[PATCH v3 0/2] Shmem mTHP controls and stats improvements

[PATCH v3 0/2] Shmem mTHP controls and stats improvements

[Ryan Roberts]

Hi All,

This is a small series to tidy up the way the shmem controls and stats are
exposed. These patches were previously part of the series at [2], but I decided
to split them out since they can go in independently.

---
This applies against today's mm-unstable (275d686abcb59). (although I had to fix
a minor build bug in stackdepot.c due to MIN() not being defined in this tree).
All mm selftests have been run; no regressions were observed.

Changes since v2 [2]
====================
  - swpout stats are now exposed for all orders supported by anon or shmem.

[2] https://lore.kernel.org/linux-mm/20240716135907.4047689-1-ryan.roberts@arm.com/

Thanks,
Ryan

Ryan Roberts (2):
  mm: Cleanup count_mthp_stat() definition
  mm: Tidy up shmem mTHP controls and stats

 include/linux/huge_mm.h |  70 +++++++++----------
 mm/huge_memory.c        | 144 +++++++++++++++++++++++++++++++---------
 mm/memory.c             |   2 -
 mm/shmem.c              |   6 --
 4 files changed, 149 insertions(+), 73 deletions(-)

--
2.43.0

[PATCH v3 1/2] mm: Cleanup count_mthp_stat() definition

[Ryan Roberts]

Let's move count_mthp_stat() so that it's always defined, even when THP
is disabled. Previously uses of the function in files such as shmem.c,
which are compiled even when THP is disabled, required ugly THP
ifdeferry. With this cleanup, we can remove those ifdefs and the
function resolves to a nop when THP is disabled.

I shortly plan to call count_mthp_stat() from more THP-invariant source
files.

Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: Barry Song <baohua@kernel.org>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Acked-by: David Hildenbrand <david@redhat.com>
---
 include/linux/huge_mm.h | 70 ++++++++++++++++++++---------------------
 mm/memory.c             |  2 --
 mm/shmem.c              |  6 ----
 3 files changed, 35 insertions(+), 43 deletions(-)

diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
index 2b5a175196e7a..969f11f360d29 100644
--- a/include/linux/huge_mm.h
+++ b/include/linux/huge_mm.h
@@ -114,6 +114,41 @@ extern struct kobj_attribute thpsize_shmem_enabled_attr;
 #define HPAGE_PUD_MASK	(~(HPAGE_PUD_SIZE - 1))
 #define HPAGE_PUD_SIZE	((1UL) << HPAGE_PUD_SHIFT)
 
+enum mthp_stat_item {
+	MTHP_STAT_ANON_FAULT_ALLOC,
+	MTHP_STAT_ANON_FAULT_FALLBACK,
+	MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE,
+	MTHP_STAT_SWPOUT,
+	MTHP_STAT_SWPOUT_FALLBACK,
+	MTHP_STAT_SHMEM_ALLOC,
+	MTHP_STAT_SHMEM_FALLBACK,
+	MTHP_STAT_SHMEM_FALLBACK_CHARGE,
+	MTHP_STAT_SPLIT,
+	MTHP_STAT_SPLIT_FAILED,
+	MTHP_STAT_SPLIT_DEFERRED,
+	__MTHP_STAT_COUNT
+};
+
+#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
+struct mthp_stat {
+	unsigned long stats[ilog2(MAX_PTRS_PER_PTE) + 1][__MTHP_STAT_COUNT];
+};
+
+DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
+
+static inline void count_mthp_stat(int order, enum mthp_stat_item item)
+{
+	if (order <= 0 || order > PMD_ORDER)
+		return;
+
+	this_cpu_inc(mthp_stats.stats[order][item]);
+}
+#else
+static inline void count_mthp_stat(int order, enum mthp_stat_item item)
+{
+}
+#endif
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 
 extern unsigned long transparent_hugepage_flags;
@@ -269,41 +304,6 @@ struct thpsize {
 
 #define to_thpsize(kobj) container_of(kobj, struct thpsize, kobj)
 
-enum mthp_stat_item {
-	MTHP_STAT_ANON_FAULT_ALLOC,
-	MTHP_STAT_ANON_FAULT_FALLBACK,
-	MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE,
-	MTHP_STAT_SWPOUT,
-	MTHP_STAT_SWPOUT_FALLBACK,
-	MTHP_STAT_SHMEM_ALLOC,
-	MTHP_STAT_SHMEM_FALLBACK,
-	MTHP_STAT_SHMEM_FALLBACK_CHARGE,
-	MTHP_STAT_SPLIT,
-	MTHP_STAT_SPLIT_FAILED,
-	MTHP_STAT_SPLIT_DEFERRED,
-	__MTHP_STAT_COUNT
-};
-
-struct mthp_stat {
-	unsigned long stats[ilog2(MAX_PTRS_PER_PTE) + 1][__MTHP_STAT_COUNT];
-};
-
-#ifdef CONFIG_SYSFS
-DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
-
-static inline void count_mthp_stat(int order, enum mthp_stat_item item)
-{
-	if (order <= 0 || order > PMD_ORDER)
-		return;
-
-	this_cpu_inc(mthp_stats.stats[order][item]);
-}
-#else
-static inline void count_mthp_stat(int order, enum mthp_stat_item item)
-{
-}
-#endif
-
 #define transparent_hugepage_use_zero_page()				\
 	(transparent_hugepage_flags &					\
 	 (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG))
diff --git a/mm/memory.c b/mm/memory.c
index 87a5c900c77f2..c96ac5b2464fa 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4761,9 +4761,7 @@ static vm_fault_t do_anonymous_page(struct vm_fault *vmf)
 
 	folio_ref_add(folio, nr_pages - 1);
 	add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr_pages);
-#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 	count_mthp_stat(folio_order(folio), MTHP_STAT_ANON_FAULT_ALLOC);
-#endif
 	folio_add_new_anon_rmap(folio, vma, addr, RMAP_EXCLUSIVE);
 	folio_add_lru_vma(folio, vma);
 setpte:
diff --git a/mm/shmem.c b/mm/shmem.c
index 68c9a31bc763b..33a146843e4c0 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1791,9 +1791,7 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
 
 			if (pages == HPAGE_PMD_NR)
 				count_vm_event(THP_FILE_FALLBACK);
-#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 			count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK);
-#endif
 			order = next_order(&suitable_orders, order);
 		}
 	} else {
@@ -1818,10 +1816,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
 				count_vm_event(THP_FILE_FALLBACK);
 				count_vm_event(THP_FILE_FALLBACK_CHARGE);
 			}
-#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 			count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK);
 			count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE);
-#endif
 		}
 		goto unlock;
 	}
@@ -2301,9 +2297,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 		if (!IS_ERR(folio)) {
 			if (folio_test_pmd_mappable(folio))
 				count_vm_event(THP_FILE_ALLOC);
-#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 			count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC);
-#endif
 			goto alloced;
 		}
 		if (PTR_ERR(folio) == -EEXIST)
-- 
2.43.0

[PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Ryan Roberts]

Previously we had a situation where shmem mTHP controls and stats were
not exposed for some supported sizes and were exposed for some
unsupported sizes. So let's clean that up.

Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
shmem controls and stats were previously being exposed for all the anon
mTHP orders, meaning order-1 was not present, and for arm64 64K base
pages, orders 12 and 13 were exposed but were not supported internally.

Tidy this all up by defining ctrl and stats attribute groups for anon
and file separately. Anon ctrl and stats groups are populated for all
orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.

Additionally, create "any" ctrl and stats attribute groups which are
populated for all orders in (THP_ORDERS_ALL_ANON |
THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
anon and shmem.

The side-effect of all this is that different hugepage-*kB directories
contain different sets of controls and stats, depending on which memory
types support that size. This approach is preferred over the
alternative, which is to populate dummy controls and stats for memory
types that do not support a given size.

Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
---
 mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
 1 file changed, 114 insertions(+), 30 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 0c3075ee00012..082d86b7c6c2f 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
 static DEFINE_SPINLOCK(huge_anon_orders_lock);
 static LIST_HEAD(thpsize_list);
 
-static ssize_t thpsize_enabled_show(struct kobject *kobj,
-				    struct kobj_attribute *attr, char *buf)
+static ssize_t anon_enabled_show(struct kobject *kobj,
+				 struct kobj_attribute *attr, char *buf)
 {
 	int order = to_thpsize(kobj)->order;
 	const char *output;
@@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
 	return sysfs_emit(buf, "%s\n", output);
 }
 
-static ssize_t thpsize_enabled_store(struct kobject *kobj,
-				     struct kobj_attribute *attr,
-				     const char *buf, size_t count)
+static ssize_t anon_enabled_store(struct kobject *kobj,
+				  struct kobj_attribute *attr,
+				  const char *buf, size_t count)
 {
 	int order = to_thpsize(kobj)->order;
 	ssize_t ret = count;
@@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
 	return ret;
 }
 
-static struct kobj_attribute thpsize_enabled_attr =
-	__ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
+static struct kobj_attribute anon_enabled_attr =
+	__ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
 
-static struct attribute *thpsize_attrs[] = {
-	&thpsize_enabled_attr.attr,
+static struct attribute *anon_ctrl_attrs[] = {
+	&anon_enabled_attr.attr,
+	NULL,
+};
+
+static const struct attribute_group anon_ctrl_attr_grp = {
+	.attrs = anon_ctrl_attrs,
+};
+
+static struct attribute *file_ctrl_attrs[] = {
 #ifdef CONFIG_SHMEM
 	&thpsize_shmem_enabled_attr.attr,
 #endif
 	NULL,
 };
 
-static const struct attribute_group thpsize_attr_group = {
-	.attrs = thpsize_attrs,
+static const struct attribute_group file_ctrl_attr_grp = {
+	.attrs = file_ctrl_attrs,
+};
+
+static struct attribute *any_ctrl_attrs[] = {
+	NULL,
+};
+
+static const struct attribute_group any_ctrl_attr_grp = {
+	.attrs = any_ctrl_attrs,
 };
 
 static const struct kobj_type thpsize_ktype = {
@@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);
 DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
 DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
 DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
+#ifdef CONFIG_SHMEM
 DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
 DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
 DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
+#endif
 DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
 DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
 DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
 
-static struct attribute *stats_attrs[] = {
+static struct attribute *anon_stats_attrs[] = {
 	&anon_fault_alloc_attr.attr,
 	&anon_fault_fallback_attr.attr,
 	&anon_fault_fallback_charge_attr.attr,
+#ifndef CONFIG_SHMEM
 	&swpout_attr.attr,
 	&swpout_fallback_attr.attr,
-	&shmem_alloc_attr.attr,
-	&shmem_fallback_attr.attr,
-	&shmem_fallback_charge_attr.attr,
+#endif
 	&split_attr.attr,
 	&split_failed_attr.attr,
 	&split_deferred_attr.attr,
 	NULL,
 };
 
