mm: thp: fix flags for pmd migration when split

When splitting a huge migrating PMD, we'll transfer all the existing PMD bits and apply them again onto the small PTEs. However we are fetching the bits unconditionally via pmd_soft_dirty(), pmd_write() or pmd_yound() while actually they don't make sense at all when it's a migration entry. Fix them up. Since at it, drop the ifdef together as not needed.

Note that if my understanding is correct about the problem then if without the patch there is chance to lose some of the dirty bits in the migrating pmd pages (on x86_64 we're fetching bit 11 which is part of swap offset instead of bit 2) and it could potentially corrupt the memory of an userspace program which depends on the dirty bit.

CC Andrea Arcangeli <[email protected]>

CC Andrew Morton <[email protected]>

CC "Kirill A. Shutemov" <[email protected]>

CC Matthew Wilcox <[email protected]>

CC Michal Hocko <[email protected]>

CC Dave Jiang <[email protected]>

CC "Aneesh Kumar K.V" <[email protected]>

CC Souptick Joarder <[email protected]>

CC Konstantin Khlebnikov <[email protected]>

CC Zi Yan <[email protected]>

CC [email protected]

CC [email protected]

Signed-off-by Peter Xu <[email protected]>

---
v2:
- fix it up for young/write/dirty bits too [Konstantin] v3:
- fetch write correctly for migration entry; drop macro [Konstantin]

---
 mm/huge_memory.c | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)

mm/huge_memory.c +11 -10

index f2d19e4fe854..aebade83cec9 100644
@@ -2145,23 +2145,25 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
         */
        old_pmd = pmdp_invalidate(vma, haddr, pmd);
 
-#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
        pmd_migration = is_pmd_migration_entry(old_pmd);
-       if (pmd_migration) {
+       if (unlikely(pmd_migration)) {
                swp_entry_t entry;
 
                entry = pmd_to_swp_entry(old_pmd);
                page = pfn_to_page(swp_offset(entry));
-       } else
-#endif
+               write = is_write_migration_entry(entry);
+               young = false;
+               soft_dirty = pmd_swp_soft_dirty(old_pmd);
+       } else {
                page = pmd_page(old_pmd);
+               if (pmd_dirty(old_pmd))
+                       SetPageDirty(page);
+               write = pmd_write(old_pmd);
+               young = pmd_young(old_pmd);
+               soft_dirty = pmd_soft_dirty(old_pmd);
+       }
        VM_BUG_ON_PAGE(!page_count(page), page);
        page_ref_add(page, HPAGE_PMD_NR - 1);
-       if (pmd_dirty(old_pmd))
-               SetPageDirty(page);
-       write = pmd_write(old_pmd);
-       young = pmd_young(old_pmd);
-       soft_dirty = pmd_soft_dirty(old_pmd);
 
        /*
         * Withdraw the table only after we mark the pmd entry invalid.
--

2.17.1

mm/huge_memory.c +11 -4

index f2d19e4fe854..aebade83cec9 100644
@@ -2145,23 +2145,25 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 	 */
 	old_pmd = pmdp_invalidate(vma, haddr, pmd);
 
-#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
 	pmd_migration = is_pmd_migration_entry(old_pmd);
-	if (pmd_migration) {
+	if (unlikely(pmd_migration)) {
 		swp_entry_t entry;
 
 		entry = pmd_to_swp_entry(old_pmd);
 		page = pfn_to_page(swp_offset(entry));
-	} else
-#endif
+		write = is_write_migration_entry(entry);
+		young = false;
+		soft_dirty = pmd_swp_soft_dirty(old_pmd);
+	} else {
 		page = pmd_page(old_pmd);
+		if (pmd_dirty(old_pmd))
+			SetPageDirty(page);
+		write = pmd_write(old_pmd);
+		young = pmd_young(old_pmd);
+		soft_dirty = pmd_soft_dirty(old_pmd);
+	}

When splitting a huge migrating PMD, we'll transfer all the existing PMD bits and apply them again onto the small PTEs. However we are fetching the bits unconditionally via pmd_soft_dirty(), pmd_write() or pmd_yound() while actually they don't make sense at all when it's a migration entry. Fix them up. Since at it, drop the ifdef together as not needed.

Note that if my understanding is correct about the problem then if without the patch there is chance to lose some of the dirty bits in the migrating pmd pages (on x86_64 we're fetching bit 11 which is part of swap offset instead of bit 2) and it could potentially corrupt the memory of an userspace program which depends on the dirty bit.

CC Andrea Arcangeli <[email protected]>

CC Andrew Morton <[email protected]>

CC "Kirill A. Shutemov" <[email protected]>

CC Matthew Wilcox <[email protected]>

CC Michal Hocko <[email protected]>

CC Dave Jiang <[email protected]>

CC "Aneesh Kumar K.V" <[email protected]>

CC Souptick Joarder <[email protected]>

CC Konstantin Khlebnikov <[email protected]>

CC Zi Yan <[email protected]>

CC [email protected]

CC [email protected]

Signed-off-by Peter Xu <[email protected]>

Quoted 13 lines · Peter Xu

When splitting a huge migrating PMD, we'll transfer all the existing PMD bits and apply them again onto the small PTEs. However we are fetching the bits unconditionally via pmd_soft_dirty(), pmd_write() or pmd_yound() while actually they don't make sense at all when it's a migration entry. Fix them up. Since at it, drop the ifdef together as not needed.

Note that if my understanding is correct about the problem then if without the patch there is chance to lose some of the dirty bits in the migrating pmd pages (on x86_64 we're fetching bit 11 which is part of swap offset instead of bit 2) and it could potentially corrupt the memory of an userspace program which depends on the dirty bit.

Quoted 13 lines

CC Andrea Arcangeli <[email protected]>

CC Andrew Morton <[email protected]>

CC "Kirill A. Shutemov" <[email protected]>

CC Matthew Wilcox <[email protected]>

CC Michal Hocko <[email protected]>

CC Dave Jiang <[email protected]>

CC "Aneesh Kumar K.V" <[email protected]>

CC Souptick Joarder <[email protected]>

CC Konstantin Khlebnikov <[email protected]>

CC Zi Yan <[email protected]>

CC [email protected]

CC [email protected]

Signed-off-by Peter Xu <[email protected]>

Acked-by Kirill A. Shutemov <[email protected]>

Stable?