Skip to contentSkip to conversation

Convert x86/mm/pat to generic page table apis

PATCH RESEND Applied Vishal Moola (Oracle) 2 participants 9 messages Mar 3 – Mar 4

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

v7 cover Convert x86/mm/pat to generic page table apis

Replace the get_free_pages() allocation sites in set_memory.c with the appropriate generic pagetable APIs, since the file already calls pagetable_free() and the mismatch prevents separate allocation of ptdescs from struct page.

Use the generic pagetable APIs consistently for both allocation and freeing to simplify and standardize the code.

Enable Matthew’s work to allocate frozen pagetables in the short term[1], and prepare for a clean split of ptdesc allocations from struct page in the long term[2].

[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/[email protected]/

------

Based on current mm-new. Re-sending because the previous set was split up for some reason.

v7:
- Collect some Acks
- Fix multi-line comment style to match the coding-style
- Reword all messages using imperative voice (used an LLM for this)
- CC all people found to have edited nearby lines (through git blame)

v6 link:
https://lore.kernel.org/linux-mm/[email protected]/

Vishal Moola (Oracle) (4):
x86/mm/pat: Convert pte code to use page table apis x86/mm/pat: Convert pmd code to use page table apis x86/mm/pat: Convert populate_pgd() to use page table apis x86/mm/pat: Convert split_large_page() to use ptdescs

 arch/x86/mm/pat/set_memory.c | 37 ++++++++++++++++++++++++------------
 1 file changed, 25 insertions(+), 12 deletions(-)
Signature · 1 line
2.53.0
#1 lore raw

PATCH RESEND v7 1/4 x86/mm/pat: Convert pte code to use page table apis

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future. Convert the PTE allocation and free sites to use the generic page table APIs, as they already use ptdescs.

Pass through init_mm since these are kernel page tables; otherwise, pte_alloc_one_kernel() becomes a no-op.

---
 arch/x86/mm/pat/set_memory.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/mm/pat/set_memory.c +2 -2
index 40581a720fe8..a4b1b329c23d 100644
@@ -1408,7 +1408,7 @@ static bool try_to_free_pte_page(pte_t *pte)
 		if (!pte_none(pte[i]))
 			return false;
 
-	free_page((unsigned long)pte);
+	pte_free_kernel(&init_mm, pte);
 	return true;
 }
 
@@ -1539,7 +1539,7 @@ static void unmap_pud_range(p4d_t *p4d, unsigned long start, unsigned long end)
 
 static int alloc_pte_page(pmd_t *pmd)
 {
-	pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
+	pte_t *pte = pte_alloc_one_kernel(&init_mm);
 	if (!pte)
 		return -1;
 
Signature · 1 line
2.53.0
#2 lore raw

[tip: x86/mm] x86/mm/pat: Convert pte code to use page table apis

The following commit has been merged into the x86/mm branch of tip:

The following commit has been merged into the x86/mm branch of tip:

Commit-ID:     c22ca4a96300c8ed01fe2282d2dd9e9a75032379
Gitweb:        https://git.kernel.org/tip/c22ca4a96300c8ed01fe2282d2dd9e9a75032379
Author:        Vishal Moola (Oracle) <[email protected]>
AuthorDate:    Tue, 03 Mar 2026 11:48:24 -08:00
Committer:     Dave Hansen <[email protected]>
CommitterDate: Wed, 04 Mar 2026 10:08:54 -08:00

x86/mm/pat: Convert pte code to use page table apis

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future. Convert the PTE allocation and free sites to use the generic page table APIs, as they already use ptdescs.

Pass through init_mm since these are kernel page tables; otherwise, pte_alloc_one_kernel() becomes a no-op.

