Анализ и оптимизация энергопотребления Rockchip: измерения, CPU, DDR и температура

    /* C is constant, V is voltage, F is frequency*/
    P(d)= C * V^2 * F

cat /sys/class/thermal/thermal_zone0/temp

/* Acquire the frequency voltage table, target column means the voltage required
by some frequency */
cat /sys/kernel/debug/opp/opp_summary
 device                rate(Hz)    target(uV)    min(uV)    max(uV)
-------------------------------------------------------------------
...
 cpu0
                      408000000       950000      950000     1350000
                      600000000       950000      950000     1350000
                      816000000      1000000     1000000     1350000
                     1008000000      1125000     1125000     1350000
                     1200000000      1275000     1275000     1350000
                     1248000000      1300000     1300000     1350000
                     1296000000      1350000     1350000     1350000

/* Check the frequency scaling strategy currently used by cpufreq, cpu frequency
scaling is enabled with the default interactive strategy */
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
interactive
/* Set userspace strategy to fix the frequency of cpu, then set different
frequencies, compare the set voltage with the measured voltage */
echo userspace > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
userspace

/* Check the frequency point supported by cpufreq */
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
408000 600000 816000 1008000 1200000 1248000 1296000

/* Set the fixed frequency */
echo 408000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed

/* Confirm current frequency*/
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
408000

/* Confirm current voltage, and compare with measured value, vdd_arm represents
the name of regulator, which is differnt for differnt projects*/
cat /sys/kernel/debug/regulator/vdd_arm/voltage
950000

/* Acquire current voltages of all regulators*/
cat /sys/kernel/debug/regulator/regulator_summary
 regulator                      use open bypass voltage current     min     max
-------------------------------------------------------------------------------
...
 vcc3v8_sys                       0   12      0  3800mV     0mA  3800mV  3800mV
    deviceless                                                      0mV     0mV
    vdd_logic                     0    4      0   950mV     0mA   950mV  1350mV
       dmc                                                        950mV  1350mV
       ff400000.gpu                                               950mV  1350mV
       bus-apll                                                   950mV  1350mV
       deviceless                                                   0mV     0mV
    vdd_arm                       0    2      0   950mV     0mA   950mV  1350mV
       cpu0                                                       950mV  1350mV
       deviceless                                                   0mV     0mV
...

/* Use top command to check the task loading, the output of top with different
versions will have difference, this version of top supports to check the thread
and the running cpu of the thread */
top -m 5 -t
User 51%, System 2%, IOW 0%, IRQ 0%
User 712 + Nice 0 + Sys 33 + Idle 634 + IOW 0 + IRQ 0 + SIRQ 0 = 1379

/* PR column represents currently running cpu of the thread, the sum of all cpu
loading percentage is equal to 100%, so the highest loading percentage of each
cpu is 100%/NR_CPU, the highest loading percentage of each CPU of SoC with 4
cores is 25% */
  PID   TID PR CPU% S     VSS     RSS PCY UID      Thread          Proc
 2631  2631  3  25% R   3104K    552K  fg root     busybox         busybox
 2632  2632  2  25% R   3104K    552K  fg root     busybox         busybox
 2633  2633  1   3% R    740K    400K  fg root     top             /data/top
  255   476  0   0% S  15492K   4988K  fg system   HwBinder:255_1  /vendor/bin/hw/android.hardware.sensors@1.0-service
  419   478  1   0% S 3770752K 256884K  fg system   SensorService   system_server

/* Use cpustats to observe the frequency change of cpu */
cpustats
Total: User 600 + Nice 0 + Sys 3 + Idle 591 + IOW 0 + IRQ 0 + SIRQ 0 = 1194
  408000kHz 0 +
  600000kHz 0 +
  816000kHz 0 +
  1008000kHz 0 +
  1200000kHz 0 +
  1248000kHz 0 +
  1296000kHz 0 +
  1416000kHz 0 +
  1512000kHz 1200 = 1200 /* within the statistic time, there are 1200 system
jiffies in total, 1512M running for 1200 jiffies */
/* from below, we can see the loading status of each cpu, including user mode,
kernel mode, interrupt and idle time */
cpu0: User 0 + Nice 0 + Sys 1 + Idle 294 + IOW 0 + IRQ 0 + SIRQ 0 = 295
cpu1: User 299 + Nice 0 + Sys 1 + Idle 0 + IOW 0 + IRQ 0 + SIRQ 0 = 300
cpu2: User 1 + Nice 0 + Sys 1 + Idle 296 + IOW 0 + IRQ 0 + SIRQ 0 = 298
cpu3: User 300 + Nice 0 + Sys 1 + Idle 0 + IOW 0 + IRQ 0 + SIRQ 0 = 301

/* check the ratio of running time for each frequency through cpufreq node, time
unit: jiffies */
cat /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state
408000 718186
600000 548
816000 368
1008000 1578
1200000 1104
1248000 84
1296000 101
1416000 678
1512000 47495

/* check the inturrupt quantity of all peripherals */
cat /proc/interrupts
           CPU0       CPU1       CPU2       CPU3
  1:          0          0          0          0     GICv2  29 Edge     arch_timer
  2:     181898     165057     636772     839244     GICv2  30 Edge     arch_timer
  5:     180743      39000      28905      65189     GICv2  62 Level     rk_timer
 13:     260634          0          0          0     GICv2  39 Level     ff180000.i2c
 14:     354805          0          0          0     GICv2  40 Level     ff190000.i2c
 15:          0          0          0          0     GICv2  41 Level     ff1a0000.i2c
...