-static struct attribute_group stats_attr_group = {
+static struct attribute_group anon_stats_attr_grp = {
+	.name = "stats",
+	.attrs = anon_stats_attrs,
+};
+
+static struct attribute *file_stats_attrs[] = {
+#ifdef CONFIG_SHMEM
+	&shmem_alloc_attr.attr,
+	&shmem_fallback_attr.attr,
+	&shmem_fallback_charge_attr.attr,
+#endif
+	NULL,
+};
+
+static struct attribute_group file_stats_attr_grp = {
+	.name = "stats",
+	.attrs = file_stats_attrs,
+};
+
+static struct attribute *any_stats_attrs[] = {
+#ifdef CONFIG_SHMEM
+	&swpout_attr.attr,
+	&swpout_fallback_attr.attr,
+#endif
+	NULL,
+};
+
+static struct attribute_group any_stats_attr_grp = {
 	.name = "stats",
-	.attrs = stats_attrs,
+	.attrs = any_stats_attrs,
 };
 
+static int sysfs_add_group(struct kobject *kobj,
+			   const struct attribute_group *grp)
+{
+	int ret = -ENOENT;
+
+	/*
+	 * If the group is named, try to merge first, assuming the subdirectory
+	 * was already created. This avoids the warning emitted by
+	 * sysfs_create_group() if the directory already exists.
+	 */
+	if (grp->name)
+		ret = sysfs_merge_group(kobj, grp);
+	if (ret)
+		ret = sysfs_create_group(kobj, grp);
+
+	return ret;
+}
+
 static struct thpsize *thpsize_create(int order, struct kobject *parent)
 {
 	unsigned long size = (PAGE_SIZE << order) / SZ_1K;
 	struct thpsize *thpsize;
-	int ret;
+	int ret = -ENOMEM;
 
 	thpsize = kzalloc(sizeof(*thpsize), GFP_KERNEL);
 	if (!thpsize)
-		return ERR_PTR(-ENOMEM);
+		goto err;
+
+	thpsize->order = order;
 
 	ret = kobject_init_and_add(&thpsize->kobj, &thpsize_ktype, parent,
 				   "hugepages-%lukB", size);
 	if (ret) {
 		kfree(thpsize);
-		return ERR_PTR(ret);
+		goto err;
 	}
 
-	ret = sysfs_create_group(&thpsize->kobj, &thpsize_attr_group);
-	if (ret) {
-		kobject_put(&thpsize->kobj);
-		return ERR_PTR(ret);
+
+	ret = sysfs_add_group(&thpsize->kobj, &any_ctrl_attr_grp);
+	if (ret)
+		goto err_put;
+
+	ret = sysfs_add_group(&thpsize->kobj, &any_stats_attr_grp);
+	if (ret)
+		goto err_put;
+
+	if (BIT(order) & THP_ORDERS_ALL_ANON) {
+		ret = sysfs_add_group(&thpsize->kobj, &anon_ctrl_attr_grp);
+		if (ret)
+			goto err_put;
+
+		ret = sysfs_add_group(&thpsize->kobj, &anon_stats_attr_grp);
+		if (ret)
+			goto err_put;
 	}
 
-	ret = sysfs_create_group(&thpsize->kobj, &stats_attr_group);
-	if (ret) {
-		kobject_put(&thpsize->kobj);
-		return ERR_PTR(ret);
+	if (BIT(order) & THP_ORDERS_ALL_FILE_DEFAULT) {
+		ret = sysfs_add_group(&thpsize->kobj, &file_ctrl_attr_grp);
+		if (ret)
+			goto err_put;
+
+		ret = sysfs_add_group(&thpsize->kobj, &file_stats_attr_grp);
+		if (ret)
+			goto err_put;
 	}
 
-	thpsize->order = order;
 	return thpsize;
+err_put:
+	kobject_put(&thpsize->kobj);
+err:
+	return ERR_PTR(ret);
 }
 
 static void thpsize_release(struct kobject *kobj)
@@ -692,7 +776,7 @@ static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
 		goto remove_hp_group;
 	}
 
-	orders = THP_ORDERS_ALL_ANON;
+	orders = THP_ORDERS_ALL_ANON | THP_ORDERS_ALL_FILE_DEFAULT;
 	order = highest_order(orders);
 	while (orders) {
 		thpsize = thpsize_create(order, *hugepage_kobj);
-- 
2.43.0

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Barry Song]

On Thu, Aug 8, 2024 at 11:19 PM Ryan Roberts <ryan.roberts@arm.com> wrote:
>
> Previously we had a situation where shmem mTHP controls and stats were
> not exposed for some supported sizes and were exposed for some
> unsupported sizes. So let's clean that up.
>
> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
> shmem controls and stats were previously being exposed for all the anon
> mTHP orders, meaning order-1 was not present, and for arm64 64K base
> pages, orders 12 and 13 were exposed but were not supported internally.
>
> Tidy this all up by defining ctrl and stats attribute groups for anon
> and file separately. Anon ctrl and stats groups are populated for all
> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>
> Additionally, create "any" ctrl and stats attribute groups which are
> populated for all orders in (THP_ORDERS_ALL_ANON |
> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
> anon and shmem.

This passage is a bit confusing to me. I'm not sure whether it's about
creating any control and stats for both file and anon, or creating them
separately depending on the situation. However, the previous passage
mentions that file and anon create control and stats separately based
on different orders, and the orders they create might differ.

But after running your patches, I understood it. For instance, I'm now
seeing 8kB folder that didn't exist before:

           /sys/kernel/mm/transparent_hugepage # ls -l
           drwxr-xr-x 3 root root    0 Aug  8 22:01 hugepages-1024kB
            ...
           drwxr-xr-x 3 root root    0 Aug  8 22:01 hugepages-8kB
           drwxr-xr-x 2 root root    0 Aug  8 22:01 khugepaged
           ...

Then, when I entered the 8kB folder, I found 'shmem_enabled'
but not 'enabled' for anon:
            /sys/kernel/mm/transparent_hugepage/hugepages-8kB # ls
            shmem_enabled  stats

However, in the 16kB folder, I found both:
            /sys/kernel/mm/transparent_hugepage/hugepages-16kB # ls
            enabled  shmem_enabled stats

In the 8kB 'stats,' I see 'shmem_alloc' but not 'anon_alloc.' Since
both shmem and anon require swapout, I am seeing 'swpout' and
'swpout_fallback':

            /sys/kernel/mm/transparent_hugepage/hugepages-8kB/stats # ls
            shmem_alloc  shmem_fallback  shmem_fallback_charge  swpout
 swpout_fallback

Everything observed seems to meet expectations, so:

