Bug#802246: Can't modprobe -r nvidia, rmmod works

[Luca Boccassi]

On Tue, 2015-10-20 at 11:37 +0300, Alexander Monakov wrote:
> On Tue, Oct 20, 2015 at 12:22 AM, Luca Boccassi <luca.boccassi at gmail.com> wrote:
> > The problem was that the nvidia's modprobe conf was moved to
> > glx-alternative and it was not configured (since on a bumblebee setup
> > glx is served by mesa). This was fixed by Andreas with a new diversion
> > specifically for bumblebee setups. I'm not seeing issues now with
> > modprobe -r nvidia.
> 
> But if nvidia-modprobe.conf was not active, how did 'modprobe nvidia'
> (insertion, not removal) work in the first place? I would understand
> if neither worked, but insertion apparently worked — otherwise you'd
> notice the issue much earlier.

I haven't inspected the event with gdb, so I cannot say with 100%
certainty. My best guess is that since the module is in the right path
and the internal name is nvidia, modprobe is able to resolve it.

Andreas, any more insight on this?

> > The reason for the change was that the old version broke when
> > nvidia-current and nvidia-legacy-YYYxx were installed.
> 
> I don't see what was the problem. Can you briefly explain?

This bug here: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=798207

Summary is, due to the old aliases depmod generates a broken
modules.dep, which means the wrong module is loaded when -current and
-legacy-YYYxx are installed and -current is active.

> Sorry to press you for answers, but on one hand, on occasion I help
> with bumblebee bugs, so I'd like to know what Debian is doing, and on
> the other hand, I don't have a machine with Debian to investigate on
> my own.

No problem at all, it's great to have more engagement with upstream
developers. Keep them coming :-)

> FWIW I don't think a "special" bumblebee-specific glx-alternative
> option is a good way to go. Bumblebee aside, consider for example
> people running server-class hardware with a small Matrox card as
> primary VGA adapter and NVIDIA GPUs for rendering or compute jobs, or
> enthusiasts trying to learn OpenCL/CUDA while leaving their desktop
> running on Intel GPU. If the issue is in kernel module availability,
> I'd recommend decoupling the concept of "GLX alternative" and "nvidia
> kernel module alternative".

This glx-alternative profile that Andreas created sets mesa as the
provider of the libraries, so it should work for those use cases. It can
be configured without bumblebee.

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/20151020/355366e6/attachment.sig>