---
 arch/x86/mm/pat/set_memory.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/mm/pat/set_memory.c +2 -2
index 40581a7..a4b1b32 100644
@@ -1408,7 +1408,7 @@ static bool try_to_free_pte_page(pte_t *pte)
 		if (!pte_none(pte[i]))
 			return false;
 
-	free_page((unsigned long)pte);
+	pte_free_kernel(&init_mm, pte);
 	return true;
 }
 
@@ -1539,7 +1539,7 @@ static void unmap_pud_range(p4d_t *p4d, unsigned long start, unsigned long end)
 
 static int alloc_pte_page(pmd_t *pmd)
 {
-	pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
+	pte_t *pte = pte_alloc_one_kernel(&init_mm);
 	if (!pte)
 		return -1;
 
#3 lore raw

PATCH RESEND v7 2/4 x86/mm/pat: Convert pmd code to use page table apis

In reply to Vishal Moola (Oracle)

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future. Convert the PMD allocation and free sites to use the generic page table APIs, as they already use ptdescs.

Pass through init_mm since these are kernel page tables, as pmd_alloc_one() requires it to identify kernel page tables. Because the generic implementation does not use the second argument, pass a placeholder to avoid reimplementing it or risking breakage on other architectures.

---
 arch/x86/mm/pat/set_memory.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/mm/pat/set_memory.c +6 -2
index a4b1b329c23d..72a260007996 100644
@@ -1420,7 +1420,7 @@ static bool try_to_free_pmd_page(pmd_t *pmd)
 		if (!pmd_none(pmd[i]))
 			return false;
 
-	free_page((unsigned long)pmd);
+	pmd_free(&init_mm, pmd);
 	return true;
 }
 
@@ -1549,7 +1549,11 @@ static int alloc_pte_page(pmd_t *pmd)
 
 static int alloc_pmd_page(pud_t *pud)
 {
-	pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
+	/*
+	 * Pass 0 as a placeholder for the second argument, since the
+	 * generic implementation of pmd_alloc_one() does not use it.
+	 */
+	pmd_t *pmd = pmd_alloc_one(&init_mm, 0);
 	if (!pmd)
 		return -1;
 
Signature · 1 line
2.53.0
#4 lore raw

[tip: x86/mm] x86/mm/pat: Convert pmd code to use page table apis

The following commit has been merged into the x86/mm branch of tip:

The following commit has been merged into the x86/mm branch of tip:

Commit-ID:     6ba1917a4c5c59415c8fc3f83019d92e3f81c87c
Gitweb:        https://git.kernel.org/tip/6ba1917a4c5c59415c8fc3f83019d92e3f81c87c
Author:        Vishal Moola (Oracle) <[email protected]>
AuthorDate:    Tue, 03 Mar 2026 11:48:25 -08:00
Committer:     Dave Hansen <[email protected]>
CommitterDate: Wed, 04 Mar 2026 10:08:54 -08:00

x86/mm/pat: Convert pmd code to use page table apis

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future. Convert the PMD allocation and free sites to use the generic page table APIs, as they already use ptdescs.

Pass through init_mm since these are kernel page tables, as pmd_alloc_one() requires it to identify kernel page tables. Because the generic implementation does not use the second argument, pass a placeholder to avoid reimplementing it or risking breakage on other architectures.

---
 arch/x86/mm/pat/set_memory.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/mm/pat/set_memory.c +6 -2
index a4b1b32..72a2600 100644
@@ -1420,7 +1420,7 @@ static bool try_to_free_pmd_page(pmd_t *pmd)
 		if (!pmd_none(pmd[i]))
 			return false;
 
-	free_page((unsigned long)pmd);
+	pmd_free(&init_mm, pmd);
 	return true;
 }
 
@@ -1549,7 +1549,11 @@ static int alloc_pte_page(pmd_t *pmd)
 
 static int alloc_pmd_page(pud_t *pud)
 {
-	pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
+	/*
+	 * Pass 0 as a placeholder for the second argument, since the
+	 * generic implementation of pmd_alloc_one() does not use it.
+	 */
+	pmd_t *pmd = pmd_alloc_one(&init_mm, 0);
 	if (!pmd)
 		return -1;
 
#5 lore raw

PATCH RESEND v7 3/4 x86/mm/pat: Convert populate_pgd() to use page table apis

In reply to Vishal Moola (Oracle)

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future. Convert the remaining get_zeroed_page() calls to the generic page table APIs, as they already use ptdescs.

