[PATCH] memcg: Ignore partial THP when moving task

[Michal Hocko]

After "mm: rework mapcount accounting to enable 4k mapping of THPs"
it is possible to have a partial THP accessible via ptes. Memcg task
migration code is not prepared for this situation and uncharges the tail
page from the original memcg while the original THP is still charged via
the head page which is not mapped to the moved task. The page counter
of the origin memcg will underflow when the whole THP is uncharged later
on and lead to:
WARNING: CPU: 0 PID: 1340 at mm/page_counter.c:26 page_counter_cancel+0x34/0x40()
reported by Minchan Kim.

This patch prevents from the underflow by skipping any partial THP pages
in mem_cgroup_move_charge_pte_range. PageTransCompound is checked when
we do pte walk. This means that a process might leave a partial THP
behind in the original memcg if there is no other process mapping it via
pmd but this is considered acceptable because it shouldn't happen often
and this is not considered a memory leak because the original THP is
still accessible and reclaimable. Moreover the task migration has always
been racy and never guaranteed to move all pages.

Reported-by: Minchan Kim <minchan@xxxxxxxxxx>
Acked-by: Johannes Weiner <hannes@xxxxxxxxxxx>
Signed-off-by: Michal Hocko <mhocko@xxxxxxxx>
---
 mm/memcontrol.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 79a29d564bff..4cecefa4a3b0 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -4895,6 +4895,14 @@ static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
 		switch (get_mctgt_type(vma, addr, ptent, &target)) {
 		case MC_TARGET_PAGE:
 			page = target.page;
+			/*
+			 * We can have a part of the split pmd here. Moving it
+			 * can be done but it would be too convoluted so simply
+			 * ignore such a partial THP and keep it in original
+			 * memcg. There should be somebody mapping the head.
+			 */
+			if (PageTransCompound(page))
+				goto put;
 			if (isolate_lru_page(page))
 				goto put;
 			if (!mem_cgroup_move_account(page, false,
-- 
2.6.2

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