[PATCH v1 0/4] Expose blk-mq tag sets through debugfs

[PATCH v1 0/4] Expose blk-mq tag sets through debugfs

[Lei Chen]

Block debugfs provides views of request queues and hardware contexts, but
not the tag set they share. This series adds a tag set view for inspecting
configuration, CPU mappings and driver tags, with links from request queue
directories to identify queues using the same tag set.

The series first removes a stale CPU mapping comment and moves driver tag
allocation and reclamation from individual request queue initialization to
the tag set level. This gives the debugfs tag files a lifetime that can be
coordinated with tag allocation and release. It then adds tagset/<id>
directories under the debugfs root and links each request queue to its tag
set directory.

Tag reads hold a debugfs active reference only while copying tag fields to
local variables. Releasing it before copying to userspace lets tag file
removal complete even if a reader faults on its user buffer and needs I/O
to a frozen queue.

Based on Linux v7.3-rc2.

Testing on a 192-CPU x86_64 host with KASAN and lockdep enabled included:

  - null_blk queue growth and shrink under concurrent direct I/O and
    debugfs reads, including 100 cycles between 1 and 192 queues.
  - Shared tag sets, shared tag bitmaps, three-map configurations, and
    repeated device teardown and recreation with old file descriptors.
  - Unmapped tag reclamation with 256 poll queues: tracing confirmed
    allocation, reclamation, reallocation and reclamation again for
    indices that became unmapped as the queue count changed.
  - Task-local allocation fault injection, including hctx allocation
    failures that rolled the hardware queue count back.
  - NVMe loop with two namespaces, 50 controller resets and data-checked
    direct I/O. These resets retained the same hardware queue count.
  - userfaultfd-controlled stalls in tag reads: null_blk queue resizing
    and NVMe loop controller deletion completed while copying to the
    user buffer was blocked.

No KASAN or lockdep reports were observed in these tests. A hardware queue
count change with NVMe queues already frozen has not been tested.

Lei Chen (4):
  blk-mq: remove stale comment about non-present CPU mapping
  blk-mq: manage driver tags at the tag set level
  block: expose blk-mq tag sets through debugfs
  block: link request queue debugfs directories to tag sets

 block/blk-mq-debugfs.c | 298 +++++++++++++++++++++++++++++++++++++++++
 block/blk-mq-debugfs.h |  22 +++
 block/blk-mq.c         | 143 +++++++++++++++-----
 include/linux/blk-mq.h |  13 ++
 4 files changed, 445 insertions(+), 31 deletions(-)

-- 
2.43.0

[PATCH v1 1/4] blk-mq: remove stale comment about non-present CPU mapping

[Lei Chen]

Commit 20e4d8139319 ("blk-mq: simplify queue mapping & schedule with each
possisble CPU") changed blk_mq_map_swqueue() to iterate over all possible
CPUs instead of only present CPUs, but left the comment claiming that
non-present CPUs are mapped to the first hctx.

The function now uses the tag set's CPU-to-hardware-queue maps for all
possible CPUs, without checking whether a CPU is present. Non-present
CPUs are therefore not necessarily mapped to hctx0. The fallback to
hctx0 in this function handles tag allocation failure, independently
of CPU presence.

Remove the outdated statement and retain the description of the mapping
loop.

Signed-off-by: Lei Chen <lei.chen@smartx.com>
---
 block/blk-mq.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index a26a11c73ee3..59452b855afe 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -4173,11 +4173,7 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 		hctx->dispatch_from = NULL;
 	}
 
-	/*
-	 * Map software to hardware queues.
-	 *
-	 * If the cpu isn't present, the cpu is mapped to first hctx.
-	 */
+	/* Map software to hardware queues. */
 	for_each_possible_cpu(i) {
 
 		ctx = per_cpu_ptr(q->queue_ctx, i);
-- 
2.43.0

[PATCH v1 2/4] blk-mq: manage driver tags at the tag set level

[Lei Chen]

blk_mq_map_swqueue() allocates and frees driver tags while initializing
an individual request queue, even though those tags are shared by all
queues using the tag set. Which hardware queues need tags is determined
by the tag set's CPU maps, so manage their lifetime at that level.

Move tag allocation and reclamation out of blk_mq_map_swqueue(). Reclaim
unmapped tags before returning a newly allocated tag set. When updating
the hardware queue count, allocate missing tags after rebuilding the CPU
maps and before rebuilding hardware contexts. Free unused tags only after
all request queues have been remapped, while they are still frozen, since
old hardware contexts may still reference them during teardown. Serialize
these runtime changes with update_nr_hwq_lock.

