--- /dev/null
+/*
+ * test_urcu_wfq.c
+ *
+ * Userspace RCU library - example RCU-based lock-free queue
+ *
+ * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define _GNU_SOURCE
+#include "../config.h"
+#include <stdio.h>
+#include <pthread.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <assert.h>
+#include <sys/syscall.h>
+#include <sched.h>
+#include <errno.h>
+
+#include <urcu/arch.h>
+
+/* hardcoded number of CPUs */
+#define NR_CPUS 16384
+
+#if defined(_syscall0)
+_syscall0(pid_t, gettid)
+#elif defined(__NR_gettid)
+static inline pid_t gettid(void)
+{
+ return syscall(__NR_gettid);
+}
+#else
+#warning "use pid as tid"
+static inline pid_t gettid(void)
+{
+ return getpid();
+}
+#endif
+
+#ifndef DYNAMIC_LINK_TEST
+#define _LGPL_SOURCE
+#endif
+#include <urcu.h>
+#include <urcu/rcuwfqueue.h>
+#include <urcu-defer.h>
+
+static volatile int test_go, test_stop;
+
+static unsigned long rduration;
+
+static unsigned long duration;
+
+/* read-side C.S. duration, in loops */
+static unsigned long wdelay;
+
+static inline void loop_sleep(unsigned long l)
+{
+ while(l-- != 0)
+ cpu_relax();
+}
+
+static int verbose_mode;
+
+#define printf_verbose(fmt, args...) \
+ do { \
+ if (verbose_mode) \
+ printf(fmt, args); \
+ } while (0)
+
+static unsigned int cpu_affinities[NR_CPUS];
+static unsigned int next_aff = 0;
+static int use_affinity = 0;
+
+pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+#ifndef HAVE_CPU_SET_T
+typedef unsigned long cpu_set_t;
+# define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
+# define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
+#endif
+
+static void set_affinity(void)
+{
+ cpu_set_t mask;
+ int cpu;
+ int ret;
+
+ if (!use_affinity)
+ return;
+
+#if HAVE_SCHED_SETAFFINITY
+ ret = pthread_mutex_lock(&affinity_mutex);
+ if (ret) {
+ perror("Error in pthread mutex lock");
+ exit(-1);
+ }
+ cpu = cpu_affinities[next_aff++];
+ ret = pthread_mutex_unlock(&affinity_mutex);
+ if (ret) {
+ perror("Error in pthread mutex unlock");
+ exit(-1);
+ }
+
+ CPU_ZERO(&mask);
+ CPU_SET(cpu, &mask);
+#if SCHED_SETAFFINITY_ARGS == 2
+ sched_setaffinity(0, &mask);
+#else
+ sched_setaffinity(0, sizeof(mask), &mask);
+#endif
+#endif /* HAVE_SCHED_SETAFFINITY */
+}
+
+/*
+ * returns 0 if test should end.
+ */
+static int test_duration_dequeue(void)
+{
+ return !test_stop;
+}
+
+static int test_duration_enqueue(void)
+{
+ return !test_stop;
+}
+
+static unsigned long long __thread nr_dequeues;
+static unsigned long long __thread nr_enqueues;
+
+static unsigned long long __thread nr_successful_dequeues;
+static unsigned long long __thread nr_successful_enqueues;
+
+static unsigned int nr_enqueuers;
+static unsigned int nr_dequeuers;
+
+static struct rcu_wfq_queue q;
+
+void *thr_enqueuer(void *_count)
+{
+ unsigned long long *count = _count;
+
+ printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
+ "enqueuer", pthread_self(), (unsigned long)gettid());
+
+ set_affinity();
+
+ rcu_register_thread();
+
+ while (!test_go)
+ {
+ }
+ smp_mb();
+
+ for (;;) {
+ struct rcu_wfq_node *node = malloc(sizeof(*node));
+ if (!node)
+ goto fail;
+ rcu_wfq_node_init(node);
+ rcu_wfq_enqueue(&q, node);
+ nr_successful_enqueues++;
+
+ if (unlikely(wdelay))
+ loop_sleep(wdelay);
+fail:
+ nr_enqueues++;
+ if (unlikely(!test_duration_enqueue()))
+ break;
+ }
+
+ rcu_unregister_thread();
+
+ count[0] = nr_enqueues;
+ count[1] = nr_successful_enqueues;
+ printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
+ "enqueues %llu successful_enqueues %llu\n",
+ pthread_self(), (unsigned long)gettid(), nr_enqueues,
+ nr_successful_enqueues);
+ return ((void*)1);
+
+}
+
+static void rcu_release_node(struct urcu_ref *ref)
+{
+ struct rcu_wfq_node *node = container_of(ref, struct rcu_wfq_node, ref);
+ defer_rcu(free, node);
+ //synchronize_rcu();
+ //free(node);
+}
+
+void *thr_dequeuer(void *_count)
+{
+ unsigned long long *count = _count;
+
+ printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
+ "dequeuer", pthread_self(), (unsigned long)gettid());
+
+ set_affinity();
+
+ rcu_defer_register_thread();
+ rcu_register_thread();
+
+ while (!test_go)
+ {
+ }
+ smp_mb();
+
+ for (;;) {
+ struct rcu_wfq_node *node =
+ rcu_wfq_dequeue_blocking(&q, rcu_release_node);
