Add support for Allwinner H6 DVFS
Subsystems (2)
v3 cover Add support for Allwinner H6 DVFS
Hi Sunxi maintainers and members,
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
This serie is based on Yangtao Li serie[0] and Ondřej Jirman work[1].
Most of the OPP tables are taken from original vendor kernel[2]. Plus there are new CPU frequencies at 1.6GHz, 1.7GHz and 1.8GHz.
I wrote a simple script to randomly set a frequency during a random time[3]. This script is quite stressfull and set some high frequency without checking temperature. This can result on behavior that whould not occurs with the real cpufreq framework. As Maxime point out I also tested with cpufreq-ljt-stress-test (found here https://github.com/ssvb/cpuburn-arm).
This script doesn't trigger any issue.
I also test that that offlining CPU0 and doing DVFS on other CPUs works. As CPU regulator is only set for CPU0.
The GPU devfreq was drop as the regulator is still not properly drive by panfrost driver[4].
I will re-introduce it later.
Ondřej Jirman has an Orange Pi 3, Jernej has a PineH64 and a Tanix TX6 boards and I have a Beelink GS1 board so I have enable these boards. But CPU Devfreq is really touchy has it depends on:
board design, SoC speed_grade and environement which can affect thermal cooling and have different behavior for different user.
If people can test this serie and give feedback, I will try to introduce this in LibreElec tree, so LE community can test it.
Thanks,
Clément
0: https://patchwork.kernel.org/cover/10815117/
1: https://megous.com/git/linux/log/?h=ths-5.7
2: https://github.com/orangepi-xunlong/OrangePiH6_Linux4_9/blob/master/arch/arm64/boot/dts/sunxi/sun50iw6p1.dtsi#L345-L517
3: https://gist.github.com/clementperon/55a055dae3f13bbd14fb39c0069fe2e2
4: https://patchwork.kernel.org/patch/11486893/
Changes since v2 (thanks to Maxime Ripard):
- Change Orange Pi boards to Orange Pi 3
- Change soc speed nvmem node name
- Fix device tree warnings
- Drop GPU opp tables
Changes since v1 (thanks to Ondřej Jirman):
- Remove Polling thermal
- Add Orange Pi boards
- Remove minimal voltage change for Beelink GS1
- Add ramp-deplay for GPU and CPU regulators
- Push to thermal point to 85°C (Allwinner set them to 100°C and 115°C)
- Added 1.6GHz and 1.7GHz to OPP table.
Clément Péron (6):
arm64: configs: Enable sun50i cpufreq nvmem
arm64: dts: allwinner: h6: Enable CPU opp tables for Beelink GS1
arm64: dts: allwinner: h6: Enable CPU opp tables for Orange Pi 3
arm64: dts: allwinner: h6: Enable CPU opp tables for Tanix TX6
arm64: dts: allwinner: Sort Pine H64 device-tree nodes
arm64: dts: allwinner: h6: Enable CPU and GPU opp tables for Pine H64
Ondrej Jirman (1):
arm64: dts: allwinner: h6: Add CPU Operating Performance Points table
.../dts/allwinner/sun50i-h6-beelink-gs1.dts | 9 +- .../boot/dts/allwinner/sun50i-h6-cpu-opp.dtsi | 121 ++++++++++++++++++ .../dts/allwinner/sun50i-h6-orangepi-3.dts | 3 + .../boot/dts/allwinner/sun50i-h6-pine-h64.dts | 41 +++--- .../dts/allwinner/sun50i-h6-tanix-tx6.dts | 13 ++ arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 4 + arch/arm64/configs/defconfig | 1 + 7 files changed, 174 insertions(+), 18 deletions(-) create mode 100644 arch/arm64/boot/dts/allwinner/sun50i-h6-cpu-opp.dtsi
Signature · 1 line
2.20.1
PATCH v3 1/7 arm64: dts: allwinner: h6: Add CPU Operating Performance Points table
From: Ondrej Jirman <[email protected]>
Add an Operating Performance Points table for the CPU cores to enable Dynamic Voltage & Frequency Scaling on the H6.
