On Tue, Sep 18, 2018 at 01:25:17PM +0200, Vincent Whitchurch wrote:

On Sun, Sep 16, 2018 at 06:24:52PM -0500, Rob Herring wrote:

quoted

On Thu, Sep 06, 2018 at 02:47:23PM +0200, Linus Walleij wrote:

quoted

On Thu, Sep 6, 2018 at 1:09 PM Bartosz Golaszewski [off-list ref] wrote:

quoted

Do we even need DT bindings for this? Device tree bindings mean that
we commit to a stable interface so that once a device with given DT
blob is released, it will work on every future kernel. Also: DT should
only include information about actual HW, not operating system
concepts. Meanwhile this is for testing purposes only and it won't end
up in any actual DT source file upstream.

Hm that relates to another discussion I have with the DT maintainers
about a virtualized display panel.

I do not know how the DT maintainers feel about supporting things
in the kernel that uses DT infrastructure and specially tailored
device trees but include elements with no formal device tree
bindings. Would be interesting to hear their thoughts on this.

I have one usecase in mind where I think it is valid, but for this case 
in particular, I think it would be better to just provide a sysfs 
interface. If there's no GPIO binding connections to this mockup 
controller, there's no real need for DT. OTOH, if this device allowed 
building a test framework for all the other GPIO based drivers and 
bindings such as LED, PWM, bitbanged buses, etc.

My original usecase for this was for leds-gpio.  I used the DT bindings
added by this patch and hooked up a leds-gpio via DT in order to develop
userspace.  Once the hardware with the GPIO expander arrived, the mockup
device in the DT was replaced with the expander.  I don't see how a
sysfs interface would allow the same thing.

Okay, I'm fine with the binding. I'm also okay with not documenting this 
at all as this should never be in any stable or upstream dts file.

Rob