diff options
Diffstat (limited to 'tools/perf/bench')
| -rw-r--r-- | tools/perf/bench/Build | 3 | ||||
| -rw-r--r-- | tools/perf/bench/bench.h | 3 | ||||
| -rw-r--r-- | tools/perf/bench/epoll-ctl.c | 4 | ||||
| -rw-r--r-- | tools/perf/bench/epoll-wait.c | 7 | ||||
| -rw-r--r-- | tools/perf/bench/futex-hash.c | 3 | ||||
| -rw-r--r-- | tools/perf/bench/futex-lock-pi.c | 3 | ||||
| -rw-r--r-- | tools/perf/bench/kallsyms-parse.c | 75 | ||||
| -rw-r--r-- | tools/perf/bench/sched-messaging.c | 2 | ||||
| -rw-r--r-- | tools/perf/bench/synthesize.c | 262 |
9 files changed, 352 insertions, 10 deletions
diff --git a/tools/perf/bench/Build b/tools/perf/bench/Build index e4e321b6f883..768e408757a0 100644 --- a/tools/perf/bench/Build +++ b/tools/perf/bench/Build @@ -6,9 +6,10 @@ perf-y += futex-wake.o perf-y += futex-wake-parallel.o perf-y += futex-requeue.o perf-y += futex-lock-pi.o - perf-y += epoll-wait.o perf-y += epoll-ctl.o +perf-y += synthesize.o +perf-y += kallsyms-parse.o perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-lib.o perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-asm.o diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h index 4aa6de1aa67d..61cae4966cae 100644 --- a/tools/perf/bench/bench.h +++ b/tools/perf/bench/bench.h @@ -41,9 +41,10 @@ int bench_futex_wake_parallel(int argc, const char **argv); int bench_futex_requeue(int argc, const char **argv); /* pi futexes */ int bench_futex_lock_pi(int argc, const char **argv); - int bench_epoll_wait(int argc, const char **argv); int bench_epoll_ctl(int argc, const char **argv); +int bench_synthesize(int argc, const char **argv); +int bench_kallsyms_parse(int argc, const char **argv); #define BENCH_FORMAT_DEFAULT_STR "default" #define BENCH_FORMAT_DEFAULT 0 diff --git a/tools/perf/bench/epoll-ctl.c b/tools/perf/bench/epoll-ctl.c index cadc18d42aa4..ca2d591aad8a 100644 --- a/tools/perf/bench/epoll-ctl.c +++ b/tools/perf/bench/epoll-ctl.c @@ -5,7 +5,7 @@ * Benchmark the various operations allowed for epoll_ctl(2). * The idea is to concurrently stress a single epoll instance */ -#ifdef HAVE_EVENTFD +#ifdef HAVE_EVENTFD_SUPPORT /* For the CLR_() macros */ #include <string.h> #include <pthread.h> @@ -412,4 +412,4 @@ int bench_epoll_ctl(int argc, const char **argv) errmem: err(EXIT_FAILURE, "calloc"); } -#endif // HAVE_EVENTFD +#endif // HAVE_EVENTFD_SUPPORT diff --git a/tools/perf/bench/epoll-wait.c b/tools/perf/bench/epoll-wait.c index f938c585d512..75dca9773186 100644 --- a/tools/perf/bench/epoll-wait.c +++ b/tools/perf/bench/epoll-wait.c @@ -1,5 +1,5 @@ // SPDX-License-Identifier: GPL-2.0 -#ifdef HAVE_EVENTFD +#ifdef HAVE_EVENTFD_SUPPORT /* * Copyright (C) 2018 Davidlohr Bueso. * @@ -519,7 +519,8 @@ int bench_epoll_wait(int argc, const char **argv) qsort(worker, nthreads, sizeof(struct worker), cmpworker); for (i = 0; i < nthreads; i++) { - unsigned long t = worker[i].ops / bench__runtime.tv_sec; + unsigned long t = bench__runtime.tv_sec > 0 ? + worker[i].ops / bench__runtime.tv_sec : 0; update_stats(&throughput_stats, t); @@ -539,4 +540,4 @@ int bench_epoll_wait(int argc, const char **argv) errmem: err(EXIT_FAILURE, "calloc"); } -#endif // HAVE_EVENTFD +#endif // HAVE_EVENTFD_SUPPORT diff --git a/tools/perf/bench/futex-hash.c b/tools/perf/bench/futex-hash.c index 65eebe06c04d..915bf3da7ce2 100644 --- a/tools/perf/bench/futex-hash.c +++ b/tools/perf/bench/futex-hash.c @@ -205,7 +205,8 @@ int bench_futex_hash(int argc, const char **argv) pthread_mutex_destroy(&thread_lock); for (i = 0; i < nthreads; i++) { - unsigned long t = worker[i].ops / bench__runtime.tv_sec; + unsigned long t = bench__runtime.tv_sec > 0 ? + worker[i].ops / bench__runtime.tv_sec : 0; update_stats(&throughput_stats, t); if (!silent) { if (nfutexes == 1) diff --git a/tools/perf/bench/futex-lock-pi.c b/tools/perf/bench/futex-lock-pi.c index 89fd8f325f38..bb25d8beb3b8 100644 --- a/tools/perf/bench/futex-lock-pi.c +++ b/tools/perf/bench/futex-lock-pi.c @@ -211,7 +211,8 @@ int bench_futex_lock_pi(int argc, const char **argv) pthread_mutex_destroy(&thread_lock); for (i = 0; i < nthreads; i++) { - unsigned long t = worker[i].ops / bench__runtime.tv_sec; + unsigned long t = bench__runtime.tv_sec > 0 ? + worker[i].ops / bench__runtime.tv_sec : 0; update_stats(&throughput_stats, t); if (!silent) diff --git a/tools/perf/bench/kallsyms-parse.c b/tools/perf/bench/kallsyms-parse.c new file mode 100644 index 000000000000..2b0d0f980ae9 --- /dev/null +++ b/tools/perf/bench/kallsyms-parse.c @@ -0,0 +1,75 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Benchmark of /proc/kallsyms parsing. + * + * Copyright 2020 Google LLC. + */ +#include <stdlib.h> +#include "bench.h" +#include "../util/stat.h" +#include <linux/time64.h> +#include <subcmd/parse-options.h> +#include <symbol/kallsyms.h> + +static unsigned int iterations = 100; + +static const struct option options[] = { + OPT_UINTEGER('i', "iterations", &iterations, + "Number of iterations used to compute average"), + OPT_END() +}; + +static const char *const bench_usage[] = { + "perf bench internals kallsyms-parse <options>", + NULL +}; + +static int bench_process_symbol(void *arg __maybe_unused, + const char *name __maybe_unused, + char type __maybe_unused, + u64 start __maybe_unused) +{ + return 0; +} + +static int do_kallsyms_parse(void) +{ + struct timeval start, end, diff; + u64 runtime_us; + unsigned int i; + double time_average, time_stddev; + int err; + struct stats time_stats; + + init_stats(&time_stats); + + for (i = 0; i < iterations; i++) { + gettimeofday(&start, NULL); + err = kallsyms__parse("/proc/kallsyms", NULL, + bench_process_symbol); + if (err) + return err; + + gettimeofday(&end, NULL); + timersub(&end, &start, &diff); + runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec; + update_stats(&time_stats, runtime_us); + } + + time_average = avg_stats(&time_stats) / USEC_PER_MSEC; + time_stddev = stddev_stats(&time_stats) / USEC_PER_MSEC; + printf(" Average kallsyms__parse took: %.3f ms (+- %.3f ms)\n", + time_average, time_stddev); + return 0; +} + +int bench_kallsyms_parse(int argc, const char **argv) +{ + argc = parse_options(argc, argv, options, bench_usage, 0); + if (argc) { + usage_with_options(bench_usage, options); + exit(EXIT_FAILURE); + } + + return do_kallsyms_parse(); +} diff --git a/tools/perf/bench/sched-messaging.c b/tools/perf/bench/sched-messaging.c index 97e4a4fb3362..71d830d7b923 100644 --- a/tools/perf/bench/sched-messaging.c +++ b/tools/perf/bench/sched-messaging.c @@ -40,7 +40,7 @@ struct sender_context { unsigned int num_fds; int ready_out; int wakefd; - int out_fds[0]; + int out_fds[]; }; struct receiver_context { diff --git a/tools/perf/bench/synthesize.c b/tools/perf/bench/synthesize.c new file mode 100644 index 000000000000..8d624aea1c5e --- /dev/null +++ b/tools/perf/bench/synthesize.c @@ -0,0 +1,262 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Benchmark synthesis of perf events such as at the