Re: [PATCH v4 06/10] dt-bindings: riscv: Add bouffalolab bl808 board compatibles

[Conor Dooley]

On Thu, May 18, 2023 at 10:31:35PM -0500, Samuel Holland wrote:
> Hi Jisheng, DT maintainers,

Sick, thanks for piping up Samuel!
Both Rob and Krzysztof are not around at the moment, so that probably
leaves it up to me.. I'm adding Arnd in case he has a take here too.

> On 5/18/23 10:22, Jisheng Zhang wrote:
> > Several SoMs and boards are available that feature the Bouffalolab
> > bl808 SoC. Document the compatible strings.
> > 
> > Signed-off-by: Jisheng Zhang <jszhang@xxxxxxxxxx>
> > Acked-by: Palmer Dabbelt <palmer@xxxxxxxxxxxx>
> > Reviewed-by: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
> > ---
> >  .../bindings/riscv/bouffalolab.yaml           | 29 +++++++++++++++++++
> >  1 file changed, 29 insertions(+)
> >  create mode 100644 Documentation/devicetree/bindings/riscv/bouffalolab.yaml
> > 
> > diff --git a/Documentation/devicetree/bindings/riscv/bouffalolab.yaml b/Documentation/devicetree/bindings/riscv/bouffalolab.yaml
> > new file mode 100644
> > index 000000000000..3b25d1a5d04a
> > --- /dev/null
> > +++ b/Documentation/devicetree/bindings/riscv/bouffalolab.yaml
> > @@ -0,0 +1,29 @@
> > +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
> > +%YAML 1.2
> > +---
> > +$id: http://devicetree.org/schemas/riscv/bouffalolab.yaml#
> > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > +
> > +title: Bouffalo Lab Technology SoC-based boards
> > +
> > +maintainers:
> > +  - Jisheng Zhang <jszhang@xxxxxxxxxx>
> > +
> > +description:
> > +  Bouffalo Lab Technology SoC-based boards
> > +
> > +properties:
> > +  $nodename:
> > +    const: '/'
> > +  compatible:
> > +    oneOf:
> > +      - description: Carrier boards for the Sipeed M1s SoM
> > +        items:
> > +          - enum:
> > +              - sipeed,m1s-dock
> > +          - const: sipeed,m1s
> > +          - const: bouffalolab,bl808
> 
> As mentioned in the message for patch 5, "The Bouffalolab bl808 SoC
> contains three riscv CPUs, namely M0, D0 and LP. The D0 is 64bit RISC-V
> GC compatible, so can run linux."
> 
> I have also been running U-Boot and NOMMU Linux on the less powerful,
> but still quite fast, "M0" core. However, this core needs a different
> DTB because:
>  1) The CPU is different (T-HEAD E907 instead of C906).
>  2) The interrupt routing is completely different.
>     a. The M0 core contains a CLIC instead of a PLIC.
>     b. The peripherals in the SoC are split between two buses. Those
>        on one bus have their IRQs directly connected to M0, and share
>        a multiplexed IRQ connection to D0; and vice versa for the
>        other bus. So each bus's interrupt-parent needs to be swapped.
> 
> Using some preprocessor magic like we did for Allwinner and Renesas, I
> was able to share most of the SoC and board DTs between the cores[1].
> However, this still ends up with two DTs for each board. So here are my
> questions:
>  - Is this acceptable?

I expected it to look worse than it actually turned out to be.
I don't think Krzysztof in particular is a fan of having conditional
bits in dts files, but for the shared arm/riscv stuff there was not
really another sensible option.

>  - Is there precedent for how we should name the two board DTs?

Arnd might have some idea about precedent here, but I like your naming
well enough.

>  - How does this affect the board and SoC compatible strings?
>    - Should there be a separate "bouffalolab,bl808-d0" in addition to
>      "bouffalolab,bl808"?

What ordering were you intending here?
"pine64,0x64" "bouffalolab,bl808" "bouffalolab,bl808-d0"?

That doesn't really seem correct though, as it does not get less specific
as you move right.

"pine64,0x64" "bouffalolab,bl808-d0" "bouffalolab,bl808" doesn't seem
right either though, for the same sort of reason.

>    - Is it acceptable to use the same board compatible string for both,
>      since the _board_ part of the DT does not change, only things
>      inside the SoC?

I think you may need to have 2 compatibles per board, depending on which
cpu. Perhaps even as verbose as:
"pine61,0x64-d0" "pine64,0x64" "bouffalolab,bl808-d0" "bouffalolab,bl808"

Not exactly straightforward though, is it!

> It would be possible to avoid having two DTs per board by guarding all
> of the differences behind "#ifdef CONFIG_64BIT", but that seems wrong
> because you would end up with two totally incompatible DTBs named the
> same thing, depending on how the DTB was built.

I think having 2 dtbs is fine, and as I mentioned, I've seen Krzysztof
complain previously about conditional bits like that.

Cheers,
Conor.

Attachment: signature.asc
Description: PGP signature