Re: [PATCH v4 3/3] vfio/type1: optimize vfio_unpin_pages_remote() for large folio

[lizhe.67@bytedance.com]

On Wed, 18 Jun 2025 14:11:43 +0800, lizhe.67@bytedance.com wrote:
 
> > > How do you think of this implementation?
> > > 
> > > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > > index 242b05671502..eb91f99ea973 100644
> > > --- a/include/linux/mm.h
> > > +++ b/include/linux/mm.h
> > > @@ -2165,6 +2165,23 @@ static inline long folio_nr_pages(const struct folio *folio)
> > >          return folio_large_nr_pages(folio);
> > >   }
> > >   
> > > +/*
> > > + * folio_remaining_pages - Counts the number of pages from a given
> > > + * start page to the end of the folio.
> > > + *
> > > + * @folio: Pointer to folio
> > > + * @start_page: The starting page from which to begin counting.
> > > + *
> > > + * Returned number includes the provided start page.
> > > + *
> > > + * The caller must ensure that @start_page belongs to @folio.
> > > + */
> > > +static inline unsigned long folio_remaining_pages(struct folio *folio,
> > > +               struct page *start_page)
> > > +{
> > > +       return folio_nr_pages(folio) - folio_page_idx(folio, start_page);
> > > +}
> > > +
> > >   /* Only hugetlbfs can allocate folios larger than MAX_ORDER */
> > >   #ifdef CONFIG_ARCH_HAS_GIGANTIC_PAGE
> > >   #define MAX_FOLIO_NR_PAGES     (1UL << PUD_ORDER)
> > > diff --git a/mm/gup.c b/mm/gup.c
> > > index 15debead5f5b..14ae2e3088b4 100644
> > > --- a/mm/gup.c
> > > +++ b/mm/gup.c
> > > @@ -242,7 +242,7 @@ static inline struct folio *gup_folio_range_next(struct page *start,
> > >   
> > >          if (folio_test_large(folio))
> > >                  nr = min_t(unsigned int, npages - i,
> > > -                          folio_nr_pages(folio) - folio_page_idx(folio, next));
> > > +                          folio_remaining_pages(folio, next));
> > >   
> > >          *ntails = nr;
> > >          return folio;
> > > diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> > > index b2fc5266e3d2..34e85258060c 100644
> > > --- a/mm/page_isolation.c
> > > +++ b/mm/page_isolation.c
> > > @@ -96,7 +96,7 @@ static struct page *has_unmovable_pages(unsigned long start_pfn, unsigned long e
> > >                                  return page;
> > >                          }
> > >   
> > > -                       skip_pages = folio_nr_pages(folio) - folio_page_idx(folio, page);
> > > +                       skip_pages = folio_remaining_pages(folio, page);
> > >                          pfn += skip_pages - 1;
> > >                          continue;
> > >                  }
> > > ---
> > 
> > Guess I would have pulled the "min" in there, but passing something like 
> > ULONG_MAX for the page_isolation case also looks rather ugly.
> 
> Yes, the page_isolation case does not require the 'min' logic. Since
> there are already places in the current kernel code where
> folio_remaining_pages() is used without needing min, we could simply
> create a custom wrapper function based on folio_remaining_pages() only
> in those specific scenarios where min is necessary.
> 
> Following this line of thinking, the wrapper function in vfio would
> look something like this.
> 
> diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
> --- a/drivers/vfio/vfio_iommu_type1.c
> +++ b/drivers/vfio/vfio_iommu_type1.c
> @@ -801,16 +801,40 @@ static long vfio_pin_pages_remote(struct vfio_dma *dma, unsigned long vaddr,
>         return pinned;
>  }
>  
> +static inline unsigned long vfio_folio_remaining_pages(
> +               struct folio *folio, struct page *start_page,
> +               unsigned long max_pages)
> +{
> +       if (!folio_test_large(folio))
> +               return 1;

The above two lines may no longer be necessary.

> +       return min(max_pages,
> +                  folio_remaining_pages(folio, start_page));
> +}

Thanks,
Zhe