Re: [RFC PATCH] mm, oom: distinguish blockable mode for mmu notifiers

[Michal Hocko <mhocko-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

On Mon 02-07-18 14:39:50, Christian König wrote:
[...]
> Not wanting to block something as important as this, so feel free to add an
> Acked-by: Christian König <christian.koenig@amd.com> to the patch.

Thanks a lot!

> Let's rather face the next topic: Any idea how to runtime test this?

This is a good question indeed. One way to do that would be triggering
the OOM killer from the context which uses each of these mmu notifiers
(one at the time) and see how that works. You would see the note in the
log whenever the notifier would block. The primary thing to test is how
often the oom reaper really had to back off completely.

> I mean I can rather easily provide a test which crashes an AMD GPU, which in
> turn then would mean that the MMU notifier would block forever without this
> patch.

Well, you do not really have to go that far. It should be sufficient to
do the above. The current code would simply back of without releasing
any memory. The patch should help to reclaim some memory.
 
> But do you know a way to let the OOM killer kill a specific process?

Yes, you can set its oom_score_adj to 1000 which means always select
that task.
-- 
Michal Hocko
SUSE Labs
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx