Re: [External] Using IIO to export laptop palm-sensor and lap-mode info to userspace?

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 11/13/20 7:58 AM, Dmitry Torokhov wrote:
> On Thu, Nov 12, 2020 at 10:50:12AM +0100, Hans de Goede wrote:
>> Hi,
>>
>> On 11/12/20 7:23 AM, Dmitry Torokhov wrote:
>>> On Wed, Oct 07, 2020 at 11:51:05AM +0200, Hans de Goede wrote:
>>>> Hi,
>>>>
>>>> On 10/7/20 10:36 AM, Jonathan Cameron wrote:
>>>>> On Mon, 5 Oct 2020 22:04:27 -0400
>>>>> Mark Pearson <markpearson@lenovo.com> wrote:
>>>>>
>>>>>> Adding Nitin, lead for this feature, to the thread
>>>>>
>>>>> +CC linux-input and Dmitry for reasons that will become clear below.
>>>>>>
>>>>>> On 2020-10-03 10:02 a.m., Hans de Goede wrote:
>>>>>>> Hi All,
>>>>>>>
>>>>>>> Modern laptops can have various sensors which are kinda
>>>>>>> like proximity sensors, but not really (they are more
>>>>>>> specific in which part of the laptop the user is
>>>>>>> proximate to).
>>>>>>>
>>>>>>> Specifically modern Thinkpad's have 2 readings which we
>>>>>>> want to export to userspace, and I'm wondering if we
>>>>>>> could use the IIO framework for this since these readings
>>>>>>> are in essence sensor readings:
>>>>>>>
>>>>>>> 1. These laptops have a sensor in the palm-rests to
>>>>>>> check if a user is physically proximate to the device's
>>>>>>> palm-rests. This info will be used by userspace for WWAN
>>>>>>> functionality to control the transmission level safely.
>>>>>>>
>>>>>>> A patch adding a thinkpad_acpi specific sysfs API for this
>>>>>>> is currently pending:
>>>>>>> https://patchwork.kernel.org/patch/11722127/
>>>>>>>
>>>>>>> But I'm wondering if it would not be better to use
>>>>>>> IIO to export this info.
>>>>>
>>>>> My first thought on this is it sounds more like a key than a sensor
>>>>> (simple proximity sensors fall into this category as well.)
>>>
>>> [ sorry for sitting on this thread for so long ]
>>>
>>> So I think the important question here is if we only ever want yes/no
>>> answer, or if we can consider adjusting behavior of the system based on
>>> the "closeness" of an object to the device, in which case I think IIO is
>>> more flexible.
>>>
>>> FWIW in Chrome OS land we name IIO proximity sensors using a scheme
>>> "proximity-lte", "proximity-wifi", "proximity-wifi-left",
>>> "proximity-wifi-right", etc, and then userspace implements various
>>> policies (SAR, etc) based off it.
>>
>> Interesting, so 2 questions:
>>
>> 1. So your encoding the location in the sensor's parent-device name
>> instead of using a new sysfs attribute for this ?
> 
> I think it depends on the kernel we use and architecture. On x86 I think
> we rely on udev, like this:
> 
> https://chromium.googlesource.com/chromiumos/overlays/board-overlays/+/master/overlay-nocturne/chromeos-base/chromeos-bsp-nocturne/files/udev/99-cros-sx-proximity.rules
> 
> DEVPATH=="*/pci0000:00/0000:00:15.1/*", SYMLINK+="proximity-wifi-right"
> DEVPATH=="*/pci0000:00/0000:00:19.1/*", SYMLINK+="proximity-wifi-left"
> ATTR{events/in_proximity1_USE_CS1_thresh_either_en}="1"

So that results in a symlink under /dev, right ? That seems like
it is not really compatible with how most modern userspace discovers
hw (through udev). Although I guess code using udev could still
lookup the symlink in the udev per device data, this just not feel
like a good way forward.

> On newer ARM we use "label" attribute in DTS:
> 
> arch/arm64/boot/dts/qcom/sc7180-trogdor.dtsi
> 
>         ap_sar_sensor: proximity@28 {
>                 compatible = "semtech,sx9310";
>                 reg = <0x28>;
>                 #io-channel-cells = <1>;
>                 pinctrl-names = "default";
>                 pinctrl-0 = <&p_sensor_int_l>;
> 
>                 interrupt-parent = <&tlmm>;
>                 interrupts = <24 IRQ_TYPE_LEVEL_LOW>;
> 
>                 vdd-supply = <&pp3300_a>;
>                 svdd-supply = <&pp1800_prox>;
> 
>                 status = "disabled";
>                 label = "proximity-wifi";
>         };

Hmm, interesting. I did not know iio-devices could
have a label sysfs attribute (nor that that could be
set through device-tree). I was thinking about adding
an in_proximity_location sysfs attribute. But using
labels (and standardizing a set of label names) will
work nicely too.

I have no real preference for this either way, so
I guess we might as well go with labels to avoid
having any unnecessary discrepancies between ChromeOS
and whatever we do for the Thinkpad sensors.

Is there a know set of labels which ChromeOS is currently
using? If we are going to use labels for this it would
be good IMHO to define a set of standard labels for
this in say Documentation/ABI/testing/sysfs-bus-iio-labels.

>> 2. Do these sensors just give a boolean value atm, or do they already
>> report a range ?  IIRC one of the objections from the iio folks in
>> the Lenovo case was that booleans are not really a good fit for iio
>> (IIRC they also said we could still use iio for this).
> 
> One of the sensors we use is sx9310 that I believe can report range, but
> I think we configure them to trigger when a threshold is crossed.
> 
> Events are handled by our powerd:
> 
> https://chromium.googlesource.com/chromiumos/platform2/+/master/power_manager/powerd/system/sar_watcher.cc
> 
>>
>> Perhaps you can provide an URL to the kernel code implementing these ?
> 
> drivers/iio/proximity/sx9310.c

If I'm reading that correctly the it exports a raw "distance"
reading and a suggested threshold value for the code interpreting
the reading to use.

So that would be a bit different then the Thinkpad sensors, but
exporting just a 0-1 range for the in_proximity_raw value for the
Thinkpad case should not be a problem. Or we could just make it
repot 0 and 100 and export a fixed in_proximity_nearlevel of 50,
that would make the userspace API more like other proximity sensors.

Regards,

Hans