sysfs: add PCIe AER counters and Root Port AER totals to PciDevice

sysfs/net_class_aer.go

@@ -133,13 +133,19 @@ func parseAerCounters(devicePath string) (*AerCounters, error) {
 // parseCorrectableAerCounters parses correctable error counters in
 // /sys/class/net/<iface>/device/aer_dev_correctable.
 func parseCorrectableAerCounters(devicePath string, counters *CorrectableAerCounters) error {
-	path := filepath.Join(devicePath, "device", "aer_dev_correctable")
+	_, err := parseCorrectableAerFile(filepath.Join(devicePath, "device", "aer_dev_correctable"), counters)
+	return err
+}
+
+// parseCorrectableAerFile parses a correctable AER counters file such as
+// aer_dev_correctable. It reports whether the file was present and readable.
+func parseCorrectableAerFile(path string, counters *CorrectableAerCounters) (bool, error) {
 	value, err := parsers.SysReadFile(path)
 	if err != nil {
 		if canIgnoreError(err) {
-			return nil
+			return false, nil
 		}
-		return fmt.Errorf("failed to read file %q: %w", path, err)
+		return false, fmt.Errorf("failed to read file %q: %w", path, err)
 	}
 
 	for line := range strings.SplitSeq(string(value), "\n") {
@@ -148,12 +154,12 @@ func parseCorrectableAerCounters(devicePath string, counters *CorrectableAerCoun
 		}
 		fields := strings.Fields(line)
 		if len(fields) != 2 {
-			return fmt.Errorf("unexpected number of fields: %v", fields)
+			return false, fmt.Errorf("unexpected number of fields: %v", fields)
 		}
 		counterName := fields[0]
 		value, err := strconv.ParseUint(fields[1], 10, 64)
 		if err != nil {
-			return fmt.Errorf("error parsing value for %s: %w", counterName, err)
+			return false, fmt.Errorf("error parsing value for %s: %w", counterName, err)
 		}
 
 		switch counterName {
@@ -178,21 +184,28 @@ func parseCorrectableAerCounters(devicePath string, counters *CorrectableAerCoun
 		}
 	}
 
-	return nil
+	return true, nil
 }
 
 // parseUncorrectableAerCounters parses uncorrectable error counters in
 // /sys/class/net/<iface>/device/aer_dev_[non]fatal.
 func parseUncorrectableAerCounters(devicePath, counterType string,
 	counters *UncorrectableAerCounters,
 ) error {
-	path := filepath.Join(devicePath, "device", "aer_dev_"+counterType)
+	_, err := parseUncorrectableAerFile(filepath.Join(devicePath, "device", "aer_dev_"+counterType), counters)
+	return err
+}
+
+// parseUncorrectableAerFile parses an uncorrectable AER counters file such as
+// aer_dev_fatal or aer_dev_nonfatal. It reports whether the file was present
+// and readable.
+func parseUncorrectableAerFile(path string, counters *UncorrectableAerCounters) (bool, error) {
 	value, err := parsers.ReadFileNoStat(path)
 	if err != nil {
 		if canIgnoreError(err) {
-			return nil
+			return false, nil
 		}
-		return fmt.Errorf("failed to read file %q: %w", path, err)
+		return false, fmt.Errorf("failed to read file %q: %w", path, err)
 	}
 
 	for line := range strings.SplitSeq(string(value), "\n") {
@@ -201,12 +214,12 @@ func parseUncorrectableAerCounters(devicePath, counterType string,
 		}
 		fields := strings.Fields(line)
 		if len(fields) != 2 {
-			return fmt.Errorf("unexpected number of fields: %v", fields)
+			return false, fmt.Errorf("unexpected number of fields: %v", fields)
 		}
 		counterName := fields[0]
 		value, err := strconv.ParseUint(fields[1], 10, 64)
 		if err != nil {
-			return fmt.Errorf("error parsing value for %s: %w", counterName, err)
+			return false, fmt.Errorf("error parsing value for %s: %w", counterName, err)
 		}
 
 		switch counterName {
@@ -251,5 +264,5 @@ func parseUncorrectableAerCounters(devicePath, counterType string,
 		}
 	}
 