--- .../boot/dts/allwinner/sun50i-h6-cpu-opp.dtsi | 121 ++++++++++++++++++ arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi | 4 + 2 files changed, 125 insertions(+) create mode 100644 arch/arm64/boot/dts/allwinner/sun50i-h6-cpu-opp.dtsi
arch/arm64/boot/dts/allwinner/sun50i-h6-cpu-opp.dtsi new file +121 -0
@@ -0,0 +1,121 @@ +// SPDX-License-Identifier: (GPL-2.0+ OR MIT) +// Copyright (C) 2020 Ondrej Jirman <[email protected]> +// Copyright (C) 2020 Clément Péron <[email protected]> + +/ { + cpu_opp_table: cpu-opp-table { + compatible = "allwinner,sun50i-h6-operating-points"; + nvmem-cells = <&cpu_speed_grade>; + opp-shared; + + opp@480000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <480000000>; + + opp-microvolt-speed0 = <880000>; + opp-microvolt-speed1 = <820000>; + opp-microvolt-speed2 = <820000>; + }; + + opp@720000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <720000000>; + + opp-microvolt-speed0 = <880000>; + opp-microvolt-speed1 = <820000>; + opp-microvolt-speed2 = <820000>; + }; + + opp@816000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <816000000>; + + opp-microvolt-speed0 = <880000>; + opp-microvolt-speed1 = <820000>; + opp-microvolt-speed2 = <820000>; + }; + + opp@888000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <888000000>; + + opp-microvolt-speed0 = <880000>; + opp-microvolt-speed1 = <820000>; + opp-microvolt-speed2 = <820000>; + }; + + opp@1080000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <1080000000>; + + opp-microvolt-speed0 = <940000>; + opp-microvolt-speed1 = <880000>; + opp-microvolt-speed2 = <880000>; + }; + + opp@1320000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <1320000000>; + + opp-microvolt-speed0 = <1000000>; + opp-microvolt-speed1 = <940000>; + opp-microvolt-speed2 = <940000>; + }; + + opp@1488000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <1488000000>; + + opp-microvolt-speed0 = <1060000>; + opp-microvolt-speed1 = <1000000>; + opp-microvolt-speed2 = <1000000>; + }; + + opp@1608000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <1608000000>; + + opp-microvolt-speed0 = <1090000>; + opp-microvolt-speed1 = <1030000>; + opp-microvolt-speed2 = <1030000>; + }; + + opp@1704000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <1704000000>; + + opp-microvolt-speed0 = <1120000>; + opp-microvolt-speed1 = <1060000>; + opp-microvolt-speed2 = <1060000>; + }; + + opp@1800000000 { + clock-latency-ns = <244144>; /* 8 32k periods */ + opp-hz = /bits/ 64 <1800000000>; + + opp-microvolt-speed0 = <1160000>; + opp-microvolt-speed1 = <1100000>; + opp-microvolt-speed2 = <1100000>; + }; + }; +}; + +&cpu0 { + operating-points-v2 = <&cpu_opp_table>; + #cooling-cells = <2>; +}; + +&cpu1 { + operating-points-v2 = <&cpu_opp_table>; + #cooling-cells = <2>; +}; + +&cpu2 { + operating-points-v2 = <&cpu_opp_table>; + #cooling-cells = <2>; +}; + +&cpu3 { + operating-points-v2 = <&cpu_opp_table>; + #cooling-cells = <2>; +};
arch/arm64/boot/dts/allwinner/sun50i-h6.dtsi +4 -0
@@ -258,6 +258,10 @@ ths_calibration: thermal-sensor-calibration@14 { reg = <0x14 0x8>; }; + + cpu_speed_grade: cpu-speed-grade@1c { + reg = <0x1c 0x4>; + }; }; watchdog: watchdog@30090a0 {
Signature · 1 line
2.20.1
PATCH v3 5/7 arm64: dts: allwinner: h6: Enable CPU opp tables for Tanix TX6
In reply to Clément PéronEnable CPU opp tables for Tanix TX6.
Also add the fixed regulator that provided vdd-cpu-gpu required for CPU opp tables.
This voltage has been found using a voltmeter and could be wrong.