Pass through init_mm since these are kernel page tables, as both functions require it to identify kernel page tables. Because the generic implementations do not use the second argument, pass a placeholder to avoid reimplementing them or risking breakage on other architectures.

It is not obvious whether these pages are freed. Regardless, convert the remaining free paths as needed, noting that the only other possible free paths have already been converted and that a frozen page table test kernel has not reported any issues.

---
 arch/x86/mm/pat/set_memory.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
arch/x86/mm/pat/set_memory.c +10 -2
index 72a260007996..17c1c28e3491 100644
@@ -1747,7 +1747,11 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
 	pgd_entry = cpa->pgd + pgd_index(addr);
 
 	if (pgd_none(*pgd_entry)) {
-		p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
+		/*
+		 * Pass 0 as a placeholder for the second argument, since the
+		 * generic implementation of p4d_alloc_one() does not use it.
+		 */
+		p4d = p4d_alloc_one(&init_mm, 0);
 		if (!p4d)
 			return -1;
 
@@ -1759,7 +1763,11 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
 	 */
 	p4d = p4d_offset(pgd_entry, addr);
 	if (p4d_none(*p4d)) {
-		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
+		/*
+		 * Pass 0 as a placeholder for the second argument, since the
+		 * generic implementation of pud_alloc_one() does not use it.
+		 */
+		pud = pud_alloc_one(&init_mm, 0);
 		if (!pud)
 			return -1;
 
Signature · 1 line
2.53.0
#6 lore raw

[tip: x86/mm] x86/mm/pat: Convert populate_pgd() to use page table apis

The following commit has been merged into the x86/mm branch of tip:

The following commit has been merged into the x86/mm branch of tip:

Commit-ID:     b2203a9bf53237368a7c7fc976c9616b5562af8e
Gitweb:        https://git.kernel.org/tip/b2203a9bf53237368a7c7fc976c9616b5562af8e
Author:        Vishal Moola (Oracle) <[email protected]>
AuthorDate:    Tue, 03 Mar 2026 11:48:26 -08:00
Committer:     Dave Hansen <[email protected]>
CommitterDate: Wed, 04 Mar 2026 10:08:54 -08:00

x86/mm/pat: Convert populate_pgd() to use page table apis

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future. Convert the remaining get_zeroed_page() calls to the generic page table APIs, as they already use ptdescs.

Pass through init_mm since these are kernel page tables, as both functions require it to identify kernel page tables. Because the generic implementations do not use the second argument, pass a placeholder to avoid reimplementing them or risking breakage on other architectures.

It is not obvious whether these pages are freed. Regardless, convert the remaining free paths as needed, noting that the only other possible free paths have already been converted and that a frozen page table test kernel has not reported any issues.

---
 arch/x86/mm/pat/set_memory.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
arch/x86/mm/pat/set_memory.c +10 -2
index 72a2600..17c1c28 100644
@@ -1747,7 +1747,11 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
 	pgd_entry = cpa->pgd + pgd_index(addr);
 
 	if (pgd_none(*pgd_entry)) {
-		p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
+		/*
+		 * Pass 0 as a placeholder for the second argument, since the
+		 * generic implementation of p4d_alloc_one() does not use it.
+		 */
+		p4d = p4d_alloc_one(&init_mm, 0);
 		if (!p4d)
 			return -1;
 
@@ -1759,7 +1763,11 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
 	 */
 	p4d = p4d_offset(pgd_entry, addr);
 	if (p4d_none(*p4d)) {
-		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
+		/*
+		 * Pass 0 as a placeholder for the second argument, since the
+		 * generic implementation of pud_alloc_one() does not use it.
+		 */
+		pud = pud_alloc_one(&init_mm, 0);
 		if (!pud)
 			return -1;
 
#7 lore raw

PATCH RESEND v7 4/4 x86/mm/pat: Convert split_large_page() to use ptdescs

In reply to Vishal Moola (Oracle)

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future.

Update split_large_page() to allocate a ptdesc instead of allocating a page for use as a page table.

