workqueue: implement benchmark
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Sun, 19 Oct 2014 15:31:58 +0000 (17:31 +0200)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Sun, 19 Oct 2014 15:31:58 +0000 (17:31 +0200)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
tests/benchmark/Makefile.am
tests/benchmark/test_urcu_workqueue.c [new file with mode: 0644]

index 85d454d829de1b64fb34db503e369d91ec294f10..1660a29e5833b4e7405c280b3ed0aec1ef202483 100644 (file)
@@ -20,7 +20,8 @@ noinst_PROGRAMS = test_urcu test_urcu_dynamic_link test_urcu_timing \
        test_urcu_wfq_dynlink test_urcu_wfs_dynlink \
        test_urcu_wfcq_dynlink \
        test_urcu_lfq_dynlink test_urcu_lfs_dynlink test_urcu_hash \
-       test_urcu_lfs_rcu_dynlink
+       test_urcu_lfs_rcu_dynlink \
+       test_urcu_workqueue
 
 URCU_COMMON_LIB=$(top_builddir)/liburcu-common.la
 URCU_LIB=$(top_builddir)/liburcu.la
@@ -189,6 +190,9 @@ test_urcu_wfs_dynlink_SOURCES = test_urcu_wfs.c
 test_urcu_wfs_dynlink_CFLAGS = -DDYNAMIC_LINK_TEST $(AM_CFLAGS)
 test_urcu_wfs_dynlink_LDADD = $(URCU_COMMON_LIB)
 
+test_urcu_workqueue_SOURCES = test_urcu_workqueue.c
+test_urcu_workqueue_LDADD = $(URCU_COMMON_LIB) $(URCU_LIB)
+
 test_urcu_hash_SOURCES = test_urcu_hash.c test_urcu_hash.h \
                test_urcu_hash_rw.c test_urcu_hash_unique.c
 test_urcu_hash_CFLAGS = -DRCU_QSBR $(AM_CFLAGS)
diff --git a/tests/benchmark/test_urcu_workqueue.c b/tests/benchmark/test_urcu_workqueue.c
new file mode 100644 (file)
index 0000000..9ac2350
--- /dev/null
@@ -0,0 +1,404 @@
+/*
+ * test_urcu_workqueue.c
+ *
+ * Userspace RCU library - workqueue test
+ *
+ * Copyright February 2010-2014 - 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 <errno.h>
+
+#include <urcu/arch.h>
+#include <urcu/tls-compat.h>
+#include <urcu/uatomic.h>
+#include "cpuset.h"
+#include "thread-id.h"
+
+/* hardcoded number of CPUs */
+#define NR_CPUS 16384
+
+#ifndef DYNAMIC_LINK_TEST
+#define _LGPL_SOURCE
+#endif
+#include <urcu.h>
+#include <urcu/wfstack.h>
+#include <urcu/workqueue-fifo.h>
+
+static volatile int test_go, test_stop_enqueue, test_stop_dequeue;
+
+static unsigned long work_loops;
+
+static unsigned long duration;
+
+static unsigned long dispatch_delay_loops;
+
+static inline void loop_sleep(unsigned long loops)
+{
+       while (loops-- != 0)
+               caa_cpu_relax();
+}
+
+static int verbose_mode;
+
+static int test_wait_empty;
+static int test_enqueue_stopped;
+
+#define printf_verbose(fmt, args...)           \
+       do {                                    \
+               if (verbose_mode)               \
+                       fprintf(stderr, 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;
+
+static void set_affinity(void)
+{
+#if HAVE_SCHED_SETAFFINITY
+       cpu_set_t mask;
+       int cpu, ret;
+#endif /* HAVE_SCHED_SETAFFINITY */
+
+       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_dequeue;
+}
+
+static int test_duration_enqueue(void)
+{
+       return !test_stop_enqueue;
+}
+
+static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
+
+static unsigned int nr_dispatchers;
+static unsigned int nr_workers;
+
+static struct urcu_workqueue workqueue;
+
+struct test_work {
+       struct urcu_work w;
+};
+
+static void *thr_dispatcher(void *_count)
+{
+       unsigned long long *count = _count;
+       bool was_nonempty;
+
+       printf_verbose("thread_begin %s, tid %lu\n",
+                       "dispatcher", urcu_get_thread_id());
+
+       set_affinity();
+
+       while (!test_go)
+       {
+       }
+       cmm_smp_mb();
+
+       for (;;) {
+               struct test_work *work = malloc(sizeof(*work));
+               if (!work)
+                       goto fail;
