Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Vlastimil Babka]

On 6/4/25 11:51, Kirill A. Shutemov wrote:
> When compiling with MEMCG enabled but VM_EVENT_COUNTERS disabled,
> BUILD_BUG_ON() is triggered in vmstat_start because the vmstat_text
> array is larger than NR_VMSTAT_ITEMS.
> 
> This issue arises because some elements of the vmstat_text array are
> present when either MEMCG or VM_EVENT_COUNTERS is enabled, but
> NR_VMSTAT_ITEMS only accounts for these elements if VM_EVENT_COUNTERS is
> enabled.
> 
> The recent change in the BUILD_BUG_ON() check made it more strict,
> disallowing extra elements in the array, which revealed the issue.
> 
> Instead of adjusting the NR_VMSTAT_ITEMS definition to account for
> MEMCG, make MEMCG select VM_EVENT_COUNTERS. VM_EVENT_COUNTERS is
> enabled in most configurations anyway.
> 
> There is no need to backport this fix to stable trees. Without the
> strict BUILD_BUG_ON(), the issue is not harmful. The elements in
> question would only be read by the memcg code, not by /proc/vmstat.
> 
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Reported-by: Randy Dunlap <rdunlap@infradead.org>
> Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")

Well in that case I think we should put Fixes: to the BUILD_BUG_ON() change.
And if it's not yet a stable sha1, squash that together with this?
It doesn't seem ebc5d83d0443 alone needs this fix.

> Cc: Konstantin Khlebnikov <koct9i@gmail.com>

Otherwise,
Acked-by: Vlastimil Babka <vbabka@suse.cz>

> ---
>  include/linux/vmstat.h | 4 ++--
>  init/Kconfig           | 1 +
>  mm/vmstat.c            | 4 ++--
>  3 files changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
> index b2ccb6845595..c287998908bf 100644
> --- a/include/linux/vmstat.h
> +++ b/include/linux/vmstat.h
> @@ -507,7 +507,7 @@ static inline const char *lru_list_name(enum lru_list lru)
>  	return node_stat_name(NR_LRU_BASE + lru) + 3; // skip "nr_"
>  }
>  
> -#if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
> +#if defined(CONFIG_VM_EVENT_COUNTERS)
>  static inline const char *vm_event_name(enum vm_event_item item)
>  {
>  	return vmstat_text[NR_VM_ZONE_STAT_ITEMS +
> @@ -516,7 +516,7 @@ static inline const char *vm_event_name(enum vm_event_item item)
>  			   NR_VM_STAT_ITEMS +
>  			   item];
>  }
> -#endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
> +#endif /* CONFIG_VM_EVENT_COUNTERS */
>  
>  #ifdef CONFIG_MEMCG
>  
> diff --git a/init/Kconfig b/init/Kconfig
> index ab83abe0fd9d..dd332cac6036 100644
> --- a/init/Kconfig
> +++ b/init/Kconfig
> @@ -989,6 +989,7 @@ config MEMCG
>  	select PAGE_COUNTER
>  	select EVENTFD
>  	select SLAB_OBJ_EXT
> +	select VM_EVENT_COUNTERS
>  	help
>  	  Provides control over the memory footprint of tasks in a cgroup.
>  
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index 27dc37168cfd..c3114b8826e4 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -1301,7 +1301,7 @@ const char * const vmstat_text[] = {
>  	[I(NR_MEMMAP_BOOT_PAGES)]		= "nr_memmap_boot_pages",
>  #undef I
>  
> -#if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
> +#if defined(CONFIG_VM_EVENT_COUNTERS)
>  	/* enum vm_event_item counters */
>  #define I(x) (NR_VM_ZONE_STAT_ITEMS + NR_VM_NUMA_EVENT_ITEMS + \
>  	     NR_VM_NODE_STAT_ITEMS + NR_VM_STAT_ITEMS + x)
> @@ -1498,7 +1498,7 @@ const char * const vmstat_text[] = {
>  #endif
>  #endif
>  #undef I
> -#endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
> +#endif /* CONFIG_VM_EVENT_COUNTERS */
>  };
>  #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA || CONFIG_MEMCG */
>  