---
 arch/x86/mm/pat/set_memory.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
arch/x86/mm/pat/set_memory.c +7 -6
index 17c1c28e3491..cba907c39718 100644
@@ -1119,9 +1119,10 @@ static void split_set_pte(struct cpa_data *cpa, pte_t *pte, unsigned long pfn,
 
 static int
 __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
-		   struct page *base)
+		   struct ptdesc *ptdesc)
 {
 	unsigned long lpaddr, lpinc, ref_pfn, pfn, pfninc = 1;
+	struct page *base = ptdesc_page(ptdesc);
 	pte_t *pbase = (pte_t *)page_address(base);
 	unsigned int i, level;
 	pgprot_t ref_prot;
@@ -1226,18 +1227,18 @@ __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
 static int split_large_page(struct cpa_data *cpa, pte_t *kpte,
 			    unsigned long address)
 {
-	struct page *base;
+	struct ptdesc *ptdesc;
 
 	if (!debug_pagealloc_enabled())
 		spin_unlock(&cpa_lock);
-	base = alloc_pages(GFP_KERNEL, 0);
+	ptdesc = pagetable_alloc(GFP_KERNEL, 0);
 	if (!debug_pagealloc_enabled())
 		spin_lock(&cpa_lock);
-	if (!base)
+	if (!ptdesc)
 		return -ENOMEM;
 
-	if (__split_large_page(cpa, kpte, address, base))
-		__free_page(base);
+	if (__split_large_page(cpa, kpte, address, ptdesc))
+		pagetable_free(ptdesc);
 
 	return 0;
 }
Signature · 1 line
2.53.0
#8 lore raw

[tip: x86/mm] x86/mm/pat: Convert split_large_page() to use ptdescs

The following commit has been merged into the x86/mm branch of tip:

The following commit has been merged into the x86/mm branch of tip:

Commit-ID:     e751303e0ad2e998f421d104193f6904df3516d1
Gitweb:        https://git.kernel.org/tip/e751303e0ad2e998f421d104193f6904df3516d1
Author:        Vishal Moola (Oracle) <[email protected]>
AuthorDate:    Tue, 03 Mar 2026 11:48:27 -08:00
Committer:     Dave Hansen <[email protected]>
CommitterDate: Wed, 04 Mar 2026 10:08:54 -08:00

x86/mm/pat: Convert split_large_page() to use ptdescs

Use the ptdesc APIs for all page table allocation and free sites to allow their separate allocation from struct page in the future.

Update split_large_page() to allocate a ptdesc instead of allocating a page for use as a page table.

---
 arch/x86/mm/pat/set_memory.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
arch/x86/mm/pat/set_memory.c +7 -6
index 17c1c28..cba907c 100644
@@ -1119,9 +1119,10 @@ set:
 
 static int
 __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
-		   struct page *base)
+		   struct ptdesc *ptdesc)
 {
 	unsigned long lpaddr, lpinc, ref_pfn, pfn, pfninc = 1;
+	struct page *base = ptdesc_page(ptdesc);
 	pte_t *pbase = (pte_t *)page_address(base);
 	unsigned int i, level;
 	pgprot_t ref_prot;
@@ -1226,18 +1227,18 @@ __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
 static int split_large_page(struct cpa_data *cpa, pte_t *kpte,
 			    unsigned long address)
 {
-	struct page *base;
+	struct ptdesc *ptdesc;
 
 	if (!debug_pagealloc_enabled())
 		spin_unlock(&cpa_lock);
-	base = alloc_pages(GFP_KERNEL, 0);
+	ptdesc = pagetable_alloc(GFP_KERNEL, 0);
 	if (!debug_pagealloc_enabled())
 		spin_lock(&cpa_lock);
-	if (!base)
+	if (!ptdesc)
 		return -ENOMEM;
 
-	if (__split_large_page(cpa, kpte, address, base))
-		__free_page(base);
+	if (__split_large_page(cpa, kpte, address, ptdesc))
+		pagetable_free(ptdesc);
 
 	return 0;
 }