Skip to contentSkip to conversation

mm: slub: move sysfs slab alloc/free interfaces to debugfs

PATCH Superseded Faiyaz Mohammed 3 participants 7 messages May 18, 2021 – May 21, 2021

Subsystems (2)

v6 of 12 Superseded Latest is v12 →

lore
Share & tools
lore
#0 lore raw

PATCH v6 1/1 mm: slub: move sysfs slab alloc/free interfaces to debugfs

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

---
 include/linux/slub_def.h |   8 ++
 mm/slab_common.c         |   9 ++
 mm/slub.c                | 359 ++++++++++++++++++++++++++++++++++++-----------
 3 files changed, 294 insertions(+), 82 deletions(-)
include/linux/slub_def.h +8 -0
index dcde82a..b413ebe 100644
@@ -159,6 +159,14 @@ static inline void sysfs_slab_release(struct kmem_cache *s)
 }
 #endif
 
+#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
+#define SLAB_SUPPORTS_DEBUGFS
+void debugfs_slab_release(struct kmem_cache *);
+#else
+static inline void debugfs_slab_release(struct kmem_cache *s)
+{
+}
+#endif
 void object_err(struct kmem_cache *s, struct page *page,
 		u8 *object, char *reason);
 
mm/slab_common.c +9 -0
index a4a5714..873dd79 100644
@@ -455,6 +455,9 @@ static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
 #else
 		slab_kmem_cache_release(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 	}
 }
 
@@ -472,6 +475,9 @@ static int shutdown_cache(struct kmem_cache *s)
 #ifdef SLAB_SUPPORTS_SYSFS
 		sysfs_slab_unlink(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
 		schedule_work(&slab_caches_to_rcu_destroy_work);
 	} else {
@@ -482,6 +488,9 @@ static int shutdown_cache(struct kmem_cache *s)
 #else
 		slab_kmem_cache_release(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 	}
 
 	return 0;
mm/slub.c +277 -82

418 lines not on this page · raw message

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation
#1 lore raw
Quoted 43 lines · Faiyaz Mohammed

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

---
 include/linux/slub_def.h |   8 ++
 mm/slab_common.c         |   9 ++
 mm/slub.c                | 359 ++++++++++++++++++++++++++++++++++++-----------
 3 files changed, 294 insertions(+), 82 deletions(-)
include/linux/slub_def.h +8 -0
index dcde82a..b413ebe 100644
@@ -159,6 +159,14 @@ static inline void sysfs_slab_release(struct kmem_cache *s)
 }
 #endif
 
+#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
+#define SLAB_SUPPORTS_DEBUGFS
+void debugfs_slab_release(struct kmem_cache *);
+#else
+static inline void debugfs_slab_release(struct kmem_cache *s)
+{
+}
+#endif
 void object_err(struct kmem_cache *s, struct page *page,
 		u8 *object, char *reason);
 
mm/slab_common.c +0 -0
index a4a5714..873dd79 100644

mm/slab_common.c static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) · line 460

Quoted 6 lines
mm/slab_common.c +3 -0
@@ -455,6 +455,9 @@ static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
 #else
 		slab_kmem_cache_release(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif

If you write your .h file correctly, there's no need for #ifdef in .c files. Please fix this up, it shoudl be trivial given the above changes in the .h file.

Quoted 36 lines
mm/slab_common.c +6 -0
 	}
 }
 
@@ -472,6 +475,9 @@ static int shutdown_cache(struct kmem_cache *s)
 #ifdef SLAB_SUPPORTS_SYSFS
 		sysfs_slab_unlink(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
 		schedule_work(&slab_caches_to_rcu_destroy_work);
 	} else {
@@ -482,6 +488,9 @@ static int shutdown_cache(struct kmem_cache *s)
 #else
 		slab_kmem_cache_release(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 	}
 
 	return 0;
mm/slub.c +1 -0
index 438fa8d..f470a29 100644
@@ -36,6 +36,7 @@
 #include <linux/memcontrol.h>
 #include <linux/random.h>
 
+#include <linux/debugfs.h>
 #include <trace/events/kmem.h>
 
 #include "internal.h"

mm/slub.c static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) · line 231

Quoted 6 lines
mm/slub.c +3 -0
@@ -225,6 +226,15 @@ static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
 							{ return 0; }
 #endif
 
+#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
+static void debugfs_slab_add(struct kmem_cache *);
+static int debugfs_slab_alias(struct kmem_cache *, const char *);

Why does this have to return an int? What are you going to do if it fails?

Quoted 10 lines
mm/slub.c +6 -0
+#else
+static inline void debugfs_slab_add(struct kmem_cache *s) { }
+static inline int debugfs_slab_alias(struct kmem_cache *s, const char *p)
+							{ return 0; }
+#endif
+
 static inline void stat(const struct kmem_cache *s, enum stat_item si)
 {
 #ifdef CONFIG_SLUB_STATS

mm/slub.c __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, · line 4538

Quoted 5 lines
mm/slub.c +2 -0
@@ -4524,6 +4534,8 @@ __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 			s->refcount--;
 			s = NULL;
 		}
+
+		debugfs_slab_alias(s, name);

See, like you do so here.