-	return nil
+	return true, nil
 }

sysfs/net_class_aer_test.go

@@ -50,7 +50,12 @@ func TestAerCounters(t *testing.T) {
 
 	ac, _ := fs.AerCounters()
 	aerCounters := AllAerCounters{
-		"enp3s0f0": AerCounters{Name: "enp3s0f0"},
+		"enp3s0f0": AerCounters{
+			Name:        "enp3s0f0",
+			Correctable: CorrectableAerCounters{RxErr: 2, BadTLP: 1, Timeout: 3},
+			Fatal:       UncorrectableAerCounters{MalfTLP: 1},
+			NonFatal:    UncorrectableAerCounters{CmpltTO: 1, UnsupReq: 4},
+		},
 		"eth0": AerCounters{
 			Name: "eth0",
 			Correctable: CorrectableAerCounters{

sysfs/pci_device.go

@@ -97,6 +97,18 @@ type PciDevice struct {
 
 	D3coldAllowed *bool          // /sys/bus/pci/devices/<Location>/d3cold_allowed
 	PowerState    *PciPowerState // /sys/bus/pci/devices/<Location>/power_state
+
+	// PCIe Advanced Error Reporting counters. These are nil when the kernel
+	// does not expose AER statistics for the device.
+	AerCorrectable *CorrectableAerCounters   // /sys/bus/pci/devices/<Location>/aer_dev_correctable
+	AerFatal       *UncorrectableAerCounters // /sys/bus/pci/devices/<Location>/aer_dev_fatal
+	AerNonFatal    *UncorrectableAerCounters // /sys/bus/pci/devices/<Location>/aer_dev_nonfatal
+
+	// Totals of the errors a PCIe Root Port has received from the devices
+	// below it. These are nil for devices that are not AER-capable Root Ports.
+	AerRootPortTotalErrCor      *uint64 // /sys/bus/pci/devices/<Location>/aer_rootport_total_err_cor
+	AerRootPortTotalErrFatal    *uint64 // /sys/bus/pci/devices/<Location>/aer_rootport_total_err_fatal
+	AerRootPortTotalErrNonFatal *uint64 // /sys/bus/pci/devices/<Location>/aer_rootport_total_err_nonfatal
 }
 
 func (pd PciDevice) Name() string {
@@ -417,5 +429,62 @@ func (fs FS) parsePciDevice(name string) (*PciDevice, error) {
 		}
 	}
 
+	// Parse AER counters (optional: only present on PCIe devices with AER
+	// support and kernels built with CONFIG_PCIEAER).
+	var correctable CorrectableAerCounters
+	found, err := parseCorrectableAerFile(filepath.Join(path, "aer_dev_correctable"), &correctable)
+	if err != nil {
+		return nil, fmt.Errorf("failed to parse AER counters %s: %w", device.Location, err)
+	}
+	if found {
+		device.AerCorrectable = &correctable
+	}
+	for _, f := range [...]string{"aer_dev_fatal", "aer_dev_nonfatal"} {
+		var uncorrectable UncorrectableAerCounters
+		found, err := parseUncorrectableAerFile(filepath.Join(path, f), &uncorrectable)
+		if err != nil {
+			return nil, fmt.Errorf("failed to parse AER counters %s: %w", device.Location, err)
+		}
+		if !found {
+			continue
+		}
+		switch f {
+		case "aer_dev_fatal":
+			device.AerFatal = &uncorrectable
+		case "aer_dev_nonfatal":
+			device.AerNonFatal = &uncorrectable
+		}
+	}
+
+	// Parse Root Port AER totals (optional: only present on Root Ports).
+	for _, f := range [...]string{"aer_rootport_total_err_cor", "aer_rootport_total_err_fatal", "aer_rootport_total_err_nonfatal"} {
+		name := filepath.Join(path, f)
+		valueStr, err := parsers.SysReadFile(name)
+		if err != nil {
+			if os.IsNotExist(err) {
+				continue
+			}
+			return nil, fmt.Errorf("failed to read AER Root Port file %q %s: %w", name, device.Location, err)
+		}
+
+		valueStr = strings.TrimSpace(valueStr)
+		if valueStr == "" {
+			continue
+		}
+		value, err := strconv.ParseUint(valueStr, 10, 64)
+		if err != nil {
+			return nil, fmt.Errorf("failed to parse %s %q %s: %w", f, valueStr, device.Location, err)
+		}
+
+		switch f {
+		case "aer_rootport_total_err_cor":
+			device.AerRootPortTotalErrCor = &value
+		case "aer_rootport_total_err_fatal":
+			device.AerRootPortTotalErrFatal = &value
+		case "aer_rootport_total_err_nonfatal":
+			device.AerRootPortTotalErrNonFatal = &value
+		}
+	}
+
 	return device, nil
 }

sysfs/pci_device_test.go

@@ -52,6 +52,11 @@ func TestPciDevices(t *testing.T) {
 		NumaNodeNeg1  = int32(-1)
 		D3coldAllowed = true
 		PowerState    = PciPowerStateD0
+
+		// Root Port AER totals on 0000:00:02.1
+		RootPortErrCor      = uint64(7)
+		RootPortErrFatal    = uint64(0)
+		RootPortErrNonFatal = uint64(2)
 	)
 	want := PciDevices{
 		"0000:00:02:1": PciDevice{
@@ -78,6 +83,10 @@ func TestPciDevices(t *testing.T) {
 
