Power management for in-vehicle infotainment systems

Transcription

1 Automotive Linux Summit 2017 Power management for in-vehicle infotainment systems 2017/05/31 Takahiko Gomi Automotive Information Solution Business Division Renesas Electronics Corporation 1

2 Who am I? Name:Takahiko Gomi Company:Renesas Electronics corp. Career:10 years experiences in power management development Estimation of power consumption for Mobile use-cases Promotion activity and market analysis in Mobile platform Development of thermal control with power management for in-vehicle infotainment systems 2

3 Today s GOAL Introducing our power management solution for in-vehicle infotainment systems Two solutions of those three our power management solutions introduce today, are already contributed to AGL DD. 3

4 Market trend of the IVI systems Now:Increasing running application In former:independent systems Car navigation system, Audio/Video feature, Analog meter cluster, Rear-view monitor, etc. Connectivity for smart-phones, Enriched graphical monitor on meter cluster systems, Surround view camera features, Smart camera features, etc. It changed its roles and quantities of the applications on the system to improve safety and user experiences. 4

5 In the future Augmented Reality cooperating with cloud-system Back mirror from camera feature Graphical meter cluster systems Information display with cloud-system Voice recognition as user interface via cloud-system It would be integrated to Connected car in cooperation with the cloud system. 5

6 Corresponding scope of Renesas Applications Depends on communities Window Manager Frameworks Sound Manager Camera Smartphone Link ARM trusted firmware Boot Loader BSP Kernel Hypervisor Device Drivers Depends on silicon vendors Hardware 6

7 Major key technology for IVI system High performance to improve parallel execution facilities Virtualization technology to have different types of systems on a single SoC System healthiness to prevent troubles on heavy thermal conditions IVI system Functional safety to lead system state to safe condition Improved user experience to provide good user interface Secure environment to protect systems from threats Scope of presentation 7

8 Efforts on power management in key technology for IVI system High performance to improve parallel execution facilities To process huge amount of running application System healthiness to prevent troubles on heavy thermal conditions To keep thermal condition within the specification Improved user experience to provide good user interface To keep system requirement of boot time More multiple computation Multi-processing with big.little methodology Reliable thermal control Controlling performance by power management features Sophisticated user experience fast boot feature using Suspend to RAM 8

9 Solution for more multiple computation High performance to improve parallel execution facilities To process huge amount of running application System healthiness to prevent troubles on heavy thermal conditions To keep thermal condition within the specification Improved user experience to provide good user interface To keep system requirement of boot time More multiple computation Multi-processing with big.little methodology Reliable thermal control Controlling performance by power management features Sophisticated user experience fast boot feature using Suspend to RAM 9

10 Background for higher performance requirement Trend of IVI systems Many type of application would be implemented. Several type of service would be simultaneously executed. IVI Use-Cases Improvement along trend of IVI systems In-vehicle network Camera features Connectivity for USB devices Multi Core Solution is required on latest multiple application environment. GPS, etc IVI system Connectivity for smart-phones Meter cluster Car navigation High-resolution video player 10 -A15 DDR3- SDRAM 25 DMIPS CPU Perf R-Car H2 170 GFLOPS GPU Perf -A7 PowerVR G GB/s Maximum memory bandwidth -A57 DDR4- SDRAM 40 DMIPS CPU Perf R-Car H3 460 GFLOPS GPU Perf -A53 PowerVR GX GB/s Maximum memory bandwidth

11 Multi Core Solution Only high performance big CPUs A57 A57 High performance High power consumption A57 big.little architecture A57 High performance A57 A57 Large HW footprint A53 A53 A53 A53 Low power consumption Smaller HW footprint Only high efficiency LITTLE CPUs A53 A53 A53 A53 A53 A53 A53 A53 Low performance Low power consumption Smaller HW footprint R-Car Starter Kit Gen3 big.little realizes appropriate combination balance of the performance, hardware footprint and power consumptions. 11