--- .../boot/dts/allwinner/sun50i-h6-tanix-tx6.dts | 13 +++++++++++++ 1 file changed, 13 insertions(+)
arch/arm64/boot/dts/allwinner/sun50i-h6-tanix-tx6.dts +13 -0
@@ -4,6 +4,7 @@ /dts-v1/; #include "sun50i-h6.dtsi" +#include "sun50i-h6-cpu-opp.dtsi" #include <dt-bindings/gpio/gpio.h>
@@ -37,6 +38,17 @@ regulator-min-microvolt = <3300000>; regulator-max-microvolt = <3300000>; }; + + reg_vdd_cpu_gpu: vdd-cpu-gpu { + compatible = "regulator-fixed"; + regulator-name = "vdd-cpu-gpu"; + regulator-min-microvolt = <1135000>; + regulator-max-microvolt = <1135000>; + }; +}; + +&cpu0 { + cpu-supply = <®_vdd_cpu_gpu>; }; &de {
@@ -56,6 +68,7 @@ }; &gpu { + mali-supply = <®_vdd_cpu_gpu>; status = "okay"; };
Signature · 1 line
2.20.1
PATCH v3 3/7 arm64: dts: allwinner: h6: Enable CPU opp tables for Beelink GS1
In reply to Clément PéronEnable CPU opp tables for Beelink GS1.
This needs to change the CPU regulator max voltage to fit the OPP table.
Also add the ramp-delay information to avoid any out of spec running as the regulator is slower at reaching the voltage requested compare to the PLL reaching the frequency.
There is no such information for AXP805 but similar PMIC (AXP813) has a DVM (Dynamic Voltage scaling Management) ramp rate equal to 2500uV/us.
--- arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/allwinner/sun50i-h6-beelink-gs1.dts +8 -1
@@ -4,6 +4,7 @@ /dts-v1/; #include "sun50i-h6.dtsi" +#include "sun50i-h6-cpu-opp.dtsi" #include <dt-bindings/gpio/gpio.h>
@@ -77,6 +78,10 @@ }; }; +&cpu0 { + cpu-supply = <®_dcdca>; +}; + &de { status = "okay"; };
@@ -234,7 +239,8 @@ reg_dcdca: dcdca { regulator-always-on; regulator-min-microvolt = <810000>; - regulator-max-microvolt = <1080000>; + regulator-max-microvolt = <1160000>; + regulator-ramp-delay = <2500>; regulator-name = "vdd-cpu"; };
@@ -242,6 +248,7 @@ regulator-enable-ramp-delay = <32000>; regulator-min-microvolt = <810000>; regulator-max-microvolt = <1080000>; + regulator-ramp-delay = <2500>; regulator-name = "vdd-gpu"; };
Signature · 1 line
2.20.1
PATCH v3 4/7 arm64: dts: allwinner: h6: Enable CPU opp tables for Orange Pi 3
In reply to Clément PéronEnable CPU opp tables for Orange Pi 3.
This needs to change the CPU regulator max voltage to fit the OPP table.
Also add the ramp-delay information to avoid any out of spec running as the regulator is slower at reaching the voltage requested compare to the PLL reaching the frequency.
There is no such information for AXP805 but similar PMIC (AXP813) has a DVM (Dynamic Voltage scaling Management) ramp rate equal to 2500uV/us.
--- arch/arm64/boot/dts/allwinner/sun50i-h6-orangepi-3.dts | 3 +++ 1 file changed, 3 insertions(+)
arch/arm64/boot/dts/allwinner/sun50i-h6-orangepi-3.dts +3 -0
@@ -4,6 +4,7 @@ /dts-v1/; #include "sun50i-h6.dtsi" +#include "sun50i-h6-cpu-opp.dtsi" #include <dt-bindings/gpio/gpio.h>
@@ -257,6 +258,7 @@ regulator-always-on; regulator-min-microvolt = <800000>; regulator-max-microvolt = <1160000>; + regulator-ramp-delay = <2500>; regulator-name = "vdd-cpu"; };
@@ -264,6 +266,7 @@ regulator-enable-ramp-delay = <32000>; regulator-min-microvolt = <810000>; regulator-max-microvolt = <1080000>; + regulator-ramp-delay = <2500>; regulator-name = "vdd-gpu"; };
Signature · 1 line
2.20.1
PATCH v3 7/7 arm64: dts: allwinner: h6: Enable CPU and GPU opp tables for Pine H64
In reply to Clément PéronEnable CPU and GPU opp tables for Pine H64.
Also add the ramp-delay information to avoid any out of spec running as the regulator is slower at reaching the voltage requested compare to the PLL reaching the frequency.
