mm: vm_normal_page*() + CoW PFNMAP improvements

[PATCH RFC 06/14] mm/huge_memory: support huge zero folio in vmf_insert_folio_pmd()

Posted by David Hildenbrand 3 months, 3 weeks ago

Just like we do for vmf_insert_page_mkwrite() -> ... ->
insert_page_into_pte_locked(), support the huge zero folio.

Signed-off-by: David Hildenbrand <david@redhat.com>
---
 mm/huge_memory.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 1ea23900b5adb..92400f3baa9ff 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1418,9 +1418,11 @@ static vm_fault_t insert_pmd(struct vm_area_struct *vma, unsigned long addr,
 	if (fop.is_folio) {
 		entry = folio_mk_pmd(fop.folio, vma->vm_page_prot);
 
-		folio_get(fop.folio);
-		folio_add_file_rmap_pmd(fop.folio, &fop.folio->page, vma);
-		add_mm_counter(mm, mm_counter_file(fop.folio), HPAGE_PMD_NR);
+		if (!is_huge_zero_folio(fop.folio)) {
+			folio_get(fop.folio);
+			folio_add_file_rmap_pmd(fop.folio, &fop.folio->page, vma);
+			add_mm_counter(mm, mm_counter_file(fop.folio), HPAGE_PMD_NR);
+		}
 	} else {
 		entry = pmd_mkhuge(pfn_pmd(fop.pfn, prot));
 		entry = pmd_mkspecial(entry);
-- 
2.49.0

Re: [PATCH RFC 06/14] mm/huge_memory: support huge zero folio in vmf_insert_folio_pmd()

Posted by Oscar Salvador 3 months, 2 weeks ago

On Tue, Jun 17, 2025 at 05:43:37PM +0200, David Hildenbrand wrote:
> Just like we do for vmf_insert_page_mkwrite() -> ... ->
> insert_page_into_pte_locked(), support the huge zero folio.

It might just be me because I don't have the full context cached, but
I might have appreciated more info here :-).

> 
> Signed-off-by: David Hildenbrand <david@redhat.com>

Reviewed-by: Oscar Salvador <osalvador@suse,de>


-- 
Oscar Salvador
SUSE Labs

Re: [PATCH RFC 06/14] mm/huge_memory: support huge zero folio in vmf_insert_folio_pmd()

Posted by David Hildenbrand 3 months, 2 weeks ago

On 25.06.25 10:20, Oscar Salvador wrote:
> On Tue, Jun 17, 2025 at 05:43:37PM +0200, David Hildenbrand wrote:
>> Just like we do for vmf_insert_page_mkwrite() -> ... ->
>> insert_page_into_pte_locked(), support the huge zero folio.
> 
> It might just be me because I don't have the full context cached, but
> I might have appreciated more info here :-).

I can add something similar to what we have in patch #8, stating that we 
will change it to be special as well soon.

Thanks!

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 06/14] mm/huge_memory: support huge zero folio in vmf_insert_folio_pmd()

Posted by Oscar Salvador 3 months, 2 weeks ago

On Tue, Jun 17, 2025 at 05:43:37PM +0200, David Hildenbrand wrote:
> Just like we do for vmf_insert_page_mkwrite() -> ... ->
> insert_page_into_pte_locked(), support the huge zero folio.
> 
> Signed-off-by: David Hildenbrand <david@redhat.com>

insert_page_into_pte_locked() creates a special pte in case it finds the
zero folio while insert_pmd() doesn't.
I know that we didn't want to create special mappings for normal refcount folios
but this seems inconsistent? I'm pretty sure there's a reason but could you
elaborate on that?

> ---
>  mm/huge_memory.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 1ea23900b5adb..92400f3baa9ff 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1418,9 +1418,11 @@ static vm_fault_t insert_pmd(struct vm_area_struct *vma, unsigned long addr,
>  	if (fop.is_folio) {
>  		entry = folio_mk_pmd(fop.folio, vma->vm_page_prot);
>  
> -		folio_get(fop.folio);
> -		folio_add_file_rmap_pmd(fop.folio, &fop.folio->page, vma);
> -		add_mm_counter(mm, mm_counter_file(fop.folio), HPAGE_PMD_NR);
> +		if (!is_huge_zero_folio(fop.folio)) {
> +			folio_get(fop.folio);
> +			folio_add_file_rmap_pmd(fop.folio, &fop.folio->page, vma);
> +			add_mm_counter(mm, mm_counter_file(fop.folio), HPAGE_PMD_NR);
> +		}
>  	} else {
>  		entry = pmd_mkhuge(pfn_pmd(fop.pfn, prot));
>  		entry = pmd_mkspecial(entry);
> -- 
> 2.49.0
> 
> 

-- 
Oscar Salvador
SUSE Labs

Re: [PATCH RFC 06/14] mm/huge_memory: support huge zero folio in vmf_insert_folio_pmd()

Posted by Oscar Salvador 3 months, 2 weeks ago

On Wed, Jun 25, 2025 at 10:15:22AM +0200, Oscar Salvador wrote:
> On Tue, Jun 17, 2025 at 05:43:37PM +0200, David Hildenbrand wrote:
> > Just like we do for vmf_insert_page_mkwrite() -> ... ->
> > insert_page_into_pte_locked(), support the huge zero folio.
> > 
> > Signed-off-by: David Hildenbrand <david@redhat.com>
> 
> insert_page_into_pte_locked() creates a special pte in case it finds the
> zero folio while insert_pmd() doesn't.
> I know that we didn't want to create special mappings for normal refcount folios
> but this seems inconsistent? I'm pretty sure there's a reason but could you
> elaborate on that?

Heh, I should have checked further. I see that happening on patch#8.


-- 
Oscar Salvador
SUSE Labs