@ 2018-11-01 21:59   ` Omar Sandoval
  2018-11-01 22:50     ` Jens Axboe
  1 sibling, 1 reply; 48+ messages in thread
From: Omar Sandoval @ 2018-11-01 21:59 UTC (permalink / raw)
  To: Jens Axboe; +Cc: linux-block, linux-nvme, linux-kernel

On Tue, Oct 30, 2018 at 12:32:42PM -0600, Jens Axboe wrote:
> It can be useful for a user to verify what type a given hardware
> queue is, expose this information in sysfs.
> 
> Reviewed-by: Hannes Reinecke <hare@suse.com>
> Reviewed-by: Bart Van Assche <bvanassche@acm.org>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  block/blk-mq-sysfs.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)

Out of curiousity, why sysfs instead of debugfs?

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory
  2018-11-01 21:59   ` Omar Sandoval
@ 2018-11-01 22:50     ` Jens Axboe
  0 siblings, 0 replies; 48+ messages in thread
From: Jens Axboe @ 2018-11-01 22:50 UTC (permalink / raw)
  To: Omar Sandoval; +Cc: linux-block, linux-nvme, linux-kernel

On 11/1/18 3:59 PM, Omar Sandoval wrote:
> On Tue, Oct 30, 2018 at 12:32:42PM -0600, Jens Axboe wrote:
>> It can be useful for a user to verify what type a given hardware
>> queue is, expose this information in sysfs.
>>
>> Reviewed-by: Hannes Reinecke <hare@suse.com>
>> Reviewed-by: Bart Van Assche <bvanassche@acm.org>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  block/blk-mq-sysfs.c | 10 ++++++++++
>>  1 file changed, 10 insertions(+)
> 
> Out of curiousity, why sysfs instead of debugfs?

I don't feel that strongly about it, but it is a property
like the mapped CPUs, for instance, so kind of makes sense to
put it there rather than in debugfs.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs
  2018-10-30 18:32 ` [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs Jens Axboe
  2018-10-31  1:17   ` Sagi Grimberg
@ 2018-11-02 14:37   ` Ming Lei
  2018-11-02 15:09     ` Keith Busch
  1 sibling, 1 reply; 48+ messages in thread
From: Ming Lei @ 2018-11-02 14:37 UTC (permalink / raw)
  To: Jens Axboe; +Cc: linux-block, linux-nvme, linux-kernel, Thomas Gleixner

