4 * Userspace RCU library - example RCU-based lock-free queue
6 * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com>
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.
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.
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.
25 #include "../config.h"
32 #include <sys/types.h>
39 #include <urcu/arch.h>
40 #include <urcu/tls-compat.h>
47 /* hardcoded number of CPUs */
50 #if defined(_syscall0)
51 _syscall0(pid_t
, gettid
)
52 #elif defined(__NR_gettid)
53 static inline pid_t
gettid(void)
55 return syscall(__NR_gettid
);
58 #warning "use pid as tid"
59 static inline pid_t
gettid(void)
65 #ifndef DYNAMIC_LINK_TEST
71 static volatile int test_go
, test_stop
;
73 static unsigned long rduration
;
75 static unsigned long duration
;
77 /* read-side C.S. duration, in loops */
78 static unsigned long wdelay
;
80 static inline void loop_sleep(unsigned long loops
)
86 static int verbose_mode
;
88 #define printf_verbose(fmt, args...) \
94 static unsigned int cpu_affinities
[NR_CPUS
];
95 static unsigned int next_aff
= 0;
96 static int use_affinity
= 0;
98 pthread_mutex_t affinity_mutex
= PTHREAD_MUTEX_INITIALIZER
;
100 static void set_affinity(void)
102 #if HAVE_SCHED_SETAFFINITY
105 #endif /* HAVE_SCHED_SETAFFINITY */
110 #if HAVE_SCHED_SETAFFINITY
111 ret
= pthread_mutex_lock(&affinity_mutex
);
113 perror("Error in pthread mutex lock");
116 cpu
= cpu_affinities
[next_aff
++];
117 ret
= pthread_mutex_unlock(&affinity_mutex
);
119 perror("Error in pthread mutex unlock");
125 #if SCHED_SETAFFINITY_ARGS == 2
126 sched_setaffinity(0, &mask
);
128 sched_setaffinity(0, sizeof(mask
), &mask
);
130 #endif /* HAVE_SCHED_SETAFFINITY */
134 * returns 0 if test should end.
136 static int test_duration_dequeue(void)
141 static int test_duration_enqueue(void)
146 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues
);
147 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues
);
149 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues
);
150 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues
);
152 static unsigned int nr_enqueuers
;
153 static unsigned int nr_dequeuers
;
156 struct cds_lfq_node_rcu list
;
160 static struct cds_lfq_queue_rcu q
;
162 void *thr_enqueuer(void *_count
)
164 unsigned long long *count
= _count
;
166 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
167 "enqueuer", (unsigned long) pthread_self(),
168 (unsigned long) gettid());
172 rcu_register_thread();
180 struct test
*node
= malloc(sizeof(*node
));
183 cds_lfq_node_init_rcu(&node
->list
);
185 cds_lfq_enqueue_rcu(&q
, &node
->list
);
187 URCU_TLS(nr_successful_enqueues
)++;
189 if (caa_unlikely(wdelay
))
192 URCU_TLS(nr_enqueues
)++;
193 if (caa_unlikely(!test_duration_enqueue()))
197 rcu_unregister_thread();
199 count
[0] = URCU_TLS(nr_enqueues
);
200 count
[1] = URCU_TLS(nr_successful_enqueues
);
201 printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
202 "enqueues %llu successful_enqueues %llu\n",
204 (unsigned long) gettid(),
205 URCU_TLS(nr_enqueues
), URCU_TLS(nr_successful_enqueues
));
211 void free_node_cb(struct rcu_head
*head
)
214 caa_container_of(head
, struct test
, rcu
);
218 void *thr_dequeuer(void *_count
)
220 unsigned long long *count
= _count
;
222 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
223 "dequeuer", (unsigned long) pthread_self(),
224 (unsigned long) gettid());
228 rcu_register_thread();
236 struct cds_lfq_node_rcu
*qnode
;
239 qnode
= cds_lfq_dequeue_rcu(&q
);
245 node
= caa_container_of(qnode
, struct test
, list
);
246 call_rcu(&node
->rcu
, free_node_cb
);
247 URCU_TLS(nr_successful_dequeues
)++;
250 URCU_TLS(nr_dequeues
)++;
251 if (caa_unlikely(!test_duration_dequeue()))
253 if (caa_unlikely(rduration
))
254 loop_sleep(rduration
);
257 rcu_unregister_thread();
258 printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
259 "dequeues %llu, successful_dequeues %llu\n",
261 (unsigned long) gettid(),
262 URCU_TLS(nr_dequeues
), URCU_TLS(nr_successful_dequeues
));
263 count
[0] = URCU_TLS(nr_dequeues
);
264 count
[1] = URCU_TLS(nr_successful_dequeues
);
268 void test_end(struct cds_lfq_queue_rcu
