Skip to contentSkip to conversation

mm/migrate_device: fix pgtable leak in migrate_vma_insert_huge_pmd_page

PATCH Applied Sunny Patel 5 participants 10 messages May 1 – May 20

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

PATCH v3 1/1 mm/migrate_device: fix pgtable leak in migrate_vma_insert_huge_pmd_page

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

---

Changes in v3:
- Added free_abort label as suggested to release the pgtable in error path.
- v2 had the pte_free() call in both error paths, this version has a single goto target for both paths to avoid code duplication.

 mm/migrate_device.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
mm/migrate_device.c +3 -1
index fbfe5715f635..0360b410067b 100644
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 20 lines · On Fri, 1 May 2026 17:21:16 +0530 Sunny Patel <[email protected]>…

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

...

mm/migrate_device.c +1 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 900

Quoted 8 lines
mm/migrate_device.c +2 -0
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

Yikes, we leak that page on several error paths.

Thanks, I'll retain David's ack from the v2 patch.

Balbir, please review?

#2 lore raw
Quoted 33 lines · Andrew Morton
Quoted 20 lines · On Fri, 1 May 2026 17:21:16 +0530 Sunny Patel <[email protected]>…

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

...

mm/migrate_device.c +1 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 900

Quoted 8 lines
mm/migrate_device.c +2 -0
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

Yikes, we leak that page on several error paths.

Quoted 1 line

Thanks, I'll retain David's ack from the v2 patch.

Yes. If we want to avoid more labels, we could do something like:

mm/migrate_device.c +14 -12
index ab49d4dcdb60..babb56c4d47f 100644
@@ -795,8 +795,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        struct folio *folio = page_folio(page);
        int ret;
        vm_fault_t csa_ret;
-       spinlock_t *ptl;
-       pgtable_t pgtable;
+       spinlock_t *ptl = NULL;
+       pgtable_t pgtable = NULL;
        pmd_t entry;
        bool flush = false;
        unsigned long i;
@@ -818,14 +818,14 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
                count_vm_event(THP_FAULT_FALLBACK);
                count_mthp_stat(HPAGE_PMD_ORDER, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
                ret = -ENOMEM;
-               goto abort;
+               goto error;
        }
 
        __folio_mark_uptodate(folio);
 
        pgtable = pte_alloc_one(vma->vm_mm);
        if (unlikely(!pgtable))
-               goto abort;
+               goto error;
 
        if (folio_is_device_private(folio)) {
                swp_entry_t swp_entry;
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        } else {
                if (folio_is_zone_device(folio) &&
                    !folio_is_device_coherent(folio)) {
-                       goto abort;
+                       goto error;
                }
                entry = folio_mk_pmd(folio, vma->vm_page_prot);
                if (vma->vm_flags & VM_WRITE)
@@ -850,21 +850,21 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        ptl = pmd_lock(vma->vm_mm, pmdp);
        csa_ret = check_stable_address_space(vma->vm_mm);
        if (csa_ret)
-               goto unlock_abort;
+               goto error;
 
        /*
         * Check for userfaultfd but do not deliver the fault. Instead,
         * just back off.
         */
        if (userfaultfd_missing(vma))
-               goto unlock_abort;
+               goto error;
 
        if (!pmd_none(*pmdp)) {
                if (!is_huge_zero_pmd(*pmdp))
-                       goto unlock_abort;
+                       goto error;
                flush = true;
        } else if (!pmd_none(*pmdp))
-               goto unlock_abort;
+               goto error;
 
        add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
        folio_add_new_anon_rmap(folio, vma, addr, RMAP_EXCLUSIVE);
@@ -891,9 +891,11 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
        return 0;
 
-unlock_abort:
-       spin_unlock(ptl);
-abort:
+error:
+       if (ptl)
+               spin_unlock(ptl);
+       if (pgtable)
+               pte_free(vma->vm_mm, pgtable);
        for (i = 0; i < HPAGE_PMD_NR; i++)
                src[i] &= ~MIGRATE_PFN_MIGRATE;
        return 0;
Signature · 3 lines
Cheers,

David
Reviewed-by
#3 lore raw
Quoted 112 lines · David Hildenbrand (Arm)
Quoted 33 lines · Andrew Morton
Quoted 20 lines · On Fri, 1 May 2026 17:21:16 +0530 Sunny Patel <[email protected]>…

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

...

mm/migrate_device.c +1 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 900

Quoted 8 lines
mm/migrate_device.c +2 -0
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

Yikes, we leak that page on several error paths.

