feat(device): implement MSR fallback for CPU power meter

cmd/kepler/main.go

@@ -233,9 +233,17 @@ func createCPUMeter(logger *slog.Logger, cfg *config.Config) (device.CPUPowerMet
 		logger.Info("rapl zones are filtered", "zones-enabled", cfg.Rapl.Zones)
 	}
 
+	// Convert config MSR settings to device MSRConfig
+	msrConfig := device.MSRConfig{
+		Enabled:    cfg.MSR.Enabled,
+		Force:      cfg.MSR.Force,
+		DevicePath: cfg.MSR.DevicePath,
+	}
+
 	return device.NewCPUPowerMeter(
 		cfg.Host.SysFS,
 		device.WithRaplLogger(logger),
 		device.WithZoneFilter(cfg.Rapl.Zones),
+		device.WithMSRConfig(msrConfig),
 	)
 }

compose/dev/kepler-dev/etc/kepler/config.yaml

@@ -35,6 +35,11 @@ host:
 rapl:
   zones: [] # zones to be enabled, empty enables all default zones
 
+msr: # MSR fallback configuration for RAPL energy reading
+  enabled: false # enable automatic MSR fallback when powercap unavailable (default: false)
+  force: false # force MSR usage even if powercap available (testing only, default: false)
+  devicePath: /dev/cpu/%d/msr # MSR device path template (default: "/dev/cpu/%d/msr")
+
 exporter:
   stdout: # stdout exporter related config
     enabled: false # disabled by default

config/config.go

@@ -33,6 +33,16 @@ type (
 		Zones []string `yaml:"zones"`
 	}
 
+	// MSR configuration for fallback power reading
+	MSR struct {
+		// Enable automatic MSR fallback when powercap unavailable
+		Enabled *bool `yaml:"enabled"`
+		// Force MSR usage even if powercap available (testing)
+		Force *bool `yaml:"force"`
+		// MSR device path template
+		DevicePath string `yaml:"devicePath"`
+	}
+
 	// Development mode settings; disabled by default
 	Dev struct {
 		FakeCpuMeter struct {
@@ -98,6 +108,7 @@ type (
 		Host     Host     `yaml:"host"`
 		Monitor  Monitor  `yaml:"monitor"`
 		Rapl     Rapl     `yaml:"rapl"`
+		MSR      MSR      `yaml:"msr"`
 		Exporter Exporter `yaml:"exporter"`
 		Web      Web      `yaml:"web"`
 		Debug    Debug    `yaml:"debug"`
@@ -168,6 +179,12 @@ const (
 	// RAPL
 	RaplZones = "rapl.zones" // not a flag
 
+	// MSR - NOTE: MSR settings are not exposed as CLI flags per proposal
+	// They should only be configured via YAML files due to security implications
+	MSREnabled    = "msr.enabled"    // not a flag
+	MSRForce      = "msr.force"      // not a flag
+	MSRDevicePath = "msr.devicePath" // not a flag
+
 	pprofEnabledFlag = "debug.pprof"
 
 	WebConfigFlag        = "web.config-file"
@@ -203,6 +220,11 @@ func DefaultConfig() *Config {
 		Rapl: Rapl{
 			Zones: []string{},
 		},
+		MSR: MSR{
+			Enabled:    ptr.To(false), // Opt-in for security
+			Force:      ptr.To(false),
+			DevicePath: "/dev/cpu/%d/msr",
+		},
 		Monitor: Monitor{
 			Interval:  5 * time.Second,
 			Staleness: 500 * time.Millisecond,
@@ -408,6 +430,9 @@ func (c *Config) sanitize() {
 		c.Rapl.Zones[i] = strings.TrimSpace(c.Rapl.Zones[i])
 	}
 
+	// MSR settings sanitization
+	c.MSR.DevicePath = strings.TrimSpace(c.MSR.DevicePath)
+
 	for i := range c.Exporter.Prometheus.DebugCollectors {
 		c.Exporter.Prometheus.DebugCollectors[i] = strings.TrimSpace(c.Exporter.Prometheus.DebugCollectors[i])
 	}
@@ -488,6 +513,16 @@ func (c *Config) Validate(skips ...SkipValidation) error {
 			errs = append(errs, fmt.Sprintf("invalid monitor min terminated energy threshold: %d can't be negative", c.Monitor.MinTerminatedEnergyThreshold))
 		}
 	}
+	{ // MSR settings
+		if c.MSR.DevicePath == "" {
+			errs = append(errs, "MSR device path cannot be empty")
+		} else {
+			// Basic validation that device path is a template
+			if !strings.Contains(c.MSR.DevicePath, "%d") {
+				errs = append(errs, "MSR device path must contain '%d' placeholder for CPU ID")
+			}
+		}
+	}
 	{ // Kubernetes
 		if ptr.Deref(c.Kube.Enabled, false) {
 			if c.Kube.Config != "" {