start of a 'perf + * record'. Synthesis is done on the current process and the 'dummy' event + * handlers are invoked that support dump_trace but otherwise do nothing. + * + * Copyright 2019 Google LLC. + */ +#include <stdio.h> +#include "bench.h" +#include "../util/debug.h" +#include "../util/session.h" +#include "../util/stat.h" +#include "../util/synthetic-events.h" +#include "../util/target.h" +#include "../util/thread_map.h" +#include "../util/tool.h" +#include "../util/util.h" +#include <linux/atomic.h> +#include <linux/err.h> +#include <linux/time64.h> +#include <subcmd/parse-options.h> + +static unsigned int min_threads = 1; +static unsigned int max_threads = UINT_MAX; +static unsigned int single_iterations = 10000; +static unsigned int multi_iterations = 10; +static bool run_st; +static bool run_mt; + +static const struct option options[] = { + OPT_BOOLEAN('s', "st", &run_st, "Run single threaded benchmark"), + OPT_BOOLEAN('t', "mt", &run_mt, "Run multi-threaded benchmark"), + OPT_UINTEGER('m', "min-threads", &min_threads, + "Minimum number of threads in multithreaded bench"), + OPT_UINTEGER('M', "max-threads", &max_threads, + "Maximum number of threads in multithreaded bench"), + OPT_UINTEGER('i', "single-iterations", &single_iterations, + "Number of iterations used to compute single-threaded average"), + OPT_UINTEGER('I', "multi-iterations", &multi_iterations, + "Number of iterations used to compute multi-threaded average"), + OPT_END() +}; + +static const char *const bench_usage[] = { + "perf bench internals synthesize <options>", + NULL +}; + +static atomic_t event_count; + +static int process_synthesized_event(struct perf_tool *tool __maybe_unused, + union perf_event *event __maybe_unused, + struct perf_sample *sample __maybe_unused, + struct machine *machine __maybe_unused) +{ + atomic_inc(&event_count); + return 0; +} + +static int do_run_single_threaded(struct perf_session *session, + struct perf_thread_map *threads, + struct target *target, bool data_mmap) +{ + const unsigned int nr_threads_synthesize = 1; + struct timeval start, end, diff; + u64 runtime_us; + unsigned int i; + double time_average, time_stddev, event_average, event_stddev; + int err; + struct stats time_stats, event_stats; + + init_stats(&time_stats); + init_stats(&event_stats); + + for (i = 0; i < single_iterations; i++) { + atomic_set(&event_count, 0); + gettimeofday(&start, NULL); + err = __machine__synthesize_threads(&session->machines.host, + NULL, + target, threads, + process_synthesized_event, + data_mmap, + nr_threads_synthesize); + if (err) + return err; + + gettimeofday(&end, NULL); + timersub(&end, &start, &diff); + runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec; + update_stats(&time_stats, runtime_us); + update_stats(&event_stats, atomic_read(&event_count)); + } + + time_average = avg_stats(&time_stats); + time_stddev = stddev_stats(&time_stats); + printf(" Average %ssynthesis took: %.3f usec (+- %.3f usec)\n", + data_mmap ? "data " : "", time_average, time_stddev); + + event_average = avg_stats(&event_stats); + event_stddev = stddev_stats(&event_stats); + printf(" Average num. events: %.3f (+- %.3f)\n", + event_average, event_stddev); + + printf(" Average time per event %.3f usec\n", + time_average / event_average); + return 0; +} + +static int run_single_threaded(void) +{ + struct perf_session *session; + struct target target = { + .pid = "self", + }; + struct perf_thread_map *threads; + int err; + + perf_set_singlethreaded(); + session = perf_session__new(NULL, false, NULL); + if (IS_ERR(session)) { + pr_err("Session creation failed.