Sched-tools June 2015

sched-tools@lists.linaro.org

11 participants
16 discussions

[PATCH] workgen: Add a video playback use case

by Pi-Cheng Chen

Signed-off-by: Pi-Cheng Chen <pi-cheng.chen(a)linaro.org> --- doc/examples/video_playback.json | 278 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 278 insertions(+) create mode 100644 doc/examples/video_playback.json diff --git a/doc/examples/video_playback.json b/doc/examples/video_playback.json new file mode 100644 index 0000000..941b20f --- /dev/null +++ b/doc/examples/video_playback.json @@ -0,0 +1,278 @@ +{ + "tasks" : { + "surfaceflinger" : { + "priority" : -7, + "loop" : -1, + "suspend" : "surfaceflinger", + "run" : 1500 + }, + + "DispSync" : { + "priority" : -7, + "loop" : -1, + "phases" : { + "p1" : { + "lock" : "mutexB", + "wait" : { "ref" : "DispSync", "mutex" : "mutexB" }, + "unlock": "mutexB", + "run" : 35, + "lock" : "mutexA", + "broad" : "EventThread", + "unlock" : "mutexA", + "run" : 40, + }, + + "p2" : { + "loop" : 2, + + "lock" : "mutexB", + "wait" : { "ref" : "DispSync", "mutex" : "mutexB" }, + "unlock": "mutexB", + "run" : 30 + } + }, + }, + + "hwc_eventmon" : { + "priority" : -19, + "loop" : -1, + "lock" : "mutexB", + "signal" : "DispSync", + "unlock": "mutexB", + "run" : 115, + "timer" : { "ref" : "timerA", "period" : 16667 } + }, + + "EventThread1" : { + "priority" : -8, + "loop" : -1, + "phases" : { + "p1" : { + "lock" : "mutexA", + "wait" : { "ref" : "EventThread", "mutex" : "mutexA" }, + "unlock" : "mutexA", + "run" : 25, + "lock" : "mutexB", + "signal" : "DispSync", + "unlock": "mutexB", + "sleep" : 9650, + "run" : 70, + "lock" : "mutexB", + "signal" : "DispSync", + "unlock": "mutexB", + "run" : 80 + }, + + "p2" : { + "lock" : "mutexA", + "wait" : { "ref" : "EventThread", "mutex" : "mutexA" }, + "unlock" : "mutexA", + "run" : 90, + "lock" : "mutexB", + "signal" : "DispSync", + "unlock": "mutexB" + } + } + }, + + "EventThread2" : { + "priority" : -8, + "loop" : -1, + "phases" : { + "p1" : { + "lock" : "mutexA", + "wait" : { "ref" : "EventThread", "mutex" : "mutexA" }, + "unlock" : "mutexA", + "run" : 30, + "resume" : "surfaceflinger" + }, + + "p2" : { + "lock" : "mutexA", + "wait" : { "ref" : "EventThread", "mutex" : "mutexA" }, + "unlock" : "mutexA", + "run" : 35, + "sleep" : 2000, + "run" : 110, + "lock" : "mutexB", + "signal" : "DispSync", + "unlock": "mutexB", + "run" : 60 + } + } + }, + + "waker" : { + "priority" : -19, + "loop" : -1, + "resume" : "NuPlayerRenderer", + "timer" : { "ref" : "timerB", "period" : 33333 } + }, + + "NuPlayerRenderer" : { + "priority" : -15, + "loop" : -1, + "phases" : { + "p1" : { + "loop" : 3, + "suspend" : "NuPlayerRenderer", + "run" : 140, + "signal" : "NuPlayerDriver1", + "run" : 95 + }, + + "p2" : { + "sleep" : 27000, + "run" : 580, + "resume" : "NPDecoder", + "resume" : "NPDecoder-CL", + "resume" : "gle.aac.decoder" + } + } + }, + + "NuPlayerDriver1" : { + "priority" : -15, + "loop" : -1, + "suspend" : "NuPlayerDriver1", + "run" : 100, + "lock" : "mutexC", + "sync" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 50, + "lock" : "mutexC", + "wait" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 80, + "lock" : "mutexC", + "sync" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 370, + "lock" : "mutexC", + "sync" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 135, + "lock" : "mutexC", + "signal" : "NuPlayerDriver", + "unlock" : "mutexC" + }, + + "NuPlayerDriver2" : { + "priority" : -15, + "loop" : -1, + "lock" : "mutexC", + "wait" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 110, + "lock" : "mutexC", + "signal" : "NuPlayerDriver", + "unlock" : "mutexC", + "lock" : "mutexD", + "signal" : "CodecLooper1", + "unlock" : "mutexD", + "sleep" : 2500, + "run" : 80, + "lock" : "mutexC", + "sync" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 50, + "lock" : "mutexC", + "sync" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 70, + "lock" : "mutexC", + "sync" : { "ref" : "NuPlayerDriver", "mutex" : "mutexC" }, + "unlock" : "mutexC", + "run" : 35 + }, + + "CodecLooper1" : { + "priority" : -15, + "loop" : -1, + "lock" : "mutexD", + "wait" : { "ref" : "CodecLooper1", "mutex" : "mutexD" }, + "unlock" : "mutexD", + "run" : 230, + "sleep" : 80, + "run" : 150, + "sleep" : 210, + "run" : 330, + "resume" : "CodecLooper2", + "sleep" : 900, + "run" : 170, + "sleep" : 670, + "run" : 125, + "resume" : "CodecLooper2" + }, + + "CodecLooper2" : { + "priority" : -1, + "loop" : -1, + "suspend" : "CodecLooper2", + "run" : 160, + "resume" : "CodecLooper3", + "sleep" : 590, + "resume" : "OMXCallbackDisp2", + "run" : 75, + "suspend" : "CodecLooper2", + "run" : 260 + }, + + "OMXCallbackDisp2" : { + "priority" : -1, + "loop" : -1, + "suspend" : "OMXCallbackDisp2", + "run" : 180 + }, + + "CodecLooper3" : { + "priority" : -1, + "loop" : -1, + "suspend" : "CodecLooper3", + "run" : 1000 + }, + + "NPDecoder" : { + "priority" : -15, + "loop" : -1, + "suspend" : "NPDecoder", + "run" : 500, + "sleep" : 680, + "resume" : "OMXCallbackDisp1", + "run" : 2000 + }, + + "NPDecoder-CL" : { + "priority" : -15, + "loop" : -1, + "suspend" : "NPDecoder-CL", + "run" : 570, + "sleep" : 570, + "run" : 2100 + }, + + "gle.aac.decoder" : { + "priority" : -1, + "loop" : -1, + "suspend" : "gle.aac.decoder", + "run" : 2400, + "sleep" : 430, + "run" : 45 + }, + + "OMXCallbackDisp1" : { + "priority" : -1, + "loop" : -1, + "suspend" : "OMXCallbackDisp1", + "run" : 135, + "sleep" : 230, + "run" : 140, + "sleep" : 330, + "run" : 190, + "sleep" : 550, + "run" : 160 + } + } +} + + -- 1.9.1

