[yocto] Compile issue with 1.7 SDK and hf

[Mark Hatle]

On 1/29/15 3:35 AM, Teresa Gámez wrote:
> Hello,
> 
> I have installed a SDK based on yocto 1.7 with
> DEFAULTTUNE ?="cortexa8thf-neon" set. 
> 
> I have tried to compile the barebox bootloader with this
> SDK like this:
> 
> source /opt/poky/1.7/environment-setup-cortexa8t2hf-vfp-neon-poky-linux-gnueabi
> unset LDFLAGS
> make am335x_defconfig
> Also enabled a barebox target tool in menuconfig: BAREBOXENV_TARGET 
> make

It looks like the tool you are trying to build is not using the full $CC,
otherwise the necessary ABI and related hardfloat parameters would be passed..
(alternatively if the component is adding it's own soft-float ABI values that
could also result in the failure.)

Generally the hard float toolchain does not have all of the code in place to
produce a soft-float binary as the executables have a different (incompatible) ABI.

> And see following compile error:
> 
> ..
>   CC      scripts/bareboxenv-target
> In file included
> from /opt/poky/1.7/sysroots/cortexa8t2hf-vfp-neon-poky-linux-gnueabi/usr/include/features.h:389:0,
> 
> from /opt/poky/1.7/sysroots/cortexa8t2hf-vfp-neon-poky-linux-gnueabi/usr/include/stdio.h:27,
>                  from scripts/bareboxenv.c:20:
> /opt/poky/1.7/sysroots/cortexa8t2hf-vfp-neon-poky-linux-gnueabi/usr/include/gnu/stubs.h:7:29: fatal error: gnu/stubs-soft.h: No such file or directory
>  # include <gnu/stubs-soft.h>
>                             ^
> compilation terminated.
> 
> I have tried the same with other non yocto hf toolchains. Compilation
> worked there. I wonder if this is a toolchain issue or more a barebox
> issue.

I suspect in this case it's barebox.  If you turn on more explicit (verbose)
logging you may see the error in the calls.

--Mark

> I also noticed:
> 
> arm-poky-linux-gnueabi-gcc -dM -E - < /dev/null | grep ARM_PCS
> #define __ARM_PCS 1
> 
> 
>  arm-poky-linux-gnueabi-gcc -mfloat-abi=hard -dM -E - < /dev/null | grep
> ARM_PCS
> #define __ARM_PCS_VFP 1
> 
> Other hf toolchains have __ARM_PCS_VFP 1 regardless if -mfloat-abi
> parameter set or not. I have no clue what the right behaviour should be.
> Can someone explain?
> 
> Regards
> Teresa
> 
> 
>