Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 00:02, Dmitry Osipenko wrote:
> On 02.10.2017 20:05, Stephen Warren wrote:
>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>
>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>
>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@xxxxxxxxx>
>>>>>>>> ---
>>>>>>>> ÂÂ .../bindings/dma/nvidia,tegra20-ahbdma.txtÂÂÂÂÂÂÂÂ | 23
>>>>>>>> ++++++++++++++++++++++
>>>>>>>> ÂÂ 1 file changed, 23 insertions(+)
>>>>>>>> ÂÂ create mode 100644
>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>
>>>>>>>> diff --git
>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> new file mode 100644
>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>> +
>>>>>>>> +Required properties:
>>>>>>>> +- compatible:ÂÂÂ Must be "nvidia,tegra20-ahbdma"
>>>>>>>> +- reg:ÂÂÂÂÂÂÂ Should contain registers base address and length.
>>>>>>>> +- interrupts:ÂÂÂ Should contain one entry, DMA controller interrupt.
>>>>>>>> +- clocks:ÂÂÂ Should contain one entry, DMA controller clock.
>>>>>>>> +- resets :ÂÂÂ Should contain one entry, DMA controller reset.
>>>>>>>> +- #dma-cells:ÂÂÂ Should be <1>. The cell represents DMA request select
>>>>>>>> value
>>>>>>>> +ÂÂÂÂÂÂÂ for the peripheral. For more details consult the Tegra TRM's
>>>>>>>> +ÂÂÂÂÂÂÂ documentation, in particular AHB DMA channel control register
>>>>>>>> +ÂÂÂÂÂÂÂ REQ_SEL field.
>>>>>>>
>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>
>>>>>>
>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>> up to
>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>
>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>
>>>>>> And I think the same applies to requester... any objections?
>>>>>
>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>
>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>> this.
>>>>
>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>> when SW desires, rather than being a combination of after SW programs the
>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>> rather than some other separate higher-level property/driver/...
>>>
>>> Thank you for the comment. If we'd want to extend the binding further with the
>>> trigger, how to differentiate trigger from the requester in a case of a single
>>> #data-cell?
>>>
>>> Of course realistically a chance that the further extension would be needed is
>>> very-very low, so we may defer the efforts to solve that question and for now
>>> make driver aware of the potential #dma-cells extension.
>>
>> The request selector cell isn't optional, so is always present. If we later add
>> an optional trig_sel cell, we'll either have:
>>
>> #dma-cells=<1>: req_sel
>>
>> or:
>>
>> #dma-cells=<2>: req_sel, trig_sel
> 
> Why request sel. couldn't be optional? Could you please elaborate a bit more?
> 
> I think possible options are:
> 
> #dma-cells=<1>: req_sel
> #dma-cells=<1>: trig_sel

With the above, how would you know that it is the req_sel or trig_sel
that is specified?

> #dma-cells=<2>: req_sel, trig_sel
> 
> The only difference between request and trigger is that trigger issues the whole
> transfer, while request only a single burst. Isn't it possible to have a case in
> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
> REQ_SEL must be mandatory.

I think that what Stephen is proposing is that for now we go with
'#dma-cells=<1>' and if we ever need to support the trigger cell we
could add support for '#dma-cells=<2>'. So with this proposal the
'req_sel' would always be required for both '#dma-cells=<1>' and
'#dma-cells=<2>'. Even if the req_sel is not actually used but the
'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
value (eg. -1) where we know to ignore it.

Cheers
Jon

-- 
nvpublic