Tested-by: Barry Song <baohua@kernel.org>

>
> The side-effect of all this is that different hugepage-*kB directories
> contain different sets of controls and stats, depending on which memory
> types support that size. This approach is preferred over the
> alternative, which is to populate dummy controls and stats for memory
> types that do not support a given size.
>
> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
> ---
>  mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>  1 file changed, 114 insertions(+), 30 deletions(-)
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 0c3075ee00012..082d86b7c6c2f 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>  static DEFINE_SPINLOCK(huge_anon_orders_lock);
>  static LIST_HEAD(thpsize_list);
>
> -static ssize_t thpsize_enabled_show(struct kobject *kobj,
> -                                   struct kobj_attribute *attr, char *buf)
> +static ssize_t anon_enabled_show(struct kobject *kobj,
> +                                struct kobj_attribute *attr, char *buf)
>  {
>         int order = to_thpsize(kobj)->order;
>         const char *output;
> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>         return sysfs_emit(buf, "%s\n", output);
>  }
>
> -static ssize_t thpsize_enabled_store(struct kobject *kobj,
> -                                    struct kobj_attribute *attr,
> -                                    const char *buf, size_t count)
> +static ssize_t anon_enabled_store(struct kobject *kobj,
> +                                 struct kobj_attribute *attr,
> +                                 const char *buf, size_t count)
>  {
>         int order = to_thpsize(kobj)->order;
>         ssize_t ret = count;
> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>         return ret;
>  }
>
> -static struct kobj_attribute thpsize_enabled_attr =
> -       __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
> +static struct kobj_attribute anon_enabled_attr =
> +       __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>
> -static struct attribute *thpsize_attrs[] = {
> -       &thpsize_enabled_attr.attr,
> +static struct attribute *anon_ctrl_attrs[] = {
> +       &anon_enabled_attr.attr,
> +       NULL,
> +};
> +
> +static const struct attribute_group anon_ctrl_attr_grp = {
> +       .attrs = anon_ctrl_attrs,
> +};
> +
> +static struct attribute *file_ctrl_attrs[] = {
>  #ifdef CONFIG_SHMEM
>         &thpsize_shmem_enabled_attr.attr,
>  #endif
>         NULL,
>  };
>
> -static const struct attribute_group thpsize_attr_group = {
> -       .attrs = thpsize_attrs,
> +static const struct attribute_group file_ctrl_attr_grp = {
> +       .attrs = file_ctrl_attrs,
> +};
> +
> +static struct attribute *any_ctrl_attrs[] = {
> +       NULL,
> +};
> +
> +static const struct attribute_group any_ctrl_attr_grp = {
> +       .attrs = any_ctrl_attrs,
>  };
>
>  static const struct kobj_type thpsize_ktype = {
> @@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);
>  DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
>  DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
>  DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
> +#ifdef CONFIG_SHMEM
>  DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
>  DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
>  DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
> +#endif
>  DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>  DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>  DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>
> -static struct attribute *stats_attrs[] = {
> +static struct attribute *anon_stats_attrs[] = {
>         &anon_fault_alloc_attr.attr,
>         &anon_fault_fallback_attr.attr,
>         &anon_fault_fallback_charge_attr.attr,
> +#ifndef CONFIG_SHMEM
>         &swpout_attr.attr,
>         &swpout_fallback_attr.attr,
> -       &shmem_alloc_attr.attr,
> -       &shmem_fallback_attr.attr,
> -       &shmem_fallback_charge_attr.attr,
> +#endif
>         &split_attr.attr,
>         &split_failed_attr.attr,
>         &split_deferred_attr.attr,
>         NULL,
>  };
>
> -static struct attribute_group stats_attr_group = {
> +static struct attribute_group anon_stats_attr_grp = {
> +       .name = "stats",
> +       .attrs = anon_stats_attrs,
> +};
> +
> +static struct attribute *file_stats_attrs[] = {
> +#ifdef CONFIG_SHMEM
> +       &shmem_alloc_attr.attr,
> +       &shmem_fallback_attr.attr,
> +       &shmem_fallback_charge_attr.attr,
> +#endif
> +       NULL,
> +};
> +
> +static struct attribute_group file_stats_attr_grp = {
> +       .name = "stats",
> +       .attrs = file_stats_attrs,
> +};
> +
> +static struct attribute *any_stats_attrs[] = {
> +#ifdef CONFIG_SHMEM
> +       &swpout_attr.attr,
> +       &swpout_fallback_attr.attr,
> +#endif
> +       NULL,
> +};
> +
> +static struct attribute_group any_stats_attr_grp = {
>         .name = "stats",
> -       .attrs = stats_attrs,
> +       .attrs = any_stats_attrs,
>  };
>
> +static int sysfs_add_group(struct kobject *kobj,
> +                          const struct attribute_group *grp)
> +{
> +       int ret = -ENOENT;
> +
> +       /*
> +        * If the group is named, try to merge first, assuming the subdirectory
> +        * was already created. This avoids the warning emitted by
> +        * sysfs_create_group() if the directory already exists.
> +        */
> +       if (grp->name)
> +               ret = sysfs_merge_group(kobj, grp);
> +       if (ret)
> +               ret = sysfs_create_group(kobj, grp);
> +
> +       return ret;
> +}
> +
>  static struct thpsize *thpsize_create(int order, struct kobject *parent)
>  {
>         unsigned long size = (PAGE_SIZE << order) / SZ_1K;
>         struct thpsize *thpsize;
> -       int ret;
> +       int ret = -ENOMEM;
>
>         thpsize = kzalloc(sizeof(*thpsize), GFP_KERNEL);
>         if (!thpsize)
> -               return ERR_PTR(-ENOMEM);
> +               goto err;
> +
> +       thpsize->order = order;
>
>         ret = kobject_init_and_add(&thpsize->kobj, &thpsize_ktype, parent,
>                                    "hugepages-%lukB", size);
>         if (ret) {
>                 kfree(thpsize);
> -               return ERR_PTR(ret);
> +               goto err;
>         }
>
> -       ret = sysfs_create_group(&thpsize->kobj, &thpsize_attr_group);
> -       if (ret) {
> -               kobject_put(&thpsize->kobj);
> -               return ERR_PTR(ret);
> +
> +       ret = sysfs_add_group(&thpsize->kobj, &any_ctrl_attr_grp);
> +       if (ret)
> +               goto err_put;
> +
> +       ret = sysfs_add_group(&thpsize->kobj, &any_stats_attr_grp);
> +       if (ret)
> +               goto err_put;
> +
> +       if (BIT(order) & THP_ORDERS_ALL_ANON) {
> +               ret = sysfs_add_group(&thpsize->kobj, &anon_ctrl_attr_grp);
> +               if (ret)
> +                       goto err_put;
> +
> +               ret = sysfs_add_group(&thpsize->kobj, &anon_stats_attr_grp);
> +               if (ret)
> +                       goto err_put;
>         }
>
> -       ret = sysfs_create_group(&thpsize->kobj, &stats_attr_group);
> -       if (ret) {
> -               kobject_put(&thpsize->kobj);
> -               return ERR_PTR(ret);
> +       if (BIT(order) & THP_ORDERS_ALL_FILE_DEFAULT) {
> +               ret = sysfs_add_group(&thpsize->kobj, &file_ctrl_attr_grp);
> +               if (ret)
> +                       goto err_put;
> +
> +               ret = sysfs_add_group(&thpsize->kobj, &file_stats_attr_grp);
> +               if (ret)
> +                       goto err_put;
>         }
>
> -       thpsize->order = order;
>         return thpsize;
> +err_put:
> +       kobject_put(&thpsize->kobj);
> +err:
> +       return ERR_PTR(ret);
>  }
>
>  static void thpsize_release(struct kobject *kobj)
> @@ -692,7 +776,7 @@ static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
>                 goto remove_hp_group;
>         }
>
> -       orders = THP_ORDERS_ALL_ANON;
> +       orders = THP_ORDERS_ALL_ANON | THP_ORDERS_ALL_FILE_DEFAULT;
>         order = highest_order(orders);
>         while (orders) {
>                 thpsize = thpsize_create(order, *hugepage_kobj);
> --
> 2.43.0
>

Thanks
Barry

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Ryan Roberts]

On 08/08/2024 23:11, Barry Song wrote:
> On Thu, Aug 8, 2024 at 11:19 PM Ryan Roberts <ryan.roberts@arm.com> wrote:
>>
>> Previously we had a situation where shmem mTHP controls and stats were
>> not exposed for some supported sizes and were exposed for some
>> unsupported sizes. So let's clean that up.
>>
>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
>> shmem controls and stats were previously being exposed for all the anon
>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
>> pages, orders 12 and 13 were exposed but were not supported internally.
>>
>> Tidy this all up by defining ctrl and stats attribute groups for anon
>> and file separately. Anon ctrl and stats groups are populated for all
>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>>
>> Additionally, create "any" ctrl and stats attribute groups which are
>> populated for all orders in (THP_ORDERS_ALL_ANON |
>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
>> anon and shmem.
> 
> This passage is a bit confusing to me. I'm not sure whether it's about
> creating any control and stats for both file and anon, or creating them
> separately depending on the situation. However, the previous passage
> mentions that file and anon create control and stats separately based
> on different orders, and the orders they create might differ.

Yes, I'll admit that I added this paragraph for the change I made since the last
version, and it's not the clearest thing I've ever written on reflection.

The idea is that there are 3 buckets; "anon", "file" and "any". controls/stats
in the "anon" bucket are populated for all orders supported by anon memory. Same
for file. controls/stats in the "any" bucket are populated for all orders
supported by either anon or file.

> 
> But after running your patches, I understood it. For instance, I'm now
> seeing 8kB folder that didn't exist before:
> 
>            /sys/kernel/mm/transparent_hugepage # ls -l
>            drwxr-xr-x 3 root root    0 Aug  8 22:01 hugepages-1024kB
>             ...
>            drwxr-xr-x 3 root root    0 Aug  8 22:01 hugepages-8kB
>            drwxr-xr-x 2 root root    0 Aug  8 22:01 khugepaged
>            ...
> 
> Then, when I entered the 8kB folder, I found 'shmem_enabled'
> but not 'enabled' for anon:
>             /sys/kernel/mm/transparent_hugepage/hugepages-8kB # ls
>             shmem_enabled  stats
> 
> However, in the 16kB folder, I found both:
>             /sys/kernel/mm/transparent_hugepage/hugepages-16kB # ls
>             enabled  shmem_enabled stats
> 
> In the 8kB 'stats,' I see 'shmem_alloc' but not 'anon_alloc.' Since
> both shmem and anon require swapout, I am seeing 'swpout' and
> 'swpout_fallback':
> 
>             /sys/kernel/mm/transparent_hugepage/hugepages-8kB/stats # ls
>             shmem_alloc  shmem_fallback  shmem_fallback_charge  swpout
>  swpout_fallback
> 
> Everything observed seems to meet expectations, so:

Yes, what you observe is exactly as intended. It gets a bit tricky if
CONFIG_SHMEM is disabled; In this case, some ifdeffery ensures that the swpout
stats are declared in the "anon" bucket instead of the "any" bucket.

> 
> Tested-by: Barry Song <baohua@kernel.org>

Thanks!

