| Message ID | 20250103101448.890946-1-peter@korsgaard.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v4,1/2] dt-bindings: hwmon: pwm-fan: Document default-pwm property | expand |
On Fri, Jan 03, 2025 at 11:14:47AM +0100, Peter Korsgaard wrote: > The pwm-fan driver uses full PWM (255) duty cycle at startup, which may not > always be desirable because of noise or power consumption peaks, so add an > optional "default-pwm" property that can be used to specify a custom default > PWM duty cycle (0..255). > > This is somewhat similar to target-rpm from fan-common.yaml, but that cannot > be used here as target-rpm specifies the target fan speed, whereas this is > the default pwm to set when the device is instantiated - And the resulting > fan RPM resulting from a given PWM duty cycle is fan dependent. I still don't agree. Quoting Guenter: > The two values are also orthogonal. The fan rpm is fan dependent. > Each fan will require a different pwm value to reach the target speed. > Trying to use target-rpm to set a default pwm value would really > not make much if any sense. But RPM is ultimately what you care about and is the fan parameter that's universal yet independent of the underlying control. RPM is what determines noise and power consumption. There's 2 cases to consider: you have a tach signal and know the fan RPM or you don't know the RPM. If you have a tach signal, we probably wouldn't be discussing this because target-rpm would be enough. So I'm assuming this is the case and you have no idea what RPM the fan runs at. The fan-common.yaml binding is a bit incomplete for this. What you need is some map of fan speed to PWM duty cycle as most likely it is not linear response. I think there are 2 options here: Use the 'cooling-levels' property. Fan "speed" is the index of the array. So you just need a 'default-cooling-level' property that's the default index. The other option is define an array of (fan RPM, PWM duty cycle) tuples. Then target-rpm can be used to select the entry. We already have something like this with 'gpio-fan,speed-map'. There's also no definition of the minimum RPM or duty cycle in the pwm-fan binding. We have min-rpm in fan-common, but that doesn't work without a tach. A map would help here as well This problem to me is similar to LEDs. Ultimately it's brightness that you care about, not the current or PWM duty cycle to get there. Finally, whatever we end up with, it should go in fan-common.yaml. That supports PWMs too, so whatever we end up with is applicable to any PWM controlled fan. Rob
On 1/3/25 20:58, Rob Herring wrote: > I still don't agree. Quoting Guenter: > >> The two values are also orthogonal. The fan rpm is fan dependent. >> Each fan will require a different pwm value to reach the target speed. >> Trying to use target-rpm to set a default pwm value would really >> not make much if any sense. > But RPM is ultimately what you care about and is the fan parameter > that's universal yet independent of the underlying control. RPM is what > determines noise and power consumption. > > There's 2 cases to consider: you have a tach signal and know the fan RPM > or you don't know the RPM. If you have a tach signal, we probably > wouldn't be discussing this because target-rpm would be enough. So I'm > assuming this is the case and you have no idea what RPM the fan runs at. Correct, no tacho. > The fan-common.yaml binding is a bit incomplete for this. What you need > is some map of fan speed to PWM duty cycle as most likely it is not > linear response. I think there are 2 options here: > > Use the 'cooling-levels' property. Fan "speed" is the index of the > array. So you just need a 'default-cooling-level' property that's the > default index. I am not sure I what you mean with the RPM reference here? The cooling-levels support in the fan-pwm.c driver is a mapping between cooling levels and PWM values, NOT RPM value. > The other option is define an array of (fan RPM, PWM duty cycle) tuples. > Then target-rpm can be used to select the entry. We already have > something like this with 'gpio-fan,speed-map'. Where should these "invented" RPM values come from when there is no tacho signal? That sounds backwards / complicated for the very trivial "what should the default PWM value be at driver probe time" use case. > There's also no definition of the minimum RPM or duty cycle in the > pwm-fan binding. We have min-rpm in fan-common, but that doesn't work > without a tach. A map would help here as well The minimum PWM is presumably 0, E.G. signal always low? > This problem to me is similar to LEDs. Ultimately it's brightness that > you care about, not the current or PWM duty cycle to get there. The use case (as described in the commit message) is to drive the fan less hard to limit noise and/or power consumption. The input to the fan drive control is a PWM setting, so it IMHO makes sense to specify that, as that is the interface provided by the fan-pwm driver - E.G. you boot up and tweak the pwm1 property in sysfs until you have a value that suits the noise/power consumption requirements and stick that value in the dts. > Finally, whatever we end up with, it should go in fan-common.yaml. That > supports PWMs too, so whatever we end up with is applicable to any PWM > controlled fan. What makes this "default-pwm" (or whatever it will be called) more generic than E.G. the recently added "fan-stop-to-start-percent" / "fan-stop-to-start-usec" properties added to pwm-fan.yaml by commit 80bc64201e78 ("dt-bindings: hwmon: pwm-fan: Document start from stopped state properties")?
On 1/3/25 11:58, Rob Herring wrote: > On Fri, Jan 03, 2025 at 11:14:47AM +0100, Peter Korsgaard wrote: >> The pwm-fan driver uses full PWM (255) duty cycle at startup, which may not >> always be desirable because of noise or power consumption peaks, so add an >> optional "default-pwm" property that can be used to specify a custom default >> PWM duty cycle (0..255). >> >> This is somewhat similar to target-rpm from fan-common.yaml, but that cannot >> be used here as target-rpm specifies the target fan speed, whereas this is >> the default pwm to set when the device is instantiated - And the resulting >> fan RPM resulting from a given PWM duty cycle is fan dependent. > > I still don't agree. Quoting Guenter: > >> The two values are also orthogonal. The fan rpm is fan dependent. >> Each fan will require a different pwm value to reach the target speed. >> Trying to use target-rpm to set a default pwm value would really >> not make much if any sense. > That is a mis-quote, really. I was talking about target-rpm. The property to be added here is default-pwm, which is completely different. > But RPM is ultimately what you care about and is the fan parameter > that's universal yet independent of the underlying control. RPM is what > determines noise and power consumption. > > There's 2 cases to consider: you have a tach signal and know the fan RPM > or you don't know the RPM. If you have a tach signal, we probably > wouldn't be discussing this because target-rpm would be enough. So I'm > assuming this is the case and you have no idea what RPM the fan runs at. > The fan-common.yaml binding is a bit incomplete for this. What you need > is some map of fan speed to PWM duty cycle as most likely it is not > linear response. I think there are 2 options here: > > Use the 'cooling-levels' property. Fan "speed" is the index of the > array. So you just need a 'default-cooling-level' property that's the > default index. > > The other option is define an array of (fan RPM, PWM duty cycle) tuples. > Then target-rpm can be used to select the entry. We already have > something like this with 'gpio-fan,speed-map'. > That won't work. Literally each individual fan (even the same model) runs at a different rpm for a given pwm value, and I am sure the rpm for a given pwm input changes depending on fan age and temperature. Quoting from the NCT6796D-S datasheet, for an example: "The default duty cycle is 7Fh, or 50% for SYSFANOUT, CPUFANOUT, AUXFANOUT0, AUXFANOUT1, AUXFANOUT2, AUXFANOUT3 and AUXFANOUT4. Note.The default speed of fan output is specified in registers CR[E0h] ~ CR[E4h] and CR[E7h] of Logical Device B, and the AUXFANOUT4 default speed of fan output is specified in register CR[E3h] of Logical Device D. " Note that the term "default speed" as used in the datasheet refers to pwm. Again, from the datasheet: CR E3h. AUXFAN4 Duty Cycle Register ^^^^^^^^^^^^^^^^^^^ Setting the default value for such a register is what we are talking about here. No rpm involved, just default pwm values. This is completely independent of rpm. Guenter
On 1/3/25 11:58, Rob Herring wrote: > > But RPM is ultimately what you care about and is the fan parameter > that's universal yet independent of the underlying control. RPM is what > determines noise and power consumption. > I forgot to add: I would argue that airflow restrictions play a substantial role for both power consumption and noise, especially when trying to achieve a specific fan speed. For power consumption, I don't even see how rpm would play a substantial role in the first place since power consumption depends on pwm, not on the resulting rpm (or airflow). Guenter
On Sun, Jan 05, 2025 at 05:10:42PM +0100, Peter Korsgaard wrote: > On 1/3/25 20:58, Rob Herring wrote: > > > I still don't agree. Quoting Guenter: > > > > > The two values are also orthogonal. The fan rpm is fan dependent. > > > Each fan will require a different pwm value to reach the target speed. > > > Trying to use target-rpm to set a default pwm value would really > > > not make much if any sense. > > > But RPM is ultimately what you care about and is the fan parameter > > that's universal yet independent of the underlying control. RPM is what > > determines noise and power consumption. > > > > There's 2 cases to consider: you have a tach signal and know the fan RPM > > or you don't know the RPM. If you have a tach signal, we probably > > wouldn't be discussing this because target-rpm would be enough. So I'm > > assuming this is the case and you have no idea what RPM the fan runs at. > > Correct, no tacho. > > > > The fan-common.yaml binding is a bit incomplete for this. What you need > > is some map of fan speed to PWM duty cycle as most likely it is not > > linear response. I think there are 2 options here: > > > > Use the 'cooling-levels' property. Fan "speed" is the index of the > > array. So you just need a 'default-cooling-level' property that's the > > default index. > > I am not sure I what you mean with the RPM reference here? The > cooling-levels support in the fan-pwm.c driver is a mapping between cooling > levels and PWM values, NOT RPM value. Did I say RPM anywhere for this option? It is the index of the array that is meaningful to anything outside of the driver. The values are opaque. They are duty cycle in some cases and RPMs in other cases. The thermal subsystem knows nothing about PWM duty cycle nor RPMs. Defining a default-cooling-level would be useful to anyone, not just your usecase. IOW, you are proposing: default-pwm = <123>; I'm proposing doing