Re: [PATCH 1/7] dt-bindings: PCI: qcom: Add IPQ9574 specific compatible

[Dmitry Baryshkov]

On Tue, 7 Mar 2023 at 16:40, Devi Priya <quic_devipriy@xxxxxxxxxxx> wrote:
>
>
>
> On 3/7/2023 6:26 PM, Manivannan Sadhasivam wrote:
> > On Tue, Mar 07, 2023 at 03:15:08PM +0530, Devi Priya wrote:
> >>
> >>
> >> On 3/3/2023 11:10 PM, Manivannan Sadhasivam wrote:
> >>> On Fri, Mar 03, 2023 at 05:16:58PM +0200, Dmitry Baryshkov wrote:
> >>>> 28 февраля 2023 г. 08:33:58 GMT+02:00, Manivannan Sadhasivam <mani@xxxxxxxxxx> пишет:
> >>>>> On Tue, Feb 28, 2023 at 10:56:53AM +0530, Devi Priya wrote:
> >>>>>>
> >>>>>>
> >>>>>> On 2/24/2023 1:53 PM, Manivannan Sadhasivam wrote:
> >>>>>>> On Tue, Feb 14, 2023 at 10:11:29PM +0530, Devi Priya wrote:
> >>>>>>>> Document the compatible for IPQ9574
> >>>>>>>>
> >>>>>> Hi Mani, Thanks for taking time to review the patch.
> >>>>>>>
> >>>>>>> You didn't mention about the "msi-parent" property that is being added
> >>>>>>> by this patch
> >>>>>> Sure, will update the commit message in the next spin
> >>>>>>>
> >>>>>>>> Signed-off-by: Devi Priya <quic_devipriy@xxxxxxxxxxx>
> >>>>>>>> ---
> >>>>>>>>     .../devicetree/bindings/pci/qcom,pcie.yaml    | 72 ++++++++++++++++++-
> >>>>>>>>     1 file changed, 70 insertions(+), 2 deletions(-)
> >>>>>>>>
> >>>>>>>> diff --git a/Documentation/devicetree/bindings/pci/qcom,pcie.yaml b/Documentation/devicetree/bindings/pci/qcom,pcie.yaml
> >>>>>>>> index 872817d6d2bd..dabdf2684e2d 100644
> >>>>>>>> --- a/Documentation/devicetree/bindings/pci/qcom,pcie.yaml
> >>>>>>>> +++ b/Documentation/devicetree/bindings/pci/qcom,pcie.yaml
> >>>>>>>> @@ -26,6 +26,7 @@ properties:
> >>>>>>>>               - qcom,pcie-ipq8064-v2
> >>>>>>>>               - qcom,pcie-ipq8074
> >>>>>>>>               - qcom,pcie-ipq8074-gen3
> >>>>>>>> +          - qcom,pcie-ipq9574
> >>>>>>>>               - qcom,pcie-msm8996
> >>>>>>>>               - qcom,pcie-qcs404
> >>>>>>>>               - qcom,pcie-sa8540p
> >>>>>>>> @@ -44,11 +45,11 @@ properties:
> >>>>>>>>       reg:
> >>>>>>>>         minItems: 4
> >>>>>>>> -    maxItems: 5
> >>>>>>>> +    maxItems: 6
> >>>>>>>>       reg-names:
> >>>>>>>>         minItems: 4
> >>>>>>>> -    maxItems: 5
> >>>>>>>> +    maxItems: 6
> >>>>>>>>       interrupts:
> >>>>>>>>         minItems: 1
> >>>>>>>> @@ -105,6 +106,8 @@ properties:
> >>>>>>>>         items:
> >>>>>>>>           - const: pciephy
> >>>>>>>> +  msi-parent: true
> >>>>>>>> +
> >>>>>>>>       power-domains:
> >>>>>>>>         maxItems: 1
> >>>>>>>> @@ -173,6 +176,27 @@ allOf:
> >>>>>>>>                 - const: parf # Qualcomm specific registers
> >>>>>>>>                 - const: config # PCIe configuration space
> >>>>>>>> +  - if:
> >>>>>>>> +      properties:
> >>>>>>>> +        compatible:
> >>>>>>>> +          contains:
> >>>>>>>> +            enum:
> >>>>>>>> +              - qcom,pcie-ipq9574
> >>>>>>>> +    then:
> >>>>>>>> +      properties:
> >>>>>>>> +        reg:
> >>>>>>>> +          minItems: 5
> >>>>>>>> +          maxItems: 6
> >>>>>>>> +        reg-names:
> >>>>>>>> +          minItems: 5
> >>>>>>>> +          items:
> >>>>>>>> +            - const: dbi # DesignWare PCIe registers
> >>>>>>>> +            - const: elbi # External local bus interface registers
> >>>>>>>> +            - const: atu # ATU address space
> >>>>>>>> +            - const: parf # Qualcomm specific registers
> >>>>>>>> +            - const: config # PCIe configuration space
> >>>>>>>> +            - const: aggr_noc #PCIe aggr_noc
> >>>>>>>
> >>>>>>> Why do you need this region unlike other SoCs? Is the driver making use of it?
> >>>>>> We have the aggr_noc region in ipq9574 to achieve higher throughput & to
> >>>>>> handle multiple PCIe instances. The driver uses it to rate adapt 1-lane PCIe
> >>>>>> clocks. My bad, missed it. Will add the driver changes in V2.
> >>>>>
> >>>>> Hmm, this is something new. How can you achieve higher throughput with this
> >>>>> region? Can you explain more on how it is used?
> >>>>
> >>>> Based on the name of the region, it looks like it is an interconnect region.
> >>>>
> >>>
> >>> Well, we only have BCM based interconnects so far. That's why I was curious
> >>> about this region and its purpose.
> >> For connected PCIe slave devices that are running at frequency lesser
> >> than the ANOC frequency (342MHz), the rate adapter of ANOC needs to be
> >> configured
> >>>
> >>>> Devi, if this is the case, then you have to handle it through the interconnect driver, rather than poking directly into these registers.
> >>>
> >>> If that so, it doesn't need to be added in this series itself. I believe that
> >>> without aggr_noc region, the PCIe controller can still function properly with
> >>> reduced performance. But you can add the interconnect support later as a
> >>> separate series.
> >> Sure, okay. The ANOC runs at a fixed frequency of 342MHz and the
> >> interconnect clocks are not scaled. The aggr_noc register is just a magic
> >> register for configuring it's rate adapter to ensure no wait cycles are
> >> inserted.
> >>
> >
> > If the purpose of the aggr_noc region is to configure the interconnect clock,
> > then it should be modeled as an interconnect driver.
> Can we use 'syscon' here, as we are not scaling the interconnect
> frequency and this is just a single register write for setting
> the rate adapter?

It should be done outside of the PCIe driver.
It is not "just a single register". It is also setting the anoc/snoc
clocks for USB. And maybe something else, which we haven't seen at
this moment. You are still setting up the NoC, even if the icc
frequency is not scaled.

-- 
With best wishes
Dmitry