[SOLVED] Nvidia driver failure on Vera because of gcc-11 on updated kernel 5.19.0-28-generic

[midwestE]

When installing nvidia-driver-525 on kernel 5.19.0-28-generic the nvidia install fails with

Code: Select all

dpkg: error processing package nvidia-dkms-525(--configure):
 installed nvidia-dkms-525 package post-installation script subprocess returned error exit status 10

Examining the /var/lib/dkms/nvidia/525.78.01/build/make.log reveals multiple
cc: error: unrecognized command-line option ‘-ftrivial-auto-var-init=zero’

Looks like kernel 5.19.0-28-generic needs gcc-12 to compile the nvidia-driver-525 correctly.

[Per SMG - The Nvidia-525 driver needs the same version gcc as the kernel you have installed used. You upgraded to the 5.19 kernel. Had you remained on the 5.15 kernel then you would not have had problems.] Switching back to 5.15 is therefore an option.

-------------------------------------

If you don't want to switch back to kernel 5.15, here's what worked for me:

Remove nvidia driver

Code: Select all

sudo apt-get purge *nvidia* && sudo apt-get autoremove

Update gcc and update alternatives
# https://linuxconfig.org/how-to-switch-b ... ocal-fossa

Code: Select all

sudo apt-get install gcc-12
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 12

Check the gcc version, should now show v 12.x.x:

Code: Select all

gcc --version
gcc (Ubuntu 12.1.0-2ubuntu1~22.04) 12.1.0
Copyright (C) 2022 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Install nvidia driver:

Code: Select all

sudo apt-get install nvidia-driver-525

[kallestar4]

I seem to have had this just fail with an updated kernel that came along, with nvidia-driver-515 installed - suddenly the Mesa driver was being used instead.

In my case, gcc-12 was installed as a package (according to dpkg), but /usr/bin/gcc-12 didn't exist, so I guess maybe something was mis-configured somewhere.

I found installing the package 'g++-12' was the key thing to getting gcc-12 fully installed and, then I could re-apply the nvidia drivers again with the new kernel.

[SMG]

Looks like nvidia-driver-525 needs gcc-12 to install correctly.

The Nvidia-525 driver needs the same version gcc as the kernel you have installed used. You upgraded to the 5.19 kernel. Had you remained on the 5.15 kernel then you would not have had problems.

[kallestar4]

You upgraded to the 5.19 kernel. Had you remained on the 5.15 kernel then you would not have had problems.

In my case, I don't remember actively switching kernel versions. I certainly haven't changed anything in the View -> Linux Kernels part of the Linux Mint Update manager, and a few weeks ago I was on 5.15 after the upgrade to 21.1, so I'm not really sure how this happened.

On another 21.1 machine with Intel GPU (so didn't have an issue), I appear to be on 5.19 there as well all of a sudden, so that's two machines that appear to have suddenly changed minor kernel versions without my obviously agreeing to it, which I don't remember that having happened before for several years.

Can the kernel version update (as a minor version, not just a patch version like normal) with a "standard" Update manager package change?

[SMG]

In my case, I don't remember actively switching kernel versions. I certainly haven't changed anything in the View -> Linux Kernels part of the Linux Mint Update manager, and a few weeks ago I was on 5.15 after the upgrade to 21.1, so I'm not really sure how this happened.

On another 21.1 machine with Intel GPU (so didn't have an issue), I appear to be on 5.19 there as well all of a sudden, so that's two machines that appear to have suddenly changed minor kernel versions without my obviously agreeing to it, which I don't remember every happening before for several years.

Did you upgrade in place from LM20.3 on those computers? I ask because maybe you were using the HWE kernel in LM20.3 and thus had the packages for it already installed. That would be my guess as to what might have happened.

The 5.15 kernel is the LTS kernel for LM21 and is supported through its entire life so change is not needed and, if one has the LTS kernel (rather than the HWE kernel), then it should not change. (At least that is my understanding how it should work.)

[kallestar4]

Did you upgrade in place from LM20.3 on those computers? I ask because maybe you were using the HWE kernel in LM20.3 and thus had the packages for it already installed. That would be my guess as to what might have happened.

On the laptop with the Intel GPU (which didn't have any issues obviously) it was an update progressively from 20.0 -> 21.0 -> 21.1, but my desktop system with an nvidia GPU which had the compiler mis-match issue, was a clean install of 21.0 in August (when it came out).

The 5.15 kernel is the LTS kernel for LM21 and is supported through its entire life so change is not needed and, if one has the LTS kernel (rather than the HWE kernel), then it should not change. (At least that is my understanding how it should work.)

Yeah, that's what's (slightly) concerning to me: it's not the end of the world as I've fixed it now, but I am curious as to how this happened on two machines, as I haven't had anything break "automatically" in Linux Mint in 4/5 years (since some Python 2/3 weirdness which broke Cinnamon back in Mint 17 or something) until now.

I'm also not sure if it's normal for a package (in this case Kernel 5.19) to be built with the non-"native" compiler version for the Ubuntu/Mint distro version, which I'm fairly sure is gcc 11 (going off the readmes of both Ubuntu LTS and Mint 21.1): I would have expected all packages to be built with the "native" compiler, rather than a slightly newer one, just for maximum compatibility (I know C ABIs should really be compatible between major compiler versions, but....)

[SMG]

On the laptop with the Intel GPU (which didn't have any issues obviously) it was an update progressively from 20.0 -> 21.0 -> 21.1

I believe you meant 20.0->20.3->21.0->21.1.

If at any time while on LM20 you upgraded to an HWE kernel then I would presume you had those packages on your computer which would explain why that one upgraded. Approximately a year ago, the HWE packages started automatically upgrading to a newer kernel when it became available. HWE kernels are short-term support kernels so if one does not upgrade, they may end up running an unsupported kernel.

but my desktop system with an nvidia GPU which had the compiler mis-match issue, was a clean install of 21.0 in August (when it came out).

I do not understand why that one automatically upgraded to the 5.19 kernel. Also, it is odd your system needed more than just the gcc-12 package installed. So far, others have only needed the one package. I do not have any guesses as to what might have happened on your install. Maybe others will have an idea of what might have caused the automatic upgrade.

I'm also not sure if it's normal for a package (in this case Kernel 5.19) to be built with the non-"native" compiler version for the Ubuntu/Mint distro version, which I'm fairly sure is gcc 11 (going off the readmes of both Ubuntu LTS and Mint 21.1): I would have expected all packages to be built with the "native" compiler, rather than a slightly newer one, just for maximum compatibility (I know C ABIs should really be compatible between major compiler versions, but....)

That is also a bit puzzling to me. The 5.19 kernel was the release kernel for Ubuntu 22.10 and, based on what someone else posted for inxi info from their install, there were some other difference I would not have expected. I do not know where might be the appropriate place to find out why that happened.

[Loron]

Hello!
Your instructions work for me as well. Had the same issues with the kernel and NVidia drivers.
Thank you!

Regards

Loroin

[midwestE]

Looks like nvidia-driver-525 needs gcc-12 to install correctly.

The Nvidia-525 driver needs the same version gcc as the kernel you have installed used. You upgraded to the 5.19 kernel. Had you remained on the 5.15 kernel then you would not have had problems.

Thank you, that makes sense. I may have previously updated my kernel in previous version of Mint before I did a mintupdate. I've updated the post to add information you provided

[visseroth]

@midwestE, thank you for the post, I was banging my head on the wall trying to figure out why I couldn't get the NVidia driver installed with the new kernel and the new kernel gave me better hardware support than the old kernel.
SM3 was kind enough to point out your fix to me.

[ggill]

Thanks for this post. Really helpful.

When I had a look at the logs, I realised that the nvidia-driver-525 install was failing only for kernel 5.19 and installing correctly for kernel 5.15, for the reasons given here. But using the Mint update manager or installing software with synaptic (for completely unrelated packages), every time it would try to finish the installation of nvidia-driver-525 and give errors.

If you want to use kernel 5.19 and keep kernel 5.15 installed as a backup, how would you go about doing that? How can you do the equivalent of installing nvidia-driver-525 just for kernel 5.19 using gcc-12, and then again for kernel 5.15 using gcc-11? Seems like you'd want to have separate packages that allow something like that.

[SMG]

If you want to use kernel 5.19 and keep kernel 5.15 installed as a backup, how would you go about doing that? How can you do the equivalent of installing nvidia-driver-525 just for kernel 5.19 using gcc-12, and then again for kernel 5.15 using gcc-11? Seems like you'd want to have separate packages that allow something like that.

You can have both gcc-11 and gcc-12 installed. When you install gcc-12, that adds that package to your system in addition to the gcc-11 package you already have. It does not replace the gcc-11 package on your system. Multiple versions of gcc / g++ affects.

[mtwebster]

Hi we need to update our 'dkms' package for Mint 21, which has the missing dependencies..

In the meantime you can run:

Code: Select all

apt install dkms=2.8.7-2ubuntu2.1

to get the updated version. Once we update our own that will be replaced.

The issue with that upstream version (and the reason we patch it for Mint) is it causes an issue with dual-gpu laptops - not breaking issues, but it can make the 'powersave' mode unavailable. For normal systems there should be no problem.

[mgarcia]

When installing nvidia-driver-525 on kernel 5.19.0-28-generic the nvidia install fails with

Code: Select all

dpkg: error processing package nvidia-dkms-525(--configure):
 installed nvidia-dkms-525 package post-installation script subprocess returned error exit status 10

Examining the /var/lib/dkms/nvidia/525.78.01/build/make.log reveals multiple
cc: error: unrecognized command-line option ‘-ftrivial-auto-var-init=zero’

Looks like kernel 5.19.0-28-generic needs gcc-12 to compile the nvidia-driver-525 correctly.

[Per SMG - The Nvidia-525 driver needs the same version gcc as the kernel you have installed used. You upgraded to the 5.19 kernel. Had you remained on the 5.15 kernel then you would not have had problems.] Switching back to 5.15 is therefore an option.

-------------------------------------

If you don't want to switch back to kernel 5.15, here's what worked for me:

Remove nvidia driver

Code: Select all

sudo apt-get purge *nvidia* && sudo apt-get autoremove

Update gcc and update alternatives
# https://linuxconfig.org/how-to-switch-b ... ocal-fossa

Code: Select all

sudo apt-get install gcc-12
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 12

Check the gcc version, should now show v 12.x.x:

Code: Select all

gcc --version
gcc (Ubuntu 12.1.0-2ubuntu1~22.04) 12.1.0
Copyright (C) 2022 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Install nvidia driver:

Code: Select all

sudo apt-get install nvidia-driver-525

thanks, that helped me too

[mtwebster]

The dkms package has been updated (2.8.7-2ubuntu2.1mint1) so this should not be an issue anymore.

[enigmista911]

Thank a lot, that solve the problem for me in Dell G16 .
I owe 1 beer to you

[SMG]

Thank a lot, that solve the problem for me in Dell G16 .
I owe 1 beer to you

Welcome to the forum, enigmista911.

What exactly solved your problem? This topic is old and updating your system to the latest files before installing the Nvidia driver is all one needs to do.