Re: [PATCH 4/4] mm,oom: Check pending victims earlier in out_of_memory().

[Michal Hocko <mhocko@kernel.org>]

On Thu 07-06-18 20:00:23, Tetsuo Handa wrote:
> The "mm, oom: cgroup-aware OOM killer" patchset is trying to introduce
> INFLIGHT_VICTIM in order to replace open-coded ((void *)-1UL). But
> (regarding CONFIG_MMU=y case) we have a list of inflight OOM victim
> threads which are connected to oom_reaper_list. Thus we can check
> whether there are inflight OOM victims before starting process/memcg
> list traversal. Since it is likely that only few threads are linked to
> oom_reaper_list, checking all victims' OOM domain will not matter.
> 
> Thus, check whether there are inflight OOM victims before starting
> process/memcg list traversal and eliminate the "abort" path.

OK, this looks like a nice shortcut. I am quite surprise that all your
NOMMU concerns are gone now while you clearly regress that case because
inflight victims are not detected anymore AFAICS. Not that I care all
that much, just sayin'.

Anyway, I would suggest splitting this into two patches. One to add an
early check for inflight oom victims and one removing the detection from
oom_evaluate_task. Just to make it easier to revert if somebody on nommu
actually notices a regression.

> Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> Cc: Roman Gushchin <guro@fb.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Cc: Johannes Weiner <hannes@cmpxchg.org>
> Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
> Cc: David Rientjes <rientjes@google.com>
> Cc: Tejun Heo <tj@kernel.org>

[...]
-- 
Michal Hocko
SUSE Labs