[PATCH v2 1/2] mm: khugepaged: convert set_huge_pmd() to take a folio

[PATCH v2 1/2] mm: khugepaged: convert set_huge_pmd() to take a folio

[Baolin Wang]

We've already gotten the stable locked folio in collapse_pte_mapped_thp(),
so just use folio for set_huge_pmd() to set the PMD entry, which is more
straightforward.

Moreover, we will check the folio size in do_set_pmd(), so we can remove
the unnecessary VM_BUG_ON() in set_huge_pmd(). While we are at it, we can
also remove the PageTransHuge(), as it currently has no callers.

Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
Changes from v1:
 - Remove the unnecessary VM_BUG_ON().
 - Remove the PageTransHuge().
---
 include/linux/page-flags.h | 15 ---------------
 mm/khugepaged.c            |  9 ++++-----
 2 files changed, 4 insertions(+), 20 deletions(-)

diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index 37b11f15dbd9..1c1d49554c71 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -907,20 +907,6 @@ FOLIO_FLAG_FALSE(partially_mapped)
 #define PG_head_mask ((1UL << PG_head))
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-/*
- * PageHuge() only returns true for hugetlbfs pages, but not for
- * normal or transparent huge pages.
- *
- * PageTransHuge() returns true for both transparent huge and
- * hugetlbfs pages, but not normal pages. PageTransHuge() can only be
- * called only in the core VM paths where hugetlbfs pages can't exist.
- */
-static inline int PageTransHuge(const struct page *page)
-{
-	VM_BUG_ON_PAGE(PageTail(page), page);
-	return PageHead(page);
-}
-
 /*
  * PageTransCompound returns true for both transparent huge pages
  * and hugetlbfs pages, so it should only be called when it's known
@@ -931,7 +917,6 @@ static inline int PageTransCompound(const struct page *page)
 	return PageCompound(page);
 }
 #else
-TESTPAGEFLAG_FALSE(TransHuge, transhuge)
 TESTPAGEFLAG_FALSE(TransCompound, transcompound)
 #endif
 
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index b04b6a770afe..aca66e7f4fd9 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1467,7 +1467,7 @@ static void collect_mm_slot(struct khugepaged_mm_slot *mm_slot)
 #ifdef CONFIG_SHMEM
 /* hpage must be locked, and mmap_lock must be held */
 static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
-			pmd_t *pmdp, struct page *hpage)
+			pmd_t *pmdp, struct folio *folio)
 {
 	struct vm_fault vmf = {
 		.vma = vma,
@@ -1476,13 +1476,12 @@ static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
 		.pmd = pmdp,
 	};
 
-	VM_BUG_ON(!PageTransHuge(hpage));
 	mmap_assert_locked(vma->vm_mm);
 
-	if (do_set_pmd(&vmf, hpage))
+	if (do_set_pmd(&vmf, &folio->page))
 		return SCAN_FAIL;
 
-	get_page(hpage);
+	folio_get(folio);
 	return SCAN_SUCCEED;
 }
 
@@ -1689,7 +1688,7 @@ int collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr,
 maybe_install_pmd:
 	/* step 5: install pmd entry */
 	result = install_pmd
-			? set_huge_pmd(vma, haddr, pmd, &folio->page)
+			? set_huge_pmd(vma, haddr, pmd, folio)
 			: SCAN_SUCCEED;
 	goto drop_folio;
 abort:
-- 
2.43.5

[PATCH v2 2/2] mm: convert do_set_pmd() to take a folio

[Baolin Wang]

In do_set_pmd(), we always use the folio->page to build PMD mappings for
the entire folio. Since all callers of do_set_pmd() already hold a stable
folio, converting do_set_pmd() to take a folio is safe and more straightforward.

In addition, to ensure the extensibility of do_set_pmd() for supporting
larger folios beyond PMD size, we keep the 'page' parameter to specify
which page within the folio should be mapped.

No functional changes expected.

Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
Changes from v1:
 - Keep the 'page' parameter of the do_set_pmd().

