Skip to contentSkip to conversation

mm: use ptep_get() instead of directly dereferencing pte_t*

PATCH Applied Ryan Roberts 5 participants 7 messages Mar 10, 2025 – Mar 11, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm: use ptep_get() instead of directly dereferencing pte_t*

It is best practice for all pte accesses to go via the arch helpers, to ensure non-torn values and to allow the arch to intervene where needed (contpte for arm64 for example). While in this case it was probably safe to directly dereference, let's tidy it up for consistency.

---
 mm/migrate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/migrate.c +1 -1
index 22e270f727ed..33a22c2d6b20 100644
@@ -202,7 +202,7 @@ static bool try_to_map_unused_to_zeropage(struct page_vma_mapped_walk *pvmw,
 		return false;
 	VM_BUG_ON_PAGE(!PageAnon(page), page);
 	VM_BUG_ON_PAGE(!PageLocked(page), page);
-	VM_BUG_ON_PAGE(pte_present(*pvmw->pte), page);
+	VM_BUG_ON_PAGE(pte_present(ptep_get(pvmw->pte)), page);
 
 	if (folio_test_mlocked(folio) || (pvmw->vma->vm_flags & VM_LOCKED) ||
 	    mm_forbids_zeropage(pvmw->vma->vm_mm))

--
2.43.0

Reviewed-by on v1 1/1
#1 lore raw
Quoted 6 lines · Ryan Roberts

It is best practice for all pte accesses to go via the arch helpers, to ensure non-torn values and to allow the arch to intervene where needed (contpte for arm64 for example). While in this case it was probably safe to directly dereference, let's tidy it up for consistency.

LGTM, so:

Quoted 20 lines
---
 mm/migrate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/migrate.c +1 -2
index 22e270f727ed..33a22c2d6b20 100644
@@ -202,7 +202,7 @@ static bool try_to_map_unused_to_zeropage(struct page_vma_mapped_walk *pvmw,
 		return false;
 	VM_BUG_ON_PAGE(!PageAnon(page), page);
 	VM_BUG_ON_PAGE(!PageLocked(page), page);
-	VM_BUG_ON_PAGE(pte_present(*pvmw->pte), page);
+	VM_BUG_ON_PAGE(pte_present(ptep_get(pvmw->pte)), page);
 
 	if (folio_test_mlocked(folio) || (pvmw->vma->vm_flags & VM_LOCKED) ||
 	    mm_forbids_zeropage(pvmw->vma->vm_mm))
--

2.43.0

Reviewed-by on v1 1/1
#2 lore raw
In reply to Ryan Roberts

Hi Ryan,

Quoted 9 lines · Ryan Roberts

It is best practice for all pte accesses to go via the arch helpers, to ensure non-torn values and to allow the arch to intervene where needed (contpte for arm64 for example). While in this case it was probably safe to directly dereference, let's tidy it up for consistency.

---
  mm/migrate.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

This looks good to me. So

BTW, there are many other places in the kernel that directly dereference pmd_t* and pud_t*, etc.

For example:

root@debian:~# grep "*vmf->pmd" . -rwn
./mm/memory.c:5113:	if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) {
./mm/memory.c:5207:	if (unlikely(!pmd_none(*vmf->pmd)))
./mm/memory.c:5339:	if (pmd_none(*vmf->pmd)) {
./mm/memory.c:5490:	if (pmd_none(*vmf->pmd)) {
./mm/memory.c:5996:	if (unlikely(pmd_none(*vmf->pmd))) {
./mm/filemap.c:3612:	if (pmd_trans_huge(*vmf->pmd)) {
./mm/filemap.c:3618:	if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
./mm/filemap.c:3628:	if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
./mm/huge_memory.c:1237:	if (unlikely(!pmd_none(*vmf->pmd))) {
./mm/huge_memory.c:1352:		if (pmd_none(*vmf->pmd)) {
./mm/huge_memory.c:1496:	if (pmd_none(*vmf->pmd)) {
./mm/huge_memory.c:1882:	if (unlikely(!pmd_same(*vmf->pmd, vmf->orig_pmd)))
./mm/huge_memory.c:1947:	if (unlikely(!pmd_same(*vmf->pmd, orig_pmd))) {
./mm/huge_memory.c:1965:		if (unlikely(!pmd_same(*vmf->pmd, orig_pmd))) {
./fs/dax.c:1935:	if (pmd_trans_huge(*vmf->pmd) || pmd_devmap(*vmf->pmd)) {
./fs/dax.c:2058:	if (!pmd_none(*vmf->pmd) && !pmd_trans_huge(*vmf->pmd) &&
./fs/dax.c:2059:			!pmd_devmap(*vmf->pmd)) {

Would it be best to clean them up as well?

