[OE-core] [PATCH v2] virglrenderer: use gold linker for x86 if cflag -fvisibility=default set

[Richard Purdie]

On Tue, 2019-04-30 at 10:23 -0400, kai.kang at windriver.com wrote:
> From: Kai Kang <kai.kang at windriver.com>
> 
> When gcc compile options '-fvisibility=default' is set in CFLAGS, it
> fails to compile virglrenderer for x86:
> 
> > i586-wrs-linux-libtool: link: i586-wrs-linux-gcc  -m32 -march=i586 ...
>   -Wl,--whole-archive ./.libs/libvrend.a
>   gallium/auxiliary/.libs/libgallium.a -Wl,--no-whole-archive ...
>   -Wl,-Bsymbolic -Wl,-O1 -Wl,--hash-style=gnu -Wl,--as-needed
>   -fstack-protector-strong -Wl,-z -Wl,relro -Wl,-z -Wl,now   -pthread
>   -Wl,-soname -Wl,libvirglrenderer.so.0 -o .libs/libvirglrenderer.so.0.2.0
> > ld: gallium/auxiliary/.libs/libgallium.a(u_cpu_detect.o): relocation
>   R_386_GOTOFF against undefined symbol `util_cpu_caps' can not be used
>   when making a shared object
> > ld: final link failed: bad value
> > collect2: error: ld returned 1 exit status
> 
> It seems GNU bfd linker doesn't work well to handle option '-Bsymbolic'
> and gcc option '-fvisibility=default'. Use gold instead under this
> circumstance.
> 
> Signed-off-by: Kai Kang <kai.kang at windriver.com>
> ---
>  meta/recipes-graphics/virglrenderer/virglrenderer_0.7.0.bb | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/meta/recipes-graphics/virglrenderer/virglrenderer_0.7.0.bb b/meta/recipes-graphics/virglrenderer/virglrenderer_0.7.0.bb
> index 225a0b8b0c..f72cfbd9b4 100644
> --- a/meta/recipes-graphics/virglrenderer/virglrenderer_0.7.0.bb
> +++ b/meta/recipes-graphics/virglrenderer/virglrenderer_0.7.0.bb
> @@ -15,6 +15,8 @@ S = "${WORKDIR}/git"
>  
>  inherit autotools pkgconfig distro_features_check
>  
> +EXTRA_OEMAKE_x86 = "${@bb.utils.contains('CFLAGS', '-fvisibility=default', 'GM_LDFLAGS+=-fuse-ld=gold', '', d)}"
> +
>  BBCLASSEXTEND = "native nativesdk"
>  
>  REQUIRED_DISTRO_FEATURES = "opengl"

Do we really want to start adding hacks to recipes to work around
linker bugs like this?

Has the bug been reported upstream? Why doesn't this happen on non-x86?

Cheers,

Richard