** Warning ** when upgrading to NVIDIA 595.58.03 Driver

Hi all, Im running FreeBSD 15-stable on my laptop and did a Uppgrade of the NVIDIA GPU Driver to 595.58.03 on the Quadro T1000 GPU in the laptop.
I ran a # pkg upgrade on the laptop that i try to do monthly and then investigated the status of the NVIDIA drivers ( shown in the attchment )
As expected the Nvidia KMOD modules were not upgraded and also the listings at the "freshport.org" site shows that 596.58.03 nvidia-drm-kmod is only avaailbel for FreeBSD 16-current. As I realised that this was a nonworking situation I upgraded my /usr/ports/ tree from GIT and built the
graphics/nvidia-drm.kmod and graphics/nvidia-drm-66-kmod from the source code package.
After that I was able to reboot the desktop cleanly.
 

Attachments

  • ksnip_20260402-231055.png
    ksnip_20260402-231055.png
    15.2 KB · Views: 11
Hi all, Im running FreeBSD 15-stable on my laptop and did a Uppgrade of the NVIDIA GPU Driver to 595.58.03 on the Quadro T1000 GPU in the laptop.
I ran a # pkg upgrade on the laptop that i try to do monthly and then investigated the status of the NVIDIA drivers ( shown in the attchment )
As expected the Nvidia KMOD modules were not upgraded and also the listings at the "freshport.org" site shows that 596.58.03 nvidia-drm-kmod is only avaailbel for FreeBSD 16-current. As I realised that this was a nonworking situation I upgraded my /usr/ports/ tree from GIT and built the
graphics/nvidia-drm.kmod and graphics/nvidia-drm-66-kmod from the source code package.
After that I was able to reboot the desktop cleanly.
This doesn't matter with the version. On EVERY upgrades, this mutually happen on official pkg repo.

This is because main pkg builders and kmod pkg builders are running independently, thus, finishes in different timings.

This is mutually unavoidable. Users are adviced to perform dry-run and confirm the versions of nvidia-kmod* and corresponding nvidia-driver* pkgs (if you're using nvidia-drm-*-kmod* pkg, it need to be, too) matches before performing actual upgrades.

And this shouldn't happen on local bulk builds/upgrades (except building related ports independently, step by step) on exactly same environment (i.e., builder jails) on exactly the same computer if all builds fine.
 
For what it's worth before any upgrade with Nvidia, as I have a system on ZFS, I use bectl to make an alternative boot from the working system. This way, if the upgrade messes up, I can boot to the working one.
I also find that I can fix problems most of the time, by upgrading ports to the very latest, and then doing portmaster nvidia-drm-kmod. (not guaranteed, but works most of the time, as drsnx60 did. I find that nvdia-drm-kmod usually pulls in the nvidia-drm-66-kmod).

I somewhat inadvisedly did an upgrade yesterday, and found that I could only get resolution of 1024x768 (monitor is usually 2560x1440) and had to log into work, with that machine. It's a time when having the bectl boot environment was very helpful. (After work, I was able to fix it with my usual method, the portmaster nvidia-drm-kmod)

I have, somehow, the linux-nvidia-libs, and at present, they're a different version, but it doesn't seem to be affecting anything. I'm not sure how I got them, maybe they got pulled in with some other nvidia stuff.

As T-Aoki says, a dry run beforehand can save problems.
 
Back
Top