Quoted 1 line

Thanks, I'll retain David's ack from the v2 patch.

Yes. If we want to avoid more labels, we could do something like:

mm/migrate_device.c +9 -10
index ab49d4dcdb60..babb56c4d47f 100644
@@ -795,8 +795,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        struct folio *folio = page_folio(page);
        int ret;
        vm_fault_t csa_ret;
-       spinlock_t *ptl;
-       pgtable_t pgtable;
+       spinlock_t *ptl = NULL;
+       pgtable_t pgtable = NULL;
        pmd_t entry;
        bool flush = false;
        unsigned long i;
@@ -818,14 +818,14 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
                count_vm_event(THP_FAULT_FALLBACK);
                count_mthp_stat(HPAGE_PMD_ORDER, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
                ret = -ENOMEM;
-               goto abort;
+               goto error;
        }
 
        __folio_mark_uptodate(folio);
 
        pgtable = pte_alloc_one(vma->vm_mm);
        if (unlikely(!pgtable))
-               goto abort;
+               goto error;
 
        if (folio_is_device_private(folio)) {
                swp_entry_t swp_entry;
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        } else {
                if (folio_is_zone_device(folio) &&
                    !folio_is_device_coherent(folio)) {
-                       goto abort;
+                       goto error;
                }
                entry = folio_mk_pmd(folio, vma->vm_page_prot);
                if (vma->vm_flags & VM_WRITE)
@@ -850,21 +850,21 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        ptl = pmd_lock(vma->vm_mm, pmdp);
        csa_ret = check_stable_address_space(vma->vm_mm);
        if (csa_ret)
-               goto unlock_abort;
+               goto error;
 
        /*
         * Check for userfaultfd but do not deliver the fault. Instead,
         * just back off.
         */
        if (userfaultfd_missing(vma))
-               goto unlock_abort;
+               goto error;
 
        if (!pmd_none(*pmdp)) {
                if (!is_huge_zero_pmd(*pmdp))
-                       goto unlock_abort;
+                       goto error;
                flush = true;
        } else if (!pmd_none(*pmdp))
-               goto unlock_abort;
+               goto error;
 
        add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
        folio_add_new_anon_rmap(folio, vma, addr, RMAP_EXCLUSIVE);
@@ -891,9 +891,11 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
        return 0;
 
-unlock_abort:

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 901

Quoted 10 lines
mm/migrate_device.c +5 -2
-       spin_unlock(ptl);
-abort:
+error:
+       if (ptl)
+               spin_unlock(ptl);
+       if (pgtable)
+               pte_free(vma->vm_mm, pgtable);
        for (i = 0; i < HPAGE_PMD_NR; i++)
                src[i] &= ~MIGRATE_PFN_MIGRATE;
        return 0;

The changes look good!

#4 lore raw
Quoted 17 lines
Quoted 4 lines
mm/migrate_device.c +0 -1
-unlock_abort:

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 901

Quoted 10 lines
mm/migrate_device.c +5 -2
-       spin_unlock(ptl);
-abort:
+error:
+       if (ptl)
+               spin_unlock(ptl);
+       if (pgtable)
+               pte_free(vma->vm_mm, pgtable);
        for (i = 0; i < HPAGE_PMD_NR; i++)
                src[i] &= ~MIGRATE_PFN_MIGRATE;
        return 0;

The changes look good!

Quoted 1 line

Do we prefer that or what we have in v3?

In case we prefer what I posted, it would be great if Sunny could send a v4!

Signature · 3 lines
Cheers,

David
reviewer Reviewed-by
#5 lore raw
In reply to David Hildenbrand (Arm)
Quoted 110 lines · David Hildenbrand (Arm)
Quoted 33 lines · Andrew Morton
Quoted 20 lines · On Fri, 1 May 2026 17:21:16 +0530 Sunny Patel <[email protected]>…

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

...

mm/migrate_device.c +1 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 900

Quoted 8 lines
mm/migrate_device.c +2 -0
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

Yikes, we leak that page on several error paths.

Quoted 1 line

Thanks, I'll retain David's ack from the v2 patch.

