On Mon 25-07-16 20:47:03, Tetsuo Handa wrote:
> Michal Hocko wrote:
> > On Mon 25-07-16 20:07:11, Tetsuo Handa wrote:
> > > Michal Hocko wrote:
> > > > > Are you planning to change the scope where the OOM victims can access memory
> > > > > reserves?
> > > >
> > > > Yes. Because we know that there are some post exit_mm allocations and I
> > > > do not want to get back to PF_EXITING and other tricks...
> > > >
> > > > > (1) If you plan to allow the OOM victims to access memory reserves until
> > > > > TASK_DEAD, tsk_is_oom_victim() will be as trivial as
> > > > >
> > > > > bool tsk_is_oom_victim(struct task_struct *task)
> > > > > {
> > > > > return task->signal->oom_mm;
> > > > > }
> > > >
> > > > yes, exactly. That's what I've tried to say above. with the oom_mm this
> > > > is trivial to implement while mm lists will not help us much due to
> > > > their life time. This also means that we know about the oom victim until
> > > > it is unhashed and become invisible to the oom killer.
> > >
> > > Then, what are advantages with allowing only OOM victims access to memory
> > > reserves after they left exit_mm()?
> >
> > Because they might need it in order to move on... Say you want to close
> > all the files which might release considerable amount of memory or any
> > other post exit_mm() resources.
>
> OOM victims might need memory reserves in order to move on, but non OOM victims
> might also need memory reserves in order to move on. And non OOM victims might
> be blocking OOM victims via locks.
Yes that might be true but OOM situations are rare events and quite
reduced in the scope. Considering all exiting tasks is more dangerous
because they might deplete those memory reserves easily.
> > > Since we assume that mm_struct is the primary source of memory consumption,
> > > we don't select threads which already left exit_mm(). Since we assume that
> > > mm_struct is the primary source of memory consumption, why should we
> > > distinguish OOM victims and non OOM victims after they left exit_mm()?
> >
> > Because we might prevent from pointless OOM killer selection that way.
>
> That "might" sounds obscure to me.
>
> If currently allocating task is not an OOM victim then not giving it
> access to memory reserves will cause OOM victim selection.
Sure, that is true. I am talking about the case where the current victim
tries to get out and exit and it needs a memory for that.
> We might prevent from pointless OOM victim selection by giving
> killed/exiting tasks access to memory reserves.
This will open risks for other problems, I am afraid. Please note that
we are only trying to reduce the damage as much as possible. There is no
100% correct thing to do.
> > If we know that the currently allocating task is an OOM victim then
> > giving it access to memory reserves is preferable to selecting another
> > oom victim.
>
> If we know that the currently allocating task is killed/exiting then
> giving it access to memory reserves is preferable to selecting another
> OOM victim.
I believe this is getting getting off topic. Can we get back to mm list
vs signal::oom_mm decision? I have expressed one aspect that would speak
for oom_mm as it provides a persistent and easy to detect oom victim
which would be tricky with the mm list approach. Could you name some
arguments which would speak for the mm list and would be a problem with
the other approach?
--
Michal Hocko
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>