Re: [PATCH][RFC] mm: memcontrol: fix a unbalance uncharged count
From: Michal Hocko <hidden>
Date: 2016-05-31 09:50:07
Also in:
linux-mm
On Tue 31-05-16 17:38:32, roy.qing.li-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org wrote:
From: Li RongQing <redacted> I see the number of page of hpage_nr_pages(page) is charged if page is transparent huge or hugetlbfs pages; but when uncharge a huge page,
hugetlb pages do not get charged to the memcg. They have their own hugetlbfscg controller.
(1<<compound_order) page is uncharged, and maybe hpage_nr_pages(page) is not same as 1<<compound_order.
This should never happen. So this is not a fix. I guess it would a clean up though.
And remove VM_BUG_ON_PAGE(!PageTransHuge(page), page); since PageTransHuge(page) always is true, when this VM_BUG_ON_PAGE is called.
I guess we can drop this BUG_ON because nobody should be be seeing the page at this stage so we cannot race with a THP split up. But I would really have to look closer. This would be a patch on its own with the full explanation though, IMHO.
quoted hunk ↗ jump to hunk
Signed-off-by: Li RongQing <redacted> --- mm/memcontrol.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 12aaadd..28c0137 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c@@ -5453,8 +5453,7 @@ static void uncharge_list(struct list_head *page_list) } if (PageTransHuge(page)) { - nr_pages <<= compound_order(page); - VM_BUG_ON_PAGE(!PageTransHuge(page), page); + nr_pages = hpage_nr_pages(page); nr_huge += nr_pages; }-- 2.1.4
-- Michal Hocko SUSE Labs