Re: [PATCH] mm: unified folio_test_anon()/folio_anon_vma() and use them

[Yajun Deng]

July 11, 2024 at 9:08 PM, "Matthew Wilcox" <willy@xxxxxxxxxxxxx> wrote:



> 
> On Thu, Jul 11, 2024 at 09:03:51PM +0800, Yajun Deng wrote:
> 
> > 
> > +++ b/include/linux/page-flags.h
> > 
> >  @@ -691,7 +691,8 @@ static __always_inline bool PageMappingFlags(const struct page *page)
> > 
> >  
> > 
> >  static __always_inline bool folio_test_anon(const struct folio *folio)
> > 
> >  {
> > 
> >  - return ((unsigned long)folio->mapping & PAGE_MAPPING_ANON) != 0;
> > 
> >  + return ((unsigned long)folio->mapping & PAGE_MAPPING_FLAGS) ==
> > 
> >  + PAGE_MAPPING_ANON;
> > 
> >  }
> > 
> 
> This is wrong. KSM pages are supposed to return true for
> 
> folio_test_anon(). I haven't looked any further at this patch, since
> 
> the premise appears to wrong and you clearly haven't tested.
>

Thank you for pointing this out. It was my mistake!

I think I need to add a new helper function, like __folio_test_movable. e.g.

static __always_inline bool __folio_test_anon(const struct folio *folio)
{
        return ((unsigned long)folio->mapping & PAGE_MAPPING_FLAGS) ==
                PAGE_MAPPING_ANON;
}