There is no such information for AXP805 but similar PMIC (AXP813) has a DVM (Dynamic Voltage scaling Management) ramp rate equal to 2500uV/us.
--- arch/arm64/boot/dts/allwinner/sun50i-h6-pine-h64.dts | 7 +++++++ 1 file changed, 7 insertions(+)
arch/arm64/boot/dts/allwinner/sun50i-h6-pine-h64.dts +7 -0
@@ -4,6 +4,7 @@ /dts-v1/; #include "sun50i-h6.dtsi" +#include "sun50i-h6-cpu-opp.dtsi" #include <dt-bindings/gpio/gpio.h>
@@ -80,6 +81,10 @@ }; }; +&cpu0 { + cpu-supply = <®_dcdca>; +}; + &de { status = "okay"; };
@@ -239,6 +244,7 @@ regulator-always-on; regulator-min-microvolt = <810000>; regulator-max-microvolt = <1080000>; + regulator-ramp-delay = <2500>; regulator-name = "vdd-cpu"; };
@@ -246,6 +252,7 @@ regulator-enable-ramp-delay = <32000>; regulator-min-microvolt = <810000>; regulator-max-microvolt = <1080000>; + regulator-ramp-delay = <2500>; regulator-name = "vdd-gpu"; };
Signature · 1 line
2.20.1
PATCH v3 6/7 arm64: dts: allwinner: Sort Pine H64 device-tree nodes
In reply to Clément PéronPine H64 device-tree have some nodes not properly sorted.
Fix this.
--- .../boot/dts/allwinner/sun50i-h6-pine-h64.dts | 34 +++++++++---------- 1 file changed, 17 insertions(+), 17 deletions(-)
arch/arm64/boot/dts/allwinner/sun50i-h6-pine-h64.dts +17 -17
@@ -80,6 +80,18 @@ }; }; +&de { + status = "okay"; +}; + +&ehci0 { + status = "okay"; +}; + +&ehci3 { + status = "okay"; +}; + &emac { pinctrl-names = "default"; pinctrl-0 = <&ext_rgmii_pins>;
@@ -91,17 +103,6 @@ status = "okay"; }; -&mdio { - ext_rgmii_phy: ethernet-phy@1 { - compatible = "ethernet-phy-ieee802.3-c22"; - reg = <1>; - }; -}; - -&de { - status = "okay"; -}; - &gpu { mali-supply = <®_dcdcc>; status = "okay";
@@ -117,12 +118,11 @@ }; }; -&ehci0 { - status = "okay"; -}; - -&ehci3 { - status = "okay"; +&mdio { + ext_rgmii_phy: ethernet-phy@1 { + compatible = "ethernet-phy-ieee802.3-c22"; + reg = <1>; + }; }; &mmc0 {
Signature · 1 line
2.20.1
PATCH v3 2/7 arm64: configs: Enable sun50i cpufreq nvmem
In reply to Clément PéronAllwinner H6 needs this driver to be able to get
the correct speed_bin required for DVFS.
Enable this option in arm64 defconfig.
--- arch/arm64/configs/defconfig | 1 + 1 file changed, 1 insertion(+)
arch/arm64/configs/defconfig +1 -0
@@ -84,6 +84,7 @@ CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m CONFIG_CPU_FREQ_GOV_SCHEDUTIL=y CONFIG_CPUFREQ_DT=y CONFIG_ACPI_CPPC_CPUFREQ=m +CONFIG_ARM_ALLWINNER_SUN50I_CPUFREQ_NVMEM=m CONFIG_ARM_ARMADA_37XX_CPUFREQ=y CONFIG_ARM_SCPI_CPUFREQ=y CONFIG_ARM_IMX_CPUFREQ_DT=m
Signature · 1 line
2.20.1
Maxime Ripard
AppliedHi,
Quoted 22 lines · Clément Péron
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
This serie is based on Yangtao Li serie[0] and Ondřej Jirman work[1].
Most of the OPP tables are taken from original vendor kernel[2]. Plus there are new CPU frequencies at 1.6GHz, 1.7GHz and 1.8GHz.
I wrote a simple script to randomly set a frequency during a random time[3]. This script is quite stressfull and set some high frequency without checking temperature. This can result on behavior that whould not occurs with the real cpufreq framework. As Maxime point out I also tested with cpufreq-ljt-stress-test (found here https://github.com/ssvb/cpuburn-arm).
