tests/benchmark/test_urcu_workqueue.c

   1 /*
   2  * test_urcu_workqueue.c
   3  *
   4  * Userspace RCU library - workqueue test
   5  *
   6  * Copyright February 2010-2014 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write to the Free Software Foundation, Inc.,
  21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22  */
  23
  24 #define _GNU_SOURCE
  25 #include "config.h"
  26 #include <stdio.h>
  27 #include <pthread.h>
  28 #include <stdlib.h>
  29 #include <stdint.h>
  30 #include <stdbool.h>
  31 #include <string.h>
  32 #include <sys/types.h>
  33 #include <sys/wait.h>
  34 #include <unistd.h>
  35 #include <stdio.h>
  36 #include <assert.h>
  37 #include <errno.h>
  38
  39 #include <urcu/arch.h>
  40 #include <urcu/tls-compat.h>
  41 #include <urcu/uatomic.h>
  42 #include "cpuset.h"
  43 #include "thread-id.h"
  44
  45 /* hardcoded number of CPUs */
  46 #define NR_CPUS 16384
  47
  48 #ifndef DYNAMIC_LINK_TEST
  49 #define _LGPL_SOURCE
  50 #endif
  51 #include <urcu.h>
  52 #include <urcu/wfstack.h>
  53 #include <urcu/workqueue-fifo.h>
  54
  55 static volatile int test_go, test_stop_enqueue;
  56
  57 static unsigned long work_loops;
  58
  59 static unsigned long duration;
  60
  61 static unsigned long dispatch_delay_loops;
  62
  63 static unsigned long max_queue_len;
  64
  65 static inline void loop_sleep(unsigned long loops)
  66 {
  67         while (loops-- != 0)
  68                 caa_cpu_relax();
  69 }
  70
  71 static int verbose_mode;
  72
  73 static int test_wait_empty;
  74 static int test_enqueue_stopped;
  75
  76 #define printf_verbose(fmt, args...)            \
  77         do {                                    \
  78                 if (verbose_mode)               \
  79                         fprintf(stderr, fmt, ## args);  \
  80         } while (0)
  81
  82 static unsigned int cpu_affinities[NR_CPUS];
  83 static unsigned int next_aff = 0;
  84 static int use_affinity = 0;
  85
  86 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  87
  88 static void set_affinity(void)
  89 {
  90 #if HAVE_SCHED_SETAFFINITY
  91         cpu_set_t mask;
  92         int cpu, ret;
  93 #endif /* HAVE_SCHED_SETAFFINITY */
  94
  95         if (!use_affinity)
  96                 return;
  97
  98 #if HAVE_SCHED_SETAFFINITY
  99         ret = pthread_mutex_lock(&affinity_mutex);
 100         if (ret) {
 101                 perror("Error in pthread mutex lock");
 102                 exit(-1);
 103         }
 104         cpu = cpu_affinities[next_aff++];
 105         ret = pthread_mutex_unlock(&affinity_mutex);
 106         if (ret) {
 107                 perror("Error in pthread mutex unlock");
 108                 exit(-1);
 109         }
 110
 111         CPU_ZERO(&mask);
 112         CPU_SET(cpu, &mask);
 113 #if SCHED_SETAFFINITY_ARGS == 2
 114         sched_setaffinity(0, &mask);
 115 #else
 116         sched_setaffinity(0, sizeof(mask), &mask);
 117 #endif
 118 #endif /* HAVE_SCHED_SETAFFINITY */
 119 }
 120
 121 /*
 122  * returns 0 if test should end.
