On Wed, May 20, 2020 at 11:25:40PM +0200, Sebastian Reichel wrote:

Hi,

On Wed, May 20, 2020 at 10:49:52AM -0700, Dmitry Torokhov wrote:

quoted

Hi Sebastian,

On Wed, May 20, 2020 at 05:39:35PM +0200, Sebastian Reichel wrote:

quoted

Expose model and fw_version via sysfs. Also query the model
in probe to make sure, that the I2C communication with the
device works before successfully probing the driver.

This is a bit complicated, since EETI devices do not have
a sync interface. Sending the commands and directly reading
does not work. Sending the command and waiting for some time
is also not an option, since there might be touch events in
the mean time.

Last but not least we do not cache the results, since this
interface can be used to check the I2C communication is still
working as expected.

Signed-off-by: Sebastian Reichel <redacted>
---
 .../ABI/testing/sysfs-driver-input-exc3000    |  15 ++
 drivers/input/touchscreen/exc3000.c           | 145 +++++++++++++++++-
 2 files changed, 159 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/ABI/testing/sysfs-driver-input-exc3000

diff --git a/Documentation/ABI/testing/sysfs-driver-input-exc3000 b/Documentation/ABI/testing/sysfs-driver-input-exc3000
new file mode 100644
index 000000000000..d79da4f869af
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-driver-input-exc3000

@@ -0,0 +1,15 @@
+What:		/sys/class/input/inputX/fw_version
+Date:		May 2020
+Contact:	linux-input@vger.kernel.org
+Description:	Reports the firmware version provided by the touchscreen, for example "00_T6" on a EXC80H60
+
+		Access: Read
+		Valid values: Represented as string
+
+What:		/sys/class/input/inputX/model
+Date:		May 2020
+Contact:	linux-input@vger.kernel.org
+Description:	Reports the model identification provided by the touchscreen, for example "Orion_1320" on a EXC80H60
+
+		Access: Read
+		Valid values: Represented as string

These are properties of the controller (i2c device), not input
abstraction class on top of it, so the attributes should be attached to
i2c_client instance.

Please use devm_device_add_group() in probe to instantiate them at the
proper level.

As written in the cover letter using devm_device_add_group() in
probe routine results in a udev race condition:

http://kroah.com/log/blog/2013/06/26/how-to-create-a-sysfs-file-correctly/

This race has been solved with the addition of KOBJ_BIND/KOBJ_UNBIND
uevents that signal when driver is bound or unbound from the device.
Granted, current systemd/udev drops them as it does not know how to
"add" to the device state, but this is on systemd to solve.

Thanks.

-- 
Dmitry