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>
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 /* 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 #ifndef HAVE_CPU_SET_T
105 typedef unsigned long cpu_set_t
;
106 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
107 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
110 static void set_affinity(void)
119 #if HAVE_SCHED_SETAFFINITY
120 ret
= pthread_mutex_lock(&affinity_mutex
);
122 perror("Error in pthread mutex lock");
125 cpu
= cpu_affinities
[next_aff
++];
126 ret
= pthread_mutex_unlock(&affinity_mutex
);
128 perror("Error in pthread mutex unlock");
134 #if SCHED_SETAFFINITY_ARGS == 2
135 sched_setaffinity(0, &mask
);
137 sched_setaffinity(0, sizeof(mask
), &mask
);
139 #endif /* HAVE_SCHED_SETAFFINITY */
143 * returns 0 if test should end.
145 static int test_duration_dequeue(void)
150 static int test_duration_enqueue(void)
155 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues
);
156 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues
);
158 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues
);
159 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues
);
161 static unsigned int nr_enqueuers
;
162 static unsigned int nr_dequeuers
;
165 struct cds_lfs_node_rcu list
;
169 static struct cds_lfs_stack_rcu s
;
171 void *thr_enqueuer(void *_count
)
173 unsigned long long *count
= _count
;
175 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
176 "enqueuer", pthread_self(), (unsigned long)gettid());
180 rcu_register_thread();
188 struct test
*node
= malloc(sizeof(*node
));
191 cds_lfs_node_init_rcu(&node
->list
);
192 /* No rcu read-side is needed for push */
193 cds_lfs_push_rcu(&s
, &node
->list
);
194 URCU_TLS(nr_successful_enqueues
)++;
196 if (caa_unlikely(wdelay
))
199 URCU_TLS(nr_enqueues
)++;
200 if (caa_unlikely(!test_duration_enqueue()))
204 rcu_unregister_thread();
206 count
[0] = URCU_TLS(nr_enqueues
);
207 count
[1] = URCU_TLS(nr_successful_enqueues
);
208 printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
209 "enqueues %llu successful_enqueues %llu\n",
210 pthread_self(), (unsigned long)gettid(),
211 URCU_TLS(nr_enqueues
), URCU_TLS(nr_successful_enqueues
));
217 void free_node_cb(struct rcu_head
*head
)
220 caa_container_of(head
, struct test
, rcu
);
224 void *thr_dequeuer(void *_count
)
226 unsigned long long *count
= _count
;
228 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
229 "dequeuer", pthread_self(), (unsigned long)gettid());
233 rcu_register_thread();
241 struct cds_lfs_node_rcu
*snode
;
244 snode
= cds_lfs_pop_rcu(&s
);
249 node
= caa_container_of(snode
, struct test
, list
);
250 call_rcu(&node
->rcu
, free_node_cb
);
251 URCU_TLS(nr_successful_dequeues
)++;
253 URCU_TLS(nr_dequeues
)++;
254 if (caa_unlikely(!test_duration_dequeue()))
256 if (caa_unlikely(rduration
))
257 loop_sleep(rduration
);
260 rcu_unregister_thread();
262 printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
263 "dequeues %llu, successful_dequeues %llu\n",
264 pthread_self(), (unsigned long)gettid(),
265 URCU_TLS(nr_dequeues
), URCU_TLS(nr_successful_dequeues
));
266 count
[0] = URCU_TLS(nr_dequeues
);
267 count
[1] = URCU_TLS(nr_successful_dequeues
);
271 void test_end(struct cds_lfs_stack_rcu
*s
, unsigned long long *nr_dequeues
)
273 struct cds_lfs_node_rcu
*snode
;
276 snode
= cds_lfs_pop_rcu(s
);
280 node
= caa_container_of(snode
, struct test
, list
);
287 void show_usage(int argc
, char **argv
)
289 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv
[0]);
290 printf(" [-d delay] (enqueuer period (in loops))");
291 printf(" [-c duration] (dequeuer period (in loops))");
292 printf(" [-v] (verbose output)");
293 printf(" [-a cpu#] [-a cpu#]... (affinity)");
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", pthread_self(), (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_lfs_init_rcu(&s
);
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(&s
, &end_dequeues
);
428 printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
429 tot_enqueues
, tot_dequeues
);
430 printf_verbose("total number of successful enqueues : %llu, "
431 "successful dequeues %llu\n",
432 tot_successful_enqueues
, tot_successful_dequeues
);
433 printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
435 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
436 "successful enqueues %12llu successful dequeues %12llu "
437 "end_dequeues %llu nr_ops %12llu\n",
438 argv
[0], duration
, nr_enqueuers
, wdelay
,
439 nr_dequeuers
, rduration
, tot_enqueues
, tot_dequeues
,
440 tot_successful_enqueues
,
441 tot_successful_dequeues
, end_dequeues
,
442 tot_enqueues
+ tot_dequeues
);
443 if (tot_successful_enqueues
!= tot_successful_dequeues
+ end_dequeues
)
444 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
445 "succ. dequeues + end dequeues %llu.\n",
446 tot_successful_enqueues
,
447 tot_successful_dequeues
+ end_dequeues
);
449 free_all_cpu_call_rcu_data();
450 free(count_enqueuer
);
451 free(count_dequeuer
);