From d4a2d7fa556d03b90d8e493ad4691b0d03e2bc7d Mon Sep 17 00:00:00 2001
From: Michael Chus <mike@mchus.pro>
Date: Thu, 5 Mar 2026 18:05:57 +0300
Subject: [PATCH] fix: use proprietary NVIDIA .run installer instead of open
 kernel modules

Builds kernel modules from the official NVIDIA installer source tree,
same as a standard NVIDIA driver install. No open-gpu-kernel-modules.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 iso/builder/build-nvidia-module.sh | 93 +++++++++++++-----------------
 1 file changed, 39 insertions(+), 54 deletions(-)

diff --git a/iso/builder/build-nvidia-module.sh b/iso/builder/build-nvidia-module.sh
index 3bd3da4..509e424 100644
--- a/iso/builder/build-nvidia-module.sh
+++ b/iso/builder/build-nvidia-module.sh
@@ -1,14 +1,16 @@
 #!/bin/sh
-# build-nvidia-module.sh — build NVIDIA open kernel modules and extract nvidia-smi
+# build-nvidia-module.sh — install NVIDIA proprietary driver into ISO overlay
 #
-# Builds NVIDIA open-gpu-kernel-modules from source against the installed linux-lts
-# kernel headers. Output is cached in DIST_DIR/nvidia-<version>-<kver>/ so
-# subsequent builds are instant unless NVIDIA_DRIVER_VERSION or kernel changes.
+# Downloads the official NVIDIA .run installer, extracts kernel modules and
+# userspace tools (nvidia-smi, libnvidia-ml). Everything is proprietary NVIDIA.
+#
+# Output is cached in DIST_DIR/nvidia-<version>-<kver>/ so subsequent builds
+# are instant unless NVIDIA_DRIVER_VERSION or kernel version changes.
 #
 # Output layout:
-#   $CACHE_DIR/modules/   — nvidia*.ko files (stripped)
-#   $CACHE_DIR/bin/       — nvidia-smi
-#   $CACHE_DIR/lib/       — libnvidia-ml.so.1, libcuda stub (for nvidia-smi)
+#   $CACHE_DIR/modules/   — nvidia*.ko files
+#   $CACHE_DIR/bin/       — nvidia-smi, nvidia-debugdump
+#   $CACHE_DIR/lib/       — libnvidia-ml.so*, libcuda.so* (for nvidia-smi)
 
 set -e
 
@@ -26,71 +28,54 @@ if [ -z "$KVER" ]; then
     KVER=$(ls /usr/src/ | grep '^linux-headers-' | sed 's/linux-headers-//' | head -1)
 fi
 KDIR="/usr/src/linux-headers-${KVER}"
-echo "=== NVIDIA ${NVIDIA_VERSION} for kernel ${KVER} ==="
+echo "=== NVIDIA ${NVIDIA_VERSION} (proprietary) for kernel ${KVER} ==="
 
 CACHE_DIR="${DIST_DIR}/nvidia-${NVIDIA_VERSION}-${KVER}"
 if [ -d "$CACHE_DIR/modules" ] && [ -f "$CACHE_DIR/bin/nvidia-smi" ]; then
-    echo "=== NVIDIA modules cached, skipping build ==="
+    echo "=== NVIDIA cached, skipping build ==="
     echo "cache: $CACHE_DIR"
-    echo "modules: $(ls $CACHE_DIR/modules/*.ko 2>/dev/null | wc -l) .ko files"
+    echo "modules: $(ls "$CACHE_DIR/modules/"*.ko 2>/dev/null | wc -l) .ko files"
     exit 0
 fi
 
 # Install build dependencies
-echo "=== installing build deps ==="
-apk add --quiet gcc make perl linux-lts-dev wget tar
+apk add --quiet gcc make perl linux-lts-dev wget
 
