Re: [PATCH v6 1/2] mm: migration: fix the FOLL_GET failure on following huge page
From: Gerald Schaefer
Date: Thu Aug 18 2022 - 07:53:54 EST
On Tue, 16 Aug 2022 22:43:22 -0700
Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, 17 Aug 2022 03:31:37 +0000 "Wang, Haiyue" <haiyue.wang@xxxxxxxxx> wrote:
>
> > > > }
> > >
> > > I would be better to fix this for real at those three client code sites?
> >
> > Then 5.19 will break for a while to wait for the final BIG patch ?
>
> If that's the proposal then your [1/2] should have had a cc:stable and
> changelog words describing the plan for 6.0.
>
> But before we do that I'd like to see at least a prototype of the final
> fixes to s390 and hugetlb, so we can assess those as preferable for
> backporting. I don't think they'll be terribly intrusive or risky?
>
The private follow_huge_pud() for s390 is just some leftover, and the
only reason is / was that the generic version was using pte_page()
instead of pud_page(), which would not work for s390. See also commit
97534127012f ("mm/hugetlb: use pmd_page() in follow_huge_pmd()").
Since commit 3a194f3f8ad01 ("mm/hugetlb: make pud_huge() and
follow_huge_pud() aware of non-present pud entry") made
follow_huge_pud() behave similar to follow_huge_pmd(), in particular
also adding pud_page(), we can now switch to the generic version.
Note that we cannot support migration / hwpoison for hugetlb or THP,
because of different layout for PTE and PMD/PUD on s390. The generic
swp_entry functions all require proper PTEs, which wouldn't work on
PMD/PUD entries. In theory, at least for hugetlb, due to the "fake
PTE" conversion logic in huge_ptep_get(), we might be able to also
fake swp_entries, but the other problem is that we do not have enough
free bits in the PMD/PUD, so there probably will never be migration
support for huge pages on s390.
Anyway, that should not matter wrt to switching to the generic
follow_huge_pud(), because is_hugetlb_entry_migration() should always
return false, and no special change to pud_huge() check should be
needed like on x86.