mm/memory: use folio lock/ref helpers in do_swap_page

PATCH In review Hongfu Li 4 participants 6 messages Jul 21 – Jul 22

Subsystems (2)

Memory management MM core

lore

Share & tools

lore

Hongfu Li

author

2026-07-21 01:34 UTC #0 lore raw

PATCH 1/1 mm/memory: use folio lock/ref helpers in do_swap_page

From: Hongfu Li <[email protected]>

Replace page-level lock/ref operations with their folio equivalents in the device_private migration path of do_swap_page().

Introduce a local fault_folio to hold page_folio(vmf->page) to avoid repeated page_folio() invocations.

Signed-off-by Hongfu Li <[email protected]>

---
 mm/memory.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

mm/memory.c +7 -4

index d5e87624f692..bd27372a1052 100644
@@ -4852,6 +4852,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			vmf->page = softleaf_to_page(entry);
 			ret = remove_device_exclusive_entry(vmf);
 		} else if (softleaf_is_device_private(entry)) {
+			struct folio *fault_folio;
+
 			if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
 				/*
 				 * migrate_to_ram is not yet ready to operate

@@ -4863,6 +4865,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			}
 
 			vmf->page = softleaf_to_page(entry);
+			fault_folio = page_folio(vmf->page);
 			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
 					vmf->address, &vmf->ptl);
 			if (unlikely(!vmf->pte ||

@@ -4874,15 +4877,15 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			 * Get a page reference while we know the page can't be
 			 * freed.
 			 */
-			if (trylock_page(vmf->page)) {
+			if (folio_trylock(fault_folio)) {
 				struct dev_pagemap *pgmap;
 
-				get_page(vmf->page);
+				folio_get(fault_folio);
 				pte_unmap_unlock(vmf->pte, vmf->ptl);
 				pgmap = page_pgmap(vmf->page);
 				ret = pgmap->ops->migrate_to_ram(vmf);
-				unlock_page(vmf->page);
-				put_page(vmf->page);
+				folio_unlock(fault_folio);
+				folio_put(fault_folio);
 			} else {
 				pte_unmap(vmf->pte);
 				softleaf_entry_wait_on_locked(entry, vmf->ptl);

Signature · 1 line

2.54.0

David Hildenbrand (Arm)

maintainer Acked-by on v1 1/1

2026-07-21 10:21 UTC #1 lore raw

Quoted 10 lines · Hongfu Li

From: Hongfu Li <[email protected]>
Replace page-level lock/ref operations with their folio equivalents in the device_private migration path of do_swap_page().
Introduce a local fault_folio to hold page_folio(vmf->page) to avoid repeated page_folio() invocations.
Signed-off-by Hongfu Li <[email protected]>
---

Acked-by David Hildenbrand (Arm) <[email protected]>

Signature · 3 lines

Cheers,

David

Andrew Morton

maintainer

2026-07-22 01:02 UTC #2 lore raw

In reply to Hongfu Li

Quoted 22 lines · On Tue, 21 Jul 2026 09:33:47 +0800 Hongfu Li <[email protected]> wrote:

From: Hongfu Li <[email protected]>
Replace page-level lock/ref operations with their folio equivalents in the device_private migration path of do_swap_page().
Introduce a local fault_folio to hold page_folio(vmf->page) to avoid repeated page_folio() invocations.
...
mm/memory.c +2 -0
@@ -4852,6 +4852,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			vmf->page = softleaf_to_page(entry);
 			ret = remove_device_exclusive_entry(vmf);
 		} else if (softleaf_is_device_private(entry)) {
+			struct folio *fault_folio;
+
 			if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
 				/*
 				 * migrate_to_ram is not yet ready to operate

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 4870

Quoted 6 lines

mm/memory.c +1 -0

@@ -4863,6 +4865,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			}
 
 			vmf->page = softleaf_to_page(entry);
+			fault_folio = page_folio(vmf->page);
 			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
 					vmf->address, &vmf->ptl);

AI review thinks we shouldn't be reading vmf->page until after taking the page table lock:

	https://sashiko.dev/#/patchset/[email protected]

Hongfu Li

author

2026-07-22 06:23 UTC #3 lore raw

Quoted 22 lines

Quoted 12 lines

mm/memory.c +2 -0

@@ -4852,6 +4852,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			vmf->page = softleaf_to_page(entry);
 			ret = remove_device_exclusive_entry(vmf);
 		} else if (softleaf_is_device_private(entry)) {
+			struct folio *fault_folio;
+
 			if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
 				/*
 				 * migrate_to_ram is not yet ready to operate

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 4870

Quoted 6 lines

mm/memory.c +1 -0

@@ -4863,6 +4865,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			}
 
 			vmf->page = softleaf_to_page(entry);
+			fault_folio = page_folio(vmf->page);
 			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
 					vmf->address, &vmf->ptl);

AI review thinks we shouldn't be reading vmf->page until after taking the page table lock:

Quoted 1 line

	https://sashiko.dev/#/patchset/[email protected]

Thanks for the review.

