1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2015 Davidlohr Bueso. 4 */ 5 6 /* For the CLR_() macros */ 7 #include <string.h> 8 #include <pthread.h> 9 10 #include <signal.h> 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> 16 #include <errno.h> 17 #include "bench.h" 18 #include "futex.h" 19 #include "cpumap.h" 20 21 #include <err.h> 22 #include <stdlib.h> 23 #include <sys/time.h> 24 25 struct worker { 26 int tid; 27 u_int32_t *futex; 28 pthread_t thread; 29 unsigned long ops; 30 }; 31 32 static u_int32_t global_futex = 0; 33 static struct worker *worker; 34 static unsigned int nsecs = 10; 35 static bool silent = false, multi = false; 36 static bool done = false, fshared = false; 37 static unsigned int nthreads = 0; 38 static int futex_flag = 0; 39 struct timeval start, end, runtime; 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; 44 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"), 51 OPT_END() 52 }; 53 54 static const char * const bench_futex_lock_pi_usage[] = { 55 "perf bench futex lock-pi <options>", 56 NULL 57 }; 58 59 static void print_summary(void) 60 { 61 unsigned long avg = avg_stats(&throughput_stats); 62 double stddev = stddev_stats(&throughput_stats); 63 64 printf("%sAveraged %ld operations/sec (+- %.2f%%), total secs = %d\n", 65 !silent ? "\n" : "", avg, rel_stddev_stats(stddev, avg), 66 (int) runtime.tv_sec); 67 } 68 69 static void toggle_done(int sig __maybe_unused, 70 siginfo_t *info __maybe_unused, 71 void *uc __maybe_unused) 72 { 73 /* inform all threads that we're done for the day */ 74 done = true; 75 gettimeofday(&end, NULL); 76 timersub(&end, &start, &runtime); 77 } 78 79 static void *workerfn(void *arg) 80 { 81 struct worker *w = (struct worker *) arg; 82 unsigned long ops = w->ops; 83 84 pthread_mutex_lock(&thread_lock); 85 threads_starting--; 86 if (!threads_starting) 87 pthread_cond_signal(&thread_parent); 88 pthread_cond_wait(&thread_worker, &thread_lock); 89 pthread_mutex_unlock(&thread_lock); 90 91 do { 92 int ret; 93 again: 94 ret = futex_lock_pi(w->futex, NULL, futex_flag); 95 96 if (ret) { /* handle lock acquisition */ 97 if (!silent) 98 warn("thread %d: Could not lock pi-lock for %p (%d)", 99 w->tid, w->futex, ret); 100 if (done) 101 break; 102 103 goto again; 104 } 105 106 usleep(1); 107 ret = futex_unlock_pi(w->futex, futex_flag); 108 if (ret && !silent) 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 */ 112 } while (!done); 113 114 w->ops = ops; 115 return NULL; 116 } 117 118 static void create_threads(struct worker *w, pthread_attr_t thread_attr, 119 struct cpu_map *cpu) 120 { 121 cpu_set_t cpuset; 122 unsigned int i; 123 124 threads_starting = nthreads; 125 126 for (i = 0; i < nthreads; i++) { 127 worker[i].tid = i; 128 129 if (multi) { 130 worker[i].futex = calloc(1, sizeof(u_int32_t)); 131 if (!worker[i].futex) 132 err(EXIT_FAILURE, "calloc"); 133 } else 134 worker[i].futex = &global_futex; 135 136 CPU_ZERO(&cpuset); 137 CPU_SET(cpu->map[i % cpu->nr], &cpuset); 138 139 if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpuset)) 140 err(EXIT_FAILURE, "pthread_attr_setaffinity_np"); 141 142 if (pthread_create(&w[i].thread, &thread_attr, workerfn, &worker[i])) 143 err(EXIT_FAILURE, "pthread_create"); 144 } 145 } 146 147 int bench_futex_lock_pi(int argc, const char **argv) 148 { 149 int ret = 0; 150 unsigned int i; 151 struct sigaction act; 152 pthread_attr_t thread_attr; 153 struct cpu_map *cpu; 154 155 argc = parse_options(argc, argv, options, bench_futex_lock_pi_usage, 0); 156 if (argc) 157 goto err; 158 159 cpu = cpu_map__new(NULL); 160 if (!cpu) 161 err(EXIT_FAILURE, "calloc"); 162 163 sigfillset(&act.sa_mask); 164 act.sa_sigaction = toggle_done; 165 sigaction(SIGINT, &act, NULL); 166 167 if (!nthreads) 168 nthreads = cpu->nr; 169 170 worker = calloc(nthreads, sizeof(*worker)); 171 if (!worker) 172 err(EXIT_FAILURE, "calloc"); 173 174 if (!fshared) 175 futex_flag = FUTEX_PRIVATE_FLAG; 176 177 printf("Run summary [PID %d]: %d threads doing pi lock/unlock pairing for %d secs.\n\n", 178 getpid(), nthreads, nsecs); 179 180 init_stats(&throughput_stats); 181 pthread_mutex_init(&thread_lock, NULL); 182 pthread_cond_init(&thread_parent, NULL); 183 pthread_cond_init(&thread_worker, NULL); 184 185 threads_starting = nthreads; 186 pthread_attr_init(&thread_attr); 187 gettimeofday(&start, NULL); 188 189 create_threads(worker, thread_attr, cpu); 190 pthread_attr_destroy(&thread_attr); 191 192 pthread_mutex_lock(&thread_lock); 193 while (threads_starting) 194 pthread_cond_wait(&thread_parent, &thread_lock); 195 pthread_cond_broadcast(&thread_worker); 196 pthread_mutex_unlock(&thread_lock); 197 198 sleep(nsecs); 199 toggle_done(0, NULL, NULL); 200 201 for (i = 0; i < nthreads; i++) { 202 ret = pthread_join(worker[i].thread, NULL); 203 if (ret) 204 err(EXIT_FAILURE, "pthread_join"); 205 } 206 207 /* cleanup & report results */ 208 pthread_cond_destroy(&thread_parent); 209 pthread_cond_destroy(&thread_worker); 210 pthread_mutex_destroy(&thread_lock); 211 212 for (i = 0; i < nthreads; i++) { 213 unsigned long t = worker[i].ops/runtime.tv_sec; 214 215 update_stats(&throughput_stats, t); 216 if (!silent) 217 printf("[thread %3d] futex: %p [ %ld ops/sec ]\n", 218 worker[i].tid, worker[i].futex, t); 219 220 if (multi) 221 zfree(&worker[i].futex); 222 } 223 224 print_summary(); 225 226 free(worker); 227 return ret; 228 err: 229 usage_with_options(bench_futex_lock_pi_usage, options); 230 exit(EXIT_FAILURE); 231 } 232