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Randy Dunlap]

On 6/4/25 2:51 AM, Kirill A. Shutemov wrote:
> When compiling with MEMCG enabled but VM_EVENT_COUNTERS disabled,
> BUILD_BUG_ON() is triggered in vmstat_start because the vmstat_text
> array is larger than NR_VMSTAT_ITEMS.
> 
> This issue arises because some elements of the vmstat_text array are
> present when either MEMCG or VM_EVENT_COUNTERS is enabled, but
> NR_VMSTAT_ITEMS only accounts for these elements if VM_EVENT_COUNTERS is
> enabled.
> 
> The recent change in the BUILD_BUG_ON() check made it more strict,
> disallowing extra elements in the array, which revealed the issue.
> 
> Instead of adjusting the NR_VMSTAT_ITEMS definition to account for
> MEMCG, make MEMCG select VM_EVENT_COUNTERS. VM_EVENT_COUNTERS is
> enabled in most configurations anyway.
> 
> There is no need to backport this fix to stable trees. Without the
> strict BUILD_BUG_ON(), the issue is not harmful. The elements in
> question would only be read by the memcg code, not by /proc/vmstat.
> 
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Reported-by: Randy Dunlap <rdunlap@infradead.org>
> Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")
> Cc: Konstantin Khlebnikov <koct9i@gmail.com>

Tested-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Randy Dunlap <rdunlap@infradead.org>

Thanks.

> ---
>  include/linux/vmstat.h | 4 ++--
>  init/Kconfig           | 1 +
>  mm/vmstat.c            | 4 ++--
>  3 files changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
> index b2ccb6845595..c287998908bf 100644
> --- a/include/linux/vmstat.h
> +++ b/include/linux/vmstat.h
> @@ -507,7 +507,7 @@ static inline const char *lru_list_name(enum lru_list lru)
>  	return node_stat_name(NR_LRU_BASE + lru) + 3; // skip "nr_"
>  }
>  
> -#if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
> +#if defined(CONFIG_VM_EVENT_COUNTERS)
>  static inline const char *vm_event_name(enum vm_event_item item)
>  {
>  	return vmstat_text[NR_VM_ZONE_STAT_ITEMS +
> @@ -516,7 +516,7 @@ static inline const char *vm_event_name(enum vm_event_item item)
>  			   NR_VM_STAT_ITEMS +
>  			   item];
>  }
> -#endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
> +#endif /* CONFIG_VM_EVENT_COUNTERS */
>  
>  #ifdef CONFIG_MEMCG
>  
> diff --git a/init/Kconfig b/init/Kconfig
> index ab83abe0fd9d..dd332cac6036 100644
> --- a/init/Kconfig
> +++ b/init/Kconfig
> @@ -989,6 +989,7 @@ config MEMCG
>  	select PAGE_COUNTER
>  	select EVENTFD
>  	select SLAB_OBJ_EXT
> +	select VM_EVENT_COUNTERS
>  	help
>  	  Provides control over the memory footprint of tasks in a cgroup.
>  
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index 27dc37168cfd..c3114b8826e4 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -1301,7 +1301,7 @@ const char * const vmstat_text[] = {
>  	[I(NR_MEMMAP_BOOT_PAGES)]		= "nr_memmap_boot_pages",
>  #undef I
>  
> -#if defined(CONFIG_VM_EVENT_COUNTERS) || defined(CONFIG_MEMCG)
> +#if defined(CONFIG_VM_EVENT_COUNTERS)
>  	/* enum vm_event_item counters */
>  #define I(x) (NR_VM_ZONE_STAT_ITEMS + NR_VM_NUMA_EVENT_ITEMS + \
>  	     NR_VM_NODE_STAT_ITEMS + NR_VM_STAT_ITEMS + x)
> @@ -1498,7 +1498,7 @@ const char * const vmstat_text[] = {
>  #endif
>  #endif
>  #undef I
> -#endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
> +#endif /* CONFIG_VM_EVENT_COUNTERS */
>  };
>  #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA || CONFIG_MEMCG */
>  