10 years

[RESEND PATCH v3] doc: measure the efficiency of cpufreq governors

by Xunlei Pang

From: Xunlei Pang <pang.xunlei(a)linaro.org> DVFS adds a latency in the execution of task because of the time to decide to move at max freq. We need to measure this latency and check that the governor stays in an acceptable range. When rt-app runs a json file, a log file is created for each thread. This log file records the number of loop that has been executed and the duration for executing these loops (per phase). We can use these figures to evaluate to latency that is added by a cpufreq governor and its "performance efficiency". We use the run+sleep pattern to do the measurement, for the run time per loop, the performance governor should run the expected duration as the CPU stays a max freq. At the opposite, the powersave governor will give use the longest duration (as it stays at lowest OPP). Other governor will be somewhere between the 2 previous duration as they will use several OPP and will go back to max frequency after a defined duration which depends on its monitoring period. The formula: duration of powersave gov - duration of the gov -------------------------------------------------------- x 100% duration of powersave gov - duration of performance gov will give the efficiency of the governor. 100% means as efficient as the perf governor and 0% means as efficient as the powersave governor. This patch offers json files and shell scripts to do the measurement. Usage: ./calibration.sh <cpu> cpu: cpu number on which you want to run the test ./test.sh <governor> <cpu> <runtime> <sleeptime> [<loops>] governor: target CPUFreq governor you want to test cpu: cpu number on which you want to run the test. Be the same as the one passed to "calibration.sh". runtime: running time in ms per loop of the workload pattern sleeptime: sleeping time in ms per loop of the workload pattern loops: repeat times of the workload pattern. default: 10 Example: "./calibration.sh 0" means to calculate the computing capacity of CPU0 which will be used in the following test. "./test.sh ondemand 0 100 100 20" means to test "ondemand" on CPU0 with workload pattern "run 100ms + sleep 100ms"(20 loops). NOTE: - Make sure there are "sed", "cut", "grep", "rt-app", etc tools on your test machine, and run the scripts under root privilege. - Run the test while the system is idle. - You can change the target governor's parameters after running "calibration.sh", but before "test.sh". Signed-off-by: Xunlei Pang <pang.xunlei(a)linaro.org> --- doc/examples/cpufreq_governor_efficiency/README | 60 ++++++++++++ .../cpufreq_governor_efficiency/calibration.json | 26 ++++++ .../cpufreq_governor_efficiency/calibration.sh | 17 ++++ doc/examples/cpufreq_governor_efficiency/dvfs.json | 27 ++++++ doc/examples/cpufreq_governor_efficiency/dvfs.sh | 38 ++++++++ doc/examples/cpufreq_governor_efficiency/test.sh | 104 +++++++++++++++++++++ 6 files changed, 272 insertions(+) create mode 100755 doc/examples/cpufreq_governor_efficiency/README create mode 100755 doc/examples/cpufreq_governor_efficiency/calibration.json create mode 100755 doc/examples/cpufreq_governor_efficiency/calibration.sh create mode 100755 doc/examples/cpufreq_governor_efficiency/dvfs.json create mode 100755 doc/examples/cpufreq_governor_efficiency/dvfs.sh create mode 100755 doc/examples/cpufreq_governor_efficiency/test.sh diff --git a/doc/examples/cpufreq_governor_efficiency/README b/doc/examples/cpufreq_governor_efficiency/README new file mode 100755 index 0000000..10482b8 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/README @@ -0,0 +1,60 @@ +Measure the efficiency of cpufreq governors using rt-app + +BACKGROUND: + DVFS adds a latency in the execution of task because of the time to + decide to move at max freq. We need to measure this latency and check + that the governor stays in an acceptable range. + + When rt-app runs a json file, a log file is created for each thread. + This log file records the number of loop that has been executed and + the duration for executing these loops (per phase). We can use these + figures to evaluate to latency that is added by a cpufreq governor + and its "performance efficiency". + + We use the run+sleep pattern to do the measurement, for the run time per + loop, the performance governor should run the expected duration as the + CPU stays a max freq. At the opposite, the powersave governor will give + use the longest duration (as it stays at lowest OPP). Other governor will + be somewhere between the 2 previous duration as they will use several OPP + and will go back to max frequency after a defined duration which depends + on its monitoring period. + + The formula: + + duration of powersave gov - duration of the gov + -------------------------------------------------------- x 100% + duration of powersave gov - duration of performance gov + + will give the efficiency of the governor. 100% means as efficient as + the perf governor and 0% means as efficient as the powersave governor. + + This test offers json files and shell scripts to do the measurement. + +Usage: + ./calibration.sh <cpu> + cpu: cpu number on which you want to run the test + + ./test.sh <governor> <cpu> <runtime> <sleeptime> [<loops>] + governor: target CPUFreq governor you want to test + cpu: cpu number on which you want to run the test. Be the same + as the one passing to "calibration.sh". + runtime: running time in ms per loop of the workload pattern + sleeptime: sleeping time in ms per loop of the workload pattern + loops: repeat times of the workload pattern. default: 10 + +Example: + "./calibration.sh 0" means to calculate the computing capacity of CPU0 which + will be used in the following test. + + "./test.sh ondemand 0 100 100 20" means to + test "ondemand" on CPU0 with workload pattern "run 100ms + sleep 100ms"(20 loops). + +NOTE: + - Make sure there are "sed", "cut", "grep", "rt-app", etc tools on + your test machine, and run the scripts under root privilege. + + - Run the test while the system is idle. + + - You can change the target governor's parameters after running "calibration.sh", + but before "test.sh". + diff --git a/doc/examples/cpufreq_governor_efficiency/calibration.json b/doc/examples/cpufreq_governor_efficiency/calibration.json new file mode 100755 index 0000000..2d5870c --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/calibration.json @@ -0,0 +1,26 @@ +{ + "tasks" : { + "thread" : { + "instance" : 1, + "loop" : 1, + "phases" : { + "run" : { + "loop" : 1, + "run" : 2000, + }, + "sleep" : { + "loop" : 1, + "sleep" : 2000, + } + } + } + }, + "global" : { + "default_policy" : "SCHED_FIFO", + "calibration" : "CPU0", + "lock_pages" : true, + "ftrace" : false, + "logdir" : "./", + } +} + diff --git a/doc/examples/cpufreq_governor_efficiency/calibration.sh b/doc/examples/cpufreq_governor_efficiency/calibration.sh new file mode 100755 index 0000000..a217487 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/calibration.sh @@ -0,0 +1,17 @@ +#!/bin/sh + +set -e + +if [ ! $1 ] ; then + echo "Please input one cpu" + exit +fi + +echo performance > /sys/devices/system/cpu/cpu$1/cpufreq/scaling_governor + +sleep 1 + +sed 's/"calibration" : "CPU.*",/"calibration" : "CPU'$1'",/' -i calibration.json +pLoad=$(rt-app calibration.json 2>&1 |grep pLoad |sed 's/.*= $.*$ns.*/\1/') +sed 's/"calibration" : .*,/"calibration" : '$pLoad',/' -i dvfs.json +echo CPU$1\'s pLoad is $pLoad diff --git a/doc/examples/cpufreq_governor_efficiency/dvfs.json b/doc/examples/cpufreq_governor_efficiency/dvfs.json new file mode 100755 index 0000000..c8447df --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/dvfs.json @@ -0,0 +1,27 @@ +{ + "tasks" : { + "thread" : { + "instance" : 1, + "cpus" : [0], +"loop" : 21, + "phases" : { + "running" : { + "loop" : 1, + "run" : 100, + }, + "sleeping" : { + "loop" : 1, + "sleep" : 100, + } + } + } + }, + "global" : { + "default_policy" : "SCHED_OTHER", + "calibration" : 90, + "lock_pages" : true, + "ftrace" : false, + "logdir" : "./", + } +} + diff --git a/doc/examples/cpufreq_governor_efficiency/dvfs.sh b/doc/examples/cpufreq_governor_efficiency/dvfs.sh new file mode 100755 index 0000000..00ce81d --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/dvfs.sh @@ -0,0 +1,38 @@ +#!/bin/sh + +# $1 $2 $3 $4 $5: governor cpu run sleep loops +set -e + +echo $1 > /sys/devices/system/cpu/cpu$2/cpufreq/scaling_governor +#echo $1 > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor +sed 's/"cpus" : \[.*\],/"cpus" : \['$2'\],/' -i dvfs.json +sleep 3 + +if [ $3 ] ; then + sed 's/"run" : .*,/"run" : '$3',/' -i dvfs.json +fi + +if [ $4 ] ; then + sed 's/"sleep" : .*,/"sleep" : '$4',/' -i dvfs.json +fi + +if [ $5 ] ; then + sed 's/^"loop" : .*,/"loop" : '$5',/' -i dvfs.json +fi + +rt-app dvfs.json 2> /dev/null + +if [ $1 ] ; then + mv -f rt-app-thread-0.log rt-app_$1_run$3us_sleep$4us.log + + sum=0 + loop=0 + for i in $(cat rt-app_$1_run$3us_sleep$4us.log | sed 'n;d' | sed '1d' |cut -f 3); do + loop=$(expr $loop + 1) + sum=$(expr $sum + $i) + done + sum=$(expr $sum / $loop) + echo $sum + rm -f rt-app_$1_run$3us_sleep$4us.log +fi + diff --git a/doc/examples/cpufreq_governor_efficiency/test.sh b/doc/examples/cpufreq_governor_efficiency/test.sh new file mode 100755 index 0000000..0160952 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/test.sh @@ -0,0 +1,104 @@ +#!/bin/sh + +set -e + +test_efficiency() { + + FILENAME="results_$RANDOM$$.txt" + + if [ -e /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors ]; then + for i in $(cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors); do + if [ $i = $1 ] ; then + gov_target=$i + fi + export gov_$i=$(echo $i) + done + else + echo "cpufreq sysfs is not available!" + exit + fi + + if [ ! $gov_target ] ; then + echo " Can't find $1 governor!" + exit + fi + + if [ ! $gov_performance ] ; then + echo "Can't find performance governor!" + exit + fi + + if [ ! $gov_powersave ] ; then + echo "Can't find powersave governor!" + exit + fi + + if [ $gov_target = $gov_performance ] || [ $gov_target = $gov_powersave ] ; then + echo "Please input a governor other than \"performance\" or \"powersave\"" + exit + fi + + # Get target gov data first + dvfs.sh $1 $2 $3 $4 $5 > $FILENAME + target=$(cat $FILENAME |sed -n '1p') + + # Get powersave data + dvfs.sh powersave $2 $3 $4 $5 > $FILENAME + powersave=$(cat $FILENAME |sed -n '1p') + + # Get performance data + dvfs.sh performance $2 $3 $4 $5 > $FILENAME + performance=$(cat $FILENAME |sed -n '1p') + + if [ $performance -ge $powersave ] ; then + echo "powersave: $powersave" + echo "performance: $performance" + echo "Error! performance spent more time than powersave!" + exit + fi + + echo "\"powersave\" efficiency: 0%" + echo "\"performance\" efficiency: 100%" + + denominator=$(expr $powersave - $performance) + + if [ $powersave -le $target ]; then + target=0 + else + numerator=$(expr $powersave - $target) + numerator=$(expr $numerator \* 100) + target=$(expr $numerator / $denominator) + if [ $target -gt 100 ]; then + target=100 + fi + fi + + echo "\"$gov_target\" efficiency: $target%" + + rm -f $FILENAME +} + +if [ $# -lt 4 ]; then + echo "Usage: ./test.sh <governor> <cpu> <runtime> <sleeptime> [<loops>]" + echo "governor: target CPUFreq governor you want to test" + echo "cpu: cpu number on which you want to run the test" + echo "runtime: running time in ms per loop of the workload pattern" + echo "sleeptime: sleeping time in ms per loop of the workload pattern" + echo "loops: repeat times of the workload pattern. default: 10" + echo "\nExample:\n\"./test.sh ondemand 0 100 100 20\" means\nTest \"ondemand\" on CPU0 with workload pattern \"run 100ms + sleep 100ms\"(20 loops).\n" + exit +fi + +if [ $# = 4 ]; then + loops=10 +else + loops=$5 +fi + +echo "Test \"$1\" on CPU$2 with workload pattern \"run $3ms + sleep $4ms\"($loops loops)." + +sleep 1 +PATH=$PATH:. + +test_efficiency $1 $2 $(expr $3 \* 1000) $(expr $4 \* 1000) $loops + -- 1.9.1

