Hamza Mahfooz <hamza.mahfooz@xxxxxxx> writes:
> On 4/4/23 06:28, Michael Ellerman wrote:
>> The amdgpu driver builds some of its code with hard-float enabled,
>> whereas the rest of the kernel is built with soft-float.
>>
>> When building with 64-bit long double, if soft-float and hard-float
>> objects are linked together, the build fails due to incompatible ABI
>> tags.
>>
>> In the past there have been build errors in the amdgpu driver caused by
>> this, some of those were due to bad intermingling of soft & hard-float
>> code, but those issues have now all been fixed since commit c92b7fe0d92a
>> ("drm/amd/display: move remaining FPU code to dml folder").
>>
>> However it's still possible for soft & hard-float objects to end up
>> linked together, if the amdgpu driver is built-in to the kernel along
>> with the test_emulate_step.c code, which uses soft-float. That happens
>> in an allyesconfig build.
>>
>> Currently those build errors are avoided because the amdgpu driver is
>> gated on 128-bit long double being enabled. But that's not a detail the
>> amdgpu driver should need to be aware of, and if another driver starts
>> using hard-float the same problem would occur.
>>
>> All versions of the 64-bit ABI specify that long-double is 128-bits.
>> However some compilers, notably the kernel.org ones, are built to use
>> 64-bit long double by default.
>>
>> Apart from this issue of soft vs hard-float, the kernel doesn't care
>> what size long double is. In particular the kernel using 128-bit long
>> double doesn't impact userspace's ability to use 64-bit long double, as
>> musl does.
>>
>> So always build the 64-bit kernel with 128-bit long double. That should
>> avoid any build errors due to the incompatible ABI tags. Excluding the
>> code that uses soft/hard-float, the vmlinux is identical with/without
>> the flag.
>>
>> It does mean any code which is incorrectly intermingling soft &
>> hard-float code will build without error, so those bugs will need to be
>> caught by testing rather than at build time.
>>
>> For more background see:
>> - commit d11219ad53dc ("amdgpu: disable powerpc support for the newer display engine")
>> - commit c653c591789b ("drm/amdgpu: Re-enable DCN for 64-bit powerpc")
>> - https://lore.kernel.org/r/dab9cbd8-2626-4b99-8098-31fe76397d2d@xxxxxxxxxxxxxxxx
>>
>> Signed-off-by: Michael Ellerman <mpe@xxxxxxxxxxxxxx>
>
> Reviewed-by: Hamza Mahfooz <hamza.mahfooz@xxxxxxx>
Thanks.
> If you'd prefer to have this go through the amdgpu branch, please let
> me know.
I think it makes more sense to go via the powerpc tree, it will get more
testing on powerpc that way.
cheers
