On Wed 12-03-25 22:48:12, Shakeel Butt wrote: > We never charge the page counters of root memcg, so there is no need to > put root memcg in the memcg stock. At the moment, refill_stock() can be > called from try_charge_memcg(), obj_cgroup_uncharge_pages() and > mem_cgroup_uncharge_skmem(). > > The try_charge_memcg() and mem_cgroup_uncharge_skmem() are never called > with root memcg, so those are fine. However obj_cgroup_uncharge_pages() > can potentially call refill_stock() with root memcg if the objcg object > has been reparented over to the root memcg. Let's just avoid > refill_stock() from obj_cgroup_uncharge_pages() for root memcg. Makes sense. It is just pointless draining of a different memcg. > Signed-off-by: Shakeel Butt <shakeel.butt@xxxxxxxxx> Acked-by: Michal Hocko <mhockoc@xxxxxxxx> Thanks! > --- > mm/memcontrol.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > index e08ce52caabd..393b73aec6dd 100644 > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -2643,7 +2643,8 @@ static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, > > mod_memcg_state(memcg, MEMCG_KMEM, -nr_pages); > memcg1_account_kmem(memcg, -nr_pages); > - refill_stock(memcg, nr_pages); > + if (!mem_cgroup_is_root(memcg)) > + refill_stock(memcg, nr_pages); > > css_put(&memcg->css); > } > -- > 2.47.1 > -- Michal Hocko SUSE Labs
