Re: [PATCH 1/2] dt-bindings: usb: tps6598x: Make the interrupts property optional

[Roger Quadros <rogerq@kernel.org>]

Hi,

On 22/04/2022 08:07, Aswath Govindraju wrote:
> Hi Roger,
> 
> On 21/04/22 00:46, Roger Quadros wrote:
>> Hi,
>>
>> On 18/04/2022 08:19, Aswath Govindraju wrote:
>>> Hi Roger,
>>>
>>> On 14/04/22 23:40, Roger Quadros wrote:
>>>> Hi,
>>>>
>>>> On 14/04/2022 11:31, Aswath Govindraju wrote:
>>>>> Support for polling has been added in the driver, which will be used by
>>>>> default if interrupts property is not populated. Therefore, remove
>>>>> interrupts and interrupt-names from the required properties and add a note
>>>>> under interrupts property describing the above support in driver.
>>>>>
>>>>> Suggested-by: Roger Quadros <rogerq@kernel.org>
>>>>
>>>> I did not suggest to make interrupts optional by default.
>>>>
>>>> What I suggested was that if a DT property exists to explicitly
>>>> indicate polling mode then interrupts are not required.
>>>>
>>>
>>> ohh okay, got it. However, may I know if adding a dt property to
>>> indicate polling for aiding the driver, is the correct approach to model it?
>>>
>>> In terms of modelling hardware, as interrupts are not connected we are
>>> not populating the interrupts property. Shouldn't that be all. If we are
>>> adding a property explicitly to indicate polling that can be used by
>>> driver, wouldn't that be a software aid being added in the device tree?
>>
>> The hardware (tps6598x chip) has an interrupt pin and is expected to be used
>> in normal case.
>>
>> Some buggy boards might have forgot to connect it. We are adding polling mode only for these buggy boards. ;)
>> So polling mode is an exception.
>>
> 
> Yes as you mentioned the interrupt line is expected to connected but
> there could be cases where there are not enough pins on the SoC and
> polling is used intentionally. In these cases this would be a feature
> rather than a bug.

I do not agree that this is a feature but a board defect. You can always use
a GPIO expander to add more GPIOs than the SoC can provide.

Type-C events are asynchronous and polling is a waste of CPU time.
What will you do if system suspends and you need to wake up on Type-C
status change?
So polling mode is just an exception for the defective boards or could
be used for debugging.

> 
> Also, I feel like not adding interrupts property in the dt nodes will
> indicate polling. My question is why are we adding an extra property
> (which is being used only as an aid in the driver) when this feature can
> be modeled by making interrupts property optional.

Because interrupt property was not originally optional for this driver.

I would like to hear what Heikki has to say about this.

Any thoughts Heikki?

cheers,
-roger

> 
> Thanks,
> Aswath
> 
>> cheers,
>> -roger
>>
>>>
>>> Thanks,
>>> Aswath
>>>
>>>>> Signed-off-by: Aswath Govindraju <a-govindraju@ti.com>
>>>>> ---
>>>>>  Documentation/devicetree/bindings/usb/ti,tps6598x.yaml | 4 ++--
>>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/usb/ti,tps6598x.yaml b/Documentation/devicetree/bindings/usb/ti,tps6598x.yaml
>>>>> index a4c53b1f1af3..1c4b8c6233e5 100644
>>>>> --- a/Documentation/devicetree/bindings/usb/ti,tps6598x.yaml
>>>>> +++ b/Documentation/devicetree/bindings/usb/ti,tps6598x.yaml
>>>>> @@ -25,6 +25,8 @@ properties:
>>>>>  
>>>>>    interrupts:
>>>>>      maxItems: 1
>>>>> +    description:
>>>>> +      If interrupts are not populated then by default polling will be used.
>>>>>  
>>>>>    interrupt-names:
>>>>>      items:
>>>>> @@ -33,8 +35,6 @@ properties:
>>>>>  required:
>>>>>    - compatible
>>>>>    - reg
>>>>> -  - interrupts
>>>>> -  - interrupt-names
>>>>>  
>>>>>  additionalProperties: true
>>>>>  
>>>>
>>>> cheers,
>>>> -roger
> 
>