This script doesn't trigger any issue.
I also test that that offlining CPU0 and doing DVFS on other CPUs works. As CPU regulator is only set for CPU0.The GPU devfreq was drop as the regulator is still not properly drive by panfrost driver[4].
I will re-introduce it later.
Quoted 8 lines
Ondřej Jirman has an Orange Pi 3, Jernej has a PineH64 and a Tanix TX6 boards and I have a Beelink GS1 board so I have enable these boards. But CPU Devfreq is really touchy has it depends on:
board design, SoCspeed_gradeand environement which can affect thermal cooling and have different behavior for different user.If people can test this serie and give feedback, I will try to introduce this in LibreElec tree, so LE community can test it.
Applied all of them, thanks!
Maxime
Hello Clément,
Quoted 5 lines · Clément Péron
Hi Sunxi maintainers and members,
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
[ snip ]
Quoted 6 lines
Changes since v2 (thanks to Maxime Ripard):
- Change Orange Pi boards to Orange Pi 3
- Change soc speed nvmem node name
- Fix device tree warnings
- Drop GPU opp tables
Looks like you may have also inadverently dropped the second patch from v2 series that implemented CPU thermal trip points.
Quoted 8 lines
Changes since v1 (thanks to Ondřej Jirman):
- Remove Polling thermal
- Add Orange Pi boards
- Remove minimal voltage change for Beelink GS1
- Add ramp-deplay for GPU and CPU regulators
- Push to thermal point to 85°C (Allwinner set them to 100°C and 115°C)
- Added 1.6GHz and 1.7GHz to OPP table.
Quoted 7 lines
Clément Péron (6):
arm64: configs: Enable sun50i cpufreq nvmem
arm64: dts: allwinner: h6: Enable CPU opp tables for Beelink GS1
arm64: dts: allwinner: h6: Enable CPU opp tables for Orange Pi 3
arm64: dts: allwinner: h6: Enable CPU opp tables for Tanix TX6
arm64: dts: allwinner: Sort Pine H64 device-tree nodes
arm64: dts: allwinner: h6: Enable CPU and GPU opp tables for Pine H64
You may also want to fix title of this patch to drop the GPU reference.
thank you and regards,
o.
Quoted 2 lines
Ondrej Jirman (1):
arm64: dts: allwinner: h6: Add CPU Operating Performance Points table
Quoted 12 lines · megous hlavni
Hello Clément,
Quoted 5 lines · Clément Péron
Hi Sunxi maintainers and members,
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
[ snip ]
Quoted 1 line
Quoted 8 lines
Quoted 5 lines
Changes since v2 (thanks to Maxime Ripard):
- Change Orange Pi boards to Orange Pi 3
- Change soc speed nvmem node name
- Fix device tree warnings
- Drop GPU opp tablesLooks like you may have also inadverently dropped the second patch from v2 series that implemented CPU thermal trip points.
Also it looks like the patch 1 from v2 implementing the clock properties is missing from v3, and I don't see it being already applied anywhere. Without that cpufreq doesn't work.
Also, thermal trip points need to be in the opp.dtsi to avoid dtc warnings during build.
https://megous.com/git/linux/commit/?h=ths-5.7&id=cacefd7decf5ae0ce42ab4d48a13a58552929ebd
regards,
o.
Quoted 23 lines
Quoted 8 lines
Changes since v1 (thanks to Ondřej Jirman):
- Remove Polling thermal
- Add Orange Pi boards
- Remove minimal voltage change for Beelink GS1
- Add ramp-deplay for GPU and CPU regulators
- Push to thermal point to 85°C (Allwinner set them to 100°C and 115°C)
- Added 1.6GHz and 1.7GHz to OPP table.Quoted 7 lines
Clément Péron (6):
arm64: configs: Enable sun50i cpufreq nvmem
arm64: dts: allwinner: h6: Enable CPU opp tables for Beelink GS1
arm64: dts: allwinner: h6: Enable CPU opp tables for Orange Pi 3
arm64: dts: allwinner: h6: Enable CPU opp tables for Tanix TX6
arm64: dts: allwinner: Sort Pine H64 device-tree nodes
arm64: dts: allwinner: h6: Enable CPU and GPU opp tables for Pine H64You may also want to fix title of this patch to drop the GPU reference.
thank you and regards,
o.Quoted 2 lines
Ondrej Jirman (1):
arm64: dts: allwinner: h6: Add CPU Operating Performance Points table
Hi Maxime,
Quoted 31 lines · Maxime Ripard
Hi,
Quoted 28 lines · Clément Péron
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
This serie is based on Yangtao Li serie[0] and Ondřej Jirman work[1].