-- 
~Randy

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Shakeel Butt]

On Wed, Jun 04, 2025 at 12:51:11PM +0300, Kirill A. Shutemov wrote:
> When compiling with MEMCG enabled but VM_EVENT_COUNTERS disabled,
> BUILD_BUG_ON() is triggered in vmstat_start because the vmstat_text
> array is larger than NR_VMSTAT_ITEMS.
> 
> This issue arises because some elements of the vmstat_text array are
> present when either MEMCG or VM_EVENT_COUNTERS is enabled, but
> NR_VMSTAT_ITEMS only accounts for these elements if VM_EVENT_COUNTERS is
> enabled.
> 
> The recent change in the BUILD_BUG_ON() check made it more strict,
> disallowing extra elements in the array, which revealed the issue.
> 
> Instead of adjusting the NR_VMSTAT_ITEMS definition to account for
> MEMCG, make MEMCG select VM_EVENT_COUNTERS. VM_EVENT_COUNTERS is
> enabled in most configurations anyway.
> 
> There is no need to backport this fix to stable trees. Without the
> strict BUILD_BUG_ON(), the issue is not harmful. The elements in
> question would only be read by the memcg code, not by /proc/vmstat.
> 
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Reported-by: Randy Dunlap <rdunlap@infradead.org>
> Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")
> Cc: Konstantin Khlebnikov <koct9i@gmail.com>

Acked-by: Shakeel Butt <shakeel.butt@linux.dev>

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Andrew Morton]

On Wed, 4 Jun 2025 11:56:42 +0200 Vlastimil Babka <vbabka@suse.cz> wrote:

> > There is no need to backport this fix to stable trees. Without the
> > strict BUILD_BUG_ON(), the issue is not harmful. The elements in
> > question would only be read by the memcg code, not by /proc/vmstat.
> > 
> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> > Reported-by: Randy Dunlap <rdunlap@infradead.org>
> > Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")
> 
> Well in that case I think we should put Fixes: to the BUILD_BUG_ON() change.
> And if it's not yet a stable sha1, squash that together with this?
> It doesn't seem ebc5d83d0443 alone needs this fix.

I shuffled things around.

I moved "mm: strictly check vmstat_text array size" from mm-hotfixes
and back into mm-new for the next cycle.

I reworked "mm/vmstat: fix build with MEMCG=y and VM_EVENT_COUNTERS=n"
so it precedes "mm: strictly check vmstat_text array size".

I reworked "mm/vmstat: utilize designated initializers for the
vmstat_text array" so it comes last.


So the applying order is now

mm-hotfixes:
mm-fix-vmstat-after-removing-nr_bounce.patch

mm-new:
mm-vmstat-fix-build-with-memcg=y-and-vm_event_counters=n.patch
mm-strictly-check-vmstat_text-array-size.patch
mm-vmstat-utilize-designated-initializers-for-the-vmstat_text-array.patch

and everything should land nicely.

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Vlastimil Babka]

On 6/4/25 23:20, Andrew Morton wrote:
> On Wed, 4 Jun 2025 11:56:42 +0200 Vlastimil Babka <vbabka@suse.cz> wrote:
> 
>> > There is no need to backport this fix to stable trees. Without the
>> > strict BUILD_BUG_ON(), the issue is not harmful. The elements in
>> > question would only be read by the memcg code, not by /proc/vmstat.
>> > 
>> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
>> > Reported-by: Randy Dunlap <rdunlap@infradead.org>
>> > Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")
>> 
>> Well in that case I think we should put Fixes: to the BUILD_BUG_ON() change.
>> And if it's not yet a stable sha1, squash that together with this?
>> It doesn't seem ebc5d83d0443 alone needs this fix.
> 
> I shuffled things around.
> 
> I moved "mm: strictly check vmstat_text array size" from mm-hotfixes
> and back into mm-new for the next cycle.
> 
> I reworked "mm/vmstat: fix build with MEMCG=y and VM_EVENT_COUNTERS=n"
> so it precedes "mm: strictly check vmstat_text array size".
> 
> I reworked "mm/vmstat: utilize designated initializers for the
> vmstat_text array" so it comes last.
> 
> 
> So the applying order is now

I checked and in general it looks good, except a nit below.

