[PATCH 1/4 v3] dt-bindings: PCI: s32g: Add NXP PCIe controller

[Vincent Guittot]

On Thu, 6 Nov 2025 at 18:38, Bjorn Helgaas <helgaas at kernel.org> wrote:
>
> [+cc Senchuan]
>
> On Thu, Nov 06, 2025 at 09:09:01AM +0100, Vincent Guittot wrote:
> > On Thu, 6 Nov 2025 at 08:12, Manivannan Sadhasivam <mani at kernel.org> wrote:
> > > On Wed, Oct 22, 2025 at 07:43:06PM +0200, Vincent Guittot wrote:
> > > > Describe the PCIe host controller available on the S32G platforms.
>
> > > > +            phys = <&serdes0 PHY_TYPE_PCIE 0 0>;
> > >
> > > PHY is a Root Port specific resource, not Root Complex. So it
> > > should be moved to the Root Port node and the controller driver
> > > should parse the Root Port node and control PHY. Most of the
> > > existing platforms still specify PHY and other Root Port
> > > properties in controller node, but they are wrong.
> >
> > Yeah, we had similar discussion on v1 and as designware core code
> > doesn't support it, the goal was to follow other implementations
> > until designware core is able to parse root port nodes.  I can add a
> > root port node for the phy and parse it in s32 probe function but
> > then If I need to restrict the number of lane to 1 instead of the
> > default 2 with num-lanes then I have to put it the controller node
> > otherwise designware core node will not get it.
>
> I think it's better to put the PHY info, including num-lanes, in Root
> Port DT nodes now even thought the DWC core doesn't explicitly support
> that yet because it's much easier to change the DWC core and the
> driver code than it is to change the DT structure.
>
> That will mean a little extra code in the s32g driver now, but we will
> be able to remove that eventually.  If we leave the PHY in the DT
> controller node, we may eventually end up having to support two s32g
> DT structures: the single RP style with PHY in the controller, and a
> multiple RP style with PHY in the RP.
>
> We'll likely have both structures for many existing drivers, but I
> think it will be simpler if new drivers can avoid the old one.

Okay, i will add a RP node

>
> The eic7700 driver is an example of num-lanes support in the driver:
> https://lore.kernel.org/linux-pci/20251030083143.1341-1-zhangsenchuan@eswincomputing.com/
>
> Bjorn