Quoting Brian Masney (2020-01-07 04:03:17)

On Sun, Jan 05, 2020 at 12:35:33AM -0800, Stephen Boyd wrote:

quoted

Quoting Brian Masney (2019-12-04 16:25:00)

quoted

+examples:
+  - |
+    #include <dt-bindings/clock/qcom,mmcc-msm8974.h>
+    #include <dt-bindings/gpio/gpio.h>
+
+    vibrator {
+        compatible = "clk-vibrator";
+
+        vcc-supply = <&pm8941_l19>;
+
+        clocks = <&mmcc CAMSS_GP1_CLK>;
+        clock-names = "core";
+        clock-frequency = <24000>;
+
+        enable-gpios = <&msmgpio 60 GPIO_ACTIVE_HIGH>;
+
+        pinctrl-names = "default";
+        pinctrl-0 = <&vibrator_pin>;

I'm still trying to wrap my head around this. I think we can have a pwm
provider in a clk controller node (so imagine &mmcc has #pwm-cells) and
then this 'clk-vibrator' binding wouldn't exist? Instead we would have
some sort of binding for a device that expects a pwm and whatever else
is required, like the enable gpio and power supply. Is there an actual
hardware block that is this way? Does it have a real product id and is
made by some company? Right now this looks a little too generic to not
just be a catch-all for something that buzzes.

So have some of the Qualcomm clocks like this one register with both the
clk and the pwm frameworks? I feel that approach would better represent
the hardware in device tree.

That is one option. Or another option would be to have another node that
"adapts" a clk signal to a pwm provider. Similar to how we adapt a gpio
to make a clk gate or mux. Something like:

	gcc: clock-controller@f00d {
		reg = <0xf00d 0xd00d>;
		#clock-cells = <1>;
	};


	pwm {
		compatible = "pwm-clk";
		#pwm-cells = <0>;
		clocks = <&gcc 45>;
		assigned-clocks = <&gcc 45>;
		assigned-clock-rates = <1400000>;
	};

And then the pwm-clk driver would adjust the duty cycle to generate a
pwm.

If we did that, then the pwm-vibra driver in the input subsystem could
be used.