Re: [PATCH v3 4/7] arm64: dts: qcom: sc8280xp: Add Venus

[Konrad Dybcio]

On 1/26/26 11:47 AM, Dmitry Baryshkov wrote:
> On Mon, Jan 26, 2026 at 11:00:25AM +0100, Konrad Dybcio wrote:
>> On 1/25/26 4:33 PM, Dmitry Baryshkov wrote:
>>> From: Konrad Dybcio <konradybcio@xxxxxxxxxx>
>>>
>>> Add the required nodes to enable Venus on sc8280xp.
>>>
>>> [ bod: added interconnect tags ]
>>>
>>> Signed-off-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxx>
>>> [ johan: use sm8350 videocc defines ]
>>> Signed-off-by: Johan Hovold <johan+linaro@xxxxxxxxxx>
>>> [ bod: dropped video encoder/decoder declarations ]
>>> Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@xxxxxxxxxx>
>>> [ db: dropped llcc icc, switched to sc8280xp compat, added more freqs ]
>>> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxxxxxxxx>
>>> ---
>>
>> [...]
>>
>>> +			venus_opp_table: opp-table {
>>> +				compatible = "operating-points-v2";
>>> +
>>> +				opp-720000000 {
>>> +					opp-hz = /bits/ 64 <720000000>;
>>
>> You need to div3 all of these, otherwise this LOW_SVS OPP
>> will actually apply a more-than-TURBO_L1-requiring frequency..
> 
> Hmm, is the OPP table for SM8250 then also incorrect?

The OPP-set clock is "vcodec0_core"

The DT references VIDEO_CC_MVS0_CLK, which is a branch-child of
video_cc_mvs0_div_clk_src, so that sounds like it makes sense..

And the frequency plan concurs that.

> 
> I cross-checked SM8250 and Kodiak against, msm-4.19 and msm-5.4
> correspondignly, the OPP tables are written in the same way.
> 
>>
>> [...]
>>> +		videocc: clock-controller@abf0000 {
>>> +			compatible = "qcom,sc8280xp-videocc";
>>> +			reg = <0 0x0abf0000 0 0x10000>;
>>> +			clocks = <&rpmhcc RPMH_CXO_CLK>,
>>> +				 <&rpmhcc RPMH_CXO_CLK_A>,
>>> +				 <&sleep_clk>;
>>> +			power-domains = <&rpmhpd SC8280XP_MMCX>;
>>> +			required-opps = <&rpmhpd_opp_low_svs>;
>>
>> AFAICS the PLLs will not turn on if *MX* is at < LOWSVS, but they still
>> need to be scaled 1:1 with MMCX as the freuqency rises, so you need to
>> wire up 2 power domains to the OPP table
> 
> I don't see this being done for SM8250, should it also be fixed? On
> SM8550 we indeed manage MMCX and MX together. Should SC7280 also scale
> both CX and MX together?

8250 yes

On kodiak, the docs strangely say that it's *all* powered by CX, maybe
+Taniya or +Jagadeesh could confirm if that's right

Konrad