Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Michael Ellerman <mpe@ellerman.id.au>
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     linuxppc-dev@lists.ozlabs.org, alexdeucher@gmail.com,
        amd-gfx@lists.freedesktop.org, linux@roeck-us.net,
        linux-kernel@vger.kernel.org, dan@danny.cz,
        tpearson@raptorengineering.com
Subject: Re: [PATCH] drm/amdgpu: Re-enable DCN for 64-bit powerpc
In-Reply-To: <CAHk-=wihON4Ytte5zLHWNQtTapUvCpkToxY06OjX-_2B+Gq6Gg@mail.gmail.com>
References: <20220725123918.1903255-1-mpe@ellerman.id.au>
 <CAHk-=wihON4Ytte5zLHWNQtTapUvCpkToxY06OjX-_2B+Gq6Gg@mail.gmail.com>
Date:   Tue, 26 Jul 2022 10:56:07 +1000
Message-ID: <87k080bsk8.fsf@mpe.ellerman.id.au>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Linus Torvalds <torvalds@linux-foundation.org> writes:
> On Mon, Jul 25, 2022 at 5:39 AM Michael Ellerman <mpe@ellerman.id.au> wrote:
>>
>> Further digging shows that the build failures only occur with compilers
>> that default to 64-bit long double.
>
> Where the heck do we have 'long double' things anywhere in the kernel?

There's one or two uses, but not in any code that's relevant to this
issue AFAICS.

> I tried to grep for it, and failed miserably. I found some constants
> that would qualify, but they were in the v4l colorspaces-details.rst
> doc file.
>
> Strange.

It doesn't seem to matter if you use long double or not. It's just that
if the long double size is 64-bits the linker refuses to link a mixture
of soft/hard-float objects.

The 64-bit ABI says long double is 128-bits, so the compilers that are
using 64-bit long double are either not built correctly, or we are not
passing the correct flags to them.

There's an -mlong-double-128 flag which we can pass at build time which
seems to do the right thing, I will probably add that to the kernel
CFLAGS, but I want that to get a bit more testing.

cheers