Most of the OPP tables are taken from original vendor kernel[2]. Plus there are new CPU frequencies at 1.6GHz, 1.7GHz and 1.8GHz.
I wrote a simple script to randomly set a frequency during a random time[3]. This script is quite stressfull and set some high frequency without checking temperature. This can result on behavior that whould not occurs with the real cpufreq framework. As Maxime point out I also tested with cpufreq-ljt-stress-test (found here https://github.com/ssvb/cpuburn-arm).
This script doesn't trigger any issue.
I also test that that offlining CPU0 and doing DVFS on other CPUs works. As CPU regulator is only set for CPU0.The GPU devfreq was drop as the regulator is still not properly drive by panfrost driver[4].
I will re-introduce it later.Ondřej Jirman has an Orange Pi 3, Jernej has a PineH64 and a Tanix TX6 boards and I have a Beelink GS1 board so I have enable these boards. But CPU Devfreq is really touchy has it depends on:
board design, SoCspeed_gradeand environement which can affect thermal cooling and have different behavior for different user.
Quoted 4 lines
Quoted 2 lines
If people can test this serie and give feedback, I will try to introduce this in LibreElec tree, so LE community can test it.
Applied all of them, thanks!
Please also apply "[PATCH v2 1/7] arm64: dts: allwinner: h6: Add clock to CPU cores" from the v2 series, otherwise cpufreq will not work.
I can also send a missing patch adding the trip points, and cpu as a cooling device, that I linked in my other reply to this patch series afterwards, if Clément wants.
regards,
o.
Quoted 1 line
Maxime
Hi Ondrej, Maxime,
Quoted 40 lines · Ondřej Jirman
Hi Maxime,
Quoted 31 lines · Maxime Ripard
Hi,
Quoted 28 lines · Clément Péron
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
This serie is based on Yangtao Li serie[0] and Ondřej Jirman work[1].
Most of the OPP tables are taken from original vendor kernel[2]. Plus there are new CPU frequencies at 1.6GHz, 1.7GHz and 1.8GHz.
I wrote a simple script to randomly set a frequency during a random time[3]. This script is quite stressfull and set some high frequency without checking temperature. This can result on behavior that whould not occurs with the real cpufreq framework. As Maxime point out I also tested with cpufreq-ljt-stress-test (found here https://github.com/ssvb/cpuburn-arm).
This script doesn't trigger any issue.
I also test that that offlining CPU0 and doing DVFS on other CPUs works. As CPU regulator is only set for CPU0.The GPU devfreq was drop as the regulator is still not properly drive by panfrost driver[4].
I will re-introduce it later.Ondřej Jirman has an Orange Pi 3, Jernej has a PineH64 and a Tanix TX6 boards and I have a Beelink GS1 board so I have enable these boards. But CPU Devfreq is really touchy has it depends on:
board design, SoCspeed_gradeand environement which can affect thermal cooling and have different behavior for different user.Quoted 4 lines
Quoted 2 lines
If people can test this serie and give feedback, I will try to introduce this in LibreElec tree, so LE community can test it.
Applied all of them, thanks!
Quoted 7 lines
Please also apply "[PATCH v2 1/7] arm64: dts: allwinner: h6: Add clock to CPU cores" from the v2 series, otherwise cpufreq will not work.
I can also send a missing patch adding the trip points, and cpu as a cooling device, that I linked in my other reply to this patch series afterwards, if Clément wants.
Indeed I have sent using the following cmd : "git send-email --to=XXXX HEAD~7" from the previous version.
I should had do Instead: "git send-email --to=XXXX next/master "
Sorry for that :(
Clément
Quoted 7 lines
regards,
o.Quoted 1 line
Maxime
Quoted 51 lines · Clément Péron
Hi Ondrej, Maxime,
Quoted 40 lines · Ondřej Jirman
Hi Maxime,
Quoted 31 lines · Maxime Ripard
Hi,
Quoted 28 lines · Clément Péron
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
This serie is based on Yangtao Li serie[0] and Ondřej Jirman work[1].