10 years, 1 month

[RESEND PATCH v2] doc: measure the efficiency of cpufreq governors

by Xunlei Pang

From: Xunlei Pang <pang.xunlei(a)linaro.org> DVFS adds a latency in the execution of task because of the time to decide to move at max freq. We need to measure this latency and check that the governor stays in an acceptable range. When workgen runs a json file, a log file is created for each thread. This log file records the number of loop that has been executed and the duration for executing these loops (per phase). We can use these figures to evaluate to latency that is added by a cpufreq governor and its "performance efficiency". We use the run+sleep pattern to do the measurement, for the run time per loop, the performance governor should run the expected duration as the CPU stays a max freq. At the opposite, the powersave governor will give use the longest duration (as it stays at lowest OPP). Other governor will be somewhere between the 2 previous duration as they will use several OPP and will go back to max frequency after a defined duration which depends on its monitoring period. The formula: duration of powersave gov - duration of the gov -------------------------------------------------------- x 100% duration of powersave gov - duration of performance gov will give the efficiency of the governor. 100% means as efficient as the perf governor and 0% means as efficient as the powersave governor. This patch offers json files and shell scripts to do the measurement, Usage: ./test.sh <cpus> <runtime> <sleeptime> cpus: number of cpus in the CPU0's frequency domain runtime: running time in ms per loop of the workload pattern sleeptime: sleeping time in ms per loop of the workload pattern Example: "./test.sh 4 100 1000" means CPU0~CPU3 sharing frequency, "100ms run + 1000ms sleep" workload pattern. test result on my machine: ~#./test.sh 4 100 1000 Frequency domain CPU0~CPU3, run 100ms, sleep 1000ms: powersave efficiency: 0% performance efficiency: 100% conservative efficiency: 28% ondemand efficiency: 95% NOTE: Make sure there are "sed", "cut", "grep", "rt-app", etc tools on your test machine, and run the script under root privilege. Signed-off-by: Xunlei Pang <pang.xunlei(a)linaro.org> --- doc/examples/cpufreq_governor_efficiency/README | 54 ++++++++++++++ .../cpufreq_governor_efficiency/calibration.json | 27 +++++++ .../cpufreq_governor_efficiency/calibration.sh | 11 +++ doc/examples/cpufreq_governor_efficiency/dvfs.json | 27 +++++++ doc/examples/cpufreq_governor_efficiency/dvfs.sh | 38 ++++++++++ doc/examples/cpufreq_governor_efficiency/test.sh | 82 ++++++++++++++++++++++ 6 files changed, 239 insertions(+) create mode 100644 doc/examples/cpufreq_governor_efficiency/README create mode 100644 doc/examples/cpufreq_governor_efficiency/calibration.json create mode 100755 doc/examples/cpufreq_governor_efficiency/calibration.sh create mode 100644 doc/examples/cpufreq_governor_efficiency/dvfs.json create mode 100755 doc/examples/cpufreq_governor_efficiency/dvfs.sh create mode 100755 doc/examples/cpufreq_governor_efficiency/test.sh diff --git a/doc/examples/cpufreq_governor_efficiency/README b/doc/examples/cpufreq_governor_efficiency/README new file mode 100644 index 0000000..cc8efe1 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/README @@ -0,0 +1,54 @@ +Measure the efficiency of cpufreq governors using rt-app + +BACKGROUND: + DVFS adds a latency in the execution of task because of the time to + decide to move at max freq. We need to measure this latency and check + that the governor stays in an acceptable range. + + When workgen runs a json file, a log file is created for each thread. + This log file records the number of loop that has been executed and + the duration for executing these loops (per phase). We can use these + figures to evaluate to latency that is added by a cpufreq governor + and its "performance efficiency". + + We use the run+sleep pattern to do the measurement, for the run time per + loop, the performance governor should run the expected duration as the + CPU stays a max freq. At the opposite, the powersave governor will give + use the longest duration (as it stays at lowest OPP). Other governor will + be somewhere between the 2 previous duration as they will use several OPP + and will go back to max frequency after a defined duration which depends + on its monitoring period. + + The formula: + + duration of powersave gov - duration of the gov + -------------------------------------------------------- x 100% + duration of powersave gov - duration of performance gov + + will give the efficiency of the governor. 100% means as efficient as + the perf governor and 0% means as efficient as the powersave governor. + + This test offers json files and shell scripts to do the measurement, + +USAGE: + ./test.sh <cpus> <runtime> <sleeptime> + cpus: number of cpus in the CPU0's frequency domain + runtime: running time in ms per loop of the workload pattern + sleeptime: sleeping time in ms per loop of the workload pattern + +Example: + "./test.sh 4 100 1000" means + CPU0~CPU3 sharing frequency, "100ms run + 1000ms sleep" workload pattern. + + test result on an Intel machine: + ~#./test.sh 4 100 1000 + Frequency domain CPU0~CPU3, run 100ms, sleep 1000ms: + powersave efficiency: 0% + performance efficiency: 100% + conservative efficiency: 28% + ondemand efficiency: 95% + +NOTE: + Make sure there are "sed", "cut", "grep", "rt-app", etc tools on your test + machine, and run the script under root privilege. + diff --git a/doc/examples/cpufreq_governor_efficiency/calibration.json b/doc/examples/cpufreq_governor_efficiency/calibration.json new file mode 100644 index 0000000..4377990 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/calibration.json @@ -0,0 +1,27 @@ +{ + "tasks" : { + "thread" : { + "instance" : 1, + "cpus" : [0], + "loop" : 1, + "phases" : { + "run" : { + "loop" : 1, + "run" : 200000, + }, + "sleep" : { + "loop" : 1, + "sleep" : 200000, + } + } + } + }, + "global" : { + "default_policy" : "SCHED_FIFO", + "calibration" : "CPU0", + "lock_pages" : true, + "ftrace" : true, + "logdir" : "./", + } +} + diff --git a/doc/examples/cpufreq_governor_efficiency/calibration.sh b/doc/examples/cpufreq_governor_efficiency/calibration.sh new file mode 100755 index 0000000..d10e644 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/calibration.sh @@ -0,0 +1,11 @@ +#!/bin/sh + +set -e + +echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor + +sleep 1 + +pLoad=$(rt-app calibration.json 2>&1 |grep pLoad |sed 's/.*= $.*$ns.*/\1/') +sed 's/"calibration" : .*,/"calibration" : '$pLoad',/' -i dvfs.json + diff --git a/doc/examples/cpufreq_governor_efficiency/dvfs.json b/doc/examples/cpufreq_governor_efficiency/dvfs.json new file mode 100644 index 0000000..b413156 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/dvfs.json @@ -0,0 +1,27 @@ +{ + "tasks" : { + "thread" : { + "instance" : 1, + "cpus" : [0], + "loop" : 5, + "phases" : { + "running" : { + "loop" : 1, + "run" : 100000, + }, + "sleeping" : { + "loop" : 1, + "sleep" : 1000000, + } + } + } + }, + "global" : { + "default_policy" : "SCHED_OTHER", + "calibration" : 90, + "lock_pages" : true, + "ftrace" : true, + "logdir" : "./", + } +} + diff --git a/doc/examples/cpufreq_governor_efficiency/dvfs.sh b/doc/examples/cpufreq_governor_efficiency/dvfs.sh new file mode 100755 index 0000000..8591fc7 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/dvfs.sh @@ -0,0 +1,38 @@ +#!/bin/sh + +#echo $1 $2 $3 +set -e + +if [ $1 ] && [ $2 ] ; then + for i in $(seq 0 1 $(expr $2 - 1)); do + echo $1 > /sys/devices/system/cpu/cpu$i/cpufreq/scaling_governor + #cat /sys/devices/system/cpu/cpu$i/cpufreq/scaling_governor + done + + sleep 3 +fi + +if [ $3 ] ; then + sed 's/"run" : .*,/"run" : '$3',/' -i dvfs.json +fi + +if [ $4 ] ; then + sed 's/"sleep" : .*,/"sleep" : '$4',/' -i dvfs.json +fi + +#cat dvfs.json + +rt-app dvfs.json 2> /dev/null + +if [ $1 ] ; then + mv -f rt-app-thread-0.log rt-app_$1_run$3us_sleep$4us.log + + sum=0 + for i in $(cat rt-app_$1_run$3us_sleep$4us.log | sed 'n;d' | sed '1d' |cut -f 3); do + sum=$(expr $sum + $i) + done + sum=$(expr $sum / 5) + echo $sum + rm -f rt-app_$1_run$3us_sleep$4us.log +fi + diff --git a/doc/examples/cpufreq_governor_efficiency/test.sh b/doc/examples/cpufreq_governor_efficiency/test.sh new file mode 100755 index 0000000..d72fc6a --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/test.sh @@ -0,0 +1,82 @@ +#!/bin/sh + +set -e + +set_calibration() { + calibration.sh +} + +test_efficiency() { + + FILENAME="results_$RANDOM$$.txt" + + if [ -e /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors ]; then + for i in $(cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors); do + export gov_$i=$(echo $i) + done + else + echo "cpufreq is not available!" + exit + fi + + if [ ! $gov_performance ] ; then + echo "Can't find performance governor!" + exit + fi + + if [ ! $gov_powersave ] ; then + echo "Can't find powersave governor!" + exit + fi + + # Get powersave data + dvfs.sh powersave $1 $2 $3 > $FILENAME + powersave=$(cat $FILENAME |sed -n '1p') + + # Get performance data + dvfs.sh performance $1 $2 $3 > $FILENAME + performance=$(cat $FILENAME |sed -n '1p') + + if [ $performance -ge $powersave ] ; then + echo "Error! Probably not input all the cpus in the same frequency domain" + exit + fi + + denominator=$(expr $powersave - $performance) + echo "powersave efficiency: 0%" + echo "performance efficiency: 100%" + + # Calcuate other governors data + for gov_next in $gov_conservative $gov_ondemand $gov_cfs; do + if [ "$gov_next" != "" ] ; then + dvfs.sh $gov_next $1 $2 $3 > $FILENAME + data=$(cat $FILENAME |sed -n '1p'); + numerator=$(expr $powersave - $data) + numerator=$(expr $numerator \* 100) + if [ $numerator -lt 0 ] ; then + let numerator=0 + fi + data=$(expr $numerator / $denominator) + echo "$gov_next efficiency: $data%" + fi + done + + rm -f $FILENAME +} + +if [ $# -lt 3 ]; then + echo "Usage: ./test.sh <cpus> <runtime> <sleeptime>" + echo "cpus: number of cpus in the CPU0's frequency domain" + echo "runtime: running time in ms per loop of the workload pattern" + echo "sleeptime: sleeping time in ms per loop of the workload pattern" + echo -e "\nExample: \n\"./test.sh 4 100 1000\" means\nCPU0~CPU3 sharing frequency, \"100ms run + 1000ms sleep\" workload pattern.\n" + exit +fi + +echo "Frequency domain CPU0~CPU$(expr $1 - 1), run $2ms, sleep $3ms:" + +sleep 1 +PATH=$PATH:. +set_calibration +test_efficiency $1 $(expr $2 \* 1000) $(expr $3 \* 1000) + -- 1.9.1

