diff mbox series

[6/7] arm64: dts: juno: Add cpu dynamic-power-coefficient information

Message ID 20190128165522.31749-7-quentin.perret@arm.com (mailing list archive)
State Changes Requested, archived
Headers show
Series Register an Energy Model for Arm reference platforms | expand

Commit Message

Quentin Perret Jan. 28, 2019, 4:55 p.m. UTC
From: Dietmar Eggemann <dietmar.eggemann@arm.com>

A CPUfreq driver, like the scpi driver used on Juno boards, which
provide the Energy Model with power cost information via the PM_OPP
of_dev_pm_opp_get_cpu_power() function, do need the
dynamic-power-coefficient (C) in the device tree.

Method used to obtain the C value:
C is computed by measuring energy (E) consumption of a frequency domain
(FD) over a 10s runtime (t) sysbench workload running at each Operating
Performance Point (OPP) affine to 1 or 2 CPUs of that FD while the other
CPUs of the system are hotplugged out.
By definition all CPUs of a FD have the the same micro-architecture. An
OPP is characterized by a certain frequency (f) and voltage (V) value.
The corresponding power values (P) are calculated by dividing the delta
of the E values between the runs with 2 and 1 CPUs by t.

With n data tuples (P, f, V), n equal to number of OPPs for this
frequency domain, we can solve C by:

P = Pstat + Pdyn

P = Pstat + CV²f

Cx = (Px - P1)/(Vx²fx - V1²f1) with x = {2, ..., n}

The C value is the arithmetic mean out of {C2, ..., Cn}.

Since DVFS is broken on Juno r1, no dynamic-power-coefficient
information has been added to its dts file.

Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Quentin Perret <quentin.perret@arm.com>
---
 arch/arm64/boot/dts/arm/juno-r2.dts | 6 ++++++
 arch/arm64/boot/dts/arm/juno.dts    | 6 ++++++
 2 files changed, 12 insertions(+)

Comments

Sudeep Holla Jan. 29, 2019, 3:27 p.m. UTC | #1
On Mon, Jan 28, 2019 at 04:55:21PM +0000, Quentin Perret wrote:
> From: Dietmar Eggemann <dietmar.eggemann@arm.com>
>
> A CPUfreq driver, like the scpi driver used on Juno boards, which
> provide the Energy Model with power cost information via the PM_OPP
> of_dev_pm_opp_get_cpu_power() function, do need the
> dynamic-power-coefficient (C) in the device tree.
>
> Method used to obtain the C value:
> C is computed by measuring energy (E) consumption of a frequency domain
> (FD) over a 10s runtime (t) sysbench workload running at each Operating
> Performance Point (OPP) affine to 1 or 2 CPUs of that FD while the other
> CPUs of the system are hotplugged out.
> By definition all CPUs of a FD have the the same micro-architecture. An
> OPP is characterized by a certain frequency (f) and voltage (V) value.
> The corresponding power values (P) are calculated by dividing the delta
> of the E values between the runs with 2 and 1 CPUs by t.
>
> With n data tuples (P, f, V), n equal to number of OPPs for this
> frequency domain, we can solve C by:
>
> P = Pstat + Pdyn
>
> P = Pstat + CV²f
>
> Cx = (Px - P1)/(Vx²fx - V1²f1) with x = {2, ..., n}
>
> The C value is the arithmetic mean out of {C2, ..., Cn}.
>
> Since DVFS is broken on Juno r1, no dynamic-power-coefficient
> information has been added to its dts file.
>

Since the binding for "dynamic-power-coefficient" property already exist,
and I don't see any dependency for this and the next patch(TC2) on this
series, I will apply them. Please shout if for any reason that's not true.

