WARNING!. Problems with the incoming update - kernel 6.1.0-18 & nvidia driver

[roseway]

I've successfully upgraded using proposed-updates. There were no problems, and everything now works as it should.

[MikeLieberman]

I added the proposed updates and the nvidia driveres were updated.

But was not re-offered the 6.1.0-18 I had run the purge of it five days ago.

So do I wait now for 6.1.0-19?

[dlu2021]

When you purged 6.1.0-18, did you do anything to stop it from coming back?

[MikeLieberman]

No, I just stopped all updates until I read about adding the "proposed" line in the sources.list file. I updated the file and get ran apt update/upgrade/dist-upgrade.

A bunch of stuff upgraded, including the nvidia code, but as of now, following the upgrades...

6.1.0-17-amd64

[appye]

Hello, I am posting here because this seems to be the main thread for this issue.

I am aware of various workarounds for this issue and my nvidia users (including myself) are happy for now. I would like to know though, how does this work on the development side? What I mean to ask is, if I want to be able to update (apt update && apt full-upgrade) these systems using standard repositories (not enabling proposed-updates, or using a backported kernel, etc) what am I waiting for before I can go back to being able to do so?

I hope this makes sense. It is not my intention to ask for any timeframes or anything like that. I guess I am asking how proposed-updates works ...? Is this something where I can just wait for some weeks or something for the fix to be included in the repositories, or do I need to wait for the next point release before I can continue to update these systems without having to implement whatever workaround I choose for this issue?

This is the first time I have run into this level of breakage on a Debian system on the stable branch!

[None1975]

Hello, I am posting here because this seems to be the main thread for this issue.

I am aware of various workarounds for this issue and my nvidia users (including myself) are happy for now. I would like to know though, how does this work on the development side? What I mean to ask is, if I want to be able to update (apt update && apt full-upgrade) these systems using standard repositories (not enabling proposed-updates, or using a backported kernel, etc) what am I waiting for before I can go back to being able to do so?

I have repeated many times and will repeat - if you use the open source nvidia video driver, you can update and live without any problems.

[appye]

I have repeated many times and will repeat - if you use the open source nvidia video driver, you can update and live without any problems.

Aside from performance, there are other issues with that one though. Resuming from hibernation/sleep has often been problematic for example. Also, I don't know of any way to limit (16-256) the color range with this driver for older displays.

My question was about how Debian handles distribution of the fixed driver though, whether or not it would be available in 12.5 without proposed-updates and a bunch of pinning, or if nvidia-driver users would have continue with workarounds until 12.6.

[arzgi]

Aside from performance, there are other issues with that one though. Resuming from hibernation/sleep has often been problematic for example. Also, I don't know of any way to limit (16-256) the color range with this driver for older displays.

I had no such problems when had nvidia gpu and used nouveau, suspend problems might be related or not, there have been many threads here about suspend/resume problems.

And the performance, how much do you need? My wife watches full screen tv-shows using nouveau, and I have said could even play some older games.

[appye]

And the performance, how much do you need? My wife watches full screen tv-shows using nouveau, and I have said could even play some older games.

I don't need the performance myself. This does not mean others do not. Stability and overly dark display issues are why I have generally avoid the open source driver for some time though. Arrogantly presenting ("I have repeated many times and will repeat ...") as a fix-all for any and all nvidia issues is also not practical or helpful for many.

Sigh. I would like to avoid a proprietary-vs-open source nvidia driver debate here. Perhaps in another thread? I would be happy to learn.

My question was about how Debian handles distribution of the fix for the issue originally presented in this thread.

[ListLess]

I added that to sources.list, and found this new module that I think is the right one for my GeForce 1050Ti:

Code: Select all

xserver-xorg-video-nvidia/proposed-updates 525.147.05-7~deb12u1 amd64

So do I apt install that and then run the NVIDIA .run file for 525? (Even though the latest is 535?)

My Ryzen-3 3200G system does include a low-end AMD GPU function within it, but I think I disabled that in the BIOS some time ago and I can't find the setting to restore it. (System76 Thelio) AT first (after the fresh install) the nouveau driver was at least working, but now something is interfering with that and all I get is 1024x768 VGA frame-buffer functionality. I removed the 'disable nouveau file from /etc/modprobe.d. It says nouveau is installed, but the system does not seem to use it.

I am doing all this because I had to wipe my system partition and reinstall from a Debian12 'stable' DVD, which gave me kernel 6.1.0-18. The previous Debian12 system used the 1050Ti just fine but apparently since I last installed that the kernel people changed something that made whatever NVIDIA was doing no longer work, and any attempt to install a driver gets the "GPL-only" error.