On Tue, Oct 30, 2018 at 12:32:49PM -0600, Jens Axboe wrote:
> A driver may have a need to allocate multiple sets of MSI/MSI-X
> interrupts, and have them appropriately affinitized. Add support for
> defining a number of sets in the irq_affinity structure, of varying
> sizes, and get each set affinitized correctly across the machine.
> 
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: linux-kernel@vger.kernel.org
> Reviewed-by: Hannes Reinecke <hare@suse.com>
> Reviewed-by: Ming Lei <ming.lei@redhat.com>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  drivers/pci/msi.c         | 14 ++++++++++++++
>  include/linux/interrupt.h |  4 ++++
>  kernel/irq/affinity.c     | 40 ++++++++++++++++++++++++++++++---------
>  3 files changed, 49 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> index af24ed50a245..e6c6e10b9ceb 100644
> --- a/drivers/pci/msi.c
> +++ b/drivers/pci/msi.c
> @@ -1036,6 +1036,13 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
>  	if (maxvec < minvec)
>  		return -ERANGE;
>  
> +	/*
> +	 * If the caller is passing in sets, we can't support a range of
> +	 * vectors. The caller needs to handle that.
> +	 */
> +	if (affd->nr_sets && minvec != maxvec)
> +		return -EINVAL;
> +
>  	if (WARN_ON_ONCE(dev->msi_enabled))
>  		return -EINVAL;
>  
> @@ -1087,6 +1094,13 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
>  	if (maxvec < minvec)
>  		return -ERANGE;
>  
> +	/*
> +	 * If the caller is passing in sets, we can't support a range of
> +	 * supported vectors. The caller needs to handle that.
> +	 */
> +	if (affd->nr_sets && minvec != maxvec)
> +		return -EINVAL;
> +
>  	if (WARN_ON_ONCE(dev->msix_enabled))
>  		return -EINVAL;
>  
> diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
> index 1d6711c28271..ca397ff40836 100644
> --- a/include/linux/interrupt.h
> +++ b/include/linux/interrupt.h
> @@ -247,10 +247,14 @@ struct irq_affinity_notify {
>   *			the MSI(-X) vector space
>   * @post_vectors:	Don't apply affinity to @post_vectors at end of
>   *			the MSI(-X) vector space
> + * @nr_sets:		Length of passed in *sets array
> + * @sets:		Number of affinitized sets
>   */
>  struct irq_affinity {
>  	int	pre_vectors;
>  	int	post_vectors;
> +	int	nr_sets;
> +	int	*sets;
>  };
>  
>  #if defined(CONFIG_SMP)
> diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c
> index f4f29b9d90ee..2046a0f0f0f1 100644
> --- a/kernel/irq/affinity.c
> +++ b/kernel/irq/affinity.c
> @@ -180,6 +180,7 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
>  	int curvec, usedvecs;
>  	cpumask_var_t nmsk, npresmsk, *node_to_cpumask;
>  	struct cpumask *masks = NULL;
> +	int i, nr_sets;
>  
>  	/*
>  	 * If there aren't any vectors left after applying the pre/post
> @@ -210,10 +211,23 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
>  	get_online_cpus();
>  	build_node_to_cpumask(node_to_cpumask);
>  
> -	/* Spread on present CPUs starting from affd->pre_vectors */
> -	usedvecs = irq_build_affinity_masks(affd, curvec, affvecs,
> -					    node_to_cpumask, cpu_present_mask,
> -					    nmsk, masks);
> +	/*
> +	 * Spread on present CPUs starting from affd->pre_vectors. If we
> +	 * have multiple sets, build each sets affinity mask separately.
> +	 */
> +	nr_sets = affd->nr_sets;
> +	if (!nr_sets)
> +		nr_sets = 1;
> +
> +	for (i = 0, usedvecs = 0; i < nr_sets; i++) {
> +		int this_vecs = affd->sets ? affd->sets[i] : affvecs;
> +		int nr;
> +
> +		nr = irq_build_affinity_masks(affd, curvec, this_vecs,
> +					      node_to_cpumask, cpu_present_mask,
> +					      nmsk, masks + usedvecs);

The last parameter of the above function should have been 'masks',
because irq_build_affinity_masks() always treats 'masks' as the base
address of the array.

> +		usedvecs += nr;
> +	}

Thinking of further, one big problem in this patch is that each set of
IRQs should have been spread on all possible CPUs, which is done via
2-stages spread now.

However, this patch only spreads each set of IRQs on present CPUs, this
way may not work in case of physical CPU hotplug.

Thanks,
Ming

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs
  2018-11-02 14:37   ` Ming Lei
@ 2018-11-02 15:09     ` Keith Busch
  2018-11-03  2:22       ` Ming Lei
  0 siblings, 1 reply; 48+ messages in thread
From: Keith Busch @ 2018-11-02 15:09 UTC (permalink / raw)
  To: Ming Lei
  Cc: Jens Axboe, linux-block, linux-nvme, linux-kernel, Thomas Gleixner

