On 8/13/14 14:51, H. Peter Anvin wrote:
> On 08/12/2014 10:32 PM, Chen Gang wrote:
>>
>> Yeah, we need.
>>
>>> The solution you suggest assumes that an arch is either little or big endian.
>>> But we cannot ignore the hybriads that can do both.
>>>
>>
>> For the architectures which can do both, for me, they are: sh, powerpc,
>> m32r, and mips (may mark CPU_LITTLE_ENDIAN), also are: arm/arm64, and
>> c6x (may mark !CPU_BIG_ENDIAN).
>>
>> For the architectures which only support little endian: x86, and ia64.
>>
>> The left, I assumes they are big endian (no any 'ENDIAN' can be found
>> in their Kconfig files). In my memory, except related with Intel, all
>> are (or support) big endian.
>>
>
> I'm quite certain that is wrong. From memory, for example, I believe
> CRIS is littleendian, and sure enough:
>
> : tazenda 103 ; less arch/cris/include/uapi/asm/byteorder.h
> #ifndef _CRIS_BYTEORDER_H
> #define _CRIS_BYTEORDER_H
>
> #include <linux/byteorder/little_endian.h>
>
> #endif
>
OK, what you said sounds reasonable to me. After search "endian" in
"./arch", xtensa also can be support both big endian or little endian.
> As far as I know, endianism is always a compile-time option on Linux, so
> we should be able to relatively easily define the architecture so that
> we either have hardwired littleendian, bigendian, or prompt.
>
Yeah, but that may break building and re-config manually. So it is still
necessary to check little endian or big endian during configuration time
("make *config").
Thanks.
--
Chen Gang
Open, share, and attitude like air, water, and life which God blessed
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
