Re: [PATCH 10/10] mm, oom: hide mm which is shared with kthread or global init

From: Tetsuo Handa
Date: Thu Jun 16 2016 - 09:36:21 EST


Tetsuo Handa wrote:
> But this is not safe for CONFIG_MMU=y kernels as well.
> can_oom_reap == false means that oom_reap_task() will not be called.
> It is possible that the TIF_MEMDIE thread falls into
>
> atomic_read(&task->signal->oom_victims) > 0 && find_lock_task_mm(task) == NULL
>
> situation. We are still risking OOM livelock. We must somehow clear (or ignore)
> TIF_MEMDIE even if oom_reap_task() is not called.

Oops. mmput() from exit_mm() does not block in that case. So, we won't livelock here.

>
> Can't we apply http://lkml.kernel.org/r/201606102323.BCC73478.FtOJHFQMSVFLOO@xxxxxxxxxxxxxxxxxxx now?
>