Re: [PATCH 1/3] mm,oom: Move last second allocation to inside the OOM killer.

[Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>]

Michal Hocko wrote:
> On Tue 05-12-17 22:17:27, Tetsuo Handa wrote:
> > Michal Hocko wrote:
> > > > I do understand the upsides you're advocating for - although you
> > > > haven't quantified them. They're just not worth the downsides.
> > > 
> > > OK, fair enough. Let's drop the patch then. There is no _strong_
> > > justification for it and what I've seen as "nice to have" is indeed
> > > really hard to quantify and not really worth merging without a full
> > > consensus.
> > 
> > Dropping "mm,oom: move last second allocation to inside the OOM killer"
> > means dropping "mm,oom: remove oom_lock serialization from the OOM reaper"
> > together, right?
> 
> No, I believe that we can drop the lock even without this patch. This
> will need more investigation though.

We cannot drop the lock without this patch.

> 
> > The latter patch helped mitigating
> > schedule_timeout_killable(1) lockup problem though...
> > 
> > Also, what is the alternative for "mm,oom: use ALLOC_OOM for OOM victim's
> > last second allocation" ? I proposed "mm, oom: task_will_free_mem(current)
> > should ignore MMF_OOM_SKIP for once." and rejected by you. I also proposed
> > "mm,oom: Set ->signal->oom_mm to all thread groups sharing the victim's mm."
> > and rejected by you.
> 
> Yes, and so far I am not really sure we have to care all that much. I
> haven't seen any real world workload actually hitting this condition.
> 

Somebody will observe what Manish Jaggi observed. OOM with mlock()ed and/or
MAP_SHARED is irrelevant. There is always possibility that the OOM reaper
fails to reclaim memory due to mmap_sem contention (and results in extra
OOM kills).

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>