Re: [PATCH v2 6/9] arm64: dts: qcom: msm8996: Add PSCI cpuidle low power states

[Niklas Cassel]

On Tue, May 21, 2019 at 03:05:16PM +0530, Amit Kucheria wrote:
> Add device bindings for cpuidle states for cpu devices.
> 
> msm8996 features 4 cpus - 2 in each cluster. However, all cpus implement
> the same microarchitecture and the two clusters only differ in the
> maximum frequency attainable by the CPUs.
> 
> Signed-off-by: Amit Kucheria <amit.kucheria@xxxxxxxxxx>
> ---
>  arch/arm64/boot/dts/qcom/msm8996.dtsi | 17 +++++++++++++++++
>  1 file changed, 17 insertions(+)
> 
> diff --git a/arch/arm64/boot/dts/qcom/msm8996.dtsi b/arch/arm64/boot/dts/qcom/msm8996.dtsi
> index c761269caf80..4f2fb7885f39 100644
> --- a/arch/arm64/boot/dts/qcom/msm8996.dtsi
> +++ b/arch/arm64/boot/dts/qcom/msm8996.dtsi
> @@ -95,6 +95,7 @@
>  			compatible = "qcom,kryo";
>  			reg = <0x0 0x0>;
>  			enable-method = "psci";
> +			cpu-idle-states = <&CPU_SLEEP_0>;
>  			next-level-cache = <&L2_0>;
>  			L2_0: l2-cache {
>  			      compatible = "cache";
> @@ -107,6 +108,7 @@
>  			compatible = "qcom,kryo";
>  			reg = <0x0 0x1>;
>  			enable-method = "psci";
> +			cpu-idle-states = <&CPU_SLEEP_0>;
>  			next-level-cache = <&L2_0>;
>  		};
>  
> @@ -115,6 +117,7 @@
>  			compatible = "qcom,kryo";
>  			reg = <0x0 0x100>;
>  			enable-method = "psci";
> +			cpu-idle-states = <&CPU_SLEEP_0>;
>  			next-level-cache = <&L2_1>;
>  			L2_1: l2-cache {
>  			      compatible = "cache";
> @@ -127,6 +130,7 @@
>  			compatible = "qcom,kryo";
>  			reg = <0x0 0x101>;
>  			enable-method = "psci";
> +			cpu-idle-states = <&CPU_SLEEP_0>;
>  			next-level-cache = <&L2_1>;
>  		};
>  
> @@ -151,6 +155,19 @@
>  				};
>  			};
>  		};
> +
> +		idle-states {
> +			entry-method = "psci";
> +
> +			CPU_SLEEP_0: cpu-sleep-0 {
> +				compatible = "arm,idle-state";
> +				idle-state-name = "standalone-power-collapse";
> +				arm,psci-suspend-param = <0x00000004>;
> +				entry-latency-us = <40>;
> +				exit-latency-us = <80>;

Hello Amit,

Looking at this line of code in msm-4.14:
https://source.codeaurora.org/quic/la/kernel/msm-4.14/tree/drivers/cpuidle/lpm-levels.c?h=LA.UM.7.1.r1-14000-sm8150.0#n993

And seeing the equivalent in msm-4.4:
https://source.codeaurora.org/quic/la/kernel/msm-4.4/tree/drivers/cpuidle/lpm-levels.c?h=msm-4.4#n1080

It becomes obvious that

qcom,time-overhead == entry-latency-us + exit-latency-us
and
qcom,latency-us == exit-latency-us

which means that

entry-latency-us == qcom,time-overhead - qcom,latency-us


Using this formula, with the numbers from downstream SDM845:
https://source.codeaurora.org/quic/la/kernel/msm-4.9/tree/arch/arm64/boot/dts/qcom/sdm845-pm.dtsi?h=msm-4.9#n123

qcom,latency-us = <621>;
qcom,time-overhead = <885>;

885 - 621 = 264

we end up with the same values that Raju
has in his submission for upstream SDM845:
https://patchwork.kernel.org/patch/10953253/

entry-latency-us = <264>;
exit-latency-us = <621>;



Which for msm8996:

qcom,latency-us = <80>;
qcom,time-overhead = <210>;

gives:

entry-latency-us = <130>
exit-latency-us = <80>;

> +				min-residency-us = <300>;
> +			};
> +		};
>  	};
>  
>  	thermal-zones {
> -- 
> 2.17.1
>