Re: [PATCH] gpio: pca953x: Add Maxim MAX7313 PWM support

[Bartosz Golaszewski <bgolaszewski@xxxxxxxxxxxx>]

pon., 4 lis 2019 o 16:11 Miquel Raynal <miquel.raynal@xxxxxxxxxxx> napisał(a):
>
> Hi Linus,
>
> Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote on Tue, 15 Oct 2019
> 17:55:33 +0300:
>
> > On Tue, Oct 15, 2019 at 5:30 PM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> > >
> > > Hi Andy,
> > >
> > > Thanks for the feedback.
> > >
> > > Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote on Mon, 14 Oct 2019
> > > 20:59:01 +0300:
> > >
> > > > On Mon, Oct 14, 2019 at 4:09 PM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> > > > >
> > > > > The MAX7313 chip is fully compatible with the PCA9535 on its basic
> > > > > functions but can also manage the intensity on each of its ports with
> > > > > PWM. Each output is independent and may be tuned with 16 values (4
> > > > > bits per output). The period is always 32kHz, only the duty-cycle may
> > > > > be changed. One can use any output as GPIO or PWM.
> > > >
> > > > Can we rather not contaminate driver with this?
> > > >
> > > > Just register a separate PWM driver and export its functionality to
> > > > GPIO, or other way around (in the latter case we actually have PCA8685
> > > > which provides a GPIO fgunctionality).
> > > >
> > >
> > > I understand your concern but I am not sure to understand which
> > > solution you have in mind. In the former case, could you explain a bit
> > > more what you are thinking about? Would linking the PWM support with
> > > the GPIO driver (code would be located in another .c file) work for
> > > you? Or maybe you would prefer an MFD on top of the GPIO driver?
> > >
> > > As for the later case, I am not willing to re-implement GPIO support in
> > > an alternate driver for an already supported chip, it is too much work
> > > for the time I can spend on it.
> >
> >
> > drivers/pwm/pwm-max7313.c:
> >
> > probe(platform_device)
> > {
> >   struct regmap = pdata;
> >   ...
> > }
> >
> > --- 8< --- 8< ---
> > drivers/gpio/gpio-pca953x.c:
> >
> > probe()
> > {
> >   struct regmap rm;
> > ...
> >   if (dev_has_pwm)
> >    pca953x_register_pwm_driver(rm);
> > ...
> > }
> >
> > In the above regmap may be replaced with some (shared) container.
> >
> > Or other way around. PWM registers GPIO (which actually I prefer since
> > we have PCA9685 case where PWM provides GPIO functionality, though via
> > different means)
> >
>
> Can I have your input on this proposal?
>
> On one hand I agree that the GPIO driver is already quite big due to
> its genericity and the amount of controllers it supports, on the other
> hand:
> 1/ Registering a PWM driver from the GPIO core seems strange. Maybe
> registering a platform device could do the trick but I am not convinced
> it is worth the trouble.
> 2/ Putting the PWM logic in the drivers/pwm/ directory is not very
> convenient as the object file will have to be embedded within the GPIO
> one; this line in drivers/gpio/Makefile would be horrible:
> ... += gpio-pca953x.o ../pwm/pwm-max7313.o (not even sure it works)
> 3/ In any cases, the regmap's ->readable_reg(), ->writable_reg()
> callbacks shall be tweaked to turn the PWM registers accessible, so we
> would still have PWM related code in the PCA953x GPIO driver.
>
> In the end, I wonder if keeping everything in one file is not better?
> Eventually I can create a separate file and fill it with just the PWM
> helpers/hooks. Please advise on the better solution for you, I'll wait
> your feedback before addressing Thierry Reding's other review and
> resubmit.
>

I'm not sure if this has been discussed, but is it possible to create
an MFD driver for this chip and conditionally plug in the GPIO part
from pca953x? I don't like the idea of having PWM functionality in a
GPIO driver either.

Bart

>
> Thanks,
> Miquèl