Hi all,
I’ve been running Qubes OS R4.2 dom0 (Fedora 37) with an NVIDIA RTX 4070 (AD104) using the nouveau driver for months without issue. About a month ago, after entering my LUKS2 passphrase, I now get only a black screen. The system boots fine otherwise — TTY works (Ctrl+Alt+F2) — but no graphical desktop appears. Lightdm start and is running
when I check status, but it does not appear graphically.
What changed:
Around early July there were updates to linux-firmware, linux-firmware-whence and the kernel. This is also when the problem began.
What I have tried so far:
(I can log into tty.)
Booting older kernels: 6.12.21, 6.12.25, 6.12.37, 6.13.7 — all show the same black screen.
Rolling back linux-firmware to multiple versions:
2025-07-08 (latest at the time)
2025-06-27
2025-04-10
Downloaded from repo with curl and transferred these older RPMs from another qube, installed in dom0, and rebuilt initramfs for each tested kernel:
sudo dracut --force --kver <kernel_version>
Verified AD104 firmware blobs are present in initramfs (lsinitrd | grep nvidia lists many .bin files).
Booted each kernel with its corresponding rebuilt initramfs — same result: black screen after LUKS2 unlock.
dmesg consistently shows:
nouveau : gsp ctor failed : -2
nouveau: probe with driver nouveau failed with error -2
Related reports:
Red Hat Bug 2375308 — GSP firmware init failure on Ada GPUs
Red Hat Bug 2375857
Similar Qubes forum thread
Summary:
Even after rolling back firmware to known working versions and rebuilding initramfs for multiple older kernels, the issue persists.
Regard.
EDIT
I found a work around!
ATTENTION - THIS FIX IS AT YOUR OWN RISK AND THERE IS NO WARRANTY ABOUT THE SECURITY.
I did not think of any threat model doing this. This is probably out of QubesOS security guidelines since we download a rpm from the web from the default personal appvm and then copy it into dom0 and install.
You will have to downgrade the firmware.
if needed, in /etc/default/grub, add 3 nomodeset modprobe.blacklist=nouveau
then
sudo grub2-mkconfig -o /boot/efi/EFI/qubes/grub.cfg
for convenience(this assume you use UEFI).
you will have to download linux-firmware, linux-firmware-whence and nvidia-gpu-firmware
from a older version and then install it and then do a dracut for each kernel, then remove the blacklist and nomodeset 3, then another grub2-mkconfig.
Personnaly I used firmware version 20250509 for both linux-firmware and nvidia-gpu-firmware and I boot with kernel 6.14.4-1. Then use dnf versionlock add linux-firmware nvidia-gpu-firmware
You need to install it with python3-dnf-plugins-extras-versionlock.
To list all package use this:
qvm-run --pass-io personal ‘curl -s
https://yum.qubes-os.org/r4.2/current/dom0/fc37/rpm/’ | grep linux-firmware
same for linux-firmware-whence and nvidia-gpu-firmware
then download with
qvm-run personal ‘curl -fL -O https://yum.qubes-os.org/r4.2/current/dom0/fc37/rpm/nvidia-gpu-firmware-20250509-1.fc37.noarch.rpm’
(optional)Verify the presence and type of rpm
qvm-run -p personal ‘file *.rpm’
They have to show RPM v3.0 bin etc.
then transfer with
qvm-run --pass-io personal ‘cat ~/nvidia-gpu-firmware-20250509-1.fc37.noarch.rpm’ > ~/nvidia-gpu-firmware-20250509-1.fc37.noarch.rpm
Do the same with linux-firmware and linux-firmware-whence.
Verify signature with :
rpm -q --qf “%{NAME}-%{VERSION}-%{RELEASE} : %{SIGPGP:pgpsig}\n” linux-firmware nvidia-gpu-firmware
compare with
gpg --show-keys /etc/pki/rpm-gpg/* | less
then install both:
sudo dnf install --disablerepo=* --setopt=install_weak_deps=False ./nvidia-gpu-firmware-20250509-1.fc37.noarch.rpm
then regenerate initramfs.
sudo dracut --force --kver 6.14.4-1.qubes.fc37.x86_64
(optional)Verify the presence of .img
ls -lh /boot/initramfs-*.img
then
sudo grub2-mkconfig -o /boot/efi/EFI/qubes/grub.cfg