On Sun, Jan 22, 2017 at 11:10:31AM +0100, Hans de Goede wrote:

Hi,

On 22-01-17 11:00, Dmitry Torokhov wrote:

quoted

On Sun, Jan 22, 2017 at 12:49 AM, Hans de Goede [off-list ref] wrote:

quoted

Hi,

On 21-01-17 20:13, Dmitry Torokhov wrote:

quoted

On Mon, Jan 09, 2017 at 06:57:06PM +0100, Hans de Goede wrote:

quoted

On some tablets using the soc_button_array driver the buttons do not
follow the standard home, power, volume_up, volume_down, rotation_lock
button order as published by Microsoft.

We can use the existing udev hwdb mechanism to fix this up, but then
the created devices must have a unique name, therefor this commit adds
a unique name for the 2 created gpio-keys input devices.

Why does it have to have unique name? You should be able to match on
other input device properties, for example ATTR{capabilities/ev} or
ATTR{capabilities/keys} to identify the device you want to adjust.

hwdb entries do not have access to full udev data, basically there
are 2 match formats:

# Supported hardware matches are:
#  - Generic input devices match:
#      evdev:input:bZZZZvYYYYpXXXXeWWWW-VVVV
#    This matches on the kernel modalias of the input-device, mainly:
#    ZZZZ is the bus-id (see /usr/include/linux/input.h BUS_*), YYYY, XXXX
and
#    WWW are the 4-digit hex uppercase vendor, product and version ID and
VVVV
#    is an arbitrary length input-modalias describing the device
capabilities.
#    The vendor, product and version ID for a device node "eventX" is listed
#    in /sys/class/input/eventX/device/id.
#
#  - Input driver device name and DMI data match:
#      evdev:name:<input device name>:dmi:bvn*:bvr*:bd*:svn<vendor>:pn*
#    <input device name> is the name device specified by the
#    driver, <vendor> is the firmware-provided string exported
#    by the kernel DMI modalias, see /sys/class/dmi/id/modalias


Since we want to match on DMI info we need to use the second, and
the info you are referring to is not available here.

Well, you can either teach hwdb new tricks or mangle the name in udev
rule. As far as I can see the original invocation is:

# device matching the input device name and the machine's DMI data
KERNELS=="input*", IMPORT{builtin}="hwdb
'evdev:name:$attr{name}:$attr{[dmi/id]modalias}'", \
RUN{builtin}+="keyboard", GOTO="evdev_end"

You can add a similar rule that also looks at ATTR{whatever}, but
instead of using "name:$attr{name}" you can use whatever string you
want.

There is no need to change kernel, it already exports all necessary data.

Ah, come one, requiring a custom udev rule for this is a pain, where as

OK, do not require custom property. Provide an extended one:

# device matching the input device properties and the machine's DMI data
KERNELS=="input*", IMPORT{builtin}="hwdb
'evdev:name:$attr{name}:phys:$attr{phys}:ev:$attr{capabilities/ev}:$attr{[dmi/id]modalias}'", \
  RUN{builtin}+="keyboard"

and either migrate old keymaps to the extended one, or keep both. In the
end, there is nothing special in 'evdev:name:...' prefix, it matches
across all lines of hwdb with whatever is provided in the rule.

this is really easy to fix on the kernel side.

Even if changing the kernel seems most convenient for you it does not
mean it is the right approach.

Besides that, the way soc_button_array works is that we currently end
up with 2 identical named (gpio_keys) input devices which is all sorts
of inconvenient, e.g. during testing the setkeycode support I had
to guess which was which when invoking evemu-record to test, they
will have the same name in "xinput list", etc.

Input device names really should be unique where ever possible, for
various reasons, and here we can easily make them unique.

We never provided any guarantees about uniqueness of names of input
devices and I do not think we should start now. If you plug 2 same USB
keyboards you'll get 2 devices with same names. Same goes for mice,
tablets, etc. I do not see why SOC buttons should be different.

Thanks.

-- 
Dmitry