10 years, 1 month

[PATCH v3] doc: measure the efficiency of cpufreq governors

by Xunlei Pang

From: Xunlei Pang <pang.xunlei(a)linaro.org> DVFS adds a latency in the execution of task because of the time to decide to move at max freq. We need to measure this latency and check that the governor stays in an acceptable range. When rt-app runs a json file, a log file is created for each thread. This log file records the number of loop that has been executed and the duration for executing these loops (per phase). We can use these figures to evaluate to latency that is added by a cpufreq governor and its "performance efficiency". We use the run+sleep pattern to do the measurement, for the run time per loop, the performance governor should run the expected duration as the CPU stays a max freq. At the opposite, the powersave governor will give use the longest duration (as it stays at lowest OPP). Other governor will be somewhere between the 2 previous duration as they will use several OPP and will go back to max frequency after a defined duration which depends on its monitoring period. The formula: duration of powersave gov - duration of the gov -------------------------------------------------------- x 100% duration of powersave gov - duration of performance gov will give the efficiency of the governor. 100% means as efficient as the perf governor and 0% means as efficient as the powersave governor. This patch offers json files and shell scripts to do the measurement, Usage: ./calibration.sh <cpu> cpu: cpu number on which you want to run the test ./test.sh <governor> <cpu> <runtime> <sleeptime> [<loops>] governor: target CPUFreq governor you want to test cpu: cpu number on which you want to run the test. Be the same as the one passed to "calibration.sh". runtime: running time in ms per loop of the workload pattern sleeptime: sleeping time in ms per loop of the workload pattern loops: repeat times of the workload pattern. default: 10 Example: "./calibration.sh 0" means to calculate the computing capacity of CPU0 which will be used in the following test. "./test.sh ondemand 0 100 100 20" means to test "ondemand" on CPU0 with workload pattern "run 100ms + sleep 100ms"(20 loops). NOTE: - Make sure there are "sed", "cut", "grep", "rt-app", etc tools on your test machine, and run the scripts under root privilege. - Run the test while the system is idle. - You can change the target governor's parameters after running "calibration.sh", but before "test.sh". Signed-off-by: Xunlei Pang <pang.xunlei(a)linaro.org> --- doc/examples/cpufreq_governor_efficiency/README | 60 ++++++++++++ .../cpufreq_governor_efficiency/calibration.json | 26 ++++++ .../cpufreq_governor_efficiency/calibration.sh | 17 ++++ doc/examples/cpufreq_governor_efficiency/dvfs.json | 27 ++++++ doc/examples/cpufreq_governor_efficiency/dvfs.sh | 36 +++++++ doc/examples/cpufreq_governor_efficiency/test.sh | 104 +++++++++++++++++++++ 6 files changed, 270 insertions(+) create mode 100644 doc/examples/cpufreq_governor_efficiency/README create mode 100644 doc/examples/cpufreq_governor_efficiency/calibration.json create mode 100755 doc/examples/cpufreq_governor_efficiency/calibration.sh create mode 100644 doc/examples/cpufreq_governor_efficiency/dvfs.json create mode 100755 doc/examples/cpufreq_governor_efficiency/dvfs.sh create mode 100755 doc/examples/cpufreq_governor_efficiency/test.sh diff --git a/doc/examples/cpufreq_governor_efficiency/README b/doc/examples/cpufreq_governor_efficiency/README new file mode 100644 index 0000000..a6dc6de --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/README @@ -0,0 +1,60 @@ +Measure the efficiency of cpufreq governors using rt-app + +BACKGROUND: + DVFS adds a latency in the execution of task because of the time to + decide to move at max freq. We need to measure this latency and check + that the governor stays in an acceptable range. + + When rt-app runs a json file, a log file is created for each thread. + This log file records the number of loop that has been executed and + the duration for executing these loops (per phase). We can use these + figures to evaluate to latency that is added by a cpufreq governor + and its "performance efficiency". + + We use the run+sleep pattern to do the measurement, for the run time per + loop, the performance governor should run the expected duration as the + CPU stays a max freq. At the opposite, the powersave governor will give + use the longest duration (as it stays at lowest OPP). Other governor will + be somewhere between the 2 previous duration as they will use several OPP + and will go back to max frequency after a defined duration which depends + on its monitoring period. + + The formula: + + duration of powersave gov - duration of the gov + -------------------------------------------------------- x 100% + duration of powersave gov - duration of performance gov + + will give the efficiency of the governor. 100% means as efficient as + the perf governor and 0% means as efficient as the powersave governor. + + This test offers json files and shell scripts to do the measurement, + +Usage: + ./calibration.sh <cpu> + cpu: cpu number on which you want to run the test + + ./test.sh <governor> <cpu> <runtime> <sleeptime> [<loops>] + governor: target CPUFreq governor you want to test + cpu: cpu number on which you want to run the test. Be the same + as the one passing to "calibration.sh". + runtime: running time in ms per loop of the workload pattern + sleeptime: sleeping time in ms per loop of the workload pattern + loops: repeat times of the workload pattern. default: 10 + +Example: + "./calibration.sh 0" means to calculate the computing capacity of CPU0 which + will be used in the following test. + + "./test.sh ondemand 0 100 100 20" means to + test "ondemand" on CPU0 with workload pattern "run 100ms + sleep 100ms"(20 loops). + +NOTE: + - Make sure there are "sed", "cut", "grep", "rt-app", etc tools on + your test machine, and run the scripts under root privilege. + + - Run the test while the system is idle. + + - You can change the target governor's parameters after running "calibration.sh", + but before "test.sh". + diff --git a/doc/examples/cpufreq_governor_efficiency/calibration.json b/doc/examples/cpufreq_governor_efficiency/calibration.json new file mode 100644 index 0000000..2d5870c --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/calibration.json @@ -0,0 +1,26 @@ +{ + "tasks" : { + "thread" : { + "instance" : 1, + "loop" : 1, + "phases" : { + "run" : { + "loop" : 1, + "run" : 2000, + }, + "sleep" : { + "loop" : 1, + "sleep" : 2000, + } + } + } + }, + "global" : { + "default_policy" : "SCHED_FIFO", + "calibration" : "CPU0", + "lock_pages" : true, + "ftrace" : false, + "logdir" : "./", + } +} + diff --git a/doc/examples/cpufreq_governor_efficiency/calibration.sh b/doc/examples/cpufreq_governor_efficiency/calibration.sh new file mode 100755 index 0000000..a217487 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/calibration.sh @@ -0,0 +1,17 @@ +#!/bin/sh + +set -e + +if [ ! $1 ] ; then + echo "Please input one cpu" + exit +fi + +echo performance > /sys/devices/system/cpu/cpu$1/cpufreq/scaling_governor + +sleep 1 + +sed 's/"calibration" : "CPU.*",/"calibration" : "CPU'$1'",/' -i calibration.json +pLoad=$(rt-app calibration.json 2>&1 |grep pLoad |sed 's/.*= $.*$ns.*/\1/') +sed 's/"calibration" : .*,/"calibration" : '$pLoad',/' -i dvfs.json +echo CPU$1\'s pLoad is $pLoad diff --git a/doc/examples/cpufreq_governor_efficiency/dvfs.json b/doc/examples/cpufreq_governor_efficiency/dvfs.json new file mode 100644 index 0000000..9a1accf --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/dvfs.json @@ -0,0 +1,27 @@ +{ + "tasks" : { + "thread" : { + "instance" : 1, + "cpus" : [0], +"loop" : 10, + "phases" : { + "running" : { + "loop" : 1, + "run" : 10000, + }, + "sleeping" : { + "loop" : 1, + "sleep" : 10000, + } + } + } + }, + "global" : { + "default_policy" : "SCHED_OTHER", + "calibration" : 90, + "lock_pages" : true, + "ftrace" : false, + "logdir" : "./", + } +} + diff --git a/doc/examples/cpufreq_governor_efficiency/dvfs.sh b/doc/examples/cpufreq_governor_efficiency/dvfs.sh new file mode 100755 index 0000000..014531f --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/dvfs.sh @@ -0,0 +1,36 @@ +#!/bin/sh + +# $1 $2 $3 $4 $5: governor cpu run sleep loops +set -e + +echo $1 > /sys/devices/system/cpu/cpu$2/cpufreq/scaling_governor +#echo $1 > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor +sed 's/"cpus" : \[.*\],/"cpus" : \['$2'\],/' -i dvfs.json +sleep 3 + +if [ $3 ] ; then + sed 's/"run" : .*,/"run" : '$3',/' -i dvfs.json +fi + +if [ $4 ] ; then + sed 's/"sleep" : .*,/"sleep" : '$4',/' -i dvfs.json +fi + +if [ $5 ] ; then + sed 's/^"loop" : .*,/"loop" : '$5',/' -i dvfs.json +fi + +rt-app dvfs.json 2> /dev/null + +if [ $1 ] ; then + mv -f rt-app-thread-0.log rt-app_$1_run$3us_sleep$4us.log + + sum=0 + for i in $(cat rt-app_$1_run$3us_sleep$4us.log | sed 'n;d' | sed '1d' |cut -f 3); do + sum=$(expr $sum + $i) + done + sum=$(expr $sum / 5) + echo $sum + rm -f rt-app_$1_run$3us_sleep$4us.log +fi + diff --git a/doc/examples/cpufreq_governor_efficiency/test.sh b/doc/examples/cpufreq_governor_efficiency/test.sh new file mode 100755 index 0000000..0160952 --- /dev/null +++ b/doc/examples/cpufreq_governor_efficiency/test.sh @@ -0,0 +1,104 @@ +#!/bin/sh + +set -e + +test_efficiency() { + + FILENAME="results_$RANDOM$$.txt" + + if [ -e /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors ]; then + for i in $(cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors); do + if [ $i = $1 ] ; then + gov_target=$i + fi + export gov_$i=$(echo $i) + done + else + echo "cpufreq sysfs is not available!" + exit + fi + + if [ ! $gov_target ] ; then + echo " Can't find $1 governor!" + exit + fi + + if [ ! $gov_performance ] ; then + echo "Can't find performance governor!" + exit + fi + + if [ ! $gov_powersave ] ; then + echo "Can't find powersave governor!" + exit + fi + + if [ $gov_target = $gov_performance ] || [ $gov_target = $gov_powersave ] ; then + echo "Please input a governor other than \"performance\" or \"powersave\"" + exit + fi + + # Get target gov data first + dvfs.sh $1 $2 $3 $4 $5 > $FILENAME + target=$(cat $FILENAME |sed -n '1p') + + # Get powersave data + dvfs.sh powersave $2 $3 $4 $5 > $FILENAME + powersave=$(cat $FILENAME |sed -n '1p') + + # Get performance data + dvfs.sh performance $2 $3 $4 $5 > $FILENAME + performance=$(cat $FILENAME |sed -n '1p') + + if [ $performance -ge $powersave ] ; then + echo "powersave: $powersave" + echo "performance: $performance" + echo "Error! performance spent more time than powersave!" + exit + fi + + echo "\"powersave\" efficiency: 0%" + echo "\"performance\" efficiency: 100%" + + denominator=$(expr $powersave - $performance) + + if [ $powersave -le $target ]; then + target=0 + else + numerator=$(expr $powersave - $target) + numerator=$(expr $numerator \* 100) + target=$(expr $numerator / $denominator) + if [ $target -gt 100 ]; then + target=100 + fi + fi + + echo "\"$gov_target\" efficiency: $target%" + + rm -f $FILENAME +} + +if [ $# -lt 4 ]; then + echo "Usage: ./test.sh <governor> <cpu> <runtime> <sleeptime> [<loops>]" + echo "governor: target CPUFreq governor you want to test" + echo "cpu: cpu number on which you want to run the test" + echo "runtime: running time in ms per loop of the workload pattern" + echo "sleeptime: sleeping time in ms per loop of the workload pattern" + echo "loops: repeat times of the workload pattern. default: 10" + echo "\nExample:\n\"./test.sh ondemand 0 100 100 20\" means\nTest \"ondemand\" on CPU0 with workload pattern \"run 100ms + sleep 100ms\"(20 loops).\n" + exit +fi + +if [ $# = 4 ]; then + loops=10 +else + loops=$5 +fi + +echo "Test \"$1\" on CPU$2 with workload pattern \"run $3ms + sleep $4ms\"($loops loops)." + +sleep 1 +PATH=$PATH:. + +test_efficiency $1 $2 $(expr $3 \* 1000) $(expr $4 \* 1000) $loops + -- 1.9.1