12 big.little Overview Hardware Structure Interrupt Controller Power performance curve Power A57 A57 A53 A53 A53 A53 A57 L2 cache L2 cache CCI (Cache Coherent Interconnect) A53 Performance The big CPU such as -A57 has higher single thread performance. The LITTLE CPU such as -A53 has better power consumptions statistics. 12

13 Necessity of big.little Software Non big.little Software big.little Software Application Music light Navigation heavy Application Music light Navigation heavy Software (BSP) big.little Software Software (BSP) big.little Software Hardware A57 A53 Hardware A57 A53 Performance instability Best practice!! In big.little SW, heavy weight process would be assigned big CPU core. And any other light process would be assigned LITTLE CPU core. 13

14 big.little Software Overview Energy Aware Scheduling(EAS) EAS have two characteristics : It assigns applications using process load statistics. User High Load task A Linux Kernel Middle Load task B Low Load task C Low Load task D It reduces power consumptions by utilizing light weight LITTLE core as possible. Scheduler EAS CPU Capacity A57 cpu0 cpu1 cpu2 cpu3 14

15 big.little - EAS Issue Lower performance at starting application Video playback Playing Lower response of application manipulation Connectivity for smart-phones Operating Load of task START END Load of task Plug in Plug out max Switch CPU by increased task load max Sporadic peak performance min A53 A57 Time min A53 Time Solution It can use taskset or cgroup features to assign such heavy weight applications into big CPU as designer s will. When we use original EAS on IVI systems, it have to solve such issues on it. 15

16 Renesas s future activity for big.little Renesas is developing additional tools which makes easy to assign applications with some hint or static enforcement. Renesas is continuing evaluation of EAS under AGL UCB DD. Renesas will promote such additional feature and tools to our BSP and contribute to AGL community in near future. Renesas will provide solution which able to used by everyone on AGL community utilize full performance result of big.little hardware architecture. 16

17 Solution for reliable thermal control High performance to improve parallel execution facilities To process huge amount of running application System healthiness to prevent troubles on heavy thermal conditions To keep thermal condition within the specification Improved user experience to provide good user interface To keep system requirement of boot time More multiple computation Multi-processing with big.little methodology Reliable thermal control Controlling performance by power management features Sophisticated user experience fast boot feature using Suspend to RAM 17

18 Impact by increased power consumption It s consuming more electricity accordance as its computing performance improvement When huge heat is generated by increased power consumption, Impact on lifespan margin Impact on failure rate R-Car H3 Thermal -A57 DDR4- SDRAM -A53 PowerVR GX6650 Impact on failure rate 25 DMIPS -A15 DDR3- SDRAM R-Car H2 170 GFLOPS -A7 PowerVR G GB/s 40 DMIPS 460 GFLOPS 50 GB/s Impact on Lifespan margin CPU Perf GPU Perf Maximum Power memory consumption bandwidth CPU Perf GPU Perf Maximum memory bandwidth 18 Power consumption Power consumption thermal control feature is needed to prevent to keeping lifespan margin and protecting failure rate.

19 How to approach to reduce thermal To prevent thermal runaway...? High cost! HW approach Thermal No heat radiation Impact on failure rate Impact on Lifespan margin With heat radiation Heat radiation devices Air-flow aware case design etc. SW approach HW Active cooling fan Huge fin HW SW 小型ファン 小型の放熱器 Heat-resistant temperature No thermal control With thermal control Clock control Power switching Lack of safety High cost Low cost 19 No heat radiation HW approach HW+SW approach Software approach is needed as it reduce heat generation just when it exceeded capability of hardware cooling devices.

