Date:   Mon, 8 Jun 2020 12:36:48 +0000
From:   Sandipan Patra <spatra@...dia.com>
To:     Uwe Kleine-König 
        <u.kleine-koenig@...gutronix.de>
CC:     Thierry Reding <treding@...dia.com>,
        Jonathan Hunter <jonathanh@...dia.com>,
        Bibek Basu <bbasu@...dia.com>,
        Laxman Dewangan <ldewangan@...dia.com>,
        Krishna Yarlagadda <kyarlagadda@...dia.com>,
        "linux-pwm@...r.kernel.org" <linux-pwm@...r.kernel.org>,
        "linux-tegra@...r.kernel.org" <linux-tegra@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: RE: [PATCH V4] pwm: tegra: dynamic clk freq configuration by PWM
 driver

Hi Uwe,


> -----Original Message-----
> From: Uwe Kleine-König <u.kleine-koenig@...gutronix.de>
> Sent: Wednesday, June 3, 2020 9:59 PM
> To: Sandipan Patra <spatra@...dia.com>
> Cc: Thierry Reding <treding@...dia.com>; Jonathan Hunter
> <jonathanh@...dia.com>; Bibek Basu <bbasu@...dia.com>; Laxman Dewangan
> <ldewangan@...dia.com>; Krishna Yarlagadda <kyarlagadda@...dia.com>;
> linux-pwm@...r.kernel.org; linux-tegra@...r.kernel.org; linux-
> kernel@...r.kernel.org
> Subject: Re: [PATCH V4] pwm: tegra: dynamic clk freq configuration by PWM
> driver
> 
> Hello,
> 
> On Mon, Jun 01, 2020 at 10:50:36AM +0530, Sandipan Patra wrote:
> > diff --git a/drivers/pwm/pwm-tegra.c b/drivers/pwm/pwm-tegra.c index
> > d26ed8f..1daf591 100644
> > --- a/drivers/pwm/pwm-tegra.c
> > +++ b/drivers/pwm/pwm-tegra.c
> > @@ -4,8 +4,36 @@
> >   *
> >   * Tegra pulse-width-modulation controller driver
> >   *
> > - * Copyright (c) 2010, NVIDIA Corporation.
> > + * Copyright (c) 2010-2020, NVIDIA Corporation.
> >   * Based on arch/arm/plat-mxc/pwm.c by Sascha Hauer
> > <s.hauer@...gutronix.de>
> > + *
> > + * Overview of Tegra Pulse Width Modulator Register:
> > + * 1. 13-bit: Frequency division (SCALE)
> > + * 2. 8-bit : Pulse division (DUTY)
> > + * 3. 1-bit : Enable bit
> > + *
> > + * The PWM clock frequency is divided by 256 before subdividing it
> > + based
> > + * on the programmable frequency division value to generate the
> > + required
> > + * frequency for PWM output. The maximum output frequency that can be
> > + * achieved is (max rate of source clock) / 256.
> > + * e.g. if source clock rate is 408 MHz, maximum output frequency can be:
> > + * 408 MHz/256 = 1.6 MHz.
> > + * This 1.6 MHz frequency can further be divided using SCALE value in PWM.
> > + *
> > + * PWM pulse width: 8 bits are usable [23:16] for varying pulse width.
> > + * To achieve 100% duty cycle, program Bit [24] of this register to
> > + * 1’b1. In which case the other bits [23:16] are set to don't care.
> > + *
> > + * Limitations:
> > + * -	When PWM is disabled, the output is driven to inactive.
> > + * -	It does not allow the current PWM period to complete and
> > + *	stops abruptly.
> > + *
> 
> I'd prefer to have no empty lines in the in Limitations paragraph to be able to get
> all infos using something like:
> 
> 	sed -rn '/\* Limitations:/,/^ \*\/?$/p' drivers/pwm/pwm-tegra.c
> 

I'll push a change for this.

> > + * -	If the register is reconfigured while PWM is running,
> > + *	it does not complete the currently running period.
> > + *
> > + * -	If the user input duty is beyond acceptible limits,
> > + *	-EINVAL is returned.
> 
> s/acceptible/acceptable/ (but in fact this isn't a limitation, so I'd drop this here,
> as pointed out in v2).
> 
> In v2 I mentioned a few things to add here.

Ok. Understood. I will make sure to follow this in new change.

Instead of this
- If the user input duty is beyond acceptible limits,
	-EINVAL is returned.
I will add below info now and follow up on the atomic API implementation eventually.
- The driver doesn't implement the right rounding rules
- The driver needs updating to the atomic API

> 
> >   */
> >
> >  #include <linux/clk.h>
> > @@ -41,6 +69,7 @@ struct tegra_pwm_chip {
> >  	struct reset_control*rst;
> >
> >  	unsigned long clk_rate;
> > +	unsigned long min_period_ns;
> >
> >  	void __iomem *regs;
> >
> > @@ -68,7 +97,7 @@ static int tegra_pwm_config(struct pwm_chip *chip,
> > struct pwm_device *pwm,  {
> >  	struct tegra_pwm_chip *pc = to_tegra_pwm_chip(chip);
> >  	unsigned long long c = duty_ns, hz;
> > -	unsigned long rate;
> > +	unsigned long rate, required_clk_rate;
> 
> In v2 I requested to move this into the if block below. You replied to want to
> move it accordingly.