10 years, 1 month

[PATCH] rt-app: remove compiler warnings during parse_config

by Jorge Ramirez-Ortiz

This patch fixes the compiler warnings caused by the invalid use of const. It also provides a certain level of encapsulation when parsing the thread_event data. Some of the macros could be merged simplifying the code a bit more but I am not sure if it would make the code too obfuscated.

10 years, 1 month

[PATCH] rt-app: remove compiler warnings during parse_config

by Jorge Ramirez-Ortiz

10 years, 1 month

[PATCH v2] idlestat: Support energy modeling of idle consumption at different OPPs

by Xunlei Pang

From: Xunlei Pang <pang.xunlei(a)linaro.org> BACKGROUND Overview: Currently idlestat only handles a energy model with a single value for energy savings when going to WFI. IOW, idlestat does not know what OPP (frequency, voltage) is this WFI energy savings calculated at. Add funtionality to allow a table of OPP and energy saving values to be listed so that the savings at various frequencies can be modeled correctly. As an example, a platform may have different power consumption when idled at different OPPs, as follow: @1989 MHz: single core in WFI consumes about 120 mW @1001 MHz: single core in WFI consumes about 45 mW @ 507 MHz: single core in WFI consumes about 30 mW This data is useful to optimise a platform's idle policy. MODIFICATION METHOD Now we want to modify the format of idlestat's energy model to accomodate the different C-states energy data at different OPPs. Let's take a simple example to explicate the way: Assuming different C-States energy data below: 1200Mhz: WFI 80 10 1200Mhz: cluster-sleep-b 30 5 1000Mhz: WFI 70 0 1000Mhz: cluster-sleep-b 25 0 Before modification: clusterA: 2 cap states 2 C states P-states: 1200 15200 6997 1000 8446 3846 C-states: WFI 70 0 cluster-sleep-b 25 0 After modification: clusterA: 2 cap states 2 C states P-state C-states pair 1: 1200 15200 6997 WFI 80 10 cluster-sleep-b 30 5 P-state C-states pair 2: 1000 8446 3846 WFI 70 0 cluster-sleep-b 25 0 In this way, we ensure P-states and C-States always appear in pairs. Signed-off-by: Xunlei Pang <pang.xunlei(a)linaro.org> --- energy_model.c | 101 ++++++++++++++++++++++----------------------------- idlestat.c | 64 ++++++++++++++++++++++++++++++-- idlestat.h | 10 +++++ tracefile_idlestat.c | 4 ++ 4 files changed, 119 insertions(+), 60 deletions(-) diff --git a/energy_model.c b/energy_model.c index 2ea6fdd..8beb6a8 100644 --- a/energy_model.c +++ b/energy_model.c @@ -75,41 +75,38 @@ static int make_energy_model_template(struct program_options *options) list_for_each_entry(s_phy, &cpu_topo->physical_head, list_physical) { unsigned int num_cap_states = 0; unsigned int num_c_states = 0; - int i; + int i, j, pair; s_core = list_entry((&s_phy->core_head)->prev, struct cpu_core, list_core); s_cpu = list_entry((&s_core->cpu_head)->prev, struct cpu_cpu, list_cpu); num_c_states = s_cpu->cstates->cstate_max + 1; for (i = 0; i < s_cpu->pstates->max; i++) { struct cpufreq_pstate *p = &s_cpu->pstates->pstate[i]; - if (p->freq == 0) continue; num_cap_states++; } - fprintf(f, "\nC-states:\n"); - fprintf(f, "cluster%c: %d cap states %d C states\n\n", cluster_number + 'A', num_cap_states, num_c_states); - fprintf(f, "P-states:\n"); - fprintf(f, "# speed, cluster power, core power\n"); + pair = 0; for (i = 0; i < s_cpu->pstates->max; i++) { struct cpufreq_pstate *p = &s_cpu->pstates->pstate[i]; if (p->freq == 0) continue; + fprintf(f, "P-state C-states pair %d:\n", ++pair); + fprintf(f, "# P-state: speed, cluster power, core power\n"); fprintf(f, "%d\t\t?\t?\n", p->freq/1000); - } - fprintf(f, "\nC-states:\n"); - fprintf(f, "# name, cluster power, core power\n"); - for (i = 0; i < s_cpu->cstates->cstate_max + 1; i++) { - struct cpuidle_cstate *c = &s_cpu->cstates->cstate[i]; + fprintf(f, "# C-states: name, cluster power, core power\n"); + for (j = 0; j < s_cpu->cstates->cstate_max + 1; j++) { + struct cpuidle_cstate *c = &s_cpu->cstates->cstate[j]; - fprintf(f, "%s\t\t?\t?\n", c->name); + fprintf(f, "%s\t\t?\t?\n", c->name); + } } fprintf(f, "\n"); cluster_number++; @@ -130,7 +127,7 @@ int parse_energy_model(struct program_options *options) struct cluster_energy_info *clustp = NULL; unsigned int number_cap_states, number_c_states; int current_cluster = -1; - unsigned int current_pstate = 0; + unsigned int current_pstate = 0, speed; unsigned int current_cstate = 0; unsigned int clust_p, core_p; char buffer[BUFSIZE]; @@ -196,12 +193,13 @@ int parse_energy_model(struct program_options *options) clustp->number_c_states = number_c_states; clustp->p_energy = calloc(number_cap_states, sizeof(struct pstate_energy_info)); - clustp->c_energy = calloc(number_c_states, + clustp->c_energy = calloc(number_c_states * number_cap_states, sizeof(struct cstate_energy_info)); clustp->state = parsed_cluster_info; + current_pstate = 0; continue; } - if (strstr(buffer, "P-states")) { + if (strstr(buffer, "P-state C-states pair")) { if (current_cluster == -1) { fprintf(stderr, "%s: unknown cluster (cap states) in %s\n", __func__, path); @@ -214,27 +212,9 @@ int parse_energy_model(struct program_options *options) fclose(f); return -1; } - current_pstate = 0; clustp->state = parsing_cap_states; continue; } - if (strstr(buffer, "C-states")) { - if (current_cluster == -1) { - fprintf(stderr, "%s: unknown cluster (c states) in %s\n", - __func__, path); - fclose(f); - return -1; - } - if (clustp->state < parsed_cluster_info) { - fprintf(stderr, "%s: number of c states for cluster%c not specified in %s\n", - __func__, current_cluster, path); - fclose(f); - return -1; - } - current_cstate = 0; - clustp->state = parsing_c_states; - continue; - } if (strstr(buffer, "wakeup")) { unsigned int clust_w, core_w; @@ -254,10 +234,9 @@ int parse_energy_model(struct program_options *options) __func__, path); fclose(f); return -1; - } + } if (clustp->state == parsing_cap_states) { struct pstate_energy_info *pp; - unsigned int speed; if (sscanf(buffer, "%d %d %d", &speed, &clust_p, &core_p) != 3) { fprintf(stderr, "%s: expected P state (speed cluster core) for cluster%c in %s\n", @@ -276,6 +255,8 @@ int parse_energy_model(struct program_options *options) pp->speed = speed; pp->cluster_power = clust_p; pp->core_power = core_p; + clustp->state = parsing_c_states; + current_cstate = 0; continue; } if (clustp->state == parsing_c_states) { @@ -295,8 +276,9 @@ int parse_energy_model(struct program_options *options) fclose(f); return -1; } - cp = &clustp->c_energy[current_cstate++]; + cp = &clustp->c_energy[(current_pstate - 1) * clustp->number_c_states + current_cstate++]; strncpy(cp->cstate_name, name, NAMELEN); + cp->speed = speed; cp->cluster_idle_power = clust_p; cp->core_idle_power = core_p; continue; @@ -309,7 +291,7 @@ int parse_energy_model(struct program_options *options) return 0; } -static struct cstate_energy_info *find_cstate_energy_info(const unsigned int cluster, const char *name) +static struct cstate_energy_info *find_cstate_energy_info(const unsigned int cluster, unsigned int speed, const char *name) { struct cluster_energy_info *clustp; struct cstate_energy_info *cp; @@ -317,8 +299,8 @@ static struct cstate_energy_info *find_cstate_energy_info(const unsigned int clu clustp = cluster_energy_table + cluster; cp = &clustp->c_energy[0]; - for (i = 0; i < clustp->number_c_states; i++, cp++) { - if (!strcmp(cp->cstate_name, name)) return cp; + for (i = 0; i < clustp->number_c_states * clustp->number_cap_states; i++, cp++) { + if (cp->speed == speed && !strcmp(cp->cstate_name, name)) return cp; } return NULL; } @@ -377,6 +359,7 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) for (j = 0; j < s_phy->cstates->cstate_max + 1; j++) { struct cpuidle_cstate *c = &s_phy->cstates->cstate[j]; + struct duration_opp *opps = c->duration_opps; if (c->nrdata == 0) { verbose_fprintf(stderr, 2, @@ -385,19 +368,20 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) continue; } - cp = find_cstate_energy_info(current_cluster, c->name); - if (!cp) { - verbose_fprintf(stderr, 2, " C%-2d no energy model for [%s] (%d hits, %f duration)\n", - j, c->name, c->nrdata, c->duration); - continue; - } + for (i = 0; i < c->nropp; i++) { + cp = find_cstate_energy_info(current_cluster, opps[i].speed/1000, c->name); + if (!cp) { + verbose_fprintf(stderr, 2, " C%-2d no energy model for [%s] (%d hits, %f duration) %d speed\n", + j, c->name, c->nrdata, opps[i].duration, opps[i].speed); + continue; + } - cluster_idl += c->duration * cp->cluster_idle_power; + cluster_idl += opps[i].duration * cp->cluster_idle_power; + } - verbose_fprintf(stderr, 1, " C%-2d +%7d hits for [%15s] | %13.0f | %7d | %7s | %12s | %12.0f | %12s |\n", + verbose_fprintf(stderr, 1, " C%-2d +%7d hits for [%15s] | %13.0f | %7s | %12s | %12.0f | %12s |\n", j, c->nrdata, c->name, c->duration, - cp->cluster_idle_power, "", "", cluster_idl, ""); @@ -442,25 +426,28 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) for (i = 0; i < s_cpu->cstates->cstate_max + 1; i++) { struct cpuidle_cstate *c = &s_cpu->cstates->cstate[i]; + struct duration_opp *opps = c->duration_opps; + if (c->nrdata == 0) { verbose_fprintf(stderr, 2, "Cpu%d C%-2d no hits for [%15s] | 0 | 0 | | | 0 | |\n", s_cpu->cpu_id, i, c->name); continue; } - cp = find_cstate_energy_info(current_cluster, c->name); - if (!cp) { - verbose_fprintf(stderr, 2, "Cpu%d C%-2d no energy model for [%s] (%d hits, %f duration)\n", - s_cpu->cpu_id, i, c->name, - c->nrdata, c->duration); - continue; + for (j = 0; j < c->nropp; j++) { + cp = find_cstate_energy_info(current_cluster, opps[j].speed/100, c->name); + if (!cp) { + verbose_fprintf(stderr, 2, "Cpu%d C%-2d no energy model for [%s] (%d hits, %f duration, %d speed)\n", + s_cpu->cpu_id, i, c->name, + c->nrdata, opps[j].duration, opps[j].speed); + continue; + } + cluster_idl += opps[j].duration * cp->core_idle_power; } - cluster_idl += c->duration * cp->core_idle_power; - verbose_fprintf(stderr, 1, "Cpu%d C%-2d +%7d hits for [%15s] | %13.0f | %7d | %7s | %12s | %12.0f | %12s |\n", + verbose_fprintf(stderr, 1, "Cpu%d C%-2d +%7d hits for [%15s] | %13.0f | %7s | %12s | %12.0f | %12s |\n", s_cpu->cpu_id, i, c->nrdata, c->name, c->duration, - cp->core_idle_power, "", "", cluster_idl, ""); diff --git a/idlestat.c b/idlestat.c index 4d773f4..75ac1fc 100644 --- a/idlestat.c +++ b/idlestat.c @@ -335,9 +335,10 @@ struct cpuidle_cstates *build_cstate_info(int nrcpus) c->avg_time = 0.; c->max_time = 0.; c->min_time = DBL_MAX; - c->duration = 0.; c->target_residency = cpuidle_get_target_residency(cpu, i); + c->duration_opps = NULL; + c->nropp = 0; } } return cstates; @@ -739,6 +740,7 @@ static int cstate_begin(double time, int state, struct cpuidle_cstates *cstates) memset(data + nrdata, 0, sizeof(*data)); data[nrdata].begin = time; + data[nrdata].speed = cstates->current_speed; cstate->data = data; cstates->cstate_max = MAX(cstates->cstate_max, state); @@ -747,7 +749,34 @@ static int cstate_begin(double time, int state, struct cpuidle_cstates *cstates) return 0; } -static void cstate_end(double time, struct cpuidle_cstates *cstates) +static int update_duration_opp(struct cpuidle_cstate *cstate, unsigned int freq, double duration) +{ + struct duration_opp *opps = cstate->duration_opps; + int i; + + for (i = 0; i < cstate->nropp; i++) { + if (opps[i].speed == freq) { + opps[i].duration += duration; + return 0; + } + } + + opps = realloc(cstate->duration_opps, sizeof(*opps) * (cstate->nropp + 1)); + if (!opps) { + free(cstate->duration_opps); + return error(__func__); + } + + cstate->duration_opps = opps; + memset(opps + cstate->nropp, 0, sizeof(*opps)); + opps[cstate->nropp].speed = freq; + opps[cstate->nropp].duration += duration; + cstate->nropp++; + + return 0; +} + +static int cstate_end(double time, struct cpuidle_cstates *cstates) { int last_cstate = cstates->current_cstate; struct cpuidle_cstate *cstate = &cstates->cstate[last_cstate]; @@ -792,12 +821,17 @@ static void cstate_end(double time, struct cpuidle_cstates *cstates) cstate->max_time = MAX(cstate->max_time, data->duration); cstate->avg_time = AVG(cstate->avg_time, data->duration, cstate->nrdata + 1); + + if (update_duration_opp(cstate, data->speed, data->duration) < 0) + return error(__func__); + cstate->duration += data->duration; cstate->nrdata++; skip_entry: /* CPU is no longer idle */ cstates->current_cstate = -1; + return 0; } int record_cstate_event(struct cpuidle_cstates *cstates, @@ -810,13 +844,35 @@ int record_cstate_event(struct cpuidle_cstates *cstates, return 0; if (cstates->current_cstate != -1) - cstate_end(time, cstates); + ret = cstate_end(time, cstates); if (state != -1) ret = cstate_begin(time, state, cstates); return ret; } +static void core_update_current_speed(struct cpu_core *core) +{ + struct cpu_cpu *cpu; + unsigned int max_speed = 0; + + core_for_each_cpu(cpu, core) + max_speed = MAX(max_speed, cpu->cstates->current_speed); + + core->cstates->current_speed = max_speed; +} + +static void cluster_update_current_speed(struct cpu_physical *clust) +{ + struct cpu_cpu *cpu; + unsigned int max_speed = 0; + + cluster_for_each_cpu(cpu, clust) + max_speed = MAX(max_speed, cpu->cstates->current_speed); + + clust->cstates->current_speed = max_speed; +} + int store_data(double time, int state, int cpu, struct cpuidle_datas *datas) { @@ -843,11 +899,13 @@ int store_data(double time, int state, int cpu, /* Update core and cluster */ aff_core = cpu_to_core(cpu, datas->topo); state = core_get_least_cstate(aff_core); + core_update_current_speed(aff_core); if (record_cstate_event(aff_core->cstates, time, state) == -1) return -1; aff_cluster = cpu_to_cluster(cpu, datas->topo); state = cluster_get_least_cstate(aff_cluster); + cluster_update_current_speed(aff_cluster); if (record_cstate_event(aff_cluster->cstates, time,state) == -1) return -1; diff --git a/idlestat.h b/idlestat.h index e030f6a..2eef6c9 100644 --- a/idlestat.h +++ b/idlestat.h @@ -48,11 +48,17 @@ "/sys/devices/system/cpu/cpu%d/cpufreq/cpuinfo_cur_freq" struct cpuidle_data { + unsigned int speed; double begin; double end; double duration; }; +struct duration_opp { + unsigned int speed; + double duration; +}; + struct cpuidle_cstate { char *name; struct cpuidle_data *data; @@ -62,7 +68,9 @@ struct cpuidle_cstate { double avg_time; double max_time; double min_time; + struct duration_opp *duration_opps; double duration; + int nropp; int target_residency; /* -1 if not available */ }; @@ -83,6 +91,7 @@ struct cpuidle_cstates { struct cpuidle_cstate cstate[MAXCSTATE]; struct wakeup_info wakeinfo; int current_cstate; + unsigned int current_speed; int cstate_max; struct wakeup_irq *wakeirq; enum {as_expected, too_long, too_short} actual_residency; @@ -157,6 +166,7 @@ struct pstate_energy_info { struct cstate_energy_info { char cstate_name[NAMELEN]; + unsigned int speed; unsigned int cluster_idle_power; unsigned int core_idle_power; double cluster_duration; diff --git a/tracefile_idlestat.c b/tracefile_idlestat.c index 3430693..79f8578 100644 --- a/tracefile_idlestat.c +++ b/tracefile_idlestat.c @@ -108,6 +108,8 @@ static struct cpuidle_cstates *load_and_build_cstate_info(FILE* f, char *buffer, c->max_time = 0.; c->min_time = DBL_MAX; c->duration = 0.; + c->duration_opps = NULL; + c->nropp = 0; c->target_residency = residency; } fgets(buffer, BUFSIZE, f); @@ -146,6 +148,8 @@ int load_text_data_line(char *buffer, struct cpuidle_datas *datas, char *format, "be wrong.\n"); return -1; } + + datas->cstates[cpu].current_speed = freq; return cpu_change_pstate(datas, cpu, freq, time); } -- 1.9.1