 			D3coldAllowed: &D3coldAllowed,
 			PowerState:    &PowerState,
+
+			AerRootPortTotalErrCor:      &RootPortErrCor,
+			AerRootPortTotalErrFatal:    &RootPortErrFatal,
+			AerRootPortTotalErrNonFatal: &RootPortErrNonFatal,
 		},
 		"0000:01:00:0": PciDevice{
 			Location: PciDeviceLocation{
@@ -106,8 +115,11 @@ func TestPciDevices(t *testing.T) {
 			CurrentLinkSpeed: &LinkSpeed8GTs,
 			CurrentLinkWidth: &LinkWidth4,
 
-			D3coldAllowed: &D3coldAllowed,
-			PowerState:    &PowerState,
+			D3coldAllowed:  &D3coldAllowed,
+			PowerState:     &PowerState,
+			AerCorrectable: &CorrectableAerCounters{},
+			AerFatal:       &UncorrectableAerCounters{},
+			AerNonFatal:    &UncorrectableAerCounters{},
 		},
 		"0000:a2:00:0": PciDevice{
 			Location: PciDeviceLocation{
@@ -141,8 +153,11 @@ func TestPciDevices(t *testing.T) {
 			SriovVfTotalMsix:      &SriovVfTotalMsix,
 
 			// Power management fields
-			D3coldAllowed: &D3coldAllowed,
-			PowerState:    &PowerState,
+			D3coldAllowed:  &D3coldAllowed,
+			PowerState:     &PowerState,
+			AerCorrectable: &CorrectableAerCounters{RxErr: 2, BadTLP: 1, Timeout: 3},
+			AerFatal:       &UncorrectableAerCounters{MalfTLP: 1},
+			AerNonFatal:    &UncorrectableAerCounters{CmpltTO: 1, UnsupReq: 4},
 		},
 	}
 

testdata/fixtures.ttar

@@ -10755,6 +10755,21 @@ Lines: 1
 0xc0a9
 Mode: 444
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: fixtures/sys/devices/pci0000:00/0000:00:02.1/aer_rootport_total_err_cor
+Lines: 1
+7
+Mode: 444
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: fixtures/sys/devices/pci0000:00/0000:00:02.1/aer_rootport_total_err_fatal
+Lines: 1
+0
+Mode: 444
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: fixtures/sys/devices/pci0000:00/0000:00:02.1/aer_rootport_total_err_nonfatal
+Lines: 1
+2
+Mode: 444
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Path: fixtures/sys/devices/pci0000:00/0000:00:02.1/ari_enabled
 Lines: 1
 0
@@ -17338,15 +17353,15 @@ Mode: 755
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Path: fixtures/sys/devices/pci0000:a2/0000:a2:00.0/aer_dev_correctable
 Lines: 9
-RxErr 0
-BadTLP 0
+RxErr 2
+BadTLP 1
 BadDLLP 0
 Rollover 0
-Timeout 0
+Timeout 3
 NonFatalErr 0
 CorrIntErr 0
 HeaderOF 0
-TOTAL_ERR_COR 0
+TOTAL_ERR_COR 6
 Mode: 444
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Path: fixtures/sys/devices/pci0000:a2/0000:a2:00.0/aer_dev_fatal
@@ -17360,7 +17375,7 @@ CmpltTO 0
 CmpltAbrt 0
 UnxCmplt 0
 RxOF 0
-MalfTLP 0
+MalfTLP 1
 ECRC 0
 UnsupReq 0
 ACSViol 0
@@ -17369,7 +17384,7 @@ BlockedTLP 0
 AtomicOpBlocked 0
 TLPBlockedErr 0
 PoisonTLPBlocked 0
-TOTAL_ERR_FATAL 0
+TOTAL_ERR_FATAL 1
 Mode: 444
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Path: fixtures/sys/devices/pci0000:a2/0000:a2:00.0/aer_dev_nonfatal
@@ -17379,20 +17394,20 @@ DLP 0
 SDES 0
 TLP 0
 FCP 0
-CmpltTO 0
+CmpltTO 1
 CmpltAbrt 0
 UnxCmplt 0
 RxOF 0
 MalfTLP 0
 ECRC 0
-UnsupReq 0
+UnsupReq 4
 ACSViol 0
 UncorrIntErr 0
 BlockedTLP 0
 AtomicOpBlocked 0
 TLPBlockedErr 0
 PoisonTLPBlocked 0
-TOTAL_ERR_NONFATAL 0
+TOTAL_ERR_NONFATAL 5
 Mode: 444
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Path: fixtures/sys/devices/pci0000:a2/0000:a2:00.0/ari_enabled