fix(node_info): Intel GPU detection misidentifying AMD GPUs when intel-gpu-tools is installed

src/modules/node_info/files/Collector/Intel.pm

@@ -5,7 +5,7 @@ use warnings;
 use Exporter 'import';
 
 use PVE::PVEMod::Config qw(%config $process_type $pve_mod_working_dir);
-use PVE::PVEMod::Utils  qw(debug check_executable setup_collector_signals safe_write_json);
+use PVE::PVEMod::Utils  qw(debug check_executable setup_collector_signals safe_write_json read_sysfs);
 use PVE::PVEMod::Store  qw(update_intel_gpu_rrd);
 
 our @EXPORT_OK = qw(
@@ -13,10 +13,30 @@ our @EXPORT_OK = qw(
     collector_for_intel_device
 );
 
+use constant INTEL_VENDOR_ID => '8086';
+
 # ============================================================================
 # Intel GPU — device discovery
 # ============================================================================
 
+# Resolve the PCI vendor ID (lowercase, e.g. "8086") from either a
+# "vendor=XXXX" descriptor or a "DDDD:BB:DD.F" address looked up via sysfs.
+sub _get_pci_vendor_id {
+    my ($path) = @_;
+
+    if ($path =~ /vendor=([0-9a-fA-F]{4})/) {
+        return lc($1);
+    }
+
+    if ($path =~ m{^pci:([0-9a-fA-F]{4}:[0-9a-fA-F]{2}:[0-9a-fA-F]{2}\.\d+)$}) {
+        my $vendor = read_sysfs("/sys/bus/pci/devices/$1/vendor");
+        $vendor =~ s/^0x//i;
+        return lc($vendor);
+    }
+
+    return undef;
+}
+
 sub get_intel_gpu_devices {
     my @devices = ();
     my $fh;
@@ -43,6 +63,11 @@ sub get_intel_gpu_devices {
         chomp;
         if (/^(card\d+)\s+(.+?)\s+(pci:[^\s]+)/) {
             my ($card, $name, $path) = ($1, $2, $3);
+            my $vendor = _get_pci_vendor_id($path);
+            if (!defined $vendor || $vendor ne INTEL_VENDOR_ID) {
+                debug(__LINE__, "Skipping non-Intel device: $card ($path) vendor=" . ($vendor // 'unknown'));
+                next;
+            }
             push @devices, {
                 card     => $card,
                 name     => $name,

src/modules/node_info/files/Config.pm

@@ -17,7 +17,7 @@ our @EXPORT_OK = qw(
 # Debug / Version
 # ============================================================================
 
-our $DEBUG_ENABLED;  # derived from %config{debug}{mod_debug} after ini load, below
+our $DEBUG_ENABLED = 1;
 our $VERSION       = 'version-placeholder';
 
 # ============================================================================
@@ -43,7 +43,6 @@ our %config = (
         gpu_history    => 0,
     },
     debug => {
-        mod_debug              => 0,
         log_enabled            => 0,
         log_file               => '/tmp/pve-mod-debug.log',
         lm_sensors_mode        => 0,
@@ -160,6 +159,4 @@ sub _load_ini_file {
 _load_ini_file($CONF_FILE);
 _load_ini_file($_) for glob("$CONFD_DIR/*.conf");
 
-$DEBUG_ENABLED = $config{debug}{mod_debug};
-
 1;

src/modules/node_info/node_info.conf

@@ -30,11 +30,7 @@ ignore_temp_below=5
 
 # Debug mode: when a collector's mode is 1, the real tool is not required.
 # Data is read from the file path instead. Useful for development/testing.
-# Note: debug settings are only read at pveproxy startup — restart pveproxy
-# after changing them. Debug output goes to the journal (journalctl -u pveproxy),
-# plus log_file if log_enabled=1.
 [debug]
-mod_debug=0
 lm_sensors_mode=0
 lm_sensors_output_file=/tmp/sensors-output.json
 intel_mode=0

src/modules/node_info/node_info.configure.sh

@@ -59,6 +59,24 @@ _check_nvidia_tool() {
     return 1
 }
 
+# Resolve the PCI vendor ID (lowercase, e.g. "8086") from either a
+# "vendor=XXXX" descriptor or a "DDDD:BB:DD.F" address looked up via sysfs.
+_get_pci_vendor_id() {
+    local path="$1" vendor=""
+    if [[ "$path" =~ vendor=([0-9a-fA-F]{4}) ]]; then
+        vendor="${BASH_REMATCH[1]}"
+    elif [[ "$path" =~ ^pci:([0-9a-fA-F]{4}:[0-9a-fA-F]{2}:[0-9a-fA-F]{2}\.[0-9]+)$ ]]; then
+        local vfile="/sys/bus/pci/devices/${BASH_REMATCH[1]}/vendor"
+        [[ -f "$vfile" ]] && vendor=$(cat "$vfile")
+        vendor="${vendor#0x}"
+    fi
+    echo "${vendor,,}"
+}
+
+_is_intel_pci_device() {
+    [[ "$(_get_pci_vendor_id "$1")" == "8086" ]]
+}
+
 # --- standard API ------------------------------------------------------------
 
 node_info_defaults() {
@@ -78,7 +96,6 @@ node_info_defaults() {
     DEBUG_AMD=0;         DEBUG_AMD_FILE="/tmp/amd-gpu-devices.json"
     DEBUG_UPS=0;         DEBUG_UPS_FILE="/tmp/ups-output.json"
     DEBUG_LOG=0;         DEBUG_LOG_FILE="/tmp/pve-mod-debug.log"
-    DEBUG_MOD=0
 }
 