Track hardware queue indices referenced by the CPU maps in a bitmap to
avoid rescanning every CPU and map for each tag being reclaimed. Keep
queue 0's tags for allocation failure fallback and rebuild the bitmap
when fallback changes the mappings.

This leaves request queue initialization to associate software and
hardware contexts without changing the shared driver tags or CPU maps.

Signed-off-by: Lei Chen <lei.chen@smartx.com>
---
 block/blk-mq.c         | 129 ++++++++++++++++++++++++++++++++---------
 include/linux/blk-mq.h |   5 ++
 2 files changed, 108 insertions(+), 26 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 59452b855afe..12e28d779e2c 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -9,6 +9,7 @@
 #include <linux/module.h>
 #include <linux/backing-dev.h>
 #include <linux/bio.h>
+#include <linux/bitmap.h>
 #include <linux/blkdev.h>
 #include <linux/blk-integrity.h>
 #include <linux/kmemleak.h>
@@ -4159,9 +4160,74 @@ static void __blk_mq_free_map_and_rqs(struct blk_mq_tag_set *set,
 	set->tags[hctx_idx] = NULL;
 }
 
+static void blk_mq_update_mapped_tags_bitmap(struct blk_mq_tag_set *set)
+{
+	unsigned int i, cpu;
+
+	bitmap_zero(set->mapped_tags_bitmap, set->nr_hw_queues);
+	if (set->nr_hw_queues == 1) {
+		__set_bit(0, set->mapped_tags_bitmap);
+		return;
+	}
+
+	for (i = 0; i < set->nr_maps; i++) {
+		struct blk_mq_queue_map *map = &set->map[i];
+
+		if (!map->nr_queues)
+			continue;
+		for_each_possible_cpu(cpu)
+			__set_bit(map->mq_map[cpu], set->mapped_tags_bitmap);
+	}
+}
+
+static void blk_mq_alloc_mapped_tags(struct blk_mq_tag_set *set)
+{
+	unsigned int cpu, i;
+	bool remapped = false;
+
+	lockdep_assert_held_write(&set->update_nr_hwq_lock);
+
+	for_each_possible_cpu(cpu) {
+		for (i = 0; i < set->nr_maps; i++) {
+			struct blk_mq_queue_map *map = &set->map[i];
+			unsigned int hctx_idx;
+
+			if (!map->nr_queues)
+				continue;
+			hctx_idx = map->mq_map[cpu];
+			if (!set->tags[hctx_idx] &&
+			    !__blk_mq_alloc_map_and_rqs(set, hctx_idx)) {
+				/* Queue 0 always has tags available for fallback. */
+				map->mq_map[cpu] = 0;
+				remapped = true;
+			}
+		}
+	}
+	if (remapped)
+		blk_mq_update_mapped_tags_bitmap(set);
+}
+
+static bool blk_mq_tagset_tags_mapped(struct blk_mq_tag_set *set,
+				      unsigned int hctx_idx)
+{
+	return test_bit(hctx_idx, set->mapped_tags_bitmap);
+}
+
+/* The queue maps must be stable and the unused tags no longer in use. */
+static void blk_mq_free_unmapped_tags(struct blk_mq_tag_set *set)
+{
+	unsigned int i;
+
+	/* Keep queue 0 as a fallback if a later tag allocation fails. */
+	for (i = 1; i < set->nr_hw_queues; i++) {
+		if (set->tags[i] && !blk_mq_tagset_tags_mapped(set, i))
+			__blk_mq_free_map_and_rqs(set, i);
+	}
+}
+
 static void blk_mq_map_swqueue(struct request_queue *q)
 {
-	unsigned int j, hctx_idx;
+	unsigned int j;
 	unsigned long i;
 	struct blk_mq_hw_ctx *hctx;
 	struct blk_mq_ctx *ctx;
@@ -4183,19 +4249,6 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 						HCTX_TYPE_DEFAULT, i);
 				continue;
 			}
-			hctx_idx = set->map[j].mq_map[i];
-			/* unmapped hw queue can be remapped after CPU topo changed */
-			if (!set->tags[hctx_idx] &&
-			    !__blk_mq_alloc_map_and_rqs(set, hctx_idx)) {
-				/*
-				 * If tags initialization fail for some hctx,
-				 * that hctx won't be brought online.  In this
-				 * case, remap the current ctx to hctx[0] which
-				 * is guaranteed to always have tags allocated
-				 */
-				set->map[j].mq_map[i] = 0;
-			}
-
 			hctx = blk_mq_map_queue_type(q, j, i);
 			ctx->hctxs[j] = hctx;
 			/*
@@ -4226,18 +4279,8 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 	queue_for_each_hw_ctx(q, hctx, i) {
 		int cpu;
 
