[RFC,0/8] Stop monitoring disabled devices

Message ID 20200407174926.23971-1-andrzej.p@collabora.com
Headers show
  • Stop monitoring disabled devices
Related show


Andrzej Pietrasiewicz April 7, 2020, 5:49 p.m.
The current kernel behavior is to keep polling the thermal zone devices
regardless of their current mode. This is not desired, as all such "disabled"
devices are meant to be handled by userspace, so polling them makes no sense.

There was an attempt to solve this issue:


and it ultimately has not succeeded:


This is a new attempt addressing all the relevant drivers, and I have
identified them with:

$ git grep "thermal_zone_device_ops" | grep "= {" | cut -f1 -d: | sort | uniq

The idea is to modify thermal_zone_device_update() and monitor_thermal_zone()
in such a way that they stop polling a disabled device. To do decide what to
do they should call ->get_mode() operation of the specialized thermal zone
device in question (e.g. drivers/acpi/thermal.c's). But here comes problem:
sometimes a thermal zone device must be initially disabled and becomes enabled
only after its sensors appear on the system. If such thermal zone's
->get_mode() /* in the context of thermal_zone_device_update() or
monitor_thermal_zone() */ is called _before_ the sensors are available, it will
be reported as "disabled" and consequently polling it will be ceased. This is
a change in behavior from userspace's perspective.

To solve the above described problem I want to introduce the third mode of a
thermal_zone_device: initial. The idea is that when the device is in its
initial mode, then its polling will be handled as it is now. This is a good
thing: should the temperature be just about hitting the critical treshnold
early during the boot process, it might be too late if we wait for the
userspace to run to save the system from overheating. The initial mode should
be reported in sysfs as "enabled" to keep the userspace interface intact.