Re: [RFC] x86: merge nmi_32-64 to nmi.c

From: Cyrill Gorcunov
Date: Sun May 18 2008 - 06:10:20 EST


[Thomas Gleixner - Sun, May 18, 2008 at 12:34:15AM +0200]
| On Sat, 17 May 2008, Mikael Pettersson wrote:
| > Maciej W. Rozycki writes:
| > > On Sat, 17 May 2008, Tom Spink wrote:
| > >
| > > > static inline unsigned int get_nmi_count(int cpu)
| > > > {
| > > > #ifdef CONFIG_X86_64
| > > > return cpu_pda(cpu)->__nmi_count;
| > > > #else
| > > > return nmi_count(cpu);
| > > > #endif
| > > > }
| > > >
| > > > I know it introduces a lot of these conditionals, but at least there
| > > > is one place to look for the get_nmi_count function, instead of
| > > > searching for all variants of the function.
| > >
| > > Well, I suppose some header should provide a definition like:
| > >
| > > #ifdef CONFIG_X86_64
| > > #define cpu_x86_64 1
| > > #else
| > > #define cpu_x86_64 0
| > > #endif
| > >
| > > and the you can remove the horrible #ifdef clutter and make the quoted
| > > function look like:
| > >
| > > static inline unsigned int get_nmi_count(int cpu)
| > > {
| > > return cpu_x86_64 ? cpu_pda(cpu)->__nmi_count : nmi_count(cpu);
| > > }
| > >
| > > Much better -- isn't it?
| >
| > IMO, no, the #ifdef is preferable.
| >
| > Why? Because the #ifdef is a very visible signal to the platform
| > people that there are (in this case) subarch differences that force
| > "clients" to behave differently on different subarchs. By removing
| > the #ifdef you're IMO making it less likely for the platform people
| > to take notice and work towards eliminating those differences.
|
| The #ifdef is a poor choice. Maciej is damned right, that the single
| function with a clear distinction of the return value is better in
| terms of readability and maintenance.
|
| As I said before, We can make this more visible with an uppercase
| CONFIG_WHATEVER instaed of the innocent cpu_x86_64 one, but both
| solutions are better than #ifdefs and provide simple grepable
| patterns.
|
| The awareness of those differences does not depend at all on an
| #ifdef. Developers who are aware of the platform differences prefer a
| readable not ifdef poluted code base. People who need to be poked into
| the difference via an #ifdef are probably not those who can actually
| clean it up.
|
| Thanks,
|
| tglx
|

Btw Thomas, is there any reason why can't we add do_panic in die_nmi
on 32bit? And that would help us to add "panic" boot option on 32bits
as well.

- Cyrill -
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/