+               printf_verbose("queue work %p\n", work);
+               urcu_queue_work(&workqueue, &work->w);
+               URCU_TLS(nr_enqueues)++;
+
+               if (caa_unlikely(dispatch_delay_loops))
+                       loop_sleep(dispatch_delay_loops);
+fail:
+               if (caa_unlikely(!test_duration_enqueue()))
+                       break;
+       }
+
+       uatomic_inc(&test_enqueue_stopped);
+       count[0] = URCU_TLS(nr_enqueues);
+       printf_verbose("dispatcher thread_end, tid %lu, "
+                       "enqueues %llu\n",
+                       urcu_get_thread_id(),
+                       URCU_TLS(nr_enqueues));
+       return ((void*)1);
+}
+
+static void *thr_worker(void *_count)
+{
+       unsigned long long *count = _count;
+       unsigned int counter = 0;
+       struct urcu_worker worker;
+       int blocking = 1;
+
+       printf_verbose("thread_begin %s, tid %lu\n",
+                       "worker", urcu_get_thread_id());
+
+       set_affinity();
+
+       rcu_register_thread();
+       urcu_worker_init(&worker, URCU_WORKER_STEAL);
+       //urcu_worker_init(&worker, 0);
+       urcu_worker_register(&workqueue, &worker);
+
+       while (!test_go)
+       {
+       }
+       cmm_smp_mb();
+
+       for (;;) {
+               int batch_work_count = 0;
+
+               urcu_accept_work(&workqueue, &worker, blocking);
+               for (;;) {
+                       struct urcu_work *work;
+                       struct test_work *t;
+
+                       work = urcu_dequeue_work(&worker);
+                       if (!work)
+                               break;
+                       t = caa_container_of(work, struct test_work, w);
+                       printf_verbose("dequeue work %p\n", t);
+                       batch_work_count++;
+                       URCU_TLS(nr_dequeues)++;
+                       if (caa_unlikely(work_loops))
+                               loop_sleep(work_loops);
+                       free(t);
+               }
+               if (!test_duration_dequeue())
+                       blocking = 0;
+               if (caa_unlikely(!test_duration_dequeue()
+                               && !batch_work_count))
+                       break;
+       }
+end:
+       urcu_worker_unregister(&workqueue, &worker);
+       rcu_unregister_thread();
+
+       printf_verbose("worker thread_end, tid %lu, "
+                       "dequeues %llu\n",
+                       urcu_get_thread_id(),
+                       URCU_TLS(nr_dequeues));
+       count[0] = URCU_TLS(nr_dequeues);
+       return ((void*)2);
+}
+
+static void show_usage(int argc, char **argv)
+{
+       printf("Usage : %s nr_workers nr_dispatchers duration (s) <OPTIONS>\n",
+               argv[0]);
+       printf("OPTIONS:\n");
+       printf("        [-d delay] (dispatcher period (in loops))\n");
+       printf("        [-c duration] (worker period (in loops))\n");
+       printf("        [-v] (verbose output)\n");
+       printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
+       printf("        [-w] Wait for worker to empty stack\n");
+       printf("\n");
+}
+
+int main(int argc, char **argv)
+{
+       int err;
+       pthread_t *tid_dispatcher, *tid_worker;
+       void *tret;
+       unsigned long long *count_dispatcher, *count_worker;
+       unsigned long long tot_enqueues = 0, tot_dequeues = 0;
+       unsigned long long end_dequeues = 0;
+       int i, a, retval = 0;
+
+       if (argc < 4) {
+               show_usage(argc, argv);
+               return -1;
+       }
+
+       err = sscanf(argv[1], "%u", &nr_workers);
+       if (err != 1) {
+               show_usage(argc, argv);
+               return -1;
+       }
+
+       err = sscanf(argv[2], "%u", &nr_dispatchers);
