Re: [PATCH 3/3] dt-bindings: PCI: Convert generic host binding to DT schema

[Rob Herring <robh@kernel.org>]

On Tue, Dec 31, 2019 at 7:31 AM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>
> Hi Rob,
>
> On Tue, Dec 31, 2019 at 9:23 AM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> > On Tue, Dec 31, 2019 at 12:30 AM Rob Herring <robh@kernel.org> wrote:
> > > On Thu, Dec 12, 2019 at 7:41 AM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> > > > On Sat, Nov 16, 2019 at 1:53 AM Rob Herring <robh@kernel.org> wrote:
> > > > > Convert the generic PCI host binding to DT schema. The derivative Juno,
> > > > > PLDA XpressRICH3-AXI, and Designware ECAM bindings all just vary in
> > > > > their compatible strings. The simplest way to convert those to
> > > > > schema is just add them into the common generic PCI host schema.
> > > > >
> > > > > Cc: Bjorn Helgaas <bhelgaas@google.com>
> > > > > Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
> > > > > Cc: Andrew Murray <andrew.murray@arm.com>
> > > > > Cc: Zhou Wang <wangzhou1@hisilicon.com>
> > > > > Cc: Will Deacon <will@kernel.org>
> > > > > Cc: David Daney <david.daney@cavium.com>
> > > > > Signed-off-by: Rob Herring <robh@kernel.org>
> > > >
> > > > > index 515b2f9542e5..000000000000
> > > > > --- a/Documentation/devicetree/bindings/pci/designware-pcie-ecam.txt
> > > > > +++ /dev/null
> > > >
> > > > > -Example:
> > > > > -
> > > > > -    pcie1: pcie@7f000000 {
> > > > > -        compatible = "socionext,synquacer-pcie-ecam", "snps,dw-pcie-ecam";
> > > > > -        device_type = "pci";
> > > > > -        reg = <0x0 0x7f000000 0x0 0xf00000>;
> > > > > -        bus-range = <0x0 0xe>;
> > > > > -        #address-cells = <3>;
> > > > > -        #size-cells = <2>;
> > > > > -        ranges = <0x1000000 0x00 0x00010000 0x00 0x7ff00000 0x0 0x00010000>,
> > > > > -                 <0x2000000 0x00 0x70000000 0x00 0x70000000 0x0 0x0f000000>,
> > > > > -                 <0x3000000 0x3f 0x00000000 0x3f 0x00000000 0x1 0x00000000>;
> > > > > -
> > > > > -        #interrupt-cells = <0x1>;
> > > > > -        interrupt-map-mask = <0x0 0x0 0x0 0x0>;
> > > >
> > > > An all-zeroes interrupt-map-mask seems to be very common on embedded
> > > > SoCs, where all devices are mapped to a single interrupt.
> > >
> > > Indeed.
> > >
> > > > However, schemas/pci/pci-bus.yaml says:
> > > >
> > > >   interrupt-map-mask:
> > > >     items:
> > > >       - description: PCI high address cell
> > > >         minimum: 0
> > > >         maximum: 0xf800
> > > >       - description: PCI mid address cell
> > > >         const: 0
> > > >       - description: PCI low address cell
> > > >         const: 0
> > > >       - description: PCI IRQ cell
> > > >         minimum: 1
> > > >         maximum: 7
> > > >
> > > > and thus complains about an all-zeroes mask, e.g.
> > > >
> > > >     arch/arm64/boot/dts/renesas/r8a7795-salvator-x.dt.yaml:
> > > > pcie@fe000000: interrupt-map-mask:0:3: 0 is less than the minimum of 1
> > >
> > > Now fixed.
> >
> > Thank you, confirmed.
>
> And with latest renesas-drivers, I started seeing:
>
>     arch/arm/boot/dts/r8a7791-koelsch.dt.yaml: pci@ee0d0000:
> interrupt-map:0: [0, 0, 0, 1] is too short
>     arch/arm/boot/dts/r8a7791-koelsch.dt.yaml: pci@ee0d0000:
> interrupt-map:1: [5, 0, 113, 4] is too short
>     arch/arm/boot/dts/r8a7791-koelsch.dt.yaml: pci@ee0d0000:
> interrupt-map:2: [2048, 0, 0, 1] is too short
>     arch/arm/boot/dts/r8a7791-koelsch.dt.yaml: pci@ee0d0000:
> interrupt-map:3: [5, 0, 113, 4] is too short
>     arch/arm/boot/dts/r8a7791-koelsch.dt.yaml: pci@ee0d0000:
> interrupt-map:4: [4096, 0, 0, 2] is too short
>     arch/arm/boot/dts/r8a7791-koelsch.dt.yaml: pci@ee0d0000:
> interrupt-map:5: [5, 0, 113, 4] is too short
>
> Looks like interrupt-map is split incorrectly: shouldn't each entry have 8
> cells?

That must be with a current dtc which now splits the array before each
phandle. That works for phandle+args, but not *-map properties. :( I
was trying to avoid a bunch of dts updates to add brackets. I think
for now, I'll just drop the interrupt-map size constraint. It's not
all that accurate anyways as it doesn't look at cell sizes.

Rob