Yes. If we want to avoid more labels, we could do something like:

mm/migrate_device.c +9 -10
index ab49d4dcdb60..babb56c4d47f 100644
@@ -795,8 +795,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        struct folio *folio = page_folio(page);
        int ret;
        vm_fault_t csa_ret;
-       spinlock_t *ptl;
-       pgtable_t pgtable;
+       spinlock_t *ptl = NULL;
+       pgtable_t pgtable = NULL;
        pmd_t entry;
        bool flush = false;
        unsigned long i;
@@ -818,14 +818,14 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
                count_vm_event(THP_FAULT_FALLBACK);
                count_mthp_stat(HPAGE_PMD_ORDER, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE);
                ret = -ENOMEM;
-               goto abort;
+               goto error;
        }
 
        __folio_mark_uptodate(folio);
 
        pgtable = pte_alloc_one(vma->vm_mm);
        if (unlikely(!pgtable))
-               goto abort;
+               goto error;
 
        if (folio_is_device_private(folio)) {
                swp_entry_t swp_entry;
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        } else {
                if (folio_is_zone_device(folio) &&
                    !folio_is_device_coherent(folio)) {
-                       goto abort;
+                       goto error;
                }
                entry = folio_mk_pmd(folio, vma->vm_page_prot);
                if (vma->vm_flags & VM_WRITE)
@@ -850,21 +850,21 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
        ptl = pmd_lock(vma->vm_mm, pmdp);
        csa_ret = check_stable_address_space(vma->vm_mm);
        if (csa_ret)
-               goto unlock_abort;
+               goto error;
 
        /*
         * Check for userfaultfd but do not deliver the fault. Instead,
         * just back off.
         */
        if (userfaultfd_missing(vma))
-               goto unlock_abort;
+               goto error;
 
        if (!pmd_none(*pmdp)) {
                if (!is_huge_zero_pmd(*pmdp))
-                       goto unlock_abort;
+                       goto error;
                flush = true;
        } else if (!pmd_none(*pmdp))
-               goto unlock_abort;
+               goto error;
 
        add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
        folio_add_new_anon_rmap(folio, vma, addr, RMAP_EXCLUSIVE);
@@ -891,9 +891,11 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
        return 0;
 
-unlock_abort:

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 901

Quoted 10 lines
mm/migrate_device.c +5 -2
-       spin_unlock(ptl);
-abort:
+error:
+       if (ptl)
+               spin_unlock(ptl);
+       if (pgtable)
+               pte_free(vma->vm_mm, pgtable);
        for (i = 0; i < HPAGE_PMD_NR; i++)
                src[i] &= ~MIGRATE_PFN_MIGRATE;
        return 0;

Both look good to me, feel free to add my

in the future versions.

---
Best Regards,
Huang, Ying

#6 lore raw
In reply to Andrew Morton
Quoted 33 lines · Andrew Morton
Quoted 20 lines · On Fri, 1 May 2026 17:21:16 +0530 Sunny Patel <[email protected]>…

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

...

mm/migrate_device.c +1 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 900

Quoted 8 lines
mm/migrate_device.c +2 -0
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

Yikes, we leak that page on several error paths.

Quoted 3 lines

Thanks, I'll retain David's ack from the v2 patch.

Balbir, please review?

The issue can be reached if we try to migrate a non private/coherent folio (I assume DAX/P2PDMA) pages. Even migrate_vma_insert_page() has a similar pattern. Creating the pud/pmd/pte structure there is unnecessary (but not allocations occur, depends on pmd_none() for example), do we want to clean up all of it (need to think through)?

I'll re-review and send fixes as needed. Looks like David sent some fixes out, will review those as well.

Balbir

#7 lore raw
Quoted 39 lines · Balbir Singh
Quoted 33 lines · Andrew Morton
Quoted 20 lines · On Fri, 1 May 2026 17:21:16 +0530 Sunny Patel <[email protected]>…

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

...

mm/migrate_device.c +1 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)

mm/migrate_device.c static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate, · line 900

Quoted 8 lines
mm/migrate_device.c +2 -0
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

Yikes, we leak that page on several error paths.

Quoted 3 lines

Thanks, I'll retain David's ack from the v2 patch.

Balbir, please review?

