On Tue, Dec 10, 2013 at 10:20:33AM +0800, Bo Shen wrote:
> Hi Thierry,
>
> On 12/03/2013 11:09 AM, Bo Shen wrote:
> >>>+ atmel_pwm->chip.of_xlate = of_pwm_xlate_with_flags;
> >>>+ atmel_pwm->chip.of_pwm_n_cells = 3;
> >>>+ atmel_pwm->chip.base = -1;
> >>>+ } else {
> >>>+ atmel_pwm->chip.base = pdev->id;
> >>
> >>That's not correct. The chip cannot be tied to pdev->id, because that ID
> >>is the instance number of the device. So typically you would have
> >>devices name like this:
> >>
> >> atmel-pwm.0
> >> atmel-pwm.1
> >> ...
> >>
> >>Now, if you have that, then you won't be able to register the second
> >>instance because the first instance will already have requested PWMs
> >>0-3, and setting .base to 1 will cause PWMs 1-4 to be requested, which
> >>intersects with the range of the first instance.
> >>
> >>The same applies of course if you have other PWM controllers in the
> >>system which have similar instance names.
> >>
> >>So the right thing to do here is to provide that number via platform
> >>data so that platform code can define it, knowing in advance all ranges
> >>for all other PWM controllers and thereby make sure there's no
> >>intersection.
> >
> >OK, I will fix this.
>
> After read deeply of PWM framework, for non device tree, I think we'd better
> let the PWM core to choose chip.base as device tree, while not pass a number
> through platform data to it. Or else, it will confuse the user to set the
> chip.base, must set it in correct value to avoid intersection. And, actually
> we won't use chip.base in driver itself.
Yes, that should work as well, if you make sure that every user actually
has the PWM lookup table and doesn't rely on a fixed global index to
retrieve the PWM channel.
Thierry
Attachment:
pgpWQU9hq8gNv.pgp
Description: PGP signature
