Message-ID: <59A79FD4-0A1C-4FFE-B4BC-D24588785717@gmail.com>
Date: Thu, 15 Jan 2026 09:45:55 -0800
From: "Derek J. Clark" <derekjohn.clark@...il.com>
To: Rong Zhang <i@...g.moe>, Kurt Borja <kuurtb@...il.com>
CC: Mark Pearson <mpearson-lenovo@...ebb.ca>, Armin Wolf <W_Armin@....de>,
 Hans de Goede <hansg@...nel.org>,
 Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>,
 Guenter Roeck <linux@...ck-us.net>, platform-driver-x86@...r.kernel.org,
 linux-kernel@...r.kernel.org, linux-hwmon@...r.kernel.org
Subject: Re: [PATCH v9 7/7] platform/x86: lenovo-wmi-other: Add HWMON for fan reporting/tuning

On January 15, 2026 8:57:42 AM PST, Rong Zhang <i@...g.moe> wrote:
>Hi Kurt,
>
>On Thu, 2026-01-15 at 08:58 -0500, Kurt Borja wrote:
>> On Thu Jan 15, 2026 at 8:03 AM -05, Rong Zhang wrote:
>> > Hi Kurt,
>> > 
>> > On Wed, 2026-01-14 at 19:16 -0500, Kurt Borja wrote:
>> > > Hi Rong,
>> > > 
>> > > On Wed Jan 14, 2026 at 7:27 AM -05, Rong Zhang wrote:
>> > > > Register an HWMON device for fan reporting/tuning according to
>> > > > Capability Data 00 (capdata00) and Fan Test Data (capdata_fan) provided
>> > > > by lenovo-wmi-capdata. The corresponding HWMON nodes are:
>> > > > 
>> > > >  - fanX_enable: enable/disable the fan (tunable)
>> > > >  - fanX_input: current RPM
>> > > >  - fanX_max: maximum RPM
>> > > >  - fanX_min: minimum RPM
>> > > >  - fanX_target: target RPM (tunable)
>> > > > 
>> > > > Information from capdata00 and capdata_fan are used to control the
>> > > > visibility and constraints of HWMON attributes. Fan info from capdata00
>> > > > is collected on bind, while fan info from capdata_fan is collected in a
>> > > > callback. Once all fan info is collected, register the HWMON device.
>> > > > 
>> > > > Signed-off-by: Rong Zhang <i@...g.moe>
>> > > > Reviewed-by: Derek J. Clark <derekjohn.clark@...il.com>
>> > > > ---
>> > > 
>> > > ...
>> > > 
>> > > > diff --git a/Documentation/wmi/devices/lenovo-wmi-other.rst b/Documentation/wmi/devices/lenovo-wmi-other.rst
>> > > > index 821282e07d93c..bd1d733ff286d 100644
>> > > > --- a/Documentation/wmi/devices/lenovo-wmi-other.rst
>> > > > +++ b/Documentation/wmi/devices/lenovo-wmi-other.rst
>> > > > @@ -31,6 +31,8 @@ under the following path:
>> > > >  
>> > > >    /sys/class/firmware-attributes/lenovo-wmi-other/attributes/<attribute>/
>> > > >  
>> > > > +Additionally, this driver also exports attributes to HWMON.
>> > > > +
>> > > >  LENOVO_CAPABILITY_DATA_00
>> > > >  -------------------------
>> > > >  
>> > > > @@ -39,6 +41,11 @@ WMI GUID ``362A3AFE-3D96-4665-8530-96DAD5BB300E``
>> > > >  The LENOVO_CAPABILITY_DATA_00 interface provides various information that
>> > > >  does not rely on the gamezone thermal mode.
>> > > >  
>> > > > +The following HWMON attributes are implemented:
>> > > > + - fanX_enable: enable/disable the fan (tunable)
>> > > 
>> > > I was testing this series and I'm a bit confused about fanX_enable.
>> > 
>> > Thanks for testing!
>> 
>> Thanks for working on this!
>> 
>> > 
>> > > Judging by this comment and also by taking a quick look at the code, it
>> > > looks like writting 0 to this attribute disables the fan.
>> > > 
>> > > This is however (per hwmon ABI documentation [1]) not how this attribute
>> > > should work. IIUC, it is intended for devices which can disable the fan
>> > > sensor, not the actual fan.

I agree, it's just for disabling the reporting of the rpm, not for disabling the fan. I didn't catch this before.

>> > 
>> > Hmm, what a confusing name :-/
>> > 
>> > > I fail to see how this feature is useful and I also find it dangerous
>> > > for this to be exposed by default, considering the same could be
>> > > achieved with the relaxed module parameter, which at least tells the
>> > > user to be careful.
>> > 
>> > Makes sense. I will remove the attribute and mention such behavior in
>> > the module parameter.
>> 
>> Also, it would be worth to mention that writting 0 to the fanY_target
>> attribute is auto mode, right?
>
>Ah, yes.
>
>> I was testing the fanX_target attribute and it does work as intended,
>> i.e. the fan speed changes to the desired speed. However, every time I
>> write to this attribute I'm getting -EIO error and it always reads 0.
>> 
>> For example:
>> 
>> 	$ echo 3550 | sudo tee fan*_target
>> 	3550
>> 	tee: fan1_target: Input/output error
>> 	tee: fan2_target: Input/output error
>> 	$ cat fan*_target
>> 	0
>> 	0
>> 
>> But as I said, the fans do reach the desired speed (an approximation of
>> it):
>> 
>> 	$ cat fan*_input
>> 	3500
>> 	3500
>> 
>> This is a bit weird, but I haven't look in depth into it. I will find
>> some time to do it later. This happens on a 83KY (Legion 7 16IAX1)
>> laptop.
>
>I'd like to fix it in the next revision. Looking forward to your
>progress in debugging :-)
>
>It seems to me that the corresponding ACPI method did not return 1 on
>success. There should be some clues in the decompiled ASL code. Could
>you attach it in your reply? The HWMON implementation was developed
>mostly based on my understanding on the decompiled ASL code of my
>device. I'd like to check the one of your device as a cross-reference
>to see if there are more potential bugs.
>
>> As it seems that you can change the RPM in 100 increments,
>
>Yeah. That's also true on my device, but I am not sure if all devices
>with the interface behave like this. I'd prefer not making such an
>assumption.
>

fanY_div maybe makes sense here, defaulting to the known 100 and adjustable with a DMI quirk table if we find devices that do 200/50/20/etc.?

>> maybe you
>> could look into the pwmY attributes [1]. I think it is a better fit for
>> this feature because pwmY_enable allows you to select between manual and
>> auto fan control [2], and I believe some user-space tools already use
>> this attribute.
>
>For pwmY, I don't see the point why the kernel should adapt to
>userspace tools... If we don't have to, don't be overfit.
>
>For pwmY_enable, it is only a good choice if we adopt pwmY. It's weird
>to mix pwmY_enable and fanY_target.
>
>@Derek, may I ask for your opinion here? Should we adopt pwmY*?
>

All the platform drivers I've written used pwmY/pwmY_enable and fanY_input for reporting speed. There is additionally a pwmY_enable mode that sets the fan to max speed. TBS, I think it's a fundamentally different interface.

>> On the implementation you can use fixp_linear_interpolate() [3] to
>> convert between and from pwm duty cycle.
>
>If we adopt this, I could imagine three ways to create a PWM-RPM curve:
>(1) PWM[0,255] => RPM[min,max]
>(2) PWM[0,255] => RPM[0,max]
>(3) PWM[0,255] => RPM[0,a_large_number]
>

I don't believe this is canonical for pwm, normally you set a fixed speed with pwmY, let the BIOS handle it with auto, or set to max speed with the enable function. Fan curves would need to be under pwmY_auto_point_pwm[X|Z], but they're usually paired with tempY_auto_point_pwm[X|Y] to set a speed at a given temperature, not necessary to restrict a min/max.

Docs: https://docs.kernel.org/hwmon/sysfs-interface.html

The nature of pwm is that a fixed pulse width determines a fixed fan speed. It is necessarily a single set value. If you want a range, fanY_[min|max] seems the way to go.

Using pwmY* will also collide with the fan curve patch I'm adding after this series since newer devices have a 10 value speed Y at temp X curve built into the WMI interface. Not insurmountable, but I'd rather avoid the conflict so we're not matching on GUID or something like that.

Cheers,
Derek

>relax_fan_constraint=1 definitely needs (3) as the fan can spin
>slower/faster than min/max_rpm. I don't want to create scale mismatches
>between relax_fan_constraint=1 and =0, so only (3) looks viable to me
>in any case. Hmm, we are creating a weird PWM-RPM curve with two
>plateau [1].
>
>> This is just a suggestion though, I know I came in too late to the
>> discussion but I just got this laptop :P
>
>Happy hacking :P
>
>ThinkBook (my device) only implements few Legion interfaces, so your
>information here is very valuable.
>
>> [1] https://elixir.bootlin.com/linux/v6.19-rc5/source/Documentation/ABI/testing/sysfs-class-hwmon#L297
>> [2] https://elixir.bootlin.com/linux/v6.19-rc5/source/Documentation/ABI/testing/sysfs-class-hwmon#L312
>> [3] https://elixir.bootlin.com/linux/v6.19-rc5/source/include/linux/fixp-arith.h#L145
>
>[1]: Typical curves look like
>https://www.overclock.net/threads/are-voltage-controllers-for-fans-bad-and-pwm-controlled-fans-better.1627409/#post-25993378
>
>Thanks,
>Rong

Powered by blists - more mailing lists