On Fri, Nov 02, 2018 at 10:37:07PM +0800, Ming Lei wrote:
> On Tue, Oct 30, 2018 at 12:32:49PM -0600, Jens Axboe wrote:
> > A driver may have a need to allocate multiple sets of MSI/MSI-X
> > interrupts, and have them appropriately affinitized. Add support for
> > defining a number of sets in the irq_affinity structure, of varying
> > sizes, and get each set affinitized correctly across the machine.
> > 
> > Cc: Thomas Gleixner <tglx@linutronix.de>
> > Cc: linux-kernel@vger.kernel.org
> > Reviewed-by: Hannes Reinecke <hare@suse.com>
> > Reviewed-by: Ming Lei <ming.lei@redhat.com>
> > Signed-off-by: Jens Axboe <axboe@kernel.dk>
> > ---
> >  drivers/pci/msi.c         | 14 ++++++++++++++
> >  include/linux/interrupt.h |  4 ++++
> >  kernel/irq/affinity.c     | 40 ++++++++++++++++++++++++++++++---------
> >  3 files changed, 49 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> > index af24ed50a245..e6c6e10b9ceb 100644
> > --- a/drivers/pci/msi.c
> > +++ b/drivers/pci/msi.c
> > @@ -1036,6 +1036,13 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
> >  	if (maxvec < minvec)
> >  		return -ERANGE;
> >  
> > +	/*
> > +	 * If the caller is passing in sets, we can't support a range of
> > +	 * vectors. The caller needs to handle that.
> > +	 */
> > +	if (affd->nr_sets && minvec != maxvec)
> > +		return -EINVAL;
> > +
> >  	if (WARN_ON_ONCE(dev->msi_enabled))
> >  		return -EINVAL;
> >  
> > @@ -1087,6 +1094,13 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
> >  	if (maxvec < minvec)
> >  		return -ERANGE;
> >  
> > +	/*
> > +	 * If the caller is passing in sets, we can't support a range of
> > +	 * supported vectors. The caller needs to handle that.
> > +	 */
> > +	if (affd->nr_sets && minvec != maxvec)
> > +		return -EINVAL;
> > +
> >  	if (WARN_ON_ONCE(dev->msix_enabled))
> >  		return -EINVAL;
> >  
> > diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
> > index 1d6711c28271..ca397ff40836 100644
> > --- a/include/linux/interrupt.h
> > +++ b/include/linux/interrupt.h
> > @@ -247,10 +247,14 @@ struct irq_affinity_notify {
> >   *			the MSI(-X) vector space
> >   * @post_vectors:	Don't apply affinity to @post_vectors at end of
> >   *			the MSI(-X) vector space
> > + * @nr_sets:		Length of passed in *sets array
> > + * @sets:		Number of affinitized sets
> >   */
> >  struct irq_affinity {
> >  	int	pre_vectors;
> >  	int	post_vectors;
> > +	int	nr_sets;
> > +	int	*sets;
> >  };
> >  
> >  #if defined(CONFIG_SMP)
> > diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c
> > index f4f29b9d90ee..2046a0f0f0f1 100644
> > --- a/kernel/irq/affinity.c
> > +++ b/kernel/irq/affinity.c
> > @@ -180,6 +180,7 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
> >  	int curvec, usedvecs;
> >  	cpumask_var_t nmsk, npresmsk, *node_to_cpumask;
> >  	struct cpumask *masks = NULL;
> > +	int i, nr_sets;
> >  
> >  	/*
> >  	 * If there aren't any vectors left after applying the pre/post
> > @@ -210,10 +211,23 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
> >  	get_online_cpus();
> >  	build_node_to_cpumask(node_to_cpumask);
> >  
> > -	/* Spread on present CPUs starting from affd->pre_vectors */
> > -	usedvecs = irq_build_affinity_masks(affd, curvec, affvecs,
> > -					    node_to_cpumask, cpu_present_mask,
> > -					    nmsk, masks);
> > +	/*
> > +	 * Spread on present CPUs starting from affd->pre_vectors. If we
> > +	 * have multiple sets, build each sets affinity mask separately.
> > +	 */
> > +	nr_sets = affd->nr_sets;
> > +	if (!nr_sets)
> > +		nr_sets = 1;
> > +
> > +	for (i = 0, usedvecs = 0; i < nr_sets; i++) {
> > +		int this_vecs = affd->sets ? affd->sets[i] : affvecs;
> > +		int nr;
> > +
> > +		nr = irq_build_affinity_masks(affd, curvec, this_vecs,
> > +					      node_to_cpumask, cpu_present_mask,
> > +					      nmsk, masks + usedvecs);
> 
> The last parameter of the above function should have been 'masks',
> because irq_build_affinity_masks() always treats 'masks' as the base
> address of the array.

We have multiple "bases" when using sets, so we have to update which
base to use by adding accordingly. If you just use 'masks', then you're
going to overwrite your masks from the previous set.

^ permalink raw reply	[flat|nested] 48+ messages in thread

* Re: [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs
  2018-11-02 15:09     ` Keith Busch
@ 2018-11-03  2:22       ` Ming Lei
  0 siblings, 0 replies; 48+ messages in thread
From: Ming Lei @ 2018-11-03  2:22 UTC (permalink / raw)
  To: Keith Busch
  Cc: Jens Axboe, linux-block, linux-nvme, linux-kernel, Thomas Gleixner

