post-352 drivers

[Luca Boccassi]

On 2 February 2016 at 21:37, Luca Boccassi <luca.boccassi at gmail.com> wrote:
> On Tue, 2016-02-02 at 13:00 +0100, Andreas Beckmann wrote:
>> On 2016-02-02 12:30, Luca Boccassi wrote:
>> > On 1 February 2016 at 19:45, Andreas Beckmann <anbe at debian.org> wrote:
>> >> Regarding conftest.sh: given the new upstream build system, can we just
>> >> switch to using upstream's conftest.sh instead of patching around it and
>> >> brewing our own conftest.h? Although it does not work out of the box ...
>> >> probably due to the path layout with the -common headers packages. But
>> >> maybe we can just patch that?
>> >
>> > It would be great if we could, as it would make maintenance so much
>> > easier. I've seen you already committed some changes to svn/355,
>> > that's great, I'll give it a spin as well tonight or tomorrow.
>>
>> This just needed some understanding how it works internally ... and that
>> we kept calling into the build system in a way that "broke" conftest.sh
>> :-( , predating my maintenance takeover.
>>
>> 355.11+svn successfully survived my build-against-50-kernel-headers on
>> amd64 test. Not sure whether any of these modules actually works :-)
>
> Latest from branches/355 in my sid chroots fails to build on amd64 amd
> armhf against all current kernel headers. The i386 ones work fine
> though, which is curious. I'll keep debugging, meanwhile I'm attaching
> the result of the dkms install for reference (this is a dkms install
> after installing nvidia-kernel-dkms).

After some debugging I figured out the problem.

DKMS is passing down ARCH=<binary ABI triplet> when it calls make. But
conftest.sh uses $ARCH for the arch-specific Linux headers directory.
I solved it by simply unsetting ARCH in the dkms.conf, and now it's
building on all architectures, as upstream's Makefile will derive ARCH
from uname if it's not set.

Kind regards,
Luca Boccassi