Fix sh4 build with __ASSUME_ST_INO_64_BIT redefinition

  Since 327792c sh4 builds fails with:

../sysdeps/unix/sysv/linux/kernel-features.h:49:0: error: "__ASSUME_ST_INO_64_BIT" redefined [-Werror]
 #define __ASSUME_ST_INO_64_BIT  1
 ^
In file included from ../sysdeps/unix/sysv/linux/sysdep.h:19:0,
                 from ../sysdeps/unix/sysv/linux/sh/sysdep.h:24,
                 from ../sysdeps/unix/sysv/linux/sh/sh4/sysdep.h:4,
                 from <stdin>:1:
../sysdeps/unix/sysv/linux/sh/kernel-features.h:47:0: note: this is the location of the previous definition
 #define __ASSUME_ST_INO_64_BIT 0

It is because sh4 kernel-features.sh is included multiple times
without guards and this patch fixes by adding them. Tested on a
sh4-linux-gnu build.

Also with this issues, is there are strong reason to *not* have include guards
on kernel-features.h?  With current approach, a architecture can't include
Linux default kernel-features.h and redefine it to a different value, only
undefine it (unless it explicit do not include default kernel-features.h,
and I think that's is not the idea).

	* sysdeps/unix/sysv/linux/sh/kernel-features.h: Add include
	guards.
---
 ChangeLog                                    | 5 +++++
 sysdeps/unix/sysv/linux/sh/kernel-features.h | 5 +++++
 2 files changed, 10 insertions(+)
  

On Wed, 9 Nov 2016, Adhemerval Zanella wrote:

> It is because sh4 kernel-features.sh is included multiple times
> without guards and this patch fixes by adding them. Tested on a
> sh4-linux-gnu build.
> 
> Also with this issues, is there are strong reason to *not* have include guards
> on kernel-features.h?  With current approach, a architecture can't include
> Linux default kernel-features.h and redefine it to a different value, only
> undefine it (unless it explicit do not include default kernel-features.h,
> and I think that's is not the idea).
> 
> 	* sysdeps/unix/sysv/linux/sh/kernel-features.h: Add include
> 	guards.

OK.  I think it's appropriate for all these headers to have include 
guards.

The #ifndef, #undef etc. could be avoided (moving this header closer to 
glibc's macro-name-typo-proof norms in this regard) if the header were 
split up into smaller pieces for individual macros or groups of related 
macros whenever a macro's definition needs to vary between architectures.  
Thus, you'd have e.g. kernel-features-st-ino.h defining 
__ASSUME_ST_INO_64_BIT to either 0 or 1, the default version of that 
header would define to 0 and the architectures overriding it would define 
to 1 (and wouldn't need #include_next).  kernel-features.h could end up as 
an architecture-independent header that just #includes the smaller 
architecture-specific headers.  (This is not a full design, especially for 
the more complicated cases such as socket syscalls.)
  

On 10/11/2016 16:07, Joseph Myers wrote:
> On Wed, 9 Nov 2016, Adhemerval Zanella wrote:
> 
>> It is because sh4 kernel-features.sh is included multiple times
>> without guards and this patch fixes by adding them. Tested on a
>> sh4-linux-gnu build.
>>
>> Also with this issues, is there are strong reason to *not* have include guards
>> on kernel-features.h?  With current approach, a architecture can't include
>> Linux default kernel-features.h and redefine it to a different value, only
>> undefine it (unless it explicit do not include default kernel-features.h,
>> and I think that's is not the idea).
>>
>> 	* sysdeps/unix/sysv/linux/sh/kernel-features.h: Add include
>> 	guards.
> 
> OK.  I think it's appropriate for all these headers to have include 
> guards.

Ok, I will push this patch.

> 
> The #ifndef, #undef etc. could be avoided (moving this header closer to 
> glibc's macro-name-typo-proof norms in this regard) if the header were 
> split up into smaller pieces for individual macros or groups of related 
> macros whenever a macro's definition needs to vary between architectures.  
> Thus, you'd have e.g. kernel-features-st-ino.h defining 
> __ASSUME_ST_INO_64_BIT to either 0 or 1, the default version of that 
> header would define to 0 and the architectures overriding it would define 
> to 1 (and wouldn't need #include_next).  kernel-features.h could end up as 
> an architecture-independent header that just #includes the smaller 
> architecture-specific headers.  (This is not a full design, especially for 
> the more complicated cases such as socket syscalls.)

This could be a feasible approach, although it would require some extensive
refactoring.  But I think we current approach, where architecture 
kernel-feature is used from sysdep and it is responsible to include default
linux one and undef and redefine value, plus guards should be also feasible
for macro-name-typo-proof as well.