Re: PWM class? (was: Re: MPC52xx simple GPIO support)

[Anton Vorontsov <avorontsov@ru.mvista.com>]

On Thu, Jun 11, 2009 at 04:00:51PM -0600, Grant Likely wrote:
> On Wed, Jun 3, 2009 at 7:22 AM, Anton Vorontsov<avorontsov@ru.mvista.com> wrote:
> > On Wed, Jun 03, 2009 at 02:42:26PM +0200, Stefan Strobl wrote:
> > [...]
> >> The led class provides support for setting the brightness, which
> >> obviously the gpio driver doesn't support. The hardware (mpc52xx_gpt)
> >> would support it in PWM mode though. I'm now wandering how this could be
> >> best implemented.
> >>
> >> 1) - Create some PWM class similar to the GPIO class
> >>    - Add support for PWM mode in mpc52xx_gpt.c that uses that PWM class
> >>    - And add an interface for the LED to use the PWM class
> >>
> >> 2) - Create an LED driver that accesses the mpc52xx_gpt directly.
> >>
> >> I think I would be overwhelmed trying to implement (1) but am confident
> >> to do (2). What do you think is the right approach?
> >
> > I'd suggest creating a generic PWM class, i.e. PWMLIB, alike to
> > GPIOLIB. (2) can be an acceptable approach for now, but for the
> > long-term solution (1) is the way to go.
> >
> > The non-lib PWM API is already there, see include/linux/pwm.h,
> > and arch/arm/mach-pxa/pwm.c as an implementation example.
> >
> > Note that PXA implementation is SOC-specific, which is not very
> > good.
> >
> > So I'd suggest creating drivers/pwm/pwmlib.c, borrowing
> > ideas from gpiolib. And then we can reuse drivers/leds/leds-pwm.c
> > driver (of course, after adding appropriate OF code into it).
> 
> Ugh.  The referenced pwm api is about as trivial as it gets; it is an
> anonymous context pointer (anonymous struct pwm_device *) with a set
> of accessor functions.  PWMs are also not nearly as common as GPIO
> pins, and I am not interested in gpiolib being duplicated for PWMs, at
> least not until there are more that just two examples of use to draw
> from.

I didn't say that we should duplicate gpiolib. I said that we
might borrow some ideas. ;-)

> If anything, I'd rather struct pwm_device be non-anonymous and contain
> a set of ops which call directly into the driver.  That way is at
> least multiplatform friendly.  I don't think the gpio API is the
> example to follow here.  But even then I think it is premature to try
> and define a PWM api.  Personally, I'd modify mpc52xx_gpt to export
> its own PWM interface for the time being using the existing GPIO
> infrastructure to find the appropriate pin.

Jon Smirl found that there were already some efforts put
into making generic PWM class:

http://thread.gmane.org/gmane.linux.kernel.embedded/1160

Briefly looking, the class is pretty cool. Somebody should just
continue this work.

> If you do decide to do a generic PWM api, then I think the way to go
> is to build it as an extension to gpiolib.

I don't think David Brownell will like this idea. But
who knows... maybe.

Thanks,

-- 
Anton Vorontsov
email: cbouatmailru@gmail.com
irc://irc.freenode.net/bd2