[PATCHv9 0/2] Add Allwinner SoCs PWM support

Olliver Schinagl oliver at schinagl.nl
Tue Nov 18 07:11:17 PST 2014

On 18-11-14 15:55, Maxime Ripard wrote:
> On Tue, Nov 18, 2014 at 03:11:26PM +0100, Olliver Schinagl wrote:
>> On 18-11-14 14:54, Maxime Ripard wrote:
>>> On Tue, Nov 18, 2014 at 02:47:33PM +0100, Olliver Schinagl wrote:
>>>> Hey Alexandre,
>>>> On 05-11-14 16:15, Alexandre Belloni wrote:
>>>>> Hi,
>>>>> This patch series adds support for the PWM controller found on the Allwinner
>>>>> SoCs.
>>>>> The first patch adds the driver itself.
>>>>> The second patch adds the DT binding documentation
>>>>> Changes in v8:
>>>>>   - renamed the driver sun4i as the PWM IP is different in the next sunxi SoCs
>>>> Why did you decide to rename it to sun4i? From your bindings document I
>>>> understand you still support sun4i and sun7i, what happened to sun5i?
>>>> I went over the datasheets of sun4i, sun5i and sun7i and the disp_lcd.c from
>>>> the latest linux-sunxi kernels and have to agree, they are all 4
>>>> inconsistent and messy, but I'm not sure what you mean with PWM IP is
>>>> different in next sunxi soc's. is sun5i different to sun4i? is sun7i
>>>> different? or is sun6i (A31), sun8i (a80) and sun9i (A23) different?
>>>> What I get from the datasheet is, that sun4i and sun5i are exactly the same,
>>>> with the exception that sun5i only has 1 PWM (~exposed~). I belive that is
>>>> easily solved with the bindings by having allwinner-sun4i and allwinner
>>>> sun5i bindings if I'm not mistaken.
>>>> As for sun7i compared to the other ones, according to disp_lcd.c sun5i and
>>>> sun7i should behave exactly the same. This is contradicting to the
>>>> datasheet, where sun4i and sun5i are the same.
>>>> So what are the major differences that I can see between the 3? sun4i
>>>> defines the PWM prescaler register value 0b1111 as being undefined, and
>>>> sun5i and sun7i as /1? Did you verify this (I haven't I admit, i bumped into
>>>> this while looking for your patch ;-) )? I wouldn't be supprised if it where
>>>> a typo on allwinners end in the datasheet ... disp_lcd.c stops at 72000 for
>>>> the last entry. We should just check sun4i, sun5i and sun7i hardware to see
>>>> if it behaves the same with a prescaler of 0b1111, which I would not be
>>>> totally surprised if it did.
>>>> The other difference I notice is that sun7i and sun5i use 16bit period
>>>> register where sun4i uses a 8bit register. This is probably the only reason
>>>> why they put a #ifdef in disp_lcd.c, calculations turn out differently. I
>>>> don't recognize this behavior at all in your driver however. I do think they
>>>> that there is a difference here, since they did split up the original driver
>>>> here because of this difference.
>>>> The pre-scaler bypass bit and pwm ready bit you all ready take care of :)
>>> A31 and later are using a different IP, that is not supported by this
>>> driver.
>> Ah, see that was not clear to me ;) Also a comment in the code would be
>> helpful how the 8bit vs 16bit period register is being handled with regards
>> to sun4i and sun7i. I can't seem to spot where this is taken into account,
>> since the disp_lcd.c code suggests there is supposedly some difference? I
>> can't tell from their code what is really going on, since they are using
>> unsigned 32 bit integers and crap that in the lower and upper half of the
>> 16bit register, so maybe the register has always been 16bit but the docs are
>> simply wrong? But then why differentiate between the two chip generations
>> ...
> Is this a bug report? Or just some theorical issues that we might
> never encounter?
No, I'm asking why it is nowhere documented or not directly clear from 
the code, that sun4i according to the datasheet uses 8bits for the 
period register and sun5i and sun7i use 16bits. Also, the only reason 
why the allwinner code differentiates between sun4i and sun5/7i is 
because of this register. It contradicts what we know at this point, so 
a comment indicating this would be nice.

I am currently building a kernel with these patches added for sun7i and 
see if it works; i'll try to find some sun4i hardware and test it there 
as well and report back then.

>>> This driver only support the controller introduced with the A10, that
>>> only saw minor differences between SoCs, like you have shown.
>>>> Finally, though I'm sure I should have replied to it inline in the actual
>>>> code, but hoping i can let it slip through here is that you define your
>>>> local data as:
>>>> + static const struct sun4i_pwm_data sun4i_pwm_data_a20 = {
>>>> which looks really strange to me, since there is no a20 using the sun4i
>>>> architecture :) I know it's just nitpicking, it just looks really odd. Maybe
>>>> the compatible naming works just as well? sun4i_pwm_data; sun7i_pwm_data
>>>> (and sun5i_pwm_data if you want to take care of only pwm0, only pwm1 (if we
>>>> ever encounter such a configuration, disabled pwm0, enabled pwm1) or both to
>>>> be used?)
>>> This driver is name sun4i_pwm, hence the prefix.
>> Ah right, unrelated, but I guess I should change my sunxi driver name to
>> sun4i as well then?
> What? Why?
> Maxime

More information about the linux-arm-kernel mailing list