[Adeos-main] fastcall and asmlinkage in x86

[Adeos-main] fastcall and asmlinkage in x86

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 574 bytes --]

Hi Philippe,

most of the code in arch/x86/ipipe_*.c can be pushed into some
ipipe-common.c, reducing the patch size by 10-20k and improving code
maintainability.

But one thing is currently blocking a copy&paste move: a lot of
functions that are declared fastcall on 32 bit are asmlinkage on 64 bit.
As both asmlinkage and fastcall are meaningless on 64 bit, can we safely
switch to fastcall? Or even better, could we drop that redundant
declarations totally, given that everything is fastcall on 32 bit anyway
(unless its marked asmlinkage)?

Thanks,
Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 254 bytes --]

Re: [Adeos-main] fastcall and asmlinkage in x86

[Philippe Gerum]

Jan Kiszka wrote:
> Hi Philippe,
> 
> most of the code in arch/x86/ipipe_*.c can be pushed into some
> ipipe-common.c, reducing the patch size by 10-20k and improving code
> maintainability.
>
> But one thing is currently blocking a copy&paste move: a lot of
> functions that are declared fastcall on 32 bit are asmlinkage on 64 bit.

2.4/x86 legacy that survived.

> As both asmlinkage and fastcall are meaningless on 64 bit, can we safely
> switch to fastcall?

Yes.

 Or even better, could we drop that redundant
> declarations totally, given that everything is fastcall on 32 bit anyway
> (unless its marked asmlinkage)?
> 

I see -mregparm=3 unconditionally set in arch/x86/Makefile_32, so this
should work. Looking at some assembly call sites (exception
handling/diversion) tends to confirm this; AFAIR we should not have any
regparm(0) code anywhere. Well, if we still do have some, you will be
the first to know...

-- 
Philippe.