20 How to approach to reduce thermal To prevent thermal runaway...? High cost! HW approach Thermal No heat radiation Impact oo failure rate Impact on Lifespan margin With heat radiation Heat radiation devices Air-flow aware case design etc. SW approach HW Active cooling fan Huge fin HW SW Small fan Small fin Heat-resistant temperature No thermal control With thermal control Clock control Power switching Lack of safety High cost Low cost Software approach is needed as it reduce heat generation just when it exceeded capability of hardware cooling devices. 20 No heat radiation HW approach HW+SW approach

21 The methodology of the thermal control feature software design Power consumption Software approach High frequency Low frequency Dynamic power Clock control (Voltage control) Clock Dynamic power Reduce ON OFF Leak power Power switching Power domain Leak power Reduce Software approach is able to reduce heat generation by declining power consumption with clock control and or power switching. 21

22 IPA(Intelligent power allocation) for software approach Power consumption Software approach High frequency Low frequency Dynamic power Clock control (Voltage control) Clock Dynamic power Reduce ON OFF Leak power Power switching Power domain Leak power Reduce 22

23 What is IPA? Input Output Referring thermal values Conversion heat into power Control power Conversion power into clock Clock control Voltage control Sensing Estimation Control Thermal Driver IPA CPU Freq IPA reduces dynamic power and keep performance as maximum as possible under the thermal condition. 23

24 EMS(Emergency shutdown ) for software approach Power consumption Software approach High frequency Low frequency Dynamic power Clock control (Voltage control) Clock Dynamic power Reduce ON OFF Leak power Power switching Power domain Leak power Reduce 24

25 What is EMS? Input Output Referring thermal values Detection Selection for power down domain Power switching Sensing Judgment Control Thermal Driver EMS CPU Hotplug EMS reduces leak power as the second safety net for heavy thermal condition. 25

26 Evaluation for thermal control Thermal [ ] Power Consumption [A] <Condition> - R-Car H3 Salvator-X - Ta=85 - CPU/GPU:Heavy load (Availability ratio:100%) Power consumption Tj IPA Check point No1 EMS Check point No3 Check point No A57 0 Frequency 1.7GHz 1.7GHz~0.5GHz 0.5GHz Number of Core 4 Core 4 Core 1 Core 0 IPA and EMS are able to prevent thermal runaway in heavy theraml condition. 26

27 Conclusion of the thermal control feature Confirmed effectivity of the IPA and the EMS even the system operated in a quite extreme thermal condition. IPA and the EMS are able to use as the safety net with the software thermal control solution. 27

28 Solution for sophisticated user experience High performance to improve parallel execution facilities To process huge amount of running application System healthiness to prevent troubles on heavy thermal conditions To keep thermal condition within the specification Improved user experience to provide good user interface To keep system requirement of boot time More multiple computation Multi-processing with big.little methodology Reliable thermal control Controlling performance by power management features Sophisticated user experience fast boot feature using Suspend to RAM 28

29 Requirement of sophisticated user experience Cooperation with cloud-solution Voice / Image recognition Fast boot from engine starting Augmented Reality 29

30 Background for requirements of fast boot facility on IVI system Operating navigation systems soon Boot up start Boot up complete Viewing rear camera image soon Manipulating audio / video play soon Boot up time is long Boot up complete I want to Boot up time is short 30

31 Overview of Suspend to RAM Running system Suspend to RAM Voltage supply No voltage supply Power area ON Power Switch Power Switch Power OFF DDR1/2C DDR1/2C DRAM DRAM DDR1/2 PRESETB BKUP_CTRL DDR IF IO Suspend DDR1/2 PRESETB BKUP_CTRL DDR IF IO Data RSTB PMIC BKUP_TRG I2C_SDA I2C_SCL DVFS, VDD R-CarH3 Resume RSTB PMIC BKUP_TRG I2C_SDA I2C_SCL DVFS, VDD R-CarH3 Backup VD33, VD18, etc. VD33, VD18, etc. Each Devices Each Devices 31

