/* cpufreq-bench CPUFreq microbenchmark * * Copyright (C) 2008 Christian Kornacker * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ #include #include #include #include "config.h" #include "system.h" #include "benchmark.h" /* Print out progress if we log into a file */ #define show_progress(total_time, progress_time) \ if (config->output != stdout) { \ fprintf(stdout, "Progress: %02lu %%\r", \ (progress_time * 100) / total_time); \ fflush(stdout); \ } /** * compute how many rounds of calculation we should do * to get the given load time * * @param load aimed load time in µs * * @retval rounds of calculation **/ unsigned int calculate_timespace(long load, struct config *config) { int i; long long now, then; unsigned int estimated = GAUGECOUNT; unsigned int rounds = 0; unsigned int timed = 0; if (config->verbose) printf("calibrating load of %lius, please wait...\n", load); /* get the initial calculation time for a specific number of rounds */ now = get_time(); ROUNDS(estimated); then = get_time(); timed = (unsigned int)(then - now); /* approximation of the wanted load time by comparing with the * initial calculation time */ for (i = 0; i < 4; i++) { rounds = (unsigned int)(load * estimated / timed); dprintf("calibrating with %u rounds\n", rounds); now = get_time(); ROUNDS(rounds); then = get_time(); timed = (unsigned int)(then - now); estimated = rounds; } if (config->verbose) printf("calibration done\n"); return estimated; } /** * benchmark * generates a specific sleep an load time with the performance * governor and compares the used time for same calculations done * with the configured powersave governor * * @param config config values for the benchmark * **/ void start_benchmark(struct config *config) { unsigned int _round, cycle; long long now, then; long sleep_time = 0, load_time = 0; long performance_time = 0, powersave_time = 0; unsigned int calculations; unsigned long total_time = 0, progress_time = 0; sleep_time = config->sleep; load_time = config->load; /* For the progress bar */ for (_round = 1; _round <= config->rounds; _round++) total_time += _round * (config->sleep + config->load); total_time *= 2; /* powersave and performance cycles */ for (_round = 0; _round < config->rounds; _round++) { performance_time = 0LL; powersave_time = 0LL; show_progress(total_time, progress_time); /* set the cpufreq governor to "performance" which disables * P-State switching. */ if (set_cpufreq_governor("performance", config->cpu) != 0) return; /* calibrate the calculation time. the resulting calculation * _rounds should produce a load which matches the configured * load time */ calculations = calculate_timespace(load_time, config); if (config->verbose) printf("_round %i: doing %u cycles with %u calculations" " for %lius\n", _round + 1, config->cycles, calculations, load_time); fprintf(config->output, "%u %li %li ", _round, load_time, sleep_time); if (config->verbose) printf("average: %lius, rps:%li\n", load_time / calculations, 1000000 * calculations / load_time); /* do some sleep/load cycles with the performance governor */ for (cycle = 0; cycle < config->cycles; cycle++) { now = get_time(); usleep(sleep_time); ROUNDS(calculations); then = get_time(); performance_time += then - now - sleep_time; if (config->verbose) printf("performance cycle took %lius, " "sleep: %lius, " "load: %lius, rounds: %u\n", (long)(then - now), sleep_time, load_time, calculations); } fprintf(config->output, "%li ", performance_time / config->cycles); progress_time += sleep_time + load_time; show_progress(total_time, progress_time); /* set the powersave governor which activates P-State switching * again */ if (set_cpufreq_governor(config->governor, config->cpu) != 0) return; /* again, do some sleep/load cycles with the * powersave governor */ for (cycle = 0; cycle < config->cycles; cycle++) { now = get_time(); usleep(sleep_time); ROUNDS(calculations); then = get_time(); powersave_time += then - now - sleep_time; if (config->verbose) printf("powersave cycle took %lius, " "sleep: %lius, " "load: %lius, rounds: %u\n", (long)(then - now), sleep_time, load_time, calculations); } progress_time += sleep_time + load_time; /* compare the average sleep/load cycles */ fprintf(config->output, "%li ", powersave_time / config->cycles); fprintf(config->output, "%.3f\n", performance_time * 100.0 / powersave_time); fflush(config->output); if (config->verbose) printf("performance is at %.2f%%\n", performance_time * 100.0 / powersave_time); sleep_time += config->sleep_step; load_time += config->load_step; } } pan='2' class='oid'>55703c2ea8584e303e342090614e0aab3509ab21 /tools/power/cpupower/utils/cpupower-info.c parent0b3589be9b98994ce3d5aeca52445d1f5627c4ba (diff)
perf/x86/intel/rapl: Make package handling more robust
The package management code in RAPL relies on package mapping being available before a CPU is started. This changed with: 9d85eb9119f4 ("x86/smpboot: Make logical package management more robust") because the ACPI/BIOS information turned out to be unreliable, but that left RAPL in broken state. This was not noticed because on a regular boot all CPUs are online before RAPL is initialized. A possible fix would be to reintroduce the mess which allocates a package data structure in CPU prepare and when it turns out to already exist in starting throw it away later in the CPU online callback. But that's a horrible hack and not required at all because RAPL becomes functional for perf only in the CPU online callback. That's correct because user space is not yet informed about the CPU being onlined, so nothing caan rely on RAPL being available on that particular CPU. Move the allocation to the CPU online callback and simplify the hotplug handling. At this point the package mapping is established and correct. This also adds a missing check for available package data in the event_init() function. Reported-by: Yasuaki Ishimatsu <yasu.isimatu@gmail.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Siewior <bigeasy@linutronix.de> Cc: Stephane Eranian <eranian@google.com> Cc: Vince Weaver <vincent.weaver@maine.edu> Fixes: 9d85eb9119f4 ("x86/smpboot: Make logical package management more robust") Link: http://lkml.kernel.org/r/20170131230141.212593966@linutronix.de Signed-off-by: Ingo Molnar <mingo@kernel.org>
Diffstat (limited to 'tools/power/cpupower/utils/cpupower-info.c')