Re: [RESEND 09/11] pwm: sti: Add PWM Capture call-back

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 4131 bytes --]

On Wed, Apr 13, 2016 at 11:25:54AM +0100, Lee Jones wrote:
> On Tue, 12 Apr 2016, Thierry Reding wrote:
> 
> > On Wed, Mar 02, 2016 at 03:32:07PM +0000, Lee Jones wrote:
> > > Once a PWM Capture has been initiated, the capture call
> > > enables a rising edge detection IRQ, then waits.  Once each
> > > of the 3 phase changes have been recorded the thread then
> > > wakes.  The remaining part of the call carries out the
> > > relevant calculations and passes back a formatted string to
> > > the caller.
> > > 
> > > Signed-off-by: Lee Jones <lee.jones@linaro.org>
> > > ---
> > >  drivers/pwm/pwm-sti.c | 72 +++++++++++++++++++++++++++++++++++++++++++++++++++
> > >  1 file changed, 72 insertions(+)
> > > 
> > > diff --git a/drivers/pwm/pwm-sti.c b/drivers/pwm/pwm-sti.c
> > > index 82a69e4..8de9b4a 100644
> > > --- a/drivers/pwm/pwm-sti.c
> > > +++ b/drivers/pwm/pwm-sti.c
> > > @@ -309,7 +309,79 @@ static void sti_pwm_free(struct pwm_chip *chip, struct pwm_device *pwm)
> > >  	clear_bit(pwm->hwpwm, &pc->configured);
> > >  }
> > >  
> > > +static int sti_pwm_capture(struct pwm_chip *chip, struct pwm_device *pwm,
> > > +			   int channel, char *buf)
> > > +{
> > > +	struct sti_pwm_chip *pc = to_sti_pwmchip(chip);
> > > +	struct sti_pwm_compat_data *cdata = pc->cdata;
> > > +	struct sti_cpt_data *d = pc->cpt_data[channel];
> > > +	struct device *dev = pc->dev;
> > > +	unsigned int f, dc;
> > > +	unsigned int high, low;
> > > +	bool level;
> > > +	int ret;
> > > +
> > > +	if (channel > cdata->cpt_num_chan - 1) {
> > > +		dev_err(dev, "Channel %d is not valid\n", channel);
> > > +		return -EINVAL;
> > > +	}
> > > +
> > > +	mutex_lock(&d->lock);
> > 
> > Should this perhaps reuse the struct pwm_device's ->lock?

That was actually a stupid suggestion by me, because that lock is being
removed in an unrelated patch series.

> > > +	/* Prepare capture measurement */
> > > +	d->index = 0;
> > > +	regmap_write(pc->regmap, PWM_CPT_EDGE(channel), CPT_EDGE_RISING);
> > > +	regmap_field_write(pc->pwm_cpt_int_en, BIT(channel));
> > > +	ret = wait_event_interruptible_timeout(d->wait, d->index > 1, HZ);
> > 
> > The timeout here should make sure callers don't hang forever. But maybe
> > you can still make sure that when the PWM gets disabled the wait queue
> > is woken and perhaps return an appropriate error code to let users know
> > that the operation was interrupted.
> 
> Sure.  I'll look into that.
> 
> > Also, how about letting callers choose the value of the timeout? In some
> > cases they may be interested in long-running signals. In other cases the
> > whole second timeout may be much too long.
> 
> I'm not opposed to it.  How do you suggest we do that?

The easiest would probably be to add an unsigned long timeout parameter
to the pwm_capture() function and ->capture() callbacks.

But thinking about this further I'm wondering if it might not be easier
and more flexible to move the timeout completely outside of this code
and into callers. I suspect that the most simple way to do that would be
to add a completion to struct pwm_capture that callers can use to wait
for completion of a capture. This would make the whole process
asynchronous and allow interesting things like making the sysfs capture
file pollable, for example.

> > > +	/*
> > > +	 * In case we woke up for another reason than completion
> > > +	 * make sure to disable the capture.
> > > +	 */
> > > +	regmap_write(pc->regmap, PWM_CPT_EDGE(channel), CPT_EDGE_DISABLED);
> > 
> > The comment here is slightly confusing because it implies that disabling
> > the capture should be done conditionally, whereas it is always disabled.
> 
> Not really.  We do it unconditionally for reason explained.
> 
> It says:
> 
>   "disable the capture just in case X happens"
> 
> rather than
> 
>   "disable the capture if X happens".
> 
> Perhaps the language is too subtle.  I can reword for clarity.

I'd be okay with just dropping the comment altogether, it seems rather
obvious to me. But clarifying is okay with me, too.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]