 node_info_load_conf() {
@@ -125,7 +142,6 @@ node_info_load_conf() {
             debug.ups_output_file)          DEBUG_UPS_FILE="$val" ;;
             debug.log_enabled)              DEBUG_LOG="$val" ;;
             debug.log_file)                 DEBUG_LOG_FILE="$val" ;;
-            debug.mod_debug)                DEBUG_MOD="$val" ;;
         esac
     done < "$NODE_INFO_CONF"
 }
@@ -340,7 +356,15 @@ node_info_configure() {
             warn "No Intel GPUs in debug file."
         fi
     elif _check_or_install_tool intel_gpu_top intel-gpu-tools "Intel GPU tools (intel-gpu-tools)"; then
-        intelCards=$(intel_gpu_top -L 2>/dev/null | grep -E '^card[0-9]+' || true)
+        local rawIntelCards
+        rawIntelCards=$(intel_gpu_top -L 2>/dev/null | grep -E '^card[0-9]+' || true)
+        # Filter out non-Intel devices (e.g. AMD cards misdetected when intel-gpu-tools is installed on non-Intel hardware)
+        while IFS= read -r line; do
+            [[ -z "$line" ]] && continue
+            if [[ "$line" =~ (pci:[^[:space:]]+) ]] && _is_intel_pci_device "${BASH_REMATCH[1]}"; then
+                intelCards+="${intelCards:+$'\n'}$line"
+            fi
+        done <<< "$rawIntelCards"
         if [[ -n "$intelCards" ]]; then
             info "Intel GPU(s) detected:"
             echo "$intelCards" | while IFS= read -r line; do echo "  $line"; done
@@ -365,7 +389,7 @@ node_info_configure() {
             echo "$json" > "$DEBUG_INTEL_FILE"
             info "Intel GPU device list saved to $DEBUG_INTEL_FILE"
         else
-            warn "No Intel GPUs detected by intel_gpu_top."
+            warn "No Intel GPUs detected by intel_gpu_top (or none had an Intel PCI vendor ID)."
         fi
     fi
     #endregion Intel GPU
@@ -514,11 +538,7 @@ ignore_temp_below=${IGNORE_TEMP_BELOW}
 
 # Debug mode: when a collector's mode is 1, the real tool is not required.
 # Data is read from the file path instead. Useful for development/testing.
-# Note: debug settings are only read at pveproxy startup — restart pveproxy
-# after changing them. Debug output goes to the journal (journalctl -u pveproxy),
-# plus log_file if log_enabled=1.
 [debug]
-mod_debug=${DEBUG_MOD}
 lm_sensors_mode=${DEBUG_LM_SENSORS}
 lm_sensors_output_file=${DEBUG_LM_SENSORS_FILE}
 intel_mode=${DEBUG_INTEL}

src/modules/node_info/readme.md

@@ -67,27 +67,3 @@ Each feature requires the corresponding tool to be installed on the Proxmox host
 ## Debug Mode
 
 Each collector supports a debug mode that reads from a local file instead of executing the real tool. Useful for development and testing without physical hardware.
-
-### Per-collector debug files
-
-Set the collector's `_mode` flag to `1` in the `[debug]` section of `/etc/pve-mod/conf.d/node_info.conf` and populate its file(s) with sample data, captured from a real host via the commands below:
-
-| Collector | Enable flag | File | Content | Example to generate it |
-|-----------|-------------|------|---------|-------------------------|
-| Temperature sensors | `lm_sensors_mode` | `lm_sensors_output_file` | Raw `sensors -j` JSON output | `sensors -j > /tmp/sensors-output.json` |
-| Intel GPU | `intel_mode` | `intel_devices_file` | Device list, one line per GPU (`intel_gpu_top -L` format) | `intel_gpu_top -L > /tmp/intel-gpu-devices.json` |
-| | | `intel_output_file` | Continuous `intel_gpu_top` stats output | `intel_gpu_top -d /dev/dri/card0 -s 1000 -l > /tmp/intel-gpu-output.txt` |
-| NVIDIA GPU | `nvidia_mode` | `nvidia_devices_file` | Device list CSV | `nvidia-smi --query-gpu=index,name --format=csv > /tmp/nvidia-smi-devices.csv` |
-| | | `nvidia_output_file` | Stats CSV | `nvidia-smi --query-gpu=index,name,temperature.gpu,utilization.gpu,utilization.memory,memory.used,memory.total,power.draw,power.limit,fan.speed --format=csv,nounits > /tmp/nvidia-smi-output.csv` |
-| AMD GPU | `amd_mode` | `amd_devices_file` | Placeholder — collector not yet implemented | — |
-| UPS | `ups_mode` | `ups_output_file` | Raw `upsc` key: value output (despite the `.json` name, it's plain text, not JSON) | `upsc ups@192.168.1.50 > /tmp/ups-output.json` |
-
-### Verbose module logging (mod_debug)
-
-Set `mod_debug=1` in the `[debug]` section of `/etc/pve-mod/conf.d/node_info.conf`, then restart `pveproxy` (`systemctl restart pveproxy`) — the setting is only read at startup. With it enabled, the mod logs each internal step (collector start/stop, cache hits, file reads, etc.) to the journal, viewable with:
-
-```sh
-journalctl -u pveproxy -f
-```
-
-Set `log_enabled=1` (and optionally `log_file`) in the same section to also persist this output to a file.