Re: [PATCH] arm64: dts: fsd: Change the reg properties from 64-bit to 32-bit

[Robin Murphy]

On 2022-11-16 12:38, Robin Murphy wrote:
> On 2022-11-16 11:16, Arnd Bergmann wrote:
> > On Wed, Nov 16, 2022, at 10:12, Vivek Yadav wrote:
> > > Change the reg properties from 64-bit to 32-bit for all IPs, as none of
> > > the nodes are above 32-bit range in the fsd SoC.
> > >
> > > Since dma-ranges length does not fit into 32-bit size, keep it 64-bit
> > > and move it to specific node where it is used instead of SoC section.
> >
> > I don't think that works, the dma-ranges property is part of the
> > bus, not a particular device:
> >
> >           mdma0: dma-controller@10100000 {
> >               compatible = "arm,pl330", "arm,primecell";
> > -            reg = <0x0 0x10100000 0x0 0x1000>;
> > +            reg = <0x10100000 0x1000>;
> >               interrupts = <GIC_SPI 424 IRQ_TYPE_LEVEL_HIGH>;
> >               #dma-cells = <1>;
> >               clocks = <&clock_imem IMEM_DMA0_IPCLKPORT_ACLK>;
> >               clock-names = "apb_pclk";
> >               iommus = <&smmu_imem 0x800 0x0>;
> > +            #address-cells = <2>;
> > +            #size-cells = <2>;
> > +            dma-ranges = <0x0 0x0 0x0 0x10 0x0>;
> >           };
> >
> > Since the dma-controller has no children, I don't see how this has
> > any effect. Also, translating a 36-bit address into a 32-bit
> > address just means it gets truncated anyway, so there is no
> > point in making it appear to have a larger address range.
>
> Yes, this is definitely bogus on both counts.

Oh, and also that PL330 can only do 32-bit DMA anyway :)

Thanks,
Robin.