On 2019/01/07 20:41, Michal Hocko wrote: > On Sun 06-01-19 15:02:24, Tetsuo Handa wrote: >> Michal and Johannes, can we please stop this stupid behavior now? > > I have proposed a patch with a much more limited scope which is still > waiting for feedback. I haven't heard it wouldn't be working so far. > You mean mutex_lock_killable would take care of exiting task already. I would then still prefer to check for mark_oom_victim because that is not racy with the exit path clearing signals. I can update my patch to use _killable lock variant if we are really going with the memcg specific fix. ? No response for two months. One memcg OOM killer kills all processes in that memcg is broken. What is the race you are referring by "racy with the exit path clearing signals" ? You are saying that a thread between clearing fatal signal and setting PF_EXITING can invoke the memcg OOM killer again, aren't you? But how likely is that? Even if it can happen, your patch can call mark_oom_victim() even if my patch bailed out upon SIGKILL. That is, your patch and my patch are not conflicting/exclusive.
