Michal Hocko wrote: > Hmm, so you are creating a separate process (from the signal point of > view) and I suspect it is one of those that holds the last reference to > the mm_struct which is released here and it has tsk_oom_victim = F
Right. > So we need a more robust test for the oom victim. Your suggestion is > basically what I came up with originally [1] and which was deemed > ineffective because we took the mmap_sem even for regular paths and > Kirill was afraid this adds some unnecessary cycles to the exit path > which is quite hot. > > So I guess we have to do something else instead. We have to store the > oom flag to the mm struct as well. Something like the patch below. Yes, adding a new flag for this purpose will work. Also, setting MMF_UNSTABLE flag between after sending SIGKILL and before victim->mm becomes NULL and testing MMF_UNSTABLE at exit_mm() should work. But I prefer simple revert + mmget()/mmput_async() approach at http://lkml.kernel.org/r/201712062037.daf90168.svfqojfmoot...@i-love.sakura.ne.jp , for my approach not only saves lines but also fixes unexpected change for nommu at http://lkml.kernel.org/r/201711091949.bdb73475.oshfomqtlfo...@i-love.sakura.ne.jp . Also, if we replace asynchronous OOM reaping by the OOM reaper kernel thread with synchronous OOM reaping by the OOM killer, we can close MMF_OOM_SKIP race window because it is guaranteed that __oom_reap_task_mm() is called before __mmput() is called.
