1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2015 Davidlohr Bueso.
6 /* For the CLR_() macros */
11 #include "../util/stat.h"
12 #include <subcmd/parse-options.h>
13 #include <linux/compiler.h>
14 #include <linux/kernel.h>
15 #include <linux/zalloc.h>
17 #include <internal/cpumap.h>
18 #include <perf/cpumap.h>
33 static u_int32_t global_futex = 0;
34 static struct worker *worker;
35 static unsigned int nsecs = 10;
36 static bool silent = false, multi = false;
37 static bool done = false, fshared = false;
38 static unsigned int nthreads = 0;
39 static int futex_flag = 0;
40 static pthread_mutex_t thread_lock;
41 static unsigned int threads_starting;
42 static struct stats throughput_stats;
43 static pthread_cond_t thread_parent, thread_worker;
45 static const struct option options[] = {
46 OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
47 OPT_UINTEGER('r', "runtime", &nsecs, "Specify runtime (in seconds)"),
48 OPT_BOOLEAN( 'M', "multi", &multi, "Use multiple futexes"),
49 OPT_BOOLEAN( 's', "silent", &silent, "Silent mode: do not display data/details"),
50 OPT_BOOLEAN( 'S', "shared", &fshared, "Use shared futexes instead of private ones"),
54 static const char * const bench_futex_lock_pi_usage[] = {
55 "perf bench futex lock-pi <options>",
59 static void print_summary(void)
61 unsigned long avg = avg_stats(&throughput_stats);
62 double stddev = stddev_stats(&throughput_stats);
64 printf("%sAveraged %ld operations/sec (+- %.2f%%), total secs = %d\n",
65 !silent ? "\n" : "", avg, rel_stddev_stats(stddev, avg),
66 (int)bench__runtime.tv_sec);
69 static void toggle_done(int sig __maybe_unused,
70 siginfo_t *info __maybe_unused,
71 void *uc __maybe_unused)
73 /* inform all threads that we're done for the day */
75 gettimeofday(&bench__end, NULL);
76 timersub(&bench__end, &bench__start, &bench__runtime);
79 static void *workerfn(void *arg)
81 struct worker *w = (struct worker *) arg;
82 unsigned long ops = w->ops;
84 pthread_mutex_lock(&thread_lock);
86 if (!threads_starting)
87 pthread_cond_signal(&thread_parent);
88 pthread_cond_wait(&thread_worker, &thread_lock);
89 pthread_mutex_unlock(&thread_lock);
94 ret = futex_lock_pi(w->futex, NULL, futex_flag);
96 if (ret) { /* handle lock acquisition */
98 warn("thread %d: Could not lock pi-lock for %p (%d)",
99 w->tid, w->futex, ret);
107 ret = futex_unlock_pi(w->futex, futex_flag);
109 warn("thread %d: Could not unlock pi-lock for %p (%d)",
110 w->tid, w->futex, ret);
111 ops++; /* account for thread's share of work */
118 static void create_threads(struct worker *w, pthread_attr_t thread_attr,
119 struct perf_cpu_map *cpu)
124 threads_starting = nthreads;
126 for (i = 0; i < nthreads; i++) {
130 worker[i].futex = calloc(1, sizeof(u_int32_t));
131 if (!worker[i].futex)
132 err(EXIT_FAILURE, "calloc");
134 worker[i].futex = &global_futex;
137 CPU_SET(cpu->map[i % cpu->nr], &cpuset);
139 if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpuset))
140 err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
142 if (pthread_create(&w[i].thread, &thread_attr, workerfn, &worker[i]))
143 err(EXIT_FAILURE, "pthread_create");
147 int bench_futex_lock_pi(int argc, const char **argv)
151 struct sigaction act;
152 pthread_attr_t thread_attr;
153 struct perf_cpu_map *cpu;
155 argc = parse_options(argc, argv, options, bench_futex_lock_pi_usage, 0);
159 cpu = perf_cpu_map__new(NULL);
161 err(EXIT_FAILURE, "calloc");
163 memset(&act, 0, sizeof(act));
164 sigfillset(&act.sa_mask);
165 act.sa_sigaction = toggle_done;
166 sigaction(SIGINT, &act, NULL);
171 worker = calloc(nthreads, sizeof(*worker));
173 err(EXIT_FAILURE, "calloc");
176 futex_flag = FUTEX_PRIVATE_FLAG;
178 printf("Run summary [PID %d]: %d threads doing pi lock/unlock pairing for %d secs.\n\n",
179 getpid(), nthreads, nsecs);
181 init_stats(&throughput_stats);
182 pthread_mutex_init(&thread_lock, NULL);
183 pthread_cond_init(&thread_parent, NULL);
184 pthread_cond_init(&thread_worker, NULL);
186 threads_starting = nthreads;
187 pthread_attr_init(&thread_attr);
188 gettimeofday(&bench__start, NULL);
190 create_threads(worker, thread_attr, cpu);
191 pthread_attr_destroy(&thread_attr);
193 pthread_mutex_lock(&thread_lock);
194 while (threads_starting)
195 pthread_cond_wait(&thread_parent, &thread_lock);
196 pthread_cond_broadcast(&thread_worker);
197 pthread_mutex_unlock(&thread_lock);
200 toggle_done(0, NULL, NULL);
202 for (i = 0; i < nthreads; i++) {
203 ret = pthread_join(worker[i].thread, NULL);
205 err(EXIT_FAILURE, "pthread_join");
208 /* cleanup & report results */
209 pthread_cond_destroy(&thread_parent);
210 pthread_cond_destroy(&thread_worker);
211 pthread_mutex_destroy(&thread_lock);
213 for (i = 0; i < nthreads; i++) {
214 unsigned long t = worker[i].ops / bench__runtime.tv_sec;
216 update_stats(&throughput_stats, t);
218 printf("[thread %3d] futex: %p [ %ld ops/sec ]\n",
219 worker[i].tid, worker[i].futex, t);
222 zfree(&worker[i].futex);
230 usage_with_options(bench_futex_lock_pi_usage, options);