> mm-hotfixes:
> mm-fix-vmstat-after-removing-nr_bounce.patch
> 
> mm-new:
> mm-vmstat-fix-build-with-memcg=y-and-vm_event_counters=n.patch
> mm-strictly-check-vmstat_text-array-size.patch

The changelogs of these two don't reflect the new ordering though, maybe
Kirill can provide updated ones?

> mm-vmstat-utilize-designated-initializers-for-the-vmstat_text-array.patch
> 
> and everything should land nicely.

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Kirill A. Shutemov]

On Thu, Jun 05, 2025 at 08:19:28AM +0200, Vlastimil Babka wrote:
> On 6/4/25 23:20, Andrew Morton wrote:
> > On Wed, 4 Jun 2025 11:56:42 +0200 Vlastimil Babka <vbabka@suse.cz> wrote:
> > 
> >> > There is no need to backport this fix to stable trees. Without the
> >> > strict BUILD_BUG_ON(), the issue is not harmful. The elements in
> >> > question would only be read by the memcg code, not by /proc/vmstat.
> >> > 
> >> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> >> > Reported-by: Randy Dunlap <rdunlap@infradead.org>
> >> > Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")
> >> 
> >> Well in that case I think we should put Fixes: to the BUILD_BUG_ON() change.
> >> And if it's not yet a stable sha1, squash that together with this?
> >> It doesn't seem ebc5d83d0443 alone needs this fix.
> > 
> > I shuffled things around.
> > 
> > I moved "mm: strictly check vmstat_text array size" from mm-hotfixes
> > and back into mm-new for the next cycle.
> > 
> > I reworked "mm/vmstat: fix build with MEMCG=y and VM_EVENT_COUNTERS=n"
> > so it precedes "mm: strictly check vmstat_text array size".
> > 
> > I reworked "mm/vmstat: utilize designated initializers for the
> > vmstat_text array" so it comes last.
> > 
> > 
> > So the applying order is now
> 
> I checked and in general it looks good, except a nit below.
> 
> > mm-hotfixes:
> > mm-fix-vmstat-after-removing-nr_bounce.patch
> > 
> > mm-new:
> > mm-vmstat-fix-build-with-memcg=y-and-vm_event_counters=n.patch
> > mm-strictly-check-vmstat_text-array-size.patch
> 
> The changelogs of these two don't reflect the new ordering though, maybe
> Kirill can provide updated ones?

Maybe something like this, for the first patch?


mm/vmstat: Make MEMCG select VM_EVENT_COUNTERS

The vmstat_text array contains labels for counters displayed in /proc/vmstat.
It is important to keep the labels in sync with the counters.

There is a BUILD_BUG_ON() check in vmstat_start() that ensures the size of the
vmstat_text is not smaller than VM_EVENT_COUNTERS. This helps to catch cases
where a new counter is added but the label is not. However, it does not help if
a counter is removed but the label remains.

It would be nice to make the BUILD_BUG_ON() check more strict to catch such
cases. However, when compiling with MEMCG enabled but VM_EVENT_COUNTERS
disabled, the vmstat_text array is larger than NR_VMSTAT_ITEMS.

This issue arises because some elements of the vmstat_text array are present
when either MEMCG or VM_EVENT_COUNTERS is enabled, but NR_VMSTAT_ITEMS only
accounts for these elements if VM_EVENT_COUNTERS is enabled.

