Quoting Gregory CLEMENT (2018-11-23 07:02:56)

Hi Stephen,
 
 On ven., oct. 12 2018, Stephen Boyd [off-list ref] wrote:

quoted

+Rob

Quoting Gregory CLEMENT (2018-09-22 11:17:09)

quoted

diff --git a/arch/arm64/boot/dts/marvell/armada-ap806.dtsi b/arch/arm64/boot/dts/marvell/armada-ap806.dtsi
index 4a65e4e830aa..27c840e91abe 100644
--- a/arch/arm64/boot/dts/marvell/armada-ap806.dtsi
+++ b/arch/arm64/boot/dts/marvell/armada-ap806.dtsi

@@ -280,6 +280,12 @@
                                #address-cells = <1>;
                                #size-cells = <1>;
 
+                               cpu_clk: clock-cpu {
+                                       compatible = "marvell,ap806-cpu-clock";
+                                       clocks = <&ap_clk 0>, <&ap_clk 1>;
+                                       #clock-cells = <1>;
+                               };

This looks like the wrong place because there isn't a reg property. It

There is no reg property because we are inside a syscon node where the
registers are shared between multiple IPs.

The node right after this node has a reg property. What's going on?

quoted

should go to the root of the tree. And then it looks like we're adding
something to DT to get a driver to probe, which is improper DT design.

There is nothing related to the driver, this subnode describes the way
the hardware is designed. Under the system controller node there are
several IPs , like the CPU clocks, but also the GPIO or the pinctrl.

Ok. And some of them have reg properties and others don't? We should
somehow deprecate the idea of a syscon with child nodes. It leads to
this weird twisting of DT. The only use for syscon from my perspective
is when we have a register region that other random devices with their
own 'reg' property need to peek/poke random things into.