[PATCH] drm/amdgpu: drop the long-double-128 powerpc check/hack
Christophe Leroy
christophe.leroy at csgroup.eu
Thu Dec 7 13:11:14 UTC 2023
Le 31/03/2023 à 12:53, Michael Ellerman a écrit :
> "Daniel Kolesa" <daniel at octaforge.org> writes:
>> Commit c653c591789b ("drm/amdgpu: Re-enable DCN for 64-bit powerpc")
>> introduced this check as a workaround for the driver not building
>> with toolchains that default to 64-bit long double.
> ...
>> In mainline, this work is now fully done, so this check is fully
>> redundant and does not do anything except preventing AMDGPU DC
>> from being built on systems such as those using musl libc. The
>> last piece of work to enable this was commit c92b7fe0d92a
>> ("drm/amd/display: move remaining FPU code to dml folder")
>> and this has since been backported to 6.1 stable (in 6.1.7).
>>
>> Relevant issue: https://gitlab.freedesktop.org/drm/amd/-/issues/2288
>
> I looked to pick this up for 6.3 but was still seeing build errors with
> some compilers. I assumed that was due to some fixes coming in
> linux-next that I didn't have.
>
> But applying the patch on v6.3-rc4 I still see build errors. This is
> building allyesconfig with the kernel.org GCC 12.2.0 / binutils 2.39
> toolchain:
>
> powerpc64le-linux-gnu-ld: drivers/gpu/drm/amd/display/dc/dml/display_mode_lib.o uses hard float, arch/powerpc/lib/test_emulate_step.o uses soft float
> powerpc64le-linux-gnu-ld: failed to merge target specific data of file drivers/gpu/drm/amd/display/dc/dml/display_mode_lib.o
>
> etc.
>
> All the conflicts are between test_emulate_step.o and some file in drivers/gpu/drm/amd/display/dc/dml.
>
> So even with all the hard-float code isolated in the dml folder, we
> still hit build errors, because allyesconfig wants to link those
> hard-float using objects with soft-float objects from elsewhere in the
> kernel.
>
> It seems like the only workable fix is to force the kernel build to use
> 128-bit long double. I'll send a patch doing that.
>
Commit 78f0929884d4 ("powerpc/64: Always build with 128-bit long
double") I guess ?
Let's drop this patch from patchwork then.
More information about the amd-gfx
mailing list