Re: [PATCH v2 1/2] dt-bindings: hwmon: Document adt7475 PWM initial duty cycle

[Chris Packham]

On 13/05/24 04:58, Guenter Roeck wrote:
> On 5/10/24 08:51, Chris Packham wrote:
>>
>> On 10/05/24 15:36, Guenter Roeck wrote:
>>> Chris,
>>>
>>> On Thu, May 09, 2024 at 06:19:12PM +0000, Chris Packham wrote:
>>>> Hi Krzysztof,
>>>>
>>>> On 9/05/24 19:06, Krzysztof Kozlowski wrote:
>>>>> On 08/05/2024 23:55, Chris Packham wrote:
>>>>>> Add documentation for the pwm-initial-duty-cycle and
>>>>>> pwm-initial-frequency properties. These allow the starting state 
>>>>>> of the
>>>>>> PWM outputs to be set to cater for hardware designs where 
>>>>>> undesirable
>>>>>> amounts of noise is created by the default hardware state.
>>>>>>
>>>>>> Signed-off-by: Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx>
>>>>>> ---
>>>>>>
>>>>>> Notes:
>>>>>>        Changes in v2:
>>>>>>        - Document 0 as a valid value (leaves hardware as-is)
>>>>>>
>>>>>>     .../devicetree/bindings/hwmon/adt7475.yaml    | 27 
>>>>>> ++++++++++++++++++-
>>>>>>     1 file changed, 26 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/hwmon/adt7475.yaml 
>>>>>> b/Documentation/devicetree/bindings/hwmon/adt7475.yaml
>>>>>> index 051c976ab711..97deda082b4a 100644
>>>>>> --- a/Documentation/devicetree/bindings/hwmon/adt7475.yaml
>>>>>> +++ b/Documentation/devicetree/bindings/hwmon/adt7475.yaml
>>>>>> @@ -51,6 +51,30 @@ properties:
>>>>>>           enum: [0, 1]
>>>>>>           default: 1
>>>>>>     +  adi,pwm-initial-duty-cycle:
>>>>>> +    description: |
>>>>>> +      Configures the initial duty cycle for the PWM outputs. The 
>>>>>> hardware
>>>>>> +      default is 100% but this may cause unwanted fan noise at 
>>>>>> startup. Set
>>>>>> +      this to a value from 0 (0% duty cycle) to 255 (100% duty 
>>>>>> cycle).
>>>>>> +    $ref: /schemas/types.yaml#/definitions/uint32-array
>>>>>> +    minItems: 3
>>>>>> +    maxItems: 3
>>>>>> +    items:
>>>>>> +      minimum: 0
>>>>>> +      maximum: 255
>>>>>> +      default: 255
>>>>>> +
>>>>>> +  adi,pwm-initial-frequency:
>>>>> Frequency usually has some units, so use appropriate unit suffix and
>>>>> drop $ref.  Maybe that's just target-rpm property?
>>>>>
>>>>> But isn't this duplicating previous property? This is fan controller,
>>>>> not PWM provider (in any case you miss proper $refs to pwm.yaml or
>>>>> fan-common.yaml), so the only thing you initially want to 
>>>>> configure is
>>>>> the fan rotation, not specific PWM waveform. If you you want to
>>>>> configure specific PWM waveform, then it's a PWM provider... but 
>>>>> it is
>>>>> not... Confused.
>>>> There's two things going on here. There's a PWM duty cycle which is
>>>> configurable from 0% to 100%. It might be nice if this was 
>>>> expressed as
>>>> a percentage instead of 0-255 but I went with the latter because 
>>>> that's
>>>> how the sysfs ABI for the duty cycle works.
>>>>
>>>> The frequency (which I'll call adi,pwm-initial-frequency-hz in v3)
>>>> affects how that duty cycle is presented to the fans. So you could 
>>>> still
>>>> have a duty cycle of 50% at any frequency. What frequency is best
>>>> depends on the kind of fans being used. In my particular case the 
>>>> lower
>>>> frequencies end up with the fans oscillating annoyingly so I use the
>>>> highest setting.
>>>>
>>> My udnerstanding is that we are supposed to use standard pwm provider
>>> properties. The property description is provider specicic, so I think
>>> we can pretty much just make it up.
>>>
>>> Essentially you'd first define a pwm provider which defines all the
>>> pwm parameters needed, such as pwm freqency, default duty cycle,
>>> and flags such as PWM_POLARITY_INVERTED. You'd then add something like
>>>
>>>     pwms = <&pwm index frequency duty_cycle ... flags>;
>>>
>>> to the node for each fan, and be done.
>>>
>>> That doesn't mean that we would actually have to register the chip
>>> as pwm provider with the pwm subsystem; all we would have to do is to
>>> interpret the property values.
>>
>> We've already got the pwm-active-state as a separate property so that
>> might be tricky to deal with, I guess it could be deprecated in favour
>> of something else. Looking at pwm.yaml and fan-common.yaml I can't quite
>> see how that'd help here. Were you thinking maybe something like
>>
>> pwm: hwmon@2e {
>>       compatible = "adi,adt7476";
>>       reg = <0x2e>;
>>       #pwm-cells = <4>;
>>       fan-0 {
>>           pwms = <&pwm 0 255 22500 PWM_POLARITY_INVERTED>;
>>           pwm-names = "PWM1";
>>           tach-ch = <0>;
>>       };
>>       fan-1 {
>>           // controlled by pwm 0
>>           tach-ch = <1>
>>       };
>>       fan-0 {
>>           pwms = <&pwm 2 255 22500 PWM_POLARITY_INVERTED>;
>>           pwm-names = "PWM3";
>>           tach-ch <2>;
>>       };
>>       fan-1 {
>>           // controlled by pwm 2
>>           tach-ch = <3>
>
> I think that would have to be
>
>     ...
>     fan-0 {
>         pwms = <&pwm 0 255 22500 PWM_POLARITY_INVERTED>;
>         tach-ch = <1 2>;
>     };
>     fan-1 {
>         tach-ch = <3>
>     };
>     ...
>
> Context: pwm-names is optional and does not add value here unless I am 
> missing
> something. Also, if I understand the bindings correctly, all 
> tachometer channels
> controlled by a single pwm are supposed to be listed in a single node. 
> With the
> above, you'd then have fan1, fan2, and fan3 plus pwm1 and pwm3 (pwm2 
> would be
> disabled/unused).
>
> Code-wise, I think you'd then call
>
>     struct of_phandle_args args;
>     ...
>     err = of_parse_phandle_with_args(np, "pwms", "#pwm-cells", 0, &args)
>
> with np pointing to the fan node. This should return the parameters in 
> 'args'.

