Re: [v5 2/2] pwm: sifive: change the PWM controlled LED algorithm

[Nylon Chen]

Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx> 於 2023年12月12日 週二 上午4:50寫道：
>
> Hello Nylon,

Hi Uwe, thanks for your feedback.
>
>
>
> On Tue, Oct 24, 2023 at 06:19:02PM +0800, Nylon Chen wrote:
> > The `frac` variable represents the pulse inactive time, and the result
> > of this algorithm is the pulse active time. Therefore, we must reverse the result.
> >
> > The reference is SiFive FU740-C000 Manual[0]
> >
> > Link: https://sifive.cdn.prismic.io/sifive/1a82e600-1f93-4f41-b2d8-86ed8b16acba_fu740-c000-manual-v1p6.pdf [0]
> >
> > Signed-off-by: Nylon Chen <nylon.chen@xxxxxxxxxx>
> > Signed-off-by: Vincent Chen <vincent.chen@xxxxxxxxxx>
> > ---
> >  drivers/pwm/pwm-sifive.c | 10 ++++++----
> >  1 file changed, 6 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/pwm/pwm-sifive.c b/drivers/pwm/pwm-sifive.c
> > index eabddb7c7820..353c2342fbf1 100644
> > --- a/drivers/pwm/pwm-sifive.c
> > +++ b/drivers/pwm/pwm-sifive.c
> > @@ -101,7 +101,7 @@ static void pwm_sifive_update_clock(struct pwm_sifive_ddata *ddata,
> >
> >       /* As scale <= 15 the shift operation cannot overflow. */
> >       num = (unsigned long long)NSEC_PER_SEC << (PWM_SIFIVE_CMPWIDTH + scale);
> > -     ddata->real_period = div64_ul(num, rate);
> > +     ddata->real_period = DIV_ROUND_UP_ULL(num, rate);
>
> It's unclear to me, why you changed that.
Because there is a gap in idempotent tests.
e.g.
root@unmatched:~# echo 110 >
/sys/devices/platform/led-controller-1/leds/d12/brightness
[  706.987712] .apply is not idempotent (ena=1 pol=0 1739692/4032985)
-> (ena=1 pol=0 1739630/4032985)
root@unmatched:~# echo 120 >
/sys/devices/platform/led-controller-1/leds/d12/brightness
[  709.817554] .apply is not idempotent (ena=1 pol=0 1897846/4032985)
-> (ena=1 pol=0 1897784/4032985)

Round the result to the nearest whole number. This ensures that
real_period is always a reasonable integer that is not lower than the
actual value.

After modification, idempotent errors can be avoided.
>
>
> >       dev_dbg(ddata->chip.dev,
> >               "New real_period = %u ns\n", ddata->real_period);
> >  }
> > @@ -121,13 +121,14 @@ static int pwm_sifive_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
> >               state->enabled = false;
> >
> >       state->period = ddata->real_period;
> > +
> > +     duty = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - duty;
>
> I would have placed that directly after
>
>         duty = readl(...);
>
> which then also influences
>
>         state->enabled = duty > 0;
>
> (as it should?).
>
Yes, you are right. I will make relevant corrections.
...
        duty = readl(ddata->regs + PWM_SIFIVE_PWMCMP(pwm->hwpwm));
+       duty = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - duty;
-       state->enabled = duty <= 65535;
+       state->enabled = duty > 0;
...
        state->period = ddata->real_period;
-       duty = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - duty;

>
> >       state->duty_cycle =
> >               (u64)duty * ddata->real_period >> PWM_SIFIVE_CMPWIDTH;
> > -     state->polarity = PWM_POLARITY_INVERSED;
> > +     state->polarity = PWM_POLARITY_NORMAL;
> >
> >       return 0;
> >  }
> > -
> >  static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> >                           const struct pwm_state *state)
> >  {
> > @@ -139,7 +140,7 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> >       int ret = 0;
> >       u32 frac;
> >
> > -     if (state->polarity != PWM_POLARITY_INVERSED)
> > +     if (state->polarity != PWM_POLARITY_NORMAL)
> >               return -EINVAL;
> >
> >       cur_state = pwm->state;
> > @@ -158,6 +159,7 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> >       num = (u64)duty_cycle * (1U << PWM_SIFIVE_CMPWIDTH);
> >       frac = DIV64_U64_ROUND_CLOSEST(num, state->period);
> >       /* The hardware cannot generate a 100% duty cycle */
> > +     frac = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - frac;
> >       frac = min(frac, (1U << PWM_SIFIVE_CMPWIDTH) - 1);
>
> frac can only be > (1U << PWM_SIFIVE_CMPWIDTH) - 1 if an overflow
> happend the line above. Is that what you want here?
I made a mistake, I pushed the wrong changes.

I want to invert it after taking the minimum value, which makes sense to me.
        frac = DIV64_U64_ROUND_CLOSEST(num, state->period);
        /* The hardware cannot generate a 100% duty cycle */
        frac = min(frac, (1U << PWM_SIFIVE_CMPWIDTH) - 1);
+       frac = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - frac;

>
> >       mutex_lock(&ddata->lock);
>
> Best regards
> Uwe
Best regards
Nylon
>
> --
> Pengutronix e.K.                           | Uwe Kleine-König            |
> Industrial Linux Solutions                 | https://www.pengutronix.de/ |