mesh-catalog/modules/asus-zephyrus-g14/cmd/zephyrus/thermals.go

package main

import (
	"context"
	"path"
	"sort"
	"strconv"
	"strings"
)

// Sensor is one temperature, fan or power reading from hwmon.
type Sensor struct {
	Chip  string  `json:"chip"`
	Label string  `json:"label"`
	Value float64 `json:"value"`
}

// DGPU is the discrete GPU as the PCI bus and its driver see it.
type DGPU struct {
	Address string   `json:"pci_address"`
	Runtime string   `json:"runtime_status"`
	Name    string   `json:"name,omitempty"`
	TempC   *float64 `json:"temp_c,omitempty"`
	PowerW  *float64 `json:"power_w,omitempty"`
	PState  string   `json:"pstate,omitempty"`
	Note    string   `json:"note,omitempty"`
}

// hwmon reads every hwmon reading of one kind: "temp" (°C), "fan" (RPM) or "power" (W).
func (m *Machine) hwmon(kind string) []Sensor {
	var out []Sensor
	for _, dir := range m.glob("/sys/class/hwmon/hwmon*") {
		chip := m.read(dir + "/name")
		inputs := m.glob(dir + "/" + kind + "*_input")
		if kind == "power" {
			inputs = append(inputs, m.glob(dir+"/power*_average")...)
		}
		for _, in := range inputs {
			v, ok := m.readInt(in)
			if !ok {
				continue
			}
			base := path.Base(in)
			stem := base[:strings.LastIndex(base, "_")]
			label := m.read(dir + "/" + stem + "_label")
			if label == "" {
				label = base
			} else if strings.HasSuffix(base, "_average") {
				label += " (average)"
			}
			value := float64(v)
			switch kind {
			case "temp":
				value = round1(value / 1000)
			case "power":
				value = round1(value / 1e6)
			}
			out = append(out, Sensor{Chip: chip, Label: label, Value: value})
		}
	}
	sort.Slice(out, func(i, j int) bool {
		if out[i].Chip != out[j].Chip {
			return out[i].Chip < out[j].Chip
		}
		return out[i].Label < out[j].Label
	})
	return out
}

// dgpu finds the NVIDIA display controller and, only when it is already awake, asks its driver for
// its temperature and draw. **Asking wakes it**: nvidia-smi brings a suspended GPU out of D3, which
// is the power a reading of power draw should not cost.
func (m *Machine) dgpu(ctx context.Context) *DGPU {
	for _, dir := range m.glob("/sys/bus/pci/devices/*") {
		if m.read(dir+"/vendor") != "0x10de" || !strings.HasPrefix(m.read(dir+"/class"), "0x03") {
			continue
		}
		g := &DGPU{Address: path.Base(dir), Runtime: m.read(dir + "/power/runtime_status")}
		if g.Runtime != "active" {
			g.Note = "the discrete GPU is " + g.Runtime + "; not woken to be read"
			return g
		}
		out, err := m.Run(ctx, "nvidia-smi", "--query-gpu=name,temperature.gpu,power.draw,pstate", "--format=csv,noheader,nounits")
		if err != nil {
			g.Note = "nvidia-smi: " + err.Error()
			return g
		}
		f := strings.Split(strings.TrimSpace(strings.SplitN(out, "\n", 2)[0]), ",")
		if len(f) >= 4 {
			g.Name = strings.TrimSpace(f[0])
			if t, err := strconv.ParseFloat(strings.TrimSpace(f[1]), 64); err == nil {
				g.TempC = &t
			}
			if w, err := strconv.ParseFloat(strings.TrimSpace(f[2]), 64); err == nil {
				w = round1(w)
				g.PowerW = &w
			}
			g.PState = strings.TrimSpace(f[3])
		}
		return g
	}
	return nil
}

// Thermals is what the thermals tool answers.
type Thermals struct {
	Temperatures []Sensor `json:"temperatures_c"`
	Fans         []Sensor `json:"fans_rpm"`
	DGPU         *DGPU    `json:"dgpu,omitempty"`
	Profile      string   `json:"platform_profile,omitempty"`
	Hottest      *Sensor  `json:"hottest,omitempty"`
}

func (m *Machine) Thermals(ctx context.Context) Thermals {
	t := Thermals{Temperatures: m.hwmon("temp"), Fans: m.hwmon("fan"), DGPU: m.dgpu(ctx),
		Profile: m.read("/sys/firmware/acpi/platform_profile")}
	if t.Temperatures == nil {
		t.Temperatures = []Sensor{}
	}
	if t.Fans == nil {
		t.Fans = []Sensor{}
	}
	for i := range t.Temperatures {
		if t.Hottest == nil || t.Temperatures[i].Value > t.Hottest.Value {
			h := t.Temperatures[i]
			t.Hottest = &h
		}
	}
	return t
}

// PowerDraw is what the power-draw tool answers.
type PowerDraw struct {
	Source      Source   `json:"source"`
	BatteryW    *float64 `json:"battery_w,omitempty"`
	BatteryFlow string   `json:"battery_flow,omitempty"`
	CPUPackageW *float64 `json:"apu_package_w,omitempty"`
	DGPU        *DGPU    `json:"dgpu,omitempty"`
	Note        string   `json:"note"`
}

func (m *Machine) PowerDraw(ctx context.Context) PowerDraw {
	p := PowerDraw{Source: PowerSource(m.Supplies()), DGPU: m.dgpu(ctx),
		Note: "on battery, battery_w is what the whole machine draws; on mains it is only what the battery takes or gives"}
	for _, b := range m.Batteries() {
		if b.PowerW != nil {
			w := *b.PowerW
			p.BatteryW = &w
			switch strings.ToLower(b.Status) {
			case "discharging":
				p.BatteryFlow = "discharging"
			case "charging":
				p.BatteryFlow = "charging"
			default:
				p.BatteryFlow = strings.ToLower(b.Status)
			}
			break
		}
	}
	// The integrated GPU's hwmon reports the whole APU's package power (PPT) on this model.
	for _, s := range m.hwmon("power") {
		if s.Chip == "amdgpu" && s.Label == "PPT" {
			w := s.Value
			p.CPUPackageW = &w
		}
	}
	return p
}