/* acquire the frequency voltage table */
cat /sys/kernel/debug/opp/opp_summary
 device                rate(Hz)    target(uV)    min(uV)    max(uV)
-------------------------------------------------------------------
...
 platform-ff400000.gpu
                      200000000       950000      950000      950000
                      300000000       950000      950000      950000
                      400000000      1025000     1025000     1025000
                      480000000      1100000     1100000     1100000
                      520000000      1150000     1150000     1150000
...

/* check the frequency scaling strategy currently used by gpu devfreq, gpu
frequency scaling is enabled with the default simple_ondemand strategy */
cat /sys/class/devfreq/ff400000.gpu/governor
simple_ondemand

/* Set userspace strategy to fix the frequency of gpu, then set different
frequencies, compare the set voltage with measured voltage */
echo userspace > /sys/class/devfreq/ff400000.gpu/governor
cat /sys/class/devfreq/ff400000.gpu/governor
userspace

/* check the frequency points supported by gpu devfreq */
cat /sys/class/devfreq/ff400000.gpu/available_frequencies
520000000 480000000 400000000 300000000 200000000

/* set the fixed frequency */
echo 200000000 > /sys/class/devfreq/ff400000.gpu/userspace/set_freq

/* confirm current frequency */
cat /sys/class/devfreq/ff400000.gpu/cur_freq
200000000

/* confirm current voltage, and compare with measued value */
cat /sys/kernel/debug/regulator/vdd_gpu/voltage
950000

/* check gpu loading */
cat /sys/class/devfreq/ff400000.gpu/load
0@200000000Hz

cat /sys/kernel/debug/clk/clk_summary
   clock                         enable_cnt  prepare_cnt        rate   accuracy phase
----------------------------------------------------------------------------------------
 xin24m                                   9           10    24000000          0 0
...
    pll_gpll                              1            1  1200000000          0 0
       gpll                               9           20  1200000000          0 0
          clk_sdio_div50                  1            1   100000000          0 0
             clk_sdio                     1            5   100000000          0 0
                sdio_sample               0            1    50000000          0 0
                sdio_drv                  0            1    50000000          0 180
          clk_emmc_div50                  1            1   300000000          0 0
             clk_emmc                     1            5   300000000          0 0
                emmc_sample               0            1   150000000          0 42
                emmc_drv                  0            1   150000000          0 180
...

cat /sys/kernel/debug/pm_genpd/pm_genpd_summary                                <
domain                          status          slaves
    /device                                             runtime status
----------------------------------------------------------------------
pd_gpu                          off
    /devices/platform/ff400000.gpu                      suspended
pd_vi                           off
    /devices/platform/ff4a8000.iommu                    suspended
pd_vo                           on
    /devices/platform/ff460f00.iommu                    active
    /devices/platform/ff470f00.iommu                    suspended
    /devices/platform/ff2e0000.video-phy                suspended
    /devices/platform/ff450000.dsi                      active
    /devices/platform/ff460000.vop                      active
    /devices/platform/ff470000.vop                      suspended
    /devices/platform/ff480000.rk_rga                   suspended
pd_vpu                          off
    /devices/platform/ff440440.iommu                    suspended
    /devices/platform/ff442800.iommu                    suspended
    /devices/platform/vpu_combo                         suspended
pd_mmc_nand                     on
    /devices/platform/ff380000.dwmmc                    unsupported
    /devices/platform/ff390000.dwmmc                    unsupported
    /devices/platform/ff3b0000.nandc                    active
pd_gmac                         off
pd_sdcard                       off
pd_usb                          on
    /devices/platform/ff300000.usb                      active

cat /sys/kernel/debug/opp/opp_summary
 device                rate(Hz)    target(uV)    min(uV)    max(uV)
-------------------------------------------------------------------
 platform-dmc
                      194000000       950000      950000      950000
                      328000000       950000      950000      950000
                      450000000       950000      950000      950000
                      528000000       975000      975000      975000
                      666000000      1000000     1000000     1000000
...

