Re: [RFC PATCH v2 35/51] mm: guestmem_hugetlb: Add support for splitting and merging pages

[Ackerley Tng]

Michael Roth <michael.roth@xxxxxxx> writes:

> On Wed, May 14, 2025 at 04:42:14PM -0700, Ackerley Tng wrote:
>> 
>> [...snip...]
>> 
>
> Hi Ackerley,
>
> We've been doing some testing with this series on top of David's
> guestmemfd-preview branch with some SNP enablement[1][2] to exercise
> this code along with the NUMA support from Shivank (BTW, I know you
> have v3 in the works so let me know if we can help with testing that
> as well).
>

Thank you for offering! I'm quite backed up now with some internal
work. Will definitely appreciate all the help I can get once I manage to
push out an RFCv3!

> One issue we hit is if you do a split->merge sequence the unstash of the
> private data will result in folio_test_hugetlb_vmemmap_optimized() reporting
> true even though the hugetlb_vmemmap_optimize_folio() call hasn't been
> performed yet, and when that does get called it will be skipped, so some HVO
> optimization can be lost in this way.
>
> More troublesome however is if you later split the folio again,
> hugetlb_vmemmap_restore_folio() may cause a BUG_ON() since the flags are in a
> state that's not consistent with the state of the folio/vmemmap.
>
> The following patch seems to resolve the issue but I'm not sure what the
> best approach would be:
>
>   https://github.com/AMDESE/linux/commit/b1f25956f18d32730b8d4ded6d77e980091eb4d3
>

I saw your reply on the other thread. Thanks for informing me :)

> Thanks,
>
> Mike
>
> [1] https://github.com/AMDESE/linux/commits/snp-hugetlb-v2-wip0/
> [2] https://github.com/AMDESE/qemu/tree/snp-hugetlb-dev-wip0