Re: [PATCH v6 04/15] dt-bindings: add BCM6328 pincontroller binding documentation

[Rob Herring <robh+dt@kernel.org>]

On Wed, Mar 10, 2021 at 12:10 PM Álvaro Fernández Rojas
<noltari@gmail.com> wrote:
>
> Hi Rob,
>
> > El 10 mar 2021, a las 19:45, Rob Herring <robh+dt@kernel.org> escribió:
> >
> > On Wed, Mar 10, 2021 at 11:03 AM Álvaro Fernández Rojas
> > <noltari@gmail.com> wrote:
> >>
> >> Hi Rob,
> >>
> >>> El 10 mar 2021, a las 18:45, Rob Herring <robh+dt@kernel.org> escribió:
> >>>
> >>> On Wed, Mar 10, 2021 at 5:55 AM Álvaro Fernández Rojas
> >>> <noltari@gmail.com> wrote:
> >>>>
> >>>> Add binding documentation for the pincontrol core found in BCM6328 SoCs.
> >>>>
> >>>> Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
> >>>> Co-developed-by: Jonas Gorski <jonas.gorski@gmail.com>
> >>>> Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
> >>>> ---
> >>>> v6: add changes suggested by Rob Herring
> >>>> v5: change Documentation to dt-bindings in commit title
> >>>> v4: no changes
> >>>> v3: add new gpio node
> >>>> v2: remove interrupts
> >>>>
> >>>> .../pinctrl/brcm,bcm6328-pinctrl.yaml         | 174 ++++++++++++++++++
> >>>> 1 file changed, 174 insertions(+)
> >>>> create mode 100644 Documentation/devicetree/bindings/pinctrl/brcm,bcm6328-pinctrl.yaml
> >>>>
> >>>> diff --git a/Documentation/devicetree/bindings/pinctrl/brcm,bcm6328-pinctrl.yaml b/Documentation/devicetree/bindings/pinctrl/brcm,bcm6328-pinctrl.yaml
> >>>> new file mode 100644
> >>>> index 000000000000..471f6efa1754
> >>>> --- /dev/null
> >>>> +++ b/Documentation/devicetree/bindings/pinctrl/brcm,bcm6328-pinctrl.yaml
> >>>> @@ -0,0 +1,174 @@
> >>>> +# SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
> >>>> +%YAML 1.2
> >>>> +---
> >>>> +$id: http://devicetree.org/schemas/pinctrl/brcm,bcm6328-pinctrl.yaml#
> >>>> +$schema: http://devicetree.org/meta-schemas/core.yaml#
> >>>> +
> >>>> +title: Broadcom BCM6328 pin controller
> >>>> +
> >>>> +maintainers:
> >>>> +  - Álvaro Fernández Rojas <noltari@gmail.com>
> >>>> +  - Jonas Gorski <jonas.gorski@gmail.com>
> >>>> +
> >>>> +description: |+
> >>>> +  The pin controller node should be the child of a syscon node.
> >>>> +
> >>>> +  Refer to the the bindings described in
> >>>> +  Documentation/devicetree/bindings/mfd/syscon.yaml
> >>>> +
> >>>> +properties:
> >>>> +  compatible:
> >>>> +    const: brcm,bcm6328-pinctrl
> >>>> +
> >>>> +  gpio:
> >>>> +    type: object
> >>>> +    properties:
> >>>> +      compatible:
> >>>> +        const: brcm,bcm6328-gpio
> >>>> +
> >>>> +      data:
> >>>> +        $ref: /schemas/types.yaml#/definitions/uint32
> >>>> +        description: |
> >>>> +          Offset in the register map for the data register (in bytes).
> >>>> +
> >>>> +      dirout:
> >>>> +        $ref: /schemas/types.yaml#/definitions/uint32
> >>>> +        description: |
> >>>> +          Offset in the register map for the dirout register (in bytes).
> >>>> +
> >>>> +      gpio-controller: true
> >>>> +
> >>>> +      "#gpio-cells":
> >>>> +        const: 2
> >>>> +
> >>>> +      gpio-ranges:
> >>>> +        maxItems: 1
> >>>> +
> >>>> +    required:
> >>>> +      - gpio-controller
> >>>> +      - gpio-ranges
> >>>> +      - '#gpio-cells'
> >>>> +
> >>>> +    additionalProperties: false
> >>>> +
> >>>> +patternProperties:
> >>>> +  '^.*-pins$':
> >>>> +    if:
> >>>> +      type: object
> >>>> +    then:
> >>>> +      properties:
> >>>> +        function:
> >>>> +          $ref: "pinmux-node.yaml#/properties/function"
> >>>> +          enum: [ serial_led_data, serial_led_clk, inet_act_led, pcie_clkreq,
> >>>> +                  led, ephy0_act_led, ephy1_act_led, ephy2_act_led,
> >>>> +                  ephy3_act_led, hsspi_cs1, usb_device_port, usb_host_port ]
> >>>> +
> >>>> +        pins:
> >>>> +          $ref: "pinmux-node.yaml#/properties/pins"
> >>>> +          enum: [ gpio6, gpio7, gpio11, gpio16, gpio17, gpio18, gpio19,
> >>>> +                  gpio20, gpio25, gpio26, gpio27, gpio28, hsspi_cs1,
> >>>> +                  usb_port1 ]
> >>>> +
> >>>> +required:
> >>>> +  - compatible
> >>>> +  - gpio
> >>>> +
> >>>> +additionalProperties: false
> >>>> +
> >>>> +examples:
> >>>> +  - |
> >>>> +    gpio_cntl@10000080 {
> >>>> +      compatible = "brcm,bcm6328-gpio-controller", "syscon", "simple-mfd";
> >>>
> >>> You just added "brcm,bcm6328-gpio-controller", it would need to be documented.
> >>
> >> I just added that because you requested me to do it ¯\_(ツ)_/¯
> >
> > I said 'syscon' by itself was not allowed, then asked about the multiple levels.
>
> Why not?