/* ddr uses dmc_ondemand frequency scaling strategy by default */
cat /sys/class/devfreq/dmc/governor
dmc_ondemand

/* the default frequency scaling strategy used is interactive, relative
parameters are as follows: */
ls -l /sys/devices/system/cpu/cpu0/cpufreq/interactive
go_hispeed_load     /* when the loading is larger than go_hispeed_load and the
frequency is smaller than hispeed_freq, directly jump to hispeed_freq */
hispeed_freq        /* when jumping from low frequency to high frequency, need to
jump to hispedd_freq first */
above_hispeed_delay /* when the frequency is larger than hispeed_freq, the time
duration before each frequency increase */
min_sample_time     /* after each frequency increase, if it is to reduce the
frequency next time , the time duration before frequency reduce */
target_loads        /* the target loading of the frequency scaling */
timer_rate          /* the loading sampling time，unit:us */
timer_slack         /* the loading sampling time after cpu entering idle */
boost               /* when the frequency is smaller than hispeed_freq, keep
boost to hispeed_freq */
boostpulse          /* when the frequency is smaller than hispeed_freq, boost to
hispeed_freq, keep a while */
boostpulse_duration /* time duration of boostpulse, unit:us */
io_is_busy          /* whether to compute io wait to cpu loading */

/* close cpu2，cpu3 */
echo 0 > /sys/devices/system/cpu/cpu2/online
echo 0 > /sys/devices/system/cpu/cpu3/online

/* set the max frequency of cpu0 to 1200MHz */
echo 1200000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq

/* create group of litte core*/
mkdir /dev/cpuset/little

/* set cpu used by group of little core */
echo 0-3 > /dev/cpuset/little/cpus

/* add pid=1111 task into the group of little core */
echo 1111 > /dev/cpuset/little/tasks

/* Android system creates several groups by default, the framework layer puts the
tasks into differenct groups, you can adjust cpus of each group, analyze the
power consumption */
ls /dev/cpuset
background
foreground
system-background
top-app

/* create the group of bandwidth limitation */
mkdir /dev/cpuctl/mygroup

/* set the cycle of bandwidth limitation as 10ms */
echo 10000 > /dev/cpuctl/mygroup/cpu.cfs_quota_us

/* within each cycle, total running time of the tasks in the group cannot exceed
5ms, this value can be larger than cfs_quota_us, because it is the total running
time of multiple cpus */
echo 5000 > /dev/cpuctl/mygroup/cpu.cfs_period_us

/* add relative tasks into the group */
echo 1111 > /dev/cpuctl/mygroup/tasks
echo 1112 > /dev/cpuctl/mygroup/tasks

/* cpu.shares means to limit the bandwidth of the task through weight, used for
performance optimization, without affecting the power consumption */
/dev/cpuctl/mygroup/cpu.shares

/* scenario definition */
include/dt-bindings/clock/rk_system_status.h
#define SYS_STATUS_NORMAL       (1<<0)
#define SYS_STATUS_SUSPEND      (1<<1)
#define SYS_STATUS_IDLE         (1<<2)
#define SYS_STATUS_REBOOT       (1<<3)
#define SYS_STATUS_VIDEO_4K     (1<<4)
#define SYS_STATUS_VIDEO_1080P  (1<<5)
...

/* configure the frequencies for different scenarios in dts */
arch/arm64/boot/dts/rockchip/px30.dtsi
dmc: dmc {
    compatible = "rockchip,px30-dmc";
    ...
    system-status-freq = <
        /*system status         freq(KHz)*/
        SYS_STATUS_NORMAL       528000
        SYS_STATUS_REBOOT       450000
        SYS_STATUS_SUSPEND      194000
        SYS_STATUS_VIDEO_1080P  450000
        SYS_STATUS_BOOST        528000
        SYS_STATUS_ISP          666000
        SYS_STATUS_PERFORMANCE  666000
    >;
    ...

/* acquire the current scenario */
cat /sys/class/devfreq/dmc/system_status
0x401

/* configure the parameter of frequency scaling with loading in dts, need to open
dfi node to monitor DDR utility ratio */
dmc: dmc {
    compatible = "rockchip,px30-dmc";
    ...

    /* use dfi to monitor the utility ratio of DDR */
    devfreq-events = <&dfi>;

        /*
     * the threshold of frequency scaling：
     * when the utility ratio is over 40%, adjust to the highest frequency.
     * when the loading is less than 40% and larger than 40%-20%, maintain
current frequency.
     * when the loading is less than 40%-20%, it will adjust the frequency to a
certain value to make the loading to be around 40%-2%/2.
     */
    upthreshold = <40>;
    downdifferential = <20>;

/* check the DDR loading of current system */
cat /sys/class/devfreq/dmc/load                                       <
33@528000000Hz

&cpu0_opp_table {
    /* when the temperature is over 85 degree, limit the max voltage of cpu to
1.1V */
    rockchip,high-temp = <85000>;
    rockchip,high-temp-max-volt = <1100000>;

    /* or directly limit the max frequency to avoid the high voltage */
    rockchip,high-temp-max-freq = <1008000>;
};