tools/perf/bench/futex-hash.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2013  Davidlohr Bueso <davidlohr@hp.com>
   4  *
   5  * futex-hash: Stress the hell out of the Linux kernel futex uaddr hashing.
   6  *
   7  * This program is particularly useful for measuring the kernel's futex hash
   8  * table/function implementation. In order for it to make sense, use with as
   9  * many threads and futexes as possible.
  10  */
  11
  12 /* For the CLR_() macros */
  13 #include <string.h>
  14 #include <pthread.h>
  15
  16 #include <errno.h>
  17 #include <signal.h>
  18 #include <stdlib.h>
  19 #include <linux/compiler.h>
  20 #include <linux/kernel.h>
  21 #include <sys/time.h>
  22
  23 #include "../util/stat.h"
  24 #include <subcmd/parse-options.h>
  25 #include "bench.h"
  26 #include "futex.h"
  27 #include "cpumap.h"
  28
  29 #include <err.h>
  30
  31 static unsigned int nthreads = 0;
  32 static unsigned int nsecs    = 10;
  33 /* amount of futexes per thread */
  34 static unsigned int nfutexes = 1024;
  35 static bool fshared = false, done = false, silent = false;
  36 static int futex_flag = 0;
  37
  38 struct timeval start, end, runtime;
  39 static pthread_mutex_t thread_lock;
  40 static unsigned int threads_starting;
  41 static struct stats throughput_stats;
  42 static pthread_cond_t thread_parent, thread_worker;
  43
  44 struct worker {
  45         int tid;
  46         u_int32_t *futex;
  47         pthread_t thread;
  48         unsigned long ops;
  49 };
  50
  51 static const struct option options[] = {
  52         OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
  53         OPT_UINTEGER('r', "runtime", &nsecs,    "Specify runtime (in seconds)"),
  54         OPT_UINTEGER('f', "futexes", &nfutexes, "Specify amount of futexes per threads"),
  55         OPT_BOOLEAN( 's', "silent",  &silent,   "Silent mode: do not display data/details"),
  56         OPT_BOOLEAN( 'S', "shared",  &fshared,  "Use shared futexes instead of private ones"),
  57         OPT_END()
  58 };
  59
  60 static const char * const bench_futex_hash_usage[] = {
  61         "perf bench futex hash <options>",
  62         NULL
  63 };
  64
  65 static void *workerfn(void *arg)
  66 {
  67         int ret;
  68         struct worker *w = (struct worker *) arg;
  69         unsigned int i;
  70         unsigned long ops = w->ops; /* avoid cacheline bouncing */
  71
  72         pthread_mutex_lock(&thread_lock);
  73         threads_starting--;
  74         if (!threads_starting)
  75                 pthread_cond_signal(&thread_parent);
  76         pthread_cond_wait(&thread_worker, &thread_lock);
  77         pthread_mutex_unlock(&thread_lock);
  78
  79         do {
  80                 for (i = 0; i < nfutexes; i++, ops++) {
  81                         /*
  82                          * We want the futex calls to fail in order to stress
  83                          * the hashing of uaddr and not measure other steps,
  84                          * such as internal waitqueue handling, thus enlarging
  85                          * the critical region protected by hb->lock.
  86                          */
  87                         ret = futex_wait(&w->futex[i], 1234, NULL, futex_flag);
  88                         if (!silent &&
  89                             (!ret || errno != EAGAIN || errno != EWOULDBLOCK))
  90                                 warn("Non-expected futex return call");
  91                 }
  92         }  while (!done);
  93
  94         w->ops = ops;
  95         return NULL;
  96 }
  97
  98 static void toggle_done(int sig __maybe_unused,
  99                         siginfo_t *info __maybe_unused,
 100                         void *uc __maybe_unused)
 101 {
 102         /* inform all threads that we're done for the day */
 103         done = true;
 104         gettimeofday(&end, NULL);
 105         timersub(&end, &start, &runtime);
 106 }
 107
 108 static void print_summary(void)
 109 {
 110         unsigned long avg = avg_stats(&throughput_stats);
 111         double stddev = stddev_stats(&throughput_stats);
 112
 113         printf("%sAveraged %ld operations/sec (+- %.2f%%), total secs = %d\n",
 114                !silent ? "\n" : "", avg, rel_stddev_stats(stddev, avg),
 115                (int) runtime.tv_sec);
 116 }
 117
 118 int bench_futex_hash(int argc, const char **argv)
 119 {
 120         int ret = 0;
 121         cpu_set_t cpuset;
 122         struct sigaction act;
 123         unsigned int i;
 124         pthread_attr_t thread_attr;
 125         struct worker *worker = NULL;
 126         struct cpu_map *cpu;
 127
 128         argc = parse_options(argc, argv, options, bench_futex_hash_usage, 0);
 129         if (argc) {
 130                 usage_with_options(bench_futex_hash_usage, options);
 131                 exit(EXIT_FAILURE);
 132         }
 133
 134         cpu = cpu_map__new(NULL);
 135         if (!cpu)
 136                 goto errmem;
 137
 138         sigfillset(&act.sa_mask);
 139         act.sa_sigaction = toggle_done;
 140         sigaction(SIGINT, &act, NULL);
 141
 142         if (!nthreads) /* default to the number of CPUs */
 143                 nthreads = cpu->nr;
 144
 145         worker = calloc(nthreads, sizeof(*worker));
 146         if (!worker)
 147                 goto errmem;
 148
 149         if (!fshared)
 150                 futex_flag = FUTEX_PRIVATE_FLAG;
 151
 152         printf("Run summary [PID %d]: %d threads, each operating on %d [%s] futexes for %d secs.\n\n",
 153                getpid(), nthreads, nfutexes, fshared ? "shared":"private", nsecs);
 154
 155         init_stats(&throughput_stats);
 156         pthread_mutex_init(&thread_lock, NULL);
 157         pthread_cond_init(&thread_parent, NULL);
 158         pthread_cond_init(&thread_worker, NULL);
 159
 160         threads_starting = nthreads;
 161         pthread_attr_init(&thread_attr);
 162         gettimeofday(&start, NULL);
 163         for (i = 0; i < nthreads; i++) {
 164                 worker[i].tid = i;
 165                 worker[i].futex = calloc(nfutexes, sizeof(*worker[i].futex));
 166                 if (!worker[i].futex)
 167                         goto errmem;
 168
 169                 CPU_ZERO(&cpuset);
 170                 CPU_SET(cpu->map[i % cpu->nr], &cpuset);
 171
 172                 ret = pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpuset);
 173                 if (ret)
 174                         err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
 175
 176                 ret = pthread_create(&worker[i].thread, &thread_attr, workerfn,
 177                                      (void *)(struct worker *) &worker[i]);
 178                 if (ret)
 179                         err(EXIT_FAILURE, "pthread_create");
 180
 181         }
 182         pthread_attr_destroy(&thread_attr);
 183
 184         pthread_mutex_lock(&thread_lock);
 185         while (threads_starting)
 186                 pthread_cond_wait(&thread_parent, &thread_lock);
 187         pthread_cond_broadcast(&thread_worker);
 188         pthread_mutex_unlock(&thread_lock);
 189
 190         sleep(nsecs);
 191         toggle_done(0, NULL, NULL);
 192
 193         for (i = 0; i < nthreads; i++) {
 194                 ret = pthread_join(worker[i].thread, NULL);
 195                 if (ret)
 196                         err(EXIT_FAILURE, "pthread_join");
 197         }
 198
 199         /* cleanup & report results */
 200         pthread_cond_destroy(&thread_parent);
 201         pthread_cond_destroy(&thread_worker);
 202         pthread_mutex_destroy(&thread_lock);
 203
 204         for (i = 0; i < nthreads; i++) {
 205                 unsigned long t = worker[i].ops/runtime.tv_sec;
 206                 update_stats(&throughput_stats, t);
 207                 if (!silent) {
 208                         if (nfutexes == 1)
 209                                 printf("[thread %2d] futex: %p [ %ld ops/sec ]\n",
 210                                        worker[i].tid, &worker[i].futex[0], t);
 211                         else
 212                                 printf("[thread %2d] futexes: %p ... %p [ %ld ops/sec ]\n",
 213                                        worker[i].tid, &worker[i].futex[0],
 214                                        &worker[i].futex[nfutexes-1], t);
 215                 }
 216
 217                 free(worker[i].futex);
 218         }
 219
 220         print_summary();
 221
 222         free(worker);
 223         free(cpu);
 224         return ret;
 225 errmem:
 226         err(EXIT_FAILURE, "calloc");
 227 }