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>
40 #include <urcu/arch.h>
41 #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
70 #include <urcu-defer.h>
72 static volatile int test_go
, test_stop
;
74 static unsigned long rduration
;
76 static unsigned long duration
;
78 /* read-side C.S. duration, in loops */
79 static unsigned long wdelay
;
81 static inline void loop_sleep(unsigned long loops
)
87 static int verbose_mode
;
89 #define printf_verbose(fmt, args...) \
95 static unsigned int cpu_affinities
[NR_CPUS
];
96 static unsigned int next_aff
= 0;
97 static int use_affinity
= 0;
99 pthread_mutex_t affinity_mutex
= PTHREAD_MUTEX_INITIALIZER
;
101 #ifndef HAVE_CPU_SET_T
102 typedef unsigned long cpu_set_t
;
103 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
104 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
107 static void set_affinity(void)
116 #if HAVE_SCHED_SETAFFINITY
117 ret
= pthread_mutex_lock(&affinity_mutex
);
119 perror("Error in pthread mutex lock");
122 cpu
= cpu_affinities
[next_aff
++];
123 ret
= pthread_mutex_unlock(&affinity_mutex
);
125 perror("Error in pthread mutex unlock");
131 #if SCHED_SETAFFINITY_ARGS == 2
132 sched_setaffinity(0, &mask
);
134 sched_setaffinity(0, sizeof(mask
), &mask
);
136 #endif /* HAVE_SCHED_SETAFFINITY */
140 * returns 0 if test should end.
142 static int test_duration_dequeue(void)
147 static int test_duration_enqueue(void)
152 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues
);
153 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues
);
155 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues
);
156 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues
);
158 static unsigned int nr_enqueuers
;
159 static unsigned int nr_dequeuers
;
162 struct cds_lfq_node_rcu list
;
166 static struct cds_lfq_queue_rcu q
;
168 void *thr_enqueuer(void *_count
)
170 unsigned long long *count
= _count
;
172 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
173 "enqueuer", pthread_self(), (unsigned long)gettid());
177 rcu_register_thread();
185 struct test
*node
= malloc(sizeof(*node
));
188 cds_lfq_node_init_rcu(&node
->list
);
190 cds_lfq_enqueue_rcu(&q
, &node
->list
);
192 URCU_TLS(nr_successful_enqueues
)++;
194 if (caa_unlikely(wdelay
))
197 URCU_TLS(nr_enqueues
)++;
198 if (caa_unlikely(!test_duration_enqueue()))
202 rcu_unregister_thread();
204 count
[0] = URCU_TLS(nr_enqueues
);
205 count
[1] = URCU_TLS(nr_successful_enqueues
);
206 printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
207 "enqueues %llu successful_enqueues %llu\n",
208 pthread_self(), (unsigned long)gettid(),
209 URCU_TLS(nr_enqueues
), URCU_TLS(nr_successful_enqueues
));
215 void free_node_cb(struct rcu_head
*head
)
218 caa_container_of(head
, struct test
, rcu
);
222 void *thr_dequeuer(void *_count
)
224 unsigned long long *count
= _count
;
227 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
228 "dequeuer", pthread_self(), (unsigned long)gettid());
232 ret
= rcu_defer_register_thread();
234 printf("Error in rcu_defer_register_thread\n");
237 rcu_register_thread();
245 struct cds_lfq_node_rcu
*qnode
;
248 qnode
= cds_lfq_dequeue_rcu(&q
);
254 node
= caa_container_of(qnode
, struct test
, list
);
255 call_rcu(&node
->rcu
, free_node_cb
);
256 URCU_TLS(nr_successful_dequeues
)++;
259 URCU_TLS(nr_dequeues
)++;
260 if (caa_unlikely(!test_duration_dequeue()))
262 if (caa_unlikely(rduration
))
263 loop_sleep(rduration
);
266 rcu_unregister_thread();
267 rcu_defer_unregister_thread();
268 printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
269 "dequeues %llu, successful_dequeues %llu\n",
270 pthread_self(), (unsigned long)gettid(),
271 URCU_TLS(nr_dequeues
), URCU_TLS(nr_successful_dequeues
));
272 count
[0] = URCU_TLS(nr_dequeues
);
273 count
[1] = URCU_TLS(nr_successful_dequeues
);
277 void test_end(struct cds_lfq_queue_rcu
*q
, unsigned long long *nr_dequeues
)
279 struct cds_lfq_node_rcu
*snode
;
282 snode
= cds_lfq_dequeue_rcu(q
);
286 node
= caa_container_of(snode
, struct test
, list
);
287 free(node
); /* no more concurrent access */
293 void show_usage(int argc
, char **argv
)
295 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv
[0]);
296 printf(" [-d delay] (enqueuer period (in loops))");
297 printf(" [-c duration] (dequeuer period (in loops))");
298 printf(" [-v] (verbose output)");
299 printf(" [-a cpu#] [-a cpu#]... (affinity)");
303 int main(int argc
, char **argv
)
306 pthread_t
*tid_enqueuer
, *tid_dequeuer
;
308 unsigned long long *count_enqueuer
, *count_dequeuer
;
309 unsigned long long tot_enqueues
= 0, tot_dequeues
= 0;
310 unsigned long long tot_successful_enqueues
= 0,
311 tot_successful_dequeues
= 0;
312 unsigned long long end_dequeues
= 0;
316 show_usage(argc
, argv
);
320 err
= sscanf(argv
[1], "%u", &nr_dequeuers
);
322 show_usage(argc
, argv
);
326 err
= sscanf(argv
[2], "%u", &nr_enqueuers
);
328 show_usage(argc
, argv
);
332 err
= sscanf(argv
[3], "%lu", &duration
);
334 show_usage(argc
, argv
);
338 for (i
= 4; i
< argc
; i
++) {
339 if (argv
[i
][0] != '-')
341 switch (argv
[i
][1]) {
344 show_usage(argc
, argv
);
348 cpu_affinities
[next_aff
++] = a
;
350 printf_verbose("Adding CPU %d affinity\n", a
);
354 show_usage(argc
, argv
);
357 rduration
= atol(argv
[++i
]);
361 show_usage(argc
, argv
);
364 wdelay
= atol(argv
[++i
]);
372 printf_verbose("running test for %lu seconds, %u enqueuers, "
374 duration
, nr_enqueuers
, nr_dequeuers
);
375 printf_verbose("Writer delay : %lu loops.\n", rduration
);
376 printf_verbose("Reader duration : %lu loops.\n", wdelay
);
377 printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
378 "main", pthread_self(), (unsigned long)gettid());
380 tid_enqueuer
= malloc(sizeof(*tid_enqueuer
) * nr_enqueuers
);
381 tid_dequeuer
= malloc(sizeof(*tid_dequeuer
) * nr_dequeuers
);
382 count_enqueuer
= malloc(2 * sizeof(*count_enqueuer
) * nr_enqueuers
);
383 count_dequeuer
= malloc(2 * sizeof(*count_dequeuer
) * nr_dequeuers
);
384 cds_lfq_init_rcu(&q
, call_rcu
);
385 err
= create_all_cpu_call_rcu_data(0);
387 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
392 for (i
= 0; i
< nr_enqueuers
; i
++) {
393 err
= pthread_create(&tid_enqueuer
[i
], NULL
, thr_enqueuer
,
394 &count_enqueuer
[2 * i
]);
398 for (i
= 0; i
< nr_dequeuers
; i
++) {
399 err
= pthread_create(&tid_dequeuer
[i
], NULL
, thr_dequeuer
,
400 &count_dequeuer
[2 * i
]);
409 for (i
= 0; i
< duration
; i
++) {
417 for (i
= 0; i
< nr_enqueuers
; i
++) {
418 err
= pthread_join(tid_enqueuer
[i
], &tret
);
421 tot_enqueues
+= count_enqueuer
[2 * i
];
422 tot_successful_enqueues
+= count_enqueuer
[2 * i
+ 1];
424 for (i
= 0; i
< nr_dequeuers
; i
++) {
425 err
= pthread_join(tid_dequeuer
[i
], &tret
);
428 tot_dequeues
+= count_dequeuer
[2 * i
];
429 tot_successful_dequeues
+= count_dequeuer
[2 * i
+ 1];
432 test_end(&q
, &end_dequeues
);
433 err
= cds_lfq_destroy_rcu(&q
);
436 printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
437 tot_enqueues
, tot_dequeues
);
438 printf_verbose("total number of successful enqueues : %llu, "
439 "successful dequeues %llu\n",
440 tot_successful_enqueues
, tot_successful_dequeues
);
441 printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
443 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
444 "successful enqueues %12llu successful dequeues %12llu "
445 "end_dequeues %llu nr_ops %12llu\n",
446 argv
[0], duration
, nr_enqueuers
, wdelay
,
447 nr_dequeuers
, rduration
, tot_enqueues
, tot_dequeues
,
448 tot_successful_enqueues
,
449 tot_successful_dequeues
, end_dequeues
,
450 tot_enqueues
+ tot_dequeues
);
451 if (tot_successful_enqueues
!= tot_successful_dequeues
+ end_dequeues
)
452 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
453 "succ. dequeues + end dequeues %llu.\n",
454 tot_successful_enqueues
,
455 tot_successful_dequeues
+ end_dequeues
);
457 free_all_cpu_call_rcu_data();
458 free(count_enqueuer
);
459 free(count_dequeuer
);