Parse inventory/output.log to extract GPU serial numbers from lspci output, expose them via serials API, and add GPU category to web UI. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
144 lines
4.5 KiB
Go
144 lines
4.5 KiB
Go
package nvidia
|
|
|
|
import (
|
|
"bufio"
|
|
"regexp"
|
|
"strings"
|
|
|
|
"git.mchus.pro/mchus/logpile/internal/models"
|
|
"git.mchus.pro/mchus/logpile/internal/parser"
|
|
)
|
|
|
|
var (
|
|
// Regex to extract GPU serial numbers from lspci output
|
|
// Example: " Capabilities: [2f0 v1] Device Serial Number 14-17-dc-65-77-2d-b0-48"
|
|
gpuSerialRegex = regexp.MustCompile(`Device Serial Number\s+([\da-fA-F-]+)`)
|
|
|
|
// Regex to extract PCI BDF from lspci header
|
|
// Example: "2a:00.0 3D controller: NVIDIA Corporation Device 2335 (rev a1)"
|
|
// Note: lspci format is bus:device.function (e.g., "2a:00.0")
|
|
pciBDFRegex = regexp.MustCompile(`^([0-9a-fA-F]{2,4}:[0-9a-fA-F]{2}\.[0-9])\s+3D controller.*NVIDIA`)
|
|
|
|
// Regex to extract devname mappings from fieldiag command line
|
|
// Example: "devname=0000:ba:00.0,SXM5_SN_1653925027099"
|
|
devnameRegex = regexp.MustCompile(`devname=([\da-fA-F:\.]+),(\w+)`)
|
|
)
|
|
|
|
// ParseInventoryLog parses inventory/output.log to extract GPU serial numbers
|
|
// from lspci output and map them to slots
|
|
func ParseInventoryLog(content []byte, result *models.AnalysisResult) error {
|
|
if result.Hardware == nil || len(result.Hardware.GPUs) == 0 {
|
|
// No GPUs to update
|
|
return nil
|
|
}
|
|
|
|
scanner := bufio.NewScanner(strings.NewReader(string(content)))
|
|
|
|
// First pass: build mapping of PCI BDF -> Slot name from fieldiag command line
|
|
pciToSlot := make(map[string]string)
|
|
for scanner.Scan() {
|
|
line := scanner.Text()
|
|
// Look for fieldiag command with devname parameters
|
|
if strings.Contains(line, "devname=") && strings.Contains(line, "fieldiag") {
|
|
matches := devnameRegex.FindAllStringSubmatch(line, -1)
|
|
for _, match := range matches {
|
|
if len(match) == 3 {
|
|
pciBDF := match[1]
|
|
slotName := match[2]
|
|
// Extract slot number from name like "SXM5_SN_1653925027099"
|
|
// We want to map to slot like "GPUSXM5"
|
|
if strings.HasPrefix(slotName, "SXM") {
|
|
parts := strings.Split(slotName, "_")
|
|
if len(parts) >= 1 {
|
|
// Convert "SXM5" to "GPUSXM5"
|
|
slot := "GPU" + parts[0]
|
|
pciToSlot[pciBDF] = slot
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Second pass: extract GPU serial numbers from lspci output
|
|
scanner = bufio.NewScanner(strings.NewReader(string(content)))
|
|
var currentPCIBDF string
|
|
var currentSlot string
|
|
|
|
for scanner.Scan() {
|
|
line := scanner.Text()
|
|
|
|
// Check if this is a new GPU device header
|
|
if match := pciBDFRegex.FindStringSubmatch(line); len(match) > 1 {
|
|
currentPCIBDF = match[1]
|
|
// Normalize BDF format - lspci uses short format (bus:device.function)
|
|
// but fieldiag uses full format (domain:bus:device.function)
|
|
// Convert "2a:00.0" to "0000:2a:00.0"
|
|
normalizedBDF := currentPCIBDF
|
|
if len(strings.Split(currentPCIBDF, ":")) == 2 {
|
|
// Short format without domain, add 0000:
|
|
normalizedBDF = "0000:" + currentPCIBDF
|
|
}
|
|
|
|
// Map to slot name if we have it
|
|
if slot, ok := pciToSlot[normalizedBDF]; ok {
|
|
currentSlot = slot
|
|
} else if slot, ok := pciToSlot[currentPCIBDF]; ok {
|
|
currentSlot = slot
|
|
} else {
|
|
currentSlot = ""
|
|
}
|
|
continue
|
|
}
|
|
|
|
// Look for Device Serial Number in capabilities
|
|
if match := gpuSerialRegex.FindStringSubmatch(line); len(match) > 1 && currentSlot != "" {
|
|
serialNumber := match[1]
|
|
// Format: 14-17-dc-65-77-2d-b0-48
|
|
// Convert to more readable format: 48:b0:2d:77:65:dc:17:14 (reversed)
|
|
serialFormatted := formatGPUSerial(serialNumber)
|
|
|
|
// Find the GPU in our results and update its serial number
|
|
for i := range result.Hardware.GPUs {
|
|
if result.Hardware.GPUs[i].Slot == currentSlot {
|
|
result.Hardware.GPUs[i].SerialNumber = serialFormatted
|
|
break
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return scanner.Err()
|
|
}
|
|
|
|
// formatGPUSerial formats GPU serial number from PCIe format to human-readable
|
|
// Input: "14-17-dc-65-77-2d-b0-48" (little-endian from PCIe)
|
|
// Output: "48:b0:2d:77:65:dc:17:14" (reversed to match GPU label)
|
|
func formatGPUSerial(serial string) string {
|
|
parts := strings.Split(serial, "-")
|
|
if len(parts) != 8 {
|
|
return serial // Return as-is if unexpected format
|
|
}
|
|
|
|
// Reverse the bytes (PCIe reports in little-endian)
|
|
reversed := make([]string, len(parts))
|
|
for i := range parts {
|
|
reversed[len(parts)-1-i] = strings.ToUpper(parts[i])
|
|
}
|
|
|
|
return strings.Join(reversed, ":")
|
|
}
|
|
|
|
// findInventoryOutputLog finds the inventory/output.log file
|
|
func findInventoryOutputLog(files []parser.ExtractedFile) *parser.ExtractedFile {
|
|
for _, f := range files {
|
|
// Look for inventory/output.log
|
|
path := strings.ToLower(f.Path)
|
|
if strings.Contains(path, "inventory/output.log") ||
|
|
strings.Contains(path, "inventory\\output.log") {
|
|
return &f
|
|
}
|
|
}
|
|
return nil
|
|
}
|