Re: [PATCH 3/5] PCI: rcar: Add R-Car PCIe endpoint device tree bindings

[Rob Herring]

On Thu, Nov 07, 2019 at 09:08:35PM +0100, Geert Uytterhoeven wrote:
> Hi Prabhakar,
> 
> On Thu, Nov 7, 2019 at 10:26 AM Lad, Prabhakar
> <prabhakar.csengg@xxxxxxxxx> wrote:
> > On Thu, Nov 7, 2019 at 8:44 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > On Wed, Nov 6, 2019 at 8:36 PM Lad Prabhakar <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > From: "Lad, Prabhakar" <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > > >
> > > > This patch adds the bindings for the R-Car PCIe endpoint driver.
> > > >
> > > > Signed-off-by: Lad, Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > >
> > > Thanks for your patch!
> > >
> > > > --- /dev/null
> > > > +++ b/Documentation/devicetree/bindings/pci/rcar-pci-ep.txt
> > > > @@ -0,0 +1,43 @@
> > > > +* Renesas R-Car PCIe Endpoint Controller DT description
> > > > +
> > > > +Required properties:
> > > > +           "renesas,pcie-ep-r8a774c0" for the R8A774C0 SoC;
> > > > +           "renesas,pcie-ep-rcar-gen3" for a generic R-Car Gen3 or
> > > > +                                    RZ/G2 compatible device.
> > >
> > > Unless I'm missing something, this is for the exact same hardware block as
> > > Documentation/devicetree/bindings/pci/rcar-pci.txt?
> > > So shouldn't you amend those bindings, instead of adding new compatible
> > > values?
> > > Please remember that DT describes hardware, not software policy.
> > > So IMHO choosing between host and endpoint is purely a configuration
> > > issue, and could be indicated by the presence or lack of some DT properties.
> > > E.g. host mode requires both "bus-range" and "device_type" properties,
> > > so their absence could indicate endpoint mode.
> > >
> > yes its the same hardware block as described in the rcar-pci.txt, I
> > did think about amending it
> > but  it might turn out to be bit messy,
> >
> > required properties host ======required properties Endpoint
> > ====================||==================
> > 1: reg                                || reg
> > 2:bus-range                      || reg names
> > 3: device_type                  || resets
> > 4: ranges                          || clocks
> > 5: dma-ranges                  || clock-names
> > 6: interrupts                      ||
> > 7: interrupt-cells               ||
> > 8: interrupt-map-mask     ||
> > 9: clocks                          ||
> > 10: clock-names             ||
> 
> We have a similar situation with SPI, where a controller can operate in
> master or slave mode, based on the absence or presence of the
> "spi-slave" DT property.
> 
> > and if I go ahead with the same compatible string that would mean to
> > add support for endpoint
> > mode in the host driver itself. I did follow the examples of
> 
> You can still have two separate drivers, binding against the same
> compatible value.  Just let the .probe() function return -ENODEV if it
> discovers (by looking at DT properties) if the node is configured for
> the other mode.
> Which brings us to my next questions: is there any code that could be
> shared between the drivers for the two modes?
> 
> > rockchip/cadence/designware where
> > its the same hardware block but has two different binding files one
> > for host mode and other for
> > endpoint mode.
> 
> Having two separate DT binding documents sounds fine to me, if unifying
> them makes things too complex.
> However, I think they should use the same compatible value, because the
> hardware block is the same, but just used in a different mode.
> 
> Rob/Mark: Any input from the DT maintainers?

Separate files makes sense because different modes will want to 
include different common schemas. We've generally been doing different 
compatibles too which makes validating the node has the right set of 
properties easier.
 
> > > > +- reg: Five register ranges as listed in the reg-names property
> > > > +- reg-names: Must include the following names
> > > > +       - "apb-base"
> > > > +       - "memory0"
> > > > +       - "memory1"
> > > > +       - "memory2"
> > > > +       - "memory3"
> > >
> > > What is the purpose of the last 4 regions?
> > > Can they be chosen by the driver, at runtime?
> > >
> > no the driver cannot choose them at runtime, as these are the only
> > PCIE memory(0/1/2/3) ranges
> > in the AXI address space where host memory can be mapped.
> 
> Are they fixed by the PCIe hardware, i.e. could they be looked up by the
> driver based on the compatible value?

That would be strange for a memory range.

Sounds like like 'ranges' though I'm not sure if 'ranges' for an EP 
makes sense or what that should look like.

Rob