Hi Daniel,

On Fri, 2015-07-03 at 18:46 +0800, Daniel Kurtz wrote:

On Fri, Jul 3, 2015 at 7:40 AM, Stephen Boyd [off-list ref] wrote:

quoted

DT nodes typically have a reg property. Not having a reg property is a
good indicator of a problem with the binding. A syscon is used when you
have a DT node with a reg property and some driver attached to it, but
you need to poke some bits in another register region that isn't part of
the reg property. Instead of having multiple nodes with two reg
properties where the second one is the same, we use a phandle and a syscon.

If clock-controller isn't acceptable maybe clock-reset-contoller would
work? Or "power-controller"? We certainly shouldn't be making up
multiple nodes for one hardware block. Of course, the subject of the
patch is "bindings for clock controllers", so it may be that the
registers are predominantly clock related and so the name is appropriate
already.

Using "clock-controller" seems to fit best with the bindings
introduced by this patch.

However, if these bindings are for hardware blocks that contain a grab
bag of various functionality that will be added in later patches, then
I think "syscon" might be best.

I referred some dt-binding documents that contain clock and reset
controller in the same unit:

qcom,gcc.txt:

	clock-controller@900000 {
		compatible = "qcom,gcc-msm8960";

nvidia,tegra124-car.txt:

	tegra_car: clock {
		compatible = "nvidia,tegra124-car";

rockchip,rk3188-cru.txt:

	cru: cru@20000000 {
		compatible = "rockchip,rk3188-cru";

Rockchip use "Clock and Reset Unit" as their controller's name. The
other two use "clock" or "clock-controller" as their controller's name.

It looks like "clock-controller" is also an acceptable name for
clock/reset controllers. So I would like to keep this name in next
patch.


Best regards,

James