On Wed, Apr 05, 2023 at 05:17:31PM +0200, David Hildenbrand wrote:
> On 05.04.23 17:12, Peter Xu wrote:
> > On Wed, Apr 05, 2023 at 04:25:34PM +0200, David Hildenbrand wrote:
> > > Looks like what we fixed for hugetlb in commit 44f86392bdd1 ("mm/hugetlb:
> > > fix uffd-wp handling for migration entries in hugetlb_change_protection()")
> > > similarly applies to THP.
> > >
> > > Setting/clearing uffd-wp on THP migration entries is not implemented
> > > properly. Further, while removing migration PMDs considers the uffd-wp
> > > bit, inserting migration PMDs does not consider the uffd-wp bit.
> > >
> > > We have to set/clear independently of the migration entry type in
> > > change_huge_pmd() and properly copy the uffd-wp bit in
> > > set_pmd_migration_entry().
> > >
> > > Verified using a simple reproducer that triggers migration of a THP, that
> > > the set_pmd_migration_entry() no longer loses the uffd-wp bit.
> > >
> > > Fixes: f45ec5ff16a7 ("userfaultfd: wp: support swap and page migration")
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
> >
> > Reviewed-by: Peter Xu <peterx@xxxxxxxxxx>
> >
> > Thanks, one trivial nitpick:
> >
> > > ---
> > > mm/huge_memory.c | 14 ++++++++++++--
> > > 1 file changed, 12 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > > index 032fb0ef9cd1..bdda4f426d58 100644
> > > --- a/mm/huge_memory.c
> > > +++ b/mm/huge_memory.c
> > > @@ -1838,10 +1838,10 @@ int change_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
> > > if (is_swap_pmd(*pmd)) {
> > > swp_entry_t entry = pmd_to_swp_entry(*pmd);
> > > struct page *page = pfn_swap_entry_to_page(entry);
> > > + pmd_t newpmd;
> > > VM_BUG_ON(!is_pmd_migration_entry(*pmd));
> > > if (is_writable_migration_entry(entry)) {
> > > - pmd_t newpmd;
> > > /*
> > > * A protection check is difficult so
> > > * just be safe and disable write
> > > @@ -1855,8 +1855,16 @@ int change_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
> > > newpmd = pmd_swp_mksoft_dirty(newpmd);
> > > if (pmd_swp_uffd_wp(*pmd))
> > > newpmd = pmd_swp_mkuffd_wp(newpmd);
> > > - set_pmd_at(mm, addr, pmd, newpmd);
> > > + } else {
> > > + newpmd = *pmd;
> > > }
> > > +
> > > + if (uffd_wp)
> > > + newpmd = pmd_swp_mkuffd_wp(newpmd);
> > > + else if (uffd_wp_resolve)
> > > + newpmd = pmd_swp_clear_uffd_wp(newpmd);
> > > + if (!pmd_same(*pmd, newpmd))
> > > + set_pmd_at(mm, addr, pmd, newpmd);
> > > goto unlock;
> > > }
> > > #endif
> > > @@ -3251,6 +3259,8 @@ int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
> > > pmdswp = swp_entry_to_pmd(entry);
> > > if (pmd_soft_dirty(pmdval))
> > > pmdswp = pmd_swp_mksoft_dirty(pmdswp);
> > > + if (pmd_swp_uffd_wp(*pvmw->pmd))
> > > + pmdswp = pmd_swp_mkuffd_wp(pmdswp);
> >
> > I think it's fine to use *pmd, but maybe still better to use pmdval? I
> > worry pmdp_invalidate()) can be something else in the future that may
> > affect the bit.
>
> Wondering how I ended up with that, I realized that it's actually
> wrong and might have worked by chance for my reproducer on x86.
>
> That should make it work:
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index f977c965fdad..fffc953fa6ea 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3257,7 +3257,7 @@ int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
> pmdswp = swp_entry_to_pmd(entry);
> if (pmd_soft_dirty(pmdval))
> pmdswp = pmd_swp_mksoft_dirty(pmdswp);
> - if (pmd_swp_uffd_wp(*pvmw->pmd))
> + if (pmd_uffd_wp(pmdval))
> pmdswp = pmd_swp_mkuffd_wp(pmdswp);
> set_pmd_at(mm, address, pvmw->pmd, pmdswp);
> page_remove_rmap(page, vma, true);
I guess pmd_swp_uffd_wp() just reads the _USER bit 2 which is also set for
a present pte, but then it sets swp uffd-wp always even if it was not set.
Yes the change must be squashed in to be correct, with that, my R-b keeps.
Thanks,
--
Peter Xu