+
+ if (node) {
+ urcu_ref_put(&node->ref, rcu_release_node);
+ nr_successful_dequeues++;
+ }
+
+ nr_dequeues++;
+ if (unlikely(!test_duration_dequeue()))
+ break;
+ if (unlikely(rduration))
+ loop_sleep(rduration);
+ }
+
+ rcu_unregister_thread();
+ rcu_defer_unregister_thread();
+
+ printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
+ "dequeues %llu, successful_dequeues %llu\n",
+ pthread_self(), (unsigned long)gettid(), nr_dequeues,
+ nr_successful_dequeues);
+ count[0] = nr_dequeues;
+ count[1] = nr_successful_dequeues;
+ return ((void*)2);
+}
+
+static void release_node(struct urcu_ref *ref)
+{
+ struct rcu_wfq_node *node = container_of(ref, struct rcu_wfq_node, ref);
+ free(node);
+}
+
+void test_end(struct rcu_wfq_queue *q, unsigned long long *nr_dequeues)
+{
+ struct rcu_wfq_node *node;
+
+ do {
+ node = rcu_wfq_dequeue_blocking(q, release_node);
+ if (node) {
+ urcu_ref_put(&node->ref, release_node);
+ (*nr_dequeues)++;
+ }
+ } while (node);
+}
+
+void show_usage(int argc, char **argv)
+{
+ printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
+ printf(" [-d delay] (enqueuer period (in loops))");
+ printf(" [-c duration] (dequeuer period (in loops))");
+ printf(" [-v] (verbose output)");
+ printf(" [-a cpu#] [-a cpu#]... (affinity)");
+ printf("\n");
+}
+
+int main(int argc, char **argv)
+{
+ int err;
+ pthread_t *tid_enqueuer, *tid_dequeuer;
+ void *tret;
+ unsigned long long *count_enqueuer, *count_dequeuer;
+ unsigned long long tot_enqueues = 0, tot_dequeues = 0;
+ unsigned long long tot_successful_enqueues = 0,
+ tot_successful_dequeues = 0;
+ unsigned long long end_dequeues = 0;
+ int i, a;
+
+ if (argc < 4) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ err = sscanf(argv[1], "%u", &nr_dequeuers);
+ if (err != 1) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ err = sscanf(argv[2], "%u", &nr_enqueuers);
+ if (err != 1) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ err = sscanf(argv[3], "%lu", &duration);
+ if (err != 1) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ for (i = 4; i < argc; i++) {
+ if (argv[i][0] != '-')
+ continue;
+ switch (argv[i][1]) {
+ case 'a':
+ if (argc < i + 2) {
+ show_usage(argc, argv);
+ return -1;
+ }
+ a = atoi(argv[++i]);
+ cpu_affinities[next_aff++] = a;
+ use_affinity = 1;
+ printf_verbose("Adding CPU %d affinity\n", a);
+ break;
+ case 'c':
+ if (argc < i + 2) {
+ show_usage(argc, argv);
+ return -1;
+ }
+ rduration = atol(argv[++i]);
+ break;
+ case 'd':
+ if (argc < i + 2) {
+ show_usage(argc, argv);
+ return -1;
+ }
+ wdelay = atol(argv[++i]);
+ break;
+ case 'v':
+ verbose_mode = 1;
+ break;
+ }
+ }
+
+ printf_verbose("running test for %lu seconds, %u enqueuers, "
+ "%u dequeuers.\n",
+ duration, nr_enqueuers, nr_dequeuers);
+ printf_verbose("Writer delay : %lu loops.\n", rduration);
+ printf_verbose("Reader duration : %lu loops.\n", wdelay);
+ printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
+ "main", pthread_self(), (unsigned long)gettid());
+
+ tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
+ tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
+ count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
+ count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
+ rcu_wfq_init(&q);
+
+ next_aff = 0;
+
+ for (i = 0; i < nr_enqueuers; i++) {
+ err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
+ &count_enqueuer[2 * i]);
+ if (err != 0)
+ exit(1);
+ }
+ for (i = 0; i < nr_dequeuers; i++) {
+ err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
+ &count_dequeuer[2 * i]);
+ if (err != 0)
+ exit(1);
+ }
+
+ smp_mb();
+
+ test_go = 1;
+
+ for (i = 0; i < duration; i++) {
+ sleep(1);
+ if (verbose_mode)
+ write (1, ".", 1);
+ }
+
+ test_stop = 1;
+
+ for (i = 0; i < nr_enqueuers; i++) {
+ err = pthread_join(tid_enqueuer[i], &tret);
+ if (err != 0)
+ exit(1);
+ tot_enqueues += count_enqueuer[2 * i];
+ tot_successful_enqueues += count_enqueuer[2 * i + 1];
+ }
+ for (i = 0; i < nr_dequeuers; i++) {
+ err = pthread_join(tid_dequeuer[i], &tret);
+ if (err != 0)
+ exit(1);
+ tot_dequeues += count_dequeuer[2 * i];
+ tot_successful_dequeues += count_dequeuer[2 * i + 1];
+ }
+
+ test_end(&q, &end_dequeues);
+
+ printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
