Am Dienstag, 23. Februar 2016, 14:17:17 CET schrieb Arnd Bergmann:

On Tuesday 23 February 2016 11:38:52 Marc Dietrich wrote:

quoted

Am Dienstag, 23. Februar 2016, 11:31:40 CET schrieb Arnd Bergmann:

quoted

On Tuesday 23 February 2016 11:15:31 Arnd Bergmann wrote:

quoted

On Thursday 18 February 2016 21:04:49 Johannes Berg wrote:

quoted

On Tue, 2016-01-26 at 09:42 +0100, Johannes Berg wrote:

Just for my curiosity: what is the difference between a rfkill-gpio
device and a gpio-keys device with a KEY_RFKILL code?

arch/arm/boot/dts/bcm4708-netgear-r6250.dts and others seem to
do the second approach in DT so they don't need to create the
platform device.

I found the answer now (after discussing on IRC): just
for reference: KEY_RFKILL is for sending the event to the kernel
when a user presses the gpio butting, this rfkill-gpio turns
the devices on or off when after an RFKILL event is received.

yes, paz00 has no hw key. rfkill is triggered via software (rfkill) only.
The problem in the past was how to describe such a device in the device
tree.

Strictly speaking, there is no rfkill device hardware (just an interface).
The only hardware that exists is in our case is connected to a hard wired
usb bus (wifi dongle), which makes it complicated to attach a DT property
or subdevice to it. And a standalone device-tree entry for a non hardware
device was rejected in the past.

Ah, so the problem of attaching DT properties to a USB device has recently
been solved, see subject "USB: core: let USB device know device node".

Would that work for you? With this, the USB driver can simply look at
the optional DT properties of the USB function to implement its rfkill
callbacks.

oh, that looks indeed interesting. The question is now if rfkill is a property 
of a device or a subdevice itself. The latter one would only require addition 
of device-tree instantiation of rfkill, while with the former one, the (or 
all) usb driver(s) need to be modified to accept device tree properties, 
especially gpios. The driver would then be responsible to add an rfkill 
"device".

IMHO (and unfortunately), it's just a property (a way to specify the relevant 
gpios), making the solution again hard to archive.

Marc