lkml.org 
[lkml]   [2008]   [Oct]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
DateFri, 3 Oct 2008 13:21:43 -0400
FromMathieu Desnoyers <>
SubjectRe: [PATCH] x86 Topology cpu_to_node parameter check
* Steven Rostedt (rostedt@goodmis.org) wrote:
>
> On Fri, 3 Oct 2008, Mathieu Desnoyers wrote:
> >
> > That's how I did it first, but then I looked at asm-generic/topology.h
> > and have seen it uses #defines. Should we change them too ?
> >
>
> The old way of doing this is with defines. But all new code should be
> static inline functions when feasible. This way we can get typechecking
> on the parameters even when the configuration is disabled.
>
> Even if the rest of the file uses defines, the new code should be
> static inlines. Eventually, even the old defines will be converted.
>
> -- Steve
>

Argh, I think topology.h is utterly broken :-(

Have you noticed the subtile interaction between the

include/asm-x86/topology.h :

#define numa_node_id() 0
#define cpu_to_node(cpu) 0
#define early_cpu_to_node(cpu) 0
...
#include <asm-generic/topology.h>


and
include/asm-generic/topology.h :
#ifndef cpu_to_node
#define cpu_to_node(cpu) ((void)(cpu),0)
#endif

If any architecture decide for some reason to use a static inline rather
than a define, as currently done with node_to_first_cpu :

include/asm-x86/topology.h :
static inline int node_to_first_cpu(int node)
{
return first_cpu(cpu_online_map);
}
...
#include <asm-generic/topology.h>
include/asm-generic/topology.h :
#ifndef node_to_first_cpu
#define node_to_first_cpu(node) ((void)(node),0)
#endif

(which will override the static inline !)

It results in an override of the arch-specific version. Nice eh ?

Mathieu

--
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68


\
 
 \ /
  Last update: 2008-10-03 19:29    [from the cache]
©2003-2010