Bug#685617: Allow co-installation of nvidia-kernel-dkms and nvidia-kernel-legacy-173xx-dkms

Gaudenz Steinlin gaudenz at debian.org
Wed Aug 22 22:18:28 UTC 2012


Andreas Beckmann <debian at abeckmann.de> writes:

> On 2012-08-22 17:51, Gaudenz Steinlin wrote:
>> If you want to use the full range of nvidia proprietry drivers on live
>> systems the kernel dkms packages must both be installed at the same time
>> and the correct module to use selected at runtime. While co-installation
>> is possible for the GLX and xorg parts, this is currently not possible
>> for the dkms packages.
>
> Thanks for your patches. That could finally solve the missing
> co-installability of the kernel modules.
>
> Switching between current and legacy is done by reconfiguring the nvidia
> alternative (/usr/lib/nvidia/nvidia) - can't that be used by the module
> load script to select the "right" module? And "right" is here not what
> fits the hardware best ... but what was configured.

In theory yes. But there are several problems:
- plain alternatives are not possible because the files (the compiled
  kernel modules) don't necesserily exist even if the package is
  installed. Beside that I don't even know if it would work with
  modprobe.
- just looking at the selected alternative without using the
  alternatives mechanisem directly in the modprobe install script looks
  even hackier and also inferior (why not just select the best module if
  we can).

>
> The initial selection of the right driver (and setting the nvidia and
> glx alternatives) is a different problem to be solved by the live system
> boot process ... nvidia-detect should be able help with this, it looks
> at PCI IDs and checks the lists. If you need machine readable output -
> patches are welcome.

Of course. I don't expect your packages to address this problem. As I
said I'm currently working on this. Like in the module loading script I
think it's easier to just parse the nvidia.ids files directly than to
use nvidia-detect because nvidia-detect gives a lot of information that
is nto really relevant. Like what version I could use on an older
distribution or with an older X server.

>
>> The attached patches modify the dkms packages to allow coinstallation.
>> The legacy packages dkms config is modified to name the module
>> nvidia-legacy-173xx instead of just nvidia.
>
> Needs to be done for -source, too.

To be honest I did not look at the -source packages too closely. I'm not
sure if they are still relevant. Dkms is IMHO the far superior and
easier alternative. But anyway the -source packages are already
co-installable AFAICS. I don't know if there is a good way to modify the
build system so that the module is renamed. As I think the -source
packages are mostly relevant for manual builds I'm not even sure if it's
a good idea to do such a modification there.

>
> If I have a card requiring legacy-173xx - do
>   modprobe nvidia
> and
>   rmmod nvidia
> still work afterwards?

In my tests it does because the kernel internal name of the module is
still nvidia.

>
> Anyway, this looks like stuff for wheezy+1.

Agreed. Backports to wheezy would be cool though.

Gaudenz

BTW: If you need a sponsor to upload this, just contact me.

-- 
Ever tried. Ever failed. No matter.
Try again. Fail again. Fail better.
~ Samuel Beckett ~



More information about the pkg-nvidia-devel mailing list