Re: Bug: Performance regression in 1013af4f585f: mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Mon, Oct 20, 2025 at 07:18:18PM +0200, David Hildenbrand wrote:
>
> > > >
> > > > So in case the page table got reused in the meantime, we should just
> > > > back off and be fine, right?
> > >
> > > The shared page table is mapped with a PUD entry, and we don't check
> > > whether the PUD entry changed here.
> >
> > Could we simply put a PUD check in there sensibly?
>
> A PUD check would only work if we are guaranteed that the page table will
> not get freed in the meantime, otherwise we might be walking garbage, trying
> to interpret garbage as PMDs etc.
>
> That would require RCU freeing of page tables, which we are not guaranteed
> to have IIRC.

Ack. Yeah Suren is working on this :) but don't know when that'll land.

>
> The easiest approach is probably to simply never reuse shared page tables.
>
> If there is consensus on that I can try to see if I can make it fly easily.

That'd be good, I'd like to see that if viable for you to put something
forward :)

>
> --
> Cheers
>
> David / dhildenb
>

Cheers, Lorenzo