Quoted 117 lines
mm/slub.c +8 -76
 	}
 
 	return s;
@@ -4545,6 +4557,8 @@ int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
 	if (err)
 		__kmem_cache_release(s);
 
+	debugfs_slab_add(s);
+
 	return err;
 }
 
@@ -4685,6 +4699,8 @@ static long validate_slab_cache(struct kmem_cache *s)
 
 	return count;
 }
+
+#ifdef CONFIG_DEBUG_FS
 /*
  * Generate lists of code addresses where slabcache objects are allocated
  * and freed.
@@ -4708,6 +4724,9 @@ struct loc_track {
 	struct location *loc;
 };
 
+static struct dentry *slab_debugfs_root;
+struct loc_track t = { 0, 0, NULL };
+
 static void free_loc_track(struct loc_track *t)
 {
 	if (t->max)
@@ -4824,82 +4843,7 @@ static void process_slab(struct loc_track *t, struct kmem_cache *s,
 			add_location(t, s, get_track(s, p, alloc));
 	put_map(map);
 }
-
-static int list_locations(struct kmem_cache *s, char *buf,
-			  enum track_item alloc)
-{
-	int len = 0;
-	unsigned long i;
-	struct loc_track t = { 0, 0, NULL };
-	int node;
-	struct kmem_cache_node *n;
-
-	if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
-			     GFP_KERNEL)) {
-		return sysfs_emit(buf, "Out of memory\n");
-	}
-	/* Push back cpu slabs */
-	flush_all(s);
-
-	for_each_kmem_cache_node(s, node, n) {
-		unsigned long flags;
-		struct page *page;
-
-		if (!atomic_long_read(&n->nr_slabs))
-			continue;
-
-		spin_lock_irqsave(&n->list_lock, flags);
-		list_for_each_entry(page, &n->partial, slab_list)
-			process_slab(&t, s, page, alloc);
-		list_for_each_entry(page, &n->full, slab_list)
-			process_slab(&t, s, page, alloc);
-		spin_unlock_irqrestore(&n->list_lock, flags);
-	}
-
-	for (i = 0; i < t.count; i++) {
-		struct location *l = &t.loc[i];
-
-		len += sysfs_emit_at(buf, len, "%7ld ", l->count);
-
-		if (l->addr)
-			len += sysfs_emit_at(buf, len, "%pS", (void *)l->addr);
-		else
-			len += sysfs_emit_at(buf, len, "<not-available>");
-
-		if (l->sum_time != l->min_time)
-			len += sysfs_emit_at(buf, len, " age=%ld/%ld/%ld",
-					     l->min_time,
-					     (long)div_u64(l->sum_time,
-							   l->count),
-					     l->max_time);
-		else
-			len += sysfs_emit_at(buf, len, " age=%ld", l->min_time);
-
-		if (l->min_pid != l->max_pid)
-			len += sysfs_emit_at(buf, len, " pid=%ld-%ld",
-					     l->min_pid, l->max_pid);
-		else
-			len += sysfs_emit_at(buf, len, " pid=%ld",
-					     l->min_pid);
-
-		if (num_online_cpus() > 1 &&
-		    !cpumask_empty(to_cpumask(l->cpus)))
-			len += sysfs_emit_at(buf, len, " cpus=%*pbl",
-					     cpumask_pr_args(to_cpumask(l->cpus)));
-
-		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
-			len += sysfs_emit_at(buf, len, " nodes=%*pbl",
-					     nodemask_pr_args(&l->nodes));
-
-		len += sysfs_emit_at(buf, len, "\n");
-	}
-
-	free_loc_track(&t);
-	if (!t.count)
-		len += sysfs_emit_at(buf, len, "No data\n");
-
-	return len;
-}
+#endif  /* CONFIG_DEBUG_FS   */
 #endif	/* CONFIG_SLUB_DEBUG */
 
 #ifdef SLUB_RESILIENCY_TEST
@@ -5351,17 +5295,25 @@ SLAB_ATTR(validate);
 

mm/slub.c SLAB_ATTR(validate); · line 5302

Quoted 10 lines
mm/slub.c +5 -3
 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
 {
-	if (!(s->flags & SLAB_STORE_USER))
-		return -ENOSYS;
-	return list_locations(s, buf, TRACK_ALLOC);
+	int len = 0;
+
+	len += sprintf(buf + len, "Deprecated, use the equvalent under\n");
+	len += sprintf(buf + len, "<debugfs_mount_point>/slab/%s/alloc_traces\n",
+			s->name);

No, just drop the sysfs file. Userspace ALWAYS has to be able to handle a sysfs file present or not present, that's the way sysfs operates and why we have "one value per file".

If userspace can not handle it, well, someone messed up and printing a huge message like this is not going to help either, just return an error.

