Add platform related scripts, faq and known issues

vllm/FAQ.md

@@ -0,0 +1,113 @@
+# Table of Contents
+
+## Installation
+- [Can I run the platform benchmark under a bare-metal Ubuntu environment?](#can-i-run-the-platform-benchmark-under-a-bare-metal-ubuntu-environment)
+- [Can I use Ubuntu 24.04 LTS as the base OS?](#can-i-use-ubuntu-2404-lts-as-the-base-os)
+- [Why can't I see the desktop even with Ubuntu 25.04 desktop version installed?](#why-cant-i-see-the-desktop-even-with-ubuntu-2504-desktop-version-installed)
+- [Can I update the kernel version or other drivers of Ubuntu to get the latest fixes?](#can-i-update-the-kernel-version-or-other-drivers-of-ubuntu-to-get-the-latest-fixes)
+- [Why do I need to run `native_bkc_setup.sh` before using the `vllm/platform` Docker image?](#why-do-i-need-to-run-native_bkc_setupsh-before-using-the-vllmplatform-docker-image)
+
+## Hardware & Firmware
+- [No re-sizable BAR configuration in my BIOS. What can I do to enable B60 with a larger BAR2 size?](#no-re-sizable-bar-configuration-in-my-bios-what-can-i-do-to-enable-b60-with-a-larger-bar2-size)
+- [Maxsun 2x GPU Card Not Detected Behind PCIe Switch](#maxsun-2x-gpu-card-not-detected-behind-pcie-switch)
+
+## Benchmarking
+- [Why do I see unusually high Device-to-Device bandwidth in `ze_peak` benchmark?](#why-do-i-see-unusually-high-device-to-device-bandwidth-in-ze_peak-benchmark)
+- [How can I verify if the benchmark data from `platform_basic_evaluation.sh` is valid?](#how-can-i-verify-if-the-benchmark-data-from-platform_basic_evaluationsh-is-valid)
+
+## Tools
+- [Why can't I see `xpu-smi` in the `vllm` Docker image?](#why-cant-i-see-xpu-smi-in-the-vllm-docker-image)
+- [Why can't I see GPU utilization with `xpu-smi`?](#why-cant-i-see-gpu-utilization-with-xpu-smi)
+
+---
+
+# Installation
+
+## Can I run the platform benchmark under a bare-metal Ubuntu environment?
+
+Yes. Please contact the Intel support team to obtain an offline installer for native setup.
+We also plan to make the offline installer publicly available on the Intel RDC website in an upcoming release.
+
+## Can I use Ubuntu 24.04 LTS as the base OS? {#can-i-use-ubuntu-2404-lts-as-the-base-os}
+
+Not yet. Support for Ubuntu 24.04 LTS is planned in future releases (targeting late 2025).
+
+## Why can't I see the desktop even with Ubuntu 25.04 desktop version installed? {#why-cant-i-see-the-desktop-even-with-ubuntu-2504-desktop-version-installed}	
+
+Some versions of Ubuntu may default to text mode (multi-user target) after installation. You can check the current mode:
+
+```bash
+sudo systemctl get-default
+```
+
+If it returns `multi-user.target`, you can switch to graphical mode:
+
+```bash
+sudo systemctl set-default graphical.target
+sudo reboot
+```
+
+## Can I update the kernel version or other drivers of Ubuntu to get the latest fixes?
+
+During the evaluation phase, we **do not recommend updating the kernel or system packages** to ensure consistency with the validated environment.
+Any updates may affect stability or introduce compatibility issues with pre-installed components.
+
+## Why do I need to run `native_bkc_setup.sh` before using the `vllm/platform` Docker image?
+
+To ensure consistent kernel and firmware behavior, `native_bkc_setup.sh` is required to unify Linux kernel version and install B60 GuC/HuC firmware directly on the host system before using the container image.
+
+---
+
+# Hardware & Firmware
+
+## No re-sizable BAR configuration in my BIOS. What can I do to enable B60 with a larger BAR2 size?
+
+Please contact your AIB (Add-In-Board) vendor to request the latest IFWI (firmware image) with max re-sizable BAR pre-configured.
+This setup has been validated on Gunnir and Maxsun B60 cards.
+
+## Maxsun 2x GPU Card Not Detected Behind PCIe Switch
+
+Many PCIe switch firmware versions do not support PCIe bifurcation, which prevents detection of dual-GPU cards like Maxsun 2x.
+
+Solution: A firmware update for the PCIe switch is required.
+The Broadcom PEX 89104 has been validated. Please contact your PCIe switch vendor for support or an updated firmware.
+
+---
+
+# Benchmarking
+
+## Why do I see unusually high Device-to-Device bandwidth in `ze_peak` benchmark?
+
+Please export the following environment variable before running ze_peak.
+
+```bash
+export NEOReadDebugKeys=1
+export RenderCompressedBuffersEnabled=0
+```
+
+## How can I verify if the benchmark data from `platform_basic_evaluation.sh` is valid?
+
+Sample benchmark results are available in:
+
+```
+/opt/intel/multi-arc/results
+```
+
+These data points are collected from internal evaluations using an Intel® Xeon® W5-2545X system with dual B60 GPUs.
+> **Disclaimer**: This reference is provided for informational purposes only and should not be interpreted as official performance indicators or guarantees. Actual results may vary depending on hardware configuration, software stack, and usage scenarios.
+
+---
+
+# Tools
+
+## Why can't I see `xpu-smi` in the `vllm` Docker image?
+
+Due to release process limitations, `xpu-smi` is currently not included in the official `vllm` Docker image.
+We plan to add it in the next release. In the meantime, you may install it manually using:
+
+[xpu-smi 1.3.1 on GitHub](https://github.com/intel/xpumanager/releases/download/V1.3.1/xpumanager_1.3.1_20250724.061629.60921e5e_u24.04_amd64.deb)
+
+## Why can't I see GPU utilization with `xpu-smi`?
+
+GPU utilization metrics are not yet fully supported by `xpu-smi` in the current release.
+This functionality is scheduled to be added in next release.

