Bug#959210: no automatic Xorg configuration for NVidia

[Eduard Bloch]

Hallo,
* Andreas Beckmann [Fri, May 01 2020, 12:49:11AM]:
> On 30/04/2020 22.18, Eduard Bloch wrote:
> > I don't know what the problem is. Looking at Xorg log, it seems that it
>
> The problem happens earlier, before xorg gets started:
> The nvidia module does not get loaded at boot.
>
> > still insist on loading NOUVEAU. So I think something is broken in your
> > package, the bit which was supposed to turn nouveau off.
>
> I think something is "special" in your long grown system, some local
> tweak long forgotten that now gets in the way of nvidia ...
>
> /etc/modules-load.d/nvidia.conf should be responsible for loading the
> nvidia-drm kernel module, while
> /etc/modprobe.d/nvidia-blacklists-nouveau.conf prevents nouveau from
> being loaded automatically.
> Does manual 'modprobe -v nvidia-drm' work? If not, dmesg may contain
> interesting error messages.

$ cat /etc/modules-load.d/nvidia.conf
nvidia-drm
$ cat /etc/modprobe.d/nvidia-blacklists-nouveau.conf
# You need to run "update-initramfs -u" after editing this file.

# see #580894
blacklist nouveau

$ modprobe -v nvidia-drm
$ modinfo nvidia-drm
modinfo: ERROR: Module nvidia-drm not found.

Now what?

$ grep nvidia-kernel /var/log/dpkg.log
2020-04-30 21:51:28 install nvidia-kernel-common:amd64 20151021+4 20151021+11
2020-04-30 21:51:28 status half-installed nvidia-kernel-common:amd64 20151021+4
2020-04-30 21:51:28 status unpacked nvidia-kernel-common:amd64 20151021+11
2020-04-30 21:51:28 install nvidia-kernel-source:amd64 <keine> 440.82-1
2020-04-30 21:51:28 status half-installed nvidia-kernel-source:amd64 440.82-1
2020-04-30 21:51:29 status unpacked nvidia-kernel-source:amd64 440.82-1
2020-04-30 21:51:29 install nvidia-kernel-support:amd64 375.82-1 440.82-1
2020-04-30 21:51:29 status half-installed nvidia-kernel-support:amd64 375.82-1
2020-04-30 21:51:29 status unpacked nvidia-kernel-support:amd64 440.82-1
2020-04-30 21:51:29 configure nvidia-kernel-common:amd64 20151021+11 <keine>
2020-04-30 21:51:29 status unpacked nvidia-kernel-common:amd64 20151021+11
2020-04-30 21:51:29 status half-configured nvidia-kernel-common:amd64 20151021+11
2020-04-30 21:51:30 status installed nvidia-kernel-common:amd64 20151021+11
2020-04-30 21:51:30 configure nvidia-kernel-source:amd64 440.82-1 <keine>
2020-04-30 21:51:30 status unpacked nvidia-kernel-source:amd64 440.82-1
2020-04-30 21:51:30 status half-configured nvidia-kernel-source:amd64 440.82-1
2020-04-30 21:51:30 status installed nvidia-kernel-source:amd64 440.82-1
2020-04-30 21:51:43 configure nvidia-kernel-support:amd64 440.82-1 <keine>
2020-04-30 21:51:43 status unpacked nvidia-kernel-support:amd64 440.82-1
2020-04-30 21:51:43 status half-configured nvidia-kernel-support:amd64 440.82-1
2020-04-30 21:51:43 status triggers-awaited nvidia-kernel-support:amd64 440.82-1
2020-04-30 21:51:52 status installed nvidia-kernel-support:amd64 440.82-1
2020-04-30 22:11:07 install nvidia-kernel-dkms:amd64 <keine> 440.82-1
2020-04-30 22:11:07 status half-installed nvidia-kernel-dkms:amd64 440.82-1
2020-04-30 22:11:08 status unpacked nvidia-kernel-dkms:amd64 440.82-1
2020-04-30 22:11:19 configure nvidia-kernel-dkms:amd64 440.82-1 <keine>
2020-04-30 22:11:19 status unpacked nvidia-kernel-dkms:amd64 440.82-1
2020-04-30 22:11:19 status half-configured nvidia-kernel-dkms:amd64 440.82-1
2020-04-30 22:11:58 status installed nvidia-kernel-dkms:amd64 440.82-1

Now that confuses me. I have run "m-a a-i nvidia -vt" yesterday and as
far as I remember, it has finished successfully, it has installed a few
packages. But I don't see a log
in /var/cache/modass/ , only nvidia-kernel-source.cur_version and
nvidia-kernel-source.avail_version files.
That's totally confusing. Why was the package

Ok, so what I did not? Run "m-a a-i nvidia -vt" again.

OTOH I don't know. The driver worked yesterday after chaning xorg.conf
so it should be there, and I cannot check afterwards what happened. I
rerun m-a again, and it did build and install
nvidia-kernel-5.6.0-trunk-amd64 this time.  But the module it contains
is called nvidia-current-drm, not nvidia-drm.

EDIT:

continuing to write this a after a couple of reboots. I tried to
recreate the situation from yesterday.

I see that installing the m-a packages was not needed, the module was
already built and injected by dkms and yes, it was loadable. But it was
not loaded yesterday, according to the log it was only loaded after I
edited xorg.conf (so it was triggerd forcibly by the driver). Also, this
looks confusing if you don't know what to look for:

# modprobe -v nvidia-drm
# modprobe -v nvidia-current-drm
insmod /lib/modules/5.6.0-trunk-amd64/updates/dkms/nvidia-current.ko
insmod /lib/modules/5.6.0-trunk-amd64/updates/dkms/nvidia-current-modeset.ko
insmod /lib/modules/5.6.0-trunk-amd64/updates/dkms/nvidia-current-drm.ko

So, your hint from above was only partly helpful. Maybe this should be
brought to modutils maintainers, this difference should not exist.

Anyway, "modinfo nvidia-current-drm" confirms that its available:

filename:       /lib/modules/5.6.0-trunk-amd64/updates/dkms/nvidia-current-drm.ko
version:        440.82
supported:      external
license:        MIT
srcversion:     D65980CAE08DFB7B82FE1D1
alias:          pci:v000010DEd*sv*sd*bc03sc02i00*
alias:          pci:v000010DEd*sv*sd*bc03sc00i00*
depends:        drm,drm_kms_helper,nvidia-modeset
retpoline:      Y
name:           nvidia_drm
vermagic:       5.6.0-trunk-amd64 SMP mod_unload modversions
parm:           modeset:Enable atomic kernel modesetting (1 = enable, 0 = disable (default)) (bool)

> I saw this in your report that makes me suspicious: why is the radeon
> module loaded?

It's a fixed entry in /etc/modules , it was added a couple of years ago
to overrule amdgpu driver because I had a RX560D which was supported by
both drivers.

> PS: if nvidia-drm.ko is loaded, xorg should use nvidia automatically

I am trying to recollect what happened yesterday since I cannot fully
explain and cannot proove much through the logs. The only plausible
explanation is that I mixed it up. Upon the first reboot, the module was
not loaded nor was nouveau blacklisted (see Bug#959208 for the reason).
Maybe that was the moment where nvidia loading did not work.

Either that or there was some subtle issue which is no longer
reproducible. Either way, that kinda feels like 1999.

Best regards,
Eduard.