[systemd-devel] Detecting inactive sessions

[David Herrmann]

Hi Bastien

On Mon, Nov 3, 2014 at 5:30 PM, Bastien Nocera <hadess at hadess.net> wrote:
> On Mon, 2014-11-03 at 17:28 +0100, David Herrmann wrote:
>> Hi
>>
>> On Wed, Oct 29, 2014 at 3:45 PM, Bastien Nocera <hadess at hadess.net> wrote:
>> > For a very specific definition of inactive.
>> >
>> > I'm looking at a way for the iio-sensor-proxy at:
>> > https://github.com/hadess/iio-sensor-proxy
>> > to suspend reading from accelerometers (or maybe to turn them off), when
>> > all the sessions are locked and the screens turned off.
>> >
>> > This would usually mean that I would enable reading from the sensor if
>> > one session is "active" and stop reading if none are "active". Is this
>> > correct? Is it up to the session manager (eg. gnome-session) to tell us
>> > whether a session is active or not, or do I have this backwards?
>>
>> For uhid (similar to uinput) you get an OPEN and CLOSE event whenever
>> the first/last user opens/closes the device you created. I think we
>> want something similar for uinput. That is, when a gnome session is
>> inactive, they should just close the input devices that were created
>> by iio-sensor-proxy (done automatically if you use the systemd-logind
>> API to access devices). This way, iio-sensor-proxy knows whenever at
>> least one session uses the data. This is also how most kernel-internal
>> APIs work.
>
> The session doesn't read from the uinput device. The iio-sensor-proxy
> just sends out a kevent, which is caught by the accelerometer helper in
> udev. The GNOME session catches the udev event and reads the changed
> property.

Ugh, you're right, of course!

So we have this user-space driver, iio-sensor-proxy, which wants to
stop reporting data if the there is no consumer of it. Still, the
obvious way is for the consumers to tell iio-sensor-proxy. Given the
current design (via uevents), this is not possible, though.
I have no idea how to fix this up. I really dislike adding heuristics
to iio-sensor-proxy to "guess" whether there is any consumer of the
data. Imagine there's a system that uses the orientation data to
control sound-output, instead of video-output. How would you know that
in iio-sensor-proxy? The system might look idle, all displays are off,
but still, someone might want this data.

btw., looking at iio-sensor-proxy: you send uevents for _each_
accelerometer event? uevents are _really_ heavy, compared to input
events. I'm not sure this is a good idea, unless the accelerometers
report data only every few seconds.

Thanks
David