Unify NVIDIA GPU recovery paths

audit/internal/platform/benchmark.go

@@ -105,6 +105,7 @@ var (
 	benchmarkSkippedPattern    = regexp.MustCompile(`^([a-z0-9_]+)(?:\[\d+\])?=SKIPPED (.+)$`)
 	benchmarkIterationsPattern = regexp.MustCompile(`^([a-z0-9_]+)_iterations=(\d+)$`)
 	benchmarkGeteuid           = os.Geteuid
+	benchmarkResetNvidiaGPU    = resetNvidiaGPU
 	benchmarkSleep             = time.Sleep
 )
 
@@ -249,6 +250,35 @@ func setBenchmarkPowerLimit(ctx context.Context, verboseLog string, gpuIndex, po
 	return nil
 }
 
+func resetBenchmarkGPU(ctx context.Context, verboseLog string, gpuIndex int, logFunc func(string)) error {
+	if logFunc != nil {
+		logFunc(fmt.Sprintf("power benchmark pre-flight: GPU %d reset via shared NVIDIA recover path", gpuIndex))
+	}
+	out, err := benchmarkResetNvidiaGPU(gpuIndex)
+	appendSATVerboseLog(verboseLog,
+		fmt.Sprintf("[%s] start power-preflight-gpu-%d-reset.log", time.Now().UTC().Format(time.RFC3339), gpuIndex),
+		"cmd: bee-nvidia-recover reset-gpu "+strconv.Itoa(gpuIndex),
+	)
+	if trimmed := strings.TrimSpace(out); trimmed != "" && logFunc != nil {
+		for _, line := range strings.Split(trimmed, "\n") {
+			line = strings.TrimSpace(line)
+			if line != "" {
+				logFunc(line)
+			}
+		}
+	}
+	rc := 0
+	if err != nil {
+		rc = 1
+	}
+	appendSATVerboseLog(verboseLog,
+		fmt.Sprintf("[%s] finish power-preflight-gpu-%d-reset.log", time.Now().UTC().Format(time.RFC3339), gpuIndex),
+		fmt.Sprintf("rc: %d", rc),
+		"",
+	)
+	return err
+}
+
 func resetBenchmarkGPUs(ctx context.Context, verboseLog string, gpuIndices []int, logFunc func(string)) []int {
 	if len(gpuIndices) == 0 {
 		return nil
@@ -266,8 +296,7 @@ func resetBenchmarkGPUs(ctx context.Context, verboseLog string, gpuIndices []int
 	}
 	var failed []int
 	for _, idx := range gpuIndices {
-		name := fmt.Sprintf("power-preflight-gpu-%d-reset.log", idx)
-		if _, err := runSATCommandCtx(ctx, verboseLog, name, []string{"nvidia-smi", "-i", strconv.Itoa(idx), "-r"}, nil, logFunc); err != nil {
+		if err := resetBenchmarkGPU(ctx, verboseLog, idx, logFunc); err != nil {
 			failed = append(failed, idx)
 			if logFunc != nil {
 				logFunc(fmt.Sprintf("power benchmark pre-flight: GPU %d reset failed: %v", idx, err))
@@ -4440,8 +4469,7 @@ func (s *System) RunNvidiaPowerBench(ctx context.Context, baseDir string, opts N
 		_ = os.MkdirAll(singleDir, 0755)
 		singleInfo := cloneBenchmarkGPUInfoMap(infoByIndex)
 		if failed := resetBenchmarkGPUs(ctx, verboseLog, []int{idx}, logFunc); len(failed) > 0 {
-			result.Findings = append(result.Findings,
-				fmt.Sprintf("GPU %d reset pre-flight did not complete before its first power test; throttle counters may contain stale state.", idx))
+			return "", fmt.Errorf("power benchmark pre-flight: failed to reset GPU %d; benchmark aborted to keep measurements clean", idx)
 		}
 		logFunc(fmt.Sprintf("power calibration: GPU %d single-card baseline", idx))
 		singlePowerStopCh := make(chan struct{})