[RESEND PATCH] mm: hugetlb: Introduce huge_pte_{page,present,young}

[steve.capper@linaro.org (Steve Capper)]

On Mon, Mar 17, 2014 at 03:07:30PM -0700, Andrew Morton wrote:
> On Mon, 17 Mar 2014 18:51:58 +0000 Steve Capper <steve.capper@linaro.org> wrote:
> 
> > Introduce huge pte versions of pte_page, pte_present and pte_young.
> > This allows ARM (without LPAE) to use alternative pte processing logic
> > for huge ptes.
> > 
> > Where these functions are not defined by architectural code they
> > fallback to the standard functions.
> > 
> > Signed-off-by: Steve Capper <steve.capper@linaro.org>
> > ---
> > Hi,
> > I'm resending this patch to provoke some discussion.
> > 
> > We already have some huge_pte_ style functions, and this patch adds a
> > few more (that simplify to the pte_ equivalents where unspecified).
> > 
> > Having separate hugetlb versions of pte_page, present and mkyoung
> > allows for a greatly simplified huge page implementation for ARM with
> > the classical MMU (which has a different bit layout for huge ptes).
> 
> Looks OK to me.  One thing...
> 
> > --- a/include/linux/hugetlb.h
> > +++ b/include/linux/hugetlb.h
> > @@ -353,6 +353,18 @@ static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
> >  }
> >  #endif
> >  
> > +#ifndef huge_pte_page
> > +#define huge_pte_page(pte)	pte_page(pte)
> > +#endif
> 
> This #ifndef x #define x thing works well, but it is 100% unclear which
> arch header file is supposed to define x if it wishes to override the
> definition.  We've had problems with that in the past where different
> architectures put it in different files and various breakages ensued.
> 
> So can we decide which arch header file is responsible for defining
> these, then document that right here in a comment and add an explicit
> #include <asm/that-file.h>?

Thanks Andrew,
Yes I see your point, this could quickly become unstable.
I'll see how these look in include/asm-generic/hugetlb.h instead.

Cheers,
-- 
Steve