[PATCH RFC] MIPS: Make UTS_MACHINE reflect big-endian/little-endian

[PATCH RFC] MIPS: Make UTS_MACHINE reflect big-endian/little-endian

[Huacai Chen]

It seems like some application use "uname" command's output to do
something different between big-endian and little-endian, so we make
UTS_MACHINE reflect both 32bit/64bit and big-endian/little-endian.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/mips/Makefile b/arch/mips/Makefile
index e2122cc..a21c3a1 100644
--- a/arch/mips/Makefile
+++ b/arch/mips/Makefile
@@ -38,11 +38,11 @@ endif
 
 ifdef CONFIG_32BIT
 tool-archpref		= $(32bit-tool-archpref)
-UTS_MACHINE		:= mips
+UTS_MACHINE		:= $(32bit-tool-archpref)
 endif
 ifdef CONFIG_64BIT
 tool-archpref		= $(64bit-tool-archpref)
-UTS_MACHINE		:= mips64
+UTS_MACHINE		:= $(64bit-tool-archpref)
 endif
 
 ifneq ($(SUBARCH),$(ARCH))
-- 
2.7.0

Re: [PATCH RFC] MIPS: Make UTS_MACHINE reflect big-endian/little-endian

[Paul Burton]

Hi Huacai,

On Wed, Jul 11, 2018 at 04:28:17PM +0800, Huacai Chen wrote:
> It seems like some application use "uname" command's output to do
> something different between big-endian and little-endian, so we make
> UTS_MACHINE reflect both 32bit/64bit and big-endian/little-endian.

Since UTS_MACHINE is exposed to userland (which is of course the point
of your patch) I'm not comfortable changing it. If some piece of code
checks whether uname -m gives "mips" then we'd break it by suddenly
giving "mipsel". This is too risky.

Which applications are you talking about that look for endianness in
uname output? Since Linux on MIPS doesn't expose endianness information
in this way these applications will always have been broken on MIPS
systems, and it would make more sense to fix the applications than to
change the kernel & probably break others.

Thanks,
    Paul

> diff --git a/arch/mips/Makefile b/arch/mips/Makefile
> index e2122cc..a21c3a1 100644
> --- a/arch/mips/Makefile
> +++ b/arch/mips/Makefile
> @@ -38,11 +38,11 @@ endif
>  
>  ifdef CONFIG_32BIT
>  tool-archpref		= $(32bit-tool-archpref)
> -UTS_MACHINE		:= mips
> +UTS_MACHINE		:= $(32bit-tool-archpref)
>  endif
>  ifdef CONFIG_64BIT
>  tool-archpref		= $(64bit-tool-archpref)
> -UTS_MACHINE		:= mips64
> +UTS_MACHINE		:= $(64bit-tool-archpref)
>  endif

Re: [PATCH RFC] MIPS: Make UTS_MACHINE reflect big-endian/little-endian

[Huacai Chen]

On Wed, Jul 11, 2018 at 11:57 PM, Paul Burton <paul.burton@mips.com> wrote:
> Hi Huacai,
>
> On Wed, Jul 11, 2018 at 04:28:17PM +0800, Huacai Chen wrote:
>> It seems like some application use "uname" command's output to do
>> something different between big-endian and little-endian, so we make
>> UTS_MACHINE reflect both 32bit/64bit and big-endian/little-endian.
>
> Since UTS_MACHINE is exposed to userland (which is of course the point
> of your patch) I'm not comfortable changing it. If some piece of code
> checks whether uname -m gives "mips" then we'd break it by suddenly
> giving "mipsel". This is too risky.
>
> Which applications are you talking about that look for endianness in
> uname output? Since Linux on MIPS doesn't expose endianness information
> in this way these applications will always have been broken on MIPS
> systems, and it would make more sense to fix the applications than to
> change the kernel & probably break others.
Hi, we are making Fedora on MIPS, we found that yum use uname's output
to adjust endianness compatibility. We haven't seen anything broken
while we changing mips64 to mips64el, do you know something about
"broken things with this changing". I also don't want to change
UTS_MACHINE, but if it is harmless, I think we can just do it.

>
> Thanks,
>     Paul
>
>> diff --git a/arch/mips/Makefile b/arch/mips/Makefile
>> index e2122cc..a21c3a1 100644
>> --- a/arch/mips/Makefile
>> +++ b/arch/mips/Makefile
>> @@ -38,11 +38,11 @@ endif
>>
>>  ifdef CONFIG_32BIT
>>  tool-archpref                = $(32bit-tool-archpref)
>> -UTS_MACHINE          := mips
>> +UTS_MACHINE          := $(32bit-tool-archpref)
>>  endif
>>  ifdef CONFIG_64BIT
>>  tool-archpref                = $(64bit-tool-archpref)
>> -UTS_MACHINE          := mips64
>> +UTS_MACHINE          := $(64bit-tool-archpref)
>>  endif

Re: [PATCH RFC] MIPS: Make UTS_MACHINE reflect big-endian/little-endian

[Paul Burton]

Hi Huacai,

On Thu, Jul 12, 2018 at 07:09:04AM +0800, Huacai Chen wrote:
> >> It seems like some application use "uname" command's output to do
> >> something different between big-endian and little-endian, so we make
> >> UTS_MACHINE reflect both 32bit/64bit and big-endian/little-endian.
> >
> > Since UTS_MACHINE is exposed to userland (which is of course the point
> > of your patch) I'm not comfortable changing it. If some piece of code
> > checks whether uname -m gives "mips" then we'd break it by suddenly
> > giving "mipsel". This is too risky.
> >
> > Which applications are you talking about that look for endianness in
> > uname output? Since Linux on MIPS doesn't expose endianness information
> > in this way these applications will always have been broken on MIPS
> > systems, and it would make more sense to fix the applications than to
> > change the kernel & probably break others.
>
> Hi, we are making Fedora on MIPS, we found that yum use uname's output
> to adjust endianness compatibility. We haven't seen anything broken
> while we changing mips64 to mips64el, do you know something about
> "broken things with this changing". I also don't want to change
> UTS_MACHINE, but if it is harmless, I think we can just do it.

A quick GitHub code search find a ton of things which check uname -m,
some of which expect the exact string "mips":

    https://github.com/search?l=&p=1&q=uname+-m+mips&type=Code

UTS_MACHINE is not something to change.

Can you just have yum detect endianness some other way? Wikipedia tells
me yum is written in python, and a quick Google tells me there's a
sys.byteorder variable which would indicate endianness. Can you just use
that?

Thanks,
    Paul