Or I may just give up and buy a Radeon card about equivalent to the 1050Ti and wipe my hands of Nvidia forever.

[Vieno]

dang - should have checked here first before messing up my system
viewtopic.php?t=158322

[csanyi5656]

I solved the problem with Liqourix kernel.

[appye]

Hot Damn! They fixed it in the main release (which answers my question)! I timeshifted back to before this fiasco, ran some full-upgrades and we've got everything back to normal.

Now this (and other) threads don't have to devolve into snark and uh ... stream of yellow liquid distance competitions!

[114514foxes]

This bug reveals a shocking fact:
NONE of the core maintainers of Debian uses NVIDIA graphics card.

[classicdeb]

I can confirm this problem has now been fixed - could someone please tag the topic as SOLVED? And many thanks to the Debian people for fixing this issue!

[None1975]

I can confirm this problem has now been fixed - could someone please tag the topic as SOLVED? And many thanks to the Debian people for fixing this issue!

Yes. It is fixed. Just update your system as usual.

Debian Stable Updates Announcement SUA 252-1 https://www.debian.org
debian-release@lists.debian.org Andreas Beckmann
February 16th, 2024
-----------------------------------------------------------------------

This update addresses problems in three non-free driver packages supporting
nVidia graphics cards.

The Linux kernel released in Debian 12.5 changed an inlined function to
call two GPL-only symbols, making that function inaccessible to non-free
kernel modules.

As a result, the nVidia kernel modules cannot be built via DKMS at
installation time for the updated kernel. This issue could not be resolved
in time for the release of Debian 12.5.

In addition, the source packages nvidia-graphics-drivers and
nvidia-settings now build binaries for ppc64el, and
nvidia-graphics-drivers-tesla builds transitional packages to ease future
updates.

The following packages have been updated to correct the problem:

Source package Fixed version
============== =============
nvidia-graphics-drivers 525.147.05-7~deb12u1
nvidia-graphics-drivers-tesla 525.147.05-7~deb12u1
nvidia-graphics-drivers-tesla-470 470.223.02-4~deb12u1
nvidia-settings 525.147.05-1~deb12u1

If you use the affected packages, we recommend you upgrade to these
versions.

[nickmdowson]

Hi,
I just enable proposed-updates, and ran the upgrade. upgrade ran fine after issues earlier in the week.
However I got the following messages as part of the upgrade and am wondering what action I still need to take?

Code: Select all

Preparing to unpack .../34-nvidia-kernel-dkms_525.147.05-7~deb12u1_amd64.deb ...
Module nvidia-current-525.147.05 for kernel 6.1.0-16-amd64 (x86_64).
Before uninstall, this module version was ACTIVE on this kernel.

nvidia-current.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-16-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-modeset.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-16-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-drm.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-16-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-uvm.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-16-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-peermem.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-16-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.
depmod....
Module nvidia-current-525.147.05 for kernel 6.1.0-17-amd64 (x86_64).
Before uninstall, this module version was ACTIVE on this kernel.

nvidia-current.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-17-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-modeset.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-17-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-drm.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-17-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-uvm.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-17-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.

nvidia-current-peermem.ko:
 - Uninstallation
   - Deleting from: /lib/modules/6.1.0-17-amd64/updates/dkms/
 - Original module
   - No original module was found for this module on this kernel.
   - Use the dkms install command to reinstall any previous module version.
depmod....
Deleting module nvidia-current-525.147.05 completely from the DKMS tree.
Unpacking nvidia-kernel-dkms (525.147.05-7~deb12u1) over (525.147.05-4~deb12u1) ...

[mlauronen]

I can confirm this problem has now been fixed - could someone please tag the topic as SOLVED? And many thanks to the Debian people for fixing this issue!

Is this issue fixed for those who are using drivers from developer's repository (for example Cuda users) ?
For example, a week ago nvidia driver version 535.104.12 did not work with kernel version 6.1.0-18 and I had to switch back to version 6.1.0-17.

[ListLess]

These fixes seem to be for driver 525.147.05-7~deb12u1. But the 'older drivers' list at the nvidia download page do not list such a version. I have no nvidia driver installed at all, and have kernel 6.1.0-18. What are the sequence of steps I need to perform to get a working driver for my GeForce 10 GPU? I have already added the 'proposed' repository to sources.list.

[ryanz]

I can confirm this problem has now been fixed - could someone please tag the topic as SOLVED? And many thanks to the Debian people for fixing this issue!

Is this issue fixed for those who are using drivers from developer's repository (for example Cuda users) ?
For example, a week ago nvidia driver version 535.104.12 did not work with kernel version 6.1.0-18 and I had to switch back to version 6.1.0-17.

Good feedback in this thread, I'll give it a few days and check once these changes are on the stable.