On Mon 30-08-21 09:24:22, Rik van Riel wrote: > On Mon, 2021-08-30 at 13:33 +0200, Michal Hocko wrote: [...] > > I must be missing something but how can cgroup_size be ever 0 when it > > is > > max(cgroup_size, protection) and protection != 0? > > Going into the condition we use if (low || min), where > it is possible for low > 0 && min == 0. > > Inside the conditional, we can end up testing against > min. Dang, I was looking at the tree without f56ce412a59d7 applied. My bad! Personally I would consider the following slightly easier to follow scan = lruvec_size - lruvec_size * protection / max(cgroup_size, 1); The code is quite tricky already and if you asked me what kind of effect cgroup_size + 1 have there I would just shrug shoulders... Anyway your fix will prevent the reported problem and I cannot see any obvious problem with it either so Acked-by: Michal Hocko <mhocko@xxxxxxxx> -- Michal Hocko SUSE Labs