On that point. How would I explain in the bindings that cell 2 is the 
duty cycle, cell 3 is the frequency and cell 4 is the flags?

The other complication is that one of the systems I have is x86 so I 
need to express this with the ACPI Device Properties compatibility code. 
I think I can figure out the ACPI table stuff but I can't call 
of_parse_phandle_with_args() directly.

>
> However, unless you have a use case, I'd suggest not to implement 
> support for
> "multiple fans controlled by single pwm" since that would require extra
> code and you would not actually be able to test it. A mandatory 1:1 
> mapping
> is fine with me. Support for 1:n mapping can be implemented if / when 
> there
> is a use case. 

The system I'm dealing with has exactly that. But we don't adjust the 
fan RPM directly so I think we're OK (just maybe some comments so people 
aren't confused by missing fans). The ADT7476 will adjust the PWM duty 
cycle based on the temperature, the fan RPM is just something we report 
(and generate an alarm if it goes too low).

> The same is true for registering the driver with the pwm
> subsystem - that would only be necessary if anyone ever uses one of the
> pwm channels for non-fan use.

Agreed. I won't plumb anything into the pwm subsystem. Although it would 
be kind of neat to see a LED that changes as the system gets hotter, 
kind of like an electronic thermochromic crystal.

>
> That makes me wonder if we actually need tach-ch in the first place or if
> something like
>
>     fan-0 {
>         pwms = <&pwm 0 255 22500 PWM_POLARITY_INVERTED>;
>     };
>     fan-1 {
>         pwms = <&pwm 1 255 22500 0>;
>     };
>     ...
> would do for this chip. 

Yeah that'd be fine for me.