*q
, unsigned long long *nr_dequeues
)
270 struct cds_lfq_node_rcu
*snode
;
273 snode
= cds_lfq_dequeue_rcu(q
);
277 node
= caa_container_of(snode
, struct test
, list
);
278 free(node
); /* no more concurrent access */
284 void show_usage(int argc
, char **argv
)
286 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
288 printf("OPTIONS:\n");
289 printf(" [-d delay] (enqueuer period (in loops))\n");
290 printf(" [-c duration] (dequeuer period (in loops))\n");
291 printf(" [-v] (verbose output)\n");
292 printf(" [-a cpu#] [-a cpu#]... (affinity)\n");
296 int main(int argc
, char **argv
)
299 pthread_t
*tid_enqueuer
, *tid_dequeuer
;
301 unsigned long long *count_enqueuer
, *count_dequeuer
;
302 unsigned long long tot_enqueues
= 0, tot_dequeues
= 0;
303 unsigned long long tot_successful_enqueues
= 0,
304 tot_successful_dequeues
= 0;
305 unsigned long long end_dequeues
= 0;
309 show_usage(argc
, argv
);
313 err
= sscanf(argv
[1], "%u", &nr_dequeuers
);
315 show_usage(argc
, argv
);
319 err
= sscanf(argv
[2], "%u", &nr_enqueuers
);
321 show_usage(argc
, argv
);
325 err
= sscanf(argv
[3], "%lu", &duration
);
327 show_usage(argc
, argv
);
331 for (i
= 4; i
< argc
; i
++) {
332 if (argv
[i
][0] != '-')
334 switch (argv
[i
][1]) {
337 show_usage(argc
, argv
);
341 cpu_affinities
[next_aff
++] = a
;
343 printf_verbose("Adding CPU %d affinity\n", a
);
347 show_usage(argc
, argv
);
350 rduration
= atol(argv
[++i
]);
354 show_usage(argc
, argv
);
357 wdelay
= atol(argv
[++i
]);
365 printf_verbose("running test for %lu seconds, %u enqueuers, "
367 duration
, nr_enqueuers
, nr_dequeuers
);
368 printf_verbose("Writer delay : %lu loops.\n", rduration
);
369 printf_verbose("Reader duration : %lu loops.\n", wdelay
);
370 printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
371 "main", (unsigned long) pthread_self(),
372 (unsigned long) gettid());
374 tid_enqueuer
= malloc(sizeof(*tid_enqueuer
) * nr_enqueuers
);
375 tid_dequeuer
= malloc(sizeof(*tid_dequeuer
) * nr_dequeuers
);
376 count_enqueuer
= malloc(2 * sizeof(*count_enqueuer
) * nr_enqueuers
);
377 count_dequeuer
= malloc(2 * sizeof(*count_dequeuer
) * nr_dequeuers
);
378 cds_lfq_init_rcu(&q
, call_rcu
);
379 err
= create_all_cpu_call_rcu_data(0);
381 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
386 for (i
= 0; i
< nr_enqueuers
; i
++) {
387 err
= pthread_create(&tid_enqueuer
[i
], NULL
, thr_enqueuer
,
388 &count_enqueuer
[2 * i
]);
392 for (i
= 0; i
< nr_dequeuers
; i
++) {
393 err
= pthread_create(&tid_dequeuer
[i
], NULL
, thr_dequeuer
,
394 &count_dequeuer
[2 * i
]);
403 for (i
= 0; i
< duration
; i
++) {
411 for (i
= 0; i
< nr_enqueuers
; i
++) {
412 err
= pthread_join(tid_enqueuer
[i
], &tret
);
415 tot_enqueues
+= count_enqueuer
[2 * i
];
416 tot_successful_enqueues
+= count_enqueuer
[2 * i
+ 1];
418 for (i
= 0; i
< nr_dequeuers
; i
++) {
419 err
= pthread_join(tid_dequeuer
[i
], &tret
);
422 tot_dequeues
+= count_dequeuer
[2 * i
];
423 tot_successful_dequeues
+= count_dequeuer
[2 * i
+ 1];
426 test_end(&q
, &end_dequeues
);
427 err
= cds_lfq_destroy_rcu(&q
);
430 printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
431 tot_enqueues
, tot_dequeues
);
432 printf_verbose("total number of successful enqueues : %llu, "
433 "successful dequeues %llu\n",
434 tot_successful_enqueues
, tot_successful_dequeues
);
435 printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
437 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
438 "successful enqueues %12llu successful dequeues %12llu "
439 "end_dequeues %llu nr_ops %12llu\n",
440 argv
[0], duration
, nr_enqueuers
, wdelay
,
441 nr_dequeuers
, rduration
, tot_enqueues
, tot_dequeues
,
442 tot_successful_enqueues
,
443 tot_successful_dequeues
, end_dequeues
,
444 tot_enqueues
+ tot_dequeues
);
445 if (tot_successful_enqueues
!= tot_successful_dequeues
+ end_dequeues
)
446 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
447 "succ. dequeues + end dequeues %llu.\n",
448 tot_successful_enqueues
,
449 tot_successful_dequeues
+ end_dequeues
);
451 free_all_cpu_call_rcu_data();
452 free(count_enqueuer
);
453 free(count_dequeuer
);