[PATCH] mm/shmem: add mTHP swpout fallback statistics in shmem_writeout()

[PATCH] mm/shmem: add mTHP swpout fallback statistics in shmem_writeout()

[Weilin Tong]

Currently, when shmem mTHPs are split and swapped out via shmem_writeout(),
there are no unified statistics to trace these mTHP swpout fallback events.
This makes it difficult to analyze the prevalence of mTHP splitting and
fallback during swap operations, which is important for memory diagnostics.

Here we add statistics counting for mTHP fallback to small pages
when splitting and swapping out in shmem_writeout().

Signed-off-by: Weilin Tong <tongweilin@linux.alibaba.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 mm/shmem.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/mm/shmem.c b/mm/shmem.c
index 3f194c9842a8..aa624c447358 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1593,11 +1593,23 @@ int shmem_writeout(struct folio *folio, struct swap_iocb **plug,
 	}
 
 	if (split) {
+		int order;
+
 try_split:
+		order = folio_order(folio);
 		/* Ensure the subpages are still dirty */
 		folio_test_set_dirty(folio);
 		if (split_folio_to_list(folio, folio_list))
 			goto redirty;
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+		if (order >= HPAGE_PMD_ORDER) {
+			count_memcg_folio_events(folio, THP_SWPOUT_FALLBACK, 1);
+			count_vm_event(THP_SWPOUT_FALLBACK);
+		}
+#endif
+		count_mthp_stat(order, MTHP_STAT_SWPOUT_FALLBACK);
+
 		folio_clear_dirty(folio);
 	}
 
-- 
2.43.7

Re: [PATCH] mm/shmem: add mTHP swpout fallback statistics in shmem_writeout()

[Anshuman Khandual]

On 15/12/25 8:16 AM, Weilin Tong wrote:
> Currently, when shmem mTHPs are split and swapped out via shmem_writeout(),
> there are no unified statistics to trace these mTHP swpout fallback events.
> This makes it difficult to analyze the prevalence of mTHP splitting and
> fallback during swap operations, which is important for memory diagnostics.
> 
> Here we add statistics counting for mTHP fallback to small pages
> when splitting and swapping out in shmem_writeout().
> 
> Signed-off-by: Weilin Tong <tongweilin@linux.alibaba.com>
> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
>  mm/shmem.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 3f194c9842a8..aa624c447358 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1593,11 +1593,23 @@ int shmem_writeout(struct folio *folio, struct swap_iocb **plug,
>  	}
>  
>  	if (split) {
> +		int order;
> +

Could this be moved up near the 'nr_pages' declaration instead ?

>  try_split:
> +		order = folio_order(folio);
>  		/* Ensure the subpages are still dirty */
>  		folio_test_set_dirty(folio);
>  		if (split_folio_to_list(folio, folio_list))
>  			goto redirty;
> +
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +		if (order >= HPAGE_PMD_ORDER) {
> +			count_memcg_folio_events(folio, THP_SWPOUT_FALLBACK, 1);
> +			count_vm_event(THP_SWPOUT_FALLBACK);
> +		}
> +#endif
> +		count_mthp_stat(order, MTHP_STAT_SWPOUT_FALLBACK);
> +
>  		folio_clear_dirty(folio);
>  	}
>  

Otherwise LGTM.

Re: [PATCH] mm/shmem: add mTHP swpout fallback statistics in shmem_writeout()

[Weilin Tong]

[-- Attachment #1: Type: text/plain, Size: 1783 bytes --]


在 2025/12/16 14:34, Anshuman Khandual 写道:
>
> On 15/12/25 8:16 AM, Weilin Tong wrote:
>> Currently, when shmem mTHPs are split and swapped out via shmem_writeout(),
>> there are no unified statistics to trace these mTHP swpout fallback events.
>> This makes it difficult to analyze the prevalence of mTHP splitting and
>> fallback during swap operations, which is important for memory diagnostics.
>>
>> Here we add statistics counting for mTHP fallback to small pages
>> when splitting and swapping out in shmem_writeout().
>>
>> Signed-off-by: Weilin Tong<tongweilin@linux.alibaba.com>
>> Reviewed-by: Baolin Wang<baolin.wang@linux.alibaba.com>
>> ---
>>   mm/shmem.c | 12 ++++++++++++
>>   1 file changed, 12 insertions(+)
>>
>> diff --git a/mm/shmem.c b/mm/shmem.c
>> index 3f194c9842a8..aa624c447358 100644
>> --- a/mm/shmem.c
>> +++ b/mm/shmem.c
>> @@ -1593,11 +1593,23 @@ int shmem_writeout(struct folio *folio, struct swap_iocb **plug,
>>   	}
>>   
>>   	if (split) {
>> +		int order;
>> +
> Could this be moved up near the 'nr_pages' declaration instead ?

Thanks for your reviewing.

IIUC, this|order|variable is only used inside the|"if (split)"|code block, so it's best to define it here. This keeps its scope limited.

>>   try_split:
>> +		order = folio_order(folio);
>>   		/* Ensure the subpages are still dirty */
>>   		folio_test_set_dirty(folio);
>>   		if (split_folio_to_list(folio, folio_list))
>>   			goto redirty;
>> +
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +		if (order >= HPAGE_PMD_ORDER) {
>> +			count_memcg_folio_events(folio, THP_SWPOUT_FALLBACK, 1);
>> +			count_vm_event(THP_SWPOUT_FALLBACK);
>> +		}
>> +#endif
>> +		count_mthp_stat(order, MTHP_STAT_SWPOUT_FALLBACK);
>> +
>>   		folio_clear_dirty(folio);
>>   	}
>>   
> Otherwise LGTM.

[-- Attachment #2: Type: text/html, Size: 6157 bytes --]

Re: [PATCH] mm/shmem: add mTHP swpout fallback statistics in shmem_writeout()

[Andrew Morton]

On Tue, 16 Dec 2025 12:04:02 +0530 Anshuman Khandual <anshuman.khandual@arm.com> wrote:

> > --- a/mm/shmem.c
> > +++ b/mm/shmem.c
> > @@ -1593,11 +1593,23 @@ int shmem_writeout(struct folio *folio, struct swap_iocb **plug,
> >  	}
> >  
> >  	if (split) {
> > +		int order;
> > +
> 
> Could this be moved up near the 'nr_pages' declaration instead ?

It could, but I think it's good practice to restrict the scope of
things to "no more than necessary"?