Message-ID: <4de798f3-069e-4028-a5b5-5e6a639277e3@alliedtelesis.co.nz>
Date: Mon, 20 May 2024 20:55:27 +0000
From: Chris Packham <Chris.Packham@...iedtelesis.co.nz>
To: Conor Dooley <conor@...nel.org>, Guenter Roeck <linux@...ck-us.net>
CC: "jdelvare@...e.com" <jdelvare@...e.com>, "robh@...nel.org"
	<robh@...nel.org>, "krzk+dt@...nel.org" <krzk+dt@...nel.org>,
	"conor+dt@...nel.org" <conor+dt@...nel.org>, "ukleinek@...nel.org"
	<ukleinek@...nel.org>, "linux-hwmon@...r.kernel.org"
	<linux-hwmon@...r.kernel.org>, "devicetree@...r.kernel.org"
	<devicetree@...r.kernel.org>, "linux-kernel@...r.kernel.org"
	<linux-kernel@...r.kernel.org>, "linux-pwm@...r.kernel.org"
	<linux-pwm@...r.kernel.org>
Subject: Re: [PATCH v3 1/3] dt-bindings: hwmon: Add adt7475 fan/pwm properties


On 21/05/24 07:41, Conor Dooley wrote:
> On Mon, May 20, 2024 at 12:03:42PM -0700, Guenter Roeck wrote:
>> On 5/20/24 09:49, Conor Dooley wrote:
>>> On Mon, May 20, 2024 at 03:03:19PM +1200, Chris Packham wrote:
>>>> Add fan child nodes that allow describing the connections for the
>>>> ADT7475 to the fans it controls. This also allows setting some
>>>> initial values for the pwm duty cycle and frequency.
>>>>
>>>> Signed-off-by: Chris Packham <chris.packham@...iedtelesis.co.nz>
>>>> ---
>>>>
>>>> Notes:
>>>>       Changes in v3:
>>>>       - Use the pwm provider/consumer bindings
>>>>       Changes in v2:
>>>>       - Document 0 as a valid value (leaves hardware as-is)
>>>>
>>>>    .../devicetree/bindings/hwmon/adt7475.yaml    | 25 ++++++++++++++++++-
>>>>    1 file changed, 24 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/hwmon/adt7475.yaml b/Documentation/devicetree/bindings/hwmon/adt7475.yaml
>>>> index 051c976ab711..99bd689ae0cd 100644
>>>> --- a/Documentation/devicetree/bindings/hwmon/adt7475.yaml
>>>> +++ b/Documentation/devicetree/bindings/hwmon/adt7475.yaml
>>>> @@ -51,6 +51,15 @@ properties:
>>>>          enum: [0, 1]
>>>>          default: 1
>>>> +  "#pwm-cells":
>>>> +    const: 4
>>>> +    description: |
>>>> +      Number of cells in a PWM specifier.
>>>> +      - 0: The pwm channel
>>>> +      - 1: The pwm frequency in hertz - 0, 11, 14, 22, 29, 35, 44, 58, 88, 22500
>>> The standard binding is period in nanoseconds, not frequency in Hz.
>>> What's gained from deviating from that?
>>>
>> I'd strongly suspect that Chris didn't know and didn't expect it,
>> just like me.
> I did point out on v2 that the information on the standard binding was
> in pwm.txt, but I also said "order it has <index freq flags duty>" which
> likely didn't help. I did however CC you both on a patch the other day
> where I fixed pwm.yaml so that it actually contained the information on
> what the cells represented.

I did see that patch thanks. And I did adjust the order of the 
parameters I had based on it.

I personally found expressing the frequency in seconds confusing which 
the is main reason why I stuck with hertz. I also couldn't grok how to 
express the duty cycle if I did express the frequency in nanoseconds. I 
think maybe I'm a bit confused as to how to express a generic PWM period 
when the controller I'm dealing with has a frequency and a duty cycle.

>>>> +      - 2: PWM flags 0 or PWM_POLARITY_INVERTED
>>>> +      - 3: The default pwm duty cycle - 0-255
>>> Same here I guess, why not match the units used for the period for the
>>> duty cycle?
>>>
>> Same here. I am used to pwm frequency in Hz and duty cycle as percentage.
>> Using the period instead is somewhat unusual, and I must admit that I
>> have never encountered it while dealing with a variety of fan controllers.
My expectation is that a duty cycle is a percentage. The only reason I'm 
using 0-255 instead of 0-100 is because that matches the sysfs ABI. But 
I'd happily change to a true percentage (if I can figure out the integer 
math), we'd lose a little bit of precision but nothing anyone would 
really notice.
> If it is what makes sense to use, because it's what everyone and their
> mother documents, then sure. My rationale I suppose was threefold:
> - Consistency with the period
> - Time would be more portable between providers, if 8 bits of precision
>    is deemed insufficient for some providers.
> - Last & least, the PWM APIs in the kernel use time for the dutycycle
>
> If you're going to use percentages rather than time, would it not
> make more sense to either use percent itself with a 0-100 range or use
> basis points if percent doesn't provide sufficient granularity?
>
> I think it'd be good of Uwe chimed in, given we're discussing a PWM
> provider binding after all.

I'd also point out I'm more than happy to go back to my v2 approach of 
adding specific properties for the frequency and duty cycle.

>> Just to make sure I understand this correctly - duty cycles would
> s/duty cycles/periods/
>
>> be (rounded):
>>
>> Hz	ns
>> 11	90,909,091
>> 14	71,428,571
>> 22	45,454,545
>> 29:	34,482,759
>> 35:	28,571,429
>> 44:	22,727,273
>> 58:	17,241,379
>> 88:	11,363,636
>> 22500	44,444
>>
>> Examples for duty cycles would be
>>
>> 20%: 0,2s or 200,000,000
>> 50%: 0.5s or 500,000,000
>> 80%: 0.8s or 800,000,000
>>
>> Is that correct ?
> Assuming a 1 second period, those look as expected.

To me that just makes things harder to understand. If you were reading 
the ADT7475 datasheet you'd see things like the supported frequencies of 
11, 14, ... , 88, 22500. Having to express that in nanoseconds requires 
making a bit of a leap. The duty cycle percentage to nanoseconds is more 
straight forward but seems unnecessarily obscure to me.

>
> Cheers,
> Conor.
>

Powered by blists - more mailing lists