Announcement

Collapse
No announcement yet.

Today's kernel update loaded multiple kernel versions, none of which work

Collapse
This topic is closed.
X
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    #16
    Originally posted by Silent Observer View Post
    Apparently, the nVidia blob isn't getting into these newer kernels (why? Dunno).
    The connector module that hooks the non-free blob to the kernel needs to be built for each kernel, but the DKMS system should take care of that automatically.
    It can fail, of course. When updating kernels on Nvidia, it might be useful to do so using apt and check the output - you will see the rebuild process and any failures there.

    It is possible to try rebuilding all the modules for each kernel, though of course, it is not straightforward.

    Here is a one-line example

    Using this, it showed me that my OpenRGB (don't ask!) modules weren't rebuilt for most of my kernels via DKMS for some reason, and the ones for 5.15.0-70 failed, though I do have the modules installed, supposedly.
    Code:
    depmod...
    [ OK ]
    * dkms: running auto installation service for kernel 5.15.0-70-generic [ OK ]
    * dkms: running auto installation service for kernel 5.15.0-70
    Error! Your kernel headers for kernel 5.15.0-70 cannot be found at /lib/modules/5.15.0-70/build or /lib/modules/5.1
    5.0-70/source.
    You can use the --kernelsourcedir option to tell DKMS where it's located.
    [ OK ]
    But the one-liner might be useful to see if there is a failure to build for some reason,. or it might just fix the issue, assuming that broken Nvidia drivers are the actual problem here.
    (Trawling through log files via KSystemLog can be useful here)
    Last edited by claydoh; Apr 20, 2023, 08:09 AM.

    Comment


      #17
      Well, it's taking some time, but using ubuntu-drivers to install the upgraded nVidia seems to be doing something useful. It rebuilt the module in the 5.15.0-67-generic kernel I'm running,. It did tell me I could "autoremove" 5.15.0-69-generic, 5.19.0-38-generic, and 5.15.0-1018-nvidia, as well as skipping 6.1.0-1008-oem due to lack of headers. I don't see anything indicating it improved the newly upgraded HWE kernel, 5.19.0-40-generic. I'm going to try a reboot and see what I get, then I'll try claydoh's one-liner link.

      EDIT: Okay, two restarts later, I'm on kernel 5.19.0-40-generic, the newest HWE version -- which failed to boot to GUI immediately before, but now seems fine after I booted to recovery mode and then selected "resume"... Same is also true of 5.15.0-69-generic, which runs after booting to recovery mode and selecting "resume". Is my computer possessed? I'm going to try booting back into 5.19.0-40-generic and rerunning the ubuntu-drivers nVidia install and see if it fixes this kernel (since it seemed to mainly work on the kernel I was running at the time).

      EDIT 2: No. It did not. Claimed "all available drivers are already installed" whether I tried autoinstall, or explicitly entered a version (even trying to backstep to the 515 version). Driver Manager in the GUI would let me mark the 515 driver but had the "apply" button grayed out. I didn't rry it from Synaptic, since I've had bad experiences using that method (I've always used Driver Manager). Meanwhile, the one-liner from claydoh's link rebuilt for 5.19.0-40, 5.19.0-38, 5.15.0-70, but failed to find headers for the older versions (removed by updates?). One more reboot...

      EDIT 3: Aaaaaand, BINGO! Booted straight to the KDE Plasma login screen in 5.19.0-40-generic HWE kernel. Now off to figure out how to tell Customize GRUB to revert to original default start on the highest version number kernel...

      Thanks a lot to both of you, Schwartzer and claydoh -- now the only question is how dkms got uninstalled? I had to explicitly install it to run the status back up thread, which is probably why I stopped getting my nVidia blob built into my kernel updates.

      EDIT 4: Well, crap. Now BOINC task Einstine@Home is reporting "GPU Missing." Looks like I didn't get the Cuda driver. That was working before this driver fix, let me check my nVidia controls. Yep, the nVidia X control panel is working the way I'm used to. The *compute* libraries are installed (says Synaptic). Going to try another restart.

      EDIT 5: GPU Missing is the case for every kernel I've tried. 5.19.0-40, 5.15.0-70, and 5.15.0-67 (which worked and GPU computing worked before all this).
      Last edited by Silent Observer; Apr 20, 2023, 04:58 PM.

      Comment


        #18
        The "GPU Missing" problem is resolved. One of two things -- either I need to update the flatpak every time I get an updated nVidia driver, or by coincidence the flatpak had an update while this was occurring -- but updating the flatpak restored the BOINC client's ability to detect my GTx 1070 when it started and computation is back to normal for GPU tasks. I think (for the first time since installing 22.04 in December) I now have everything working the way I expect it to.

        Comment


          #19
          And now it's back. I had an update this morning that included nvidia 535 drivers and ancillary software, as well as an X update -- and about the time I'd have expected the update to complete, the screen went black. I waited several minutes (ten or so), then used the reset button to restart the machine; after restarting, I'd get a boot hang at the point where it should have switched to a graphics mode (not even a full boot through to black screen with text cursor in upper left, as used to happen if X and nvidia were mismatched). Restarted in recovery mode, checked a couple things (I'm NOT a Linux guru by any stretch), and selected "resume" and the system booted to GUI, seemingly using the Nouveau driver, since BOINC reports "GPU missing" and the nVidia control panel doesn't show any hardware information. Running lspci does show the GPU (as expected).

          Since then, I've updated everything available in flatpak (BOINC is a flatpak, I also have MEGA file transfer, MESA, nVidia, openh264, and GNOME Application Platform, the latter two were already up to date) without changing anything, used ubuntu-drivers to manually install that version of nvidia-535 -- and still not fixed. I've probably rebooted a dozen time as I went.

          Suggestions?

          Comment


            #20
            You could try this - replace 470 with 535 (which seems to be the Nvidia driver version you need):

            https://www.kubuntuforums.net/forum/...896#post671896

            Additionally do a sudo apt update && sudo apt full-upgrade, a flatpak update and a flatpak uninstall --unused in Konsole after the restart.
            Last edited by Schwarzer Kater; Jul 03, 2023, 02:12 PM.
            Debian KDE & LXQt • Kubuntu & Lubuntu • openSUSE KDE • Windows • macOS X
            Desktop: Lenovo ThinkCentre M75s • Laptop: Apple MacBook Pro 13" • and others

            get rid of Snap script (20.04 +)reinstall Snap for release-upgrade script (20.04 +)
            install traditional Firefox script (22.04 +)​ • install traditional Thunderbird script (24.04)

            Comment


              #21
              I started that, but then got this:
              Code:
              E: dpkg was interrupted, you must manually run 'sudo dpkg --configure -a' to correct the problem.
              When I issued the suggested command, installation of nvidia-535 picked up where it had apparently been when the screen went black originally (no wonder it wasn't working). After rebooting, everything seems fine...
              Last edited by Silent Observer; Jul 03, 2023, 03:31 PM.

              Comment

              Working...
              X