[PATCH v2 2/4] dt-bindings: PCI: mediatek,mt7621: add missing child node reg

[Rob Herring]

On Thu, Apr 11, 2024 at 11:11 AM Bjorn Helgaas <helgaas at kernel.org> wrote:
>
> On Thu, Apr 11, 2024 at 09:21:07AM -0500, Rob Herring wrote:
> > On Thu, Apr 11, 2024 at 07:39:17AM -0500, Bjorn Helgaas wrote:
> > > On Thu, Apr 11, 2024 at 08:13:18AM +0200, Sergio Paracuellos wrote:
> > > > On Thu, Apr 11, 2024 at 8:01 AM Krzysztof Kozlowski
> > > > <krzysztof.kozlowski at linaro.org> wrote:
> > > > > On 10/04/2024 23:26, Bjorn Helgaas wrote:
> > > > > > On Wed, Apr 10, 2024 at 08:15:19PM +0200, Krzysztof Kozlowski wrote:
> > > > > >> MT7621 PCI host bridge has children which apparently are also PCI host
> > > > > >> bridges, at least that's what the binding suggest.
> > > > > >
> > > > > > What does it even mean for a PCI host bridge to have a child that is
> > > > > > also a PCI host bridge?
> >
> > It should say 'root port' instead as the binding description correctly
> > says.
>
> OK, that makes a lot more sense, and we should fix the commit log.
>
> > > > > I think the question should be towards Mediatek folks. I don't know what
> > > > > this hardware is exactly, just looks like pci-pci-bridge. The driver
> > > > > calls the children host bridges as "ports".
> > > >
> > > > You can see the topology here in my first driver submit cover letter
> > > > message [0].
> > > >
> > > >  [0]: https://lore.kernel.org/all/CAMhs-H-BA+KzEwuDPzcmrDPdgJBFA2XdYTBvT4R4MEOUB=WQ1g@mail.gmail.com/t/
> > >
> > > Nothing unusual here, this looks like the standard PCIe topology.
> > >
> > > What *might* be unusual is describing the Root Ports in DT.  Since
> > > they are standard PCI devices, they shouldn't need DT description
> > > unless there's some unusual power/clock/reset control or something
> > > that is not discoverable via PCI enumeration.
> >
> > It's only unusual because typically there's only 1 RP per host bridge
> > and properties which really apply to the RP get stuck in the host bridge
> > node because we don't have a RP node. An example is perst-gpios. That's
> > not a property of the RP either, but the RP is the upstream side of a
> > slot and we often don't have a node for the device either.
>
> Makes sense.
>
> I'm still confused about one thing, maybe just because I don't really
> know how to read these bindings.  The binding now looks like this:
>
>   properties:
>     compatible:
>       const: mediatek,mt7621-pci
>
>     reg:
>       items:
>         - description: host-pci bridge registers
>         - description: pcie port 0 RC control registers       # A
>         - description: pcie port 1 RC control registers       # A
>         - description: pcie port 2 RC control registers       # A
>
>   patternProperties:
>     '^pcie@[0-2],0$':
>       type: object
>       $ref: /schemas/pci/pci-pci-bridge.yaml#
>
>       properties:
>         reg:                                                  # B
>           maxItems: 1
>
> It looks like the "A" items are separate things from the "B" items?
>
> But I think the relevant code is here:
>
>   mt7621_pcie_probe
>     mt7621_pcie_parse_dt
>       pcie->base = devm_platform_ioremap_resource(pdev, 0)             # 1
>       for_each_available_child_of_node(node, child)
>         mt7621_pcie_parse_port
>           port->base = devm_platform_ioremap_resource(pdev, slot + 1)  # 2
>
> where it looks like both "1" and "2" use the items in the "A" list,
> i.e., resources 0, 1, 2, 3, all from the same platform device.  Is
> there code that uses the "B" item that this patch adds?

The A items are in the host address space. The B item is a PCI
address. Specifically, for PCI devices, the first entry is config
space with just the device and function (devfn). The format is defined
in the OpenFirmware PCI bus supplement.

Rob