vllm/KNOWN_ISSUES.md

@@ -0,0 +1,13 @@
+
+# 01. System Hang During Ubuntu 25.04 Installation with B60 Card Plugged In
+The issue is caused by an outdated GPU GuC firmware bundled in the official Ubuntu 25.04 Desktop ISO image.
+
+Workaround: Remove the B60 card before starting the Ubuntu installation, and plug it back in once the installation is complete.
+We are also working with the Ubuntu team to address this issue upstream.
+
+# 02. Limited 33 GB/s Bi-Directional P2P Bandwidth with 1x GPU Card
+When using a single GPU card over a x16 PCIe connection without a PCIe switch, the observed bi-directional P2P bandwidth is limited to 33 GB/s.
+
+Workaround: Change the PCIe slot configuration in BIOS from Auto/x16 to x8/x8.
+With this change, over 40 GB/s bi-directional P2P bandwidth can be achieved.
+Root cause analysis is still in progress.

vllm/tools/platform/config/disable_apparmor.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# disable-snap-apparmor-logs.sh
+# Quiet AppArmor DENIED messages from snapd
+
+set -e
+
+CONFIG="/etc/apparmor/parser.conf"
+
+echo "[1/2] Updating AppArmor config to disable audit logs..."
+if ! grep -q "^no-audit" "$CONFIG"; then
+    echo "no-audit" | sudo tee -a "$CONFIG"
+fi
+
+echo "[2/2] Restarting AppArmor..."
+sudo systemctl restart apparmor
+
+echo "✅ AppArmor snap-confine DENIED logs have been silenced."

