Re: [PATCH v4 1/2] dt-bindings: arm: qcom: Document the sc7280 CRD Pro boards

[Rajendra Nayak <quic_rjendra@quicinc.com>]

On 1/10/2023 3:12 AM, Doug Anderson wrote:
> Hi,
> 
> On Mon, Jan 9, 2023 at 1:36 PM Dmitry Baryshkov
> <dmitry.baryshkov@linaro.org> wrote:
>>
>> On 09/01/2023 23:00, Doug Anderson wrote:
>>> Hi,
>>>
>>> On Tue, Dec 20, 2022 at 9:12 AM Dmitry Baryshkov
>>> <dmitry.baryshkov@linaro.org> wrote:
>>>>
>>>> On 20/12/2022 18:20, Rajendra Nayak wrote:
>>>>>
>>>>>
>>>>> On 12/20/2022 8:00 PM, Matthias Kaehlcke wrote:
>>>>>> On Tue, Dec 20, 2022 at 10:30:32AM +0530, Rajendra Nayak wrote:
>>>>>>>
>>>>>>> On 12/16/2022 7:49 PM, Matthias Kaehlcke wrote:
>>>>>>>> On Fri, Dec 16, 2022 at 04:59:17PM +0530, Rajendra Nayak wrote:
>>>>>>>>> Add compatibles for the Pro SKU of the sc7280 CRD boards
>>>>>>>>> which come with a Pro variant of the qcard.
>>>>>>>>> The Pro qcard variant has smps9 from pm8350c ganged up with
>>>>>>>>> smps7 and smps8.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Rajendra Nayak <quic_rjendra@quicinc.com>
>>>>>>>>> Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
>>>>>>>>> Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
>>>>>>>>> ---
>>>>>>>>> v4 changes:
>>>>>>>>> Added the zoglin-sku1536 compatible along with hoglin-sku1536.
>>>>>>>>> Zoglin is same as the Hoglin variant, with the SPI Flash reduced
>>>>>>>>> from 64MB to 8MB
>>>>>>>>>
>>>>>>>>>      Documentation/devicetree/bindings/arm/qcom.yaml | 6 ++++++
>>>>>>>>>      1 file changed, 6 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/Documentation/devicetree/bindings/arm/qcom.yaml
>>>>>>>>> b/Documentation/devicetree/bindings/arm/qcom.yaml
>>>>>>>>> index 1b5ac6b02bc5..07771d4c91bd 100644
>>>>>>>>> --- a/Documentation/devicetree/bindings/arm/qcom.yaml
>>>>>>>>> +++ b/Documentation/devicetree/bindings/arm/qcom.yaml
>>>>>>>>> @@ -558,6 +558,12 @@ properties:
>>>>>>>>>                - const: google,hoglin
>>>>>>>>>                - const: qcom,sc7280
>>>>>>>>> +      - description: Qualcomm Technologies, Inc. sc7280 CRD Pro
>>>>>>>>> platform (newest rev)
>>>>>>>>> +        items:
>>>>>>>>> +          - const: google,zoglin-sku1536
>>>>>>>>> +          - const: google,hoglin-sku1536
>>>>>>>>
>>>>>>>> Is there actually such a thing as a 'hoglin-sku1536', i.e. the Pro
>>>>>>>> qcard
>>>>>>>> with 64MB of SPI flash, or do they all have 8MB of flash?
>>>>>>>
>>>>>>> The SPI flash is on the CRD mother-board and not on the qcards, so if
>>>>>>> you replace
>>>>>>> the qcards on the CRDs with 64MB flash you would need the
>>>>>>> hoglin-sku1536 to
>>>>>>> boot on those.
>>>>>>
>>>>>> With such a configuration how does the bootloader know it should pass
>>>>>> the kernel
>>>>>> the device tree for 'hoglin-sku1536' (pro) and not the non-pro
>>>>>> variant? IIUC the
>>>>>> device tree is selected based on pin strappings on the mother-board,
>>>>>> not the
>>>>>> qcard.
>>>>>
>>>>> The device tree is selected based on the pin strappings _and_ additional
>>>>> logic
>>>>> to dynamically identify modem/non-modem(wifi) as well as pro/non-pro
>>>>> SKUs which
>>>>> was added in the bootloaders.
>>>>
>>>> Just to clarify things, when you mention pro SKU, is it a separate SoC
>>>> revision (like sc7280-pro vs bare sc7280), or is it a CRD revision (CRD
>>>> Pro vs bare CRD)?
>>>
>>> I guess Rajendra never responded, but since I know the answer: it's a

Thanks Doug for the clarifications, I seem to have missed responding to this
once I was back from vacation,

>>> different SoC revision. ...but the SoC in this case is on a daughter
>>> card, so you could remove the daughter card containing the SoC and put
>>> a new daughtercard on. That would have the effect of making an old CRD
>>> revision have the new Pro SKU SoC.
>>
>> So, this is a new SoC. Is it 100% compatible with the sc7280? In other
>> words: does it require any additional customizations (in OPP tables, in
>> frequences, speed bins, etc)?

Yes, the OPP differences are taken care of with no changes needed in kernel.
We describe a superset of *all* OPPs supported by a SoC family in DT and the
cpufreq driver then queries the firmware for supported OPPs on a given
SoC variant and ends up disabling the rest.

> 
> If I understand correctly, the OPP customizations have been accounted
> for since the beginning. I believe that the GPU operating table
> already has some fairly high operating points. Maybe commit
> 3bfef00d7671 ("arm64: dts: qcom: sc7280: Support gpu speedbin") was
> for pro? Similarly, the CPU operating table also has some fairly high
> operating points (probably for PRO?) and I think the higher points are
> dynamically disabled for CPUs that don't support them. That's how it
> was on sc7180, in any case.
> 
> ...I will say it's not 100% compatible, though. Patch #2 in this
> series deletes "vreg_s9c_0p676" on pro SKUs. As far as I know, that's
> the only needed change, though.
> 
> -Doug