Hello Daniel,

On Fri, Oct 20, 2023 at 12:27:27PM +0100, Daniel Thompson wrote:
> On Wed, Oct 18, 2023 at 11:07:41PM +0200, Uwe Kleine-König wrote:
> > Hello Philipp,
> >
> > On Thu, Jun 08, 2023 at 04:11:14PM +0200, Philipp Zabel wrote:
> > > The initial PWM state returned by pwm_init_state() has a duty cycle
> > > of 0 ns.
> >
> > This is only true for drivers without a .get_state() callback, isn't it?
> 
> pwm_init_state() explicitly zeros the duty-cycle in order to avoid
> problems when the default args have a different period to the currently
> applied config:
> https://elixir.bootlin.com/linux/latest/source/include/linux/pwm.h#L174

Ah right, pwm_init_state() is strange in a different way than I
remembered :-) pwm_get_state() is only called to get .enabled set
appropriately.

Looking at the callers:

 - drivers/gpu/drm/solomon/ssd130x.c
   It does:
 	pwm_init_state(ssd130x->pwm, &pwmstate);
	pwm_set_relative_duty_cycle(&pwmstate, 50, 100);
	pwm_apply_state(ssd130x->pwm, &pwmstate);
	pwm_enable(ssd130x->pwm);

   A probably better result can be reached quicker using:
 	pwm_init_state(ssd130x->pwm, &pwmstate);
	pwm_set_relative_duty_cycle(&pwmstate, 50, 100);
	pwmstate.enabled = true;
	pwm_apply_state(ssd130x->pwm, &pwmstate);
 - drivers/hwmon/pwm-fan.c
   __set_pwm should probably explicitly set .enabled.
   All other calls to pwm_apply_state set .enabled explicitly.

 - drivers/input/misc/da7280.c
   explicitly sets .enabled after calling pwm_init_state()

 - drivers/input/misc/pwm-beeper.c
   explicitly sets .enabled after calling pwm_init_state()

 - drivers/input/misc/pwm-vibra.c
   explicitly sets .enabled after calling pwm_init_state()

 - drivers/leds/leds-pwm.c
   explictily sets .enabled before calling pwm_apply_state()

 - drivers/leds/rgb/leds-pwm-multicolor.c
   explictily sets .enabled before calling pwm_apply_state()

 - drivers/media/rc/ir-rx51.c
   explictily sets .enabled before calling pwm_apply_state()

 - drivers/media/rc/pwm-ir-tx.c
   explictily sets .enabled before calling pwm_apply_state()

 - drivers/regulator/pwm-regulator.c
   never sets .enabled, probably a bug

 - drivers/video/backlight/lm3630a_bl.c
   explictily sets .enabled before calling pwm_apply_state()

 - drivers/video/backlight/lp855x_bl.c
   explictily sets .enabled before calling pwm_apply_state()

 - drivers/video/backlight/pwm_bl.c
   This is the one we currently discuss. I think even with the patch
   applied it uses the .enabled value returned by pwm_init_state() but
   it shouldn't.

 - drivers/video/fbdev/ssd1307fb.c
   Similar to drivers/gpu/drm/solomon/ssd130x.c. Probably the one was
   copied to the other given that it seems to handle the same hardware.

So all consumers using pwm_init_state() either don't use the .enabled
value returned by pwm_init_state() or at least shouldn't do that.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |