diff --git a/misc/tools.func b/misc/tools.func index 7e18d475d..29a95c8aa 100644 --- a/misc/tools.func +++ b/misc/tools.func @@ -2691,7 +2691,7 @@ function setup_hwaccel() { else # Multiple GPUs - show selection menu echo "" - msg_info "Multiple GPUs detected:" + msg_custom "⚠" "${YW}" "Multiple GPUs detected:" echo "" for i in "${!GPU_LIST[@]}"; do local type_display="${GPU_TYPES[$i]}" @@ -3010,16 +3010,24 @@ _setup_nvidia_gpu() { msg_info "Installing NVIDIA GPU drivers" + # Prevent interactive dialogs (e.g., "Mismatching nvidia kernel module" whiptail) + export DEBIAN_FRONTEND=noninteractive + export NEEDRESTART_MODE=a + # Detect host driver version (passed through via /proc) + # Format varies by driver type: + # Proprietary: "NVRM version: NVIDIA UNIX x86_64 Kernel Module 550.54.14 Thu..." + # Open: "NVRM version: NVIDIA UNIX Open Kernel Module for x86_64 590.48.01 Release..." + # Use regex to extract version number (###.##.## pattern) local nvidia_host_version="" if [[ -f /proc/driver/nvidia/version ]]; then - nvidia_host_version=$(grep "NVRM version:" /proc/driver/nvidia/version 2>/dev/null | awk '{print $8}') + nvidia_host_version=$(grep -oP '\d{3,}\.\d+\.\d+' /proc/driver/nvidia/version 2>/dev/null | head -1) fi if [[ -z "$nvidia_host_version" ]]; then msg_warn "NVIDIA host driver version not found in /proc/driver/nvidia/version" msg_warn "Ensure NVIDIA drivers are installed on host and GPU passthrough is enabled" - $STD apt -y install va-driver-all vainfo 2>/dev/null || true + $STD apt-get -y install va-driver-all vainfo 2>/dev/null || true return 0 fi @@ -3032,53 +3040,115 @@ _setup_nvidia_gpu() { sed -i -E 's/Components: (.*)$/Components: \1 contrib non-free non-free-firmware/g' /etc/apt/sources.list.d/debian.sources 2>/dev/null || true fi fi + $STD apt-get -y update 2>/dev/null || msg_warn "apt update failed - continuing anyway" - # Determine CUDA repository - local cuda_repo="debian12" - case "$os_codename" in - bullseye) cuda_repo="debian11" ;; - bookworm) cuda_repo="debian12" ;; - trixie | sid) cuda_repo="debian12" ;; # Forward compatible - esac + # For Debian 13 Trixie/Sid: Use Debian's own nvidia packages first (better compatibility) + # NVIDIA's CUDA repo targets Debian 12 and may not have amd64 packages for Trixie + if [[ "$os_codename" == "trixie" || "$os_codename" == "sid" ]]; then + msg_info "Debian ${os_codename}: Using Debian's NVIDIA packages" - # Add NVIDIA CUDA repository - if [[ ! -f /usr/share/keyrings/cuda-archive-keyring.gpg ]]; then - msg_info "Adding NVIDIA CUDA repository (${cuda_repo})" - local cuda_keyring - cuda_keyring="$(mktemp)" - if curl -fsSL -o "$cuda_keyring" "https://developer.download.nvidia.com/compute/cuda/repos/${cuda_repo}/x86_64/cuda-keyring_1.1-1_all.deb" 2>/dev/null; then - $STD dpkg -i "$cuda_keyring" 2>/dev/null || true + # Extract major version for flexible matching (580.126.09 -> 580) + local nvidia_major_version="${nvidia_host_version%%.*}" + + # Check what versions are actually available + local available_version="" + available_version=$(apt-cache madison libcuda1 2>/dev/null | awk '{print $3}' | grep -E "^${nvidia_major_version}\." | head -1 || true) + + if [[ -n "$available_version" ]]; then + msg_info "Found available NVIDIA version: ${available_version}" + local nvidia_pkgs="libcuda1=${available_version} libnvcuvid1=${available_version} libnvidia-encode1=${available_version} libnvidia-ml1=${available_version}" + if $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends $nvidia_pkgs 2>/dev/null; then + msg_ok "Installed NVIDIA libraries (${available_version})" + else + msg_warn "Failed to install NVIDIA ${available_version} - trying unversioned" + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null || true + fi else - msg_warn "Failed to download NVIDIA CUDA