Re: [PATCH] arm64: dts: qcom: add missing cache node for cpu1

[Konrad Dybcio]

On 10.01.2025 10:49 AM, Neil Armstrong wrote:
> On 10/01/2025 10:44, Neil Armstrong wrote:
>> On 09/01/2025 19:30, Konrad Dybcio wrote:
>>> On 9.01.2025 4:24 PM, Neil Armstrong wrote:
>>>> Add the missing l2-cache node for the cpu1
>>>>
>>>> Fixes: 20eb2057b3e4 ("arm64: dts: qcom: sm8650: change labels to lower-case")
>>>> Fixes: d2350377997f ("arm64: dts: qcom: add initial SM8650 dtsi")
>>>> Signed-off-by: Neil Armstrong <neil.armstrong@xxxxxxxxxx>
>>>> ---
>>>
>>> subject: missing `sm8650:`
>>
>> Damn
>>
>>>
>>>>   arch/arm64/boot/dts/qcom/sm8650.dtsi | 7 +++++++
>>>>   1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/boot/dts/qcom/sm8650.dtsi b/arch/arm64/boot/dts/qcom/sm8650.dtsi
>>>> index 86684cb9a9325618ddb74458621cf4bbdc1cc0d1..832f3a2c400e8348847bc24b27397e2a0dc08db8 100644
>>>> --- a/arch/arm64/boot/dts/qcom/sm8650.dtsi
>>>> +++ b/arch/arm64/boot/dts/qcom/sm8650.dtsi
>>>> @@ -119,6 +119,13 @@ cpu1: cpu@100 {
>>>>               qcom,freq-domain = <&cpufreq_hw 0>;
>>>>               #cooling-cells = <2>;
>>>> +
>>>> +            l2_100: l2-cache {
>>>> +                compatible = "cache";
>>>> +                cache-level = <2>;
>>>> +                cache-unified;
>>>> +                next-level-cache = <&l3_0>;
>>>> +            };
>>>>           };
>>>
>>> You likely wanted to hook up this new node to CPU1 as well.
>>>
>>> Reading some Arm docs [1], it seems like with A520 specifically, both shared
>>> and unique cache slices are permitted, depending on whether they're
>>> implemented as single- or dual-core complexes (not to be confused with
>>> multi-threading)
>>>
>>> [2] suggests CA720s always have their own cache pools
>>>
>>> In 8650's case, the slowest cluster has a shared L2 cache, whereas cores 2-7
>>> have their own pools, so this patch is incorrect.
>>
>> Damn you're right, so the cpu1 cache should be linked to the cpu0 cache somehow
> 
> Well, stupid me, it's already done... sorry for the noise and thx for your review

cpu3 should still get its own one

Konrad