Re: [PATCH 4/4] mm: numa: Do not mark PTEs pte_numa when splitting huge pages

[Sasha Levin <sasha.levin@oracle.com>]

On 10/02/2014 12:36 PM, Linus Torvalds wrote:
> On Thu, Oct 2, 2014 at 6:29 AM, Mel Gorman <mgorman@suse.de> wrote:
>> > This patch reverts 1ba6e0b50b ("mm: numa: split_huge_page: transfer the
>> > NUMA type from the pmd to the pte"). If a huge page is being split due
>> > a protection change and the tail will be in a PROT_NONE vma then NUMA
>> > hinting PTEs are temporarily created in the protected VMA.
> So this is the particular bug I was worried about when tracing through the code.
> 
> Should I just apply this as-is? And mark it for stable, since this has
> been around since 3.8 or so. It would seem to be a very safe change to
> do, regardless of whether this is actually the issue that Dave and
> maybe Sasha are seeing.
> 
> Sasha, I notice that you weren't on the cc for Mel's patches (probably
> because you got added later to the other thread), but they were all
> cc'd to lkml so you should see them there. Or I can forward them
> separately.

I grabbed them and will keep them in my tree for now instead of your
NUMA-chainsaw-massacre patch.

You've also mentioned that while I can tell you if nothing dies, I can't
really tell you if everything is working well. Is there a reasonable way
to easily say if NUMA is working properly? Even something that would just
tell me "your NUMA balancing seems to be sane" would be good.


Thanks,
Sasha