Re: [PATCH 00/11] Input: synaptics-rmi4: various fixes for the existing rmi4 branch

On Wed, Nov 4, 2015 at 1:38 AM, Andrew Duggan [off-list ref] wrote:

On 11/03/2015 06:01 AM, Linus Walleij wrote:

quoted

On Mon, Nov 2, 2015 at 11:14 PM, Andrew Duggan [off-list ref]
wrote:

quoted

I have been continuing to work on the synaptics-rmi4 driver and was just
trying to figure out what the next step should be. I recently uploaded my
changes here https://github.com/aduggan/linux.

I just tested this patch set on the Ux500 with a TVK UI board.

I added this:

         i2c@80110000 {
             synaptics@4b {
                 /* Synaptics RMI4 TM1217 touchscreen */
                 compatible = "syna,rmi-i2c";
                 #address-cells = <1>;
                 #size-cells = <0>;
                 reg = <0x4b>;
                 pinctrl-names = "default";
                 pinctrl-0 = <&synaptics_tvk_mode>;
                 interrupt-parent = <&gpio2>;
                 interrupts = <20 IRQ_TYPE_EDGE_FALLING>;
                 syna,sensor-name="TM1217";

                 rmi-f01@1 {
                     reg = <0x1>;
                     syna,nosleep = <1>;
                 };
                 rmi-f11@11 {
                     reg = <0x11>;
                     syna,f11-flip-x = <1>;
                     syna,sensor-type = <1>;
                 };
             };
         };

Bootlog:
[    2.143127] rmi_f01 sensor00.fn01: found RMI device, manufacturer:
Synaptics, product: TM1217
[    2.155242] input: Synaptics RMI4 Touch Sensor as
/devices/sensor00/input/input2
[    2.165466] rmi_i2c 3-004b: registered rmi i2c driver at 0x4b.

Doing cat /dev/input/event2 gives noise on screen, yay.

Some quick questions I see immediately:

- The DT examples in Documentation/devicetree/bindings/input/*
   omit
   #address-cells = <1>;
   #size-cells = <0>;
   for the I2C device children (i.e. the function nodes), it needs to look
   like my example above to work.

The devices which I have tested on haven't needed #address-cells or
#size-cells. It looks like they are inheriting values defined at higher
levels. I can add them to the example for completeness and update the
documentation to say that they may be needed.

quoted

- All things boolean like syna,nosleep and syna,f11-flip-x
   should just be like:
   syna,nosleep;
   syna,f11-flip-x;
   in the device tree. Use of_property_read_bool() for these.

That makes sense. I will switch them to booleans.

quoted

- syna,sensor-name = "FOO";
   Why?
   The bootlog clearly states that f01 can autodetect the sensor
   type. And f01 is always compiled in, right? So just cut this
   binding and handling, if the two don't match it's just
   super-confusing.

The sensor name in the platform data is used by some debug messages before
F01 is loaded. But, I agree it can be confusing having multiple names for
the device. Especially, when the sensor name is the same as the product id.
I'll see if it makes sense to use another name like the transport device's
name in the logs.

quoted

- /proc/interrupts say this:
  206:        114          0  nmk2-64-95  20 Edge      sensor00
  sensor00? Unhelpful. Why can't it say "TM1217", give take
  an instance number, with the detected sensor name?

Currently, rmi_dev's device name is also being set before F01 is loaded.
I'll see if we can simply set it later when we have the product id from F01.

quoted

But to me the code seems overall pretty mature. It just works.
I'll be happy to give a more detailed review if you post it.

Great! I'll look into the issues you highlighted above and start posting
patches soon. Thanks for reviewing!

Guys,

sorry for being a little bit unresponsive on this matter. I just moved
across the Atlantic (again :-P ) and am in a rough setup for a few
days (hopefully not weeks). I don't currently have any Synaptics RMI4
over SMBus device with me, but Lyude (AKA Chandler) does have access
to the ones we have in Westford.

In addition to the patches I published in this series (the ones Andrew
has), our latest tree
(https://github.com/bentiss/linux/commits/synaptics-rmi4-smbus-v4.3-rc6%2B)
contains the SMBus support and some attempts to fix suspend/resume
with SMBus touchpads.

The problem we have with suspend/resume on SMBus is that the device is
both seen as a PS/2 device (which we unbind) and the SMBus one.
Problem is, during resume, the serial port is reset, which disable the
SMBus part.

Our solution currently consists in disabling the PM resume function
from the rmi_driver code, and export it through an internal API. Once
the low level driver receive the resume callback (either from the PM
code or the serial driver which just busted our init through its own
reset), this low level driver calls the rmi_driver resume function
which does all the RMI4 resume handling.

I think this should be needed for i2c-hid devices too given that most
of them also enumerates as PS/2.

I always refrained to send this resume fixes because we were not able
to test the rmi4_i2c driver as it doesn't register itself with the PM
subsystem.

These are hopefully the only changes that needs to be done on the
rmi4_core part for having rmi_smbus working.

Cheers,
Benjamin