Quoted 5 lines
mm/slub.c +2 -0
+
+	return len;
 }
 SLAB_ATTR_RO(alloc_calls);
 

mm/slub.c SLAB_ATTR(validate); · line 5314

Quoted 10 lines
mm/slub.c +5 -3
 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
 {
-	if (!(s->flags & SLAB_STORE_USER))
-		return -ENOSYS;
-	return list_locations(s, buf, TRACK_FREE);
+	int len = 0;
+
+	len += sprintf(buf + len, "Deprecated, use the equvalent under\n");
+	len += sprintf(buf + len, "<debugfs_mount_point>/slab/%s/alloc_traces\n",
+			s->name);

Same here, drop it please.

Quoted 17 lines
mm/slub.c +10 -0
+
+	return len;
 }
 SLAB_ATTR_RO(free_calls);
 #endif /* CONFIG_SLUB_DEBUG */
@@ -5817,6 +5769,249 @@ static int __init slab_sysfs_init(void)
 __initcall(slab_sysfs_init);
 #endif /* CONFIG_SYSFS */
 
+#if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
+static int debugfs_slab_alias(struct kmem_cache *s, const char *name)
+{
+	struct saved_alias *al;
+	struct dentry *slab_cache_dir;
+
+	if (slab_state == FULL) {
+		/*

mm/slub.c static int __init slab_sysfs_init(void) · line 5789

Quoted 10 lines
mm/slub.c +10 -0
+		 * If we have a leftover link then remove it.
+		 */
+		slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+		debugfs_remove(slab_cache_dir);
+		debugfs_create_symlink(name, slab_debugfs_root, NULL);
+	}
+
+	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
+	if (!al)
+		return -ENOMEM;

Why do you care if you have no memory or not?

Quoted 187 lines
mm/slub.c +187 -0
+
+	al->s = s;
+	al->name = name;
+	al->next = alias_list;
+	alias_list = al;
+	return 0;
+}
+
+static int slab_debugfs_show(struct seq_file *seq, void *v)
+{
+	struct location *l;
+	unsigned int idx = *(unsigned int *)v;
+
+	if (idx < t.count) {
+		l = &t.loc[idx];
+
+		seq_printf(seq, "%7ld ", l->count);
+
+		if (l->addr)
+			seq_printf(seq, "%pS", (void *)l->addr);
+		else
+			seq_puts(seq, "<not-available>");
+
+		if (l->sum_time != l->min_time) {
+			seq_printf(seq, " age=%ld/%ld/%ld",
+				l->min_time,
+				(long)div_u64(l->sum_time, l->count),
+				l->max_time);
+		} else
+			seq_printf(seq, " age=%ld",
+				l->min_time);
+
+		if (l->min_pid != l->max_pid)
+			seq_printf(seq, " pid=%ld-%ld",
+				l->min_pid, l->max_pid);
+		else
+			seq_printf(seq, " pid=%ld",
+				l->min_pid);
+
+		if (num_online_cpus() > 1 &&
+				!cpumask_empty(to_cpumask(l->cpus)))
+			seq_printf(seq, " cpus=%*pbl",
+				 cpumask_pr_args(to_cpumask(l->cpus)));
+
+		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
+			seq_printf(seq, " nodes=%*pbl",
+				 nodemask_pr_args(&l->nodes));
+
+		seq_puts(seq, "\n");
+	}
+
+	if (t.count == 0)
+		seq_puts(seq, "No data\n");
+
+	return 0;
+}
+
+static void slab_debugfs_stop(struct seq_file *seq, void *v)
+{
+	if (!v && t.max) {
+		free_loc_track(&t);
+		t.max = 0;
+	}
+}
+
+static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
+{
+	loff_t *spos = v;
+
+	if (*ppos < t.count) {
+		*spos = *spos + 1;
+		*ppos = *spos;
+		return spos;
+	}
+
+	return NULL;
+}
+
+static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
+{
+	struct kmem_cache_node *n;
+	struct kmem_cache *s;
+	enum track_item alloc;
+	int node;
+	loff_t *spos = kmalloc(sizeof(loff_t), GFP_KERNEL);
+
+	s = seq->file->f_inode->i_private;
+
+	if (!spos)
+		return NULL;
+
+	if (!(s->flags & SLAB_STORE_USER)) {
+		kfree(spos);
+		return ERR_PTR(-EOPNOTSUPP);
+	}
+
+	if (*ppos == 0) {
+
+		t.count = 0;
+		t.max = 0;
+		t.loc = NULL;
+		if (strcmp(seq->file->f_path.dentry->d_name.name, "alloc_traces") == 0)
+			alloc =  TRACK_ALLOC;
+		else
+			alloc =  TRACK_FREE;
+
+		if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
+			     GFP_KERNEL)) {
+			seq_puts(seq, "Out of memory\n");
+			kfree(spos);
+			return ERR_PTR(-ENOMEM);
+		}
+		/* Push back cpu slabs */
+		flush_all(s);
+
+		for_each_kmem_cache_node(s, node, n) {
+			unsigned long flags;
+			struct page *page;
+
+			if (!atomic_long_read(&n->nr_slabs))
+				continue;
+
+			spin_lock_irqsave(&n->list_lock, flags);
+			list_for_each_entry(page, &n->partial, slab_list)
+				process_slab(&t, s, page, alloc);
+			list_for_each_entry(page, &n->full, slab_list)
+				process_slab(&t, s, page, alloc);
+			spin_unlock_irqrestore(&n->list_lock, flags);
+		}
+	}
+
+	if (*ppos < t.count) {
+		*spos = *ppos;
+		return spos;
+	}
+
+	kfree(spos);
+	return NULL;
+}
+
+static const struct seq_operations slab_debugfs_sops = {
+	.start  = slab_debugfs_start,
+	.next   = slab_debugfs_next,
+	.stop   = slab_debugfs_stop,
+	.show   = slab_debugfs_show
+};
+DEFINE_SEQ_ATTRIBUTE(slab_debugfs);
+
+static void debugfs_slab_add(struct kmem_cache *s)
+{
+	const char *name;
+	struct dentry *slab_cache_dir;
+	int unmergeable = slab_unmergeable(s);
+
+	if (unlikely(!slab_debugfs_root))
+		return;
+
+	if (!unmergeable && disable_higher_order_debug &&
+			(slub_debug & DEBUG_METADATA_FLAGS))
+		unmergeable = 1;
+
+	if (unmergeable) {
+		/*
+		 * Slabcache can never be merged so we can use the name proper.
+		 * This is typically the case for debug situations. In that
+		 * case we can catch duplicate names easily.
+		 */
+		slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+		debugfs_remove_recursive(slab_cache_dir);
+		name = s->name;
+	} else {
+		/*
+		 * Create a unique name for the slab as a target
+		 * for the symlinks.
+		 */
+		name = create_unique_id(s);
+	}
+
+	slab_cache_dir = debugfs_create_dir(name, slab_debugfs_root);
+
+	debugfs_create_file("alloc_traces", 0400,
+		slab_cache_dir, s, &slab_debugfs_fops);
+
+	debugfs_create_file("free_traces", 0400,
+		slab_cache_dir, s, &slab_debugfs_fops);
+
+	if (!unmergeable)