-		/*
-		 * If no software queues are mapped to this hardware queue,
-		 * disable it and free the request entries.
-		 */
+		/* Disable hardware queues with no mapped software queues. */
 		if (!hctx->nr_ctx) {
-			/* Never unmap queue 0.  We need it as a
-			 * fallback in case of a new remap fails
-			 * allocation
-			 */
-			if (i)
-				__blk_mq_free_map_and_rqs(set, i);
-
 			hctx->tags = NULL;
 			continue;
 		}
@@ -4778,12 +4821,18 @@ static void blk_mq_update_queue_map(struct blk_mq_tag_set *set)
 		BUG_ON(set->nr_maps > 1);
 		blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
 	}
+	blk_mq_update_mapped_tags_bitmap(set);
 }
 
+/*
+ * On successful growth, also install a larger mapped_tags_bitmap. The caller
+ * rebuilds its contents when updating the queue maps under update_nr_hwq_lock.
+ */
 static struct blk_mq_tags **blk_mq_prealloc_tag_set_tags(
 				struct blk_mq_tag_set *set,
 				int new_nr_hw_queues)
 {
+	unsigned long *new_mapped_tags_bitmap;
 	struct blk_mq_tags **new_tags;
 	int i;
 
@@ -4795,6 +4844,13 @@ static struct blk_mq_tags **blk_mq_prealloc_tag_set_tags(
 	if (!new_tags)
 		return ERR_PTR(-ENOMEM);
 
+	new_mapped_tags_bitmap = bitmap_zalloc_node(new_nr_hw_queues, GFP_KERNEL,
+						    set->numa_node);
+	if (!new_mapped_tags_bitmap) {
+		kfree(new_tags);
+		return ERR_PTR(-ENOMEM);
+	}
+
 	if (set->tags)
 		memcpy(new_tags, set->tags, set->nr_hw_queues *
 		       sizeof(*set->tags));
@@ -4811,12 +4867,15 @@ static struct blk_mq_tags **blk_mq_prealloc_tag_set_tags(
 		cond_resched();
 	}
 
+	bitmap_free(set->mapped_tags_bitmap);
+	set->mapped_tags_bitmap = new_mapped_tags_bitmap;
 	return new_tags;
 out_unwind:
 	while (--i >= set->nr_hw_queues) {
 		if (!blk_mq_is_shared_tags(set->flags))
 			blk_mq_free_map_and_rqs(set, new_tags[i], i);
 	}
+	bitmap_free(new_mapped_tags_bitmap);
 	kfree(new_tags);
 	return ERR_PTR(-ENOMEM);
 }
@@ -4880,9 +4939,17 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		if (ret)
 			goto out_free_srcu;
 	}
+
+	set->mapped_tags_bitmap = bitmap_zalloc_node(set->nr_hw_queues, GFP_KERNEL,
+						     set->numa_node);
+	if (!set->mapped_tags_bitmap) {
+		ret = -ENOMEM;
+		goto out_cleanup_srcu;
+	}
+
 	ret = init_srcu_struct(&set->tags_srcu);
 	if (ret)
-		goto out_cleanup_srcu;
+		goto out_free_mapped_tags_bitmap;
 
 	init_rwsem(&set->update_nr_hwq_lock);
 
@@ -4907,6 +4974,7 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 	ret = blk_mq_alloc_set_map_and_rqs(set);
 	if (ret)
 		goto out_free_mq_map;
+	blk_mq_free_unmapped_tags(set);
 
 	mutex_init(&set->tag_list_lock);
 	INIT_LIST_HEAD(&set->tag_list);
@@ -4922,6 +4990,9 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 	set->tags = NULL;
 out_cleanup_tags_srcu:
 	cleanup_srcu_struct(&set->tags_srcu);
+out_free_mapped_tags_bitmap:
+	bitmap_free(set->mapped_tags_bitmap);
+	set->mapped_tags_bitmap = NULL;
 out_cleanup_srcu:
 	if (set->flags & BLK_MQ_F_BLOCKING)
 		cleanup_srcu_struct(set->srcu);
@@ -4968,6 +5039,9 @@ void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
 	kfree(set->tags);
 	set->tags = NULL;
 
+	bitmap_free(set->mapped_tags_bitmap);
+	set->mapped_tags_bitmap = NULL;
+
 	srcu_barrier(&set->tags_srcu);
 	cleanup_srcu_struct(&set->tags_srcu);
 	if (set->flags & BLK_MQ_F_BLOCKING) {
@@ -5158,6 +5232,7 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 
 fallback:
 	blk_mq_update_queue_map(set);
+	blk_mq_alloc_mapped_tags(set);
 	list_for_each_entry(q, &set->tag_list, tag_set_list) {
 		__blk_mq_realloc_hw_ctxs(set, q);
 
@@ -5174,6 +5249,8 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 		}
 		blk_mq_map_swqueue(q);
 	}
+	/* All queues have stopped using tags excluded by the new maps. */
+	blk_mq_free_unmapped_tags(set);
 switch_back:
 	/* The blk_mq_elv_switch_back unfreezes queue for us. */
 	list_for_each_entry(q, &set->tag_list, tag_set_list) {
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index af878597afb8..3ef989dc4f99 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -517,6 +517,10 @@ enum hctx_type {
  *		   tag set.
  * @tags:	   Tag sets. One tag set per hardware queue. Has @nr_hw_queues
  *		   elements.
+ * @mapped_tags_bitmap: Bitmap of hardware queue indices referenced by active
+ *		   CPU maps. Rebuilt before publication or with update_nr_hwq_lock
+ *		   held for writing. Tags for index 0 are retained as a fallback
+ *		   even when its bit is clear.
  * @shared_tags:
  *		   Shared set of tags. Has @nr_hw_queues elements. If set,
  *		   shared by all @tags.
@@ -545,6 +549,7 @@ struct blk_mq_tag_set {
 	void			*driver_data;
 
 	struct blk_mq_tags	**tags;
+	unsigned long		*mapped_tags_bitmap;
 
 	struct blk_mq_tags	*shared_tags;
 
-- 
2.43.0

[PATCH v1 3/4] block: expose blk-mq tag sets through debugfs

[Lei Chen]

Block debugfs exposes request queues and hardware contexts, but lacks a
view of the tag set they share. Add a tagset/<id> directory under the
debugfs root to expose tag set configuration, flags, CPU-to-hardware-queue
mappings and tag counts. Represent hardware queues using shared tags with
symlinks to the shared_tags file.

Tie the directory lifetime to tag set allocation and teardown, and rebuild
the tags directory when the hardware queue count changes. Remove tag files
before freeing the objects they reference.

Hardware queue updates may run with queues frozen, so removing tag files
must not wait for a reader that needs I/O to those queues. Use a custom
read operation with debugfs_create_file_unsafe() to hold a debugfs active
reference only while copying tag fields into local variables. Drop the
reference before formatting the output and copying it to userspace, where
a page fault may require I/O. The removal-before-free ordering protects
the tag objects during this brief access; reads after removal fail before
dereferencing them.

Serialize other tag set attribute reads with hardware queue updates using
update_nr_hwq_lock. Use a NOIO allocation scope during debugfs registration
to prevent memory reclaim from issuing I/O to frozen queues.

Signed-off-by: Lei Chen <lei.chen@smartx.com>
---
 block/blk-mq-debugfs.c | 290 +++++++++++++++++++++++++++++++++++++++++
 block/blk-mq-debugfs.h |  22 ++++
 block/blk-mq.c         |   8 ++
 include/linux/blk-mq.h |   8 ++
 4 files changed, 328 insertions(+)

diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index 6754d8f9449c..298387607049 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -7,6 +7,7 @@
 #include <linux/blkdev.h>
 #include <linux/build_bug.h>
 #include <linux/debugfs.h>
+#include <linux/idr.h>
 
