Re: [RFC PATCH 06/10] pwm: ntxec: Add driver for PWM function in Netronix EC

[Jonathan NeuschÃfer]

On Mon, Jun 22, 2020 at 10:18:02AM +0200, Uwe Kleine-KÃnig wrote:
> Hello,
> 
> On Sun, Jun 21, 2020 at 12:42:17AM +0200, Jonathan NeuschÃfer wrote:
> > The Netronix EC provides a PWM output, which is used for the backlight
> 
> s/,//
> 
> > on ebook readers. This patches adds a driver for the PWM output.
> 
> on *some* ebook readers

Ok, I'll fix these.

> 
> > +#define NTXEC_UNK_A		0xa1
> > +#define NTXEC_UNK_B		0xa2
> > +#define NTXEC_ENABLE		0xa3
> > +#define NTXEC_PERIOD_LOW	0xa4
> > +#define NTXEC_PERIOD_HIGH	0xa5
> > +#define NTXEC_DUTY_LOW		0xa6
> > +#define NTXEC_DUTY_HIGH		0xa7
> > +
> > +/*
> > + * The time base used in the EC is 8MHz, or 125ns. Period and duty cycle are
> > + * measured in this unit.
> > + */
> > +static int ntxec_pwm_config(struct pwm_chip *chip, struct pwm_device *pwm_dev,
> > +				 int duty_ns, int period_ns)
> > +{
> > +	struct ntxec_pwm *pwm = pwmchip_to_pwm(chip);
> > +	uint64_t duty = duty_ns;
> > +	uint64_t period = period_ns;
> 
> As you cannot use values bigger than 8191999 anyhow, I wonder why you
> use a 64 bit type here.

No particular reason; I possibly got confused by the division API. I'll
use uint32_t instead.

> > +	int res = 0;
> > +
> > +	do_div(period, 125);
> 
> Please use a define instead of plain 125.

Will do.

> > +	if (period > 0xffff) {
> > +		dev_warn(pwm->dev,
> > +			 "Period is not representable in 16 bits: %llu\n", period);
> > +		return -ERANGE;
> > +	}
> > +
> > +	do_div(duty, 125);
> > +	if (duty > 0xffff) {
> > +		dev_warn(pwm->dev, "Duty cycle is not representable in 16 bits: %llu\n",
> > +			duty);
> > +		return -ERANGE;
> > +	}
> 
> This check isn't necessary as the pwm core ensures that duty <= period.

Ok, I'll remove it.
> 
> > +	res |= ntxec_write8(pwm->ec, NTXEC_PERIOD_HIGH, period >> 8);
> > +	res |= ntxec_write8(pwm->ec, NTXEC_PERIOD_LOW, period);
> > +	res |= ntxec_write8(pwm->ec, NTXEC_DUTY_HIGH, duty >> 8);
> > +	res |= ntxec_write8(pwm->ec, NTXEC_DUTY_LOW, duty);
> 
> Does this complete the currently running period? Can it happen that a
> new period starts between the first and the last write and so a mixed
> period can be seen at the output?

Good question. I haven't measured it, and also don't have the code
running on the EC.

> 
> > +
> > +	return (res < 0) ? -EIO : 0;
> > +}
> > +
> > +static int ntxec_pwm_enable(struct pwm_chip *chip,
> > +				 struct pwm_device *pwm_dev)
> > +{
> > +	struct ntxec_pwm *pwm = pwmchip_to_pwm(chip);
> > +
> > +	return ntxec_write8(pwm->ec, NTXEC_ENABLE, 1);
> > +}
> > +
> > +static void ntxec_pwm_disable(struct pwm_chip *chip,
> > +				   struct pwm_device *pwm_dev)
> > +{
> > +	struct ntxec_pwm *pwm = pwmchip_to_pwm(chip);
> > +
> > +	ntxec_write8(pwm->ec, NTXEC_ENABLE, 0);
> > +}
> > +
> > +static struct pwm_ops ntxec_pwm_ops = {
> > +	.config		= ntxec_pwm_config,
> > +	.enable		= ntxec_pwm_enable,
> > +	.disable	= ntxec_pwm_disable,
> > +	.owner		= THIS_MODULE,
> 
> Please don't align the =, just a single space before them is fine.

Ok

> More important: Please implement .apply() (and .get_state()) instead of
> the old API. Also please enable PWM_DEBUG which might save us a review
> iteration.

Will do!

> 
> > +};
> > +
> > +static int ntxec_pwm_probe(struct platform_device *pdev)
> > +{
> > +	struct ntxec *ec = dev_get_drvdata(pdev->dev.parent);
> > +	struct ntxec_pwm *pwm;
> > +	struct pwm_chip *chip;
> > +	int res;
> > +
> > +	pwm = devm_kzalloc(&pdev->dev, sizeof(*pwm), GFP_KERNEL);
> > +	if (!pwm)
> > +		return -ENOMEM;
> > +
> > +	pwm->ec = ec;
> > +	pwm->dev = &pdev->dev;
> > +
> > +	chip = &pwm->chip;
> > +	chip->dev = &pdev->dev;
> > +	chip->ops = &ntxec_pwm_ops;
> > +	chip->base = -1;
> > +	chip->npwm = 1;
> > +
> > +	res = pwmchip_add(chip);
> > +	if (res < 0)
> > +		return res;
> > +
> > +	platform_set_drvdata(pdev, pwm);
> > +
> > +	res |= ntxec_write8(pwm->ec, NTXEC_ENABLE, 0);
> > +	res |= ntxec_write8(pwm->ec, NTXEC_UNK_A, 0xff);
> > +	res |= ntxec_write8(pwm->ec, NTXEC_UNK_B, 0xff);
> > +
> > +	return (res < 0) ? -EIO : 0;
> 
> This is broken for several reasons:
> 
>  - You're not supposed to modify the output in .probe
>  - if ntxec_write8 results in an error you keep the pwm registered.
>  - From the moment on pwmchip_add returns the callbacks can be called.
>    The calls to ntxec_write8 probably interfere here.

Ok, I'll rework the probe function to avoid these issues.


Thanks for the review,
Jonathan NeuschÃfer

Attachment: signature.asc
Description: PGP signature