> 
>>
>> The side-effect of all this is that different hugepage-*kB directories
>> contain different sets of controls and stats, depending on which memory
>> types support that size. This approach is preferred over the
>> alternative, which is to populate dummy controls and stats for memory
>> types that do not support a given size.
>>
>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>> ---
>>  mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>>  1 file changed, 114 insertions(+), 30 deletions(-)
>>
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 0c3075ee00012..082d86b7c6c2f 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>>  static DEFINE_SPINLOCK(huge_anon_orders_lock);
>>  static LIST_HEAD(thpsize_list);
>>
>> -static ssize_t thpsize_enabled_show(struct kobject *kobj,
>> -                                   struct kobj_attribute *attr, char *buf)
>> +static ssize_t anon_enabled_show(struct kobject *kobj,
>> +                                struct kobj_attribute *attr, char *buf)
>>  {
>>         int order = to_thpsize(kobj)->order;
>>         const char *output;
>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>         return sysfs_emit(buf, "%s\n", output);
>>  }
>>
>> -static ssize_t thpsize_enabled_store(struct kobject *kobj,
>> -                                    struct kobj_attribute *attr,
>> -                                    const char *buf, size_t count)
>> +static ssize_t anon_enabled_store(struct kobject *kobj,
>> +                                 struct kobj_attribute *attr,
>> +                                 const char *buf, size_t count)
>>  {
>>         int order = to_thpsize(kobj)->order;
>>         ssize_t ret = count;
>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>         return ret;
>>  }
>>
>> -static struct kobj_attribute thpsize_enabled_attr =
>> -       __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
>> +static struct kobj_attribute anon_enabled_attr =
>> +       __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>>
>> -static struct attribute *thpsize_attrs[] = {
>> -       &thpsize_enabled_attr.attr,
>> +static struct attribute *anon_ctrl_attrs[] = {
>> +       &anon_enabled_attr.attr,
>> +       NULL,
>> +};
>> +
>> +static const struct attribute_group anon_ctrl_attr_grp = {
>> +       .attrs = anon_ctrl_attrs,
>> +};
>> +
>> +static struct attribute *file_ctrl_attrs[] = {
>>  #ifdef CONFIG_SHMEM
>>         &thpsize_shmem_enabled_attr.attr,
>>  #endif
>>         NULL,
>>  };
>>
>> -static const struct attribute_group thpsize_attr_group = {
>> -       .attrs = thpsize_attrs,
>> +static const struct attribute_group file_ctrl_attr_grp = {
>> +       .attrs = file_ctrl_attrs,
>> +};
>> +
>> +static struct attribute *any_ctrl_attrs[] = {
>> +       NULL,
>> +};
>> +
>> +static const struct attribute_group any_ctrl_attr_grp = {
>> +       .attrs = any_ctrl_attrs,
>>  };
>>
>>  static const struct kobj_type thpsize_ktype = {
>> @@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);
>>  DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
>>  DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
>>  DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
>> +#ifdef CONFIG_SHMEM
>>  DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
>>  DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
>>  DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>> +#endif
>>  DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>  DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>  DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>>
>> -static struct attribute *stats_attrs[] = {
>> +static struct attribute *anon_stats_attrs[] = {
>>         &anon_fault_alloc_attr.attr,
>>         &anon_fault_fallback_attr.attr,
>>         &anon_fault_fallback_charge_attr.attr,
>> +#ifndef CONFIG_SHMEM
>>         &swpout_attr.attr,
>>         &swpout_fallback_attr.attr,
>> -       &shmem_alloc_attr.attr,
>> -       &shmem_fallback_attr.attr,
>> -       &shmem_fallback_charge_attr.attr,
>> +#endif
>>         &split_attr.attr,
>>         &split_failed_attr.attr,
>>         &split_deferred_attr.attr,
>>         NULL,
>>  };
>>
>> -static struct attribute_group stats_attr_group = {
>> +static struct attribute_group anon_stats_attr_grp = {
>> +       .name = "stats",
>> +       .attrs = anon_stats_attrs,
>> +};
>> +
>> +static struct attribute *file_stats_attrs[] = {
>> +#ifdef CONFIG_SHMEM
>> +       &shmem_alloc_attr.attr,
>> +       &shmem_fallback_attr.attr,
>> +       &shmem_fallback_charge_attr.attr,
>> +#endif
>> +       NULL,
>> +};
>> +
>> +static struct attribute_group file_stats_attr_grp = {
>> +       .name = "stats",
>> +       .attrs = file_stats_attrs,
>> +};
>> +
>> +static struct attribute *any_stats_attrs[] = {
>> +#ifdef CONFIG_SHMEM
>> +       &swpout_attr.attr,
>> +       &swpout_fallback_attr.attr,
>> +#endif
>> +       NULL,
>> +};
>> +
>> +static struct attribute_group any_stats_attr_grp = {
>>         .name = "stats",
>> -       .attrs = stats_attrs,
>> +       .attrs = any_stats_attrs,
>>  };
>>
>> +static int sysfs_add_group(struct kobject *kobj,
>> +                          const struct attribute_group *grp)
>> +{
>> +       int ret = -ENOENT;
>> +
>> +       /*
>> +        * If the group is named, try to merge first, assuming the subdirectory
>> +        * was already created. This avoids the warning emitted by
>> +        * sysfs_create_group() if the directory already exists.
>> +        */
>> +       if (grp->name)
>> +               ret = sysfs_merge_group(kobj, grp);
>> +       if (ret)
>> +               ret = sysfs_create_group(kobj, grp);
>> +
>> +       return ret;
>> +}
>> +
>>  static struct thpsize *thpsize_create(int order, struct kobject *parent)
>>  {
>>         unsigned long size = (PAGE_SIZE << order) / SZ_1K;
>>         struct thpsize *thpsize;
>> -       int ret;
>> +       int ret = -ENOMEM;
>>
>>         thpsize = kzalloc(sizeof(*thpsize), GFP_KERNEL);
>>         if (!thpsize)
>> -               return ERR_PTR(-ENOMEM);
>> +               goto err;
>> +
>> +       thpsize->order = order;
>>
>>         ret = kobject_init_and_add(&thpsize->kobj, &thpsize_ktype, parent,
>>                                    "hugepages-%lukB", size);
>>         if (ret) {
>>                 kfree(thpsize);
>> -               return ERR_PTR(ret);
>> +               goto err;
>>         }
>>
>> -       ret = sysfs_create_group(&thpsize->kobj, &thpsize_attr_group);
>> -       if (ret) {
>> -               kobject_put(&thpsize->kobj);
>> -               return ERR_PTR(ret);
>> +
>> +       ret = sysfs_add_group(&thpsize->kobj, &any_ctrl_attr_grp);
>> +       if (ret)
>> +               goto err_put;
>> +
>> +       ret = sysfs_add_group(&thpsize->kobj, &any_stats_attr_grp);
>> +       if (ret)
>> +               goto err_put;
>> +
>> +       if (BIT(order) & THP_ORDERS_ALL_ANON) {
>> +               ret = sysfs_add_group(&thpsize->kobj, &anon_ctrl_attr_grp);
>> +               if (ret)
>> +                       goto err_put;
>> +
>> +               ret = sysfs_add_group(&thpsize->kobj, &anon_stats_attr_grp);
>> +               if (ret)
>> +                       goto err_put;
>>         }
>>
>> -       ret = sysfs_create_group(&thpsize->kobj, &stats_attr_group);
>> -       if (ret) {
>> -               kobject_put(&thpsize->kobj);
>> -               return ERR_PTR(ret);
>> +       if (BIT(order) & THP_ORDERS_ALL_FILE_DEFAULT) {
>> +               ret = sysfs_add_group(&thpsize->kobj, &file_ctrl_attr_grp);
>> +               if (ret)
>> +                       goto err_put;
>> +
>> +               ret = sysfs_add_group(&thpsize->kobj, &file_stats_attr_grp);
>> +               if (ret)
>> +                       goto err_put;
>>         }
>>
>> -       thpsize->order = order;
>>         return thpsize;
>> +err_put:
>> +       kobject_put(&thpsize->kobj);
>> +err:
>> +       return ERR_PTR(ret);
>>  }
>>
>>  static void thpsize_release(struct kobject *kobj)
>> @@ -692,7 +776,7 @@ static int __init hugepage_init_sysfs(struct kobject **hugepage_kobj)
>>                 goto remove_hp_group;
>>         }
>>
>> -       orders = THP_ORDERS_ALL_ANON;
>> +       orders = THP_ORDERS_ALL_ANON | THP_ORDERS_ALL_FILE_DEFAULT;
>>         order = highest_order(orders);
>>         while (orders) {
>>                 thpsize = thpsize_create(order, *hugepage_kobj);
>> --
>> 2.43.0
>>
> 
> Thanks
> Barry

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Baolin Wang]

On 2024/8/8 19:18, Ryan Roberts wrote:
> Previously we had a situation where shmem mTHP controls and stats were
> not exposed for some supported sizes and were exposed for some
> unsupported sizes. So let's clean that up.
> 
> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
> shmem controls and stats were previously being exposed for all the anon
> mTHP orders, meaning order-1 was not present, and for arm64 64K base
> pages, orders 12 and 13 were exposed but were not supported internally.
> 
> Tidy this all up by defining ctrl and stats attribute groups for anon
> and file separately. Anon ctrl and stats groups are populated for all
> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
> 
> Additionally, create "any" ctrl and stats attribute groups which are
> populated for all orders in (THP_ORDERS_ALL_ANON |
> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
> anon and shmem.
> 
> The side-effect of all this is that different hugepage-*kB directories
> contain different sets of controls and stats, depending on which memory
> types support that size. This approach is preferred over the
> alternative, which is to populate dummy controls and stats for memory
> types that do not support a given size.
> 
> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
> ---
>   mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>   1 file changed, 114 insertions(+), 30 deletions(-)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 0c3075ee00012..082d86b7c6c2f 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>   static DEFINE_SPINLOCK(huge_anon_orders_lock);
>   static LIST_HEAD(thpsize_list);
>   
> -static ssize_t thpsize_enabled_show(struct kobject *kobj,
> -				    struct kobj_attribute *attr, char *buf)
> +static ssize_t anon_enabled_show(struct kobject *kobj,
> +				 struct kobj_attribute *attr, char *buf)
>   {
>   	int order = to_thpsize(kobj)->order;
>   	const char *output;
> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>   	return sysfs_emit(buf, "%s\n", output);
>   }
>   
> -static ssize_t thpsize_enabled_store(struct kobject *kobj,
> -				     struct kobj_attribute *attr,
> -				     const char *buf, size_t count)
> +static ssize_t anon_enabled_store(struct kobject *kobj,
> +				  struct kobj_attribute *attr,
> +				  const char *buf, size_t count)
>   {
>   	int order = to_thpsize(kobj)->order;
>   	ssize_t ret = count;
> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>   	return ret;
>   }
>   
> -static struct kobj_attribute thpsize_enabled_attr =
> -	__ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
> +static struct kobj_attribute anon_enabled_attr =
> +	__ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>   
> -static struct attribute *thpsize_attrs[] = {
> -	&thpsize_enabled_attr.attr,
> +static struct attribute *anon_ctrl_attrs[] = {
> +	&anon_enabled_attr.attr,
> +	NULL,
> +};
> +
> +static const struct attribute_group anon_ctrl_attr_grp = {
> +	.attrs = anon_ctrl_attrs,
> +};
> +
> +static struct attribute *file_ctrl_attrs[] = {
>   #ifdef CONFIG_SHMEM
>   	&thpsize_shmem_enabled_attr.attr,
>   #endif
>   	NULL,
>   };
>   
> -static const struct attribute_group thpsize_attr_group = {
> -	.attrs = thpsize_attrs,
> +static const struct attribute_group file_ctrl_attr_grp = {
> +	.attrs = file_ctrl_attrs,
> +};
> +
> +static struct attribute *any_ctrl_attrs[] = {
> +	NULL,
> +};
> +
> +static const struct attribute_group any_ctrl_attr_grp = {
> +	.attrs = any_ctrl_attrs,
>   };

I wonder why adding a NULL group?

>   
>   static const struct kobj_type thpsize_ktype = {
> @@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);
>   DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
>   DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
>   DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
> +#ifdef CONFIG_SHMEM
>   DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
>   DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
>   DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
> +#endif
>   DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>   DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>   DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>   
> -static struct attribute *stats_attrs[] = {
> +static struct attribute *anon_stats_attrs[] = {
>   	&anon_fault_alloc_attr.attr,
>   	&anon_fault_fallback_attr.attr,
>   	&anon_fault_fallback_charge_attr.attr,
> +#ifndef CONFIG_SHMEM
>   	&swpout_attr.attr,
>   	&swpout_fallback_attr.attr,
> -	&shmem_alloc_attr.attr,
> -	&shmem_fallback_attr.attr,
> -	&shmem_fallback_charge_attr.attr,
> +#endif
>   	&split_attr.attr,
>   	&split_failed_attr.attr,
>   	&split_deferred_attr.attr,
>   	NULL,
>   };
>   
> -static struct attribute_group stats_attr_group = {
> +static struct attribute_group anon_stats_attr_grp = {
> +	.name = "stats",
> +	.attrs = anon_stats_attrs,
> +};
> +
> +static struct attribute *file_stats_attrs[] = {
> +#ifdef CONFIG_SHMEM
> +	&shmem_alloc_attr.attr,
> +	&shmem_fallback_attr.attr,
> +	&shmem_fallback_charge_attr.attr,
> +#endif
> +	NULL,
> +};
> +
> +static struct attribute_group file_stats_attr_grp = {
> +	.name = "stats",
> +	.attrs = file_stats_attrs,
> +};
> +
> +static struct attribute *any_stats_attrs[] = {
> +#ifdef CONFIG_SHMEM
> +	&swpout_attr.attr,
> +	&swpout_fallback_attr.attr,
> +#endif

Sorry I did not point it out in early version. I think file pages and 
shmem can also be split, while 'split_deferred' is only for anonymous 
page. So I think the any_stats_attrs should be:
static struct attribute *any_stats_attrs[] = {
#ifdef CONFIG_SHMEM
	&swpout_attr.attr,
	&swpout_fallback_attr.attr,
#endif
	&split_attr.attr,
	&split_failed_attr.attr,
	NULL,
};

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Ryan Roberts]

Hi Baolin,

Thanks for the review - I've been out on Paternity leave so only getting around
to replying now...

On 09/08/2024 09:31, Baolin Wang wrote:
> 
> 
> On 2024/8/8 19:18, Ryan Roberts wrote:
>> Previously we had a situation where shmem mTHP controls and stats were
>> not exposed for some supported sizes and were exposed for some
>> unsupported sizes. So let's clean that up.
>>
>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
>> shmem controls and stats were previously being exposed for all the anon
>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
>> pages, orders 12 and 13 were exposed but were not supported internally.
>>
>> Tidy this all up by defining ctrl and stats attribute groups for anon
>> and file separately. Anon ctrl and stats groups are populated for all
>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>>
>> Additionally, create "any" ctrl and stats attribute groups which are
>> populated for all orders in (THP_ORDERS_ALL_ANON |
>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
>> anon and shmem.
>>
>> The side-effect of all this is that different hugepage-*kB directories
>> contain different sets of controls and stats, depending on which memory
>> types support that size. This approach is preferred over the
>> alternative, which is to populate dummy controls and stats for memory
>> types that do not support a given size.
>>
>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>> ---
>>   mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>>   1 file changed, 114 insertions(+), 30 deletions(-)
>>
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 0c3075ee00012..082d86b7c6c2f 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>>   static DEFINE_SPINLOCK(huge_anon_orders_lock);
>>   static LIST_HEAD(thpsize_list);
>>   -static ssize_t thpsize_enabled_show(struct kobject *kobj,
>> -                    struct kobj_attribute *attr, char *buf)
>> +static ssize_t anon_enabled_show(struct kobject *kobj,
>> +                 struct kobj_attribute *attr, char *buf)
>>   {
>>       int order = to_thpsize(kobj)->order;
>>       const char *output;
>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>       return sysfs_emit(buf, "%s\n", output);
>>   }
>>   -static ssize_t thpsize_enabled_store(struct kobject *kobj,
>> -                     struct kobj_attribute *attr,
>> -                     const char *buf, size_t count)
>> +static ssize_t anon_enabled_store(struct kobject *kobj,
>> +                  struct kobj_attribute *attr,
>> +                  const char *buf, size_t count)
>>   {
>>       int order = to_thpsize(kobj)->order;
>>       ssize_t ret = count;
>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>       return ret;
>>   }
>>   -static struct kobj_attribute thpsize_enabled_attr =
>> -    __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
>> +static struct kobj_attribute anon_enabled_attr =
>> +    __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>>   -static struct attribute *thpsize_attrs[] = {
>> -    &thpsize_enabled_attr.attr,
>> +static struct attribute *anon_ctrl_attrs[] = {
>> +    &anon_enabled_attr.attr,
>> +    NULL,
>> +};
>> +
>> +static const struct attribute_group anon_ctrl_attr_grp = {
>> +    .attrs = anon_ctrl_attrs,
>> +};
>> +
>> +static struct attribute *file_ctrl_attrs[] = {
>>   #ifdef CONFIG_SHMEM
>>       &thpsize_shmem_enabled_attr.attr,
>>   #endif
>>       NULL,
>>   };
>>   -static const struct attribute_group thpsize_attr_group = {
>> -    .attrs = thpsize_attrs,
>> +static const struct attribute_group file_ctrl_attr_grp = {
>> +    .attrs = file_ctrl_attrs,
>> +};
>> +
>> +static struct attribute *any_ctrl_attrs[] = {
>> +    NULL,
>> +};
>> +
>> +static const struct attribute_group any_ctrl_attr_grp = {
>> +    .attrs = any_ctrl_attrs,
>>   };
> 
> I wonder why adding a NULL group?

It made everything a bit more generic and therefore extensible. Its my
preference to leave it as is, but will remove it if you insist.

> 
>>     static const struct kobj_type thpsize_ktype = {
>> @@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback,
>> MTHP_STAT_ANON_FAULT_FALLBACK);
>>   DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge,
>> MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
>>   DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
>>   DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
>> +#ifdef CONFIG_SHMEM
>>   DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
>>   DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
>>   DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>> +#endif
>>   DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>   DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>   DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>>   -static struct attribute *stats_attrs[] = {
>> +static struct attribute *anon_stats_attrs[] = {
>>       &anon_fault_alloc_attr.attr,
>>       &anon_fault_fallback_attr.attr,
>>       &anon_fault_fallback_charge_attr.attr,
>> +#ifndef CONFIG_SHMEM
>>       &swpout_attr.attr,
>>       &swpout_fallback_attr.attr,
>> -    &shmem_alloc_attr.attr,
>> -    &shmem_fallback_attr.attr,
>> -    &shmem_fallback_charge_attr.attr,
>> +#endif
>>       &split_attr.attr,
>>       &split_failed_attr.attr,
>>       &split_deferred_attr.attr,
>>       NULL,
>>   };
>>   -static struct attribute_group stats_attr_group = {
>> +static struct attribute_group anon_stats_attr_grp = {
>> +    .name = "stats",
>> +    .attrs = anon_stats_attrs,
>> +};
>> +
>> +static struct attribute *file_stats_attrs[] = {
>> +#ifdef CONFIG_SHMEM
>> +    &shmem_alloc_attr.attr,
>> +    &shmem_fallback_attr.attr,
>> +    &shmem_fallback_charge_attr.attr,
>> +#endif
>> +    NULL,
>> +};
>> +
>> +static struct attribute_group file_stats_attr_grp = {
>> +    .name = "stats",
>> +    .attrs = file_stats_attrs,
>> +};
>> +
>> +static struct attribute *any_stats_attrs[] = {
>> +#ifdef CONFIG_SHMEM
>> +    &swpout_attr.attr,
>> +    &swpout_fallback_attr.attr,
>> +#endif
> 
> Sorry I did not point it out in early version. I think file pages and shmem can
> also be split, while 'split_deferred' is only for anonymous page. So I think the
> any_stats_attrs should be:
> static struct attribute *any_stats_attrs[] = {
> #ifdef CONFIG_SHMEM
>     &swpout_attr.attr,
>     &swpout_fallback_attr.attr,
> #endif
>     &split_attr.attr,
>     &split_failed_attr.attr,
>     NULL,
> };

Ahh yes, good point. I'll fix this in the next version (I'm back in office on
Wednesday so will send it then).

Thanks,
Ryan

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Baolin Wang]

On 2024/9/2 17:58, Ryan Roberts wrote:
> Hi Baolin,
> 
> Thanks for the review - I've been out on Paternity leave so only getting around
> to replying now...

No worries :)

> On 09/08/2024 09:31, Baolin Wang wrote:
>>
>>
>> On 2024/8/8 19:18, Ryan Roberts wrote:
>>> Previously we had a situation where shmem mTHP controls and stats were
>>> not exposed for some supported sizes and were exposed for some
>>> unsupported sizes. So let's clean that up.
>>>
>>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
>>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
>>> shmem controls and stats were previously being exposed for all the anon
>>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
>>> pages, orders 12 and 13 were exposed but were not supported internally.
>>>
>>> Tidy this all up by defining ctrl and stats attribute groups for anon
>>> and file separately. Anon ctrl and stats groups are populated for all
>>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
>>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>>>
>>> Additionally, create "any" ctrl and stats attribute groups which are
>>> populated for all orders in (THP_ORDERS_ALL_ANON |
>>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
>>> anon and shmem.
>>>
>>> The side-effect of all this is that different hugepage-*kB directories
>>> contain different sets of controls and stats, depending on which memory
>>> types support that size. This approach is preferred over the
>>> alternative, which is to populate dummy controls and stats for memory
>>> types that do not support a given size.
>>>
>>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>>> ---
>>>    mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>>>    1 file changed, 114 insertions(+), 30 deletions(-)
>>>
>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>> index 0c3075ee00012..082d86b7c6c2f 100644
>>> --- a/mm/huge_memory.c
>>> +++ b/mm/huge_memory.c
>>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>>>    static DEFINE_SPINLOCK(huge_anon_orders_lock);
>>>    static LIST_HEAD(thpsize_list);
>>>    -static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>> -                    struct kobj_attribute *attr, char *buf)
>>> +static ssize_t anon_enabled_show(struct kobject *kobj,
>>> +                 struct kobj_attribute *attr, char *buf)
>>>    {
>>>        int order = to_thpsize(kobj)->order;
>>>        const char *output;
>>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>>        return sysfs_emit(buf, "%s\n", output);
>>>    }
>>>    -static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>> -                     struct kobj_attribute *attr,
>>> -                     const char *buf, size_t count)
>>> +static ssize_t anon_enabled_store(struct kobject *kobj,
>>> +                  struct kobj_attribute *attr,
>>> +                  const char *buf, size_t count)
>>>    {
>>>        int order = to_thpsize(kobj)->order;
>>>        ssize_t ret = count;
>>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>>        return ret;
>>>    }
>>>    -static struct kobj_attribute thpsize_enabled_attr =
>>> -    __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
>>> +static struct kobj_attribute anon_enabled_attr =
>>> +    __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>>>    -static struct attribute *thpsize_attrs[] = {
>>> -    &thpsize_enabled_attr.attr,
>>> +static struct attribute *anon_ctrl_attrs[] = {
>>> +    &anon_enabled_attr.attr,
>>> +    NULL,
>>> +};
>>> +
>>> +static const struct attribute_group anon_ctrl_attr_grp = {
>>> +    .attrs = anon_ctrl_attrs,
>>> +};
>>> +
>>> +static struct attribute *file_ctrl_attrs[] = {
>>>    #ifdef CONFIG_SHMEM
>>>        &thpsize_shmem_enabled_attr.attr,
>>>    #endif
>>>        NULL,
>>>    };
>>>    -static const struct attribute_group thpsize_attr_group = {
>>> -    .attrs = thpsize_attrs,
>>> +static const struct attribute_group file_ctrl_attr_grp = {
>>> +    .attrs = file_ctrl_attrs,
>>> +};
>>> +
>>> +static struct attribute *any_ctrl_attrs[] = {
>>> +    NULL,
>>> +};
>>> +
>>> +static const struct attribute_group any_ctrl_attr_grp = {
>>> +    .attrs = any_ctrl_attrs,
>>>    };
>>
>> I wonder why adding a NULL group?
> 
> It made everything a bit more generic and therefore extensible. Its my
> preference to leave it as is, but will remove it if you insist.

