Re: [PATCH 4/6] mm,oom: exclude oom_task_origin processes if they are OOM-unkillable.

[Michal Hocko]

On Wed 17-02-16 19:33:07, Tetsuo Handa wrote:
> >From 4924ca3031444bfb831b2d4f004e5a613ad48d68 Mon Sep 17 00:00:00 2001
> From: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> Date: Wed, 17 Feb 2016 16:35:12 +0900
> Subject: [PATCH 4/6] mm,oom: exclude oom_task_origin processes if they are OOM-unkillable.
> 
> oom_scan_process_thread() returns OOM_SCAN_SELECT when there is a
> thread which returns oom_task_origin() == true. But it is possible
> that that thread is marked as OOM-unkillable.
> 
> This patch changes oom_scan_process_thread() not to select it
> if it is marked as OOM-unkillable.

oom_task_origin is only swapoff and ksm_store right now. I seriously
doubt anybody sane will run them as OOM disabled (directly or
indirectly).

But you have a point that returing anything but OOM_SCAN_CONTINUE for
OOM_SCORE_ADJ_MIN from oom_scan_process_thread sounds suboptimal.
Sure such a check would be racy but do we actually care about a OOM vs.
oom_score_adj_write. I am dubious to say the least.

So wouldn't it make more sense to check for OOM_SCORE_ADJ_MIN at the
very top of oom_scan_process_thread instead?
 
> Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> ---
>  mm/oom_kill.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index b0c327d..ebc6764 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -308,7 +308,8 @@ enum oom_scan_t oom_scan_process_thread(struct oom_control *oc,
>  	 * If task is allocating a lot of memory and has been marked to be
>  	 * killed first if it triggers an oom, then select it.
>  	 */
> -	if (oom_task_origin(task) && !test_tsk_thread_flag(task, TIF_MEMDIE))
> +	if (oom_task_origin(task) && !test_tsk_thread_flag(task, TIF_MEMDIE) &&
> +	    task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN)
>  		return OOM_SCAN_SELECT;
>  
>  	return OOM_SCAN_OK;
> -- 
> 1.8.3.1

-- 
Michal Hocko
SUSE Labs