 #include "blk.h"
 #include "blk-mq.h"
@@ -827,3 +828,292 @@ void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
 	debugfs_remove_recursive(hctx->sched_debugfs_dir);
 	hctx->sched_debugfs_dir = NULL;
 }
+
+/* Tag set debugfs --------------------------------------------------------- */
+
+static DEFINE_IDA(blk_mq_tagset_debugfs_ida);
+static DEFINE_MUTEX(blk_mq_tagset_debugfs_mutex);
+static struct dentry *blk_mq_tagset_debugfs_root;
+
+/* Prevent debugfs allocation reclaim from issuing I/O to frozen queues. */
+static unsigned int __must_check blk_mq_tagset_debugfs_lock(void)
+	__acquires(&blk_mq_tagset_debugfs_mutex)
+{
+	unsigned int memflags = memalloc_noio_save();
+
+	mutex_lock(&blk_mq_tagset_debugfs_mutex);
+	return memflags;
+}
+
+static void blk_mq_tagset_debugfs_unlock(unsigned int memflags)
+	__releases(&blk_mq_tagset_debugfs_mutex)
+{
+	mutex_unlock(&blk_mq_tagset_debugfs_mutex);
+	memalloc_noio_restore(memflags);
+}
+
+#define BLK_MQ_F_NAME(name) \
+	[ilog2(BLK_MQ_F_##name)] = #name
+static const char *const tagset_flag_name[] = {
+	BLK_MQ_F_NAME(TAG_QUEUE_SHARED),
+	BLK_MQ_F_NAME(STACKING),
+	BLK_MQ_F_NAME(TAG_HCTX_SHARED),
+	BLK_MQ_F_NAME(BLOCKING),
+	BLK_MQ_F_NAME(TAG_RR),
+	BLK_MQ_F_NAME(NO_SCHED_BY_DEFAULT),
+};
+
+#undef BLK_MQ_F_NAME
+
+static int blk_mq_tagset_flags_show(struct seq_file *m, void *v)
+{
+	struct blk_mq_tag_set *set = m->private;
+
+	BUILD_BUG_ON(ARRAY_SIZE(tagset_flag_name) != ilog2(BLK_MQ_F_MAX));
+
+	down_read(&set->update_nr_hwq_lock);
+
+	blk_flags_show(m, set->flags, tagset_flag_name,
+		       ARRAY_SIZE(tagset_flag_name));
+	seq_putc(m, '\n');
+
+	up_read(&set->update_nr_hwq_lock);
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(blk_mq_tagset_flags);
+
+#define BLK_MQ_TAGSET_VALUE_SHOW(_name, _type, _value, _format) \
+static int blk_mq_tagset_##_name##_show(struct seq_file *m, void *v) \
+{ \
+	struct blk_mq_tag_set *set = m->private; \
+	_type value; \
+\
+	down_read(&set->update_nr_hwq_lock); \
+	value = (_value); \
+	up_read(&set->update_nr_hwq_lock); \
+\
+	seq_printf(m, _format, value); \
+	return 0; \
+} \
+DEFINE_SHOW_ATTRIBUTE(blk_mq_tagset_##_name)
+
+BLK_MQ_TAGSET_VALUE_SHOW(nr_maps, unsigned int, set->nr_maps, "%u\n");
+BLK_MQ_TAGSET_VALUE_SHOW(nr_hw_queues, unsigned int, set->nr_hw_queues, "%u\n");
+BLK_MQ_TAGSET_VALUE_SHOW(queue_depth, unsigned int, set->queue_depth, "%u\n");
+BLK_MQ_TAGSET_VALUE_SHOW(reserved_tags, unsigned int, set->reserved_tags, "%u\n");
+BLK_MQ_TAGSET_VALUE_SHOW(cmd_size, unsigned int, set->cmd_size, "%u\n");
+BLK_MQ_TAGSET_VALUE_SHOW(numa_node, int, set->numa_node, "%d\n");
+BLK_MQ_TAGSET_VALUE_SHOW(timeout, unsigned int, set->timeout, "%u\n");
+
+#undef BLK_MQ_TAGSET_VALUE_SHOW
+
+static int blk_mq_tagset_hctx_tags_shared_show(struct seq_file *m, void *v)
+{
+	struct blk_mq_tag_set *set = m->private;
+	bool shared;
+
+	down_read(&set->update_nr_hwq_lock);
+	shared = blk_mq_is_shared_tags(set->flags);
+	up_read(&set->update_nr_hwq_lock);
+
+	seq_printf(m, "%d\n", shared);
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(blk_mq_tagset_hctx_tags_shared);
+
+static int blk_mq_tagset_map_show(struct seq_file *m, void *v)
+{
+	struct blk_mq_tag_set *set = m->private;
+	unsigned long index = debugfs_get_aux_num(m->file);
+	unsigned int map = index / nr_cpu_ids;
+	unsigned int cpu = index % nr_cpu_ids;
+
+	down_read(&set->update_nr_hwq_lock);
+
+	seq_printf(m, "%u\n", set->map[map].mq_map[cpu]);
+
+	up_read(&set->update_nr_hwq_lock);
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(blk_mq_tagset_map);
+
+static void blk_mq_tagset_debugfs_create_maps(struct blk_mq_tag_set *set)
+{
+	struct dentry *dir, *map_dir;
+	unsigned int cpu, i;
+	char name[20];
+
+	dir = debugfs_create_dir("map", set->debugfs_dir);
+	if (IS_ERR_OR_NULL(dir))
+		return;
+
+	for (i = 0; i < set->nr_maps; i++) {
+		snprintf(name, sizeof(name), "%u", i);
+		map_dir = debugfs_create_dir(name, dir);
+		if (IS_ERR_OR_NULL(map_dir))
+			continue;
+
+		for_each_possible_cpu(cpu) {
+			/* Encode the map and CPU indices in the auxiliary data. */
+			unsigned long index = (unsigned long)i * nr_cpu_ids + cpu;
+
+			snprintf(name, sizeof(name), "cpu%u", cpu);
+			debugfs_create_file_aux_num(name, 0444, map_dir, set,
+						    index, &blk_mq_tagset_map_fops);
+		}
+	}
+}
+
+static ssize_t blk_mq_tagset_tags_read(struct file *file, char __user *user_buf,
+				       size_t count, loff_t *ppos)
+{
+	struct dentry *dentry = file->f_path.dentry;
+	unsigned int nr_tags, nr_reserved_tags, active_queues;