Because 'syscon' alone doesn't mean anything and doesn't describe what
registers it contains. You need something that says this is the XYZ
block in the ABC SoC.

> What if you have several controllers inside a syscon?

You either just add properties (e.g. just add #clock-cells and it's a
clock provider) or you have child nodes. Which one you do generally
depends on if the providers have DT resources themselves.

> The root should also have “something" in it?
>
> >
> >> What should I do to document it?
> >> I still don’t get most of this .yaml stuff...
> >>
> >>>
> >>>> +      reg = <0x10000080 0x80>;
> >>>> +
> >>>> +      pinctrl: pinctrl {
> >>>> +        compatible = "brcm,bcm6328-pinctrl";
> >>>> +
> >>>> +        gpio {
> >>>> +          compatible = "brcm,bcm6328-gpio";
> >>>
> >>> I'm still trying to understand why you need 3 levels of nodes here?
> >>> The gpio controller contains a pin controller plus other undefined
> >>> functions (because of 'syscon') and the pin controller contains a gpio
> >>> controller?
> >>
> >> In previous versions the gpio controller was registered along with the pin controller, but @Linus requested me to register the gpio pin controller ranges through device tree by using gpio-ranges and I decided to use this approach, which was already used by other pin controllers.
> >> However, there aren’t any pinctrl drivers using gpio-regmap, so this is kind of new…
> >>
> >>>
> >>> I think "brcm,bcm6328-gpio-controller" and "brcm,bcm6328-pinctrl"
> >>> should be a single node.
> >>
> >> I agree, but does it make sense to add gpio-ranges to a pinctrl node referencing itself?
> >
> > It wouldn't be. I wasn't saying the pinctrl and gpio controller are
> > the same node. My suggestion was combining syscon and pinctrl.
>
> But that wouldn’t be correct if there were more “things” inside the syscon, right?

Right.