-# Download and build open kernel modules
-BUILD_TMP="/var/tmp/nvidia-build"
-rm -rf "$BUILD_TMP"
-mkdir -p "$BUILD_TMP"
-
-SRC_TGZ="/var/tmp/nvidia-open-${NVIDIA_VERSION}.tar.gz"
-if [ ! -f "$SRC_TGZ" ]; then
-    echo "=== downloading NVIDIA open kernel modules source ==="
-    wget -q -O "$SRC_TGZ" \
-        "https://github.com/NVIDIA/open-gpu-kernel-modules/archive/refs/tags/${NVIDIA_VERSION}.tar.gz"
-fi
-
-echo "=== extracting source ==="
-tar -xzf "$SRC_TGZ" -C "$BUILD_TMP"
-SRC_DIR="$BUILD_TMP/open-gpu-kernel-modules-${NVIDIA_VERSION}"
-
-echo "=== building kernel modules ($(nproc) cores) ==="
-cd "$SRC_DIR"
-make -j$(nproc) \
-    KERNEL_SOURCE_PATH="$KDIR" \
-    IGNORE_MISSING_MODULE_SYMVERS=1 \
-    modules 2>&1 | tail -5
-
-# Collect .ko files
-mkdir -p "$CACHE_DIR/modules"
-find "$SRC_DIR" -name '*.ko' -exec cp {} "$CACHE_DIR/modules/" \;
-# Strip debug info from modules to reduce size
-for ko in "$CACHE_DIR"/modules/*.ko; do
-    strip --strip-debug "$ko" 2>/dev/null || true
-done
-echo "modules: $(ls $CACHE_DIR/modules/*.ko | wc -l) .ko files"
-
-# Extract nvidia-smi and required libraries from the .run installer
+# Download official NVIDIA .run installer (proprietary)
 RUN_FILE="/var/tmp/NVIDIA-Linux-x86_64-${NVIDIA_VERSION}.run"
 if [ ! -f "$RUN_FILE" ]; then
-    echo "=== downloading NVIDIA installer (for nvidia-smi) ==="
-    wget -q -O "$RUN_FILE" \
+    echo "=== downloading NVIDIA ${NVIDIA_VERSION} installer ==="
+    wget -q --show-progress \
+        -O "$RUN_FILE" \
         "https://download.nvidia.com/XFree86/Linux-x86_64/${NVIDIA_VERSION}/NVIDIA-Linux-x86_64-${NVIDIA_VERSION}.run"
 fi
 
-echo "=== extracting nvidia-smi ==="
+# Extract installer contents
+echo "=== extracting installer ==="
 chmod +x "$RUN_FILE"
 EXTRACT_DIR="/var/tmp/nvidia-extract-${NVIDIA_VERSION}"
 rm -rf "$EXTRACT_DIR"
-"$RUN_FILE" --extract-only --target "$EXTRACT_DIR" 2>/dev/null || true
+"$RUN_FILE" --extract-only --target "$EXTRACT_DIR"
 
-mkdir -p "$CACHE_DIR/bin" "$CACHE_DIR/lib"
-cp "$EXTRACT_DIR/nvidia-smi"               "$CACHE_DIR/bin/"
-cp "$EXTRACT_DIR/libnvidia-ml.so.1"        "$CACHE_DIR/lib/" 2>/dev/null || true
-cp "$EXTRACT_DIR/libnvidia-ml.so."*        "$CACHE_DIR/lib/" 2>/dev/null || true
+# Build kernel modules from extracted source
+echo "=== building kernel modules ($(nproc) cores) ==="
+cd "$EXTRACT_DIR/kernel"
+make -j$(nproc) KERNEL_UNAME="$KVER" SYSSRC="$KDIR" modules 2>&1 | tail -5
+
+# Collect outputs
+mkdir -p "$CACHE_DIR/modules" "$CACHE_DIR/bin" "$CACHE_DIR/lib"
+
+find "$EXTRACT_DIR/kernel" -name '*.ko' -exec cp {} "$CACHE_DIR/modules/" \;
+for ko in "$CACHE_DIR/modules/"*.ko; do
+    strip --strip-debug "$ko" 2>/dev/null || true
+done
+
+cp "$EXTRACT_DIR/nvidia-smi"         "$CACHE_DIR/bin/"
+cp "$EXTRACT_DIR/libnvidia-ml.so."*  "$CACHE_DIR/lib/" 2>/dev/null || true
+# libcuda stub needed by nvidia-smi at runtime
+cp "$EXTRACT_DIR/libcuda.so."*       "$CACHE_DIR/lib/" 2>/dev/null || true
 
 echo "=== NVIDIA build complete ==="
 echo "cache: $CACHE_DIR"
-ls -lh "$CACHE_DIR/bin/" "$CACHE_DIR/modules/"
+echo "modules: $(ls "$CACHE_DIR/modules/"*.ko | wc -l) .ko files"
+ls -lh "$CACHE_DIR/bin/"