[cc->to: Rob for RC/RP separation conversation]
On Thu, Dec 05, 2024 at 02:41:26PM +0100, Christian Bruel wrote:
> On 12/3/24 23:25, Bjorn Helgaas wrote:
> > On Tue, Nov 26, 2024 at 04:51:15PM +0100, Christian Bruel wrote:
> > > Document the bindings for STM32MP25 PCIe Controller configured in
> > > root complex mode.
> > >
> > > Supports 4 legacy interrupts and MSI interrupts from the ARM
> > > GICv2m controller.
> > > + wake-gpios:
> > > + description: GPIO controlled connection to WAKE# input signal
> >
> > I'm not a hardware guy, but this sounds like a GPIO that *reads*
> > WAKE#, not controls it.
>
> Rephrasing as
> "GPIO used as WAKE# input signal" (output for the endpoint bindings)
Perfect, that makes a lot of sense.
> > > + pcie@48400000 {
> > > + compatible = "st,stm32mp25-pcie-rc";
> > > + device_type = "pci";
> > > + num-lanes = <1>;
> >
> > num-lanes applies to a Root Port, not to a Root Complex. I know most
> > bindings conflate Root Ports with the Root Complex, maybe because many
> > of these controllers only support a single Root Port?
> >
> > But are we ever going to separate these out? I assume someday
> > controllers will support multiple Root Ports and/or additional devices
> > on the root bus, like RCiEPs, RCECs, etc., and we'll need per-RP phys,
> > max-link-speed, num-lanes, reset-gpios, etc.
> >
> > Seems like it would be to our benefit to split out the Root Ports when
> > we can, even if the current hardware only supports one, so we can
> > start untangling the code and data structures.
>
> OK. and we support only 1 lane anyway, so drop it.
Makes sense. What about phys, resets, etc? I'm pretty sure a PHY
would be a per-Root Port thing, and some resets and wakeup signals
also.
For new drivers, I think we should start adding Root Port stanzas to
specifically associate those things with the Root Port, e.g.,
something like this?
pcie@48400000 {
compatible = "st,stm32mp25-pcie-rc";
pcie@0,0 {
reg = <0x0000 0 0 0 0>;
phys = <&combophy PHY_TYPE_PCIE>;
phy-names = "pcie-phy";
};
};
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/devicetree/bindings/pci/mediatek,mt7621-pcie.yaml?id=v6.12#n111
is one binding that does this, others include apple,pcie.yaml,
brcm,stb-pcie.yaml, hisilicon,kirin-pcie.yaml.
Bjorn
