Re: [PATCH v5 1/2] dt-bindings: hwmon: Add ina3221 documentation

On 09/25/2018 08:08 PM, Nicolin Chen wrote:

Hello Guenter,

On Tue, Sep 25, 2018 at 06:52:29PM -0700, Guenter Roeck wrote:

quoted

quoted

+2) child nodes
+  Required properties:
+  - reg: Must be 0, 1 or 2, corresponding to IN1, IN2 or IN3 port of INA3221
+
+  Optional properties:
+  - label: Name of the input source
+  - shunt-resistor-micro-ohms: Shunt resistor value in micro-Ohm
+  - status: Should be "disabled" if no input source
+
+  Example:
+
+  input@0 {
+          reg = <0x0>;
+          status = "disabled";

I kind of feel embarrassed that I asked for the reg change ... especially while
saying at the same time that I would like to see this work for other chips
as well.

Well, though I didn't mention it, yet I changed it to "reg" is more
likely an agreement than a compromise: I searched in the mail list
and then found this mail (a year ago though):
     https://www.spinics.net/lists/kernel/msg2455439.html

I feel it very similar to my case. So rather than betting Rob won't
tell me the same, changing to "reg" may reduce a turnaround time :)

quoted

Other chips have different kinds of sensors. Voltage, current, power, temperature,
and others. Whatever we come up with should support that.

I see two possibilities right now. We can stick with reg and add a "type" property,
or we can make the index something like
	{voltage,current,power,temperature,humidity}-{id,index}

One small concern is a case of being multi-type. For example, I saw
ina2xx driver having voltage, current and power at the same time...

Yes, with that we would have something like

	voltage@0 {
		type = "voltage";
		reg = <0>;
	};
	current@0 {
		type = "current";
		reg = <0>;
	};
	...

or

	voltage@0 {
		voltage-id = <0>;
	};
	current@0 {
		current-id = <0>;
	};
	...

quoted

I personally prefer "type", but I don't really have a strong opinion.
What do you think ?

I also like this over "reg" -- "reg" requires two extra properties,
and itself sounds slightly unnatural to me for situations like this
one where we don't use it as a register address, although I know it
is convenient and common to use.

With "type", we would still need two properties.

	reg = <0>;
	type = "voltage";

and type could be optional or not required for a chip only supporting
a single sensor type (like the ina3221).

This would be equivalent to, say,
	voltage-index = <0>;
when using a single property.

With the "reg" approach, we would be ok for now - however, I would like
to get feedback from Rob if introducing a "type" property will be
acceptable when the time comes to do so.

quoted

Or maybe we should really wait for feedback from Rob.

Personally I don't mind it all to change the doc and code and then
send a v6. But eventually we'll still need the final Acked-by from
Rob right? Then I guess it's the only option.

Yes. At this point I'd rather have input from Rob than moving forward
with v6.

By the way, I have two ina3221 hwmon patches that rebase upon this
binding series. And I'd like to send them out to go through review
first, but I am not sure if you'd be okay for it -- I don't really
like to change their rebase order as it might mess up something.

Are those bug fixes or further enhancements ? For enhancements,
it is your call when to send them; I am fine either way. If they are
bug fixes, they should come first so we can apply them to -stable.

Thanks,
Guenter