10 years, 1 month

[PATCH v2] doc: measure the efficiency of cpufreq governors

by Xunlei Pang

10 years, 1 month

[RFC PATCH] idlestat: Support energy modeling of idle consumption at different OPPs

by Xunlei Pang

From: Xunlei Pang <pang.xunlei(a)linaro.org> BACKGROUND Overview: Currently idlestat only handles a energy model with a single value for energy savings when going to WFI. IOW, idlestat does not know what OPP (frequency, voltage) is this WFI energy savings calculated at. Add funtionality to allow a table of OPP and energy saving values to be listed so that the savings at various frequencies can be modeled correctly. As an example, a platform may have different power consumption when idled at different OPPs, as follow: @1989 MHz: single core in WFI consumes about 120 mW @1001 MHz: single core in WFI consumes about 45 mW @ 507 MHz: single core in WFI consumes about 30 mW This data is useful to optimise a platform's idle policy. MODIFICATION METHOD Now we want to modify the format of idlestat's energy model to accomodate the different C-states energy data at different OPPs. Let's take a simple example to explicate the way: Assuming different C-States energy data below: 1200Mhz: WFI 80 10 1200Mhz: cluster-sleep-b 30 5 1000Mhz: WFI 70 0 1000Mhz: cluster-sleep-b 25 0 Before modification: clusterA: 2 cap states 2 C states P-states: 1200 15200 6997 1000 8446 3846 C-states: WFI 70 0 cluster-sleep-b 25 0 After modification: clusterA: 2 cap states 2 C states P-state C-states pair 1: 1200 15200 6997 WFI 80 10 cluster-sleep-b 30 5 P-state C-states pair 2: 1000 8446 3846 WFI 70 0 cluster-sleep-b 25 0 In this way, we ensure P-states and C-States always appear in pairs. Signed-off-by: Xunlei Pang <pang.xunlei(a)linaro.org> --- energy_model.c | 104 ++++++++++++++++++++++----------------------------- idlestat.c | 64 +++++++++++++++++++++++++++++-- idlestat.h | 10 +++++ trace_ops.h | 1 - tracefile_idlestat.c | 4 ++ 5 files changed, 120 insertions(+), 63 deletions(-) diff --git a/energy_model.c b/energy_model.c index 2ea6fdd..3932c23 100644 --- a/energy_model.c +++ b/energy_model.c @@ -75,7 +75,7 @@ static int make_energy_model_template(struct program_options *options) list_for_each_entry(s_phy, &cpu_topo->physical_head, list_physical) { unsigned int num_cap_states = 0; unsigned int num_c_states = 0; - int i; + int i, j; s_core = list_entry((&s_phy->core_head)->prev, struct cpu_core, list_core); s_cpu = list_entry((&s_core->cpu_head)->prev, struct cpu_cpu, list_cpu); @@ -89,27 +89,23 @@ static int make_energy_model_template(struct program_options *options) num_cap_states++; } - fprintf(f, "\nC-states:\n"); - fprintf(f, "cluster%c: %d cap states %d C states\n\n", cluster_number + 'A', num_cap_states, num_c_states); - fprintf(f, "P-states:\n"); - fprintf(f, "# speed, cluster power, core power\n"); - for (i = 0; i < s_cpu->pstates->max; i++) { struct cpufreq_pstate *p = &s_cpu->pstates->pstate[i]; if (p->freq == 0) continue; + fprintf(f, "P-state C-states pair %d:\n", i + 1); + fprintf(f, "# P-state: speed, cluster power, core power\n"); fprintf(f, "%d\t\t?\t?\n", p->freq/1000); - } - fprintf(f, "\nC-states:\n"); - fprintf(f, "# name, cluster power, core power\n"); - for (i = 0; i < s_cpu->cstates->cstate_max + 1; i++) { - struct cpuidle_cstate *c = &s_cpu->cstates->cstate[i]; + fprintf(f, "# C-states: name, cluster power, core power\n"); + for (j = 0; j < s_cpu->cstates->cstate_max + 1; j++) { + struct cpuidle_cstate *c = &s_cpu->cstates->cstate[j]; - fprintf(f, "%s\t\t?\t?\n", c->name); + fprintf(f, "%s\t\t?\t?\n", c->name); + } } fprintf(f, "\n"); cluster_number++; @@ -130,8 +126,8 @@ int parse_energy_model(struct program_options *options) struct cluster_energy_info *clustp = NULL; unsigned int number_cap_states, number_c_states; int current_cluster = -1; - unsigned int current_pstate = 0; - unsigned int current_cstate = 0; + unsigned int current_pstate, speed; + unsigned int current_cstate; unsigned int clust_p, core_p; char buffer[BUFSIZE]; char *path = options->energy_model_filename; @@ -196,12 +192,13 @@ int parse_energy_model(struct program_options *options) clustp->number_c_states = number_c_states; clustp->p_energy = calloc(number_cap_states, sizeof(struct pstate_energy_info)); - clustp->c_energy = calloc(number_c_states, + clustp->c_energy = calloc(number_c_states*number_cap_states, sizeof(struct cstate_energy_info)); clustp->state = parsed_cluster_info; + current_pstate = 0; continue; } - if (strstr(buffer, "P-states")) { + if (strstr(buffer, "P-state C-states pair")) { if (current_cluster == -1) { fprintf(stderr, "%s: unknown cluster (cap states) in %s\n", __func__, path); @@ -214,27 +211,9 @@ int parse_energy_model(struct program_options *options) fclose(f); return -1; } - current_pstate = 0; clustp->state = parsing_cap_states; continue; } - if (strstr(buffer, "C-states")) { - if (current_cluster == -1) { - fprintf(stderr, "%s: unknown cluster (c states) in %s\n", - __func__, path); - fclose(f); - return -1; - } - if (clustp->state < parsed_cluster_info) { - fprintf(stderr, "%s: number of c states for cluster%c not specified in %s\n", - __func__, current_cluster, path); - fclose(f); - return -1; - } - current_cstate = 0; - clustp->state = parsing_c_states; - continue; - } if (strstr(buffer, "wakeup")) { unsigned int clust_w, core_w; @@ -254,10 +233,9 @@ int parse_energy_model(struct program_options *options) __func__, path); fclose(f); return -1; - } + } if (clustp->state == parsing_cap_states) { struct pstate_energy_info *pp; - unsigned int speed; if (sscanf(buffer, "%d %d %d", &speed, &clust_p, &core_p) != 3) { fprintf(stderr, "%s: expected P state (speed cluster core) for cluster%c in %s\n", @@ -276,6 +254,8 @@ int parse_energy_model(struct program_options *options) pp->speed = speed; pp->cluster_power = clust_p; pp->core_power = core_p; + clustp->state = parsing_c_states; + current_cstate = 0; continue; } if (clustp->state == parsing_c_states) { @@ -295,8 +275,9 @@ int parse_energy_model(struct program_options *options) fclose(f); return -1; } - cp = &clustp->c_energy[current_cstate++]; + cp = &clustp->c_energy[(current_pstate - 1) * clustp->number_c_states + current_cstate++]; strncpy(cp->cstate_name, name, NAMELEN); + cp->speed = speed; cp->cluster_idle_power = clust_p; cp->core_idle_power = core_p; continue; @@ -309,7 +290,7 @@ int parse_energy_model(struct program_options *options) return 0; } -static struct cstate_energy_info *find_cstate_energy_info(const unsigned int cluster, const char *name) +static struct cstate_energy_info *find_cstate_energy_info(const unsigned int cluster, unsigned int speed, const char *name) { struct cluster_energy_info *clustp; struct cstate_energy_info *cp; @@ -317,8 +298,8 @@ static struct cstate_energy_info *find_cstate_energy_info(const unsigned int clu clustp = cluster_energy_table + cluster; cp = &clustp->c_energy[0]; - for (i = 0; i < clustp->number_c_states; i++, cp++) { - if (!strcmp(cp->cstate_name, name)) return cp; + for (i = 0; i < clustp->number_c_states * clustp->number_cap_states; i++, cp++) { + if (cp->speed == speed && !strcmp(cp->cstate_name, name)) return cp; } return NULL; } @@ -377,6 +358,7 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) for (j = 0; j < s_phy->cstates->cstate_max + 1; j++) { struct cpuidle_cstate *c = &s_phy->cstates->cstate[j]; + struct duration_opp *opps = c->duration_opps; if (c->nrdata == 0) { verbose_fprintf(stderr, 2, @@ -385,19 +367,20 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) continue; } - cp = find_cstate_energy_info(current_cluster, c->name); - if (!cp) { - verbose_fprintf(stderr, 2, " C%-2d no energy model for [%s] (%d hits, %f duration)\n", - j, c->name, c->nrdata, c->duration); - continue; - } + for (i = 0; i < c->nropp; i++) { + cp = find_cstate_energy_info(current_cluster, opps[i].speed/1000, c->name); + if (!cp) { + verbose_fprintf(stderr, 2, " C%-2d no energy model for [%s] (%d hits, %f duration) %d