Re: [PATCH 4/9] oom: oom_kill_process() need to check p isunkillable

[Minchan Kim]

On Mon, Jun 21, 2010 at 08:45:45PM +0900, KOSAKI Motohiro wrote:
> > On Thu, 17 Jun 2010, KOSAKI Motohiro wrote:
> > 
> > > When oom_kill_allocating_task is enabled, an argument task of
> > > oom_kill_process is not selected by select_bad_process(), It's
> > > just out_of_memory() caller task. It mean the task can be
> > > unkillable. check it first.
> > > 
> > 
> > This should be unnecessary if oom_kill_process() appropriately returns 
> > non-zero when it cannot kill a task.  What problem are you addressing with 
> > this fix?
> 
> oom_kill_process() only check its children are unkillable, not its own.
> To add check oom_kill_process() also solve the issue. as my previous patch does.
> but Minchan pointed out it's unnecessary. because when !oom_kill_allocating_task
> case, we have the same check in select_bad_process(). 
> 
> 
> 

If kthread doesn't use other process's mm, oom_kill_process can return non-zero.
and it might be no problem. 
but let's consider following case that kthread use use_mm. 

if (oom_kill_allocating_task)
        oom_kill_process
                pr_err("kill process.."); <-- false alarm
                oom_kill_task
                        find_lock_task_mm if kthread use use_mm
                        kill kernel thread

Yes. it's a just theory that kthread use use_mm and is selected as victim.
But although kthread doesn't use use_mm, oom_kill_process emits false alarm.
As a matter of fact, it doesn't kill itself or sacrifice child.

I think victim process selection should be done before calling 
oom_kill_process. oom_kill_process and oom_kill_task's role is  
just to try to kill the process or process's children by best effort 
as function's name.

-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/