Hi Ralf,
2014-08-26 3:41 GMT+08:00 Ralf Baechle <ralf@xxxxxxxxxxxxxx>:
> On Mon, Aug 25, 2014 at 08:35:17PM +0100, Maciej W. Rozycki wrote:
>
>> On Mon, 25 Aug 2014, Ralf Baechle wrote:
>>
>> > > >> +cflags-$(CONFIG_CPU_LOONGSON3) += -D_MIPS_ARCH_LOONGSON3A
>> > > >
>> > > > The _MIPS_ARCH_* namespace belongs to GCC. While it seems current GCC
>> > > > does not define this symbol _MIPS_ARCH_LOONGSON3A runs into the danger
>> > > > of causing a conflict when GCC eventually will define the symbol.
>> > > When this symbol will be defined? With option '-march=loongson3a'?
>> >
>> > Well, not currently (at least not in my gcc 4.9.0) - but it might. In
>> > fact, I'm wondering why it doesn't. Maciej?
>>
>> No idea, a _MIPS_ARCH_foo macro gets defined automagically by GCC
>> whenever `-march=foo' is in effect (be it implicitly or with the use of a
>> command-line option), so there should be one.
>>
>> Has support for "loongson3a" been present in 4.9.x (it is in trunk)? If
>> so, then what _MIPS_ARCH_* macro gets defined for `-march=loongson3a'?
>
> Hmm - I must have fatfingered something. Now I'm getting:
>
> $ mips-linux-gcc < /dev/null -E -C -dM -march=loongson3a - | grep _MIPS_ARCH
> #define _MIPS_ARCH_LOONGSON3A 1
> #define _MIPS_ARCH "loongson3a"
> $
>
> So that would conflict with a manual definition, thus the patch would not
> be acceptable as it because:
>
> $ cat > c.c << EOF
> foo(){}
> EOF
> $ mips-linux-gcc -D_MIPS_ARCH_LOONGSON3A -march=loongson3a -Wall -c c.c
> c.c:1:1: warning: return type defaults to ‘int’ [-Wreturn-type]
> foo(){}
> ^
> c.c: In function ‘foo’:
> c.c:1:1: warning: control reaches end of non-void function [-Wreturn-type]
> foo(){}
> ^
> $
Thanks for the explanation. We will enable "-march=loongson3a" and do
some test internally to make sure doing so wouldn't cause any
problems.
To James:
> Any reason not to just refer directly to CONFIG_CPU_LOONGSON3 from the
> source rather than adding an intermediate definition?
>
> Cheers
> James
I guess it's because "arch/mips/include/uapi/asm/swab.h" is in "uapi/"
directory(http://lwn.net/Articles/507794/), that means it contains
"user-space API-related definitions", hence using CONFIG_CPU_LOONGSON3
is not suitable there.