+       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;
+                       }
+                       work_loops = atol(argv[++i]);
+                       break;
+               case 'd':
+                       if (argc < i + 2) {
+                               show_usage(argc, argv);
+                               return -1;
+                       }
+                       dispatch_delay_loops = atol(argv[++i]);
+                       break;
+               case 'v':
+                       verbose_mode = 1;
+                       break;
+               case 'w':
+                       test_wait_empty = 1;
+                       break;
+               }
+       }
+
+       printf_verbose("running test for %lu seconds, %u dispatchers, "
+                      "%u workers.\n",
+                      duration, nr_dispatchers, nr_workers);
+       if (test_wait_empty)
+               printf_verbose("Wait for workers to empty workqueue.\n");
+       printf_verbose("Work duration: %lu loops.\n", work_loops);
+       printf_verbose("Dispatcher arrival delay: %lu loops.\n", dispatch_delay_loops);
+       printf_verbose("thread %-6s, tid %lu\n",
+                       "main", urcu_get_thread_id());
+
+       tid_dispatcher = calloc(nr_dispatchers, sizeof(*tid_dispatcher));
+       tid_worker = calloc(nr_workers, sizeof(*tid_worker));
+       count_dispatcher = calloc(nr_dispatchers, sizeof(*count_dispatcher));
+       count_worker = calloc(nr_workers, sizeof(*count_worker));
+       urcu_workqueue_init(&workqueue);
+
+       next_aff = 0;
+
+       for (i = 0; i < nr_dispatchers; i++) {
+               err = pthread_create(&tid_dispatcher[i], NULL, thr_dispatcher,
+                                    &count_dispatcher[i]);
+               if (err != 0)
+                       exit(1);
+       }
+       for (i = 0; i < nr_workers; i++) {
+               err = pthread_create(&tid_worker[i], NULL, thr_worker,
+                                    &count_worker[i]);
+               if (err != 0)
+                       exit(1);
+       }
+
+       cmm_smp_mb();
+
+       test_go = 1;
+
+       for (i = 0; i < duration; i++) {
+               sleep(1);
+               if (verbose_mode)
+                       (void) write(1, ".", 1);
+       }
+
+       test_stop_enqueue = 1;
+       while (nr_dispatchers != uatomic_read(&test_enqueue_stopped)) {
+               sleep(1);
+       }
+
+       if (test_wait_empty) {
+               while (!cds_wfcq_empty(&workqueue.head, &workqueue.tail)) {
+                       sleep(1);
+               }
+       }
+       test_stop_dequeue = 1;
+
+       /* Send finish to all workers */
+       urcu_workqueue_wakeup_all(&workqueue);
+
+       for (i = 0; i < nr_dispatchers; i++) {
+               err = pthread_join(tid_dispatcher[i], &tret);
+               if (err != 0)
+                       exit(1);
+               tot_enqueues += count_dispatcher[i];
+       }
+       for (i = 0; i < nr_workers; i++) {
+               err = pthread_join(tid_worker[i], &tret);
+               if (err != 0)
+                       exit(1);
+               tot_dequeues += count_worker[i];
+       }
+
+       printf("SUMMARY %-25s testdur %4lu nr_dispatchers %3u dispatch_delay_loops %6lu "
+               "work_loops %lu nr_workers %3u "
+               "nr_enqueues %12llu nr_dequeues %12llu\n",
+               argv[0], duration, nr_dispatchers, dispatch_delay_loops, work_loops,
+               nr_workers, tot_enqueues, tot_dequeues);
+       free(count_dispatcher);
+       free(count_worker);
+       free(tid_dispatcher);
+       free(tid_worker);
+       return retval;
+}
This page took 0.029458 seconds and 4 git commands to generate.