From: Naoya Horiguchi on
For the migration of PAGE_SIZE pages, we can choose to continue to do
migration with "force" switch even if the old page has page lock held.
But for hugepage, I/O of subpages are not necessarily completed
in ascending order, which can cause race.
So we make migration fail then for safety.

Signed-off-by: Naoya Horiguchi <n-horiguchi(a)ah.jp.nec.com>
---
mm/migrate.c | 15 +++++++++------
1 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/mm/migrate.c b/mm/migrate.c
index 7f9a37c..43347e1 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -820,11 +820,14 @@ static int unmap_and_move_huge_page(new_page_t get_new_page,

rc = -EAGAIN;

- if (!trylock_page(hpage)) {
- if (!force)
- goto out;
- lock_page(hpage);
- }
+ /*
+ * If some subpages are locked, it can cause race condition.
+ * So then we return from migration and try again.
+ */
+ if (!trylock_huge_page(hpage))
+ goto out;
+
+ wait_on_huge_page_writeback(hpage);

if (PageAnon(hpage)) {
rcu_read_lock();
@@ -855,7 +858,7 @@ static int unmap_and_move_huge_page(new_page_t get_new_page,
if (rcu_locked)
rcu_read_unlock();
out:
- unlock_page(hpage);
+ unlock_huge_page(hpage);

if (rc != -EAGAIN)
put_page(hpage);
--
1.7.2.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/