 123  */
 124 static int test_duration_enqueue(void)
 125 {
 126         return !test_stop_enqueue;
 127 }
 128
 129 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 130 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 131
 132 static unsigned int nr_dispatchers;
 133 static unsigned int nr_workers;
 134
 135 static struct urcu_workqueue workqueue;
 136
 137 struct test_work {
 138         struct urcu_work w;
 139 };
 140
 141 static void *thr_dispatcher(void *_count)
 142 {
 143         unsigned long long *count = _count;
 144         bool was_nonempty;
 145
 146         printf_verbose("thread_begin %s, tid %lu\n",
 147                         "dispatcher", urcu_get_thread_id());
 148
 149         set_affinity();
 150
 151         while (!test_go)
 152         {
 153         }
 154         cmm_smp_mb();
 155
 156         for (;;) {
 157                 struct test_work *work = malloc(sizeof(*work));
 158                 enum urcu_enqueue_ret ret;
 159
 160                 if (!work)
 161                         goto fail;
 162 retry:
 163                 printf_verbose("attempt queue work %p\n", work);
 164                 ret = urcu_queue_work(&workqueue, &work->w);
 165                 if (ret == URCU_ENQUEUE_FULL) {
 166                         printf_verbose("queue work %p (queue full)\n", work);
 167                         (void) poll(NULL, 0, 10);
 168                         goto retry;
 169                 }
 170                 printf_verbose("queue work %p (ok)\n", work);
 171                 URCU_TLS(nr_enqueues)++;
 172
 173                 if (caa_unlikely(dispatch_delay_loops))
 174                         loop_sleep(dispatch_delay_loops);
 175 fail:
 176                 if (caa_unlikely(!test_duration_enqueue()))
 177                         break;
 178         }
 179
 180         uatomic_inc(&test_enqueue_stopped);
 181         count[0] = URCU_TLS(nr_enqueues);
 182         printf_verbose("dispatcher thread_end, tid %lu, "
 183                         "enqueues %llu\n",
 184                         urcu_get_thread_id(),
 185                         URCU_TLS(nr_enqueues));
 186         return ((void*)1);
 187 }
 188
 189 static void *thr_worker(void *_count)
 190 {
 191         unsigned long long *count = _count;
 192         unsigned int counter = 0;
 193         struct urcu_worker worker;
 194
 195         printf_verbose("thread_begin %s, tid %lu\n",
 196                         "worker", urcu_get_thread_id());
 197
 198         set_affinity();
 199
 200         rcu_register_thread();
 201         urcu_worker_init(&workqueue, &worker, URCU_WORKER_STEAL);
 202         //urcu_worker_init(&workqueue, &worker, 0);
 203         urcu_worker_register(&workqueue, &worker);
 204
 205         while (!test_go)
 206         {
 207         }
 208         cmm_smp_mb();
 209
 210         for (;;) {
 211                 enum urcu_accept_ret ret;
 212
 213                 ret = urcu_accept_work(&worker);
 214                 if (ret == URCU_ACCEPT_SHUTDOWN)
 215                         break;
 216                 for (;;) {
 217                         struct urcu_work *work;
 218                         struct test_work *t;
 219
 220                         work = urcu_dequeue_work(&worker);
 221                         if (!work)
 222                                 break;
 223                         t = caa_container_of(work, struct test_work, w);
 224                         printf_verbose("dequeue work %p\n", t);
 225                         URCU_TLS(nr_dequeues)++;
 226                         if (caa_unlikely(work_loops))
 227                                 loop_sleep(work_loops);
 228                         free(t);
 229                 }
 230         }
 231 end:
 232         urcu_worker_unregister(&workqueue, &worker);
 233         rcu_unregister_thread();
 234
 235         printf_verbose("worker thread_end, tid %lu, "
 236                         "dequeues %llu\n",
 237                         urcu_get_thread_id(),
 238                         URCU_TLS(nr_dequeues));
 239         count[0] = URCU_TLS(nr_dequeues);
 240         return ((void*)2);
 241 }
 242
 243 static void show_usage(int argc, char **argv)
 244 {
 245         printf("Usage : %s nr_workers nr_dispatchers duration (s) <OPTIONS>\n",
 246                 argv[0]);
 247         printf("OPTIONS:\n");
 248         printf("        [-d delay] (dispatcher period (in loops))\n");
 249         printf("        [-c duration] (worker period (in loops))\n");
 250         printf("        [-v] (verbose output)\n");
 251         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 252         printf("        [-w] Wait for worker to empty stack\n");
 253         printf("        [-m len] (Max queue length. 0 means infinite.))\n");
 254         printf("\n");
 255 }
 256
 257 int main(int argc, char **argv)
 258 {
 259         int err;
 260         pthread_t *tid_dispatcher, *tid_worker;
 261         void *tret;
 262         unsigned long long *count_dispatcher, *count_worker;
 263         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 264         unsigned long long end_dequeues = 0;
 265         int i, a, retval = 0;
 266
 267         if (argc < 4) {
 268                 show_usage(argc, argv);
 269                 return -1;
 270         }
 271
 272         err = sscanf(argv[1], "%u", &nr_workers);
 273         if (err != 1) {
 274                 show_usage(argc, argv);
 275                 return -1;
 276         }
 277