My preference is we should add it when necessary, but but I don't have a 
strong opinion. Let's see what other guys prefer, David, Barry?

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Barry Song]

On Tue, Sep 3, 2024 at 1:15 PM Baolin Wang
<baolin.wang@linux.alibaba.com> wrote:
>
>
>
> On 2024/9/2 17:58, Ryan Roberts wrote:
> > Hi Baolin,
> >
> > Thanks for the review - I've been out on Paternity leave so only getting around
> > to replying now...
>
> No worries :)
>
> > On 09/08/2024 09:31, Baolin Wang wrote:
> >>
> >>
> >> On 2024/8/8 19:18, Ryan Roberts wrote:
> >>> Previously we had a situation where shmem mTHP controls and stats were
> >>> not exposed for some supported sizes and were exposed for some
> >>> unsupported sizes. So let's clean that up.
> >>>
> >>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
> >>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
> >>> shmem controls and stats were previously being exposed for all the anon
> >>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
> >>> pages, orders 12 and 13 were exposed but were not supported internally.
> >>>
> >>> Tidy this all up by defining ctrl and stats attribute groups for anon
> >>> and file separately. Anon ctrl and stats groups are populated for all
> >>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
> >>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
> >>>
> >>> Additionally, create "any" ctrl and stats attribute groups which are
> >>> populated for all orders in (THP_ORDERS_ALL_ANON |
> >>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
> >>> anon and shmem.
> >>>
> >>> The side-effect of all this is that different hugepage-*kB directories
> >>> contain different sets of controls and stats, depending on which memory
> >>> types support that size. This approach is preferred over the
> >>> alternative, which is to populate dummy controls and stats for memory
> >>> types that do not support a given size.
> >>>
> >>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
> >>> ---
> >>>    mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
> >>>    1 file changed, 114 insertions(+), 30 deletions(-)
> >>>
> >>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> >>> index 0c3075ee00012..082d86b7c6c2f 100644
> >>> --- a/mm/huge_memory.c
> >>> +++ b/mm/huge_memory.c
> >>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
> >>>    static DEFINE_SPINLOCK(huge_anon_orders_lock);
> >>>    static LIST_HEAD(thpsize_list);
> >>>    -static ssize_t thpsize_enabled_show(struct kobject *kobj,
> >>> -                    struct kobj_attribute *attr, char *buf)
> >>> +static ssize_t anon_enabled_show(struct kobject *kobj,
> >>> +                 struct kobj_attribute *attr, char *buf)
> >>>    {
> >>>        int order = to_thpsize(kobj)->order;
> >>>        const char *output;
> >>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
> >>>        return sysfs_emit(buf, "%s\n", output);
> >>>    }
> >>>    -static ssize_t thpsize_enabled_store(struct kobject *kobj,
> >>> -                     struct kobj_attribute *attr,
> >>> -                     const char *buf, size_t count)
> >>> +static ssize_t anon_enabled_store(struct kobject *kobj,
> >>> +                  struct kobj_attribute *attr,
> >>> +                  const char *buf, size_t count)
> >>>    {
> >>>        int order = to_thpsize(kobj)->order;
> >>>        ssize_t ret = count;
> >>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
> >>>        return ret;
> >>>    }
> >>>    -static struct kobj_attribute thpsize_enabled_attr =
> >>> -    __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
> >>> +static struct kobj_attribute anon_enabled_attr =
> >>> +    __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
> >>>    -static struct attribute *thpsize_attrs[] = {
> >>> -    &thpsize_enabled_attr.attr,
> >>> +static struct attribute *anon_ctrl_attrs[] = {
> >>> +    &anon_enabled_attr.attr,
> >>> +    NULL,
> >>> +};
> >>> +
> >>> +static const struct attribute_group anon_ctrl_attr_grp = {
> >>> +    .attrs = anon_ctrl_attrs,
> >>> +};
> >>> +
> >>> +static struct attribute *file_ctrl_attrs[] = {
> >>>    #ifdef CONFIG_SHMEM
> >>>        &thpsize_shmem_enabled_attr.attr,
> >>>    #endif
> >>>        NULL,
> >>>    };
> >>>    -static const struct attribute_group thpsize_attr_group = {
> >>> -    .attrs = thpsize_attrs,
> >>> +static const struct attribute_group file_ctrl_attr_grp = {
> >>> +    .attrs = file_ctrl_attrs,
> >>> +};
> >>> +
> >>> +static struct attribute *any_ctrl_attrs[] = {
> >>> +    NULL,
> >>> +};
> >>> +
> >>> +static const struct attribute_group any_ctrl_attr_grp = {
> >>> +    .attrs = any_ctrl_attrs,
> >>>    };
> >>
> >> I wonder why adding a NULL group?
> >
> > It made everything a bit more generic and therefore extensible. Its my
> > preference to leave it as is, but will remove it if you insist.
>
> My preference is we should add it when necessary, but but I don't have a
> strong opinion. Let's see what other guys prefer, David, Barry?

I'm fine with either option. Adding a NULL control group makes it
easier for lazy
people like me to understand the current status, as it clearly
indicates that there
isn't a shared control group for file, shmem, and anon at the moment. :-)

Thanks
Barry

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Ryan Roberts]

On 03/09/2024 02:53, Barry Song wrote:
> On Tue, Sep 3, 2024 at 1:15 PM Baolin Wang
> <baolin.wang@linux.alibaba.com> wrote:
>>
>>
>>
>> On 2024/9/2 17:58, Ryan Roberts wrote:
>>> Hi Baolin,
>>>
>>> Thanks for the review - I've been out on Paternity leave so only getting around
>>> to replying now...
>>
>> No worries :)
>>
>>> On 09/08/2024 09:31, Baolin Wang wrote:
>>>>
>>>>
>>>> On 2024/8/8 19:18, Ryan Roberts wrote:
>>>>> Previously we had a situation where shmem mTHP controls and stats were
>>>>> not exposed for some supported sizes and were exposed for some
>>>>> unsupported sizes. So let's clean that up.
>>>>>
>>>>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
>>>>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
>>>>> shmem controls and stats were previously being exposed for all the anon
>>>>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
>>>>> pages, orders 12 and 13 were exposed but were not supported internally.
>>>>>
>>>>> Tidy this all up by defining ctrl and stats attribute groups for anon
>>>>> and file separately. Anon ctrl and stats groups are populated for all
>>>>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
>>>>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>>>>>
>>>>> Additionally, create "any" ctrl and stats attribute groups which are
>>>>> populated for all orders in (THP_ORDERS_ALL_ANON |
>>>>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
>>>>> anon and shmem.
>>>>>
>>>>> The side-effect of all this is that different hugepage-*kB directories
>>>>> contain different sets of controls and stats, depending on which memory
>>>>> types support that size. This approach is preferred over the
>>>>> alternative, which is to populate dummy controls and stats for memory
>>>>> types that do not support a given size.
>>>>>
>>>>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>>>>> ---
>>>>>    mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>>>>>    1 file changed, 114 insertions(+), 30 deletions(-)
>>>>>
>>>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>>>> index 0c3075ee00012..082d86b7c6c2f 100644
>>>>> --- a/mm/huge_memory.c
>>>>> +++ b/mm/huge_memory.c
>>>>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>>>>>    static DEFINE_SPINLOCK(huge_anon_orders_lock);
>>>>>    static LIST_HEAD(thpsize_list);
>>>>>    -static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>>>> -                    struct kobj_attribute *attr, char *buf)
>>>>> +static ssize_t anon_enabled_show(struct kobject *kobj,
>>>>> +                 struct kobj_attribute *attr, char *buf)
>>>>>    {
>>>>>        int order = to_thpsize(kobj)->order;
>>>>>        const char *output;
>>>>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>>>>        return sysfs_emit(buf, "%s\n", output);
>>>>>    }
>>>>>    -static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>>>> -                     struct kobj_attribute *attr,
>>>>> -                     const char *buf, size_t count)
>>>>> +static ssize_t anon_enabled_store(struct kobject *kobj,
>>>>> +                  struct kobj_attribute *attr,
>>>>> +                  const char *buf, size_t count)
>>>>>    {
>>>>>        int order = to_thpsize(kobj)->order;
>>>>>        ssize_t ret = count;
>>>>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>>>>        return ret;
>>>>>    }
>>>>>    -static struct kobj_attribute thpsize_enabled_attr =
>>>>> -    __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
>>>>> +static struct kobj_attribute anon_enabled_attr =
>>>>> +    __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>>>>>    -static struct attribute *thpsize_attrs[] = {
>>>>> -    &thpsize_enabled_attr.attr,
>>>>> +static struct attribute *anon_ctrl_attrs[] = {
>>>>> +    &anon_enabled_attr.attr,
>>>>> +    NULL,
>>>>> +};
>>>>> +
>>>>> +static const struct attribute_group anon_ctrl_attr_grp = {
>>>>> +    .attrs = anon_ctrl_attrs,
>>>>> +};
>>>>> +
>>>>> +static struct attribute *file_ctrl_attrs[] = {
>>>>>    #ifdef CONFIG_SHMEM
>>>>>        &thpsize_shmem_enabled_attr.attr,
>>>>>    #endif
>>>>>        NULL,
>>>>>    };
>>>>>    -static const struct attribute_group thpsize_attr_group = {
>>>>> -    .attrs = thpsize_attrs,
>>>>> +static const struct attribute_group file_ctrl_attr_grp = {
>>>>> +    .attrs = file_ctrl_attrs,
>>>>> +};
>>>>> +
>>>>> +static struct attribute *any_ctrl_attrs[] = {
>>>>> +    NULL,
>>>>> +};
>>>>> +
>>>>> +static const struct attribute_group any_ctrl_attr_grp = {
>>>>> +    .attrs = any_ctrl_attrs,
>>>>>    };
>>>>
>>>> I wonder why adding a NULL group?
>>>
>>> It made everything a bit more generic and therefore extensible. Its my
>>> preference to leave it as is, but will remove it if you insist.
>>
>> My preference is we should add it when necessary, but but I don't have a
>> strong opinion. Let's see what other guys prefer, David, Barry?
> 
> I'm fine with either option. Adding a NULL control group makes it
> easier for lazy
> people like me to understand the current status, as it clearly
> indicates that there
> isn't a shared control group for file, shmem, and anon at the moment. :-)

Thanks for the feedback, I'm going to leave it as is in the next version then.

