testing kernel module build in branches/304 and branches/340

[Luca Boccassi]

On Fri, 2016-02-26 at 09:41 +0000, Luca Boccassi wrote:
> On Feb 26, 2016 00:56, "Andreas Beckmann" <anbe at debian.org> wrote:
> >
> > On 2016-02-26 01:44, Luca Boccassi wrote:
> > > DKMS builds fine on amd64 and i386, but fails on ARM because the
> Xen
> > > test fails, despite the patch and that ARCH is set to arm:
> > >
> > > ...
> > > if ! /bin/sh /var/lib/dkms/nvidia-current/340.96/build/conftest.sh
> "
> > > gcc-4.9" " gcc-4.9" arm ...
> >
> > > Also, the dkms.conf on arm has the UVM bits enabled:
> > >
> > > BUILT_MODULE_NAME[1]="nvidia-uvm"
> > > BUILT_MODULE_LOCATION[1]="uvm/"
> > > ...
> >
> > 340 still supports uvm everywhere, but it comes with a copy of
> > conftest.sh in uvm/ ... that does not get patched. I'll fix this.
> 
> Ah, of course, completely forgot about that :-) Thanks for fixing,
> I'll try again.

That worked as expected, thanks!

> > > The kernel-source builds on i386, but curiously not on amd64,
> > > conftest.sh is failing.
> >
> > That's extremely weird, since amd64 worked for me.
> > Which kernel? Log?
> 
> It's really weird, it was on all kernel versions mentioned in the
> previous mail. I don't have the logs here now, it might have been due
> to the uvm conftest. I'll try again tonight from scratch and with the
> updated uvm symlink you committed, if it still happens I'll paste the
> logs.
> 
> By the way, I'm doing the source build as described in method 3 of the
> readme.

Ah, got it now :-)

Remember the fix to dkms.conf I did a short while back, to unset ARCH
(NVIDIA expects kernel arch, but Debian tools pass ABI triplet)?

Same thing had to be applied to debian/module/debian/rules.in too, since
the source build does not use dkms obviously. I noticed because the arm
conftest.sh fix to avoid the Xen problem was not working, and then I saw
that ARCH was armhf instead of arm.

Tested and committed to branches/340.

Both DKMS and source build works on all architectures on kernels 3.16,
4.3 and 4.4 (plus amd64/i386 on 3.2).

I'll make the same commit to trunk and test it, and then next on the
list is 304.

Kind regards,
Luca Boccassi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part
URL: <http://lists.alioth.debian.org/pipermail/pkg-nvidia-devel/attachments/20160228/106a1882/attachment.sig>