+	struct blk_mq_tags *tags;
+	char buf[128];
+	int ret, len;
+
+	/*
+	 * Tag updates remove these files before freeing the tags. Shared tags
+	 * remain valid until tag set teardown removes the shared_tags file.
+	 */
+	ret = debugfs_file_get(dentry);
+	if (ret)
+		return ret;
+
+	tags = file->private_data;
+	if (!tags) {
+		debugfs_file_put(dentry);
+		return -ENODEV;
+	}
+
+	nr_tags = tags->nr_tags;
+	nr_reserved_tags = tags->nr_reserved_tags;
+	active_queues = READ_ONCE(tags->active_queues);
+	debugfs_file_put(dentry);
+
+	/*
+	 * A fault on the user buffer may need I/O to a frozen queue. Drop the
+	 * active reference first so file removal does not wait for that I/O.
+	 */
+	len = scnprintf(buf, sizeof(buf),
+			".nr_tags=%u\n.nr_reserved_tags=%u\n.active_queues=%u\n",
+			nr_tags, nr_reserved_tags, active_queues);
+	return simple_read_from_buffer(user_buf, count, ppos, buf, len);
+}
+
+static const struct file_operations blk_mq_tagset_tags_fops = {
+	.owner		= THIS_MODULE,
+	.open		= simple_open,
+	.read		= blk_mq_tagset_tags_read,
+	.llseek		= default_llseek,
+};
+
+void blk_mq_tagset_debugfs_remove_tags(struct blk_mq_tag_set *set)
+{
+	if (IS_ERR_OR_NULL(set->debugfs_dir))
+		return;
+
+	debugfs_lookup_and_remove("tags", set->debugfs_dir);
+}
+
+void blk_mq_tagset_debugfs_create_tags(struct blk_mq_tag_set *set)
+{
+	struct dentry *dir;
+	unsigned int i;
+	char name[20];
+
+	if (IS_ERR_OR_NULL(set->debugfs_dir))
+		return;
+
+	dir = debugfs_create_dir("tags", set->debugfs_dir);
+	if (IS_ERR_OR_NULL(dir))
+		return;
+
+	for (i = 0; i < set->nr_hw_queues; i++) {
+		struct blk_mq_tags *tags = set->tags[i];
+
+		snprintf(name, sizeof(name), "%u", i);
+		if (tags && tags == set->shared_tags)
+			debugfs_create_symlink(name, dir, "../shared_tags");
+		else
+			debugfs_create_file_unsafe(name, 0444, dir, tags,
+						   &blk_mq_tagset_tags_fops);
+	}
+}
+
+static void blk_mq_tagset_debugfs_create_files(struct blk_mq_tag_set *set)
+{
+	debugfs_create_file("nr_maps", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_nr_maps_fops);
+	debugfs_create_file("nr_hw_queues", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_nr_hw_queues_fops);
+	debugfs_create_file("queue_depth", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_queue_depth_fops);
+	debugfs_create_file("reserved_tags", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_reserved_tags_fops);
+	debugfs_create_file("cmd_size", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_cmd_size_fops);
+	debugfs_create_file("numa_node", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_numa_node_fops);
+	debugfs_create_file("timeout", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_timeout_fops);
+	debugfs_create_file("hctx_tags_shared", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_hctx_tags_shared_fops);
+	debugfs_create_file("flags", 0444, set->debugfs_dir, set,
+			    &blk_mq_tagset_flags_fops);
+	blk_mq_tagset_debugfs_create_maps(set);
+	debugfs_create_file_unsafe("shared_tags", 0444, set->debugfs_dir,
+				   set->shared_tags, &blk_mq_tagset_tags_fops);
+	blk_mq_tagset_debugfs_create_tags(set);
+}
+
+void blk_mq_tagset_debugfs_register(struct blk_mq_tag_set *set)
+{
+	struct dentry *dir;
+	unsigned int memflags;
+	char name[24];
+	int id;
+
+	memflags = blk_mq_tagset_debugfs_lock();
+
+	if (IS_ERR_OR_NULL(blk_mq_tagset_debugfs_root))
+		blk_mq_tagset_debugfs_root = debugfs_create_dir("tagset", NULL);
+
+	if (set->debugfs_dir || IS_ERR_OR_NULL(blk_mq_tagset_debugfs_root))
+		goto out_unlock;
+
+	id = ida_alloc(&blk_mq_tagset_debugfs_ida, GFP_KERNEL);
+	if (id < 0)
+		goto out_unlock;
+
+	snprintf(name, sizeof(name), "%d", id);
+	dir = debugfs_create_dir(name, blk_mq_tagset_debugfs_root);
+	if (IS_ERR_OR_NULL(dir)) {
+		ida_free(&blk_mq_tagset_debugfs_ida, id);
+		goto out_unlock;
+	}
+
+	set->debugfs_id = id;
+	set->debugfs_dir = dir;
+	blk_mq_tagset_debugfs_create_files(set);
+
+out_unlock:
+	blk_mq_tagset_debugfs_unlock(memflags);
+}
+
+void blk_mq_tagset_debugfs_unregister(struct blk_mq_tag_set *set)
+{
+	mutex_lock(&blk_mq_tagset_debugfs_mutex);
+
+	debugfs_remove_recursive(set->debugfs_dir);
+	set->debugfs_dir = NULL;
+
+	if (set->debugfs_id >= 0) {
+		ida_free(&blk_mq_tagset_debugfs_ida, set->debugfs_id);
+		set->debugfs_id = -1;
+	}
+
+	mutex_unlock(&blk_mq_tagset_debugfs_mutex);
+}
diff --git a/block/blk-mq-debugfs.h b/block/blk-mq-debugfs.h
index 49bb1aaa83dc..aea278bdc4e7 100644
--- a/block/blk-mq-debugfs.h
+++ b/block/blk-mq-debugfs.h
@@ -7,6 +7,7 @@
 #include <linux/seq_file.h>
 