keyring" + # No matching major version - try latest available or unversioned + msg_warn "No NVIDIA packages for version ${nvidia_major_version}.x found in repos" + available_version=$(apt-cache madison libcuda1 2>/dev/null | awk '{print $3}' | head -1 || true) + if [[ -n "$available_version" ]]; then + msg_info "Trying latest available: ${available_version} (may cause version mismatch)" + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends \ + libcuda1="${available_version}" libnvcuvid1="${available_version}" \ + libnvidia-encode1="${available_version}" libnvidia-ml1="${available_version}" 2>/dev/null || + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends \ + libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null || + msg_warn "NVIDIA library installation failed - GPU compute may not work" + else + msg_warn "No NVIDIA packages available in Debian repos - GPU support disabled" + fi fi - rm -f "$cuda_keyring" - fi + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends nvidia-smi 2>/dev/null || true - # Pin NVIDIA repo for version matching - cat <<'NVIDIA_PIN' >/etc/apt/preferences.d/nvidia-cuda-pin + else + # Debian 11/12: Use NVIDIA CUDA repository for version matching + local cuda_repo="debian12" + case "$os_codename" in + bullseye) cuda_repo="debian11" ;; + bookworm) cuda_repo="debian12" ;; + esac + + # Add NVIDIA CUDA repository + if [[ ! -f /usr/share/keyrings/cuda-archive-keyring.gpg ]]; then + msg_info "Adding NVIDIA CUDA repository (${cuda_repo})" + local cuda_keyring + cuda_keyring="$(mktemp)" + if curl -fsSL -o "$cuda_keyring" "https://developer.download.nvidia.com/compute/cuda/repos/${cuda_repo}/x86_64/cuda-keyring_1.1-1_all.deb" 2>/dev/null; then + $STD dpkg -i "$cuda_keyring" 2>/dev/null || true + else + msg_warn "Failed to download NVIDIA CUDA keyring" + fi + rm -f "$cuda_keyring" + fi + + # Pin NVIDIA repo for version matching + cat <<'NVIDIA_PIN' >/etc/apt/preferences.d/nvidia-cuda-pin Package: * Pin: origin developer.download.nvidia.com Pin-Priority: 1001 NVIDIA_PIN - $STD apt -y update + $STD apt-get -y update 2>/dev/null || msg_warn "apt update failed - continuing anyway" - # Install version-matched NVIDIA libraries - local nvidia_pkgs="libcuda1=${nvidia_host_version}* libnvcuvid1=${nvidia_host_version}* libnvidia-encode1=${nvidia_host_version}* libnvidia-ml1=${nvidia_host_version}*" + # Extract major version for flexible matching (580.126.09 -> 580) + local nvidia_major_version="${nvidia_host_version%%.*}" - msg_info "Installing NVIDIA libraries (version ${nvidia_host_version})" - if $STD apt -y install --no-install-recommends $nvidia_pkgs 2>/dev/null; then - msg_ok "Installed version-matched NVIDIA libraries" - else - msg_warn "Version-pinned install failed - trying unpinned" - if $STD apt -y install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null; then - msg_warn "Installed NVIDIA libraries (unpinned) - version mismatch may occur" + # Check what versions are actually available in CUDA repo + local available_version="" + available_version=$(apt-cache madison libcuda1 2>/dev/null | awk '{print $3}' | grep -E "^${nvidia_major_version}\." | head -1 || true) + + if [[ -n "$available_version" ]]; then + msg_info "Installing NVIDIA libraries (version ${available_version})" + local nvidia_pkgs="libcuda1=${available_version} libnvcuvid1=${available_version} libnvidia-encode1=${available_version} libnvidia-ml1=${available_version}" + if $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends $nvidia_pkgs 2>/dev/null; then + msg_ok "Installed version-matched NVIDIA libraries" + else + msg_warn "Version-pinned install failed - trying unpinned" + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null || + msg_warn "NVIDIA library installation failed" + fi else - msg_warn "NVIDIA library