Thanks,
Qi

Quoted 17 lines
mm/migrate.c +1 -2
index 22e270f727ed..33a22c2d6b20 100644
@@ -202,7 +202,7 @@ static bool try_to_map_unused_to_zeropage(struct page_vma_mapped_walk *pvmw,
  		return false;
  	VM_BUG_ON_PAGE(!PageAnon(page), page);
  	VM_BUG_ON_PAGE(!PageLocked(page), page);
-	VM_BUG_ON_PAGE(pte_present(*pvmw->pte), page);
+	VM_BUG_ON_PAGE(pte_present(ptep_get(pvmw->pte)), page);
 
  	if (folio_test_mlocked(folio) || (pvmw->vma->vm_flags & VM_LOCKED) ||
  	    mm_forbids_zeropage(pvmw->vma->vm_mm))
--

2.43.0

#3 lore raw
Quoted 8 lines · Qi Zheng

Hi Ryan,

Quoted 5 lines · Ryan Roberts

It is best practice for all pte accesses to go via the arch helpers, to ensure non-torn values and to allow the arch to intervene where needed (contpte for arm64 for example). While in this case it was probably safe to directly dereference, let's tidy it up for consistency.

Quoted 8 lines
Quoted 4 lines

---
  mm/migrate.c | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)

This looks good to me. So

Thanks!

Quoted 3 lines

BTW, there are many other places in the kernel that directly dereference pmd_t* and pud_t*, etc.

It's all a little bit murky. For now, from arm64's perspective at least, there is only a hard requirement for ptes to be accessed through the arch helper. This is because arm64's contpte layer may apply a transform when reading the pte.

In general there are also potential issues with tearing, if you don't at least read with READ_ONCE(). But often to consumer of the value is tolerant to tearing (e.g. pmd_none(), etc). Also, in practice on arm64 the compiler will emit instructions that ensure single-copy-atomicity for direct dereferences, so it all works out.

That said, Anshuman (cc'ed) has been looking at supporting FEAT_D128 (128 bit page table descriptors) on arm64. The compiler does not emit single-copy-atomic loads for direct dereferences of 128 bit data, so he has been working on converting the other levels to use the accessors for that reason.

But that has some potentially problematic interactions with level folding that need to be solved. Some arches rely on the compiler optimizing away the direct dereferences when folded. But it can't do that for a READ_ONCE().

I believe Anshuman is aiming to post a series to do this at some point in the future.

Thanks,
Ryan

Quoted 47 lines

For example:

root@debian:~# grep "*vmf->pmd" . -rwn
./mm/memory.c:5113:    if (pmd_none(*vmf->pmd) && !vmf->prealloc_pte) {
./mm/memory.c:5207:    if (unlikely(!pmd_none(*vmf->pmd)))
./mm/memory.c:5339:    if (pmd_none(*vmf->pmd)) {
./mm/memory.c:5490:    if (pmd_none(*vmf->pmd)) {
./mm/memory.c:5996:    if (unlikely(pmd_none(*vmf->pmd))) {
./mm/filemap.c:3612:    if (pmd_trans_huge(*vmf->pmd)) {
./mm/filemap.c:3618:    if (pmd_none(*vmf->pmd) &&
folio_test_pmd_mappable(folio)) {
./mm/filemap.c:3628:    if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
./mm/huge_memory.c:1237:    if (unlikely(!pmd_none(*vmf->pmd))) {
./mm/huge_memory.c:1352:        if (pmd_none(*vmf->pmd)) {
./mm/huge_memory.c:1496:    if (pmd_none(*vmf->pmd)) {
./mm/huge_memory.c:1882:    if (unlikely(!pmd_same(*vmf->pmd, vmf->orig_pmd)))
./mm/huge_memory.c:1947:    if (unlikely(!pmd_same(*vmf->pmd, orig_pmd))) {
./mm/huge_memory.c:1965:        if (unlikely(!pmd_same(*vmf->pmd, orig_pmd))) {
./fs/dax.c:1935:    if (pmd_trans_huge(*vmf->pmd) || pmd_devmap(*vmf->pmd)) {
./fs/dax.c:2058:    if (!pmd_none(*vmf->pmd) && !pmd_trans_huge(*vmf->pmd) &&
./fs/dax.c:2059:            !pmd_devmap(*vmf->pmd)) {

Would it be best to clean them up as well?

