This commit is contained in:
CanbiZ
2025-03-27 15:22:43 +01:00
parent 8730d651d8
commit c88bea8220
4 changed files with 130 additions and 150 deletions

View File

@@ -1,7 +1,6 @@
#!/usr/bin/env bash
# NVIDIA GPU Integration for Proxmox LXC
# modular nvidia.func for LXC passthrough
# Author: CanbiZ
# License: MIT
@@ -12,11 +11,19 @@ function nvidia_exit() {
exit 0
}
function msg() {
local type="$1"
shift
case "$type" in
info) printf " \033[36m➤\033[0m %s\n" "$@" ;;
ok) printf " \033[32m✔\033[0m %s\n" "$@" ;;
warn) printf " \033[33m⚠\033[0m %s\n" "$@" >&2 ;;
err) printf " \033[31m✘\033[0m %s\n" "$@" >&2 ;;
esac
}
function nvidia_check_driver_installed() {
if ! command -v nvidia-smi &>/dev/null; then
return 1
fi
return 0
command -v nvidia-smi &>/dev/null
}
function nvidia_get_driver_version() {
@@ -29,30 +36,30 @@ function nvidia_get_cuda_version() {
function nvidia_validate_driver_version() {
if ! nvidia_check_driver_installed; then
printf "✘ NVIDIA drivers not found on host\n"
msg err "NVIDIA drivers not found"
nvidia_exit
fi
local version
version=$(nvidia_get_driver_version)
local major=${version%%.*}
local ver major
ver=$(nvidia_get_driver_version)
major=${ver%%.*}
if ((major < 500)); then
printf "⚠ Detected old NVIDIA driver version: %s\n" "$version"
read -r -p "Proceed anyway? [y/N] " confirm
msg warn "Detected old NVIDIA driver version: $ver"
read -rp "Continue anyway? [y/N]: " confirm
[[ "${confirm,,}" =~ ^(y|yes)$ ]] || nvidia_exit
fi
}
function nvidia_validate_cuda_version() {
if ! nvidia_check_driver_installed; then
printf "✘ NVIDIA drivers not found on host\n"
msg err "NVIDIA drivers not found"
nvidia_exit
fi
local version
version=$(nvidia_get_cuda_version)
local major=${version%%.*}
local ver major
ver=$(nvidia_get_cuda_version)
major=${ver%%.*}
if ((major < 11)); then
printf "⚠ Detected old CUDA version: %s\n" "$version"
read -r -p "Proceed anyway? [y/N] " confirm
msg warn "Detected old CUDA version: $ver"
read -rp "Continue anyway? [y/N]: " confirm
[[ "${confirm,,}" =~ ^(y|yes)$ ]] || nvidia_exit
fi
}
@@ -68,43 +75,34 @@ KERNEL=="nvidia", RUN+="/bin/bash -c '/usr/bin/nvidia-smi -L && chmod 666 /dev/n
KERNEL=="nvidia_uvm", RUN+="/bin/bash -c '/usr/bin/nvidia-modprobe -c0 -u && chmod 0666 /dev/nvidia-uvm*'"
EOF
printf "✔ NVIDIA kernel modules and udev rules applied\n"
printf "⚠ Please reboot the Proxmox host for changes to take effect\n"
msg ok "NVIDIA modules configured"
msg warn "Reboot the host to apply kernel changes"
}
function nvidia_select_gpu_minor() {
local menu=()
local maxwidth=0
while IFS= read -r devdir; do
local pci=${devdir##*/}
local info="/proc/driver/nvidia/gpus/${pci}/information"
local menu=() max=0
while IFS= read -r path; do
local dev="${path##*/}"
local info="/proc/driver/nvidia/gpus/${dev}/information"
[[ -f "$info" ]] || continue
local model minor
model=$(awk -F': ' '/Model:/ {print $2}' "$info")
minor=$(awk '/Device Minor/ {print $NF}' "$info")
menu+=("$minor" "$model" "OFF")
((${#model} > maxwidth)) && maxwidth=${#model}
(( ${#model} > max )) && max=${#model}
done < <(find /proc/driver/nvidia/gpus -mindepth 1 -type d)
if ((${#menu[@]} == 0)); then
printf "✘ No NVIDIA GPU found\n"
return 1
fi
[[ ${#menu[@]} -eq 0 ]] && msg err "No NVIDIA GPU found" && return 1
[[ ${#menu[@]} -eq 3 ]] && printf "%s" "${menu[0]}" && return
if ((${#menu[@]} == 3)); then
printf "%s\n" "${menu[0]}"
else
whiptail --title "NVIDIA GPU Selection" --radiolist \
"Select the GPU to passthrough:" 15 $((maxwidth + 40)) 6 \
"${menu[@]}" 3>&1 1>&2 2>&3
fi
whiptail --title "NVIDIA GPU Selection" --radiolist \
"Select GPU for passthrough:" 15 $((max + 40)) 6 \
"${menu[@]}" 3>&1 1>&2 2>&3
}
function nvidia_lxc_passthrough() {
local container_id="$1"
local minor="$2"
local config="/etc/pve/lxc/${container_id}.conf"
local ctid="$1" minor="$2"
local conf="/etc/pve/lxc/${ctid}.conf"
local devices=(
"/dev/nvidia${minor}"
"/dev/nvidiactl"
@@ -112,22 +110,19 @@ function nvidia_lxc_passthrough() {
"/dev/nvidia-uvm-tools"
)
local devnum_list=()
local devnums=()
for dev in "${devices[@]}"; do
[[ -e "$dev" ]] || continue
local major
major=$(stat -c '%t' "$dev")
local dec_major=$((16#$major))
devnum_list+=("$dec_major")
echo "lxc.mount.entry: $dev ${dev##*/} none bind,optional,create=file" >>"$config"
local major_hex
major_hex=$(stat -c '%t' "$dev")
devnums+=($((16#$major_hex)))
echo "lxc.mount.entry: $dev ${dev##*/} none bind,optional,create=file" >>"$conf"
done
echo "lxc.mount.entry: /dev/dri dev/dri none bind,optional,create=dir" >>"$config"
echo "lxc.mount.entry: /dev/dri dev/dri none bind,optional,create=dir" >>"$conf"
for num in "${devnum_list[@]}"; do
echo "lxc.cgroup2.devices.allow: c ${num}:* rwm" >>"$config"
for n in "${devnums[@]}"; do
echo "lxc.cgroup2.devices.allow: c ${n}:* rwm" >>"$conf"
done
printf "✔ NVIDIA passthrough configured for CT %s\n" "$container_id"
}
msg ok "NVIDIA passt