Re: [patch -mmotm] oom: only oom kill exiting tasks with attached memory

[Minchan Kim <minchan.kim@gmail.com>]

Hi, David.

On Mon, Jun 8, 2009 at 8:18 AM, David Rientjes<rientjes@google.com> wrote:
> On Sun, 7 Jun 2009, Balbir Singh wrote:
>
>> Seems like a good change to make. Did you encounter this situation on
>> a real machine?
>>
>
> I did, as the result of the chosen task remaining in PF_EXITING state with
> a detached mm following a previous oom kill.
> --

Let me have a question.
If I understand your situation properly, you mean

Time order
t1 < t2 < t3 < t4.

t1 : oom kill A process - send signal. it doesn't destroy mm_struct yet.
t2 : destroy A's mm_struct but it is in task list.
t3 : It happens OOM, again. Process A is selected again since it still
remain in task list
t4 : Now A's mm_struct is destroyed => situation you said.

Is right ?
Do you turn on sysctl oom_kill_allocating_task ?

-- 
Kinds regards,
Minchan Kim