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

[Luca Boccassi]

On 19 October 2015 at 00:15, Andreas Beckmann <anbe at debian.org> wrote:
> On 2015-10-19 00:24, Luca Boccassi wrote:
>>> always run
>>>   dpkg-reconfigure glx-alternative-nvidia
>>> after
>>>   update-alternatives --config glx
>
>>> to activate and run the triggers
>>>
>>> (THIS NEEDS TO BE DOCUMENTED)
>
> Maybe we should write script (update-glx) that is a wrapper around
> update-alternatives nvidia/glx and triggers+runs the neccessary followup
> actions. ENOTIME - suggestions + patches welcome.

I'll try to find some time to have a look, but haven't played much
with update-alternatives before.

>> This works with 340.93-5. Also tested on the desktop for collateral
>> and it seems fine.
>
> Thanks.
> Does CUDA/OpenCL work with Bumblebee at all? Does it work with this
> configuration?

It does and yes it works fine, I tried one of the Cuda examples.
Modules are removed and card is turned off correctly.

>> Would this mean that a bumblebee user has to run this manually when
>> the package is installed? Is there a way we can automate this and make
>> the bumblebee package trigger it?
>
> Right now, yes. Should I upload this now?

Given that the current state is broken I'd say yes, but what about
warning the users at upgrade time with a .NEWS file with the
instructions?

> How can I detect that a system is to be run under bumblebee? I could
> skip creation of the "regular" /usr/lib/nvidia alternative there. Or
> lower its priority.
> From a live-system point of view having stuff installed does not mean
> that it is being used (thats why we aim for co-installability of nvidia,
> nvidia-legacy-*, fglrx, using the nvidia and glx alternatives for
> switching). If the bumblebee packages can provide some kind of switch,
> all that would be needed after toggling it is to run
> dpkg-trigger register-glx-alternative-nvidia from their maintainer scripts.

Extremely tricky I'm afraid. The only way to be somewhat sure would be
if bumblebee is installed and the daemon is enabled and running, but
it can be installed separately from the drivers so it's not that easy.

>>> since that will 'modprobe -r nvidia'/'modprobe nvidia' you may end up
>>> with the wrong^Wproblematic permissions - are you in the video group?
>>
>> The devices are created with 666:
>
> I won't trust the output from ls since you recently reported that the
> device node switched from root:root 0666 to root:video 0660 after gdm3
> crashed (so probably the device node in the /dev tmpfs got updated after
> being accessed). :-)

What's the best way to check?

Kind regards,
Luca Boccassi