> 
> Thanks
> Barry

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Ryan Roberts]

Hi Andrew,


On 09/08/2024 09:31, Baolin Wang wrote:
> 
> 
> On 2024/8/8 19:18, Ryan Roberts wrote:
>> Previously we had a situation where shmem mTHP controls and stats were
>> not exposed for some supported sizes and were exposed for some
>> unsupported sizes. So let's clean that up.
>>
>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
>> shmem controls and stats were previously being exposed for all the anon
>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
>> pages, orders 12 and 13 were exposed but were not supported internally.
>>
>> Tidy this all up by defining ctrl and stats attribute groups for anon
>> and file separately. Anon ctrl and stats groups are populated for all
>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>>
>> Additionally, create "any" ctrl and stats attribute groups which are
>> populated for all orders in (THP_ORDERS_ALL_ANON |
>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
>> anon and shmem.
>>
>> The side-effect of all this is that different hugepage-*kB directories
>> contain different sets of controls and stats, depending on which memory
>> types support that size. This approach is preferred over the
>> alternative, which is to populate dummy controls and stats for memory
>> types that do not support a given size.
>>
>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>> ---
>>   mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>>   1 file changed, 114 insertions(+), 30 deletions(-)
>>
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 0c3075ee00012..082d86b7c6c2f 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>>   static DEFINE_SPINLOCK(huge_anon_orders_lock);
>>   static LIST_HEAD(thpsize_list);
>>   -static ssize_t thpsize_enabled_show(struct kobject *kobj,
>> -                    struct kobj_attribute *attr, char *buf)
>> +static ssize_t anon_enabled_show(struct kobject *kobj,
>> +                 struct kobj_attribute *attr, char *buf)
>>   {
>>       int order = to_thpsize(kobj)->order;
>>       const char *output;
>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>       return sysfs_emit(buf, "%s\n", output);
>>   }
>>   -static ssize_t thpsize_enabled_store(struct kobject *kobj,
>> -                     struct kobj_attribute *attr,
>> -                     const char *buf, size_t count)
>> +static ssize_t anon_enabled_store(struct kobject *kobj,
>> +                  struct kobj_attribute *attr,
>> +                  const char *buf, size_t count)
>>   {
>>       int order = to_thpsize(kobj)->order;
>>       ssize_t ret = count;
>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>       return ret;
>>   }
>>   -static struct kobj_attribute thpsize_enabled_attr =
>> -    __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
>> +static struct kobj_attribute anon_enabled_attr =
>> +    __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>>   -static struct attribute *thpsize_attrs[] = {
>> -    &thpsize_enabled_attr.attr,
>> +static struct attribute *anon_ctrl_attrs[] = {
>> +    &anon_enabled_attr.attr,
>> +    NULL,
>> +};
>> +
>> +static const struct attribute_group anon_ctrl_attr_grp = {
>> +    .attrs = anon_ctrl_attrs,
>> +};
>> +
>> +static struct attribute *file_ctrl_attrs[] = {
>>   #ifdef CONFIG_SHMEM
>>       &thpsize_shmem_enabled_attr.attr,
>>   #endif
>>       NULL,
>>   };
>>   -static const struct attribute_group thpsize_attr_group = {
>> -    .attrs = thpsize_attrs,
>> +static const struct attribute_group file_ctrl_attr_grp = {
>> +    .attrs = file_ctrl_attrs,
>> +};
>> +
>> +static struct attribute *any_ctrl_attrs[] = {
>> +    NULL,
>> +};
>> +
>> +static const struct attribute_group any_ctrl_attr_grp = {
>> +    .attrs = any_ctrl_attrs,
>>   };
> 
> I wonder why adding a NULL group?
> 
>>     static const struct kobj_type thpsize_ktype = {
>> @@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback,
>> MTHP_STAT_ANON_FAULT_FALLBACK);
>>   DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge,
>> MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
>>   DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
>>   DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
>> +#ifdef CONFIG_SHMEM
>>   DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
>>   DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
>>   DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>> +#endif
>>   DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>   DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>   DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>>   -static struct attribute *stats_attrs[] = {
>> +static struct attribute *anon_stats_attrs[] = {
>>       &anon_fault_alloc_attr.attr,
>>       &anon_fault_fallback_attr.attr,
>>       &anon_fault_fallback_charge_attr.attr,
>> +#ifndef CONFIG_SHMEM
>>       &swpout_attr.attr,
>>       &swpout_fallback_attr.attr,
>> -    &shmem_alloc_attr.attr,
>> -    &shmem_fallback_attr.attr,
>> -    &shmem_fallback_charge_attr.attr,
>> +#endif
>>       &split_attr.attr,
>>       &split_failed_attr.attr,
>>       &split_deferred_attr.attr,
>>       NULL,
>>   };
>>   -static struct attribute_group stats_attr_group = {
>> +static struct attribute_group anon_stats_attr_grp = {
>> +    .name = "stats",
>> +    .attrs = anon_stats_attrs,
>> +};
>> +
>> +static struct attribute *file_stats_attrs[] = {
>> +#ifdef CONFIG_SHMEM
>> +    &shmem_alloc_attr.attr,
>> +    &shmem_fallback_attr.attr,
>> +    &shmem_fallback_charge_attr.attr,
>> +#endif
>> +    NULL,
>> +};
>> +
>> +static struct attribute_group file_stats_attr_grp = {
>> +    .name = "stats",
>> +    .attrs = file_stats_attrs,
>> +};
>> +
>> +static struct attribute *any_stats_attrs[] = {
>> +#ifdef CONFIG_SHMEM
>> +    &swpout_attr.attr,
>> +    &swpout_fallback_attr.attr,
>> +#endif
> 
> Sorry I did not point it out in early version. I think file pages and shmem can
> also be split, while 'split_deferred' is only for anonymous page. So I think the
> any_stats_attrs should be:
> static struct attribute *any_stats_attrs[] = {
> #ifdef CONFIG_SHMEM
>     &swpout_attr.attr,
>     &swpout_fallback_attr.attr,
> #endif
>     &split_attr.attr,
>     &split_failed_attr.attr,
>     NULL,
> };

Could you please squash the following into this patch, which is already in
mm-unstable? I'm hoping this sufficient and I don't need to send a whole new
revision since there are changes on top of this in mm-unstable, which makes
things tricky.

----8<-----

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 382938e46f96f..5905957b1f70d 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -631,8 +631,6 @@ static struct attribute *anon_stats_attrs[] = {
        &swpout_attr.attr,
        &swpout_fallback_attr.attr,
 #endif
-       &split_attr.attr,
-       &split_failed_attr.attr,
        &split_deferred_attr.attr,
        &nr_anon_attr.attr,
        &nr_anon_partially_mapped_attr.attr,
@@ -663,6 +661,8 @@ static struct attribute *any_stats_attrs[] = {
        &swpout_attr.attr,
        &swpout_fallback_attr.attr,
 #endif
+       &split_attr.attr,
+       &split_failed_attr.attr,
        NULL,
 };

----8<-----

Thanks,
Ryan

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Andrew Morton]

On Wed, 4 Sep 2024 11:47:29 +0100 Ryan Roberts <ryan.roberts@arm.com> wrote:

> >> +static struct attribute *any_ctrl_attrs[] = {
> >> +    NULL,
> >> +};
> >> +
> >> +static const struct attribute_group any_ctrl_attr_grp = {
> >> +    .attrs = any_ctrl_attrs,
> >>   };
> > 
> > I wonder why adding a NULL group?

Was this review comment addressed?

>
> ...
>
> >> +    &shmem_alloc_attr.attr,
> >> +    &shmem_fallback_attr.attr,
> >> +    &shmem_fallback_charge_attr.attr,
> >> +#endif
> >> +    NULL,
> >> +};
> >> +
> >> +static struct attribute_group file_stats_attr_grp = {
> >> +    .name = "stats",
> >> +    .attrs = file_stats_attrs,
> >> +};
> >> +
> >> +static struct attribute *any_stats_attrs[] = {
> >> +#ifdef CONFIG_SHMEM
> >> +    &swpout_attr.attr,
> >> +    &swpout_fallback_attr.attr,
> >> +#endif
> > 
> > Sorry I did not point it out in early version. I think file pages and shmem can
> > also be split, while 'split_deferred' is only for anonymous page. So I think the
> > any_stats_attrs should be:
> > static struct attribute *any_stats_attrs[] = {
> > #ifdef CONFIG_SHMEM
> >     &swpout_attr.attr,
> >     &swpout_fallback_attr.attr,
> > #endif
> >     &split_attr.attr,
> >     &split_failed_attr.attr,
> >     NULL,
> > };
> 
> Could you please squash the following into this patch, which is already in
> mm-unstable? I'm hoping this sufficient and I don't need to send a whole new
> revision since there are changes on top of this in mm-unstable, which makes
> things tricky.

I did that.  Please send along a Signoff and a changelog?

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Ryan Roberts]

On 04/09/2024 21:46, Andrew Morton wrote:
> On Wed, 4 Sep 2024 11:47:29 +0100 Ryan Roberts <ryan.roberts@arm.com> wrote:
> 
>>>> +static struct attribute *any_ctrl_attrs[] = {
>>>> +    NULL,
>>>> +};
>>>> +
>>>> +static const struct attribute_group any_ctrl_attr_grp = {
>>>> +    .attrs = any_ctrl_attrs,
>>>>   };
>>>
>>> I wonder why adding a NULL group?
> 
> Was this review comment addressed?

Yes, we had discussion at [1] and concluded to leave it as is; Baolin had a
preference to remove it (but no strong opinion) and Barry preferred it as is to
explicitly indicate that there are no controls shared by file/shmem/anon currently.

