Files
logpile/internal/parser/vendors/inspur/hgx_hwinfo.go

176 lines
3.8 KiB
Go

package inspur
import (
"fmt"
"regexp"
"strconv"
"strings"
"git.mchus.pro/mchus/logpile/internal/models"
)
type hgxGPUAssemblyInfo struct {
Model string
Part string
Serial string
}
// Logical GPU index mapping used by HGX B200 UI ordering.
// Example from real logs/UI:
// GPU0->SXM5, GPU1->SXM7, GPU2->SXM6, GPU3->SXM8, GPU4->SXM2, GPU5->SXM4, GPU6->SXM3, GPU7->SXM1.
var hgxLogicalToSXM = map[int]int{
0: 5,
1: 7,
2: 6,
3: 8,
4: 2,
5: 4,
6: 3,
7: 1,
}
var (
reHGXGPUBlock = regexp.MustCompile(`(?s)/redfish/v1/Chassis/HGX_GPU_SXM_(\d+)/Assembly.*?"Name":\s*"GPU Board Assembly".*?"Model":\s*"([^"]+)".*?"PartNumber":\s*"([^"]+)".*?"SerialNumber":\s*"([^"]+)"`)
reSlotGPU = regexp.MustCompile(`(?i)gpu\s*#?\s*(\d+)`)
)
func enrichGPUsFromHGXHWInfo(content []byte, hw *models.HardwareConfig) {
if hw == nil || len(hw.GPUs) == 0 || len(content) == 0 {
return
}
bySXM := parseHGXGPUAssembly(content)
if len(bySXM) == 0 {
return
}
normalizeHGXGPUInventory(hw, bySXM)
for i := range hw.GPUs {
gpu := &hw.GPUs[i]
logicalIdx, ok := extractLogicalGPUIndex(gpu.Slot)
if !ok {
// Keep existing info if slot index cannot be determined.
continue
}
sxm := resolveSXMIndex(logicalIdx, bySXM)
info, found := bySXM[sxm]
if !found {
continue
}
if strings.TrimSpace(gpu.SerialNumber) == "" {
gpu.SerialNumber = info.Serial
}
if shouldReplaceGPUModel(gpu.Model) {
gpu.Model = info.Model
}
if strings.TrimSpace(gpu.PartNumber) == "" {
gpu.PartNumber = info.Part
}
if strings.TrimSpace(gpu.Manufacturer) == "" {
gpu.Manufacturer = "NVIDIA"
}
}
}
func parseHGXGPUAssembly(content []byte) map[int]hgxGPUAssemblyInfo {
result := make(map[int]hgxGPUAssemblyInfo)
matches := reHGXGPUBlock.FindAllSubmatch(content, -1)
for _, m := range matches {
if len(m) != 5 {
continue
}
sxmIdx, err := strconv.Atoi(string(m[1]))
if err != nil || sxmIdx <= 0 {
continue
}
result[sxmIdx] = hgxGPUAssemblyInfo{
Model: strings.TrimSpace(string(m[2])),
Part: strings.TrimSpace(string(m[3])),
Serial: strings.TrimSpace(string(m[4])),
}
}
return result
}
func extractLogicalGPUIndex(slot string) (int, bool) {
m := reSlotGPU.FindStringSubmatch(slot)
if len(m) < 2 {
return 0, false
}
idx, err := strconv.Atoi(m[1])
if err != nil || idx < 0 {
return 0, false
}
return idx, true
}
func resolveSXMIndex(logicalIdx int, bySXM map[int]hgxGPUAssemblyInfo) int {
if sxm, ok := hgxLogicalToSXM[logicalIdx]; ok {
if _, exists := bySXM[sxm]; exists {
return sxm
}
}
identity := logicalIdx + 1
if _, exists := bySXM[identity]; exists {
return identity
}
return identity
}
func shouldReplaceGPUModel(model string) bool {
trimmed := strings.TrimSpace(model)
if trimmed == "" {
return true
}
switch strings.ToLower(trimmed) {
case "vga", "3d controller", "display controller", "unknown":
return true
default:
return false
}
}
func normalizeHGXGPUInventory(hw *models.HardwareConfig, bySXM map[int]hgxGPUAssemblyInfo) {
// Keep only logical HGX GPUs (#GPU0..#GPU7) and remove BMC VGA entries.
filtered := make([]models.GPU, 0, len(hw.GPUs))
present := make(map[int]bool)
for _, gpu := range hw.GPUs {
idx, ok := extractLogicalGPUIndex(gpu.Slot)
if !ok || idx < 0 || idx > 7 {
continue
}
present[idx] = true
filtered = append(filtered, gpu)
}
// If some logical GPUs are missing in asset.json, add placeholders from HGX Redfish assembly.
for logicalIdx := 0; logicalIdx <= 7; logicalIdx++ {
if present[logicalIdx] {
continue
}
sxm := resolveSXMIndex(logicalIdx, bySXM)
info, ok := bySXM[sxm]
if !ok {
continue
}
filtered = append(filtered, models.GPU{
Slot: fmt.Sprintf("#GPU%d", logicalIdx),
Model: info.Model,
Manufacturer: "NVIDIA",
SerialNumber: info.Serial,
PartNumber: info.Part,
})
}
hw.GPUs = filtered
}