 278         err = sscanf(argv[2], "%u", &nr_dispatchers);
 279         if (err != 1) {
 280                 show_usage(argc, argv);
 281                 return -1;
 282         }
 283
 284         err = sscanf(argv[3], "%lu", &duration);
 285         if (err != 1) {
 286                 show_usage(argc, argv);
 287                 return -1;
 288         }
 289
 290         for (i = 4; i < argc; i++) {
 291                 if (argv[i][0] != '-')
 292                         continue;
 293                 switch (argv[i][1]) {
 294                 case 'a':
 295                         if (argc < i + 2) {
 296                                 show_usage(argc, argv);
 297                                 return -1;
 298                         }
 299                         a = atoi(argv[++i]);
 300                         cpu_affinities[next_aff++] = a;
 301                         use_affinity = 1;
 302                         printf_verbose("Adding CPU %d affinity\n", a);
 303                         break;
 304                 case 'm':
 305                         if (argc < i + 2) {
 306                                 show_usage(argc, argv);
 307                                 return -1;
 308                         }
 309                         max_queue_len = atol(argv[++i]);
 310                         break;
 311                 case 'c':
 312                         if (argc < i + 2) {
 313                                 show_usage(argc, argv);
 314                                 return -1;
 315                         }
 316                         work_loops = atol(argv[++i]);
 317                         break;
 318                 case 'd':
 319                         if (argc < i + 2) {
 320                                 show_usage(argc, argv);
 321                                 return -1;
 322                         }
 323                         dispatch_delay_loops = atol(argv[++i]);
 324                         break;
 325                 case 'v':
 326                         verbose_mode = 1;
 327                         break;
 328                 case 'w':
 329                         test_wait_empty = 1;
 330                         break;
 331                 }
 332         }
 333
 334         printf_verbose("running test for %lu seconds, %u dispatchers, "
 335                        "%u workers.\n",
 336                        duration, nr_dispatchers, nr_workers);
 337         if (test_wait_empty)
 338                 printf_verbose("Wait for workers to empty workqueue.\n");
 339         printf_verbose("Work duration: %lu loops.\n", work_loops);
 340         printf_verbose("Dispatcher arrival delay: %lu loops.\n", dispatch_delay_loops);
 341         printf_verbose("thread %-6s, tid %lu\n",
 342                         "main", urcu_get_thread_id());
 343
 344         tid_dispatcher = calloc(nr_dispatchers, sizeof(*tid_dispatcher));
 345         tid_worker = calloc(nr_workers, sizeof(*tid_worker));
 346         count_dispatcher = calloc(nr_dispatchers, sizeof(*count_dispatcher));
 347         count_worker = calloc(nr_workers, sizeof(*count_worker));
 348         urcu_workqueue_init(&workqueue, max_queue_len);
 349
 350         next_aff = 0;
 351
 352         for (i = 0; i < nr_dispatchers; i++) {
 353                 err = pthread_create(&tid_dispatcher[i], NULL, thr_dispatcher,
 354                                      &count_dispatcher[i]);
 355                 if (err != 0)
 356                         exit(1);
 357         }
 358         for (i = 0; i < nr_workers; i++) {
 359                 err = pthread_create(&tid_worker[i], NULL, thr_worker,
 360                                      &count_worker[i]);
 361                 if (err != 0)
 362                         exit(1);
 363         }
 364
 365         cmm_smp_mb();
 366
 367         test_go = 1;
 368
 369         for (i = 0; i < duration; i++) {
 370                 sleep(1);
 371                 if (verbose_mode)
 372                         (void) write(1, ".", 1);
 373         }
 374
 375         test_stop_enqueue = 1;
 376         while (nr_dispatchers != uatomic_read(&test_enqueue_stopped)) {
 377                 sleep(1);
 378         }
 379
 380         if (test_wait_empty) {
 381                 while (!cds_wfcq_empty(&workqueue.head, &workqueue.tail)) {
 382                         sleep(1);
 383                 }
 384         }
 385         urcu_workqueue_shutdown(&workqueue);
 386
 387         for (i = 0; i < nr_dispatchers; i++) {
 388                 err = pthread_join(tid_dispatcher[i], &tret);
 389                 if (err != 0)
 390                         exit(1);
 391                 tot_enqueues += count_dispatcher[i];
 392         }
 393         for (i = 0; i < nr_workers; i++) {
 394                 err = pthread_join(tid_worker[i], &tret);
 395                 if (err != 0)
 396                         exit(1);
 397                 tot_dequeues += count_worker[i];
 398         }
 399
 400         printf("SUMMARY %-25s testdur %4lu nr_dispatchers %3u dispatch_delay_loops %6lu "
 401                 "work_loops %lu nr_workers %3u "
 402                 "nr_enqueues %12llu nr_dequeues %12llu max_queue_len %lu\n",
 403                 argv[0], duration, nr_dispatchers, dispatch_delay_loops, work_loops,
 404                 nr_workers, tot_enqueues, tot_dequeues, max_queue_len);
 405         free(count_dispatcher);
 406         free(count_worker);
 407         free(tid_dispatcher);
 408         free(tid_worker);
 409         return retval;
 410 }