vllm/tools/platform/config/disable_auto_update.sh

@@ -0,0 +1,29 @@
+#!/bin/bash
+# disable-auto-upgrade.sh
+# Permanently disable automatic updates on Ubuntu
+
+set -e
+
+echo "[1/4] Disable unattended-upgrades service..."
+sudo systemctl stop unattended-upgrades.service || true
+sudo systemctl disable unattended-upgrades.service || true
+
+echo "[2/4] Disable apt-daily timers..."
+sudo systemctl stop apt-daily.timer apt-daily-upgrade.timer || true
+sudo systemctl disable apt-daily.timer apt-daily-upgrade.timer || true
+
+echo "[3/4] Update APT config to disable periodic upgrades..."
+CONFIG_FILE="/etc/apt/apt.conf.d/20auto-upgrades"
+if [ -f "$CONFIG_FILE" ]; then
+    sudo sed -i 's/^\(APT::Periodic::Update-Package-Lists\).*/\1 "0";/' "$CONFIG_FILE"
+    sudo sed -i 's/^\(APT::Periodic::Unattended-Upgrade\).*/\1 "0";/' "$CONFIG_FILE"
+else
+    echo 'APT::Periodic::Update-Package-Lists "0";' | sudo tee "$CONFIG_FILE"
+    echo 'APT::Periodic::Unattended-Upgrade "0";' | sudo tee -a "$CONFIG_FILE"
+fi
+
+echo "[4/4] Disable Snap auto-refresh..."
+sudo systemctl stop snapd.snap-repair.timer || true
+sudo systemctl disable snapd.snap-repair.timer || true
+
+echo "✅ Automatic updates have been disabled permanently."

vllm/tools/platform/debug/collect_sysinfo.sh

@@ -0,0 +1,64 @@
+#!/bin/bash
+set -euo pipefail
+
+is_docker() {
+  grep -qaE 'docker|kubepods|containerd' /proc/1/cgroup && return 0
+  [[ "$(hostname)" =~ ^[0-9a-f]{12}$ ]] && return 0
+  return 1
+}
+
+# Check for root privileges
+if [[ "$EUID" -ne 0 ]]; then
+  echo "[ERROR] This script must be run as root."
+  exit 1
+fi
+
+if is_docker; then
+  echo "[ERROR] Please run this script under native environment, not in docker"
+  exit 1
+fi
+
+# Prepare output directory
+TIMESTAMP=$(date +%Y%m%d_%H%M%S)
+OUTDIR="sysinfo_$TIMESTAMP"
+mkdir -p "$OUTDIR"
+
+echo "[INFO] Collecting system information into $OUTDIR..."
+
+# 1. CPU governor
+cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor > "$OUTDIR/scaling_governor.txt" 2>/dev/null || echo "Not available" > "$OUTDIR/scaling_governor.txt"
+
+# 2. CPU architecture
+lscpu > "$OUTDIR/lscpu.txt"
+
+# 3. PCI topology
+lspci -tv > "$OUTDIR/lspci_tree.txt"
+lspci -vvv > "$OUTDIR/lspci_verbose.txt"
+
+# 4. Kernel messages
+dmesg > "$OUTDIR/dmesg.txt"
+
+# 5. DRI tree
+tree /sys/kernel/debug/dri/ > "$OUTDIR/dri_tree.txt" 2>/dev/null || echo "Not available" > "$OUTDIR/dri_tree.txt"
+
+# 6. Memory usage
+free -h > "$OUTDIR/memory.txt"
+
+# 7. Hardware info
+dmidecode > "$OUTDIR/dmidecode.txt"
+
+# 8. libze info
+dpkg -l | grep libze > "$OUTDIR/libze_version.txt"
+
+# Create tar archive first
+TAR_FILE="sysinfo_$TIMESTAMP.tar"
+XZ_FILE="$TAR_FILE.xz"
+
+echo "[INFO] Creating archive $TAR_FILE..."
+tar -cf "$TAR_FILE" "$OUTDIR"
+
+echo "[INFO] Compressing with xz -9..."
+xz -9 "$TAR_FILE"
+
+echo "[INFO] Done. Output file: $XZ_FILE"
+

