On Fri, Sep 08, 2023 at 08:49:25AM +0000, Billy Tsai wrote:

On Thu, Sep 07, 2023 at 07:17:55AM +0000, Billy Tsai wrote:

quoted

quoted

On Wed, Aug 30, 2023 at 08:32:00PM +0800, Billy Tsai wrote:

quoted

quoted

From: Naresh Solanki <redacted>

Add common fan properties bindings to a schema.

Bindings for fan controllers can reference the common schema for the
fan

quoted

quoted

quoted

quoted

+properties:
+  max-rpm:
+    description:
+      Max RPM supported by fan.
+    $ref: /schemas/types.yaml#/definitions/uint32

quoted

Physics will limit this to something much less than 2^32. Add some 
constraints. 10000?

quoted

quoted

+
+  min-rpm:
+    description:
+      Min RPM supported by fan.
+    $ref: /schemas/types.yaml#/definitions/uint32

quoted

ditto

quoted

quoted

+
+  pulses-per-revolution:
+    description:
+      The number of pulse from fan sensor per revolution.
+    $ref: /schemas/types.yaml#/definitions/uint32

quoted

Needs constraints. I assume this is never more than 4 (or 2 even)?

Do you think we should add the contraint in the common binding?
In my option, the limit of the max/min rpm should be declared by
the binding if necessary, because the usage of each fan monitor is
based on the connection of the tach pin.

quoted

Yes, I think we should have default limits.

quoted

Unless we go as far as a schema for every specific fan model, then there 
is actually no way we can have specific limits unless the fan 
controllers have some limits.

quoted

The most I see in tree for pulses-per-revolution is 2. There's no value 
in more. So set the max to 4 and then if anyone needs more they can bump 
the value.

quoted

Or maybe there's some electrical/mechanical design reason fans are 1 or 
2 pulses and we'll never see anything else? This document[1] seems to 
indicate that is indeed the case. (First hit googling "fan tach signal 
pulses")

OK, I will add the maximum value for the max-rpm, min-rpm and pulses-per-revolution.

quoted

quoted

quoted

quoted

+  div:

quoted

Too generic of a name.

quoted

quoted

+    description:
+      Fan clock divisor

quoted

But what is a fan clock?

This is the divisor for the tachometer sampling clock, which determines the sensitivity of the tach pin.
So, if the name of the property changes to 'tach-div,' is it acceptable to you?

quoted

That sounds like a property of the controller, not the fan, so it 
belongs in the controller binding. Is this really a common thing?

Yes, I believe this is a common feature for fans. You can refer to the Documentation/hwmon/sysfs-interface.rst,
where the fan divisor is defined for users, determining the fan's sensitivity.

It's a feature of the controller, but I guess since it may be a per 
fan or tach channel setting having it in the fan node is fine. 
'tach-div' is fine.

quoted

quoted

quoted

quoted

+    $ref: /schemas/types.yaml#/definitions/uint32
+
+  target-rpm:
+    description:
+      Target RPM the fan should be configured during driver probe.

quoted

What driver? By the time the OS driver runs, a bunch of other boot 
software has already run on modern systems. So this value would likely 
be used much earlier. The point is that when exactly is outside the 
scope of DT. This is "what RPM do I use in case of no other information 
(e.g. temperature)".

So, the description should be changed to 'The default desired fan speed in RPM,'
and we shouldn't mention the timing of the property's operation in the DT, is that correct?

quoted

Correct.

quoted

quoted

quoted

quoted

+    $ref: /schemas/types.yaml#/definitions/uint32
+
+  mode:

quoted

Too generic.

quoted

quoted

+    description:
+      Select the operational mode of the fan.

quoted

What are modes? Spin and don't spin?

The mode is used to indicate the driving mode of the fan (DC, PWM and so on).
So, if the name of the property changes to 'fan-driving-mode,' is it acceptable to you?

quoted

I tend to think that should be implied from the parent node and/or other 
properties. PWM if "pwms" property is present. DC if the supply is 
variable. We could also use compatible strings in the fan nodes if 
there's a need.

So, it looks that this property isn't necessary for the fan. And it should be determined by the
present of the driving source. is that correct?

Looking back, I see Guenter asked for something like this. So I guess it 
is fine to keep if he still thinks we need it. I think it could also be 
implied a bit more clearly if we do 'cooling-levels' as below. The fan 
controller should know what modes it can do. Duty cycle vs. voltage in 
uV is pretty easy to distinguish.

quoted

That reminds me, both of these modes probably need a table of 
voltage/duty-cycle to RPMs. I imagine it's not always a linear response.  
Naresh also privately sent me (don't do that) an updated common binding 
which we discussed the need for this. I expect him to comment further 
with details.

For this purpose, we should add the speed-map like the usage of the gpio-fan, right?
https://github.com/torvalds/linux/blob/master/Documentation/devicetree/bindings/hwmon/gpio-fan.txt

That should be fine. I guess it comes down to whether we want one 
property or 3 depending on whether the control is voltage, duty-cycle, 
GPIO pins state, etc. Unfortunately, it's "gpio-fan,speed-map", not 
"speed-map", so we can't use that as-is. There's also "cooling-levels" 
which can accomplish the same thing. We don't know RPMs with that, but 
do we need to? Maybe, as it seems some drivers want to set speed in RPMs 
from userspace. Could also do something like this:

cooling-levels = < 33   66  100>;
rpm-levels =     <500 1000 1500>;

"rpm-levels" would replace the min-rpm/max-rpm properties.


Also, looking at gpio-fan.txt, it looks like you should add 
"#cooling-cells" as well.

Rob

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel