Re: [PATCH v8 12/26] pwm: jz4740: Allow selection of PWM channels 0 and 1

[Paul Cercueil]

Hi,

Le jeu. 13 déc. 2018 à 21:32, Uwe Kleine-König 
<u.kleine-koenig@xxxxxxxxxxxxxx> a écrit :
> On Thu, Dec 13, 2018 at 02:58:31PM +0100, Paul Cercueil wrote:
> >  Hi,
> >
> >  Le jeu. 13 déc. 2018 à 10:18, Uwe Kleine-König
> >  <u.kleine-koenig@xxxxxxxxxxxxxx> a écrit :
> >  > On Wed, Dec 12, 2018 at 11:09:07PM +0100, Paul Cercueil wrote:
> >  > >  The TCU channels 0 and 1 were previously reserved for system 
> > tasks,
> >  > > and
> >  > >  thus unavailable for PWM.
> >  > >
> >  > >  The driver will now only allow a PWM channel to be requested if
> >  > > memory
> >  > >  resources corresponding to the register area of the channel 
> > were
> >  > >  supplied to the driver. This allows the TCU channels to be 
> > reserved
> >  > > for
> >  > >  system tasks from within the devicetree.
> >  > >
> >  > >  Signed-off-by: Paul Cercueil <paul@xxxxxxxxxxxxxxx>
> >  >
> >  > While there is someone caring for this driver I'd like to 
> > complete (a
> >  > bit) my picture about the different capabilities and specialities 
> > of the
> >  > supported PWMs. So I have a few questions:
> >  >
> >  > Is there a publicly available reference manual for this device? 
> > (If
> >  > yes, adding a link to the driver would be great.)
> >
> >  I have them here: https://zcrc.me/~paul/jz_docs/
>
> Is this link good enough to add it to the driver? From a quick view 
> I'd
> say this is another pwm implementation that gets active on 
> pwm_disable().
> Can you confirm this?

It's my website, so if these get moved, I can update the link.

What do you mean it gets active on pwm_disable()? If pwm_disable() gets 
called
the PWM line goes back to inactive state, which is what it should do.

> >  > jz4740_pwm_config looks as if the currently running period isn't
> >  > completed before the new config is in effect. Is that correct? If 
> > yes,
> >  > can this be fixed? A similar question for set_polarity: Does 
> > setting the
> >  > JZ_TIMER_CTRL_PWM_ACTIVE_LOW bit in the control register take 
> > effect
> >  > immediately or is this shadowed until the next period starts?
> >
> >  I don't really know. We only use this driver for a rumble motor and
> >  backlight.
> >  Somebody would have to check with a logic analyzer.
>
> depending on the possible timings you might also be able to test this
> e.g. by setting:
>
> 	duty_cycle=1ms, period=5s
>
> and then
>
> 	duty_cycle=5s, period=5s
>
> . If the implementation is right your display should be dark for 
> nearly
> 5 seconds. (And the second call to pwm_apply should also block until 
> the
> display is on.)

So it switches to full ON as soon as I set the duty cycle to 5s. Same 
for
the polarity, it is updated as soon as the register is written. So the
registers are not shadowed.

> >  > How does the device's output behave after the PWM is disabled?
> >  > Does it complete the currently running period? How does the output
> >  > behave then? (active/inactive/high/low/high-z?)
> >
> >  There's a bit to toggle between "graceful" shutdown (bit clear) and 
> > "abrupt"
> >  shutdown (bit set). TCSR bit 9. I think that graceful shutdown will 
> > complete
> >  the running period, then keep the level active. Abrupt shutdown 
> > will keep
> >  the
> >  current level of the line.
>
> Can you confirm the things you think above? I'd like to have them
> eventually documented in the driver.

From what I can see, with "abrupt" shutdown the line will always 
return to
its inactive state (be it low or high, depending on the polarity). 
Setting
this bit to "graceful" shutdown, the only difference is that the 
hardware
will keep its current state, active or inactive. That's why we use the
"abrupt" shutdown in the PWM driver.

> >  > >  @@ -42,11 +68,7 @@ static int jz4740_pwm_request(struct 
> > pwm_chip
> >  > > *chip, struct pwm_device *pwm)
> >  > >   	char clk_name[16];
> >  > >   	int ret;
> >  > >
> >  > >  -	/*
> >  > >  -	 * Timers 0 and 1 are used for system tasks, so they are
> >  > > unavailable
> >  > >  -	 * for use as PWMs.
> >  > >  -	 */
> >  > >  -	if (pwm->hwpwm < 2)
> >  > >  +	if (!jz4740_pwm_can_use_chn(jz, pwm->hwpwm))
> >  > >   		return -EBUSY;
> >  >
> >  > Maybe EBUSY isn't the best choice here. If the needed register 
> > space for
> >  > the requested pwm is not included in the memory resources 
> > provided to
> >  > the device I'd prefer ENXIO or ENODEV.
> >
> >  The idea was that if we don't get the register space we need, that 
> > means
> >  the channel is used for something else, hence the EBUSY. Should I 
> > switch
> >  it to ENXIO?
>
> I understand your reasoning, but I think it's misleading. If I get
> -EBUSY from a PWM driver I'd start searching for another user of said
> resource. With ENXIO or ENODEV it's more obvious that the driver isn't
> responsible for the resource that was requested.

OK.

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