[mako] upower consuming over 17% cpu when in idle

[Pat McGowan]

From the meta battery bug # 1372413
mako running 191

As device was not suspending and battery level running down, upower is reported at 17%+ cpu by the cpustat tool
 mpdecision and systemd-udevd also change their behavior.

The upower cpu consumption is just a side effect of the amount of battery related events the kernel send, and specific to mako.

[Ricardo Salveti (rsalveti)]

It seems the problem relies more on the kernel side and with the amount of events the battery driver sends to the userspace (even when there is no change).

udevadm monitor is already enough to show the amount of kernel events the battery driver sends to the userspace.

[Ricardo Salveti (rsalveti)]

As an example, the amount of upower events when connecting/disconnecting the usb cable (charger) on mako:
root@ubuntu-phablet:/sys# upower -m
Monitoring activity from the power daemon. Press Ctrl+C to cancel.
[03:54:38.613] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.616] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.620] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.621] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.621] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.622] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.624] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.632] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.633] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.634] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.636] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:38.636] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.122] device changed: /org/freedesktop/UPower/devices/line_power_pm8921_dc
[03:54:39.128] device changed: /org/freedesktop/UPower/devices/line_power_pm8921_dc
[03:54:39.129] device changed: /org/freedesktop/UPower/devices/line_power_pm8921_dc
[03:54:39.129] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.130] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.131] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.131] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.131] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.132] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.132] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.133] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.133] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.134] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.134] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.143] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.144] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.144] daemon changed:
[03:54:39.427] device changed: /org/freedesktop/UPower/devices/line_power_pm8921_dc
[03:54:39.427] device changed: /org/freedesktop/UPower/devices/line_power_pm8921_dc
[03:54:39.428] device changed: /org/freedesktop/UPower/devices/line_power_pm8921_dc
[03:54:39.430] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.431] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.432] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.432] device changed: /org/freedesktop/UPower/devices/battery_battery
[03:54:39.433] device cha...

[Ricardo Salveti (rsalveti)]

After profiling upower, this is actually not an issue in there, as the only thing it is doing is getting the battery change event from the kernel, updating the internal struct and sending another dbus message so the clients could see the changes.

The main issue here is that the kernel is sending a big amount of battery events (all valid, and different changes), which results in udev, upower and dbus consuming a bit of your cpu.

Verified this bug against both krillin and arale and that is not the case there, as I already did the profiling and reduced the amount of battery events.

As this bug is just specific to mako, and would require a kernel change to reduce the amount of events, we can lower the priority and fix in vivid only.

[Pat McGowan (pat-mcgowan)]

not seen in some time