mm/slub.c static int __init slab_sysfs_init(void) · line 5986

Quoted 10 lines
mm/slub.c +10 -0
+		/* Setup first alias */
+		debugfs_slab_alias(s, s->name);
+}
+
+void debugfs_slab_release(struct kmem_cache *s)
+{
+	struct dentry *slab_cache_dir;
+
+	slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+	debugfs_remove_recursive(slab_cache_dir);

This whole function can be 1 line long, right?

Quoted 7 lines
mm/slub.c +7 -0
+}
+
+static int __init slab_debugfs_init(void)
+{
+	struct kmem_cache *s;
+
+	slab_debugfs_root = debugfs_create_dir("slab", NULL);

mm/slub.c static int __init slab_sysfs_init(void) · line 6003

Quoted 10 lines
mm/slub.c +10 -0
+
+	slab_state = FULL;
+
+	list_for_each_entry(s, &slab_caches, list)
+		debugfs_slab_add(s);
+
+	while (alias_list) {
+		struct saved_alias *al = alias_list;
+
+		alias_list = alias_list->next;

Why isn't this a "normal" linked list?

thanks,

greg k-h

#2 lore raw
Quoted 39 lines · Greg KH
Quoted 38 lines · Faiyaz Mohammed

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

---
 include/linux/slub_def.h |   8 ++
 mm/slab_common.c         |   9 ++
 mm/slub.c                | 359 ++++++++++++++++++++++++++++++++++++-----------
 3 files changed, 294 insertions(+), 82 deletions(-)
include/linux/slub_def.h +8 -0
index dcde82a..b413ebe 100644
@@ -159,6 +159,14 @@ static inline void sysfs_slab_release(struct kmem_cache *s)
 }
 #endif
 
+#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
+#define SLAB_SUPPORTS_DEBUGFS
+void debugfs_slab_release(struct kmem_cache *);
+#else
+static inline void debugfs_slab_release(struct kmem_cache *s)
+{
+}
+#endif
 void object_err(struct kmem_cache *s, struct page *page,
 		u8 *object, char *reason);
 
Quoted 17 lines

mm/slab_common.c static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) · line 460

Quoted 11 lines
mm/slab_common.c +3 -0
index a4a5714..873dd79 100644
@@ -455,6 +455,9 @@ static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
 #else
 		slab_kmem_cache_release(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif

If you write your .h file correctly, there's no need for #ifdef in .c files. Please fix this up, it shoudl be trivial given the above changes in the .h file.

Yes, there's no need for #ifdef in .c file but debugfs_slab_release decleration is there in slub_def.h like sysfs_slab_release and slub_def.h will be include in slab_common.c if CONFIG_SLUB is enable or it will include slab_def.h if CONFIG_SLAB is enable, this inclusion happend from mm/slab.h,

#ifdef CONFIG_SLAB
#include <linux/slab_def.h>
#endif

#ifdef CONFIG_SLUB
#include <linux/slub_def.h>
#endif

so mm/slab.h decide which .h has to include, to avoid undefined reference error added SLAB_SUPPORTS_DEBUGFS like SLAB_SUPPORTS_SYSFS.