Note: I did mm selftests and built kernel on tmpfs/xfs filesystems, and
did not find any regression.
---
 include/linux/mm.h |  2 +-
 mm/filemap.c       |  2 +-
 mm/khugepaged.c    |  2 +-
 mm/memory.c        | 11 +++++------
 4 files changed, 8 insertions(+), 9 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 43748c8f3454..d5f578c91e77 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1237,7 +1237,7 @@ static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
 	return pte;
 }
 
-vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page);
+vm_fault_t do_set_pmd(struct vm_fault *vmf, struct folio *folio, struct page *page);
 void set_pte_range(struct vm_fault *vmf, struct folio *folio,
 		struct page *page, unsigned int nr, unsigned long addr);
 
diff --git a/mm/filemap.c b/mm/filemap.c
index 7b90cbeb4a1a..09d005848f0d 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -3533,7 +3533,7 @@ static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
 
 	if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
 		struct page *page = folio_file_page(folio, start);
-		vm_fault_t ret = do_set_pmd(vmf, page);
+		vm_fault_t ret = do_set_pmd(vmf, folio, page);
 		if (!ret) {
 			/* The page is mapped successfully, reference consumed. */
 			folio_unlock(folio);
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index aca66e7f4fd9..c4b5031f3254 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1478,7 +1478,7 @@ static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
 
 	mmap_assert_locked(vma->vm_mm);
 
-	if (do_set_pmd(&vmf, &folio->page))
+	if (do_set_pmd(&vmf, folio, &folio->page))
 		return SCAN_FAIL;
 
 	folio_get(folio);
diff --git a/mm/memory.c b/mm/memory.c
index 68c1d962d0ad..9c202c32ca66 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -5176,9 +5176,8 @@ static void deposit_prealloc_pte(struct vm_fault *vmf)
 	vmf->prealloc_pte = NULL;
 }
 
-vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
+vm_fault_t do_set_pmd(struct vm_fault *vmf, struct folio *folio, struct page *page)
 {
-	struct folio *folio = page_folio(page);
 	struct vm_area_struct *vma = vmf->vma;
 	bool write = vmf->flags & FAULT_FLAG_WRITE;
 	unsigned long haddr = vmf->address & HPAGE_PMD_MASK;
@@ -5251,7 +5250,7 @@ vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
 	return ret;
 }
 #else
-vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
+vm_fault_t do_set_pmd(struct vm_fault *vmf, struct folio *folio, struct page *page)
 {
 	return VM_FAULT_FALLBACK;
 }
@@ -5345,6 +5344,7 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 	else
 		page = vmf->page;
 
+	folio = page_folio(page);
 	/*
 	 * check even for read faults because we might have lost our CoWed
 	 * page
@@ -5356,8 +5356,8 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 	}
 
 	if (pmd_none(*vmf->pmd)) {
-		if (PageTransCompound(page)) {
-			ret = do_set_pmd(vmf, page);
+		if (folio_test_pmd_mappable(folio)) {
+			ret = do_set_pmd(vmf, folio, page);
 			if (ret != VM_FAULT_FALLBACK)
 				return ret;
 		}
@@ -5368,7 +5368,6 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 			return VM_FAULT_OOM;
 	}
 
-	folio = page_folio(page);
 	nr_pages = folio_nr_pages(folio);
 
 	/*
-- 
2.43.5

Re: [PATCH v2 2/2] mm: convert do_set_pmd() to take a folio

[Zi Yan]

On 9 May 2025, at 9:01, Baolin Wang wrote:

> In do_set_pmd(), we always use the folio->page to build PMD mappings for
> the entire folio. Since all callers of do_set_pmd() already hold a stable
> folio, converting do_set_pmd() to take a folio is safe and more straightforward.
>
> In addition, to ensure the extensibility of do_set_pmd() for supporting
> larger folios beyond PMD size, we keep the 'page' parameter to specify
> which page within the folio should be mapped.
>
> No functional changes expected.
>
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
> Changes from v1:
>  - Keep the 'page' parameter of the do_set_pmd().
>
> Note: I did mm selftests and built kernel on tmpfs/xfs filesystems, and
> did not find any regression.
> ---
>  include/linux/mm.h |  2 +-
>  mm/filemap.c       |  2 +-
>  mm/khugepaged.c    |  2 +-
>  mm/memory.c        | 11 +++++------
>  4 files changed, 8 insertions(+), 9 deletions(-)
>
LGTM. Reviewed-by: Zi Yan <ziy@nvidia.com>