--
Regards,
Sudeep
Quentin Perret Jan. 30, 2019, 10:23 a.m. UTC | #2
On Tuesday 29 Jan 2019 at 15:27:35 (+0000), Sudeep Holla wrote:
> On Mon, Jan 28, 2019 at 04:55:21PM +0000, Quentin Perret wrote:
> > From: Dietmar Eggemann <dietmar.eggemann@arm.com>
> >
> > A CPUfreq driver, like the scpi driver used on Juno boards, which
> > provide the Energy Model with power cost information via the PM_OPP
> > of_dev_pm_opp_get_cpu_power() function, do need the
> > dynamic-power-coefficient (C) in the device tree.
> >
> > Method used to obtain the C value:
> > C is computed by measuring energy (E) consumption of a frequency domain
> > (FD) over a 10s runtime (t) sysbench workload running at each Operating
> > Performance Point (OPP) affine to 1 or 2 CPUs of that FD while the other
> > CPUs of the system are hotplugged out.
> > By definition all CPUs of a FD have the the same micro-architecture. An
> > OPP is characterized by a certain frequency (f) and voltage (V) value.
> > The corresponding power values (P) are calculated by dividing the delta
> > of the E values between the runs with 2 and 1 CPUs by t.
> >
> > With n data tuples (P, f, V), n equal to number of OPPs for this
> > frequency domain, we can solve C by:
> >
> > P = Pstat + Pdyn
> >
> > P = Pstat + CV²f
> >
> > Cx = (Px - P1)/(Vx²fx - V1²f1) with x = {2, ..., n}
> >
> > The C value is the arithmetic mean out of {C2, ..., Cn}.
> >
> > Since DVFS is broken on Juno r1, no dynamic-power-coefficient
> > information has been added to its dts file.
> >
> 
> Since the binding for "dynamic-power-coefficient" property already exist,
> and I don't see any dependency for this and the next patch(TC2) on this
> series, I will apply them. Please shout if for any reason that's not true.

Thanks Sudeep !
Quentin
diff mbox series

Patch

diff --git a/arch/arm64/boot/dts/arm/juno-r2.dts b/arch/arm64/boot/dts/arm/juno-r2.dts
index ab77adb4f3c2..66f0ec79c864 100644
--- a/arch/arm64/boot/dts/arm/juno-r2.dts
+++ b/arch/arm64/boot/dts/arm/juno-r2.dts
@@ -99,6 +99,7 @@ 
 			clocks = <&scpi_dvfs 0>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <1024>;
+			dynamic-power-coefficient = <450>;
 		};
 
 		A72_1: cpu@1 {
@@ -116,6 +117,7 @@ 
 			clocks = <&scpi_dvfs 0>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <1024>;
+			dynamic-power-coefficient = <450>;
 		};
 
 		A53_0: cpu@100 {
@@ -133,6 +135,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <485>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A53_1: cpu@101 {
@@ -150,6 +153,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <485>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A53_2: cpu@102 {
@@ -167,6 +171,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <485>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A53_3: cpu@103 {
@@ -184,6 +189,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <485>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A72_L2: l2-cache0 {
diff --git a/arch/arm64/boot/dts/arm/juno.dts b/arch/arm64/boot/dts/arm/juno.dts
index 08d4ba1716c3..9890afdda77b 100644
--- a/arch/arm64/boot/dts/arm/juno.dts
+++ b/arch/arm64/boot/dts/arm/juno.dts
@@ -98,6 +98,7 @@ 
 			clocks = <&scpi_dvfs 0>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <1024>;
+			dynamic-power-coefficient = <530>;
 		};
 
 		A57_1: cpu@1 {
@@ -115,6 +116,7 @@ 
 			clocks = <&scpi_dvfs 0>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <1024>;
+			dynamic-power-coefficient = <530>;
 		};
 
 		A53_0: cpu@100 {
@@ -132,6 +134,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <578>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A53_1: cpu@101 {
@@ -149,6 +152,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <578>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A53_2: cpu@102 {
@@ -166,6 +170,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <578>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A53_3: cpu@103 {
@@ -183,6 +188,7 @@ 
 			clocks = <&scpi_dvfs 1>;
 			cpu-idle-states = <&CPU_SLEEP_0 &CLUSTER_SLEEP_0>;
 			capacity-dmips-mhz = <578>;
+			dynamic-power-coefficient = <140>;
 		};
 
 		A57_L2: l2-cache0 {