Quoted 39 lines
Quoted 39 lines
mm/slab_common.c +6 -0
 	}
 }
 
@@ -472,6 +475,9 @@ static int shutdown_cache(struct kmem_cache *s)
 #ifdef SLAB_SUPPORTS_SYSFS
 		sysfs_slab_unlink(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
 		schedule_work(&slab_caches_to_rcu_destroy_work);
 	} else {
@@ -482,6 +488,9 @@ static int shutdown_cache(struct kmem_cache *s)
 #else
 		slab_kmem_cache_release(s);
 #endif
+#ifdef SLAB_SUPPORTS_DEBUGFS
+		debugfs_slab_release(s);
+#endif
 	}
 
 	return 0;
mm/slub.c +1 -0
index 438fa8d..f470a29 100644
@@ -36,6 +36,7 @@
 #include <linux/memcontrol.h>
 #include <linux/random.h>
 
+#include <linux/debugfs.h>
 #include <trace/events/kmem.h>
 
 #include "internal.h"
@@ -225,6 +226,15 @@ static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
 							{ return 0; }
 #endif
 
Quoted 7 lines

mm/slub.c static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) · line 231

Quoted 3 lines
mm/slub.c +3 -0
+#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
+static void debugfs_slab_add(struct kmem_cache *);
+static int debugfs_slab_alias(struct kmem_cache *, const char *);

Why does this have to return an int? What are you going to do if it fails?

Hmmm, will update the in next patch.

Quoted 150 lines
Quoted 10 lines
mm/slub.c +6 -0
+#else
+static inline void debugfs_slab_add(struct kmem_cache *s) { }
+static inline int debugfs_slab_alias(struct kmem_cache *s, const char *p)
+							{ return 0; }
+#endif
+
 static inline void stat(const struct kmem_cache *s, enum stat_item si)
 {
 #ifdef CONFIG_SLUB_STATS

mm/slub.c __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, · line 4538

Quoted 5 lines
mm/slub.c +2 -0
@@ -4524,6 +4534,8 @@ __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
 			s->refcount--;
 			s = NULL;
 		}
+
+		debugfs_slab_alias(s, name);

See, like you do so here.

Quoted 117 lines
mm/slub.c +8 -76
 	}
 
 	return s;
@@ -4545,6 +4557,8 @@ int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
 	if (err)
 		__kmem_cache_release(s);
 
+	debugfs_slab_add(s);
+
 	return err;
 }
 
@@ -4685,6 +4699,8 @@ static long validate_slab_cache(struct kmem_cache *s)
 
 	return count;
 }
+
+#ifdef CONFIG_DEBUG_FS
 /*
  * Generate lists of code addresses where slabcache objects are allocated
  * and freed.
@@ -4708,6 +4724,9 @@ struct loc_track {
 	struct location *loc;
 };
 
+static struct dentry *slab_debugfs_root;
+struct loc_track t = { 0, 0, NULL };
+
 static void free_loc_track(struct loc_track *t)
 {
 	if (t->max)
@@ -4824,82 +4843,7 @@ static void process_slab(struct loc_track *t, struct kmem_cache *s,
 			add_location(t, s, get_track(s, p, alloc));
 	put_map(map);
 }
-
-static int list_locations(struct kmem_cache *s, char *buf,
-			  enum track_item alloc)
-{
-	int len = 0;
-	unsigned long i;
-	struct loc_track t = { 0, 0, NULL };
-	int node;
-	struct kmem_cache_node *n;
-
-	if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
-			     GFP_KERNEL)) {
-		return sysfs_emit(buf, "Out of memory\n");
-	}
-	/* Push back cpu slabs */
-	flush_all(s);
-
-	for_each_kmem_cache_node(s, node, n) {
-		unsigned long flags;
-		struct page *page;
-
-		if (!atomic_long_read(&n->nr_slabs))
-			continue;
-
-		spin_lock_irqsave(&n->list_lock, flags);
-		list_for_each_entry(page, &n->partial, slab_list)
-			process_slab(&t, s, page, alloc);
-		list_for_each_entry(page, &n->full, slab_list)
-			process_slab(&t, s, page, alloc);
-		spin_unlock_irqrestore(&n->list_lock, flags);
-	}
-
-	for (i = 0; i < t.count; i++) {
-		struct location *l = &t.loc[i];
-
-		len += sysfs_emit_at(buf, len, "%7ld ", l->count);
-
-		if (l->addr)
-			len += sysfs_emit_at(buf, len, "%pS", (void *)l->addr);
-		else
-			len += sysfs_emit_at(buf, len, "<not-available>");
-
-		if (l->sum_time != l->min_time)
-			len += sysfs_emit_at(buf, len, " age=%ld/%ld/%ld",
-					     l->min_time,
-					     (long)div_u64(l->sum_time,
-							   l->count),
-					     l->max_time);
-		else
-			len += sysfs_emit_at(buf, len, " age=%ld", l->min_time);
-
-		if (l->min_pid != l->max_pid)
-			len += sysfs_emit_at(buf, len, " pid=%ld-%ld",
-					     l->min_pid, l->max_pid);
-		else
-			len += sysfs_emit_at(buf, len, " pid=%ld",
-					     l->min_pid);
-
-		if (num_online_cpus() > 1 &&
-		    !cpumask_empty(to_cpumask(l->cpus)))
-			len += sysfs_emit_at(buf, len, " cpus=%*pbl",
-					     cpumask_pr_args(to_cpumask(l->cpus)));
-
-		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
-			len += sysfs_emit_at(buf, len, " nodes=%*pbl",
-					     nodemask_pr_args(&l->nodes));
-
-		len += sysfs_emit_at(buf, len, "\n");
-	}
-
-	free_loc_track(&t);
-	if (!t.count)
-		len += sysfs_emit_at(buf, len, "No data\n");
-
-	return len;
-}
+#endif  /* CONFIG_DEBUG_FS   */
 #endif	/* CONFIG_SLUB_DEBUG */
 