speed\n", + j, c->name, c->nrdata, opps[i].duration, opps[i].speed); + continue; + } - cluster_idl += c->duration * cp->cluster_idle_power; + cluster_idl += opps[i].duration * cp->cluster_idle_power; + } - verbose_fprintf(stderr, 1, " C%-2d +%7d hits for [%15s] | %13.0f | %7d | %7s | %12s | %12.0f | %12s |\n", + verbose_fprintf(stderr, 1, " C%-2d +%7d hits for [%15s] | %13.0f | %7s | %12s | %12.0f | %12s |\n", j, c->nrdata, c->name, c->duration, - cp->cluster_idle_power, "", "", cluster_idl, ""); @@ -415,7 +398,7 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) pp = find_pstate_energy_info(current_cluster, p->freq/1000); if (!pp) { - verbose_fprintf(stderr, 2, "Cluster %c frequency %u MHz no energy model for [%d] (%d hits, %f duration)\n", + verbose_fprintf(stderr, 1, "Cluster %c frequency %u MHz no energy model for [%d] (%d hits, %f duration)\n", s_phy->physical_id + 'A', p->freq/1000, p->count, p->duration); continue; @@ -442,25 +425,28 @@ void calculate_energy_consumption(struct cpu_topology *cpu_topo) for (i = 0; i < s_cpu->cstates->cstate_max + 1; i++) { struct cpuidle_cstate *c = &s_cpu->cstates->cstate[i]; + struct duration_opp *opps = c->duration_opps; + if (c->nrdata == 0) { verbose_fprintf(stderr, 2, "Cpu%d C%-2d no hits for [%15s] | 0 | 0 | | | 0 | |\n", s_cpu->cpu_id, i, c->name); continue; } - cp = find_cstate_energy_info(current_cluster, c->name); - if (!cp) { - verbose_fprintf(stderr, 2, "Cpu%d C%-2d no energy model for [%s] (%d hits, %f duration)\n", - s_cpu->cpu_id, i, c->name, - c->nrdata, c->duration); - continue; + for (j = 0; j < c->nropp; j++) { + cp = find_cstate_energy_info(current_cluster, opps[j].speed/100, c->name); + if (!cp) { + verbose_fprintf(stderr, 2, "Cpu%d C%-2d no energy model for [%s] (%d hits, %f duration, %d speed)\n", + s_cpu->cpu_id, i, c->name, + c->nrdata, opps[j].duration, opps[j].speed); + continue; + } + cluster_idl += opps[j].duration * cp->core_idle_power; } - cluster_idl += c->duration * cp->core_idle_power; - verbose_fprintf(stderr, 1, "Cpu%d C%-2d +%7d hits for [%15s] | %13.0f | %7d | %7s | %12s | %12.0f | %12s |\n", + verbose_fprintf(stderr, 2, "Cpu%d C%-2d +%7d hits for [%15s] | %13.0f | %7s | %12s | %12.0f | %12s |\n", s_cpu->cpu_id, i, c->nrdata, c->name, c->duration, - cp->core_idle_power, "", "", cluster_idl, ""); diff --git a/idlestat.c b/idlestat.c index 4d773f4..75ac1fc 100644 --- a/idlestat.c +++ b/idlestat.c @@ -335,9 +335,10 @@ struct cpuidle_cstates *build_cstate_info(int nrcpus) c->avg_time = 0.; c->max_time = 0.; c->min_time = DBL_MAX; - c->duration = 0.; c->target_residency = cpuidle_get_target_residency(cpu, i); + c->duration_opps = NULL; + c->nropp = 0; } } return cstates; @@ -739,6 +740,7 @@ static int cstate_begin(double time, int state, struct cpuidle_cstates *cstates) memset(data + nrdata, 0, sizeof(*data)); data[nrdata].begin = time; + data[nrdata].speed = cstates->current_speed; cstate->data = data; cstates->cstate_max = MAX(cstates->cstate_max, state); @@ -747,7 +749,34 @@ static int cstate_begin(double time, int state, struct cpuidle_cstates *cstates) return 0; } -static void cstate_end(double time, struct cpuidle_cstates *cstates) +static int update_duration_opp(struct cpuidle_cstate *cstate, unsigned int freq, double duration) +{ + struct duration_opp *opps = cstate->duration_opps; + int i; + + for (i = 0; i < cstate->nropp; i++) { + if (opps[i].speed == freq) { + opps[i].duration += duration; + return 0; + } + } + + opps = realloc(cstate->duration_opps, sizeof(*opps) * (cstate->nropp + 1)); + if (!opps) { + free(cstate->duration_opps); + return error(__func__); + } + + cstate->duration_opps = opps; + memset(opps + cstate->nropp, 0, sizeof(*opps)); + opps[cstate->nropp].speed = freq; + opps[cstate->nropp].duration += duration; + cstate->nropp++; + + return 0; +} + +static int cstate_end(double time, struct cpuidle_cstates *cstates) { int last_cstate = cstates->current_cstate; struct cpuidle_cstate *cstate = &cstates->cstate[last_cstate]; @@ -792,12 +821,17 @@ static void cstate_end(double time, struct cpuidle_cstates *cstates) cstate->max_time = MAX(cstate->max_time, data->duration); cstate->avg_time = AVG(cstate->avg_time, data->duration, cstate->nrdata + 1); + + if (update_duration_opp(cstate, data->speed, data->duration) < 0) + return error(__func__); + cstate->duration += data->duration; cstate->nrdata++; skip_entry: /* CPU is no longer idle */ cstates->current_cstate = -1; + return 0; } int record_cstate_event(struct cpuidle_cstates *cstates, @@ -810,13 +844,35 @@ int record_cstate_event(struct cpuidle_cstates *cstates, return 0; if (cstates->current_cstate != -1) - cstate_end(time, cstates); + ret = cstate_end(time, cstates); if (state != -1) ret = cstate_begin(time, state, cstates); return ret; } +static void core_update_current_speed(struct cpu_core *core) +{ + struct cpu_cpu *cpu; + unsigned int max_speed = 0; + + core_for_each_cpu(cpu, core) + max_speed = MAX(max_speed, cpu->cstates->current_speed); + + core->cstates->current_speed = max_speed; +} + +static void cluster_update_current_speed(struct cpu_physical *clust) +{ + struct cpu_cpu *cpu; + unsigned int max_speed = 0; + + cluster_for_each_cpu(cpu, clust) + max_speed = MAX(max_speed, cpu->cstates->current_speed); + + clust->cstates->current_speed = max_speed; +} + int store_data(double time, int state, int cpu, struct cpuidle_datas *datas) { @@ -843,11 +899,13 @@ int store_data(double time, int state, int cpu, /* Update core and cluster */ aff_core = cpu_to_core(cpu, datas->topo); state = core_get_least_cstate(aff_core); + core_update_current_speed(aff_core); if (record_cstate_event(aff_core->cstates, time, state) == -1) return -1; aff_cluster = cpu_to_cluster(cpu, datas->topo); state = cluster_get_least_cstate(aff_cluster); + cluster_update_current_speed(aff_cluster); if (record_cstate_event(aff_cluster->cstates, time,state) == -1) return -1; diff --git a/idlestat.h b/idlestat.h index e030f6a..2eef6c9 100644 --- a/idlestat.h +++ b/idlestat.h @@ -48,11 +48,17 @@ "/sys/devices/system/cpu/cpu%d/cpufreq/cpuinfo_cur_freq" struct cpuidle_data { + unsigned int speed; double begin; double end; double duration; }; +struct duration_opp { + unsigned int speed; + double duration; +}; + struct cpuidle_cstate { char *name; struct cpuidle_data *data; @@ -62,7 +68,9 @@ struct cpuidle_cstate { double avg_time; double max_time; double min_time; + struct duration_opp *duration_opps; double duration; + int nropp; int target_residency; /* -1 if not available */ }; @@ -83,6 +91,7 @@ struct cpuidle_cstates { struct cpuidle_cstate cstate[MAXCSTATE]; struct wakeup_info wakeinfo; int current_cstate; + unsigned int current_speed; int cstate_max; struct wakeup_irq *wakeirq; enum {as_expected, too_long, too_short} actual_residency; @@ -157,6 +166,7 @@ struct pstate_energy_info { struct cstate_energy_info { char cstate_name[NAMELEN]; + unsigned int speed; unsigned int cluster_idle_power; unsigned int core_idle_power; double cluster_duration; diff --git a/trace_ops.h b/trace_ops.h index 6d097ff..41258b6 100644 --- a/trace_ops.h +++ b/trace_ops.h @@ -46,5 +46,4 @@ extern void load_text_data_lines(FILE *f, char *buffer, struct cpuidle_datas *da * tracetype_name ## _trace_ptr = &tracetype_name##_trace_ops extern const struct trace_ops *trace_ops_head; - #endif diff --git a/tracefile_idlestat.c b/tracefile_idlestat.c index 3430693..79f8578 100644 --- a/tracefile_idlestat.c +++ b/tracefile_idlestat.c @@ -108,6 +108,8 @@ static struct cpuidle_cstates *load_and_build_cstate_info(FILE* f, char *buffer, c->max_time = 0.; c->min_time = DBL_MAX; c->duration = 0.; + c->duration_opps = NULL; + c->nropp = 0; c->target_residency = residency; } fgets(buffer, BUFSIZE, f); @@ -146,6 +148,8 @@ int load_text_data_line(char *buffer, struct cpuidle_datas *datas, char *format, "be wrong.\n"); return -1; } + + datas->cstates[cpu].current_speed = freq; return cpu_change_pstate(datas, cpu, freq, time); } -- 1.9.1

