Thierry Reding wrote at Saturday, March 03, 2012 3:54 PM:

* Stephen Warren wrote:

quoted

Thierry Reding wrote at Wednesday, February 22, 2012 8:17 AM:

quoted

Add auxdata to instantiate a device tree for the PWFM controller and
include a corresponding node in the device tree.

quoted

diff --git a/arch/arm/boot/dts/tegra30.dtsi b/arch/arm/boot/dts/tegra30.dtsi

quoted

+	pwm: pwm@7000a000 {
+		compatible = "nvidia,tegra20-pwm";
+		reg = <0x7000a000 0x100>;
+		#pwm-cells = <2>;
+	};

The compatible value probably should list both Tegra30 and Tegra20, so
we can know exactly which HW is present, just in case we need to turn
on some bug-fix only for one of the variants:

    compatible = "nvidia,tegra30-pwm", "nvidia,tegra20-pwm";

I'm confused. If I know exactly that the hardware is Tegra30 (which it
definitely should be if I include tegra30.dtsi), then why list "tegra20-pwm"
as compatible?

Or did you mean to list tegra30-pwm as compatible value in the PWM driver?

Standard practice is to list the exact model of the HW as the first entry
in compatible in the .dts file:

    nvidia,tegra30-pwm

This is so that the DT always describes exactly which HW model is actually
present, so that if HW-model-specific WARs/... are required in the future,
the DT already lists that information up-front.

Then additionally list any older HW models that this HW is also compatible
with:

    nvidia,tegra20-pwm


This allows the driver to list just nvidia,tegra20-pwm but still bind
to DT nodes that are for later HW.

So, in other words, you end up with the following in the .dts/.dtsi file:

    compatible = "nvidia,tegra30-pwm", "nvidia,tegra20-pwm";

quoted

Could you also write binding documentation, in particular explaining
what the two pwm-cells are specifically for Tegra:

Documentation/devicetree/bindings/pwm/nvidia,tegra20-pwm.txt

(although perhaps that'd be part of the previous patch which implements
the driver)

Actually for Tegra the values would be those documented in the generic
binding because Tegra uses of_pwm_simple_xlate(). Does it still make sense to
add a Tegra-specific binding?

There should still be a Tegra-specific binding. Without it, there's no
definite way to know whether the "standard" properties actually apply,
or someone simply forgot to document it.

-- 
nvpublic