On Wednesday 23 January 2013 17:45:39 Alex Courbot wrote:
quoted
I'm confused. Why would you want to call into pwm_bl directly? If we're
going to split this up into separate platform devices, why not look up a
given backlight device and use the backlight API on that? The pieces of
the puzzle are all there: you can use of_find_backlight_by_node() to
obtain a backlight device from a device tree node, so I'd expect the DT
to look something like this:
backlight: backlight {
compatible = "pwm-backlight";
...
};
This would still prevent any power control from the backlight driver. I.e.
if someone sets the brightness to 0 through sysfs, we cannot power the
backlight off as pwm-backlight cannot control more than the PWM without
platform callbacks. Backlight could only be powered off as a result of a fb
blank event.
In order to solve this, how about adding a backlight notifier call chain to
broadcast backlight events in a fashion similar to what is done in
fbmem/fbcon? Then backlight_update_status() could send events like
BACKLIGHT_EARLY_EVENT_UPDATE and BACKLIGHT_EVENT_UPDATE to which panel drivers
could subscribe in order to power the backlight up and down as needed.
Then as the backlight is mentioned in the panel's DT node,
panel: panel {
compatible = "...";
...
backlight = <&backlight>;
...
};
the panel's driver could listen to backlight-related events and do its stuff
transparently, without changing anything to the backlight drivers. This would
be a good way to replace pwm-backlight's callbacks for platforms that use the
DT, and would also be applicable to any backlight class device.
Generally speaking, having a mean to monitor backlights state in the kernel
does not seem too crazy, especially since we already have a way to notify user
space through backlight_generate_event().
Richard, does that sound ok to you?
Alex.