Re: [PATCH 06/12] riscv: dts: allwinner: Add the D1 SoC base devicetree

From: Andre Przywara
Date: Mon Aug 22 2022 - 06:50:23 EST


On Sat, 20 Aug 2022 12:24:55 -0500
Samuel Holland <samuel@xxxxxxxxxxxx> wrote:

Hi,

> On 8/15/22 12:01 PM, Conor.Dooley@xxxxxxxxxxxxx wrote:
> > On 15/08/2022 14:11, Andre Przywara wrote:
> >> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> >>
> >> On Mon, 15 Aug 2022 00:08:09 -0500
> >> Samuel Holland <samuel@xxxxxxxxxxxx> wrote:
> >>
> >> Hi,
> >>
> >> thanks for all the efforts in getting those SoC peripherals supported!
> >>
> >>> D1 is a SoC containing a single-core T-HEAD Xuantie C906 CPU, as well as
> >>> one HiFi 4 DSP. The SoC is based on a design that additionally contained
> >>> a pair of Cortex A7's. For that reason, some peripherals are duplicated.
> >>
> >> So because of this, the Allwinner R528 and T113 SoCs would share almost
> >> everything in this file. Would it be useful to already split this DT up?
> >> To have a base .dtsi, basically this file without /cpus and /soc/plic,
> >> then have a RISC-V specific file with just those, including the base?
> >> There is precedence for this across-arch(-directories) sharing with the
> >> Raspberry Pi and Allwinner H3/H5 SoCs.
> >
> > For those playing along at home, one example is the arm64 bananapi m2
> > dts which looks like:
> >> /dts-v1/;
> >> #include "sun50i-h5.dtsi"
> >> #include "sun50i-h5-cpu-opp.dtsi"
> >> #include <arm/sunxi-bananapi-m2-plus-v1.2.dtsi>
> >>
> >> / {
> >> model = "Banana Pi BPI-M2-Plus v1.2 H5";
> >> compatible = "bananapi,bpi-m2-plus-v1.2", "allwinner,sun50i-h5";
> >> };
> >
> > I think this is a pretty good idea, and putting in the modularity up
> > front seems logical to me, so when the arm one does eventually get
> > added it can be done by only touching a single arch.
>
> This is not feasible, due to the different #interrupt-cells. See
> https://lore.kernel.org/linux-riscv/CAMuHMdXHSMcrVOH+vcrdRRF+i2TkMcFisGxHMBPUEa8nTMFpzw@xxxxxxxxxxxxxx/
>
> Even if we share some file across architectures, you still have to update files
> in both places to get the interrupts properties correct.

There are interrupt-maps for that:
sun8i-r528.dtsi:
soc {
#interrupt-cells = <1>;
interrupt-map = <0 18 &gic GIC_SPI 2 IRQ_TYPE_LEVEL_HIGH>,
<0 19 &gic GIC_SPI 3 IRQ_TYPE_LEVEL_HIGH>,
....

sun20i-d1.dtsi:
soc {
#interrupt-cells = <1>;
interrupt-map = <0 18 &plic 18 IRQ_TYPE_LEVEL_HIGH>,
<0 19 &plic 19 IRQ_TYPE_LEVEL_HIGH>,

then, in the shared .dtsi:
uart0: serial@2500000 {
compatible = "snps,dw-apb-uart";
...
interrupts = <18>;

This is completely untested, but I have all the files spelt out there, and
dtc seems happy for both architectures (outside of the kernel tree for now).

> I get the desire to deduplicate things, but we already deal with updating the
> same/similar nodes across several SoCs, so that is nothing new. I think it would
> be more confusing/complicated to have all of the interrupts properties
> overridden in a separate file.

So is this the only thing that prevents sharing? The above paragraph
sounds a bit you are not very fond of the idea to begin with?

Cheers,
Andre