Most of the OPP tables are taken from original vendor kernel[2]. Plus there are new CPU frequencies at 1.6GHz, 1.7GHz and 1.8GHz.
I wrote a simple script to randomly set a frequency during a random time[3]. This script is quite stressfull and set some high frequency without checking temperature. This can result on behavior that whould not occurs with the real cpufreq framework. As Maxime point out I also tested with cpufreq-ljt-stress-test (found here https://github.com/ssvb/cpuburn-arm).
This script doesn't trigger any issue.
I also test that that offlining CPU0 and doing DVFS on other CPUs works. As CPU regulator is only set for CPU0.The GPU devfreq was drop as the regulator is still not properly drive by panfrost driver[4].
I will re-introduce it later.Ondřej Jirman has an Orange Pi 3, Jernej has a PineH64 and a Tanix TX6 boards and I have a Beelink GS1 board so I have enable these boards. But CPU Devfreq is really touchy has it depends on:
board design, SoCspeed_gradeand environement which can affect thermal cooling and have different behavior for different user.Quoted 4 lines
Quoted 2 lines
If people can test this serie and give feedback, I will try to introduce this in LibreElec tree, so LE community can test it.
Applied all of them, thanks!
Quoted 7 lines
Please also apply "[PATCH v2 1/7] arm64: dts: allwinner: h6: Add clock to CPU cores" from the v2 series, otherwise cpufreq will not work.
I can also send a missing patch adding the trip points, and cpu as a cooling device, that I linked in my other reply to this patch series afterwards, if Clément wants.
Quoted 6 lines
Indeed I have sent using the following cmd : "git send-email --to=XXXX HEAD~7" from the previous version.
I should had do Instead: "git send-email --to=XXXX next/master "
Sorry for that :(
I haven't pushed anything yet, can you resend the proper branch?
Thanks!
Maxime
Hi Maxime
Quoted 53 lines · Maxime Ripard
Quoted 51 lines · Clément Péron
Hi Ondrej, Maxime,
Quoted 47 lines · Ondřej Jirman
Hi Maxime,
Quoted 31 lines · Maxime Ripard
Hi,
Quoted 28 lines · Clément Péron
Now that required drivers are merged we can contibute on DVFS support for Allwinner H6.
This serie is based on Yangtao Li serie[0] and Ondřej Jirman work[1].
Most of the OPP tables are taken from original vendor kernel[2]. Plus there are new CPU frequencies at 1.6GHz, 1.7GHz and 1.8GHz.
I wrote a simple script to randomly set a frequency during a random time[3]. This script is quite stressfull and set some high frequency without checking temperature. This can result on behavior that whould not occurs with the real cpufreq framework. As Maxime point out I also tested with cpufreq-ljt-stress-test (found here https://github.com/ssvb/cpuburn-arm).
This script doesn't trigger any issue.
I also test that that offlining CPU0 and doing DVFS on other CPUs works. As CPU regulator is only set for CPU0.The GPU devfreq was drop as the regulator is still not properly drive by panfrost driver[4].
I will re-introduce it later.Ondřej Jirman has an Orange Pi 3, Jernej has a PineH64 and a Tanix TX6 boards and I have a Beelink GS1 board so I have enable these boards. But CPU Devfreq is really touchy has it depends on:
board design, SoCspeed_gradeand environement which can affect thermal cooling and have different behavior for different user.Quoted 4 lines
Quoted 2 lines
If people can test this serie and give feedback, I will try to introduce this in LibreElec tree, so LE community can test it.
Applied all of them, thanks!
Please also apply "[PATCH v2 1/7] arm64: dts: allwinner: h6: Add clock to CPU cores" from the v2 series, otherwise cpufreq will not work.
I can also send a missing patch adding the trip points, and cpu as a cooling device, that I linked in my other reply to this patch series afterwards, if Clément wants.
Quoted 8 lines
Quoted 6 lines
Indeed I have sent using the following cmd : "git send-email --to=XXXX HEAD~7" from the previous version.
I should had do Instead: "git send-email --to=XXXX next/master "
Sorry for that :(
I haven't pushed anything yet, can you resend the proper branch?
Yes, I will,
Thanks!
Quoted 3 lines
Thanks!
Maxime