[musl] ./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative on PowerPC

[musl] ./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative on PowerPC

[Thomas Petazzoni]

Hello,

We've "recently" started in Buildroot the build testing of fully random
configurations, so we're now testing many more cases and seeing new
build failures.

As part of this, we're now seeing build failures of musl on PowerPC
32-bit architectures. So far, we've seen the problem on power4
(-mcpu=power4), power5 (-mcpu=power5) and power6 (-mcpu=power6), 970
(-mcpu=970), power8 (-mcpu=power8), 620 (-mcpu=620), 630 (-mcpu=630).

The failure is:

./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative
  229 | typedef char a_cas_p_undefined_but_pointer_not_32bit[-sizeof(char) == 0xffffffff ? 1 : -1];
      |              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

You can see the full musl build log at:

  http://autobuild.buildroot.net/results/3f1956fbc1746ac25b537ed023e6393ff715f4c9/build-end.log (power4)
  http://autobuild.buildroot.net/results/eeb3d66480e6fb4cd031e7803859480b6d0c72f3/build-end.log (power5)
  http://autobuild.buildroot.net/results/4e41d2ce80ba56f933b1a4ac2dec367d634b0b54/build-end.log (power6)
  http://autobuild.buildroot.net/results/5c7/5c7f71852b8ad201c0615838b7fa6f8bb15cc55b/build-end.log (970)
  http://autobuild.buildroot.net/results/bac/bac4489b489f9b62b5f00be6a3aaac2db90e4ca1/build-end.log (power8)
  http://autobuild.buildroot.net/results/97c/97c6c4bd683e6c7022a62343888dfa8b9d3bab04/build-end.log (620)
  http://autobuild.buildroot.net/results/e2e/e2ed03711a8f2ae7b80efd56114457f8476449cc/build-end.log (630)

I have not investigated the problem so far. I'm not super familiar with
PowerPC, but it seems like these cores are all 64-bit capable, but here
we build for them in 32-bit mode. Maybe this is the case that isn't
handled or supported by musl?

Thanks a lot for your support!

Best regards,

Thomas
-- 
Thomas Petazzoni, co-owner and CEO, Bootlin
Embedded Linux and Kernel engineering and training
https://bootlin.com

Re: [musl] ./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative on PowerPC

[Szabolcs Nagy]

* Thomas Petazzoni <thomas.petazzoni@bootlin.com> [2022-07-22 09:12:24 +0200]:
> Hello,
> 
> We've "recently" started in Buildroot the build testing of fully random
> configurations, so we're now testing many more cases and seeing new
> build failures.
> 
> As part of this, we're now seeing build failures of musl on PowerPC
> 32-bit architectures. So far, we've seen the problem on power4
> (-mcpu=power4), power5 (-mcpu=power5) and power6 (-mcpu=power6), 970
> (-mcpu=970), power8 (-mcpu=power8), 620 (-mcpu=620), 630 (-mcpu=630).
> 
> The failure is:
> 
> ./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative
>   229 | typedef char a_cas_p_undefined_but_pointer_not_32bit[-sizeof(char) == 0xffffffff ? 1 : -1];
>       |              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

the compiler is not configured for 32-bit target abi, but the triplet
(powerpc-buildroot-linux-musl) selects the 32-bit target in musl.

please use powerpc64* --target triplet when configuring musl or configure
gcc to default to 32-bit abi if that's what you want.

(the bug would be more obvious looking at gcc -v)

> 
> You can see the full musl build log at:
> 
>   http://autobuild.buildroot.net/results/3f1956fbc1746ac25b537ed023e6393ff715f4c9/build-end.log (power4)
>   http://autobuild.buildroot.net/results/eeb3d66480e6fb4cd031e7803859480b6d0c72f3/build-end.log (power5)
>   http://autobuild.buildroot.net/results/4e41d2ce80ba56f933b1a4ac2dec367d634b0b54/build-end.log (power6)
>   http://autobuild.buildroot.net/results/5c7/5c7f71852b8ad201c0615838b7fa6f8bb15cc55b/build-end.log (970)
>   http://autobuild.buildroot.net/results/bac/bac4489b489f9b62b5f00be6a3aaac2db90e4ca1/build-end.log (power8)
>   http://autobuild.buildroot.net/results/97c/97c6c4bd683e6c7022a62343888dfa8b9d3bab04/build-end.log (620)
>   http://autobuild.buildroot.net/results/e2e/e2ed03711a8f2ae7b80efd56114457f8476449cc/build-end.log (630)
> 
> I have not investigated the problem so far. I'm not super familiar with
> PowerPC, but it seems like these cores are all 64-bit capable, but here
> we build for them in 32-bit mode. Maybe this is the case that isn't
> handled or supported by musl?
> 
> Thanks a lot for your support!
> 
> Best regards,
> 
> Thomas
> -- 
> Thomas Petazzoni, co-owner and CEO, Bootlin
> Embedded Linux and Kernel engineering and training
> https://bootlin.com

Re: [musl] ./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative on PowerPC

[Thomas Petazzoni]

Hello,

On Fri, 22 Jul 2022 18:29:00 +0200
Szabolcs Nagy <nsz@port70.net> wrote:

> * Thomas Petazzoni <thomas.petazzoni@bootlin.com> [2022-07-22 09:12:24 +0200]:
> > Hello,
> > 
> > We've "recently" started in Buildroot the build testing of fully random
> > configurations, so we're now testing many more cases and seeing new
> > build failures.
> > 
> > As part of this, we're now seeing build failures of musl on PowerPC
> > 32-bit architectures. So far, we've seen the problem on power4
> > (-mcpu=power4), power5 (-mcpu=power5) and power6 (-mcpu=power6), 970
> > (-mcpu=970), power8 (-mcpu=power8), 620 (-mcpu=620), 630 (-mcpu=630).
> > 
> > The failure is:
> > 
> > ./src/internal/atomic.h:229:14: error: size of array 'a_cas_p_undefined_but_pointer_not_32bit' is negative
> >   229 | typedef char a_cas_p_undefined_but_pointer_not_32bit[-sizeof(char) == 0xffffffff ? 1 : -1];
> >       |              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
> 
> the compiler is not configured for 32-bit target abi, but the triplet
> (powerpc-buildroot-linux-musl) selects the 32-bit target in musl.
> 
> please use powerpc64* --target triplet when configuring musl or configure
> gcc to default to 32-bit abi if that's what you want.
> 
> (the bug would be more obvious looking at gcc -v)

Thanks a lot for the feedback. I have added in Cc Joel and Cédric, who
help maintain the PowerPC support in Buildroot. Joel has submitted just
today a patch series to Buildroot that I believe should help address
this particular issue. Joel, do you confirm?

Thanks a lot!

Thomas
-- 
Thomas Petazzoni, co-owner and CEO, Bootlin
Embedded Linux and Kernel engineering and training
https://bootlin.com