[resend rfc v3] pwm: add BCM2835 PWM driver
Bart Tanghe
bart.tanghe at thomasmore.be
Mon Sep 29 06:37:50 PDT 2014
On 2014-09-29 07:33, Thierry Reding wrote:
> On Fri, Sep 26, 2014 at 08:45:57AM -0600, Stephen Warren wrote:
>> On 09/26/2014 01:11 AM, Thierry Reding wrote:
>>> On Thu, Sep 04, 2014 at 09:06:48AM -0600, Stephen Warren wrote:
> [...]
>>>> Oh dear. It sounds like we need at least some form of clock driver for the
>>>> platform then. I still don't think there's complete documentation for the
>>>> HW, even though a lot of register docs were published which presumably cover
>>>> the clock HW? Equally, given that the VC firmware assumes it owns most of
>>>> the HW, it seems best to manipulate the clocks through the firmware
>>>> interface rather than directly touching the HW. Unfortunately, I don't
>>>> believe there's any ABI guarantee on the firmware interface. Perhaps we can
>>>> get one?
>>>
>>> Urgs... this VC firmware seems to be more of a headache that I thought
>>> it was. How is this handled in other drivers? Surely PWM isn't the first
>>> one that needs clocks?
>>
>> For the other clocks, I've set up dummy fixed-rate clocks in the DT and/or
>> "clock driver" code to satisfy references by phandle or clock name
>> respectively. Since the other drivers don't actually manipulate the clock
>> rates etc., this is enough for the drivers.
>
> Given that this driver only queries the clock frequency adding a fixed-
> rate clock to the device tree should work as well. Then the calls to
> clk_prepare_enable() and clk_disable_unprepare() can still be added as
> appropriate so that the driver doesn't need to change if a proper clock
> driver ever gets added.
>
> Or am I missing anything? Perhaps the issue is that the default clock
> rate for the PWM clock isn't usable? That would still not prevent the
> driver from being merged.
>
> Thierry
>
The pwm clock is default unusable. To let the pwm clock run, It's necessary to change some
clock registers. I've added the clk_prepare_enable and the clk_disable_unprepare functions. So the
driver is able to work with a proper clock driver in the future.
Bart
More information about the linux-rpi-kernel
mailing list