Re: [PATCH v2 5/8] mm/internal: rename swap offset helpers to softleaf offset

[Barry Song]

On Mon, Sep 7, 2026 at 1:38 PM Dev Jain <dev.jain@xxxxxxx> wrote:
>
>
>
> On 05/09/26 4:12 pm, Barry Song wrote:
> > On Tue, Sep 1, 2026 at 1:44 PM Dev Jain <dev.jain@xxxxxxx> wrote:
> >>
> >> In preparation for adding a helper to set softleaf ptes in one go,
> >> generalize the swap entry helpers shifting the swap offset by delta,
> >> for softleaves.
> >>
> >> Note that the soft-dirty bit, exclusive bit and uffd bit preservation
> >> will still work for non-swap softleaves, since a softleaf entry is
> >> constructed out of a type and offset, and those bits are ahead of
> >> the soft-dirty, exclusive and uffd bits.
> >>
> >> For example, for a migration entry, pte_swp_exclusive() will return
> >> false, as the exclusivity is encoded in the type itself
> >> (SOFTLEAF_MIGRATION_READ_EXCLUSIVE).
> >
> > I don't quite understand why you mention this. Is anyone calling
> > `pte_swp_exclusive()` on a migration entry? Shouldn't it only be called
> > when `softleaf_is_swap()` is true?
>
> You are right. I just wanted to emphasize the second paragraph - that the
> pte_move_swp_offset will also work for softleaf entries. But I think
> the third para confuses more, I'll drop it.

I would rather interpret this as meaning that
`pte_move_softleaf_offset` will also work for swap softleafs,
since `pte_move_softleaf_offset` will call some pure-swap
functions？

Would it be possible to use `if (softleaf_is_swap())` for
these cases to make the intent clearer? or we can
keep both pte_move_softleaf_offset and pte_move_swap_offset?

As `pte_move_softleaf_offset()` is getting a bit weird now,
a generic softleaf function has a lot of swap-specific code in it:

static inline pte_t pte_move_softleaf_offset(pte_t pte, long delta)
{
        const softleaf_t entry = softleaf_from_pte(pte);
        pte_t new = __swp_entry_to_pte(__swp_entry(swp_type(entry),
                                                   (swp_offset(entry)
+ delta)));

        if (pte_swp_soft_dirty(pte))
                new = pte_swp_mksoft_dirty(new);
        if (pte_swp_exclusive(pte))
                new = pte_swp_mkexclusive(new);
        if (pte_swp_uffd(pte))
                new = pte_swp_mkuffd(new);

        return new;
}

Am I missing something here?

>
> >
> >>
> >> Signed-off-by: Dev Jain <dev.jain@xxxxxxx>
> >
> > Reviewed-by: Barry Song <baohua@xxxxxxxxxx>
>
> Thanks.
>
>
> >
> >> ---
> >>  mm/internal.h | 29 +++++++++++++++--------------
> >>  mm/memory.c   |  4 ++--
> >>  2 files changed, 17 insertions(+), 16 deletions(-)
> >>
> > [...]
> >>
> >>  /**
> >> @@ -523,7 +524,7 @@ static inline pte_t pte_next_swp_offset(pte_t pte)
> >>   */
> >>  static inline int swap_pte_batch(pte_t *start_ptep, int max_nr, pte_t pte)
> >>  {
> >
> > We might find a user for this in the future, in which case we might
> > want to rename `swap_pte_batch()` to `swap_softleaf_batch()`?
>
> That is what is being done here:
> https://lore.kernel.org/all/20260813-migrate-rmap-batch-v2-1-3c5424c555c7@xxxxxxx/
>
> I don't have a strong opinion, I can also generalize this right now.

I don't know how you and Shivank are collaborating,
since you're both changing the same thing at the same time :-)

Hopefully, we can get one of yours into mm-new first,
and the other one can find a way to resolve the conflicts.

>
>
> >
> >> -       pte_t expected_pte = pte_next_swp_offset(pte);
> >> +       pte_t expected_pte = pte_next_softleaf_offset(pte);
> >

Best Regards
Barry