This got skipped. My mistake. I will push a change to move the definition of
required_clk_rate inside if-block as a new change.

> 
> >  	u32 val = 0;
> >  	int err;
> >
> > @@ -83,9 +112,47 @@ static int tegra_pwm_config(struct pwm_chip *chip,
> struct pwm_device *pwm,
> >  	val = (u32)c << PWM_DUTY_SHIFT;
> >
> >  	/*
> > +	 *  min period = max clock limit >> PWM_DUTY_WIDTH
> > +	 */
> > +	if (period_ns < pc->min_period_ns)
> > +		return -EINVAL;
> > +
> > +	/*
> >  	 * Compute the prescaler value for which (1 << PWM_DUTY_WIDTH)
> >  	 * cycles at the PWM clock rate will take period_ns nanoseconds.
> > +	 *
> > +	 * num_channels: If single instance of PWM controller has multiple
> > +	 * channels (e.g. Tegra210 or older) then it is not possible to
> > +	 * configure separate clock rates to each of the channels, in such
> > +	 * case the value stored during probe will be referred.
> > +	 *
> > +	 * If every PWM controller instance has one channel respectively, i.e.
> > +	 * nums_channels == 1 then only the clock rate can be modified
> > +	 * dynamically (e.g. Tegra186 or Tegra194).
> >  	 */
> > +	if (pc->soc->num_channels == 1) {
> > +		/*
> > +		 * Rate is multiplied with 2^PWM_DUTY_WIDTH so that it
> matches
> > +		 * with the maximum possible rate that the controller can
> > +		 * provide. Any further lower value can be derived by setting
> > +		 * PFM bits[0:12].
> 
> It looks a bit strange that the algorithm to calculate the clock settings depends
> on the number of channels. Looks like a wrong abstraction.

The calculation is added based on number of channels to support legacy HW.
When the PWM controller has multiple channels and all of them are sourced by
single clock, then dynamic clock configuration can not be supported. So the changes are
allowed only when PWM controller has one channel.
 
> 
> > +		 *
> > +		 * required_clk_rate is a reference rate for source clock and
> > +		 * it is derived based on user requested period. By setting the
> > +		 * source clock rate as required_clk_rate, PWM controller will
> > +		 * be able to configure the requested period.
> > +		 */
> > +		required_clk_rate =
> > +			(NSEC_PER_SEC / period_ns) << PWM_DUTY_WIDTH;
> > +
> > +		err = clk_set_rate(pc->clk, required_clk_rate);
> > +		if (err < 0)
> > +			return -EINVAL;
> > +
> > +		/* Store the new rate for further references */
> > +		pc->clk_rate = clk_get_rate(pc->clk);
> > +	}
> > +
> >  	rate = pc->clk_rate >> PWM_DUTY_WIDTH;
> >
> >  	/* Consider precision in PWM_SCALE_WIDTH rate calculation */ @@
> > -94,7 +161,7 @@ static int tegra_pwm_config(struct pwm_chip *chip,
> > struct pwm_device *pwm,
> >
> >  	/*
> >  	 * Since the actual PWM divider is the register's frequency divider
> > -	 * field minus 1, we need to decrement to get the correct value to
> > +	 * field plus 1, we need to decrement to get the correct value to
> 
> I would have put this in a separate change.
> 
> >  	 * write to the register.
> >  	 */
> >  	if (rate > 0)
> > @@ -205,6 +272,10 @@ static int tegra_pwm_probe(struct platform_device
> *pdev)
> >  	 */
> >  	pwm->clk_rate = clk_get_rate(pwm->clk);
> >
> > +	/* Set minimum limit of PWM period for the IP */
> > +	pwm->min_period_ns =
> > +	    (NSEC_PER_SEC / (pwm->soc->max_frequency >>
> PWM_DUTY_WIDTH)) +
> > +1;
> 
> To ensure that required_clk_rate in tegra_pwm_config doesn't get bigger than
> pwm->soc->max_frequency this isn't the right formula I think. I'd use
> 
> 	pwm->min_period_ns = DIV_ROUNDUP(NSEC_PER_SEC, pwm->soc-
> >max_frequency >> PWM_DUTY_WIDTH);
> 
> . Can you confirm?

As per formula, "+1" is required in the calculation.
Hence to reflect this in the code, manual division is done which by default
ignores the fractional part of the integer division. And then +1 is added.

Your suggestion to use DIV_ROUNDUP is for a crisp calculation.
But it mandates to ignore "+1" part as the division calculation is taken care by considering celling value.
When the division yields quotient with 0 fraction, then +1 is not considered.

In this patch, raw calculation is kept for a better readability, which can also help understanding
about the accurate derivation of min_period_ns as per specification.

Please help me understanding if I am not aligned with your suggestion and if is there a better
case where DIV_ROUNDUP is required in this case.

Thanks & Regards,
Sandipan

> 
> Best regards
> Uwe

Powered by blists - more mailing lists