+ tot_enqueues, tot_dequeues);
+ printf_verbose("total number of successful enqueues : %llu, "
+ "successful dequeues %llu\n",
+ tot_successful_enqueues, tot_successful_dequeues);
+ printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
+ "nr_dequeuers %3u "
+ "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
+ "successful enqueues %12llu successful dequeues %12llu "
+ "end_dequeues %llu nr_ops %12llu\n",
+ argv[0], duration, nr_enqueuers, wdelay,
+ nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
+ tot_successful_enqueues,
+ tot_successful_dequeues, end_dequeues,
+ tot_enqueues + tot_dequeues);
+ if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
+ printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
+ "succ. dequeues + end dequeues %llu.\n",
+ tot_successful_enqueues,
+ tot_successful_dequeues + end_dequeues);
+
+ free(count_enqueuer);
+ free(count_dequeuer);
+ free(tid_enqueuer);
+ free(tid_dequeuer);
+ return 0;
+}
--- /dev/null
+/*
+ * rcuwfqueue.h
+ *
+ * Userspace RCU library - RCU Queue with Wait-Free Enqueue/Blocking Dequeue
+ *
+ * Copyright 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <urcu/urcu_ref.h>
+#include <assert.h>
+
+#if (!defined(_GNU_SOURCE) && !defined(_LGPL_SOURCE))
+#error "Dynamic loader LGPL wrappers not implemented yet"
+#endif
+
+/*
+ * RCU queue with wait-free enqueue/blocking dequeue using reference counting.
+ * Enqueue and dequeue operations hold a RCU read lock to deal with cmpxchg ABA
+ * problem. This implementation keeps a dummy head node to ensure we can always
+ * update the queue locklessly. Given that this is a queue, the dummy head node
+ * must always advance as we dequeue entries. Therefore, we keep a reference
+ * count on each entry we are dequeueing, so they can be kept as dummy head node
+ * until the next dequeue, at which point their reference count will be
+ * decremented.
+ */
+
+#define URCU_WFQ_PERMANENT_REF 128
+
+struct rcu_wfq_node {
+ struct rcu_wfq_node *next;
+ struct urcu_ref ref;
+};
+
+struct rcu_wfq_queue {
+ struct rcu_wfq_node *head, *tail;
+ struct rcu_wfq_node init; /* Dummy initialization node */
+};
+
+void rcu_wfq_node_init(struct rcu_wfq_node *node)
+{
+ node->next = NULL;
+ urcu_ref_init(&node->ref);
+}
+
+void rcu_wfq_init(struct rcu_wfq_queue *q)
+{
+ rcu_wfq_node_init(&q->init);
+ /* Make sure the initial node is never freed. */
+ urcu_ref_set(&q->init.ref, URCU_WFQ_PERMANENT_REF);
+ /* Set queue end */
+ q->head = q->tail = &q->init;
+}
+
+void rcu_wfq_enqueue(struct rcu_wfq_queue *q, struct rcu_wfq_node *node)
+{
+ struct rcu_wfq_node *old_tail;
+
+ urcu_ref_get(&node->ref);
+ /*
+ * uatomic_xchg() implicit memory barrier orders earlier stores to node
+ * (setting it to NULL and incrementing the refcount) before
+ * publication.
+ */
+ old_tail = uatomic_xchg(&q->tail, node);
+ /*
+ * At this point, dequeuers see a NULL old_tail->next, which indicates
+ * end of queue. The following store will append "node" to the queue
+ * from a dequeuer perspective.
+ */
+ STORE_SHARED(old_tail->next, node);
+}
+
+/*
+ * The entry returned by dequeue must be taken care of by doing a urcu_ref_put,
+ * which calls the release primitive when the reference count drops to zero. A
+ * grace period must be waited before performing the actual memory reclamation
+ * in the release primitive. The wfq node returned by dequeue must not be
+ * modified/re-used/freed until the reference count reaches zero and a grace
+ * period has elapsed (after the refcount reached 0).
+ *
+ * TODO: implement adaptative busy-wait and wait/wakeup scheme rather than busy
+ * loops. Better for UP.
+ */
+struct rcu_wfq_node *
+rcu_wfq_dequeue_blocking(struct rcu_wfq_queue *q,
+ void (*release)(struct urcu_ref *))
+{
+ for (;;) {
+ struct rcu_wfq_node *head, *next;
+
+ rcu_read_lock();
+ head = rcu_dereference(q->head);
+ next = rcu_dereference(head->next);
+ if (next) {
+ if (uatomic_cmpxchg(&q->head, head, next) == head) {
+ rcu_read_unlock();
+ urcu_ref_put(&head->ref, release);
+ return next;
+ } else {
+ /* Concurrently pushed, retry */
+ rcu_read_unlock();
+ continue;
+ }
+ } else {
+ /* Empty */
+ rcu_read_unlock();
+ return NULL;
+ }
+ }
+}
projects / urcu.git / commitdiff
