Skip to contentSkip to conversation

mm/migrate: fix remove_migration_pte() of hugetlb entry

PATCH In review Hugh Dickins 2 participants 3 messages Feb 27, 2022 – Feb 28, 2022

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

PATCH 1/1 mm/migrate: fix remove_migration_pte() of hugetlb entry

The foliation of remove_migration_pte() is currently wrong on hugetlb anon entries, causing LTP move_pages12 to crash on BUG_ON(!PageLocked) in hugepage_add_anon_rmap().

Fixes b4010e88f071 ("mm/migrate: Convert remove_migration_ptes() to folios")

---
Please just fold in if you agree.

 mm/migrate.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
mm/migrate.c +2 -1
@@ -182,7 +182,8 @@ static bool remove_migration_pte(struct
 		struct page *new;
 		unsigned long idx = 0;
 
-		if (!folio_test_ksm(folio))
+		/* Skip call in common case, plus .pgoff is invalid for KSM */
+		if (pvmw.nr_pages != 1 && !folio_test_hugetlb(folio))
 			idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff;
 		new = folio_page(folio, idx);
 
#1 lore raw

mm/migrate.c static bool remove_migration_pte(struct · line 187

Quoted 4 lines · Hugh Dickins
mm/migrate.c +2 -1
-		if (!folio_test_ksm(folio))
+		/* Skip call in common case, plus .pgoff is invalid for KSM */
+		if (pvmw.nr_pages != 1 && !folio_test_hugetlb(folio))
 			idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff;

How do you feel about this instead?

-               if (!folio_test_ksm(folio))
+               /* pgoff is invalid for ksm pages, but they are never large */
+               if (folio_test_large(folio) && !folio_test_hugetlb(folio))
                        idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff;
#2 lore raw

mm/migrate.c static bool remove_migration_pte(struct · line 187

Quoted 13 lines · Matthew Wilcox

mm/migrate.c static bool remove_migration_pte(struct · line 187

Quoted 4 lines · Hugh Dickins
mm/migrate.c +2 -1
-		if (!folio_test_ksm(folio))
+		/* Skip call in common case, plus .pgoff is invalid for KSM */
+		if (pvmw.nr_pages != 1 && !folio_test_hugetlb(folio))
 			idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff;

How do you feel about this instead?

mm/migrate.c +2 -1
-               if (!folio_test_ksm(folio))
+               /* pgoff is invalid for ksm pages, but they are never large */
+               if (folio_test_large(folio) && !folio_test_hugetlb(folio))
                        idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff;

That looks nicer to me too. I'll assume that's what you will add or squash in your tree, and no need for me to resend - thanks.

Hugh