Re: [PATCH v2 1/8] dt-bindings: pinctrl: Add thead,th1520-pinctrl bindings

[Emil Renner Berthing <emil.renner.berthing@xxxxxxxxxxxxx>]

Rob Herring wrote:
> On Fri, May 17, 2024 at 07:48:17AM -0500, Emil Renner Berthing wrote:
> > Rob Herring wrote:
> > > On Wed, Jan 03, 2024 at 02:28:38PM +0100, Emil Renner Berthing wrote:
> > > > Add bindings for the pin controllers on the T-Head TH1520 RISC-V SoC.
> > > >
> > > > Signed-off-by: Emil Renner Berthing <emil.renner.berthing@xxxxxxxxxxxxx>
> > > > ---
> > > >  .../pinctrl/thead,th1520-pinctrl.yaml         | 372 ++++++++++++++++++
> > > >  1 file changed, 372 insertions(+)
> > > >  create mode 100644 Documentation/devicetree/bindings/pinctrl/thead,th1520-pinctrl.yaml
> > > >
> > > > diff --git a/Documentation/devicetree/bindings/pinctrl/thead,th1520-pinctrl.yaml b/Documentation/devicetree/bindings/pinctrl/thead,th1520-pinctrl.yaml
> > > > new file mode 100644
> > > > index 000000000000..d3ad7a7cfdd1
> > > > --- /dev/null
> > > > +++ b/Documentation/devicetree/bindings/pinctrl/thead,th1520-pinctrl.yaml
> > > > @@ -0,0 +1,372 @@
> > > > +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
> > > > +%YAML 1.2
> > > > +---
> > > > +$id: http://devicetree.org/schemas/pinctrl/thead,th1520-pinctrl.yaml#
> > > > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > > +
> > > > +title: T-Head TH1520 SoC pin controller
> > > > +
> > > > +maintainers:
> > > > +  - Emil Renner Berthing <emil.renner.berthing@xxxxxxxxxxxxx>
> > > > +
> > > > +description: |
> > > > +  Pinmux and pinconf controller in the T-Head TH1520 RISC-V SoC.
> > > > +
> > > > +  The TH1520 has 3 groups of pads each controlled from different memory ranges.
> > > > +  Confusingly the memory ranges are named
> > > > +    PADCTRL_AOSYS  -> PAD Group 1
> > > > +    PADCTRL1_APSYS -> PAD Group 2
> > > > +    PADCTRL0_APSYS -> PAD Group 3
> > > > +
> > > > +  Each pad can be muxed individually to up to 6 different functions. For most
> > > > +  pads only a few of those 6 configurations are valid though, and a few pads in
> > > > +  group 1 does not support muxing at all.
> > > > +
> > > > +  Pinconf is fairly regular except for a few pads in group 1 that either can't
> > > > +  be configured or has some special functions. The rest have configurable drive
> > > > +  strength, input enable, schmitt trigger, slew rate, pull-up and pull-down in
> > > > +  addition to a special strong pull up.
> > > > +
> > > > +  Certain pads in group 1 can be muxed to AUDIO_PA0 - AUDIO_PA30 functions and
> > > > +  are then meant to be used by the audio co-processor. Each such pad can then
> > > > +  be further muxed to either audio GPIO or one of 4 functions such as UART, I2C
> > > > +  and I2S. If the audio pad is muxed to one of the 4 functions then pinconf is
> > > > +  also configured in different registers. All of this is done from a different
> > > > +  AUDIO_IOCTRL memory range and is left to the audio co-processor for now.
> > >
> > > It is still not clear to me if each instance is a different programming
> > > model or the same with just different connections. The latter should
> > > be the same compatible string. That needs to be answered in *this*
> > > patch, not a reply.
> >
> > Hi Rob,
> >
> > Sorry for the late response. I honestly don't know exactly what you mean by
> > differenty programming models and what the difference is, so I'll need a bit of
> > help with what you want me to write here.
>
> Is the register interface of each instance the same? Looks like it is
> from the driver. So normally that's 3 instances of the same compatible.
>
> > Any driver for the TH1520 SoC (not just Linux) would need some way to discern
> > between the 3 pin controllers so they know how many pins to control and what
> > pinmux settings are valid. Basically they'd need the data in the three
> > th1520_group{1,2,3}_pins arrays in the driver and a way to know which of them
> > to use.
> >
> > https://lore.kernel.org/linux-riscv/20240103132852.298964-3-emil.renner.berthing@xxxxxxxxxxxxx/
>
> Why do you need to know how many pins? The DT says configure a pin and
> you just configure it. It's not the kernel's job to validate that the DT
> is correct.
>
> Aren't the pin names globally unique? So you just look up the pin name
> across all the arrays. Or you can just look up one pin from each
> instance to find which th1520_groupN_pins array goes with the instance.
> Or just have 1 array.

Just to be clear, do you mean to add just one node for all 3 instances like
this?

padctrl: pinctrl@ffe7f3c000 {
    compatible = "thead,th1520-pinctrl";
    reg = <0xff 0xe7f3c000 0x0 0x1000>,
          <0xff 0xec007000 0x0 0x1000>,
          <0xff 0xfff4a000 0x0 0x2000>;
    reg-names = "group2", "group3", "group1";
    clocks = <&apb_clk>, <&apb_clk>, <&aonsys_clk>;
    clock-names = "group2", "group3", "group1";
};

That will work since we can then register all pins at the same time. But I
can't see how it will work when keeping the 3 different nodes since each
instance needs to register the pins they control at probe time. So there will
need to be some way to tell the 3 different nodes apart.

/Emil