Differences

This shows you the differences between two versions of the page.

--- linux64 [2026/09/17 08:40] – [Kernel 7.2.5 boot issues] bengt
+++ linux64 [2026/09/17 09:49] (current) – [ollama in podman for ai] bengt
@@ Line 892: / Line 892: @@
 Configure podman:
   # dnf install nvidia-container-toolkit
-  # nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml
+  # nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml # Needs to be rerun after nvidia driver updates and when switching hdmi port between cards
   # nvidia-ctk cdi list
   # podman run --rm --security-opt label=disable --device nvidia.com/gpu=1 ubuntu nvidia-smi # Check gpu index to the one you want to use
@@ Line 1306: / Line 1306: @@
 ==== Kernel 7.2.5 boot issues ====
-Increased the number of stored kernels in /etc/dnf/dnf.conf to 5 and set the default to the previous kernel. You will need to switch manually to latest kernel during boot until this is resolved upstead. When fixed use "grubby --set-default-index=0" to restore normal boot behaviour.
+This is the likely issue, root= disappearing from the kernel line triggering auto-detection:
+https://discussion.fedoraproject.org/t/error-at-boot-with-kernel-6-17-12-300-fc43-x86-64/177533
-This is an old ticket that the AI found and some others:
+It also suggested adding the root UUID to the kernel paremeters, the others have it:
-https://bugzilla.redhat.com/show_bug.cgi?id=2350605
+  # grep -E -R 'title|options' /boot/loader/entries/
-https://discussion.fedoraproject.org/t/fedora-44-kde-kernel-7-2-5-black-screen/201990/8
+  # grubby --update-kernel=/boot/vmlinuz-7.2.5-100.fc43.x86_64 --args="root=UUID=7cbb7c6a-5996-48d4-8378-4ba59e5f1f52 ro rhgb quiet"
+It turns out this is lacking from the new boot entry.
-Then it suggested to use kernel boot option systemd.gpt_auto=0 to fix it. Did not work. Some more failed attempts:
+Possible permanent fix:
-  # akmods --rebuild --force --kernel 7.2.5-100.fc43.x86_64
+  # echo "root=UUID=7cbb7c6a-5996-48d4-8378-4ba59e5f1f52 ro rhgb quiet rd.driver.blacklist=nouveau,nova_core modprobe.blacklist=nouveau,nova_core" > /etc/kernel/cmdline
-  # dracut --force --kver 7.2.5-100.fc43.x86_64
+Or just remove that file and trust modern grubenv kernel-install handling:\
-It also suggested adding the root UUID to the kernel p1rameters. Will wait for an update instead I think...
+If /etc/kernel/cmdline does not exist, the installer automatically falls back to reading the kernelopts line straight out of your grubenv file to build new kernel configurations.If /etc/kernel/cmdline does exist, the modern BLS (Boot Loader Specification) standard dictates that it takes absolute priority. The script reads only /etc/kernel/cmdline and completely ignores the grubenv file.
-  systemd.gpt_auto=0 root=UUID=7cbb7c6a-5996-48d4-8378-4ba59e5f1f52
-The Exact Technical CauseWhen you cloned your drive using dd, your new NVMe drive inherited a hidden GPT Partition Attribute Flag called the Discoverable Partitions Specification (DPS) [INDEX].Here is why your old kernel works but the new 7.2.5 kernel fails:
+For ollama to work again:
-The Systemd Change: In the newest systemd updates shipped with your 7.2.5 package stack, systemd-gpt-auto-generator shifts from being a passive fallback tool to an aggressive early-boot tool [INDEX, INDEX]. It searches the raw disk geometry for that cloned GPT root tag before reading your healthy /etc/fstab file [INDEX].
+  # nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml
-The Race Condition: The generator tries to assemble a virtual mapping called /dev/gpt-auto-root [INDEX]. However, because your system has multiple data hard drives and complex bind mounts attached, the automated module triggers a structural race condition [INDEX]. It takes too long to map the drive paths, hits an exact 90-second systemd target timer, and panics into emergency mode [INDEX, INDEX].
 ==== Upgrade Fedora ... -> 35 -> 36 -> 37 -> 38 -> 39 -> 40 -> 41 -> 42 -> 43 ====