4 * Userspace RCU library - example RCU-based lock-free stack
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
70 /* Remove deprecation warnings from test build. */
71 #define CDS_LFS_RCU_DEPRECATED
75 static volatile int test_go
, test_stop
;
77 static unsigned long rduration
;
79 static unsigned long duration
;
81 /* read-side C.S. duration, in loops */
82 static unsigned long wdelay
;
84 static inline void loop_sleep(unsigned long loops
)
90 static int verbose_mode
;
92 #define printf_verbose(fmt, args...) \
98 static unsigned int cpu_affinities
[NR_CPUS
];
99 static unsigned int next_aff
= 0;
100 static int use_affinity
= 0;
102 pthread_mutex_t affinity_mutex
= PTHREAD_MUTEX_INITIALIZER
;
104 static void set_affinity(void)
113 #if HAVE_SCHED_SETAFFINITY
114 ret
= pthread_mutex_lock(&affinity_mutex
);
116 perror("Error in pthread mutex lock");
119 cpu
= cpu_affinities
[next_aff
++];
120 ret
= pthread_mutex_unlock(&affinity_mutex
);
122 perror("Error in pthread mutex unlock");
128 #if SCHED_SETAFFINITY_ARGS == 2
129 sched_setaffinity(0, &mask
);
131 sched_setaffinity(0, sizeof(mask
), &mask
);
133 #endif /* HAVE_SCHED_SETAFFINITY */
137 * returns 0 if test should end.
139 static int test_duration_dequeue(void)
144 static int test_duration_enqueue(void)
149 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues
);
150 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues
);
152 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues
);
153 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues
);
155 static unsigned int nr_enqueuers
;
156 static unsigned int nr_dequeuers
;
159 struct cds_lfs_node_rcu list
;
163 static struct cds_lfs_stack_rcu s
;
165 void *thr_enqueuer(void *_count
)
167 unsigned long long *count
= _count
;
169 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
170 "enqueuer", (unsigned long) pthread_self(),
171 (unsigned long) gettid());
175 rcu_register_thread();
183 struct test
*node
= malloc(sizeof(*node
));
186 cds_lfs_node_init_rcu(&node
->list
);
187 /* No rcu read-side is needed for push */
188 cds_lfs_push_rcu(&s
, &node
->list
);
189 URCU_TLS(nr_successful_enqueues
)++;
191 if (caa_unlikely(wdelay
))
194 URCU_TLS(nr_enqueues
)++;
195 if (caa_unlikely(!test_duration_enqueue()))
199 rcu_unregister_thread();
201 count
[0] = URCU_TLS(nr_enqueues
);
202 count
[1] = URCU_TLS(nr_successful_enqueues
);
203 printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
204 "enqueues %llu successful_enqueues %llu\n",
206 (unsigned long) gettid(),
207 URCU_TLS(nr_enqueues
), URCU_TLS(nr_successful_enqueues
));
213 void free_node_cb(struct rcu_head
*head
)
216 caa_container_of(head
, struct test
, rcu
);
220 void *thr_dequeuer(void *_count
)
222 unsigned long long *count
= _count
;
224 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
225 "dequeuer", (unsigned long) pthread_self(),
226 (unsigned long) gettid());
230 rcu_register_thread();
238 struct cds_lfs_node_rcu
*snode
;
241 snode
= cds_lfs_pop_rcu(&s
);
246 node
= caa_container_of(snode
, struct test
, list
);
247 call_rcu(&node
->rcu
, free_node_cb
);
248 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();
259 printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
260 "dequeues %llu, successful_dequeues %llu\n",
262 (unsigned long) gettid(),
263 URCU_TLS(nr_dequeues
), URCU_TLS(nr_successful_dequeues
));
264 count
[0] = URCU_TLS(nr_dequeues
);
265 count
[1] = URCU_TLS(nr_successful_dequeues
);
269 void test_end(struct cds_lfs_stack_rcu
*s
, unsigned long long *nr_dequeues
)
271 struct cds_lfs_node_rcu
*snode
;
274 snode
= cds_lfs_pop_rcu(s
);
278 node
= caa_container_of(snode
, struct test
, list
);
285 void show_usage(int argc
, char **argv
)
287 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
