4 * Userspace RCU library - example lock-free stack
6 * Copyright 2010-2012 - 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
72 * External synchronization used.
79 static enum test_sync test_sync
;
81 static volatile int test_go
, test_stop
;
83 static unsigned long rduration
;
85 static unsigned long duration
;
87 /* read-side C.S. duration, in loops */
88 static unsigned long wdelay
;
90 static inline void loop_sleep(unsigned long loops
)
96 static int verbose_mode
;
98 static int test_pop
, test_pop_all
;
100 #define printf_verbose(fmt, args...) \
103 printf(fmt, ## args); \
106 static unsigned int cpu_affinities
[NR_CPUS
];
107 static unsigned int next_aff
= 0;
108 static int use_affinity
= 0;
110 pthread_mutex_t affinity_mutex
= PTHREAD_MUTEX_INITIALIZER
;
112 #ifndef HAVE_CPU_SET_T
113 typedef unsigned long cpu_set_t
;
114 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
115 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
118 static void set_affinity(void)
120 #if HAVE_SCHED_SETAFFINITY
123 #endif /* HAVE_SCHED_SETAFFINITY */
128 #if HAVE_SCHED_SETAFFINITY
129 ret
= pthread_mutex_lock(&affinity_mutex
);
131 perror("Error in pthread mutex lock");
134 cpu
= cpu_affinities
[next_aff
++];
135 ret
= pthread_mutex_unlock(&affinity_mutex
);
137 perror("Error in pthread mutex unlock");
143 #if SCHED_SETAFFINITY_ARGS == 2
144 sched_setaffinity(0, &mask
);
146 sched_setaffinity(0, sizeof(mask
), &mask
);
148 #endif /* HAVE_SCHED_SETAFFINITY */
152 * returns 0 if test should end.
154 static int test_duration_dequeue(void)
159 static int test_duration_enqueue(void)
164 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues
);
165 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues
);
167 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues
);
168 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues
);
170 static unsigned int nr_enqueuers
;
171 static unsigned int nr_dequeuers
;
174 struct cds_lfs_node list
;
178 static struct cds_lfs_stack s
;
180 static void *thr_enqueuer(void *_count
)
182 unsigned long long *count
= _count
;
184 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
185 "enqueuer", (unsigned long) pthread_self(),
186 (unsigned long) gettid());
190 rcu_register_thread();
198 struct test
*node
= malloc(sizeof(*node
));
201 cds_lfs_node_init(&node
->list
);
202 cds_lfs_push(&s
, &node
->list
);
203 URCU_TLS(nr_successful_enqueues
)++;
205 if (caa_unlikely(wdelay
))
208 URCU_TLS(nr_enqueues
)++;
209 if (caa_unlikely(!test_duration_enqueue()))
213 rcu_unregister_thread();
215 count
[0] = URCU_TLS(nr_enqueues
);
216 count
[1] = URCU_TLS(nr_successful_enqueues
);
217 printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
218 "enqueues %llu successful_enqueues %llu\n",
220 (unsigned long) gettid(),
221 URCU_TLS(nr_enqueues
), URCU_TLS(nr_successful_enqueues
));
227 void free_node_cb(struct rcu_head
*head
)
230 caa_container_of(head
, struct test
, rcu
);
235 void do_test_pop(enum test_sync sync
)
237 struct cds_lfs_node
*snode
;
239 if (sync
== TEST_SYNC_RCU
)
241 snode
= __cds_lfs_pop(&s
);
242 if (sync
== TEST_SYNC_RCU
)
247 node
= caa_container_of(snode
,
249 if (sync
== TEST_SYNC_RCU
)
250 call_rcu(&node
->rcu
, free_node_cb
);
253 URCU_TLS(nr_successful_dequeues
)++;
255 URCU_TLS(nr_dequeues
)++;
259 void do_test_pop_all(enum test_sync sync
)
261 struct cds_lfs_node
*snode
;
262 struct cds_lfs_head
*head
;
263 struct cds_lfs_node
*n
;
265 head
= __cds_lfs_pop_all(&s
);
266 cds_lfs_for_each_safe(head
, snode
, n
) {
269 node
= caa_container_of(snode
, struct test
, list
);
270 if (sync
== TEST_SYNC_RCU
)
271 call_rcu(&node
->rcu
, free_node_cb
);
274 URCU_TLS(nr_successful_dequeues
)++;
275 URCU_TLS(nr_dequeues
)++;
280 static void *thr_dequeuer(void *_count
)
282 unsigned long long *count
= _count
;
284 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
285 "dequeuer", (unsigned long) pthread_self(),
286 (unsigned long) gettid());
290 rcu_register_thread();
297 assert(test_pop
|| test_pop_all
);
300 unsigned int counter
= 0;
302 if (test_pop
&& test_pop_all
) {
303 /* both pop and pop all */
305 do_test_pop(test_sync
);
307 do_test_pop_all(test_sync
);
311 do_test_pop(test_sync
);
313 do_test_pop_all(test_sync
);
316 if (caa_unlikely(!test_duration_dequeue()))
318 if (caa_unlikely(rduration
))
319 loop_sleep(rduration
);
322 rcu_unregister_thread();
324 printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
325 "dequeues %llu, successful_dequeues %llu\n",
327 (unsigned long) gettid(),
328 URCU_TLS(nr_dequeues
), URCU_TLS(nr_successful_dequeues
));
329 count
[0] = URCU_TLS(nr_dequeues
