Re: [PATCH v5 2/2] pwm: sifive: Add a driver for SiFive SoC PWM

[Uwe Kleine-König]

Hello,

On Thu, Feb 14, 2019 at 04:59:03PM +0100, Uwe Kleine-König wrote:
> On Wed, Feb 13, 2019 at 01:37:03PM +0100, Thierry Reding wrote:
> > On Thu, Feb 07, 2019 at 11:16:57AM +0100, Uwe Kleine-König wrote:
> > > On Tue, Jan 29, 2019 at 05:13:19PM +0530, Yash Shah wrote:
> > [...]
> > > > diff --git a/drivers/pwm/pwm-sifive.c b/drivers/pwm/pwm-sifive.c
> > [...]
> > > > +	writel(val, pwm->regs + PWM_SIFIVE_PWMCFG);
> > > > +
> > > > +	writel(frac, pwm->regs + PWM_SIFIVE_PWMCMP0 + dev->hwpwm * SIZE_PWMCMP);
> > > > +
> > > > +	val &= ~(1 << PWM_SIFIVE_PWMCFG_DEGLITCH);
> > > > +	writel(val, pwm->regs + PWM_SIFIVE_PWMCFG);
> > > > +
> > > > +	pwm_sifive_get_state(chip, dev, state);
> > > 
> > > Thierry: This changes the pwm_state. Is this how this should be done?
> > 
> > Yes, I think that's fine. The PWM state should always reflect the
> > current hardware state. If the configuration that we program does not
> > reflect the state that was requested, that should be reflected in the
> > PWM state.
> 
> I'm not sure you blessed what is really done here. If I do:
> 
> 	state.duty_cycle = state.period;
> 	pwm_apply_state(pwm, &state);
> 
> the call in question doesn't only result in pwm->state.duty_cycle <
> pwm->state.period, but it also corrects my local state variable (i.e. I
> have state.duty_cycle < state.period afterwards).
> 
> Is this what you thought to be fine?

I thought a bit about this and I'm convinced that updating struct
pwm_device::state is/might be fine, but changing the caller's struct pwm_state
that was passed to pwm_apply_state is not.

Consider a consumer who does:

	#define PERIOD 5000000
	#define DUTY_LITTLE 10
	...
	struct pwm_state state = {
		.period = PERIOD,
		.duty_cycle = DUTY_LITTLE,
		.polarity = PWM_POLARITY_NORMAL,
		.enabled = true,
	};

	pwm_apply_state(mypwm, &state);
	...
	state.duty_cycle = PERIOD / 2;
	pwm_apply_state(mypwm, &state);

I think the second request should have state.period = 5000000 and not
some other value (that might only have chosen by the respective driver
because the first duty cycle was so short).

Best regards
Uwe

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