Skip to contentSkip to conversation

mm: cma: support sysfs

PATCH Superseded Minchan Kim 2 participants 5 messages Mar 3, 2021 – Mar 4, 2021

Subsystems (3)

v3 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 mm: cma: support sysfs

Since CMA is getting used more widely, it's more important to keep monitoring CMA statistics for system health since it's directly related to user experience.

This patch introduces sysfs statistics for CMA, in order to provide some basic monitoring of the CMA allocator.

* the number of CMA page allocation attempts
* the number of CMA page allocation failures

These two values allow the user to calcuate the allocation failure rate for each CMA area.

e.g.)
/sys/kernel/mm/cma/WIFI/cma_alloc_pages_[attempts|fails] /sys/kernel/mm/cma/SENSOR/cma_alloc_pages_[attempts|fails] /sys/kernel/mm/cma/BLUETOOTH/cma_alloc_pages_[attempts|fails]

---
From v2 - https://lore.kernel.org/linux-mm/[email protected]/
* sysfs doc and description modification - jhubbard

From v1 - https://lore.kernel.org/linux-mm/[email protected]/
* fix sysfs build and refactoring - willy
* rename and drop some attributes - jhubbard

 Documentation/ABI/testing/sysfs-kernel-mm-cma |  25 ++++
 mm/Kconfig                                    |   7 ++
 mm/Makefile                                   |   1 +
 mm/cma.c                                      |   6 +-
 mm/cma.h                                      |  18 +++
 mm/cma_sysfs.c                                | 114 ++++++++++++++++++
 6 files changed, 170 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/ABI/testing/sysfs-kernel-mm-cma
 create mode 100644 mm/cma_sysfs.c
Documentation/ABI/testing/sysfs-kernel-mm-cma new file +25 -0
new file mode 100644
index 000000000000..f518af819cee
@@ -0,0 +1,25 @@
+What:		/sys/kernel/mm/cma/
+Date:		Feb 2021
+Contact:	Minchan Kim <[email protected]>
+Description:
+		/sys/kernel/mm/cma/ contains a subdirectory for each CMA
+		heap name (also sometimes called CMA areas).
+
+		Each CMA heap subdirectory (that is, each
+		/sys/kernel/mm/cma/<cma-heap-name> directory) contains the
+		following items:
+
+			cma_alloc_pages_attempts
+			cma_alloc_pages_fails
+
+What:		/sys/kernel/mm/cma/<cma-heap-name>/cma_alloc_pages_attempts
+Date:		Feb 2021
+Contact:	Minchan Kim <[email protected]>
+Description:
+		the number of pages CMA API tried to allocate
+
+What:		/sys/kernel/mm/cma/<cma-heap-name>/cma_alloc_pages_fails
+Date:		Feb 2021
+Contact:	Minchan Kim <[email protected]>
+Description:
+		the number of pages CMA API failed to allocate
mm/Kconfig +7 -0
index 24c045b24b95..febb7e8e24de 100644
@@ -513,6 +513,13 @@ config CMA_DEBUGFS
 	help
 	  Turns on the DebugFS interface for CMA.
 
+config CMA_SYSFS
+	bool "CMA information through sysfs interface"
+	depends on CMA && SYSFS
+	help
+	  This option exposes some sysfs attributes to get information
+	  from CMA.
+
 config CMA_AREAS
 	int "Maximum count of the CMA areas"
 	depends on CMA
mm/Makefile +1 -0
index 72227b24a616..56968b23ed7a 100644
@@ -109,6 +109,7 @@ obj-$(CONFIG_CMA)	+= cma.o
 obj-$(CONFIG_MEMORY_BALLOON) += balloon_compaction.o
 obj-$(CONFIG_PAGE_EXTENSION) += page_ext.o
 obj-$(CONFIG_CMA_DEBUGFS) += cma_debug.o
+obj-$(CONFIG_CMA_SYSFS) += cma_sysfs.o
 obj-$(CONFIG_USERFAULTFD) += userfaultfd.o
 obj-$(CONFIG_IDLE_PAGE_TRACKING) += page_idle.o
 obj-$(CONFIG_DEBUG_PAGE_REF) += debug_page_ref.o
mm/cma.c +5 -1
index 54eee2119822..551b704faeaf 100644
@@ -447,9 +447,10 @@ struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align,
 	offset = cma_bitmap_aligned_offset(cma, align);
 	bitmap_maxno = cma_bitmap_maxno(cma);
 	bitmap_count = cma_bitmap_pages_to_bits(cma, count);
+	cma_sysfs_alloc_count(cma, count);
 
 	if (bitmap_count > bitmap_maxno)
-		return NULL;
+		goto out;
 
 	for (;;) {
 		mutex_lock(&cma->lock);
@@ -504,6 +505,9 @@ struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align,
 		       __func__, cma->name, count, ret);
 		cma_debug_show_areas(cma);
 	}
+out:
+	if (!page)
+		cma_sysfs_fail_count(cma, count);
 
 	pr_debug("%s(): returned %p\n", __func__, page);
 	return page;
mm/cma.h +18 -0
index 42ae082cb067..24a1d61eabc7 100644
@@ -3,6 +3,14 @@
 #define __MM_CMA_H__
 
 #include <linux/debugfs.h>
+#include <linux/kobject.h>
+
+struct cma_stat {
+	spinlock_t lock;
+	unsigned long pages_attempts;	/* the number of CMA page allocation attempts */
+	unsigned long pages_fails;	/* the number of CMA page allocation failures */
+	struct kobject kobj;
+};
 
 struct cma {
 	unsigned long   base_pfn;
@@ -16,6 +24,9 @@ struct cma {
 	struct debugfs_u32_array dfs_bitmap;
 #endif
 	char name[CMA_MAX_NAME];
+#ifdef CONFIG_CMA_SYSFS
+	struct cma_stat	*stat;
+#endif
 };
 
 extern struct cma cma_areas[MAX_CMA_AREAS];
@@ -26,4 +37,11 @@ static inline unsigned long cma_bitmap_maxno(struct cma *cma)
 	return cma->count >> cma->order_per_bit;
 }
 
+#ifdef CONFIG_CMA_SYSFS
+void cma_sysfs_alloc_count(struct cma *cma, size_t count);
+void cma_sysfs_fail_count(struct cma *cma, size_t count);
+#else
+static inline void cma_sysfs_alloc_count(struct cma *cma, size_t count) {};
+static inline void cma_sysfs_fail_count(struct cma *cma, size_t count) {};
+#endif
 #endif
mm/cma_sysfs.c new file +114 -0

114 lines not on this page · raw message

Signature · 1 line
2.30.1.766.gb4fecdf3b7-goog
maintainer
#1 lore raw
Quoted 38 lines · On Wed, 3 Mar 2021 12:50:53 -0800 Minchan Kim <[email protected]> wrote:

Since CMA is getting used more widely, it's more important to keep monitoring CMA statistics for system health since it's directly related to user experience.

This patch introduces sysfs statistics for CMA, in order to provide some basic monitoring of the CMA allocator.

* the number of CMA page allocation attempts
* the number of CMA page allocation failures

These two values allow the user to calcuate the allocation failure rate for each CMA area.

e.g.)
/sys/kernel/mm/cma/WIFI/cma_alloc_pages_[attempts|fails] /sys/kernel/mm/cma/SENSOR/cma_alloc_pages_[attempts|fails] /sys/kernel/mm/cma/BLUETOOTH/cma_alloc_pages_[attempts|fails]

...

mm/cma.h +8 -0
@@ -3,6 +3,14 @@
 #define __MM_CMA_H__
 
 #include <linux/debugfs.h>
+#include <linux/kobject.h>
+
+struct cma_stat {
+	spinlock_t lock;
+	unsigned long pages_attempts;	/* the number of CMA page allocation attempts */
+	unsigned long pages_fails;	/* the number of CMA page allocation failures */
+	struct kobject kobj;
+};
 
 struct cma {
 	unsigned long   base_pfn;

mm/cma.h struct cma { · line 30

Quoted 7 lines
mm/cma.h +3 -0
@@ -16,6 +24,9 @@ struct cma {
 	struct debugfs_u32_array dfs_bitmap;
 #endif
 	char name[CMA_MAX_NAME];
+#ifdef CONFIG_CMA_SYSFS
+	struct cma_stat	*stat;
+#endif
 };

Why aren't the stat fields simply placed directly into struct cma_stat?

Quoted 5 lines

...

mm/cma.h +3 -0
+static int __init cma_sysfs_init(void)
+{
+	int i = 0;

mm/cma.h

Quoted 10 lines
mm/cma.h +10 -0
+	struct cma *cma;
+
+	cma_kobj = kobject_create_and_add("cma", mm_kobj);
+	if (!cma_kobj) {
+		pr_err("failed to create cma kobject\n");
+		return -ENOMEM;
+	}
+
+	cma_stats = kzalloc(array_size(sizeof(struct cma_stat),
+				cma_area_count), GFP_KERNEL);

kmalloc_array(..., GFP_KERNEL|__GFP_ZERO);

?

mm/cma.h

Quoted 2 lines
mm/cma.h +2 -0
+	if (!cma_stats) {
+		pr_err("failed to create cma_stats\n");

Probably unneeded - the ENOMEM stack backtrace will point straight here.

Quoted 27 lines
mm/cma.h +27 -0
+		goto out;
+	}
+
+	do {
+		cma = &cma_areas[i];
+		cma->stat = &cma_stats[i];
+		spin_lock_init(&cma->stat->lock);
+		if (kobject_init_and_add(&cma->stat->kobj, &cma_ktype,
+					cma_kobj, "%s", cma->name)) {
+			kobject_put(&cma->stat->kobj);
+			goto out;
+		}
+	} while (++i < cma_area_count);
+
+	return 0;
+out:
+	while (--i >= 0) {
+		cma = &cma_areas[i];
+		kobject_put(&cma->stat->kobj);
+	}
+
+	kfree(cma_stats);
+	kobject_put(cma_kobj);
+
+	return -ENOMEM;
+}
+subsys_initcall(cma_sysfs_init);
#2 lore raw
Quoted 37 lines · Andrew Morton
Quoted 35 lines · On Wed, 3 Mar 2021 12:50:53 -0800 Minchan Kim <[email protected]> wrote:

Since CMA is getting used more widely, it's more important to keep monitoring CMA statistics for system health since it's directly related to user experience.

This patch introduces sysfs statistics for CMA, in order to provide some basic monitoring of the CMA allocator.

* the number of CMA page allocation attempts
* the number of CMA page allocation failures

These two values allow the user to calcuate the allocation failure rate for each CMA area.

e.g.)
/sys/kernel/mm/cma/WIFI/cma_alloc_pages_[attempts|fails] /sys/kernel/mm/cma/SENSOR/cma_alloc_pages_[attempts|fails] /sys/kernel/mm/cma/BLUETOOTH/cma_alloc_pages_[attempts|fails]

...

mm/cma.h +8 -0
@@ -3,6 +3,14 @@
 #define __MM_CMA_H__
 
 #include <linux/debugfs.h>
+#include <linux/kobject.h>
+
+struct cma_stat {
+	spinlock_t lock;
+	unsigned long pages_attempts;	/* the number of CMA page allocation attempts */
+	unsigned long pages_fails;	/* the number of CMA page allocation failures */
+	struct kobject kobj;
+};
 
Quoted 12 lines

mm/cma.h struct cma { · line 30

Quoted 10 lines
mm/cma.h +3 -0
 struct cma {
 	unsigned long   base_pfn;
@@ -16,6 +24,9 @@ struct cma {
 	struct debugfs_u32_array dfs_bitmap;
 #endif
 	char name[CMA_MAX_NAME];
+#ifdef CONFIG_CMA_SYSFS
+	struct cma_stat	*stat;
+#endif
 };

Why aren't the stat fields simply placed directly into struct cma_stat?