10 years, 1 month

[PATCH rt-app v2 00/20] fixes of rt-app

by pi-cheng.chen

This patchset includes fixes/changes that have been done. These patches mainly fixed some typos or moves code but should not change the behavior of rt-app. changes from v1: - split style fix patches in smaller ones - clarify some changelogs - remove patches not suitable for this patchset Those patches are also found in the branch: https://git.linaro.org/people/picheng.chen/rt-app.git fixes_v2 Chris Muller (1): Update thread name Vincent Guittot (10): fix deadline print format consolidate trace and debug point update .gitignore fix inconsistency in delay unit fix cpu affinity string info deadline: set deadline field to deadline parameter reorder the start sequence of threads remove the yaml example as we don't support it remove useless json_object_put rt-app: remove use of deprecated json interface pi-cheng.chen (9): remove useless space and add blank lines to make the code more readable fix some comments align parameters and indents some style fixes Rename variable to improve readability do not sleep if we have run longer than expected fix debugfs path remove unused function add missed code snip to get deadline parameter .gitignore | 8 ++ doc/taskset.yml | 53 ------------- src/rt-app.c | 195 +++++++++++++++++++++++++++------------------- src/rt-app_args.c | 57 +++++++------- src/rt-app_parse_config.c | 49 +++++++----- src/rt-app_utils.c | 3 + 6 files changed, 185 insertions(+), 180 deletions(-) delete mode 100644 doc/taskset.yml -- 1.9.1

10 years, 1 month

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

Sched-tools June 2015