installation failed" + msg_warn "No NVIDIA packages for version ${nvidia_major_version}.x in CUDA repo (host: ${nvidia_host_version})" + # Try latest available version + available_version=$(apt-cache madison libcuda1 2>/dev/null | awk '{print $3}' | head -1 || true) + if [[ -n "$available_version" ]]; then + msg_info "Trying latest available: ${available_version} (version mismatch warning)" + if $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends \ + libcuda1="${available_version}" libnvcuvid1="${available_version}" \ + libnvidia-encode1="${available_version}" libnvidia-ml1="${available_version}" 2>/dev/null; then + msg_ok "Installed NVIDIA libraries (${available_version}) - version differs from host" + else + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends libcuda1 libnvcuvid1 libnvidia-encode1 libnvidia-ml1 2>/dev/null || + msg_warn "NVIDIA library installation failed" + fi + else + msg_warn "No NVIDIA packages available in CUDA repo - GPU support disabled" + fi fi - fi - $STD apt -y install --no-install-recommends nvidia-smi 2>/dev/null || true + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends nvidia-smi 2>/dev/null || true + fi elif [[ "$os_id" == "ubuntu" ]]; then # Ubuntu versioning @@ -3102,20 +3172,33 @@ NVIDIA_PIN rm -f "$cuda_keyring" fi - $STD apt -y update + $STD apt-get -y update 2>/dev/null || msg_warn "apt update failed - continuing anyway" - # Try version-matched install - local nvidia_pkgs="libcuda1=${nvidia_host_version}* libnvcuvid1=${nvidia_host_version}* libnvidia-encode1=${nvidia_host_version}* libnvidia-ml1=${nvidia_host_version}*" - if $STD apt -y install --no-install-recommends $nvidia_pkgs 2>/dev/null; then - msg_ok "Installed version-matched NVIDIA libraries" + # Extract major version for flexible matching + local nvidia_major_version="${nvidia_host_version%%.*}" + + # Check what versions are available + local available_version="" + available_version=$(apt-cache madison libcuda1 2>/dev/null | awk '{print $3}' | grep -E "^${nvidia_major_version}\." | head -1 || true) + + if [[ -n "$available_version" ]]; then + msg_info "Installing NVIDIA libraries (version ${available_version})" + local nvidia_pkgs="libcuda1=${available_version} libnvcuvid1=${available_version} libnvidia-encode1=${available_version} libnvidia-ml1=${available_version}" + if $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends $nvidia_pkgs 2>/dev/null; then + msg_ok "Installed version-matched NVIDIA libraries" + else + # Fallback to Ubuntu repo packages + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends libnvidia-decode libnvidia-encode nvidia-utils 2>/dev/null || msg_warn "NVIDIA installation failed" + fi else + msg_warn "No NVIDIA packages for version ${nvidia_major_version}.x in CUDA repo" # Fallback to Ubuntu repo packages - $STD apt -y install --no-install-recommends libnvidia-decode libnvidia-encode nvidia-utils 2>/dev/null || msg_warn "NVIDIA installation failed" + $STD apt-get -y -o Dpkg::Options::="--force-confold" install --no-install-recommends libnvidia-decode libnvidia-encode nvidia-utils 2>/dev/null || msg_warn "NVIDIA installation failed" fi fi # VA-API for hybrid setups (Intel + NVIDIA) - $STD apt -y install va-driver-all vainfo 2>/dev/null || true + $STD apt-get -y install va-driver-all vainfo 2>/dev/null || true msg_ok "NVIDIA GPU configured" } @@ -3215,6 +3298,10 @@ EOF $STD apt -y update } +# ══════════════════════════════════════════════════════════════════════════════ +# Helper: Setup GPU device permissions +# ══════════════════════════════════════════════════════════════════════════════ + # ══════════════════════════════════════════════════════════════════════════════ # Helper: Setup GPU device permissions # ══════════════════════════════════════════════════════════════════════════════