The AI review makes a valid observation: vmf->page should not be accessed before holding the page table lock.

I've moved fault_folio = page_folio(vmf->page) after pte_offset_map_lock() and the PTE validation check, so vmf->page is now read under the page table lock. This fix will be included in v2.

Best regards,
Hongfu

Matthew Wilcox

2026-07-22 01:25 UTC #4 lore raw

In reply to Hongfu Li

Quoted 4 lines · Hongfu Li

From: Hongfu Li <[email protected]>
Replace page-level lock/ref operations with their folio equivalents in the device_private migration path of do_swap_page().

When I do folio conversions, I like to mention the actual advantage of the patch. Here, I'd say something like ...

mm: Use a folio in the softleaf_is_device_private path

Use the folio APIs, replacing four calls to compound_head() with one.

Quoted 13 lines

Introduce a local fault_folio to hold page_folio(vmf->page) to avoid repeated page_folio() invocations.
Signed-off-by Hongfu Li <[email protected]>
---
 mm/memory.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
mm/memory.c +0 -0
index d5e87624f692..bd27372a1052 100644

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 4855

Quoted 4 lines

mm/memory.c +1 -0

@@ -4852,6 +4852,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			vmf->page = softleaf_to_page(entry);
 			ret = remove_device_exclusive_entry(vmf);
 		} else if (softleaf_is_device_private(entry)) {
+			struct folio *fault_folio;

You can just use the existing 'folio' variable.

Although this function is far too long (400 lines! 14 function-scope variables!) so I don't blame you for not noticing that. Pulling a do_non_swap_page() out of it might be a kindness.

Quoted 12 lines

mm/memory.c +1 -0

 			if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
 				/*
 				 * migrate_to_ram is not yet ready to operate

@@ -4863,6 +4865,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			}
 
 			vmf->page = softleaf_to_page(entry);
+			fault_folio = page_folio(vmf->page);
 			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
 					vmf->address, &vmf->ptl);
 			if (unlikely(!vmf->pte ||

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 4878

Quoted 2 lines

mm/memory.c +0 -0

@@ -4874,15 +4877,15 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			 * Get a page reference while we know the page can't be
 			 * freed.

You should also replace 'page' with 'folio' here. Pages don't have refcounts, folios do.

Hongfu Li

author

2026-07-22 07:46 UTC #5 lore raw

Quoted 11 lines

Quoted 4 lines
From: Hongfu Li <[email protected]>
Replace page-level lock/ref operations with their folio equivalents in the device_private migration path of do_swap_page().
When I do folio conversions, I like to mention the actual advantage of the patch. Here, I'd say something like ...
mm: Use a folio in the softleaf_is_device_private path
Use the folio APIs, replacing four calls to compound_head() with one.

Many thanks for your patient guidance.
I will update the commit message per your suggestion for v2.

Quoted 20 lines

Quoted 13 lines
Introduce a local fault_folio to hold page_folio(vmf->page) to avoid repeated page_folio() invocations.
Signed-off-by Hongfu Li <[email protected]>
---
 mm/memory.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
mm/memory.c +0 -0
index d5e87624f692..bd27372a1052 100644
mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 4855
Quoted 4 lines
mm/memory.c +1 -0
@@ -4852,6 +4852,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			vmf->page = softleaf_to_page(entry);
 			ret = remove_device_exclusive_entry(vmf);
 		} else if (softleaf_is_device_private(entry)) {
+			struct folio *fault_folio;
You can just use the existing 'folio' variable.

Quoted 3 lines

Although this function is far too long (400 lines! 14 function-scope variables!) so I don't blame you for not noticing that. Pulling a do_non_swap_page() out of it might be a kindness.

Thanks, will drop the local fault_folio and reuse the existing folio variable. This change will be included in v2.

Agreed that we can split out helpers from do_swap_page, and I will add a new patch for this refactoring in v2.

Quoted 6 lines

mm/memory.c +0 -0

 			if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
 				/*
 				 * migrate_to_ram is not yet ready to operate

@@ -4863,6 +4865,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			}

Quoted 11 lines

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 4878
Quoted 8 lines
mm/memory.c +1 -0
 			vmf->page = softleaf_to_page(entry);
+			fault_folio = page_folio(vmf->page);
 			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
 					vmf->address, &vmf->ptl);
 			if (unlikely(!vmf->pte ||
@@ -4874,15 +4877,15 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			 * Get a page reference while we know the page can't be
 			 * freed.
You should also replace 'page' with 'folio' here. Pages don't have refcounts, folios do.

Right, I'll update the comment to "folio" to match. Will include in v2.

Best regards,
Hongfu