Comment 22 for bug 1769236

Revision history for this message
Glen Ditchfield (gjditchfield) wrote :

I will attach /var/run/thermald/thermal-conf.xml.auto, and
thermal.log (generated by sudo thermald --no-daemon --loglevel=debug)
to illustrate thermald's bad behaviour.

Thermal.log up to line 500 shows the output when the laptop was idling:
sensor B0D4 stayed cooler than 40°C, the CPUs ran at 900 MHz, and
thermald is (I think) using rapl_controller for thermal control.

At that point I ran c-ray. CPU speed rose to 2400 MHz, and the
temperature began to rise (53°C at line 528, 64°C at line 632).

At line 656 thermald decided to switch from rapl_controller to
intel_pstate. At line 699 intel_pstate disabled turbo; CPU speed
dropped to 1800 MHz, and the temperature began to fall, from 71°C at
line 686 down to 54°C at line 799.

At that point, even though temperatures were falling and no fans were
running, thermald switched from intel_pstate to intel_powerclamp. CPU
speed dropped to 400 MHz, and performance tanked.

I think this switch to intel_powerclamp is the problem. It seems to
happen whenever the temperature exceeds 40°C for some number of polling
periods, which is not a useful thermal policy. Most web sites drive
the CPU hard enough to trigger it.