tools/perf/bench/futex-hash.c

   1 /*
   2  * Copyright (C) 2013  Davidlohr Bueso <davidlohr@hp.com>
   3  *
   4  * futex-hash: Stress the hell out of the Linux kernel futex uaddr hashing.
   5  *
   6  * This program is particularly useful for measuring the kernel's futex hash
   7  * table/function implementation. In order for it to make sense, use with as
   8  * many threads and futexes as possible.
   9  */
  10
  11 /* For the CLR_() macros */
  12 #include <pthread.h>
  13
  14 #include <errno.h>
  15 #include <signal.h>
  16 #include <stdlib.h>
  17 #include <linux/compiler.h>
  18 #include <linux/kernel.h>
  19 #include <sys/time.h>
  20
  21 #include "../util/stat.h"
  22 #include <subcmd/parse-options.h>
  23 #include "bench.h"
  24 #include "futex.h"
  25
  26 #include <err.h>
  27 #include <sys/time.h>
  28
  29 static unsigned int nthreads = 0;
  30 static unsigned int nsecs    = 10;
  31 /* amount of futexes per thread */
  32 static unsigned int nfutexes = 1024;
  33 static bool fshared = false, done = false, silent = false;
  34 static int futex_flag = 0;
  35
  36 struct timeval start, end, runtime;
  37 static pthread_mutex_t thread_lock;
  38 static unsigned int threads_starting;
  39 static struct stats throughput_stats;
  40 static pthread_cond_t thread_parent, thread_worker;
  41
  42 struct worker {
  43         int tid;
  44         u_int32_t *futex;
  45         pthread_t thread;
  46         unsigned long ops;
  47 };
  48
  49 static const struct option options[] = {
  50         OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
  51         OPT_UINTEGER('r', "runtime", &nsecs,    "Specify runtime (in seconds)"),
  52         OPT_UINTEGER('f', "futexes", &nfutexes, "Specify amount of futexes per threads"),
  53         OPT_BOOLEAN( 's', "silent",  &silent,   "Silent mode: do not display data/details"),
  54         OPT_BOOLEAN( 'S', "shared",  &fshared,  "Use shared futexes instead of private ones"),
  55         OPT_END()
  56 };
  57
  58 static const char * const bench_futex_hash_usage[] = {
  59         "perf bench futex hash <options>",
  60         NULL
  61 };
  62
  63 static void *workerfn(void *arg)
  64 {
  65         int ret;
  66         struct worker *w = (struct worker *) arg;
  67         unsigned int i;
  68         unsigned long ops = w->ops; /* avoid cacheline bouncing */
  69
  70         pthread_mutex_lock(&thread_lock);
  71         threads_starting--;
  72         if (!threads_starting)
  73                 pthread_cond_signal(&thread_parent);
  74         pthread_cond_wait(&thread_worker, &thread_lock);
  75         pthread_mutex_unlock(&thread_lock);
  76
  77         do {
  78                 for (i = 0; i < nfutexes; i++, ops++) {
  79                         /*
  80                          * We want the futex calls to fail in order to stress
  81                          * the hashing of uaddr and not measure other steps,
  82                          * such as internal waitqueue handling, thus enlarging
  83                          * the critical region protected by hb->lock.
  84                          */
  85                         ret = futex_wait(&w->futex[i], 1234, NULL, futex_flag);
  86                         if (!silent &&
  87                             (!ret || errno != EAGAIN || errno != EWOULDBLOCK))
  88                                 warn("Non-expected futex return call");
  89                 }
  90         }  while (!done);
  91
  92         w->ops = ops;
  93         return NULL;
  94 }
  95
  96 static void toggle_done(int sig __maybe_unused,
  97                         siginfo_t *info __maybe_unused,
  98                         void *uc __maybe_unused)
  99 {
 100         /* inform all threads that we're done for the day */
 101         done = true;
 102         gettimeofday(&end, NULL);
 103         timersub(&end, &start, &runtime);
 104 }
 105
 106 static void print_summary(void)
 107 {
 108         unsigned long avg = avg_stats(&throughput_stats);
 109         double stddev = stddev_stats(&throughput_stats);
 110
 111         printf("%sAveraged %ld operations/sec (+- %.2f%%), total secs = %d\n",
 112                !silent ? "\n" : "", avg, rel_stddev_stats(stddev, avg),
 113                (int) runtime.tv_sec);
 114 }
 115
 116 int bench_futex_hash(int argc, const char **argv,
 117                      const char *prefix __maybe_unused)
 118 {
 119         int ret = 0;
 120         cpu_set_t cpu;
 121         struct sigaction act;
 122         unsigned int i, ncpus;
 123         pthread_attr_t thread_attr;
 124         struct worker *worker = NULL;
 125
 126         argc = parse_options(argc, argv, options, bench_futex_hash_usage, 0);
 127         if (argc) {
 128                 usage_with_options(bench_futex_hash_usage, options);
 129                 exit(EXIT_FAILURE);
 130         }
 131
 132         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
 133         nsecs = futexbench_sanitize_numeric(nsecs);
 134         nfutexes = futexbench_sanitize_numeric(nfutexes);
 135
 136         sigfillset(&act.sa_mask);
 137         act.sa_sigaction = toggle_done;
 138         sigaction(SIGINT, &act, NULL);
 139
 140         if (!nthreads) /* default to the number of CPUs */
 141                 nthreads = ncpus;
 142         else
 143                 nthreads = futexbench_sanitize_numeric(nthreads);
 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(&cpu);
 170                 CPU_SET(i % ncpus, &cpu);
 171
 172                 ret = pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpu);
 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         return ret;
 224 errmem:
 225         err(EXIT_FAILURE, "calloc");
 226 }