[1]
https://lore.kernel.org/linux-mm/CAGsJ_4z+yERPLwzm-8Mkx8MsNZAz0zZWycZfuGRjOc4kxS=HwQ@mail.gmail.com/


> 
>>
>> ...
>>
>>>> +    &shmem_alloc_attr.attr,
>>>> +    &shmem_fallback_attr.attr,
>>>> +    &shmem_fallback_charge_attr.attr,
>>>> +#endif
>>>> +    NULL,
>>>> +};
>>>> +
>>>> +static struct attribute_group file_stats_attr_grp = {
>>>> +    .name = "stats",
>>>> +    .attrs = file_stats_attrs,
>>>> +};
>>>> +
>>>> +static struct attribute *any_stats_attrs[] = {
>>>> +#ifdef CONFIG_SHMEM
>>>> +    &swpout_attr.attr,
>>>> +    &swpout_fallback_attr.attr,
>>>> +#endif
>>>
>>> Sorry I did not point it out in early version. I think file pages and shmem can
>>> also be split, while 'split_deferred' is only for anonymous page. So I think the
>>> any_stats_attrs should be:
>>> static struct attribute *any_stats_attrs[] = {
>>> #ifdef CONFIG_SHMEM
>>>     &swpout_attr.attr,
>>>     &swpout_fallback_attr.attr,
>>> #endif
>>>     &split_attr.attr,
>>>     &split_failed_attr.attr,
>>>     NULL,
>>> };
>>
>> Could you please squash the following into this patch, which is already in
>> mm-unstable? I'm hoping this sufficient and I don't need to send a whole new
>> revision since there are changes on top of this in mm-unstable, which makes
>> things tricky.
> 
> I did that.  Please send along a Signoff and a changelog?

Ahh sorry, so even though you are squashing into an existing patch which has the
change log and sign-off, the process still requires a separate one here? My bad.

----8<----

Fixup commit ("mm: Tidy up shmem mTHP controls and stats") by moving split_attr
and split_failed_attr stats to the any_stats_attrs group. All memory types can
have their folios split, so it was incorrect to only show this stat for anon memory.


Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>

----8<----

Thanks,
Ryan

Re: [PATCH v3 2/2] mm: Tidy up shmem mTHP controls and stats

[Baolin Wang]

On 2024/9/4 18:47, Ryan Roberts wrote:
> Hi Andrew,
> 
> 
> On 09/08/2024 09:31, Baolin Wang wrote:
>>
>>
>> On 2024/8/8 19:18, Ryan Roberts wrote:
>>> Previously we had a situation where shmem mTHP controls and stats were
>>> not exposed for some supported sizes and were exposed for some
>>> unsupported sizes. So let's clean that up.
>>>
>>> Anon mTHP can support all large orders [2, PMD_ORDER]. But shmem can
>>> support all large orders [1, MAX_PAGECACHE_ORDER]. However, per-size
>>> shmem controls and stats were previously being exposed for all the anon
>>> mTHP orders, meaning order-1 was not present, and for arm64 64K base
>>> pages, orders 12 and 13 were exposed but were not supported internally.
>>>
>>> Tidy this all up by defining ctrl and stats attribute groups for anon
>>> and file separately. Anon ctrl and stats groups are populated for all
>>> orders in THP_ORDERS_ALL_ANON and file ctrl and stats groups are
>>> populated for all orders in THP_ORDERS_ALL_FILE_DEFAULT.
>>>
>>> Additionally, create "any" ctrl and stats attribute groups which are
>>> populated for all orders in (THP_ORDERS_ALL_ANON |
>>> THP_ORDERS_ALL_FILE_DEFAULT). swpout stats use this since they apply to
>>> anon and shmem.
>>>
>>> The side-effect of all this is that different hugepage-*kB directories
>>> contain different sets of controls and stats, depending on which memory
>>> types support that size. This approach is preferred over the
>>> alternative, which is to populate dummy controls and stats for memory
>>> types that do not support a given size.
>>>
>>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>>> ---
>>>    mm/huge_memory.c | 144 +++++++++++++++++++++++++++++++++++++----------
>>>    1 file changed, 114 insertions(+), 30 deletions(-)
>>>
>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>> index 0c3075ee00012..082d86b7c6c2f 100644
>>> --- a/mm/huge_memory.c
>>> +++ b/mm/huge_memory.c
>>> @@ -482,8 +482,8 @@ static void thpsize_release(struct kobject *kobj);
>>>    static DEFINE_SPINLOCK(huge_anon_orders_lock);
>>>    static LIST_HEAD(thpsize_list);
>>>    -static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>> -                    struct kobj_attribute *attr, char *buf)
>>> +static ssize_t anon_enabled_show(struct kobject *kobj,
>>> +                 struct kobj_attribute *attr, char *buf)
>>>    {
>>>        int order = to_thpsize(kobj)->order;
>>>        const char *output;
>>> @@ -500,9 +500,9 @@ static ssize_t thpsize_enabled_show(struct kobject *kobj,
>>>        return sysfs_emit(buf, "%s\n", output);
>>>    }
>>>    -static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>> -                     struct kobj_attribute *attr,
>>> -                     const char *buf, size_t count)
>>> +static ssize_t anon_enabled_store(struct kobject *kobj,
>>> +                  struct kobj_attribute *attr,
>>> +                  const char *buf, size_t count)
>>>    {
>>>        int order = to_thpsize(kobj)->order;
>>>        ssize_t ret = count;
>>> @@ -544,19 +544,35 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
>>>        return ret;
>>>    }
>>>    -static struct kobj_attribute thpsize_enabled_attr =
>>> -    __ATTR(enabled, 0644, thpsize_enabled_show, thpsize_enabled_store);
>>> +static struct kobj_attribute anon_enabled_attr =
>>> +    __ATTR(enabled, 0644, anon_enabled_show, anon_enabled_store);
>>>    -static struct attribute *thpsize_attrs[] = {
>>> -    &thpsize_enabled_attr.attr,
>>> +static struct attribute *anon_ctrl_attrs[] = {
>>> +    &anon_enabled_attr.attr,
>>> +    NULL,
>>> +};
>>> +
>>> +static const struct attribute_group anon_ctrl_attr_grp = {
>>> +    .attrs = anon_ctrl_attrs,
>>> +};
>>> +
>>> +static struct attribute *file_ctrl_attrs[] = {
>>>    #ifdef CONFIG_SHMEM
>>>        &thpsize_shmem_enabled_attr.attr,
>>>    #endif
>>>        NULL,
>>>    };
>>>    -static const struct attribute_group thpsize_attr_group = {
>>> -    .attrs = thpsize_attrs,
>>> +static const struct attribute_group file_ctrl_attr_grp = {
>>> +    .attrs = file_ctrl_attrs,
>>> +};
>>> +
>>> +static struct attribute *any_ctrl_attrs[] = {
>>> +    NULL,
>>> +};
>>> +
>>> +static const struct attribute_group any_ctrl_attr_grp = {
>>> +    .attrs = any_ctrl_attrs,
>>>    };
>>
>> I wonder why adding a NULL group?
>>
>>>      static const struct kobj_type thpsize_ktype = {
>>> @@ -595,64 +611,132 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback,
>>> MTHP_STAT_ANON_FAULT_FALLBACK);
>>>    DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge,
>>> MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
>>>    DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT);
>>>    DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK);
>>> +#ifdef CONFIG_SHMEM
>>>    DEFINE_MTHP_STAT_ATTR(shmem_alloc, MTHP_STAT_SHMEM_ALLOC);
>>>    DEFINE_MTHP_STAT_ATTR(shmem_fallback, MTHP_STAT_SHMEM_FALLBACK);
>>>    DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>>> +#endif
>>>    DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>>    DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>>    DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>>>    -static struct attribute *stats_attrs[] = {
>>> +static struct attribute *anon_stats_attrs[] = {
>>>        &anon_fault_alloc_attr.attr,
>>>        &anon_fault_fallback_attr.attr,
>>>        &anon_fault_fallback_charge_attr.attr,
>>> +#ifndef CONFIG_SHMEM
>>>        &swpout_attr.attr,
>>>        &swpout_fallback_attr.attr,
>>> -    &shmem_alloc_attr.attr,
>>> -    &shmem_fallback_attr.attr,
>>> -    &shmem_fallback_charge_attr.attr,
>>> +#endif
>>>        &split_attr.attr,
>>>        &split_failed_attr.attr,
>>>        &split_deferred_attr.attr,
>>>        NULL,
>>>    };
>>>    -static struct attribute_group stats_attr_group = {
>>> +static struct attribute_group anon_stats_attr_grp = {
>>> +    .name = "stats",
>>> +    .attrs = anon_stats_attrs,
>>> +};
>>> +
>>> +static struct attribute *file_stats_attrs[] = {
>>> +#ifdef CONFIG_SHMEM
>>> +    &shmem_alloc_attr.attr,
>>> +    &shmem_fallback_attr.attr,
>>> +    &shmem_fallback_charge_attr.attr,
>>> +#endif
>>> +    NULL,
>>> +};
>>> +
>>> +static struct attribute_group file_stats_attr_grp = {
>>> +    .name = "stats",
>>> +    .attrs = file_stats_attrs,
>>> +};
>>> +
>>> +static struct attribute *any_stats_attrs[] = {
>>> +#ifdef CONFIG_SHMEM
>>> +    &swpout_attr.attr,
>>> +    &swpout_fallback_attr.attr,
>>> +#endif
>>
>> Sorry I did not point it out in early version. I think file pages and shmem can
>> also be split, while 'split_deferred' is only for anonymous page. So I think the
>> any_stats_attrs should be:
>> static struct attribute *any_stats_attrs[] = {
>> #ifdef CONFIG_SHMEM
>>      &swpout_attr.attr,
>>      &swpout_fallback_attr.attr,
>> #endif
>>      &split_attr.attr,
>>      &split_failed_attr.attr,
>>      NULL,
>> };
> 
> Could you please squash the following into this patch, which is already in
> mm-unstable? I'm hoping this sufficient and I don't need to send a whole new
> revision since there are changes on top of this in mm-unstable, which makes
> things tricky.
> 
> ----8<-----
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 382938e46f96f..5905957b1f70d 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -631,8 +631,6 @@ static struct attribute *anon_stats_attrs[] = {
>          &swpout_attr.attr,
>          &swpout_fallback_attr.attr,
>   #endif
> -       &split_attr.attr,
> -       &split_failed_attr.attr,
>          &split_deferred_attr.attr,
>          &nr_anon_attr.attr,
>          &nr_anon_partially_mapped_attr.attr,
> @@ -663,6 +661,8 @@ static struct attribute *any_stats_attrs[] = {
>          &swpout_attr.attr,
>          &swpout_fallback_attr.attr,
>   #endif
> +       &split_attr.attr,
> +       &split_failed_attr.attr,
>          NULL,
>   };

The change looks good to me. Feel free to add
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>