Re: [PATCH v2 03/13] mm: Add generic p?d_large() macros

From: Mike Rapoport
Date: Fri Mar 01 2019 - 06:50:11 EST


Hi Kirill,

On Thu, Feb 21, 2019 at 05:57:06PM +0300, Kirill A. Shutemov wrote:
> On Thu, Feb 21, 2019 at 02:46:18PM +0000, Steven Price wrote:
> > On 21/02/2019 14:28, Kirill A. Shutemov wrote:
> > > On Thu, Feb 21, 2019 at 11:34:52AM +0000, Steven Price wrote:
> > >> From: James Morse <james.morse@xxxxxxx>
> > >>
> > >> Exposing the pud/pgd levels of the page tables to walk_page_range() means
> > >> we may come across the exotic large mappings that come with large areas
> > >> of contiguous memory (such as the kernel's linear map).
> > >>
> > >> For architectures that don't provide p?d_large() macros, provided a
> > >> does nothing default.
> > >
> > > Nak, sorry.
> > >
> > > Power will get broken by the patch. It has pmd_large() inline function,
> > > that will be overwritten by the define from this patch.
> > >
> > > I believe it requires more ground work on arch side in general.
> > > All architectures that has huge page support has to provide these helpers
> > > (and matching defines) before you can use it in a generic code.
> >
> > Sorry about that, I had compile tested on power, but obviously not the
> > right config to actually see the breakage.
>
> I don't think you'll catch it at compile-time. It would silently override
> the helper with always-false.

Can you explain why the compiler would override the helper define in, e.g.
arch/powerpc/include/asm/pgtable.h with the generic (0)?
Actually, I've tried to compile this on power with deliberately adding
errors to both power-specific and the generic definition of pmd_large and
the compilation failed the way I expected in the power-specific helper.

> > I'll do some grepping - hopefully this is just a case of exposing the
> > functions/defines that already exist for those architectures.
>
> I see the same type of breakage on s390 and sparc.
>
> > Note that in terms of the new page walking code, these new defines are
> > only used when walking a page table without a VMA (which isn't currently
> > done), so architectures which don't use p?d_large currently will work
> > fine with the generic versions. They only need to provide meaningful
> > definitions when switching to use the walk-without-a-VMA functionality.
>
> How other architectures would know that they need to provide the helpers
> to get walk-without-a-VMA functionality? This looks very fragile to me.
>
> --
> Kirill A. Shutemov

--
Sincerely yours,
Mike.