this instead: cooling-levels = <0 123 255>; default-cooling-level = <1>; > > The other option is define an array of (fan RPM, PWM duty cycle) tuples. > > Then target-rpm can be used to select the entry. We already have > > something like this with 'gpio-fan,speed-map'. > > Where should these "invented" RPM values come from when there is no tacho > signal? That sounds backwards / complicated for the very trivial "what > should the default PWM value be at driver probe time" use case. Every fan has at least a maximum RPM spec'ed. Probably a minium too. For anything in between, you're correct that you don't know. I guess you just assume a linear response. > > There's also no definition of the minimum RPM or duty cycle in the > > pwm-fan binding. We have min-rpm in fan-common, but that doesn't work > > without a tach. A map would help here as well > > The minimum PWM is presumably 0, E.G. signal always low? I'm talking about the duty cycle needed to start the fan spinning and to keep it spinning. I'm sure that value is not 1 for any fan except one in a physics textbook (the only place friction does not exist). Maybe the minimum is index 0 of cooling-levels? > > This problem to me is similar to LEDs. Ultimately it's brightness that > > you care about, not the current or PWM duty cycle to get there. > > The use case (as described in the commit message) is to drive the fan less > hard to limit noise and/or power consumption. The input to the fan drive > control is a PWM setting, so it IMHO makes sense to specify that, as that is > the interface provided by the fan-pwm driver - E.G. you boot up and tweak > the pwm1 property in sysfs until you have a value that suits the noise/power > consumption requirements and stick that value in the dts. I understand what you want. I'm trying to think ahead about what's the next thing someone will want to add to the binding. Just adding 1 property at a time does not result in the best binding design. There's plenty of examples of that. Second, we have these fan bindings like pwm-fan which predate coming up with a common binding. Any further evolution of these bindings should not further diverge from the common binding. If your process do this once for a given platform, then having this in DT is fine. If the process is every user of the platform does this, then I don't think it should be in DT. Having users tweak the DT is not a great experience compared to just putting the setting in a file on the rootfs. > > Finally, whatever we end up with, it should go in fan-common.yaml. That > > supports PWMs too, so whatever we end up with is applicable to any PWM > > controlled fan. > > What makes this "default-pwm" (or whatever it will be called) more generic > than E.G. the recently added "fan-stop-to-start-percent" / > "fan-stop-to-start-usec" properties added to pwm-fan.yaml by commit > 80bc64201e78 ("dt-bindings: hwmon: pwm-fan: Document start from stopped > state properties")? Nothing. Those should probably be moved. Really, pwm-fan.yaml should reference fan-common.yaml and drop all the duplicate properties. Rob
On 1/6/25 09:38, Rob Herring wrote: >> The minimum PWM is presumably 0, E.G. signal always low? > > I'm talking about the duty cycle needed to start the fan spinning and to > keep it spinning. I'm sure that value is not 1 for any fan except one in > a physics textbook (the only place friction does not exist). > > Maybe the minimum is index 0 of cooling-levels? > Fan controllers often have a separate value for the lowest cooling level and for the PWM necessary to start a fan. The pwm associated with the lowest cooling level may be lower than the pwm necessary to start a fan. The fan controller would start a fan by applying the "start fan" pwm value for a period of time, then reduce the pwm to the desired pwm value for the running fan. I'll be happy to dig up datasheet references if needed. Guenter
diff --git a/Documentation/devicetree/bindings/hwmon/pwm-fan.yaml b/Documentation/devicetree/bindings/hwmon/pwm-fan.yaml index 8b4ed5ee962f..873c4c32e608 100644 --- a/Documentation/devicetree/bindings/hwmon/pwm-fan.yaml +++ b/Documentation/devicetree/bindings/hwmon/pwm-fan.yaml @@ -20,6 +20,13 @@ properties: items: maximum: 255 + default-pwm: + description: Default PWM duty cycle value to use at startup + $ref: /schemas/types.yaml#/definitions/uint32 + minimum: 0 + maximum: 255 + default: 255 + fan-supply: description: Phandle to the regulator that provides power to the fan. @@ -100,6 +107,7 @@ examples: pwm-fan { compatible = "pwm-fan"; pwms = <&pwm 0 40000 0>; + default-pwm = <75>; fan-supply = <®_fan>; interrupt-parent = <&gpio5>; interrupts = <1 IRQ_TYPE_EDGE_FALLING>;
The pwm-fan driver uses full PWM (255) duty cycle at startup, which may not always be desirable because of noise or power consumption peaks, so add an optional "default-pwm" property that can be used to specify a custom default PWM duty cycle (0..255). This is somewhat similar to target-rpm from fan-common.yaml, but that cannot be used here as target-rpm specifies the target fan speed, whereas this is the default pwm to set when the device is instantiated - And the resulting fan RPM resulting from a given PWM duty cycle is fan dependent. Signed-off-by: Peter Korsgaard <peter@korsgaard.com> --- Changes since v3: - Fix example syntax - Extend description of why target-rpm cannot be used Changes since v2: - Recreated/resent Changes since v1: - Rename to default-pwm Documentation/devicetree/bindings/hwmon/pwm-fan.yaml | 8 ++++++++ 1 file changed, 8 insertions(+)