On Thu, 19 Nov 2020 16:39:07 +0100 Hans de Goede <hdegoede@xxxxxxxxxx> wrote: > 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@xxxxxxxxxx> 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. It's fairly new. Note we also have per channel labels though they are 'very new'. Might be handy if the sensors appear as a single device despite being spread over the laptop. > > 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. If you do want to do this, please just put it under sysfs-bus-iio doc. I want this to be in the top level doc and there is an issue we are currently trying to sort out with autogenerated docs and repeats of a given filename in the ABI docs. (basically it doesn't work and generates lots of warnings!) Thanks, Jonathan > > >> 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 > > > >