> >> Something like:
> >> syscon {
> >
> > Again with the syscon. If pinctrl and GPIO are the only functions
> > within this h/w block, then this is not a syscon. You are just abusing
> > that having 'syscon' compatible means you get a regmap created
> > automagically for you. Nothing here looks like a 'system controller'
> > to me. A 'system controller' is a random collection of register bits
> > with functions that don't fit anywhere else.
>
> pinctrl and GPIO aren’t the only functions within this HW block.
> Maybe I didn’t document/code it properly, but I’m sure I’m not abusing what a system controller is.

Okay, that's the detail missing from this patch.

> Please, take a look at http://www.datashed.science/misc/bcm/gpl/broadcom-sdk-416L05/shared/opensource/include/bcm963xx/6328_map_part.h:
> typedef struct GpioControl {
>     uint32      GPIODirHi;                  /* 0 */
>     uint32      GPIODir;                    /* 4 */
>     uint32      GPIOioHi;                   /* 8 */
>     uint32      GPIOio;                     /* C */
>     uint32      unused0;                    /* 10 */
>     uint32      SpiSlaveCfg;                /* 14 */
>     uint32      GPIOMode;                   /* 18 */
>     uint64      PinMuxSel;                  /* 1C */
>     uint32      PinMuxSelOther;             /* 24 */
>     uint32      TestControl;                /* 28 */
>     uint32      unused2;                    /* 2C */
>     uint32      RoboSWLEDControl;           /* 30 */
>     uint32      RoboSWLEDLSR;               /* 34 */
>     uint32      unused3;                    /* 38 */
>     uint32      RoboswEphyCtrl;             /* 3C */
>     uint32      RoboswSwitchCtrl;           /* 40 */
>     uint32      RegFileTmCtl;               /* 44 */
>     uint32      RingOscCtrl0;               /* 48 */
>     uint32      RingOscCtrl1;               /* 4C */
>     uint32      unused4[6];                 /* 50 - 64 */
>     uint32      DieRevID;                   /* 68 */
>     uint32      unused5;                    /* 6c */
>     uint32      DiagSelControl;             /* 70 */
>     uint32      DiagReadBack;               /* 74 */
>     uint32      DiagReadBackHi;             /* 78 */
>     uint32      DiagMiscControl;            /* 7c */
> } GpioControl;
>
> So we’re using GPIODirHi, GPIODir, GPIOioHi and GPIOio registers for GPIO regmap driver.
> And we’re using GPIOMode, PinMuxSel (u64 -> x2 u32), PinMuxSelOther for pinctrl driver.
> And this is for BCM6328, but some of the other SoCs are even more scattered.

So based on this I'd do something like this:

syscon {
  reg = <base 0x80>;
  ranges = <0 base 0x80>;
  pinctrl@18 {
    reg = <0x18 0x10>;
    foo-pins {};
  gpio@0 {
    reg = <0x0 0x10>;
  };
};

If things are more scattered within gpio or pinctrl, then maybe you
need multiple reg entries. Whether the OS uses 'reg' and mmio or a
regmap from the syscon is up to you. That's independent of the
binding.

> >>        pinctrl: pinctrl {
> >>                compatible …
> >>
> >>                gpio-controller;
> >>                gpio-ranges = <&pinctrl 0 0 32>;
> >>                #gpio-cells = <2>;
> >
> > I was assuming you have multiple GPIO controllers within 1 pinctlr?
> > The pinctrl and gpio could be a single node like above if there's only
> > 1 GPIO controller. But I'm still somewhat guessing what the h/w looks
> > like because I have to go searching thru the driver to decipher.
> > Please describe the h/w in the binding.
>
> GPIO dirout and data rely on 2x u32 registers or a single u64 register.
> This is can be either be implemented as a single GPIO controller, or as 2 separate GPIO controllers.
> However, since I’m overriding reg_mask_xlate with bcm63xx_reg_mask_xlate I can register it as a single GPIO controller, which makes more sense to me.

I think 1 makes more sense.

Rob