32 Process flow of Suspend to RAM Suspend Resume Application Application Detect Suspend trigger - Stop Application - Close Drivers by App Application Call Suspend to RAM via sysfs-if (1) Drivers & Kernel All process freezing Power Manager stop - If needs, backup register by driver - If needs, stop clock for device ARM Trusted Firmware - Non-boot CPU OFF - Self-refresh setting - Notify to PMIC via I2C - Boot CPU OFF - Re-start Application - Initialize, re-detect devices and open Drivers Drivers & Kernel Framework start to work All process wakeup Power Manager start - If needs, restore register by driver - If needs, supply clock for device - If needs, initialize driver ARM Trusted Firmware/Loader - Non-boot CPU ON - Jump to Linux entry point - Boot CPU ON - Stop Self-refresh by Loader Detect Resume trigger User need to care for the Suspend to RAM Renesas provided for the Suspend to RAM 1: # echo mem > /sys/power/state 32

33 Mechanism for reduction of boot-up period Cold boot IPL u-boot Linux Kernel Application Warm boot(resume) Application IPL Linux Kernel Skip re-load the kernel Skip kernel initialization IPL Linux Kernel Application Skip most portion of those initialization for application It s able to expect precise reduction of boot-up period with this Suspend to RAM facility. 33

34 Suspend to RAM on AGL environment AGL DD Home screen Boot up start Boot up complete R-Car Starter Kit Pro Measurement period Home screen Board R-Car Starter Kit Pro Linux Kernel v4.9 AGL Daring Dab(As of May 9) Cold boot Warm boot Boot up time 12.8 [s] 3.9 [s] 34

35 Conclusion of Suspend to RAM facility Confirmed the effort of the Suspend to RAM to get shorter boot-up period. Suspend to RAM is able to use as a effective candidate for provide fast boot facility in IVI systems. 35

36 Current status of three solution on AGL environment High performance to improve parallel execution facilities To process huge amount of running application System healthiness to prevent troubles on heavy thermal conditions To keep thermal condition within the specification Improved user experience to provide good user interface To keep system requirement of boot time More multiple computation Multi-processing with big.little methodology Reliable thermal control Controlling performance by power management features Sophisticated user experience fast boot feature using Suspend to RAM 36

37 In final conclusion Renesas is providing many sort of solutions for customers and communities as a solution provider company. Renesas will be contributing solutions to AGL and supporting it strongly. 37

38 ARM, AGL and any other logos, product names and marks are registered of there respective owners. 38

39 Appendix: Status of EAS EAS has not been officially contributed to upstream community yet. Currently, ARM and Linaro are promoting this work, and it will take more time. If upstream community accepts EAS, it will be available in all Linux environments including AGL. On the other hand, Renesas thinks as follows. BSP provided by Renesas is based on upstream first. So, EAS would be supported in BSP after acceptance by upstream community. However, Renesas has a plan to provide EAS to AGL in advance in order for AGL members to avail EAS as soon as possible. Renesas will announce this in near the future. 39

40 Appendix: Operation of Thermal control Even you have to care about cause of HW failure, you are able to see effectiveness of IPA or EMS via sysfs under extreme thermal condition. [IPA] /* Checking current frequency of CPU */ $ cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq [EMS] /* Checking output log information */ thermal emergency notifier: state=1 / Execute EMS thermal emergency notifier: state=0 / Release EMS 40

41 Appendix: Operation of Suspend to RAM on M3 SK [Suspend] 1. Set to PMIC to backup mode via i2c-tools command: $ i2cset -f -y 7 0x30 0x20 0x0F 2. Suspend to RAM can be operated via sysfs: $ echo mem > /sys/power/state [Resume] 1. Push SW8 (Power switch) 41

42 Appendix: Reference IPA EAS CPU Hotplug <Linux Kernel>/Documentation/cputopology.txt <Linux Kernel>/Documentation/cpu-hotplug.txt CPU Freq <Linux Kernel>/Documentation/cpu-freq/user-guide.txt Suspend to RAM <Linux Kernel>/Documentation/power/states.txt 42