pwm-sun4i: PWM backlight is not turned off on shutdown

[Daniel Thompson]

On Thu, Aug 27, 2020 at 10:55:28PM +0300, Andrey Lebedev wrote:
> Hello,
> 
> I think I'm experiencing problem with pwm-sun4i module. I'll describe
> the symptoms first.
> 
> I have a device, based on Allwinner A20 (Cubieboard 2) with LVDS display
> that has a PWM-based backlight. The problem is: when linux shuts down,
> the backlight stays on. I expect it to be turned off. This used to work
> as expected on kernel 5.2-rc2, but after upgrade to 5.8 the backlight
> does not turn off anymore (most of the times, see below).
> 
> The backlight is configured in the device tree [1]. The brightness can
> be changed by writing to "brightness" file on sysfs. So, linux can
> control the PWM line. Backlight sysfs directory also has a "bl_power"
> file, which can accept "0" to power on or "4" to power off the backlight
> (according to [2]).
> 
> Now, writing "4" to bl_power sometimes turns the backlight off and
> sometimes not. I've found that the probability of backlight turning off
> pretty much correlates with the current screen brightness: on 100%
> brightness it will never turn off, on 50% brightness it will turn off on
> about half of the times. When backlight does not turn off, it goes on
> full brightness. It feels like the line, controlled by pwm stays in
> whatever state it was the moment backlight was powered down - either
> full 1 or 0.
> 
> The pwm backlight device driver (pwm_bl) requests to set the duty cycle
> to 0 and disable the pwm with the same request [3], but I suspect the
> implementation driver (pwm-sun4i) does not actually set the duty cycle
> to 0 before disabling the pulse width modulation.
> 
> Is there anything that can be done to fix this?

There's some rather odd logic in sun4i_pwm_apply() that results in the
PWM being disabled twice... once when it applies the initial config
and again after waiting for a duty_cycle.

I suspect disabling the initial disable would solve your issue... but it
might provoke some new ones!

Anyhow, try removing the else clause starting at line 299 and see what
happens:
https://elixir.bootlin.com/linux/v5.8/source/drivers/pwm/pwm-sun4i.c#L299


Daniel.