[PATCH] mm/huge_memory: add pmd folio to ds_queue in do_huge_zero_wp_pmd()

[PATCH] mm/huge_memory: add pmd folio to ds_queue in do_huge_zero_wp_pmd()

[Wei Yang]

We add pmd folio into ds_queue on the first page fault in
__do_huge_pmd_anonymous_page(), so that we can split it in case of
memory pressure. This should be the same for a pmd folio during wp
page fault.

Add it as expected.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Dev Jain <dev.jain@arm.com>
---
 mm/huge_memory.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 1b81680b4225..963cba933cb6 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1878,6 +1878,7 @@ static vm_fault_t do_huge_zero_wp_pmd(struct vm_fault *vmf)
 		goto release;
 	(void)pmdp_huge_clear_flush(vma, haddr, vmf->pmd);
 	map_anon_folio_pmd(folio, vmf->pmd, vma, haddr);
+	deferred_split_folio(folio, false);
 	goto unlock;
 release:
 	folio_put(folio);
-- 
2.34.1

Re: [PATCH] mm/huge_memory: add pmd folio to ds_queue in do_huge_zero_wp_pmd()

[David Hildenbrand]

On 01.10.25 16:28, Wei Yang wrote:
> We add pmd folio into ds_queue on the first page fault in
> __do_huge_pmd_anonymous_page(), so that we can split it in case of
> memory pressure. This should be the same for a pmd folio during wp
> page fault.
> 
> Add it as expected.
> 
> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Lance Yang <lance.yang@linux.dev>
> Cc: Dev Jain <dev.jain@arm.com>

Fixes: please, and I think we want CC stable.

> ---
>   mm/huge_memory.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 1b81680b4225..963cba933cb6 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1878,6 +1878,7 @@ static vm_fault_t do_huge_zero_wp_pmd(struct vm_fault *vmf)
>   		goto release;
>   	(void)pmdp_huge_clear_flush(vma, haddr, vmf->pmd);
>   	map_anon_folio_pmd(folio, vmf->pmd, vma, haddr);
> +	deferred_split_folio(folio, false);
>   	goto unlock;
>   release:
>   	folio_put(folio);

Which raises the question whether the deferred_split_folio() should 
actually go into the map_anon_folio_pmd(). I think you should just move 
the one from __do_huge_pmd_anonymous_page() in there.

-- 
Cheers

David / dhildenb

Re: [PATCH] mm/huge_memory: add pmd folio to ds_queue in do_huge_zero_wp_pmd()

[Lance Yang]

On 2025/10/1 22:34, David Hildenbrand wrote:
> On 01.10.25 16:28, Wei Yang wrote:
>> We add pmd folio into ds_queue on the first page fault in
>> __do_huge_pmd_anonymous_page(), so that we can split it in case of
>> memory pressure. This should be the same for a pmd folio during wp
>> page fault.
>>
>> Add it as expected.

Good catch!

>>
>> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
>> Cc: David Hildenbrand <david@redhat.com>
>> Cc: Lance Yang <lance.yang@linux.dev>
>> Cc: Dev Jain <dev.jain@arm.com>
> 
> Fixes: please, and I think we want CC stable.
> 
>> ---
>>   mm/huge_memory.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 1b81680b4225..963cba933cb6 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -1878,6 +1878,7 @@ static vm_fault_t do_huge_zero_wp_pmd(struct 
>> vm_fault *vmf)
>>           goto release;
>>       (void)pmdp_huge_clear_flush(vma, haddr, vmf->pmd);
>>       map_anon_folio_pmd(folio, vmf->pmd, vma, haddr);
>> +    deferred_split_folio(folio, false);
>>       goto unlock;
>>   release:
>>       folio_put(folio);
> 
> Which raises the question whether the deferred_split_folio() should 
> actually go into the map_anon_folio_pmd(). I think you should just move 
> the one from __do_huge_pmd_anonymous_page() in there.

+1 Moving the call into map_anon_folio_pmd() feels way cleaner :)

Cheers,
Lance

Re: [PATCH] mm/huge_memory: add pmd folio to ds_queue in do_huge_zero_wp_pmd()

[Andrew Morton]

On Wed, 1 Oct 2025 16:34:45 +0200 David Hildenbrand <david@redhat.com> wrote:

> On 01.10.25 16:28, Wei Yang wrote:
> > We add pmd folio into ds_queue on the first page fault in
> > __do_huge_pmd_anonymous_page(), so that we can split it in case of
> > memory pressure. This should be the same for a pmd folio during wp
> > page fault.
> > 
> > Add it as expected.
> > 
> > Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> > Cc: David Hildenbrand <david@redhat.com>
> > Cc: Lance Yang <lance.yang@linux.dev>
> > Cc: Dev Jain <dev.jain@arm.com>
> 
> Fixes: please, and I think we want CC stable.

And a reason for Fixes: and Cc:stable, please.  ie, a description of
the userspace visible runtime effects.

Cheers, akpmbot.