Re: [PATCH 1/3] dt-bindings: PCI: cadence: Deprecate inbound/outbound specific bindings

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Rob,

On 3/31/2020 10:15 PM, Rob Herring wrote:
> On Tue, Mar 31, 2020 at 09:08:12AM +0530, Kishon Vijay Abraham I wrote:
>> Hi Rob,
>>
>> On 3/30/2020 9:31 PM, Rob Herring wrote:
>>> On Fri, Mar 27, 2020 at 04:17:25PM +0530, Kishon Vijay Abraham I wrote:
>>>> Deprecate cdns,max-outbound-regions and cdns,no-bar-match-nbits for
>>>> host mode as both these could be derived from "ranges" and "dma-ranges"
>>>> property. "cdns,max-outbound-regions" property would still be required
>>>> for EP mode.
>>>>
>>>> Signed-off-by: Kishon Vijay Abraham I <kishon@xxxxxx>
>>>> ---
>>>>  .../bindings/pci/cdns,cdns-pcie-ep.yaml       |  2 +-
>>>>  .../bindings/pci/cdns,cdns-pcie-host.yaml     |  3 +--
>>>>  .../devicetree/bindings/pci/cdns-pcie-ep.yaml | 25 +++++++++++++++++++
>>>>  .../bindings/pci/cdns-pcie-host.yaml          | 10 ++++++++
>>>>  .../devicetree/bindings/pci/cdns-pcie.yaml    |  8 ------
>>>>  5 files changed, 37 insertions(+), 11 deletions(-)
>>>>  create mode 100644 Documentation/devicetree/bindings/pci/cdns-pcie-ep.yaml
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-ep.yaml b/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-ep.yaml
>>>> index 2996f8d4777c..50ce5d79d2c7 100644
>>>> --- a/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-ep.yaml
>>>> +++ b/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-ep.yaml
>>>> @@ -10,7 +10,7 @@ maintainers:
>>>>    - Tom Joseph <tjoseph@xxxxxxxxxxx>
>>>>  
>>>>  allOf:
>>>> -  - $ref: "cdns-pcie.yaml#"
>>>> +  - $ref: "cdns-pcie-ep.yaml#"
>>>>    - $ref: "pci-ep.yaml#"
>>>>  
>>>>  properties:
>>>> diff --git a/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-host.yaml b/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-host.yaml
>>>> index cabbe46ff578..84a8f095d031 100644
>>>> --- a/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-host.yaml
>>>> +++ b/Documentation/devicetree/bindings/pci/cdns,cdns-pcie-host.yaml
>>>> @@ -45,8 +45,6 @@ examples:
>>>>              #size-cells = <2>;
>>>>              bus-range = <0x0 0xff>;
>>>>              linux,pci-domain = <0>;
>>>> -            cdns,max-outbound-regions = <16>;
>>>> -            cdns,no-bar-match-nbits = <32>;
>>>>              vendor-id = <0x17cd>;
>>>>              device-id = <0x0200>;
>>>>  
>>>> @@ -57,6 +55,7 @@ examples:
>>>>  
>>>>              ranges = <0x02000000 0x0 0x42000000  0x0 0x42000000  0x0 0x1000000>,
>>>>                       <0x01000000 0x0 0x43000000  0x0 0x43000000  0x0 0x0010000>;
>>>> +            dma-ranges = <0x02000000 0x0 0x0 0x0 0x0 0x1 0x00000000>;
>>>>  
>>>>              #interrupt-cells = <0x1>;
>>>>  
>>>> diff --git a/Documentation/devicetree/bindings/pci/cdns-pcie-ep.yaml b/Documentation/devicetree/bindings/pci/cdns-pcie-ep.yaml
>>>> new file mode 100644
>>>> index 000000000000..6150a7a7bdbf
>>>> --- /dev/null
>>>> +++ b/Documentation/devicetree/bindings/pci/cdns-pcie-ep.yaml
>>>> @@ -0,0 +1,25 @@
>>>> +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
>>>> +%YAML 1.2
>>>> +---
>>>> +$id: "http://devicetree.org/schemas/pci/cdns-pcie-ep.yaml#";
>>>> +$schema: "http://devicetree.org/meta-schemas/core.yaml#";
>>>> +
>>>> +title: Cadence PCIe Device
>>>> +
>>>> +maintainers:
>>>> +  - Tom Joseph <tjoseph@xxxxxxxxxxx>
>>>> +
>>>> +allOf:
>>>> +  - $ref: "cdns-pcie.yaml#"
>>>> +
>>>> +properties:
>>>> +  cdns,max-outbound-regions:
>>>> +    description: maximum number of outbound regions
>>>> +    allOf:
>>>> +      - $ref: /schemas/types.yaml#/definitions/uint32
>>>> +    minimum: 1
>>>> +    maximum: 32
>>>> +    default: 32
>>>
>>> I have a feeling that as the PCI endpoint binding evolves this won't be 
>>> necessary. I can see a common need to define the number of BARs for an 
>>> endpoint and then this will again just be error checking.
>>
>> For every buffer given by the host, we have to create a new outbound
>> translation. If there are no outbound regions, we have to report the error to
>> the endpoint function driver. At-least for reporting the error, we'd need to
>> have this binding no?
> 
> But isn't the endpoint defined to have some number of BARs? The PCI host 
> doesn't decide that.

cdns,max-outbound-regions defined here doesn't configure the BARs. BARs provide
an interface for the host to access the endpoints memory. IOW for BARs we
configure the inbound address translation unit.

cdns,max-outbound-regions is used while configuring the outbound address
translation unit. Outbound regions are used while the endpoint access host
memory and in that path endpoint BARs doesn't come.
> 
>>>
>>> What's the result if you write to a non-existent region in register 
>>> CDNS_PCIE_AT_OB_REGION_PCI_ADDR0/1? If the register is non-existent and 
>>> doesn't abort, you could detect this instead.
>>
>> I'm not sure if we should ever try to write to a non-existent register though
>> the behavior could be different in different platforms. IMHO maximum number of
>> outbound regions is a HW property and is best described in device tree.
> 
> AIUI, PCI defines non-existent (config space) registers to return all 
> 1s. Not sure if this register is in PCI config space or the host SoC bus 
> (e.g. AXI). It seems PCI bridges get done both ways from what I've seen.

All of that is correct for the Host or RC. However here
cdns,max-outbound-regions is an endpoint specific property (defined only in
cdns-pcie-ep.yaml) and is useful while configuring OB address translation unit
for the endpoint to access host memory.

Thanks
Kishon



[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux