On Fri, 6 Mar 2020 08:43:45 +0100
Michal Vokáč <michal.vokac@xxxxxxxxx> wrote:
> On 05. 03. 20 14:36, Uwe Kleine-König wrote:
> > Hello,
> >
> > On Thu, Mar 05, 2020 at 01:22:32PM +0000, Paul Barker wrote:
> >> I recently ran into an issue using the pwm-fan driver with an inverted
> >> PWM output on iMX6.
> >>
> >> The fan is defined in the device tree as follows:
> >>
> >> fan0: pwm-fan {
> >> compatible = "pwm-fan";
> >> pwms = <&pwm2 0 25000 PWM_POLARITY_INVERTED>;
> >> ...
> >> }
> >>
> >> In pwm_imx27_probe() the support for a third `flags` argument in a pwm
> >> reference is enabled:
> >>
> >> imx->chip.of_xlate = of_pwm_xlate_with_flags;
> >> imx->chip.of_pwm_n_cells = 3;
> >>
> >> However, the flag is ignored and the output is not inverted.
> >>
> >> By adding some prints I saw that when of_pwm_xlate_with_flags() is
> >> called, args->args_count is 2 instead of 3.
> >>
> >> Looking at the definition of the pwm device itself in imx6qdl.dtsi I
> >> can see that the number of cells in a pwm reference is set to 2 not 3:
> >>
> >> pwm2: pwm@2084000 {
> >> #pwm-cells = <2>;
> >> ...
> >> };
> >>
> >> That seems to be preventing a third argument from being passed.
> >>
> >> I can change `#pwm-cells` to <3> and then everything works for my
> >> device but I'm not sure that is the correct solution for everyone. That
> >> would require all pwm references on iMX6 devices to use 3 cells. The
> >> code in of_pwm_xlate_with_flags() seems to be built to handle either 2
> >> or 3 argument cells but I can't see any way to allow this choice in the
> >> device tree.
> >>
> >> If the solution is to set `#pwm-cells` to <3> I'm happy to send a patch
> >> which does this and updates all pwm references in device trees which
> >> include `imx6dql.dtsi`. Before I do that I'd like to know that it's the
> >> correct approach though.
> >>
> >> For context I've confirmed this is the case in Linux 5.4 and that the
> >> relevant files haven't changed between that release and 5.6.0-rc4.
> >
> > I think changing that is fine. However you'd have to care that all
> > in-tree users that rely on #pwm-cells = <2> are fixed accordingly.
> >
> > I'd do: add #pwm-cells = <3> in the cpu.dtsi and then adapt all
> > machine.dts to add #pwm-cells = <2> until there are no more changes to
> > the generated files compared to the current state.
>
> I solved that in the past on our board the same way as other imx6 boards
> did. Just override the #pwm-cells property in your board specific devicetree
> and you are fine:
>
> linux-src$ git grep pwm-cells -- arch/arm/boot/dts/imx6*-*
> arch/arm/boot/dts/imx6dl-yapp4-common.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6q-display5.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6q-mccmon6.dts: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6qdl-tx6.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6qdl-tx6.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6ul-tx6ul.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6ull-colibri.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6ull-colibri.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6ull-colibri.dtsi: #pwm-cells = <3>;
> arch/arm/boot/dts/imx6ull-colibri.dtsi: #pwm-cells = <3>;
Michal,
That approach would be good enough for my limited use case but it doesn't
prevent others from hitting the same confusion in the future. I'd rather
fix the base imx6 dtsi files to have #pwm-cells = <3> if the driver supports
the third cell containing flags.
I'll send a patch once I've done a bit more investigation.
Thanks,
--
Paul Barker
Konsulko Group
