Re: mmotm 2009-02-02-17-12 uploaded (x86/nopmd etc.)

[Andrea Righi <righi.andrea@gmail.com>]

On Wed, Feb 4, 2009 at 10:25 PM, Andrew Morton
<akpm@linux-foundation.org> wrote:
> On Wed, 4 Feb 2009 21:08:46 +0100
> Ingo Molnar <mingo@elte.hu> wrote:
>
>>
>> * Andrew Morton <akpm@linux-foundation.org> wrote:
>>
>> > > > This is getting painful.
>> > >
>> > > the include file spaghetti is ... interesting there, and it's historic.
>> > >
>> > > I could blame it on highmem, PAE or paravirt - but i'll only blame it on
>> > > paravirt for now because those developers are still around! ;-)
>> > >
>> > > Jeremy, any ideas how to reduce the historic dependency mess in that area?
>> > > I think we should go on three routes at once:
>> > >
>> > >  - agressive splitup and separation of type definitions from method
>> > >    declaration (+ inline definitions). The spinlock_types.h / spinlock.h
>> > >    splitup was really nice in solving such dependency problems.
>> >
>> > I like this one.  The mixing up of declare-something with use-something
>> > is often the source of our woes.
>>
>> yes. I mapped this problem area once and this is how the include file
>> spaghetti gets generated in practice:
>>
>>   - type A gets declared
>>   - type A gets _used_ in the same file in an inline method, BUT,
>>
>>       that usage also brings in instantiated use of type X1, X2 and X3.
>>
>> if all types are declared like that everywhere, it can be seen (and it's a
>> mathematical certainty) that the only conflict-free way of doing this is to:
>>
>>   - initially add random #include lines to bring in type X1, X2 and X3.
>>     Which brings in recursive dependencies from those X1 X2 and X3 files.
>>
>>   - when the stuff hits the fan then folks are in a big mess already and
>>     only a deep restructuring could gets them out of it - which they rarely
>>     do in an iterative environment. So they work it around iteratively:
>>     instead of new nice inline methods [which we really prefer] they delay
>>     all the 'usage' instantiation to .c file via the use of CPP macros
>>     [which we hate because they hide bugs and cause bugs].
>
> None of which would happen if we didn't also have an inlining fetish.
>

So, about the pmd_* functions, why not move the nopmd implementations
in mm/memory.c (for example) always with the #ifndef's, and move the
specific arch implementations in arch/../mm/pgtable.c (for example)?

-Andrea