 #ifdef SLUB_RESILIENCY_TEST
@@ -5351,17 +5295,25 @@ SLAB_ATTR(validate);
 

mm/slub.c SLAB_ATTR(validate); · line 5302

Quoted 10 lines
mm/slub.c +5 -3
 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
 {
-	if (!(s->flags & SLAB_STORE_USER))
-		return -ENOSYS;
-	return list_locations(s, buf, TRACK_ALLOC);
+	int len = 0;
+
+	len += sprintf(buf + len, "Deprecated, use the equvalent under\n");
+	len += sprintf(buf + len, "<debugfs_mount_point>/slab/%s/alloc_traces\n",
+			s->name);

No, just drop the sysfs file. Userspace ALWAYS has to be able to handle a sysfs file present or not present, that's the way sysfs operates and why we have "one value per file".

Quoted 4 lines

If userspace can not handle it, well, someone messed up and printing a huge message like this is not going to help either, just return an error.

Vlastimil's suggested to keep it, and it suggest's user for the new implementation path, or will remove it in next version.

Quoted 39 lines
Quoted 5 lines
mm/slub.c +2 -0
+
+	return len;
 }
 SLAB_ATTR_RO(alloc_calls);
 

mm/slub.c SLAB_ATTR(validate); · line 5314

Quoted 10 lines
mm/slub.c +5 -3
 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
 {
-	if (!(s->flags & SLAB_STORE_USER))
-		return -ENOSYS;
-	return list_locations(s, buf, TRACK_FREE);
+	int len = 0;
+
+	len += sprintf(buf + len, "Deprecated, use the equvalent under\n");
+	len += sprintf(buf + len, "<debugfs_mount_point>/slab/%s/alloc_traces\n",
+			s->name);

Same here, drop it please.

Quoted 21 lines
mm/slub.c +14 -0
+
+	return len;
 }
 SLAB_ATTR_RO(free_calls);
 #endif /* CONFIG_SLUB_DEBUG */
@@ -5817,6 +5769,249 @@ static int __init slab_sysfs_init(void)
 __initcall(slab_sysfs_init);
 #endif /* CONFIG_SYSFS */
 
+#if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
+static int debugfs_slab_alias(struct kmem_cache *s, const char *name)
+{
+	struct saved_alias *al;
+	struct dentry *slab_cache_dir;
+
+	if (slab_state == FULL) {
+		/*
+		 * If we have a leftover link then remove it.
+		 */
+		slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+		debugfs_remove(slab_cache_dir);
Quoted 9 lines

mm/slub.c static int __init slab_sysfs_init(void) · line 5789

Quoted 6 lines
mm/slub.c +6 -0
+		debugfs_create_symlink(name, slab_debugfs_root, NULL);
+	}
+
+	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
+	if (!al)
+		return -ENOMEM;

Why do you care if you have no memory or not?

yes, we can ignore, will update in new patch.

