Re: [HMM 08/15] mm/ZONE_DEVICE: special case put_page() for device private pages

[Hugh Dickins]

On Mon, 22 May 2017, Jerome Glisse wrote:
> On Mon, May 22, 2017 at 12:29:53PM -0700, Dan Williams wrote:
> > On Mon, May 22, 2017 at 9:51 AM, Jerome Glisse <jglisse@xxxxxxxxxx> wrote:
> > > A ZONE_DEVICE page that reach a refcount of 1 is free ie no longer
> > > have any user. For device private pages this is important to catch
> > > and thus we need to special case put_page() for this.
> > >
> > > Signed-off-by: Jerome Glisse <jglisse@xxxxxxxxxx>
> > > Cc: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > > Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
> > > Cc: Ross Zwisler <ross.zwisler@xxxxxxxxxxxxxxx>
> > > ---
> > >  include/linux/mm.h | 30 ++++++++++++++++++++++++++++++
> > >  kernel/memremap.c  |  1 -
> > >  2 files changed, 30 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > > index a825dab..11f7bac 100644
> > > --- a/include/linux/mm.h
> > > +++ b/include/linux/mm.h
> > > @@ -23,6 +23,7 @@
> > >  #include <linux/page_ext.h>
> > >  #include <linux/err.h>
> > >  #include <linux/page_ref.h>
> > > +#include <linux/memremap.h>
> > >
> > >  struct mempolicy;
> > >  struct anon_vma;
> > > @@ -795,6 +796,20 @@ static inline bool is_device_private_page(const struct page *page)
> > >         return ((page_zonenum(page) == ZONE_DEVICE) &&
> > >                 (page->pgmap->type == MEMORY_DEVICE_PRIVATE));
> > >  }
> > > +
> > > +static inline void put_zone_device_private_page(struct page *page)
> > > +{
> > > +       int count = page_ref_dec_return(page);
> > > +
> > > +       /*
> > > +        * If refcount is 1 then page is freed and refcount is stable as nobody
> > > +        * holds a reference on the page.
> > > +        */
> > > +       if (count == 1)
> > > +               page->pgmap->page_free(page, page->pgmap->data);
> > > +       else if (!count)
> > > +               __put_page(page);
> > > +}

Is there something else in this patchset that guarantees
that get_page_unless_zero() is never used on thse pages?
We have plenty of code that knows that refcount 0 is special:
having to know that refcount 1 may be special is worrying.

Hugh

> > >  #else
> > >  static inline bool is_zone_device_page(const struct page *page)
> > >  {
> > > @@ -805,6 +820,10 @@ static inline bool is_device_private_page(const struct page *page)
> > >  {
> > >         return false;
> > >  }
> > > +
> > > +static inline void put_zone_device_private_page(struct page *page)
> > > +{
> > > +}
> > >  #endif
> > >
> > >  static inline void get_page(struct page *page)
> > > @@ -822,6 +841,17 @@ static inline void put_page(struct page *page)
> > >  {
> > >         page = compound_head(page);
> > >
> > > +       /*
> > > +        * For private device pages we need to catch refcount transition from
> > > +        * 2 to 1, when refcount reach one it means the private device page is
> > > +        * free and we need to inform the device driver through callback. See
> > > +        * include/linux/memremap.h and HMM for details.
> > > +        */
> > > +       if (unlikely(is_device_private_page(page))) {
> > 
> > Since I presume HMM is a niche use case can we make this a
> > "static_branch_unlikely(&hmm_key) && is_device_private_page(page))"?
> > That way non-hmm platforms see minimal overhead.
> 
> Like i said in the cover letter i am bit anxious about doing for
> an inline function. I don't see any existing case for inline
> function and static key. Is that suppose to work ?
> 
> How widespread HMM use will be is hard to guess. Usual chicken
> and egg plus adoption thing. If GPGPU compte keeps growing and
> it seems it does then HMM likely gonna be enable and actively
> use for large chunk of those computer that have GPGPU workload.
> 
> I will test a static key of that branch and see if it explodes
> because put_page() is an inline function.
> 
> Cheers,
> Jerome