Re: [PATCH] mm/swap: Fix release_pages() when releasing devmap pages

[Dan Williams]

On Thu, May 23, 2019 at 8:58 PM Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
>
> On Thu, May 23, 2019 at 3:37 PM <ira.weiny@xxxxxxxxx> wrote:
> >
> > From: Ira Weiny <ira.weiny@xxxxxxxxx>
> >
> > Device pages can be more than type MEMORY_DEVICE_PUBLIC.
> >
> > Handle all device pages within release_pages()
> >
> > This was found via code inspection while determining if release_pages()
> > and the new put_user_pages() could be interchangeable.
> >
> > Cc: JÃrÃme Glisse <jglisse@xxxxxxxxxx>
> > Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
> > Cc: Michal Hocko <mhocko@xxxxxxxx>
> > Cc: John Hubbard <jhubbard@xxxxxxxxxx>
> > Signed-off-by: Ira Weiny <ira.weiny@xxxxxxxxx>
> > ---
> >  mm/swap.c | 7 +++----
> >  1 file changed, 3 insertions(+), 4 deletions(-)
> >
> > diff --git a/mm/swap.c b/mm/swap.c
> > index 3a75722e68a9..d1e8122568d0 100644
> > --- a/mm/swap.c
> > +++ b/mm/swap.c
> > @@ -739,15 +739,14 @@ void release_pages(struct page **pages, int nr)
> >                 if (is_huge_zero_page(page))
> >                         continue;
> >
> > -               /* Device public page can not be huge page */
> > -               if (is_device_public_page(page)) {
> > +               if (is_zone_device_page(page)) {
> >                         if (locked_pgdat) {
> >                                 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
> >                                                        flags);
> >                                 locked_pgdat = NULL;
> >                         }
> > -                       put_devmap_managed_page(page);
> > -                       continue;
> > +                       if (put_devmap_managed_page(page))
>
> This "shouldn't" fail, and if it does the code that follows might get
> confused by a ZONE_DEVICE page. If anything I would make this a
> WARN_ON_ONCE(!put_devmap_managed_page(page)), but always continue
> unconditionally.

As discussed offline, I'm wrong here. It needs to fall through to
put_page_testzero() for the device-dax case, but perhaps a comment for
the next time I forget that subtlety.