\n"); + return PTR_ERR(session); + } + threads = thread_map__new_by_pid(getpid()); + if (!threads) { + pr_err("Thread map creation failed.\n"); + err = -ENOMEM; + goto err_out; + } + + puts( +"Computing performance of single threaded perf event synthesis by\n" +"synthesizing events on the perf process itself:"); + + err = do_run_single_threaded(session, threads, &target, false); + if (err) + goto err_out; + + err = do_run_single_threaded(session, threads, &target, true); + +err_out: + if (threads) + perf_thread_map__put(threads); + + perf_session__delete(session); + return err; +} + +static int do_run_multi_threaded(struct target *target, + unsigned int nr_threads_synthesize) +{ + struct timeval start, end, diff; + u64 runtime_us; + unsigned int i; + double time_average, time_stddev, event_average, event_stddev; + int err; + struct stats time_stats, event_stats; + struct perf_session *session; + + init_stats(&time_stats); + init_stats(&event_stats); + for (i = 0; i < multi_iterations; i++) { + session = perf_session__new(NULL, false, NULL); + if (!session) + return -ENOMEM; + + atomic_set(&event_count, 0); + gettimeofday(&start, NULL); + err = __machine__synthesize_threads(&session->machines.host, + NULL, + target, NULL, + process_synthesized_event, + false, + nr_threads_synthesize); + if (err) { + perf_session__delete(session); + return err; + } + + gettimeofday(&end, NULL); + timersub(&end, &start, &diff); + runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec; + update_stats(&time_stats, runtime_us); + update_stats(&event_stats, atomic_read(&event_count)); + perf_session__delete(session); + } + + time_average = avg_stats(&time_stats); + time_stddev = stddev_stats(&time_stats); + printf(" Average synthesis took: %.3f usec (+- %.3f usec)\n", + time_average, time_stddev); + + event_average = avg_stats(&event_stats); + event_stddev = stddev_stats(&event_stats); + printf(" Average num. events: %.3f (+- %.3f)\n", + event_average, event_stddev); + + printf(" Average time per event %.3f usec\n", + time_average / event_average); + return 0; +} + +static int run_multi_threaded(void) +{ + struct target target = { + .cpu_list = "0" + }; + unsigned int nr_threads_synthesize; + int err; + + if (max_threads == UINT_MAX) + max_threads = sysconf(_SC_NPROCESSORS_ONLN); + + puts( +"Computing performance of multi threaded perf event synthesis by\n" +"synthesizing events on CPU 0:"); + + for (nr_threads_synthesize = min_threads; + nr_threads_synthesize <= max_threads; + nr_threads_synthesize++) { + if (nr_threads_synthesize == 1) + perf_set_singlethreaded(); + else + perf_set_multithreaded(); + + printf(" Number of synthesis threads: %u\n", + nr_threads_synthesize); + + err = do_run_multi_threaded(&target, nr_threads_synthesize); + if (err) + return err; + } + perf_set_singlethreaded(); + return 0; +} + +int bench_synthesize(int argc, const char **argv) +{ + int err = 0; + + argc = parse_options(argc, argv, options, bench_usage, 0); + if (argc) { + usage_with_options(bench_usage, options); + exit(EXIT_FAILURE); + } + + /* + * If neither single threaded or multi-threaded are specified, default + * to running just single threaded. + */ + if (!run_st && !run_mt) + run_st = true; + + if (run_st) + err = run_single_threaded(); + + if (!err && run_mt) + err = run_multi_threaded(); + + return err; +} |