It have a related long discussion.
https://lore.kernel.org/linux-mm/[email protected]/
https://lore.kernel.org/linux-mm/[email protected]/

TLDR - Greg really want to see kobject stuff working as dynamic property.

Quoted 18 lines
Quoted 6 lines

...

mm/cma.h +3 -0
+static int __init cma_sysfs_init(void)
+{
+	int i = 0;

mm/cma.h

Quoted 10 lines
mm/cma.h +10 -0
+	struct cma *cma;
+
+	cma_kobj = kobject_create_and_add("cma", mm_kobj);
+	if (!cma_kobj) {
+		pr_err("failed to create cma kobject\n");
+		return -ENOMEM;
+	}
+
+	cma_stats = kzalloc(array_size(sizeof(struct cma_stat),
+				cma_area_count), GFP_KERNEL);

kmalloc_array(..., GFP_KERNEL|__GFP_ZERO);

Yub.

Quoted 7 lines

?

mm/cma.h

Quoted 2 lines
mm/cma.h +2 -0
+	if (!cma_stats) {
+		pr_err("failed to create cma_stats\n");

Probably unneeded - the ENOMEM stack backtrace will point straight here.

I failed to find the point you mentioned to print backtrace. Where code do you mean to dump the backtrace?

maintainer
#3 lore raw
Quoted 9 lines · On Wed, 3 Mar 2021 17:38:12 -0800 Minchan Kim <[email protected]> wrote:
Quoted 8 lines

mm/cma.h struct cma { · line 30

Quoted 6 lines

#endif

mm/cma.h +3 -0
 	char name[CMA_MAX_NAME];
+#ifdef CONFIG_CMA_SYSFS
+	struct cma_stat	*stat;
+#endif
 };

Why aren't the stat fields simply placed directly into struct cma_stat?

Quoted 6 lines

It have a related long discussion.
https://lore.kernel.org/linux-mm/[email protected]/
https://lore.kernel.org/linux-mm/[email protected]/

TLDR - Greg really want to see kobject stuff working as dynamic property.

Please add to changelog?

Quoted 10 lines
Quoted 7 lines

?

mm/cma_sysfs.c line 87

Quoted 2 lines
mm/cma_sysfs.c new file +2 -0
+	if (!cma_stats) {
+		pr_err("failed to create cma_stats\n");

Probably unneeded - the ENOMEM stack backtrace will point straight here.

I failed to find the point you mentioned to print backtrace. Where code do you mean to dump the backtrace?

The thing which __GFP_NOWARN disables.

#4 lore raw
Quoted 11 lines · Andrew Morton
Quoted 9 lines · On Wed, 3 Mar 2021 17:38:12 -0800 Minchan Kim <[email protected]> wrote:
Quoted 8 lines

mm/cma.h struct cma { · line 30

Quoted 6 lines

#endif

mm/cma.h +3 -0
 	char name[CMA_MAX_NAME];
+#ifdef CONFIG_CMA_SYSFS
+	struct cma_stat	*stat;
+#endif
 };

Why aren't the stat fields simply placed directly into struct cma_stat?

Quoted 8 lines
Quoted 6 lines

It have a related long discussion.
https://lore.kernel.org/linux-mm/[email protected]/
https://lore.kernel.org/linux-mm/[email protected]/

TLDR - Greg really want to see kobject stuff working as dynamic property.

Please add to changelog?

Sure.

Quoted 9 lines
Quoted 9 lines
Quoted 8 lines

?

mm/cma_sysfs.c line 87

Quoted 2 lines
mm/cma_sysfs.c new file +2 -0
+	if (!cma_stats) {
+		pr_err("failed to create cma_stats\n");

Probably unneeded - the ENOMEM stack backtrace will point straight here.

Quoted 4 lines
Quoted 2 lines

I failed to find the point you mentioned to print backtrace. Where code do you mean to dump the backtrace?

The thing which __GFP_NOWARN disables.

Got the point.