Hello, Richard Palethorpe <rpalethorpe@xxxxxxx> writes: > Hello Michal, > > Michal Koutný <mkoutny@xxxxxxxx> writes: > >> Hello. >> >> On Wed, Oct 14, 2020 at 08:07:49PM +0100, Richard Palethorpe <rpalethorpe@xxxxxxxx> wrote: >>> SLAB objects which outlive their memcg are moved to their parent >>> memcg where they may be uncharged. However if they are moved to the >>> root memcg, uncharging will result in negative page counter values as >>> root has no page counters. >> Why do you think those are reparented objects? If those are originally >> charged in a non-root cgroup, then the charge value should be propagated up the >> hierarchy, including root memcg, so if they're later uncharged in root >> after reparenting, it should still break even. (Or did I miss some stock >> imbalance?) > > I traced it and can see they are reparented objects and that the root > groups counters are zero (or negative if I run madvise06 multiple times) > before a drain takes place. I'm guessing this is because the root group > has 'use_hierachy' set to false so that the childs page_counter parents > are set to NULL. However I will check, because I'm not sure about > either. Yes, it appears that use_hierarchy=0 which is probably because the test mounts cgroup v1, creates a child group within that and does not set use_hierarchy on the root. On v2 root always has use_hierarchy enabled. > >> >> (But the patch seems justifiable to me as objects (not)charged directly to >> root memcg may be incorrectly uncharged.) >> >> Thanks, >> Michal I'm don't know if that could happen without reparenting. I suppose if use_hierarchy=1 then actually this patch will result in root being overcharged, so perhaps it should also check for use_hierarchy? -- Thank you, Richard.