289 printf("OPTIONS:\n");
290 printf(" [-d delay] (enqueuer period (in loops))\n");
291 printf(" [-c duration] (dequeuer period (in loops))\n");
292 printf(" [-v] (verbose output)\n");
293 printf(" [-a cpu#] [-a cpu#]... (affinity)\n");
297 int main(int argc
, char **argv
)
300 pthread_t
*tid_enqueuer
, *tid_dequeuer
;
302 unsigned long long *count_enqueuer
, *count_dequeuer
;
303 unsigned long long tot_enqueues
= 0, tot_dequeues
= 0;
304 unsigned long long tot_successful_enqueues
= 0,
305 tot_successful_dequeues
= 0;
306 unsigned long long end_dequeues
= 0;
310 show_usage(argc
, argv
);
314 err
= sscanf(argv
[1], "%u", &nr_dequeuers
);
316 show_usage(argc
, argv
);
320 err
= sscanf(argv
[2], "%u", &nr_enqueuers
);
322 show_usage(argc
, argv
);
326 err
= sscanf(argv
[3], "%lu", &duration
);
328 show_usage(argc
, argv
);
332 for (i
= 4; i
< argc
; i
++) {
333 if (argv
[i
][0] != '-')
335 switch (argv
[i
][1]) {
338 show_usage(argc
, argv
);
342 cpu_affinities
[next_aff
++] = a
;
344 printf_verbose("Adding CPU %d affinity\n", a
);
348 show_usage(argc
, argv
);
351 rduration
= atol(argv
[++i
]);
355 show_usage(argc
, argv
);
358 wdelay
= atol(argv
[++i
]);
366 printf_verbose("running test for %lu seconds, %u enqueuers, "
368 duration
, nr_enqueuers
, nr_dequeuers
);
369 printf_verbose("Writer delay : %lu loops.\n", rduration
);
370 printf_verbose("Reader duration : %lu loops.\n", wdelay
);
371 printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
372 "main", (unsigned long) pthread_self(),
373 (unsigned long) gettid());
375 tid_enqueuer
= malloc(sizeof(*tid_enqueuer
) * nr_enqueuers
);
376 tid_dequeuer
= malloc(sizeof(*tid_dequeuer
) * nr_dequeuers
);
377 count_enqueuer
= malloc(2 * sizeof(*count_enqueuer
) * nr_enqueuers
);
378 count_dequeuer
= malloc(2 * sizeof(*count_dequeuer
) * nr_dequeuers
);
379 cds_lfs_init_rcu(&s
);
380 err
= create_all_cpu_call_rcu_data(0);
382 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
387 for (i
= 0; i
< nr_enqueuers
; i
++) {
388 err
= pthread_create(&tid_enqueuer
[i
], NULL
, thr_enqueuer
,
389 &count_enqueuer
[2 * i
]);
393 for (i
= 0; i
< nr_dequeuers
; i
++) {
394 err
= pthread_create(&tid_dequeuer
[i
], NULL
, thr_dequeuer
,
395 &count_dequeuer
[2 * i
]);
404 for (i
= 0; i
< duration
; i
++) {
412 for (i
= 0; i
< nr_enqueuers
; i
++) {
413 err
= pthread_join(tid_enqueuer
[i
], &tret
);
416 tot_enqueues
+= count_enqueuer
[2 * i
];
417 tot_successful_enqueues
+= count_enqueuer
[2 * i
+ 1];
419 for (i
= 0; i
< nr_dequeuers
; i
++) {
420 err
= pthread_join(tid_dequeuer
[i
], &tret
);
423 tot_dequeues
+= count_dequeuer
[2 * i
];
424 tot_successful_dequeues
+= count_dequeuer
[2 * i
+ 1];
427 test_end(&s
, &end_dequeues
);
429 printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
430 tot_enqueues
, tot_dequeues
);
431 printf_verbose("total number of successful enqueues : %llu, "
432 "successful dequeues %llu\n",
433 tot_successful_enqueues
, tot_successful_dequeues
);
434 printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
436 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
437 "successful enqueues %12llu successful dequeues %12llu "
438 "end_dequeues %llu nr_ops %12llu\n",
439 argv
[0], duration
, nr_enqueuers
, wdelay
,
440 nr_dequeuers
, rduration
, tot_enqueues
, tot_dequeues
,
441 tot_successful_enqueues
,
442 tot_successful_dequeues
, end_dequeues
,
443 tot_enqueues
+ tot_dequeues
);
444 if (tot_successful_enqueues
!= tot_successful_dequeues
+ end_dequeues
)
445 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
446 "succ. dequeues + end dequeues %llu.\n",
447 tot_successful_enqueues
,
448 tot_successful_dequeues
+ end_dequeues
);
450 free_all_cpu_call_rcu_data();
451 free(count_enqueuer
);
452 free(count_dequeuer
);