Re: [PATCH 2/3] arm64: dts: qcom: Introduce Eliza Soc base dtsi

From: Konrad Dybcio

Date: Tue Mar 10 2026 - 10:22:33 EST


On 3/6/26 10:02 AM, Krzysztof Kozlowski wrote:
> On 24/02/2026 14:06, Konrad Dybcio wrote:
>> On 2/24/26 1:13 PM, Abel Vesa wrote:
>>> Introduce the initial support for the Qualcomm Eliza SoC.
>>> It is a high-tier SoC designed for mobile platforms.
>>>
>>> The initial submission enables support for:
>>> - CPU nodes with cpufreq and cpuidle support
>>> - Global Clock Controller (GCC)
>>> - Resource State Coordinator (RSC) with clock controller & genpd provider
>>> - Interrupt controller
>>> - Power Domain Controller (PDC)
>>> - Vendor specific SMMU
>>> - SPMI bus arbiter
>>> - Top Control and Status Register (TCSR)
>>> - Top Level Mode Multiplexer (TLMM)
>>> - Debug UART
>>> - Reserved memory nodes
>>> - Interconnect providers
>>> - System timer
>>> - UFS
>>>
>>> Co-developed-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxxxxxxxx>
>>> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxxxxxxxx>
>>> Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx>
>>> ---
>>
>> [...]
>>
>>> + cpu-map {
>>> + cluster0 {
>>> + core0 {
>>> + cpu = <&cpu0>;
>>
>> The values of the MPIDR register (also present in 'reg' of CPU nodes)
>> suggest all these CPUs form a single logical cluster
>
> For reference:
>
> Data presented by firmware lies and different, e.g. fixed, firmware will
> give you different results. All resources I found (several of them -
> multiple docs, Block diagram from datasheet) say these are separate
> clusters, so this is the topology we should be using currently.

FWIW we discussed *MIDR*_EL1 (i.e. ""CPU ID"") with regards to the firmware
lying. M*P*IDR represents the position of the core within the topology, and
many, many recent Arm platforms have been reporting a single big cluster
because of:

https://developer.arm.com/documentation/100453/0401/The-DynamIQ-Shared-Unit/About-the-DSU

We tended to agree with that and you'll notice almost all recent platforms
have all the cores in a single cluster. I don't know if it was the right
decision, I believe all these years no one cared to check/benchmark the
impact that makes (if any)

Konrad