The straight backport of 5.12's e1baddf8475b ("mm/memcg: set memcg when splitting page") works fine in 5.11, but turned out to be wrong for 5.10: because that relies on a separate flag, which must also be set for the memcg to be recognized and uncharged and cleared when freeing. Fix that.
Signed-off-by: Hugh Dickins hughd@google.com ---
mm/memcontrol.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
--- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -3274,13 +3274,17 @@ void obj_cgroup_uncharge(struct obj_cgro void split_page_memcg(struct page *head, unsigned int nr) { struct mem_cgroup *memcg = head->mem_cgroup; + int kmemcg = PageKmemcg(head); int i;
if (mem_cgroup_disabled() || !memcg) return;
- for (i = 1; i < nr; i++) + for (i = 1; i < nr; i++) { head[i].mem_cgroup = memcg; + if (kmemcg) + __SetPageKmemcg(head + i); + } css_get_many(&memcg->css, nr - 1); }
On Sun, Mar 28, 2021 at 05:13:13PM -0700, Hugh Dickins wrote:
The straight backport of 5.12's e1baddf8475b ("mm/memcg: set memcg when splitting page") works fine in 5.11, but turned out to be wrong for 5.10: because that relies on a separate flag, which must also be set for the memcg to be recognized and uncharged and cleared when freeing. Fix that.
Signed-off-by: Hugh Dickins hughd@google.com
mm/memcontrol.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
--- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -3274,13 +3274,17 @@ void obj_cgroup_uncharge(struct obj_cgro void split_page_memcg(struct page *head, unsigned int nr) { struct mem_cgroup *memcg = head->mem_cgroup;
- int kmemcg = PageKmemcg(head); int i;
if (mem_cgroup_disabled() || !memcg) return;
- for (i = 1; i < nr; i++)
- for (i = 1; i < nr; i++) { head[i].mem_cgroup = memcg;
if (kmemcg)__SetPageKmemcg(head + i);- } css_get_many(&memcg->css, nr - 1);
}
Thanks for the fix, now queued up.
greg k-h
linux-stable-mirror@lists.linaro.org
-
Greg KH -
Hugh Dickins