Thanks,
Qi

Quoted 18 lines
mm/migrate.c +1 -1
index 22e270f727ed..33a22c2d6b20 100644
@@ -202,7 +202,7 @@ static bool try_to_map_unused_to_zeropage(struct
 page_vma_mapped_walk *pvmw,
           return false;
       VM_BUG_ON_PAGE(!PageAnon(page), page);
       VM_BUG_ON_PAGE(!PageLocked(page), page);
-    VM_BUG_ON_PAGE(pte_present(*pvmw->pte), page);
+    VM_BUG_ON_PAGE(pte_present(ptep_get(pvmw->pte)), page);
 

      if (folio_test_mlocked(folio) || (pvmw->vma->vm_flags & VM_LOCKED) || mm_forbids_zeropage(pvmw->vma->vm_mm))

Signature · 1 line
2.43.0
#4 lore raw
Quoted 39 lines · Ryan Roberts
Quoted 8 lines · Qi Zheng

Hi Ryan,

Quoted 5 lines · Ryan Roberts

It is best practice for all pte accesses to go via the arch helpers, to ensure non-torn values and to allow the arch to intervene where needed (contpte for arm64 for example). While in this case it was probably safe to directly dereference, let's tidy it up for consistency.

Quoted 8 lines
Quoted 4 lines
---
   mm/migrate.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)

This looks good to me. So

Thanks!

Quoted 3 lines

BTW, there are many other places in the kernel that directly dereference pmd_t* and pud_t*, etc.

It's all a little bit murky. For now, from arm64's perspective at least, there is only a hard requirement for ptes to be accessed through the arch helper. This is because arm64's contpte layer may apply a transform when reading the pte.

In general there are also potential issues with tearing, if you don't at least read with READ_ONCE(). But often to consumer of the value is tolerant to tearing (e.g. pmd_none(), etc). Also, in practice on arm64 the compiler will emit instructions that ensure single-copy-atomicity for direct dereferences, so it all works out.

That said, Anshuman (cc'ed) has been looking at supporting FEAT_D128 (128 bit page table descriptors) on arm64. The compiler does not emit single-copy-atomic loads for direct dereferences of 128 bit data, so he has been working on converting the other levels to use the accessors for that reason.

Quoted 6 lines

But that has some potentially problematic interactions with level folding that need to be solved. Some arches rely on the compiler optimizing away the direct dereferences when folded. But it can't do that for a READ_ONCE().

I believe Anshuman is aiming to post a series to do this at some point in the future.

Got it. Thanks!

Reviewed-by on v1 1/1
#5 lore raw
In reply to Ryan Roberts
Quoted 24 lines · Ryan Roberts

It is best practice for all pte accesses to go via the arch helpers, to ensure non-torn values and to allow the arch to intervene where needed (contpte for arm64 for example). While in this case it was probably safe to directly dereference, let's tidy it up for consistency.

---
 mm/migrate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/migrate.c +1 -2
index 22e270f727ed..33a22c2d6b20 100644
@@ -202,7 +202,7 @@ static bool try_to_map_unused_to_zeropage(struct page_vma_mapped_walk *pvmw,
 		return false;
 	VM_BUG_ON_PAGE(!PageAnon(page), page);
 	VM_BUG_ON_PAGE(!PageLocked(page), page);
-	VM_BUG_ON_PAGE(pte_present(*pvmw->pte), page);
+	VM_BUG_ON_PAGE(pte_present(ptep_get(pvmw->pte)), page);
 
 	if (folio_test_mlocked(folio) || (pvmw->vma->vm_flags & VM_LOCKED) ||
 	    mm_forbids_zeropage(pvmw->vma->vm_mm))
--
Quoted 3 lines

2.43.0

LGTM.