-
-
Notifications
You must be signed in to change notification settings - Fork 14
Inkbox Power Daemon IPD
All configuration is stored in the /data/config/20-sleep_daemon
directory. Below is a list and description of the files that are present in that directory.
This is a list of apps that will be frozen (SIGSTOP
) by the power daemon by default before going to sleep, it's something like this:
inkbox-bin
oobe-inkbox-bin
lockscreen-bin
qalculate-bin
qreversi-bin
2048-bin
scribble
lightmaps
It will be created as a fallback, but it should be created before.
This file contains an integer that determines how long will be the delay between switching one percent of brightness.
By default, it's 50
. If the file doesn't exist, it will be created.
Experimental feature, sets a cpuGovernor at boot, or when variables are updated.
By default, it's ondemand
. the options are:
interactive
conservative
userspace
powersave
ondemand
performance
If this file's content is false
, the device will not go to sleep when it's charged (because some devices can't). By default it's different for many devices, because it happens in devices from 2011 to ~2017 (N905B, N905B, N705, N613, N236, N437, KT )
If this file's content is true
and if the program detects that between going to sleep and after suspending that the device's charger state changed, it will try to sleep anyway.
By default it's set to false
. If the 3-whenChargerSleep
option is set to false
, then this option should also be false.
If this file's content is true
, Wi-Fi will be reconnected on wake-up.
By default, this option is set to true
.
If this file's content is true
, the following will happen: the power LED will light up for indicating that the device is going to sleep. The blinking goes as follows:
- Firstly, slow led blinking indicates invidual steps in preparing to sleep (turn off Wi-Fi, show screensaver, etc.)
- Secondly, fast blinking indicates write to state-extended, so all devices components are preparing to sleep
- Thirdly, blinking (but faster than in the first step) indicates that the device is trying to sleep, but one of the device's components (touchscreen propably) refused to. That's normal behavior, so don't worry It will also turn it on when the device is charging and will be turned off when the device finishes charging.
This file contains an integer that determines, in seconds, when to go to sleep when no touch input is received from the screen. 0
means 'Never', minimum is 15
seconds.
By default, it's set to 300
.
This option is complicated, I will try my best to explain it.
If you have a 3D-printed case with a magnet in it, there is a big chance that the magnet will put the device to sleep when you flip the case to the back. This option skips the every one magnet call.
A little visualization:
- The device boots up, the case is opened, no magnet contact. This option is enabled.
- You flip the case to the back, magnet is close to the hall sensor and triggers it. You don't want that. With this option, this call will be ignored.
- Now you close the case, magnet contacts, the device goes to sleep.
A bit of synchronization is needed, it's a bit complicated to use but it works, and is great (on the Kobo Nia). It's just an option, don't care about it if you don't want to :)
By default, it's set to false
.
This file is only read by the GUI. If it's set to true
(from the GUI power daemon settings) the power menu option of the eReader called 'Suspend' will be replaced by 'Deep sleep'. If this is clicked, /run/ipd/sleepCall
will be created and deepSleep
will be written to it instead of sleep
(as it usually would). In this way inotify will be triggered, and the device will go to sleep, but with extra things, like:
-
powersave
CPU governor, which will extend the time the device takes to wake-up, but the battery will live longer.
If this file exists and contains a valid path, it will be run every time the charger is run, and it will wait for that program to stop. Useful if you are crazy and want control your charger current using i2c, or just want to switch USB to 'Host' mode. https://github.com/Szybet/Charger-controller
If true
is written to this file, the daemon will update all its variables on-the-fly, then write false
to this file.
If the environment variable argument DEBUG
is set to true
, the daemon will output logs to stdio and /var/log/ipd.log
.
You can also enable debug logging from the OS itself, by creating /boot/flags/IPD_DEBUG
with the value 'true' in it.
In /run/ipd
(which is an 8K temporary filesystem) is a named pipe called fifo
. It's bind-mounted between the main root filesystem and the GUI's root filesystem (to /kobo/dev/ipd
). User applications can access it at /dev/ipd/fifo
, it's read-only.
Messages sent in this named pipe always have a length of 5 bytes. It makes them easier to parse.
Current messages sent by the power daemon:
-
start
indicates that the device is going to sleep. The application has 300 ms to react to this message. Because Qt applications get all touch input at wake-up even if the device is suspended (buffer), which may cause confusion and problems, it's advised to show a QDialog withexec()
, which eats all touch input until itshide()
is called. For code examples look intosrc/widgets/powerDaemon
in InkBox source code. -
stop0
is called after a device wakes up.
Any image you put in /data/onboard/.screensaver
(yes, /data/onboard
is the directory where you would usually put books) with the extension .png
will be randomly chosen and used as a screensaver. The "Sleeping" text or the image's size needs to be handled by the user itself. Additional notes:
- Yes, you create that directory and yes, it is a hidden directory
- If you want a single screensaver, put only one image there
- If no image is found, the normal screensaver will be used
-
/kobo/tmp/currentlyRunningUserApplication
is a file which has the current user application process name, to freeze it - "Why use
sleepCall
when you have the named pipe?" I would need to create another thread to watch the named pipe, which is bad because there are already 4 other threads and it gets slowly CPU-heavy. - The daemon grabs the
/dev/input/event0
input device for sanity reasons.