linux-kernel - [PATCH] arm64: dts: qcom: sm8250: correct dynamic power coefficients

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 15 Jun 2023 17:48:52 +0200
From:   Vincent Guittot <vincent.guittot@...aro.org>
To:     agross@...nel.org, andersson@...nel.org, konrad.dybcio@...aro.org,
        robh+dt@...nel.org, krzysztof.kozlowski+dt@...aro.org,
        conor+dt@...nel.org, danny@...ag0n.dev,
        linux-arm-msm@...r.kernel.org, devicetree@...r.kernel.org,
        linux-kernel@...r.kernel.org, srinivas.kandagatla@...aro.org
Cc:     Vincent Guittot <vincent.guittot@...aro.org>
Subject: [PATCH] arm64: dts: qcom: sm8250: correct dynamic power coefficients

sm8250 faces the same problem with its Energy Model as sdm845. The energy
cost of LITTLE cores is reported to be higher than medium or big cores

EM computes the energy with formula:

energy = OPP's cost / maximum cpu capacity * utilization

On v6.4-rc6 we have:
max capacity of CPU0 = 284
capacity of CPU0's OPP(1612800 Hz) = 253
cost of CPU0's OPP(1612800 Hz) = 191704

max capacity of CPU4 = 871
capacity of CPU4's OPP(710400 Hz) = 255
cost of CPU4's OPP(710400 Hz) = 343217

Both OPPs have almost the same compute capacity but the estimated energy
per unit of utilization will be estimated to:

energy CPU0 = 191704 / 284 * 1 = 675
energy CPU4 = 343217 / 871 * 1 = 394

EM estimates that little CPU0 will consume 71% more than medium CPU4 for
the same compute capacity. According to [1], little consumes 25% less than
medium core for Coremark benchmark at those OPPs for the same duration.

Set the dynamic-power-coefficient of CPU0-3 to 105 to fix the energy model
for little CPUs.

[1] https://github.com/kdrag0n/freqbench/tree/master/results/sm8250/k30s

Fixes: 6aabed5526ee ("arm64: dts: qcom: sm8250: Add CPU capacities and energy model")
Signed-off-by: Vincent Guittot <vincent.guittot@...aro.org>
---

Srini gave me access to sm8250 rb5 board so I checked the EM and found
that it has similar problem as sdm845.

 arch/arm64/boot/dts/qcom/sm8250.dtsi | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/boot/dts/qcom/sm8250.dtsi b/arch/arm64/boot/dts/qcom/sm8250.dtsi
index 7bea916900e2..4b0835109b01 100644
--- a/arch/arm64/boot/dts/qcom/sm8250.dtsi
+++ b/arch/arm64/boot/dts/qcom/sm8250.dtsi
@@ -100,7 +100,7 @@ CPU0: cpu@0 {
 			clocks = <&cpufreq_hw 0>;
 			enable-method = "psci";
 			capacity-dmips-mhz = <448>;
-			dynamic-power-coefficient = <205>;
+			dynamic-power-coefficient = <105>;
 			next-level-cache = <&L2_0>;
 			power-domains = <&CPU_PD0>;
 			power-domain-names = "psci";
@@ -131,7 +131,7 @@ CPU1: cpu@100 {
 			clocks = <&cpufreq_hw 0>;
 			enable-method = "psci";
 			capacity-dmips-mhz = <448>;
-			dynamic-power-coefficient = <205>;
+			dynamic-power-coefficient = <105>;
 			next-level-cache = <&L2_100>;
 			power-domains = <&CPU_PD1>;
 			power-domain-names = "psci";
@@ -156,7 +156,7 @@ CPU2: cpu@200 {
 			clocks = <&cpufreq_hw 0>;
 			enable-method = "psci";
 			capacity-dmips-mhz = <448>;
-			dynamic-power-coefficient = <205>;
+			dynamic-power-coefficient = <105>;
 			next-level-cache = <&L2_200>;
 			power-domains = <&CPU_PD2>;
 			power-domain-names = "psci";
@@ -181,7 +181,7 @@ CPU3: cpu@300 {
 			clocks = <&cpufreq_hw 0>;
 			enable-method = "psci";
 			capacity-dmips-mhz = <448>;
-			dynamic-power-coefficient = <205>;
+			dynamic-power-coefficient = <105>;
 			next-level-cache = <&L2_300>;
 			power-domains = <&CPU_PD3>;
 			power-domain-names = "psci";
-- 
2.34.1