Instead of adjusting the NR_VMSTAT_ITEMS definition to account for MEMCG, make
MEMCG select VM_EVENT_COUNTERS. VM_EVENT_COUNTERS is enabled in most
configurations anyway.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Vlastimil Babka]

On 6/5/25 13:53, Kirill A. Shutemov wrote:
> On Thu, Jun 05, 2025 at 08:19:28AM +0200, Vlastimil Babka wrote:
>> On 6/4/25 23:20, Andrew Morton wrote:
>> > On Wed, 4 Jun 2025 11:56:42 +0200 Vlastimil Babka <vbabka@suse.cz> wrote:
>> > 
>> >> > There is no need to backport this fix to stable trees. Without the
>> >> > strict BUILD_BUG_ON(), the issue is not harmful. The elements in
>> >> > question would only be read by the memcg code, not by /proc/vmstat.
>> >> > 
>> >> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
>> >> > Reported-by: Randy Dunlap <rdunlap@infradead.org>
>> >> > Fixes: ebc5d83d0443 ("mm/memcontrol: use vmstat names for printing statistics")
>> >> 
>> >> Well in that case I think we should put Fixes: to the BUILD_BUG_ON() change.
>> >> And if it's not yet a stable sha1, squash that together with this?
>> >> It doesn't seem ebc5d83d0443 alone needs this fix.
>> > 
>> > I shuffled things around.
>> > 
>> > I moved "mm: strictly check vmstat_text array size" from mm-hotfixes
>> > and back into mm-new for the next cycle.
>> > 
>> > I reworked "mm/vmstat: fix build with MEMCG=y and VM_EVENT_COUNTERS=n"
>> > so it precedes "mm: strictly check vmstat_text array size".
>> > 
>> > I reworked "mm/vmstat: utilize designated initializers for the
>> > vmstat_text array" so it comes last.
>> > 
>> > 
>> > So the applying order is now
>> 
>> I checked and in general it looks good, except a nit below.
>> 
>> > mm-hotfixes:
>> > mm-fix-vmstat-after-removing-nr_bounce.patch
>> > 
>> > mm-new:
>> > mm-vmstat-fix-build-with-memcg=y-and-vm_event_counters=n.patch
>> > mm-strictly-check-vmstat_text-array-size.patch
>> 
>> The changelogs of these two don't reflect the new ordering though, maybe
>> Kirill can provide updated ones?
> 
> Maybe something like this, for the first patch?

LGTM, thanks!

> > mm/vmstat: Make MEMCG select VM_EVENT_COUNTERS
> 
> The vmstat_text array contains labels for counters displayed in /proc/vmstat.
> It is important to keep the labels in sync with the counters.
> 
> There is a BUILD_BUG_ON() check in vmstat_start() that ensures the size of the
> vmstat_text is not smaller than VM_EVENT_COUNTERS. This helps to catch cases
> where a new counter is added but the label is not. However, it does not help if
> a counter is removed but the label remains.
> 
> It would be nice to make the BUILD_BUG_ON() check more strict to catch such
> cases. However, when compiling with MEMCG enabled but VM_EVENT_COUNTERS
> disabled, the vmstat_text array is larger than NR_VMSTAT_ITEMS.
> 
> This issue arises because some elements of the vmstat_text array are present
> when either MEMCG or VM_EVENT_COUNTERS is enabled, but NR_VMSTAT_ITEMS only
> accounts for these elements if VM_EVENT_COUNTERS is enabled.
> 
> Instead of adjusting the NR_VMSTAT_ITEMS definition to account for MEMCG, make
> MEMCG select VM_EVENT_COUNTERS. VM_EVENT_COUNTERS is enabled in most
> configurations anyway.
> 

Re: [PATCH] mm/vmstat: Fix build with MEMCG=y and VM_EVENT_COUNTERS=n

[Andrew Morton]

On Thu, 5 Jun 2025 14:53:59 +0300 "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> wrote:

> > The changelogs of these two don't reflect the new ordering though, maybe
> > Kirill can provide updated ones?
> 
> Maybe something like this, for the first patch?
> 
> 
> mm/vmstat: Make MEMCG select VM_EVENT_COUNTERS
> 
> The vmstat_text array contains labels for counters displayed in /proc/vmstat.
> It is important to keep the labels in sync with the counters.
>
> ...
>

Updated, thanks.