fix broken nvidia install ubuntu 26.04

fix-nvidia.sh

#!/usr/bin/env bash
# Fix NVIDIA RTX 3050 Mobile on Ubuntu 26.04 (XPS 15 9520)
# Root causes found:
#  1) GRUB has nomodeset -> disables KMS (i915 + nvidia-drm)
#  2) DKMS nvidia modules signed with a private key obscure Canonical-signed modules
#  3) Secure Boot is ON and MOK only has Canonical CA -> DKMS modules cannot load
#  4) Partial server/headless stack installed instead of full desktop driver
set -euo pipefail

if [[ ${EUID:-$(id -u)} -ne 0 ]]; then
  echo "Re-running with sudo..."
  exec sudo -E bash "$0" "$@"
fi

export DEBIAN_FRONTEND=noninteractive
KVER="$(uname -r)"
LOG="/home/tim/fix-nvidia-driver.log"
exec > >(tee -a "$LOG") 2>&1

echo "===== NVIDIA driver fix starting $(date) ====="
echo "Kernel: $KVER"
echo "Secure Boot: $(mokutil --sb-state 2>/dev/null || true)"

echo
echo "[1/7] Remove nomodeset from GRUB"
if grep -q 'nomodeset' /etc/default/grub; then
  cp -a /etc/default/grub "/etc/default/grub.bak.$(date +%Y%m%d%H%M%S)"
  sed -i 's/\bnomodeset\b//g' /etc/default/grub
  # collapse leftover double spaces inside the quotes
  sed -i -E 's/(GRUB_CMDLINE_LINUX_DEFAULT=")[ ]+/\1/; s/[ ]+(")/\1/; s/  +/ /g' /etc/default/grub
  grep '^GRUB_CMDLINE' /etc/default/grub || true
else
  echo "nomodeset already absent"
fi
# also scrub any leftover copies
if [[ -f /etc/default/grub.ucf-dist ]]; then
  sed -i 's/\bnomodeset\b//g' /etc/default/grub.ucf-dist || true
fi
update-grub

echo
echo "[2/7] Remove broken DKMS NVIDIA tree that overrides signed modules"
if dkms status 2>/dev/null | grep -qi nvidia; then
  dkms status || true
  # remove all installed nvidia dkms versions for this and any kernel
  while read -r line; do
    # formats like: nvidia/595.71.05, 7.0.0-28-generic, x86_64: installed
    ver="$(echo "$line" | awk -F'[/ ,]' '{print $2}')"
    k="$(echo "$line" | awk -F'[, ]+' '{print $2}')"
    if [[ -n "${ver:-}" ]]; then
      echo "dkms remove nvidia/$ver -k ${k:-$KVER} || dkms remove nvidia/$ver --all"
      dkms remove "nvidia/$ver" -k "${k:-$KVER}" 2>/dev/null \
        || dkms remove "nvidia/$ver" --all 2>/dev/null \
        || true
    fi
  done < <(dkms status 2>/dev/null | grep -i '^nvidia' || true)
fi
rm -rf /var/lib/dkms/nvidia || true
rm -rf /usr/src/nvidia-595.71.05 || true
# wipe updates/dkms overrides so depmod prefers Canonical-signed modules
if [[ -d /lib/modules/$KVER/updates/dkms ]]; then
  rm -f /lib/modules/$KVER/updates/dkms/nvidia*.ko* || true
  rmdir /lib/modules/$KVER/updates/dkms 2>/dev/null || true
fi
depmod -a "$KVER"

echo
echo "[3/7] Purge partial server/headless NVIDIA packages"
apt-get update -y
# Keep firmware if possible; remove the broken stack metapackages
apt-get purge -y \
  'nvidia-headless*' \
  'nvidia-kernel-source-*-server*' \
  'linux-modules-nvidia-*-server*' \
  'nvidia-compute-utils-*-server' \
  'libnvidia-compute-*-server' \
  'libnvidia-cfg1-*-server' \
  'nvidia-kernel-common-*-server' \
  'nvidia-firmware-*-server*' \
  2>/dev/null || true
# Also catch any leftover nvidia packages that would conflict
apt-get autoremove -y --purge || true

echo
echo "[4/7] Install Ubuntu-recommended signed desktop driver (595-open + prebuilt signed modules)"
# Prefer prebuilt signed modules for this HWE kernel flavour so Secure Boot works
# without enrolling a MOK key.
apt-get install -y \
  linux-modules-nvidia-595-open-generic-hwe-26.04 \
  nvidia-driver-595-open \
  nvidia-prime \
  nvidia-settings

echo
echo "[5/7] Rebuild module deps and ensure blacklists are sane"
depmod -a "$KVER"
# nouveau should stay blacklisted (package conf handles it)
if [[ ! -f /etc/modprobe.d/nvidia-blacklist.conf ]]; then
  cat >/etc/modprobe.d/nvidia-blacklist.conf <<'EOF'
blacklist nouveau
options nouveau modeset=0
EOF
fi
update-initramfs -u -k "$KVER"

echo
echo "[6/7] Configure prime for on-demand (laptop hybrid Intel+NVIDIA)"
if command -v prime-select >/dev/null; then
  # on-demand is the right default for XPS 15 hybrid graphics
  prime-select on-demand || prime-select nvidia || true
  prime-select query || true
fi

echo
echo "[7/7] Try loading modules now (may still need reboot after nomodeset removal)"
# With nomodeset currently active in THIS boot, full modeset may still fail
# until reboot — but Secure Boot + signed module path should work post-reboot.
modprobe -r nouveau 2>/dev/null || true
if modprobe nvidia; then
  echo "nvidia module loaded"
  modprobe nvidia-modeset || true
  modprobe nvidia-drm modeset=1 || modprobe nvidia-drm || true
  modprobe nvidia-uvm || true
  sleep 1
  ls -la /dev/nvidia* 2>/dev/null || true
  nvidia-smi || true
else
  echo "nvidia module failed to load on this live boot (expected if nomodeset still active)."
  echo "dmesg tail:"
  dmesg | tail -40 || true
fi

echo
echo "===== Post-check ====="
echo "Signed package modules:"
modinfo -F signer "/lib/modules/$KVER/kernel/nvidia-"*"/nvidia.ko" 2>/dev/null || \
  find "/lib/modules/$KVER" -path '*nvidia*/nvidia.ko' -exec modinfo -F filename {} \; -exec modinfo -F signer {} \; 2>/dev/null | head -20
echo "DKMS overrides left:"
ls -la "/lib/modules/$KVER/updates/dkms/" 2>/dev/null || echo "(none)"
echo "Current cmdline: $(cat /proc/cmdline)"
echo "GRUB default cmdline:"
grep '^GRUB_CMDLINE' /etc/default/grub || true
echo
echo "DONE. A reboot is required for nomodeset removal and clean driver bring-up."
echo "After reboot run:  nvidia-smi"
echo "Log saved to: $LOG"