vllm/tools/platform/debug/get_bkc_version.sh

@@ -0,0 +1,67 @@
+#!/bin/bash
+
+# Output header
+echo "Category,Version"
+
+# 1. Ubuntu version
+UBUNTU_VERSION=$(grep '^VERSION=' /etc/os-release | cut -d '"' -f 2)
+echo "Ubuntu,$UBUNTU_VERSION"
+
+# 2. Linux kernel version
+KERNEL_VERSION=$(uname -r)
+echo "Linux Kernel,$KERNEL_VERSION"
+
+# 3. Intel GPU firmware versions from dmesg
+
+# Extract GuC firmware version
+guc_ver=$(dmesg | grep -i 'Using GuC firmware' | head -n1 | grep -oP 'version \K[\d\.]+')
+if [[ -n "$guc_ver" ]]; then
+    echo "GPU Firmware (guc),$guc_ver"
+else
+    echo "GPU Firmware (guc),Not Found"
+fi
+
+# Extract HuC firmware version
+huc_ver=$(dmesg | grep -i 'Using HuC firmware' | head -n1 | grep -oP 'version \K[\d\.]+')
+if [[ -n "$huc_ver" ]]; then
+    echo "GPU Firmware (huc),$huc_ver"
+fi
+
+# Extract DMC firmware version
+dmc_ver=$(dmesg | grep -i 'Finished loading DMC firmware' | head -n1 | grep -oP '\(v\K[\d\.]+')
+if [[ -n "$dmc_ver" ]]; then
+    echo "GPU Firmware (dmc),$dmc_ver"
+else
+    echo "GPU Firmware (dmc),Not Found"
+fi
+
+# 4. OneAPI version (offline installed)
+ONEAPI_LOG=$(ls /opt/intel/oneapi/logs/installer.install.intel.oneapi.lin.basekit.product,v=* 2>/dev/null | head -n1)
+if [[ -n "$ONEAPI_LOG" ]]; then
+    oneapi_ver=$(basename "$ONEAPI_LOG" | sed -n 's/.*basekit\.product,v=\(.*\)\..*/\1/p')
+    echo "oneapi,oneapi-base-toolkit=$oneapi_ver"
+else
+    echo "oneapi,oneapi-base-toolkit=Not Installed"
+fi
+
+# 5. Parse passed-in package files
+for file in "$@"; do
+    [[ ! -f "$file" ]] && continue
+
+    category=$(basename "$file" .txt)
+    first=1
+
+    while IFS= read -r pkg; do
+        [[ -z "$pkg" || "$pkg" =~ ^# ]] && continue
+
+        version=$(dpkg-query -W -f='${Version}\n' "$pkg" 2>/dev/null)
+        version_output="$pkg=${version:-Not Installed}"
+
+        if [[ $first -eq 1 ]]; then
+            echo "$category,$version_output"
+            first=0
+        else
+            echo ",$version_output"
+        fi
+    done < "$file"
+done

vllm/tools/platform/docker/build_ubuntu_image.sh

@@ -0,0 +1,59 @@
+#!/bin/bash
+set -e
+
+# Help message
+usage() {
+    echo "Usage: $0 [-n image_name:tag]"
+    echo "Default image name: ubuntu:25.04-custom"
+    exit 1
+}
+
+# Default image name
+IMAGE_NAME="ubuntu:25.04-custom"
+
+# Parse options
+while getopts ":n:h" opt; do
+  case ${opt} in
+    n )
+      IMAGE_NAME=$OPTARG
+      ;;
+    h )
+      usage
+      ;;
+    \? )
+      echo "Invalid option: -$OPTARG" >&2
+      usage
+      ;;
+  esac
+done
+
+TAR_NAME="ubuntu-2504-rootfs.tar.gz"
+
+echo "[+] Image name: $IMAGE_NAME"
+echo "[+] Creating root filesystem archive..."
+
+sudo tar --numeric-owner -czpf "$TAR_NAME" \
+    --exclude=/proc \
+    --exclude=/sys \
+    --exclude=/dev \
+    --exclude=/tmp/* \
+    --exclude=/run/* \
+    --exclude=/mnt \
+    --exclude=/media \
+    --exclude=/lost+found \
+    --exclude=/var/tmp/* \
+    --exclude=/home \
+    --exclude=/root \
+    --exclude=/etc/ssh \
+    --exclude=/etc/hostname \
+    --exclude=/etc/hosts \
+    /
+
+echo "[+] Archive created: $TAR_NAME"
+
+echo "[+] Importing into Docker as image: $IMAGE_NAME"
+cat "$TAR_NAME" | docker import - "$IMAGE_NAME"
+
+echo "[✔] Done!"
+echo "You can run the image using:"
+echo "    docker run -it $IMAGE_NAME bash"