Quoted 191 lines
Quoted 191 lines
mm/slub.c +191 -0
+
+	al->s = s;
+	al->name = name;
+	al->next = alias_list;
+	alias_list = al;
+	return 0;
+}
+
+static int slab_debugfs_show(struct seq_file *seq, void *v)
+{
+	struct location *l;
+	unsigned int idx = *(unsigned int *)v;
+
+	if (idx < t.count) {
+		l = &t.loc[idx];
+
+		seq_printf(seq, "%7ld ", l->count);
+
+		if (l->addr)
+			seq_printf(seq, "%pS", (void *)l->addr);
+		else
+			seq_puts(seq, "<not-available>");
+
+		if (l->sum_time != l->min_time) {
+			seq_printf(seq, " age=%ld/%ld/%ld",
+				l->min_time,
+				(long)div_u64(l->sum_time, l->count),
+				l->max_time);
+		} else
+			seq_printf(seq, " age=%ld",
+				l->min_time);
+
+		if (l->min_pid != l->max_pid)
+			seq_printf(seq, " pid=%ld-%ld",
+				l->min_pid, l->max_pid);
+		else
+			seq_printf(seq, " pid=%ld",
+				l->min_pid);
+
+		if (num_online_cpus() > 1 &&
+				!cpumask_empty(to_cpumask(l->cpus)))
+			seq_printf(seq, " cpus=%*pbl",
+				 cpumask_pr_args(to_cpumask(l->cpus)));
+
+		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
+			seq_printf(seq, " nodes=%*pbl",
+				 nodemask_pr_args(&l->nodes));
+
+		seq_puts(seq, "\n");
+	}
+
+	if (t.count == 0)
+		seq_puts(seq, "No data\n");
+
+	return 0;
+}
+
+static void slab_debugfs_stop(struct seq_file *seq, void *v)
+{
+	if (!v && t.max) {
+		free_loc_track(&t);
+		t.max = 0;
+	}
+}
+
+static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
+{
+	loff_t *spos = v;
+
+	if (*ppos < t.count) {
+		*spos = *spos + 1;
+		*ppos = *spos;
+		return spos;
+	}
+
+	return NULL;
+}
+
+static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
+{
+	struct kmem_cache_node *n;
+	struct kmem_cache *s;
+	enum track_item alloc;
+	int node;
+	loff_t *spos = kmalloc(sizeof(loff_t), GFP_KERNEL);
+
+	s = seq->file->f_inode->i_private;
+
+	if (!spos)
+		return NULL;
+
+	if (!(s->flags & SLAB_STORE_USER)) {
+		kfree(spos);
+		return ERR_PTR(-EOPNOTSUPP);
+	}
+
+	if (*ppos == 0) {
+
+		t.count = 0;
+		t.max = 0;
+		t.loc = NULL;
+		if (strcmp(seq->file->f_path.dentry->d_name.name, "alloc_traces") == 0)
+			alloc =  TRACK_ALLOC;
+		else
+			alloc =  TRACK_FREE;
+
+		if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
+			     GFP_KERNEL)) {
+			seq_puts(seq, "Out of memory\n");
+			kfree(spos);
+			return ERR_PTR(-ENOMEM);
+		}
+		/* Push back cpu slabs */
+		flush_all(s);
+
+		for_each_kmem_cache_node(s, node, n) {
+			unsigned long flags;
+			struct page *page;
+
+			if (!atomic_long_read(&n->nr_slabs))
+				continue;
+
+			spin_lock_irqsave(&n->list_lock, flags);
+			list_for_each_entry(page, &n->partial, slab_list)
+				process_slab(&t, s, page, alloc);
+			list_for_each_entry(page, &n->full, slab_list)
+				process_slab(&t, s, page, alloc);
+			spin_unlock_irqrestore(&n->list_lock, flags);
+		}
+	}
+
+	if (*ppos < t.count) {
+		*spos = *ppos;
+		return spos;
+	}
+
+	kfree(spos);
+	return NULL;
+}
+
+static const struct seq_operations slab_debugfs_sops = {
+	.start  = slab_debugfs_start,
+	.next   = slab_debugfs_next,
+	.stop   = slab_debugfs_stop,
+	.show   = slab_debugfs_show
+};
+DEFINE_SEQ_ATTRIBUTE(slab_debugfs);
+
+static void debugfs_slab_add(struct kmem_cache *s)
+{
+	const char *name;
+	struct dentry *slab_cache_dir;
+	int unmergeable = slab_unmergeable(s);
+
+	if (unlikely(!slab_debugfs_root))
+		return;
+
+	if (!unmergeable && disable_higher_order_debug &&
+			(slub_debug & DEBUG_METADATA_FLAGS))
+		unmergeable = 1;
+
+	if (unmergeable) {
+		/*
+		 * Slabcache can never be merged so we can use the name proper.
+		 * This is typically the case for debug situations. In that
+		 * case we can catch duplicate names easily.
+		 */
+		slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+		debugfs_remove_recursive(slab_cache_dir);
+		name = s->name;
+	} else {
+		/*
+		 * Create a unique name for the slab as a target
+		 * for the symlinks.
+		 */
+		name = create_unique_id(s);
+	}
+
+	slab_cache_dir = debugfs_create_dir(name, slab_debugfs_root);
+
+	debugfs_create_file("alloc_traces", 0400,
+		slab_cache_dir, s, &slab_debugfs_fops);
+
+	debugfs_create_file("free_traces", 0400,
+		slab_cache_dir, s, &slab_debugfs_fops);
+
+	if (!unmergeable)
+		/* Setup first alias */
+		debugfs_slab_alias(s, s->name);
+}
+
Quoted 9 lines

mm/slub.c static int __init slab_sysfs_init(void) · line 5986

Quoted 6 lines
mm/slub.c +6 -0
+void debugfs_slab_release(struct kmem_cache *s)
+{
+	struct dentry *slab_cache_dir;
+
+	slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+	debugfs_remove_recursive(slab_cache_dir);

This whole function can be 1 line long, right?

yes, will update in new patch.

