🛠 Surveillance thermique big.LITTLE et ventilateur

Objectives

Monitor and control the thermal behavior of the Exynos5422: under mainline, the PWM fan (levels 0/130/170/230) increases in steps at 50/60/70 °C, the A15 cluster is passively throttled at 70 °C (−2 DVFS steps, ≈1.8/1.2 GHz) then at 85 °C (≈600 MHz), critical at 120 °C, with a separate thermal zone for the Mali-T628. At 2.0 GHz, active cooling is essential.

Hardware

• ODROID-XU4 with its original heatsink + fan (the passive XU4Q loses ~16% of performance under sysbench at 2.0 GHz)
• sudo apt install stress-ng for load generation

Full code

#!/usr/bin/env bash
# xu4-thermique.sh — surveillance big.LITTLE + test de chauffe
set -euo pipefail

# 1) Temperatures : un thermal_zone par capteur TMU (cpu0..cpu3 + gpu)
paste <(cat /sys/class/thermal/thermal_zone*/type) \
      <(awk '{printf "%.1f C\n", $1/1000}' /sys/class/thermal/thermal_zone*/temp)

# 2) Fréquences courantes par cluster (A15 <= 2.0 GHz, A7 <= 1.4 GHz)
grep . /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq

# 3) Politique du ventilateur : pwm-fan, cooling-levels 0/130/170/230
cat /sys/class/thermal/thermal_zone0/trip_point_*_temp 2>/dev/null || true

# 4) Test de chauffe 60 s puis observation du decrochage A15
stress-ng --cpu 8 --timeout 60s &
for i in $(seq 1 12); do sleep 5
  echo "--- T+$((i*5))s"; awk '{printf "%.1f C ", $1/1000}' \
    /sys/class/thermal/thermal_zone*/temp; echo
grep . /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq
done

Expected verification

After 20-40 s of load, the A15 cluster frequencies drop below 2.0 GHz (steps 1.8 → 1.2 GHz) while the fan ramps up; upon release, everything returns to 2000000 kHz. Without a functioning fan, permanent throttling confirms overheating.

Common errors

• Running the board without a functional fan at 2.0 GHz: permanent throttling, total silence requires limiting to 1.2 GHz.
• Writing to /sys/class/pwm/pwm0: the channel is claimed by pwm-fan, use the thermal governor instead.
• Confusing the cpu and gpu zones: the Mali has its own thermal zone with the same scale.
• Weak 5 V power supply causing frequency drops independently of temperature.