Quoted 8 lines

The issue can be reached if we try to migrate a non private/coherent folio (I assume DAX/P2PDMA) pages. Even migrate_vma_insert_page() has a similar pattern. Creating the pud/pmd/pte structure there is unnecessary (but not allocations occur, depends on pmd_none() for example), do we want to clean up all of it (need to think through)?

I'll re-review and send fixes as needed. Looks like David sent some fixes out, will review those as well.

I checked migrate_vma_insert_page() is no affected.

Balbir

maintainer Acked-by David Hildenbrand (Arm) Reviewed-by Huang Ying
#8 lore raw
Quoted 6 lines · On Sat, 2 May 2026 10:59:33 +1000 Balbir Singh <[email protected]> wrote:
Quoted 6 lines

The issue can be reached if we try to migrate a non private/coherent folio (I assume DAX/P2PDMA) pages. Even migrate_vma_insert_page() has a similar pattern. Creating the pud/pmd/pte structure there is unnecessary (but not allocations occur, depends on pmd_none() for example), do we want to clean up all of it (need to think through)?

Quoted 6 lines
Quoted 3 lines

I'll re-review and send fixes as needed. Looks like David sent some fixes out, will review those as well.

I checked migrate_vma_insert_page() is no affected.

Thanks.

What to do here. We have a cc:stable hotfix which is adequately tested and seems to have passed review. And a cleanup patch from David.

I'm inclined to upstream this patch as-is and leave the cleanup on David's TODO list (one more can't make any difference ;))

?

From: Sunny Patel <[email protected]>
Subject: mm/migrate_device: fix pgtable leak in migrate_vma_insert_huge_pmd_page
Date: Fri, 1 May 2026 17:21:16 +0530

When migrate_vma_insert_huge_pmd_page() jumps to unlock_abort due to a PMD check failure, the pgtable allocated earlier via pte_alloc_one() is never freed, causing a memory leak.

Added free_abort label to release the pgtable in error path.

Fixes a30b48bf1b24 ("mm/migrate_device: implement THP migration of zone device pages")
Cc Alistair Popple <[email protected]>
Cc Balbir Singh <[email protected]>
Cc Byungchul Park <[email protected]>
Cc Gregory Price <[email protected]>
Cc Joshua Hahn <[email protected]>
Cc Matthew Brost <[email protected]>
Cc Rakie Kim <[email protected]>
Cc Zi Yan <[email protected]>

---

 mm/migrate_device.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
mm/migrate_device.c renamed from mm/migrate_device.c~mm-migrate_device-fix-pgtable-leak-in-migrate_vma_insert_huge_pmd_page +3 -1
@@ -840,7 +840,7 @@ static int migrate_vma_insert_huge_pmd_p
 	} else {
 		if (folio_is_zone_device(folio) &&
 		    !folio_is_device_coherent(folio)) {
-			goto abort;
+			goto free_abort;
 		}
 		entry = folio_mk_pmd(folio, vma->vm_page_prot);
 		if (vma->vm_flags & VM_WRITE)
@@ -893,6 +893,8 @@ static int migrate_vma_insert_huge_pmd_p
 
 unlock_abort:
 	spin_unlock(ptl);
+free_abort:
+	pte_free(vma->vm_mm, pgtable);
 abort:
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		src[i] &= ~MIGRATE_PFN_MIGRATE;

_

#9 lore raw
Quoted 15 lines · Andrew Morton
Quoted 6 lines · On Sat, 2 May 2026 10:59:33 +1000 Balbir Singh <[email protected]> wrote:
Quoted 6 lines

The issue can be reached if we try to migrate a non private/coherent folio (I assume DAX/P2PDMA) pages. Even migrate_vma_insert_page() has a similar pattern. Creating the pud/pmd/pte structure there is unnecessary (but not allocations occur, depends on pmd_none() for example), do we want to clean up all of it (need to think through)?

Quoted 6 lines
Quoted 3 lines

I'll re-review and send fixes as needed. Looks like David sent some fixes out, will review those as well.

I checked migrate_vma_insert_page() is no affected.

Quoted 7 lines

Thanks.

What to do here. We have a cc:stable hotfix which is adequately tested and seems to have passed review. And a cleanup patch from David.

I'm inclined to upstream this patch as-is and leave the cleanup on David's TODO list (one more can't make any difference ;))

No strong opinion from my side.

Signature · 3 lines
Cheers,

David