Message-ID: <20191105072922.rku2of3cfphpfirq@pengutronix.de>
Date:   Tue, 5 Nov 2019 08:29:22 +0100
From:   Uwe Kleine-König 
        <u.kleine-koenig@...gutronix.de>
To:     Clément Péron <peron.clem@...il.com>
Cc:     Thierry Reding <thierry.reding@...il.com>,
        Rob Herring <robh+dt@...nel.org>,
        Mark Rutland <mark.rutland@....com>,
        Maxime Ripard <mripard@...nel.org>,
        Chen-Yu Tsai <wens@...e.org>, linux-pwm@...r.kernel.org,
        devicetree <devicetree@...r.kernel.org>,
        linux-arm-kernel <linux-arm-kernel@...ts.infradead.org>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        Jernej Skrabec <jernej.skrabec@...l.net>
Subject: Re: [PATCH v2 4/7] pwm: sun4i: Add support to output source clock
 directly

Hi Clément,

On Mon, Nov 04, 2019 at 10:28:54PM +0100, Clément Péron wrote:
> On Mon, 4 Nov 2019 at 09:38, Uwe Kleine-König
> <u.kleine-koenig@...gutronix.de> wrote:
> > On Sun, Nov 03, 2019 at 09:33:31PM +0100, Clément Péron wrote:
> > > From: Jernej Skrabec <jernej.skrabec@...l.net>
> > >
> > > PWM core has an option to bypass whole logic and output unchanged source
> > > clock as PWM output. This is achieved by enabling bypass bit.
> > >
> > > Note that when bypass is enabled, no other setting has any meaning, not
> > > even enable bit.
> > >
> > > This mode of operation is needed to achieve high enough frequency to
> > > serve as clock source for AC200 chip, which is integrated into same
> > > package as H6 SoC.
> >
> > I think the , should be dropped.
> >
> > > Signed-off-by: Jernej Skrabec <jernej.skrabec@...l.net>
> > > Signed-off-by: Clément Péron <peron.clem@...il.com>
> > > ---
> > >  drivers/pwm/pwm-sun4i.c | 39 ++++++++++++++++++++++++++++++++++++++-
> > >  1 file changed, 38 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/pwm/pwm-sun4i.c b/drivers/pwm/pwm-sun4i.c
> > > index b5e7ac364f59..2441574674d9 100644
> > > --- a/drivers/pwm/pwm-sun4i.c
> > > +++ b/drivers/pwm/pwm-sun4i.c
> > > @@ -3,6 +3,10 @@
> > >   * Driver for Allwinner sun4i Pulse Width Modulation Controller
> > >   *
> > >   * Copyright (C) 2014 Alexandre Belloni <alexandre.belloni@...e-electrons.com>
> > > + *
> > > + * Limitations:
> > > + * - When outputing the source clock directly, the PWM logic will be bypassed
> > > + *   and the currently running period is not guaranted to be completed
> >
> > Typo: guaranted  -> guaranteed
> >
> > >   */
> > >
> > >  #include <linux/bitops.h>
> > > @@ -73,6 +77,7 @@ static const u32 prescaler_table[] = {
> > >
> > >  struct sun4i_pwm_data {
> > >       bool has_prescaler_bypass;
> > > +     bool has_direct_mod_clk_output;
> > >       unsigned int npwm;
> > >  };
> > >
> > > @@ -118,6 +123,20 @@ static void sun4i_pwm_get_state(struct pwm_chip *chip,
> > >
> > >       val = sun4i_pwm_readl(sun4i_pwm, PWM_CTRL_REG);
> > >
> > > +     /*
> > > +      * PWM chapter in H6 manual has a diagram which explains that if bypass
> > > +      * bit is set, no other setting has any meaning. Even more, experiment
> > > +      * proved that also enable bit is ignored in this case.
> > > +      */
> > > +     if ((val & BIT_CH(PWM_BYPASS, pwm->hwpwm)) &&
> > > +         data->has_direct_mod_clk_output) {
> > > +             state->period = DIV_ROUND_CLOSEST_ULL(NSEC_PER_SEC, clk_rate);
> > > +             state->duty_cycle = state->period / 2;
> > > +             state->polarity = PWM_POLARITY_NORMAL;
> > > +             state->enabled = true;
> > > +             return;
> > > +     }
> >
> > Not sure how the rest of sun4i_pwm_get_state behaves, but I would prefer
> > to let .get_state() round up which together with .apply_state() rounding
> > down yields sound behaviour.
> Ok
> >
> > > +
> > >       if ((PWM_REG_PRESCAL(val, pwm->hwpwm) == PWM_PRESCAL_MASK) &&
> > >           sun4i_pwm->data->has_prescaler_bypass)
> > >               prescaler = 1;
> > > @@ -203,7 +222,8 @@ static int sun4i_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> > >  {
> > >       struct sun4i_pwm_chip *sun4i_pwm = to_sun4i_pwm_chip(chip);
> > >       struct pwm_state cstate;
> > > -     u32 ctrl;
> > > +     u32 ctrl, clk_rate;
> > > +     bool bypass;
> > >       int ret;
> > >       unsigned int delay_us;
> > >       unsigned long now;
> > > @@ -218,6 +238,16 @@ static int sun4i_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> > >               }
> > >       }
> > >
> > > +     /*
> > > +      * Although it would make much more sense to check for bypass in
> > > +      * sun4i_pwm_calculate(), value of bypass bit also depends on "enabled".
> > > +      * Period is allowed to be rounded up or down.
> > > +      */
> > > +     clk_rate = clk_get_rate(sun4i_pwm->clk);
> > > +     bypass = ((state->period * clk_rate >= NSEC_PER_SEC &&
> > > +                state->period * clk_rate < NSEC_PER_SEC + clk_rate) &&
> > > +               state->enabled);
> >
> > I guess the compiler is smart enough here, but checking for
> > state->enabled is cheaper than the other checks, so putting this at the
> > start of the expression seems sensible.
> >
> > The comment doesn't match the code. You don't round up state->period.
> > (This is good, please fix the comment.) I think dropping the check
> >
> >         state->period * clk_rate < NSEC_PER_SEC + clk_rate
> >
> > would be fine, too.
> Ok
> 
> >
> > I'd like to have a check for
> >
> >         state->duty_cycle * clk_rate >= NSEC_PER_SEC / 2 &&
> >         state->duty_cycle * clk_rate < NSEC_PER_SEC
> >
> > here. If this isn't true rather disable the PWM or output a 100% duty
> > cycle with a larger period.
> 
> Why not just having the duty_cycle is 50% only ?
> state->duty_cycle * 2 == state->period;

Yeah, for the bypass case you can only provide a 50% duty cycle. The
problem you have to address is that you cannot rely on your consumer to
request only 50% duty cycles. So you have to implement some behaviour if
your consumer requests period = 1 / clk_rate and 20% duty cycle.

Where I want to get the pwm framework as a whole is to let lowlevel
drivers round down both duty_cycle and period to the next possible values
in their .apply callback to be able to provide a more uniform behaviour
for consumers. So here this would mean:

 - 1 / clk_rate <= state->period < smallest value without bypass &&
   0 <= state->duty_cycle < state->period / 2
   	=> provide a constant 0

 - 1 / clk_rate <= state->period < smallest value without bypass &&
   state->period / 2 <= state->duty_cycle < state->period
   	=> use bypass mode providing 50% duty cycle

 - 1 / clk_rate <= state->period < smallest value without bypass &&
   state->period == state->duty_cycle
   	=> provide a constant 1

Best regards
Uwe

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

Powered by blists - more mailing lists