On Fri, Nov 02, 2018 at 09:09:50AM -0600, Keith Busch wrote:
> On Fri, Nov 02, 2018 at 10:37:07PM +0800, Ming Lei wrote:
> > On Tue, Oct 30, 2018 at 12:32:49PM -0600, Jens Axboe wrote:
> > > A driver may have a need to allocate multiple sets of MSI/MSI-X
> > > interrupts, and have them appropriately affinitized. Add support for
> > > defining a number of sets in the irq_affinity structure, of varying
> > > sizes, and get each set affinitized correctly across the machine.
> > > 
> > > Cc: Thomas Gleixner <tglx@linutronix.de>
> > > Cc: linux-kernel@vger.kernel.org
> > > Reviewed-by: Hannes Reinecke <hare@suse.com>
> > > Reviewed-by: Ming Lei <ming.lei@redhat.com>
> > > Signed-off-by: Jens Axboe <axboe@kernel.dk>
> > > ---
> > >  drivers/pci/msi.c         | 14 ++++++++++++++
> > >  include/linux/interrupt.h |  4 ++++
> > >  kernel/irq/affinity.c     | 40 ++++++++++++++++++++++++++++++---------
> > >  3 files changed, 49 insertions(+), 9 deletions(-)
> > > 
> > > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> > > index af24ed50a245..e6c6e10b9ceb 100644
> > > --- a/drivers/pci/msi.c
> > > +++ b/drivers/pci/msi.c
> > > @@ -1036,6 +1036,13 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
> > >  	if (maxvec < minvec)
> > >  		return -ERANGE;
> > >  
> > > +	/*
> > > +	 * If the caller is passing in sets, we can't support a range of
> > > +	 * vectors. The caller needs to handle that.
> > > +	 */
> > > +	if (affd->nr_sets && minvec != maxvec)
> > > +		return -EINVAL;
> > > +
> > >  	if (WARN_ON_ONCE(dev->msi_enabled))
> > >  		return -EINVAL;
> > >  
> > > @@ -1087,6 +1094,13 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
> > >  	if (maxvec < minvec)
> > >  		return -ERANGE;
> > >  
> > > +	/*
> > > +	 * If the caller is passing in sets, we can't support a range of
> > > +	 * supported vectors. The caller needs to handle that.
> > > +	 */
> > > +	if (affd->nr_sets && minvec != maxvec)
> > > +		return -EINVAL;
> > > +
> > >  	if (WARN_ON_ONCE(dev->msix_enabled))
> > >  		return -EINVAL;
> > >  
> > > diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
> > > index 1d6711c28271..ca397ff40836 100644
> > > --- a/include/linux/interrupt.h
> > > +++ b/include/linux/interrupt.h
> > > @@ -247,10 +247,14 @@ struct irq_affinity_notify {
> > >   *			the MSI(-X) vector space
> > >   * @post_vectors:	Don't apply affinity to @post_vectors at end of
> > >   *			the MSI(-X) vector space
> > > + * @nr_sets:		Length of passed in *sets array
> > > + * @sets:		Number of affinitized sets
> > >   */
> > >  struct irq_affinity {
> > >  	int	pre_vectors;
> > >  	int	post_vectors;
> > > +	int	nr_sets;
> > > +	int	*sets;
> > >  };
> > >  
> > >  #if defined(CONFIG_SMP)
> > > diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c
> > > index f4f29b9d90ee..2046a0f0f0f1 100644
> > > --- a/kernel/irq/affinity.c
> > > +++ b/kernel/irq/affinity.c
> > > @@ -180,6 +180,7 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
> > >  	int curvec, usedvecs;
> > >  	cpumask_var_t nmsk, npresmsk, *node_to_cpumask;
> > >  	struct cpumask *masks = NULL;
> > > +	int i, nr_sets;
> > >  
> > >  	/*
> > >  	 * If there aren't any vectors left after applying the pre/post
> > > @@ -210,10 +211,23 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
> > >  	get_online_cpus();
> > >  	build_node_to_cpumask(node_to_cpumask);
> > >  
> > > -	/* Spread on present CPUs starting from affd->pre_vectors */
> > > -	usedvecs = irq_build_affinity_masks(affd, curvec, affvecs,
> > > -					    node_to_cpumask, cpu_present_mask,
> > > -					    nmsk, masks);
> > > +	/*
> > > +	 * Spread on present CPUs starting from affd->pre_vectors. If we
> > > +	 * have multiple sets, build each sets affinity mask separately.
> > > +	 */
> > > +	nr_sets = affd->nr_sets;
> > > +	if (!nr_sets)
> > > +		nr_sets = 1;
> > > +
> > > +	for (i = 0, usedvecs = 0; i < nr_sets; i++) {
> > > +		int this_vecs = affd->sets ? affd->sets[i] : affvecs;
> > > +		int nr;
> > > +
> > > +		nr = irq_build_affinity_masks(affd, curvec, this_vecs,
> > > +					      node_to_cpumask, cpu_present_mask,
> > > +					      nmsk, masks + usedvecs);
> > 
> > The last parameter of the above function should have been 'masks',
> > because irq_build_affinity_masks() always treats 'masks' as the base
> > address of the array.
> 
> We have multiple "bases" when using sets, so we have to update which
> base to use by adding accordingly. If you just use 'masks', then you're
> going to overwrite your masks from the previous set.

For irq_build_affinity_masks(), the passed 'startvec' is always relative
to the absolute 1st element, so the passed 'masks' should be always the
absolute base too. Not mention 'curvec' isn't updated in this patch too.

If you test this patchset on one machine which possible CPUs is bigger
than present CPUs, you will see the problems I mentioned.

Thanks,
Ming

^ permalink raw reply	[flat|nested] 48+ messages in thread

end of thread, other threads:[~2018-11-03  2:22 UTC | newest]

Thread overview: 48+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-10-30 18:32 [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Jens Axboe
2018-10-30 18:32 ` [PATCH 01/16] blk-mq: kill q->mq_map Jens Axboe
2018-10-31  0:28   ` Sagi Grimberg
2018-10-30 18:32 ` [PATCH 02/16] blk-mq: abstract out queue map Jens Axboe
2018-10-31  0:31   ` Sagi Grimberg
2018-10-31 14:17     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 03/16] blk-mq: provide dummy blk_mq_map_queue_type() helper Jens Axboe
2018-10-31  0:32   ` Sagi Grimberg
2018-10-30 18:32 ` [PATCH 04/16] blk-mq: pass in request/bio flags to queue mapping Jens Axboe
2018-10-31  0:37   ` Sagi Grimberg
2018-10-31 14:18     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 05/16] blk-mq: allow software queue to map to multiple hardware queues Jens Axboe
2018-10-31  0:49   ` Sagi Grimberg
2018-10-31 14:19     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 06/16] blk-mq: add 'type' attribute to the sysfs hctx directory Jens Axboe
2018-10-31  0:53   ` Sagi Grimberg
2018-10-31 14:21     ` Jens Axboe
2018-11-01 21:59   ` Omar Sandoval
2018-11-01 22:50     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 07/16] blk-mq: support multiple hctx maps Jens Axboe
2018-10-31  0:59   ` Sagi Grimberg
2018-10-31 14:23     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 08/16] blk-mq: separate number of hardware queues from nr_cpu_ids Jens Axboe
2018-10-31  1:00   ` Sagi Grimberg
2018-10-30 18:32 ` [PATCH 09/16] blk-mq: cache request hardware queue mapping Jens Axboe
2018-10-31  1:01   ` Sagi Grimberg
2018-11-01  9:27   ` Hannes Reinecke
2018-11-01 12:22     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 10/16] blk-mq: cleanup and improve list insertion Jens Axboe
2018-10-31  1:03   ` Sagi Grimberg
2018-11-01  9:28   ` Hannes Reinecke
2018-10-30 18:32 ` [PATCH 11/16] blk-mq: improve plug list sorting Jens Axboe
2018-10-31  1:04   ` Sagi Grimberg
2018-11-01  9:30   ` Hannes Reinecke
2018-10-30 18:32 ` [PATCH 12/16] blk-mq: initial support for multiple queue maps Jens Axboe
2018-10-31  1:14   ` Sagi Grimberg
2018-10-30 18:32 ` [PATCH 13/16] irq: add support for allocating (and affinitizing) sets of IRQs Jens Axboe
2018-10-31  1:17   ` Sagi Grimberg
2018-11-02 14:37   ` Ming Lei
2018-11-02 15:09     ` Keith Busch
2018-11-03  2:22       ` Ming Lei
2018-10-30 18:32 ` [PATCH 14/16] nvme: utilize two queue maps, one for reads and one for writes Jens Axboe
2018-10-31  1:57   ` Sagi Grimberg
2018-10-31 14:32     ` Jens Axboe
2018-10-30 18:32 ` [PATCH 15/16] block: add REQ_HIPRI and inherit it from IOCB_HIPRI Jens Axboe
2018-10-31  1:58   ` Sagi Grimberg
2018-10-30 18:32 ` [PATCH 16/16] nvme: add separate poll queue map Jens Axboe
2018-10-30 18:35 ` [PATCHSET v3 0/16] blk-mq: Add support for multiple queue maps Keith Busch

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).