Re: More OOM problems

[Michal Hocko <mhocko@suse.cz>]

On Sun 18-09-16 14:18:22, Linus Torvalds wrote:
> On Sun, Sep 18, 2016 at 2:00 PM, Vlastimil Babka <vbabka@suse.cz> wrote:
> >
> > Sounds like SLUB. SLAB would use order-0 as long as things fit. I would
> > hope for SLUB to fallback to order-0 (or order-1 for 8kB) instead of
> > OOM, though. Guess not...
> 
> SLUB it is - and I think that's pretty much all the world these days.
> SLAB is largely deprecated.

It seems that this is not a general consensus
http://lkml.kernel.org/r/20160823153807.GN23577@dhcp22.suse.cz

> We should probably start to remove SLAB entirely, and I definitely
> hope that no oom people run with it. SLUB is marked default in our
> config files, and I think most distros follow that (I know Fedora
> does, didn't check others).
> 
> > Well, order-3 is actually PAGE_ALLOC_COSTLY_ORDER, and costly orders
> > have to be strictly larger in all the tests. So order-3 is in fact still
> > considered "small", and thus it actually results in OOM instead of
> > allocation failure.
> 
> Yeah, but I do think that "oom when you have 156MB free and 7GB
> reclaimable, and haven't even tried swapping" counts as obviously
> wrong.

The thing is that swapping doesn't really help. You can easily migrate
anonymous memory to create larger blocks even without reclaiming them.
So I still believe compaction is giving up too easily.

> I'm not saying the code should fail and return NULL either, of course.
> 
> So  PAGE_ALLOC_COSTLY_ORDER should *not* mean "oom rather than return
> NULL". It really has to mean "try a _lot_ harder".

Agreed and Vlastimil's patches go that route. We just do not try
sufficiently hard with the compaction.
-- 
Michal Hocko
SUSE Labs

--
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>