--
Best Regards,
Yan, Zi

Re: [PATCH v2 1/2] mm: khugepaged: convert set_huge_pmd() to take a folio

[David Hildenbrand]

> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index b04b6a770afe..aca66e7f4fd9 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -1467,7 +1467,7 @@ static void collect_mm_slot(struct khugepaged_mm_slot *mm_slot)
>   #ifdef CONFIG_SHMEM
>   /* hpage must be locked, and mmap_lock must be held */

^ that comment probably needs some love.

>   static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
> -			pmd_t *pmdp, struct page *hpage)
> +			pmd_t *pmdp, struct folio *folio)
>   {
>   	struct vm_fault vmf = {
>   		.vma = vma,
> @@ -1476,13 +1476,12 @@ static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
>   		.pmd = pmdp,
>   	};
>   
> -	VM_BUG_ON(!PageTransHuge(hpage));
>   	mmap_assert_locked(vma->vm_mm);
>   
> -	if (do_set_pmd(&vmf, hpage))
> +	if (do_set_pmd(&vmf, &folio->page))
>   		return SCAN_FAIL;
>   
> -	get_page(hpage);
> +	folio_get(folio);
>   	return SCAN_SUCCEED;
>   }
>   
> @@ -1689,7 +1688,7 @@ int collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr,
>   maybe_install_pmd:
>   	/* step 5: install pmd entry */
>   	result = install_pmd
> -			? set_huge_pmd(vma, haddr, pmd, &folio->page)
> +			? set_huge_pmd(vma, haddr, pmd, folio)

Wondering why we are not passing in the folio+page pair in here as well. 
I assume in the foreseeable future this code will not be able to work 
with folios large than PMDs?

Apart from that LGTM.

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

-- 
Cheers,

David / dhildenb

Re: [PATCH v2 1/2] mm: khugepaged: convert set_huge_pmd() to take a folio

[Baolin Wang]

On 2025/5/9 22:58, David Hildenbrand wrote:
>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>> index b04b6a770afe..aca66e7f4fd9 100644
>> --- a/mm/khugepaged.c
>> +++ b/mm/khugepaged.c
>> @@ -1467,7 +1467,7 @@ static void collect_mm_slot(struct 
>> khugepaged_mm_slot *mm_slot)
>>   #ifdef CONFIG_SHMEM
>>   /* hpage must be locked, and mmap_lock must be held */
> 
> ^ that comment probably needs some love.

Ah, missed that. Will update the comments in next version.

> 
>>   static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
>> -            pmd_t *pmdp, struct page *hpage)
>> +            pmd_t *pmdp, struct folio *folio)
>>   {
>>       struct vm_fault vmf = {
>>           .vma = vma,
>> @@ -1476,13 +1476,12 @@ static int set_huge_pmd(struct vm_area_struct 
>> *vma, unsigned long addr,
>>           .pmd = pmdp,
>>       };
>> -    VM_BUG_ON(!PageTransHuge(hpage));
>>       mmap_assert_locked(vma->vm_mm);
>> -    if (do_set_pmd(&vmf, hpage))
>> +    if (do_set_pmd(&vmf, &folio->page))
>>           return SCAN_FAIL;
>> -    get_page(hpage);
>> +    folio_get(folio);
>>       return SCAN_SUCCEED;
>>   }
>> @@ -1689,7 +1688,7 @@ int collapse_pte_mapped_thp(struct mm_struct 
>> *mm, unsigned long addr,
>>   maybe_install_pmd:
>>       /* step 5: install pmd entry */
>>       result = install_pmd
>> -            ? set_huge_pmd(vma, haddr, pmd, &folio->page)
>> +            ? set_huge_pmd(vma, haddr, pmd, folio)
> 
> Wondering why we are not passing in the folio+page pair in here as well. 
> I assume in the foreseeable future this code will not be able to work 
> with folios large than PMDs?

OK. Will do in next version.

> Apart from that LGTM.
> 
> Acked-by: David Hildenbrand <david@redhat.com>

Thanks.