[PATCH 1/4] dt-bindings: pcie: Add the NXP PCIe controller

[Vincent Guittot]

On Wed, 17 Sept 2025 at 23:28, Bjorn Helgaas <helgaas at kernel.org> wrote:
>
> On Wed, Sep 17, 2025 at 10:41:08PM +0530, Manivannan Sadhasivam wrote:
> > On Tue, Sep 16, 2025 at 09:23:13AM GMT, Bjorn Helgaas wrote:
> > > On Tue, Sep 16, 2025 at 10:10:31AM +0200, Vincent Guittot wrote:
> > > > On Sun, 14 Sept 2025 at 14:35, Vincent Guittot
> > > > <vincent.guittot at linaro.org> wrote:
> > > > > On Sat, 13 Sept 2025 at 00:50, Bjorn Helgaas <helgaas at kernel.org> wrote:
> > > > > > On Fri, Sep 12, 2025 at 04:14:33PM +0200, Vincent Guittot wrote:
> > > > > > > Describe the PCIe controller available on the S32G platforms.
> > >
> > > > > > > +                  num-lanes = <2>;
> > > > > > > +                  phys = <&serdes0 PHY_TYPE_PCIE 0 0>;
> > > > > >
> > > > > > num-lanes and phys are properties of a Root Port, not the host bridge.
> > > > > > Please put them in a separate stanza.  See this for details and
> > > > > > examples:
> > > > > >
> > > > > >   https://lore.kernel.org/linux-pci/20250625221653.GA1590146@bhelgaas/
> > > > >
> > > > > Ok, I'm going to have a look
> > > >
> > > > This driver relies on dw_pcie_host_init() to get common resources like
> > > > num-lane which doesn't look at childs to get num-lane.
> > > >
> > > > I have to keep num-lane in the pcie node. Having this in mind should I
> > > > keep phys as well as they are both linked ?
>
> > > Huh, that sounds like an issue in the DWC core.  Jingoo, Mani?
> > >
> > > dw_pcie_host_init() includes several things that assume a single Root
> > > Port: num_lanes, of_pci_get_equalization_presets(),
> > > dw_pcie_start_link() are all per-Root Port things.
> >
> > Yeah, it is a gap right now. We only recently started moving the DWC
> > platforms to per Root Port binding (like Qcom).
>
> Do you need num-lanes in the devicetree?
> dw_pcie_link_get_max_link_width() will read it from PCI_EXP_LNKCAP, so
> if that works maybe you can omit it from the binding?

num-lane is not mandatory but we can have 1 or 2 lanes so should be
able to restrict to only 1 lane for some platform
the "num-lanes = <2>;" in the example is wrong as we don't need it in
case of 2 lanes but only if we want to restrict to 1 lane

>
> If you do need num-lanes in the binding, maybe you could make a Root
> Port parser similar to mvebu_pcie_parse_port() or
> qcom_pcie_parse_port() that would get num-lanes, the PHY, and
> nxp,phy-mode from a Root Port node?
>
> Then all this would be in one place, and if you set ->num_lanes there
> it looks like the DWC core wouldn't do anything with it.