Quoted 11 lines
Quoted 11 lines
mm/slub.c +11 -0
+}
+
+static int __init slab_debugfs_init(void)
+{
+	struct kmem_cache *s;
+
+	slab_debugfs_root = debugfs_create_dir("slab", NULL);
+
+	slab_state = FULL;
+
+	list_for_each_entry(s, &slab_caches, list)
Quoted 9 lines

mm/slub.c static int __init slab_sysfs_init(void) · line 6003

Quoted 6 lines
mm/slub.c +6 -0
+		debugfs_slab_add(s);
+
+	while (alias_list) {
+		struct saved_alias *al = alias_list;
+
+		alias_list = alias_list->next;

Why isn't this a "normal" linked list?

Using the alias_list as it was use in slab_sysfs_init.

Quoted 4 lines

thanks,

greg k-h

#3 lore raw
In reply to Faiyaz Mohammed
Quoted 10 lines · Faiyaz Mohammed

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

Quoted 8 lines
---
 include/linux/slub_def.h |   8 ++
 mm/slab_common.c         |   9 ++
 mm/slub.c                | 359 ++++++++++++++++++++++++++++++++++++-----------
 3 files changed, 294 insertions(+), 82 deletions(-)

If this is v6, what happened in v1-5? That should always be below the --- line, right?

#4 lore raw

Hi,

Quoted 20 lines · Greg KH
Quoted 10 lines · Faiyaz Mohammed

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

Quoted 8 lines
---
 include/linux/slub_def.h |   8 ++
 mm/slab_common.c         |   9 ++
 mm/slub.c                | 359 ++++++++++++++++++++++++++++++++++++-----------
 3 files changed, 294 insertions(+), 82 deletions(-)
Quoted 3 lines

If this is v6, what happened in v1-5? That should always be below the --- line, right?

Will update in next patch version.

Thanks and regards,
Mohammed Faiyaz

#5 lore raw
In reply to Faiyaz Mohammed
Quoted 22 lines · Faiyaz Mohammed

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

mm/slub.c +3 -0
@@ -5817,6 +5769,249 @@ static int __init slab_sysfs_init(void)
 __initcall(slab_sysfs_init);
 #endif /* CONFIG_SYSFS */
 
+#if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
+static int debugfs_slab_alias(struct kmem_cache *s, const char *name)
+{

mm/slub.c static int __init slab_sysfs_init(void) · line 5784

Quoted 10 lines
mm/slub.c +10 -0
+	struct saved_alias *al;
+	struct dentry *slab_cache_dir;
+
+	if (slab_state == FULL) {
+		/*
+		 * If we have a leftover link then remove it.
+		 */
+		slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+		debugfs_remove(slab_cache_dir);
+		debugfs_create_symlink(name, slab_debugfs_root, NULL);

v6 got stuck on boot for me, unlike v5, seems like here you should return 0? That helped

Quoted 13 lines
mm/slub.c +13 -0
+	}
+
+	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
+	if (!al)
+		return -ENOMEM;
+
+	al->s = s;
+	al->name = name;
+	al->next = alias_list;
+	alias_list = al;
+	return 0;
+}
+
#6 lore raw
Quoted 20 lines · Vlastimil Babka
Quoted 19 lines · Faiyaz Mohammed

alloc_calls and free_calls implementation in sysfs have two issues, one is PAGE_SIZE limitiation of sysfs and other is it does not adhere to "one value per file" rule.

To overcome this issues, move the alloc_calls and free_calls implemeation to debugfs.

Rename the alloc_calls/free_calls to alloc_traces/free_traces, to be inline with what it does.

mm/slub.c +0 -0
@@ -5817,6 +5769,249 @@ static int __init slab_sysfs_init(void)
 __initcall(slab_sysfs_init);
 #endif /* CONFIG_SYSFS */
 
Quoted 17 lines

mm/slub.c static int __init slab_sysfs_init(void) · line 5784

Quoted 13 lines
mm/slub.c +13 -0
+#if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
+static int debugfs_slab_alias(struct kmem_cache *s, const char *name)
+{
+	struct saved_alias *al;
+	struct dentry *slab_cache_dir;
+
+	if (slab_state == FULL) {
+		/*
+		 * If we have a leftover link then remove it.
+		 */
+		slab_cache_dir = debugfs_lookup(s->name, slab_debugfs_root);
+		debugfs_remove(slab_cache_dir);
+		debugfs_create_symlink(name, slab_debugfs_root, NULL);

v6 got stuck on boot for me, unlike v5, seems like here you should return 0? That helped

oh, it worked me without return, but I think yes, have to return from here once the symlink is created like sysfs_slab_alias do. I will update in next patch version.

Quoted 5 lines
Quoted 5 lines
mm/slub.c +5 -0
+	}
+
+	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
+	if (!al)
+		return -ENOMEM;
Quoted 8 lines

mm/slub.c static int __init slab_sysfs_init(void) · line 5797

Quoted 8 lines
mm/slub.c +8 -0
+
+	al->s = s;
+	al->name = name;
+	al->next = alias_list;
+	alias_list = al;
+	return 0;
+}
+

Thanks and regards,
Mohammed Faiyaz