 struct blk_mq_hw_ctx;
+struct blk_mq_tag_set;
 
 struct blk_mq_debugfs_attr {
 	const char *name;
@@ -34,6 +35,11 @@ void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx);
 
 void blk_mq_debugfs_register_rq_qos(struct request_queue *q);
+
+void blk_mq_tagset_debugfs_register(struct blk_mq_tag_set *set);
+void blk_mq_tagset_debugfs_unregister(struct blk_mq_tag_set *set);
+void blk_mq_tagset_debugfs_create_tags(struct blk_mq_tag_set *set);
+void blk_mq_tagset_debugfs_remove_tags(struct blk_mq_tag_set *set);
 #else
 static inline void blk_mq_debugfs_register(struct request_queue *q)
 {
@@ -77,6 +83,22 @@ static inline void blk_mq_debugfs_register_rq_qos(struct request_queue *q)
 {
 }
 
+static inline void blk_mq_tagset_debugfs_register(struct blk_mq_tag_set *set)
+{
+}
+
+static inline void blk_mq_tagset_debugfs_unregister(struct blk_mq_tag_set *set)
+{
+}
+
+static inline void blk_mq_tagset_debugfs_create_tags(struct blk_mq_tag_set *set)
+{
+}
+
+static inline void blk_mq_tagset_debugfs_remove_tags(struct blk_mq_tag_set *set)
+{
+}
+
 #endif
 
