Re: [PATCH v2 1/5] dt-bindings: iio: at91-sama5d2_adc: add optional dma property

[Eugen Hristev <eugen.hristev@microchip.com>]

On 20.10.2017 00:58, Rob Herring wrote:
> On Thu, Oct 19, 2017 at 6:14 AM, Alexandre Belloni
> <alexandre.belloni@free-electrons.com> wrote:
>> On 13/10/2017 at 16:51:42 -0500, Rob Herring wrote:
>>> On Wed, Oct 11, 2017 at 09:35:28AM +0300, Eugen Hristev wrote:
>>>> Added property for DMA configuration of the device.
>>>>
>>>> Signed-off-by: Eugen Hristev <eugen.hristev@microchip.com>
>>>> ---
>>>>   Documentation/devicetree/bindings/iio/adc/at91-sama5d2_adc.txt | 7 +++++++
>>>>   1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/iio/adc/at91-sama5d2_adc.txt b/Documentation/devicetree/bindings/iio/adc/at91-sama5d2_adc.txt
>>>> index 552e7a8..5f94d479 100644
>>>> --- a/Documentation/devicetree/bindings/iio/adc/at91-sama5d2_adc.txt
>>>> +++ b/Documentation/devicetree/bindings/iio/adc/at91-sama5d2_adc.txt
>>>> @@ -17,6 +17,11 @@ Required properties:
>>>>     This property uses the IRQ edge types values: IRQ_TYPE_EDGE_RISING ,
>>>>     IRQ_TYPE_EDGE_FALLING or IRQ_TYPE_EDGE_BOTH
>>>>
>>>> +Optional properties:
>>>> +  - dmas: Phandle to dma channel for the ADC.
>>>> +  See ../../dma/dma.txt for details.
>>>> +  - dma-names: Must be "rx" when dmas property is being used.
>>>
>>> -names is pointless when there is only one.
>>>
>>
>> You didn't reply to the question I had previously about that: What if at
>> some point, we have multiple dmas in the same binding?
> 
> Then add dma-names at that point and rx has to be first. If you know
> there's other channels, then add them now. Don't evolve the bindings
> needlessly based on what a driver supports.
> 
> Would another channel make sense here? Maybe multi-channel rx in which
> case your naming wouldn't be setup for that. But "tx" on an ADC?
> 
> Rob

Hello Rob,

I can keep only "dmas" and remove "dma-names", but then, the API used by
the drivers that request channels requires a name parameter
(dma_request_slave_channel), and it will look always inside the
"dma-names" property to match the name and find the proper channel, and
it will fail in this case.

Looking in the general dma binding (dma.txt) I can see that both
properties are marked as required for a client driver, so that's why
I added both.

I can either keep dma-names; or try to find the channel properties by
looking directly into the OF node and create the dma_chan struct inside
my driver (maybe call some xlate function for the DMA controller), but
still need to get the DT property. A complicated option would be to
actually create an API inside the dmaengine to find a channel without
name.

Which approach you think it's best to pursue ?

Thanks,
Eugen



>