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

This commit implements EP-002 MSR fallback power meter enhancement proposal.
Add MSR (Model Specific Register) support as fallback when powercap interface
is unavailable. This enhancement improves Kepler's compatibility across
different systems and kernel configurations.

Key changes:
- Add MSR reader implementation with Intel RAPL register support
- Create raplReader interface abstracting powercap and MSR backends
- Extract existing powercap logic into dedicated reader component
- Enhance RAPL power meter with automatic fallback detection
- Add MSR configuration with security-conscious opt-in defaults
- Implement comprehensive test coverage with mock MSR data

The MSR fallback is disabled by default due to PLATYPUS attack vectors
(CVE-2020-8694/8695) and must be explicitly enabled via configuration.
When enabled, the system automatically falls back to MSR if powercap
is unavailable, maintaining transparent operation.

Signed-off-by: Sunil Thaha <sthaha@redhat.com>

Author: sthaha

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 != "" {

docs/user/configuration.md

@@ -95,6 +95,11 @@ host:
 rapl:
   zones: []     # RAPL 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
+  force: false      # Force MSR usage even if powercap available (testing)
+  devicePath: "/dev/cpu/%d/msr"  # MSR device path template
+
 exporter:
   stdout:       # stdout exporter related config
     enabled: false # disabled by default
@@ -195,6 +200,41 @@ rapl:
   zones: ["package", "core", "uncore"]
 ```
 
+### 🔌 MSR Configuration
+
+```yaml
+msr:
+  enabled: false    # Enable automatic MSR fallback
+  force: false      # Force MSR usage for testing
+  devicePath: "/dev/cpu/%d/msr"  # MSR device path template
+```
+
+Model Specific Register (MSR) support provides a fallback mechanism for reading Intel RAPL energy counters when the Linux powercap sysfs interface is unavailable.
+
+- **enabled**: Enable automatic MSR fallback when powercap is unavailable
+  - Default: `false` (opt-in for security reasons)
+  - When enabled, Kepler will automatically fall back to MSR if powercap fails
+  - Requires appropriate permissions and hardware support
+
+- **force**: Force MSR usage even when powercap is available
+  - Default: `false`
+  - Primarily for testing and development purposes
+  - When `true`, MSR will be used regardless of powercap availability
+
+- **devicePath**: Template for MSR device file paths
+  - Default: `"/dev/cpu/%d/msr"`
+  - The `%d` placeholder is replaced with the CPU number
+  - Must be accessible with appropriate permissions
+
+⚠️ **Security Note**: MSR access requires elevated privileges and may be restricted on some systems due to security considerations (PLATYPUS attacks, CVE-2020-8694/8695). Use MSR configuration only when necessary and ensure proper system security measures are in place.
+
+**Prerequisites for MSR support:**
+
+- Intel CPU with RAPL support
+- `msr` kernel module loaded (`modprobe msr`)
+- Read access to `/dev/cpu/*/msr` files
+- Elevated privileges (typically root)
+
 ### 📦 Exporter Configuration
 
 ```yaml

hack/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

internal/device/mock_cpu_power_meter.go

@@ -5,14 +5,6 @@ package device
 
 // TODO: Move this mock to a separate testutil package
 
-import (
-	"slices"
-	"testing"
-
-	"github.com/prometheus/procfs/sysfs"
-	"github.com/stretchr/testify/require"
-)
-
 const (
 	validSysFSPath = "testdata/sys"
 	badSysFSPath   = "testdata/bad_sysfs"
@@ -67,27 +59,3 @@ func (m *MockRaplZone) OnEnergy(j Energy, err error) {
 func (m *MockRaplZone) Inc(delta Energy) {
 	m.energy = (m.energy + delta) % m.maxMicroJoules
 }
-
-func validSysFSFixtures(t *testing.T) sysfs.FS {
-	t.Helper()
-	fs, err := sysfs.NewFS(validSysFSPath)
-	require.NoError(t, err, "Failed to create sysfs test FS")
-	return fs
-}
-
-func invalidSysFSFixtures(t *testing.T) sysfs.FS {
-	t.Helper()
-	fs, err := sysfs.NewFS(badSysFSPath)
-	require.NoError(t, err, "Failed to create sysfs test FS")
-	return fs
-}
-
-func sortedZoneNames(zones []EnergyZone) []string {
-	names := make([]string, len(zones))
-	for i, zone := range zones {
-		names[i] = zone.Name()
-	}
-	slices.Sort(names)
-
-	return names
-}