 #if defined(CONFIG_BLK_DEV_ZONED) && defined(CONFIG_BLK_DEBUG_FS)
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 12e28d779e2c..f32d2df044dc 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -4978,6 +4978,10 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 
 	mutex_init(&set->tag_list_lock);
 	INIT_LIST_HEAD(&set->tag_list);
+#ifdef CONFIG_BLK_DEBUG_FS
+	set->debugfs_id = -1;
+#endif
+	blk_mq_tagset_debugfs_register(set);
 
 	return 0;
 
@@ -5023,6 +5027,8 @@ void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
 {
 	int i, j;
 
+	blk_mq_tagset_debugfs_unregister(set);
+
 	for (i = 0; i < set->nr_hw_queues; i++)
 		__blk_mq_free_map_and_rqs(set, i);
 
@@ -5207,6 +5213,7 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 		blk_mq_debugfs_unregister_hctxs(q);
 		blk_mq_sysfs_unregister_hctxs(q);
 	}
+	blk_mq_tagset_debugfs_remove_tags(set);
 
 	/*
 	 * Switch IO scheduler to 'none', cleaning up the data associated
@@ -5267,6 +5274,7 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 		blk_mq_remove_hw_queues_cpuhp(q);
 		blk_mq_add_hw_queues_cpuhp(q);
 	}
+	blk_mq_tagset_debugfs_create_tags(set);
 
 out_free_ctx:
 	blk_mq_free_sched_ctx_batch(&elv_tbl);
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 3ef989dc4f99..a80bc63a6e71 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -14,6 +14,7 @@
 struct blk_mq_tags;
 struct blk_flush_queue;
 struct io_comp_batch;
+struct dentry;
 
 #define BLKDEV_MIN_RQ	4
 #define BLKDEV_DEFAULT_RQ	128
@@ -534,6 +535,8 @@ enum hctx_type {
  * @update_nr_hwq_lock:
  * 		   Synchronize updating nr_hw_queues with add/del disk &
  * 		   switching elevator.
+ * @debugfs_dir:   Debugfs directory for this tag set.
+ * @debugfs_id:    ID used as part of the debugfs directory name.
  */
 struct blk_mq_tag_set {
 	const struct blk_mq_ops	*ops;
@@ -559,6 +562,11 @@ struct blk_mq_tag_set {
 	struct srcu_struct	tags_srcu;
 
 	struct rw_semaphore	update_nr_hwq_lock;
+
+#ifdef CONFIG_BLK_DEBUG_FS
+	struct dentry		*debugfs_dir;
+	int			debugfs_id;
+#endif
 };
 
 /**
-- 
2.43.0

[PATCH v1 4/4] block: link request queue debugfs directories to tag sets

[Lei Chen]

Tag set debugfs directories use numeric IDs that do not identify their
attached disks. Starting from a disk's debugfs directory therefore gives
no direct way to find its tag set configuration and CPU mappings.

Add a tagset symlink in each blk-mq request queue's debugfs directory,
pointing to ../../tagset/<id>. Queues sharing a tag set link to the same
directory. Create the link only when both debugfs directories exist.

Signed-off-by: Lei Chen <lei.chen@smartx.com>
---
 block/blk-mq-debugfs.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index 298387607049..78a91a1095bb 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -644,11 +644,19 @@ static void debugfs_create_files(struct request_queue *q, struct dentry *parent,
 
 void blk_mq_debugfs_register(struct request_queue *q)
 {
+	struct blk_mq_tag_set *set = q->tag_set;
 	struct blk_mq_hw_ctx *hctx;
 	unsigned long i;
+	char target[32];
 
 	debugfs_create_files(q, q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
 
+	if (!IS_ERR_OR_NULL(q->debugfs_dir) &&
+	    !IS_ERR_OR_NULL(set->debugfs_dir) && set->debugfs_id >= 0) {
+		snprintf(target, sizeof(target), "../../tagset/%d", set->debugfs_id);
+		debugfs_create_symlink("tagset", q->debugfs_dir, target);
+	}
+
 	queue_for_each_hw_ctx(q, hctx, i) {
 		if (!hctx->debugfs_dir)
 			blk_mq_debugfs_register_hctx(q, hctx);
-- 
2.43.0