[PATCH]: highest_possible_processor_id() has to be a macro
Al Viro [Sun, 16 Oct 2005 07:17:33 +0000 (00:17 -0700)]
... otherwise, things like alpha and sparc64 break and break
badly.  They define cpu_possible_map to something else in smp.h
*AFTER* having included cpumask.h.

If that puppy is a macro, expansion will happen at the actual
caller, when we'd already seen #define cpu_possible_map ... and we will
get the right thing used.

As an inline helper it will be tokenized before we get to that
define and that's it; no matter what we define later, it won't affect
anything.  We get modules with dependency on cpu_possible_map instead
of the right symbol (phys_cpu_present_map in case of sparc64), or outright
link errors if they are built-in.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/cpumask.h

index fe97783..9bdba81 100644 (file)
@@ -393,15 +393,13 @@ extern cpumask_t cpu_present_map;
 #define for_each_present_cpu(cpu) for_each_cpu_mask((cpu), cpu_present_map)
 
 /* Find the highest possible smp_processor_id() */
-static inline unsigned int highest_possible_processor_id(void)
-{
-       unsigned int cpu, highest = 0;
-
-       for_each_cpu_mask(cpu, cpu_possible_map)
-               highest = cpu;
-
-       return highest;
-}
+#define highest_possible_processor_id() \
+({ \
+       unsigned int cpu, highest = 0; \
+       for_each_cpu_mask(cpu, cpu_possible_map) \
+               highest = cpu; \
+       highest; \
+})
 
 
 #endif /* __LINUX_CPUMASK_H */