);
330 count
[1] = URCU_TLS(nr_successful_dequeues
);
334 static void test_end(struct cds_lfs_stack
*s
, unsigned long long *nr_dequeues
)
336 struct cds_lfs_node
*snode
;
339 snode
= __cds_lfs_pop(s
);
343 node
= caa_container_of(snode
, struct test
, list
);
350 static void show_usage(int argc
, char **argv
)
352 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv
[0]);
353 printf(" [-d delay] (enqueuer period (in loops))");
354 printf(" [-c duration] (dequeuer period (in loops))");
355 printf(" [-v] (verbose output)");
356 printf(" [-a cpu#] [-a cpu#]... (affinity)");
357 printf(" [-p] (test pop)");
358 printf(" [-P] (test pop_all, enabled by default)");
359 printf(" [-R] (use RCU external synchronization)");
360 printf(" Note: default: no external synchronization used.");
364 int main(int argc
, char **argv
)
367 pthread_t
*tid_enqueuer
, *tid_dequeuer
;
369 unsigned long long *count_enqueuer
, *count_dequeuer
;
370 unsigned long long tot_enqueues
= 0, tot_dequeues
= 0;
371 unsigned long long tot_successful_enqueues
= 0,
372 tot_successful_dequeues
= 0;
373 unsigned long long end_dequeues
= 0;
377 show_usage(argc
, argv
);
381 err
= sscanf(argv
[1], "%u", &nr_dequeuers
);
383 show_usage(argc
, argv
);
387 err
= sscanf(argv
[2], "%u", &nr_enqueuers
);
389 show_usage(argc
, argv
);
393 err
= sscanf(argv
[3], "%lu", &duration
);
395 show_usage(argc
, argv
);
399 for (i
= 4; i
< argc
; i
++) {
400 if (argv
[i
][0] != '-')
402 switch (argv
[i
][1]) {
405 show_usage(argc
, argv
);
409 cpu_affinities
[next_aff
++] = a
;
411 printf_verbose("Adding CPU %d affinity\n", a
);
415 show_usage(argc
, argv
);
418 rduration
= atol(argv
[++i
]);
422 show_usage(argc
, argv
);
425 wdelay
= atol(argv
[++i
]);
437 test_sync
= TEST_SYNC_RCU
;
442 /* activate pop_all test by default */
443 if (!test_pop
&& !test_pop_all
)
446 printf_verbose("running test for %lu seconds, %u enqueuers, "
448 duration
, nr_enqueuers
, nr_dequeuers
);
450 printf_verbose("pop test activated.\n");
452 printf_verbose("pop_all test activated.\n");
453 if (test_sync
== TEST_SYNC_RCU
)
454 printf_verbose("External sync: RCU.\n");
456 printf_verbose("External sync: none.\n");
457 printf_verbose("Writer delay : %lu loops.\n", rduration
);
458 printf_verbose("Reader duration : %lu loops.\n", wdelay
);
459 printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
460 "main", (unsigned long) pthread_self(),
461 (unsigned long) gettid());
463 tid_enqueuer
= malloc(sizeof(*tid_enqueuer
) * nr_enqueuers
);
464 tid_dequeuer
= malloc(sizeof(*tid_dequeuer
) * nr_dequeuers
);
465 count_enqueuer
= malloc(2 * sizeof(*count_enqueuer
) * nr_enqueuers
);
466 count_dequeuer
= malloc(2 * sizeof(*count_dequeuer
) * nr_dequeuers
);
468 err
= create_all_cpu_call_rcu_data(0);
470 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
475 for (i
= 0; i
< nr_enqueuers
; i
++) {
476 err
= pthread_create(&tid_enqueuer
[i
], NULL
, thr_enqueuer
,
477 &count_enqueuer
[2 * i
]);
481 for (i
= 0; i
< nr_dequeuers
; i
++) {
482 err
= pthread_create(&tid_dequeuer
[i
], NULL
, thr_dequeuer
,
483 &count_dequeuer
[2 * i
]);
492 for (i
= 0; i
< duration
; i
++) {
500 for (i
= 0; i
< nr_enqueuers
; i
++) {
501 err
= pthread_join(tid_enqueuer
[i
], &tret
);
504 tot_enqueues
+= count_enqueuer
[2 * i
];
505 tot_successful_enqueues
+= count_enqueuer
[2 * i
+ 1];
507 for (i
= 0; i
< nr_dequeuers
; i
++) {
508 err
= pthread_join(tid_dequeuer
[i
], &tret
);
511 tot_dequeues
+= count_dequeuer
[2 * i
];
512 tot_successful_dequeues
+= count_dequeuer
[2 * i
+ 1];
515 test_end(&s
, &end_dequeues
);
517 printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
518 tot_enqueues
, tot_dequeues
);
519 printf_verbose("total number of successful enqueues : %llu, "
520 "successful dequeues %llu\n",
521 tot_successful_enqueues
, tot_successful_dequeues
);
522 printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
524 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
525 "successful enqueues %12llu successful dequeues %12llu "
526 "end_dequeues %llu nr_ops %12llu\n",
527 argv
[0], duration
, nr_enqueuers
, wdelay
,
528 nr_dequeuers
, rduration
, tot_enqueues
, tot_dequeues
,
529 tot_successful_enqueues
,
530 tot_successful_dequeues
, end_dequeues
,
531 tot_enqueues
+ tot_dequeues
);
532 if (tot_successful_enqueues
!= tot_successful_dequeues
+ end_dequeues
)
533 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
534 "succ. dequeues + end dequeues %llu.\n",
535 tot_successful_enqueues
,
536 tot_successful_dequeues
+ end_dequeues
);
538 free_all_cpu_call_rcu_data();
539 free(count_enqueuer
);
540 free(count_dequeuer
);