Re: [PATCH] bindings: PCI: artpec: correct pci binding example

From: Arnd Bergmann
Date: Thu Jul 14 2016 - 04:31:40 EST


On Thursday, July 14, 2016 1:54:20 AM CEST Niklas Cassel wrote:
> On 07/08/2016 11:39 AM, Arnd Bergmann wrote:
> > On Friday, July 8, 2016 1:28:10 AM CEST Niklas Cassel wrote:
> >> From: Niklas Cassel <niklas.cassel@xxxxxxxx>
> >>
> >> - Increase config size. When using a PCIe switch,
> >> the previous config size only had room for one device.
> >> - Add bus range. Inherited optional property.
> >> - Map downstream I/O to PCI address 0. We can map it to any
> >> address, but let's be consistent with other drivers.
> >>
> >> Signed-off-by: Niklas Cassel <niklas.cassel@xxxxxxxx>
> >> ---
> >> Documentation/devicetree/bindings/pci/axis,artpec6-pcie.txt | 7 ++++---
> >> 1 file changed, 4 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/Documentation/devicetree/bindings/pci/axis,artpec6-pcie.txt b/Documentation/devicetree/bindings/pci/axis,artpec6-pcie.txt
> >> index 330a45b..5ecaea1 100644
> >> --- a/Documentation/devicetree/bindings/pci/axis,artpec6-pcie.txt
> >> +++ b/Documentation/devicetree/bindings/pci/axis,artpec6-pcie.txt
> >> @@ -24,16 +24,17 @@ Example:
> >> compatible = "axis,artpec6-pcie", "snps,dw-pcie";
> >> reg = <0xf8050000 0x2000
> >> 0xf8040000 0x1000
> >> - 0xc0000000 0x1000>;
> >> + 0xc0000000 0x2000>;
> > If this is your config space size
> >
> >> num-lanes = <2>;
> >> + bus-range = <0x00 0xff>;
> > then the bus range looks too large. These two are typically connected.
> > I couldn't immediately see which config space access function is
> > used, but if you have 0x1000 bytes per bus, then the bus range matching
> > a 0x2000 byte config space would be either <0x00 0x01> or <0x00 0x02>
> > depending whether the root bus is part of that range.
>
> I see your point, a config space size of 0x2000 is only enough to
> hold two configuration space headers.
>
> However, all other PCIe controllers based on Synopsys DesignWare IP
> with a config space size of 0x2000, uses a bus-range of 0x0-0xff.
> (Except hisilicon which for some reason uses 0x0-0xf).
> Isn't it better to be consistent with the other DesignWare based controllers?

No, if we found a bug, we should fix it for all of them.

> Also, isn't is possible to have a device behind bus 1, then nothing behind
> bus 2, and then a device behind bus 3?
> Not sure if this could actually happen though, since Linux appears to assign
> the bus numbers, unless already defined by BIOS.

I think this might happen with hotplugged devices.

Arnd