Re: [PATCH v2] khugepaged: Reduce race probability between migration and khugepaged

[David Hildenbrand]

On 03.07.25 11:52, Dev Jain wrote:
> On 03/07/25 2:55 pm, David Hildenbrand wrote:
> > On 03.07.25 07:48, Dev Jain wrote:
> > > Suppose a folio is under migration, and khugepaged is also trying to
> > > collapse it. collapse_pte_mapped_thp() will retrieve the folio from the
> > > page cache via filemap_lock_folio(), thus taking a reference on the
> > > folio
> > > and sleeping on the folio lock, since the lock is held by the migration
> > > path. Migration will then fail in
> > > __folio_migrate_mapping -> folio_ref_freeze. Reduce the probability of
> > > such a race happening (leading to migration failure) by bailing out
> > > if we detect a PMD is marked with a migration entry.
> > >
> > > This fixes the migration-shared-anon-thp testcase failure on Apple M3.
> > >
> > > Note that, this is not a "fix" since it only reduces the chance of
> > > interference of khugepaged with migration, wherein both the kernel
> > > functionalities are deemed "best-effort".
> > >
> > > Signed-off-by: Dev Jain <dev.jain@xxxxxxx>
> > > ---
> > >
> > > v1->v2:
> > >    - Remove SCAN_PMD_MIGRATION, merge into SCAN_PMD_MAPPED (David,
> > > Anshuman)
> > >    - Add a comment (Lorenzo)
> > >
> > > v1:
> > >    - https://lore.kernel.org/all/20250630044837.4675-1-dev.jain@xxxxxxx/
> > >
> > >    mm/khugepaged.c | 9 +++++++++
> > >    1 file changed, 9 insertions(+)
> > >
> > > diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> > > index 1aa7ca67c756..3fdefc4f4984 100644
> > > --- a/mm/khugepaged.c
> > > +++ b/mm/khugepaged.c
> > > @@ -941,6 +941,15 @@ static inline int check_pmd_state(pmd_t *pmd)
> > >          if (pmd_none(pmde))
> > >            return SCAN_PMD_NONE;
> > > +
> > > +    /*
> > > +     * The folio may be under migration when khugepaged is trying to
> > > +     * collapse it. Migration success or failure will eventually end
> > > +     * up with the PMD still pointing to a PMD-order folio, so return
> > > +     * SCAN_PMD_MAPPED.
> >
> > Nit: the last part (, so return ..) is obvious from the code.
> >
> > I would have written
> >
> > /*
> >   * The folio may be under migration when khugepaged is trying to
> >   * collapse it. Migration success or failure will eventually end
> >   * up with a present PMD entry again.
> >   */
> Thanks for the suggestion, but
>
> PMD pointing to PMD-order folio necessarily implies present PMD entry,
> > but the converse is not true? For example it may point to a PTE table.

I see, talking about orders is confusing though.

"with a present PMD mapping a folio again."

--
Cheers,

David / dhildenb