4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
41 #include <sys/queue.h>
43 #include <rte_common.h>
45 #include <rte_debug.h>
46 #include <rte_memory.h>
47 #include <rte_memzone.h>
48 #include <rte_launch.h>
49 #include <rte_cycles.h>
51 #include <rte_per_lcore.h>
52 #include <rte_lcore.h>
53 #include <rte_atomic.h>
54 #include <rte_branch_prediction.h>
55 #include <rte_mempool.h>
56 #include <rte_spinlock.h>
57 #include <rte_malloc.h>
65 * Each core get *n_keep* objects per bulk of *n_get_bulk*. Then,
66 * objects are put back in the pool per bulk of *n_put_bulk*.
68 * This sequence is done during TIME_S seconds.
70 * This test is done on the following configurations:
72 * - Cores configuration (*cores*)
74 * - One core with cache
75 * - Two cores with cache
76 * - Max. cores with cache
77 * - One core without cache
78 * - Two cores without cache
79 * - Max. cores without cache
80 * - One core with user-owned cache
81 * - Two cores with user-owned cache
82 * - Max. cores with user-owned cache
84 * - Bulk size (*n_get_bulk*, *n_put_bulk*)
86 * - Bulk get from 1 to 32
87 * - Bulk put from 1 to 32
89 * - Number of kept objects (*n_keep*)
97 #define MEMPOOL_ELT_SIZE 2048
99 #define MEMPOOL_SIZE ((rte_lcore_count()*(MAX_KEEP+RTE_MEMPOOL_CACHE_MAX_SIZE))-1)
101 #define LOG_ERR() printf("test failed at %s():%d\n", __func__, __LINE__)
102 #define RET_ERR() do { \
106 #define GOTO_ERR(var, label) do { \
112 static struct rte_mempool *mp;
113 static struct rte_mempool *mp_cache, *mp_nocache;
114 static int use_external_cache;
115 static unsigned external_cache_size = RTE_MEMPOOL_CACHE_MAX_SIZE;
117 static rte_atomic32_t synchro;
119 /* number of objects in one bulk operation (get or put) */
120 static unsigned n_get_bulk;
121 static unsigned n_put_bulk;
123 /* number of objects retrived from mempool before putting them back */
124 static unsigned n_keep;
126 /* number of enqueues / dequeues */
127 struct mempool_test_stats {
129 } __rte_cache_aligned;
131 static struct mempool_test_stats stats[RTE_MAX_LCORE];
134 * save the object number in the first 4 bytes of object data. All
135 * other bytes are set to 0.
138 my_obj_init(struct rte_mempool *mp, __attribute__((unused)) void *arg,
139 void *obj, unsigned i)
141 uint32_t *objnum = obj;
142 memset(obj, 0, mp->elt_size);
147 per_lcore_mempool_test(__attribute__((unused)) void *arg)
149 void *obj_table[MAX_KEEP];
151 unsigned lcore_id = rte_lcore_id();
153 uint64_t start_cycles, end_cycles;
154 uint64_t time_diff = 0, hz = rte_get_timer_hz();
155 struct rte_mempool_cache *cache;
157 if (use_external_cache) {
158 /* Create a user-owned mempool cache. */
159 cache = rte_mempool_cache_create(external_cache_size,
164 /* May be NULL if cache is disabled. */
165 cache = rte_mempool_default_cache(mp, lcore_id);
168 /* n_get_bulk and n_put_bulk must be divisors of n_keep */
169 if (((n_keep / n_get_bulk) * n_get_bulk) != n_keep)
171 if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)
174 stats[lcore_id].enq_count = 0;
176 /* wait synchro for slaves */
177 if (lcore_id != rte_get_master_lcore())
178 while (rte_atomic32_read(&synchro) == 0);
180 start_cycles = rte_get_timer_cycles();
182 while (time_diff/hz < TIME_S) {
183 for (i = 0; likely(i < (N/n_keep)); i++) {
184 /* get n_keep objects by bulk of n_bulk */
186 while (idx < n_keep) {
187 ret = rte_mempool_generic_get(mp,
191 if (unlikely(ret < 0)) {
192 rte_mempool_dump(stdout, mp);
193 /* in this case, objects are lost... */
199 /* put the objects back */
201 while (idx < n_keep) {
202 rte_mempool_generic_put(mp, &obj_table[idx],
208 end_cycles = rte_get_timer_cycles();
209 time_diff = end_cycles - start_cycles;
210 stats[lcore_id].enq_count += N;
214 if (use_external_cache) {
215 rte_mempool_cache_flush(cache, mp);
216 rte_mempool_cache_free(cache);
222 /* launch all the per-lcore test, and display the result */
224 launch_cores(unsigned cores)
229 unsigned cores_save = cores;
231 rte_atomic32_set(&synchro, 0);
234 memset(stats, 0, sizeof(stats));
236 printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u "
237 "n_put_bulk=%u n_keep=%u ",
239 external_cache_size : (unsigned) mp->cache_size,
240 cores, n_get_bulk, n_put_bulk, n_keep);
242 if (rte_mempool_avail_count(mp) != MEMPOOL_SIZE) {
243 printf("mempool is not full\n");
247 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
251 rte_eal_remote_launch(per_lcore_mempool_test,
255 /* start synchro and launch test on master */
256 rte_atomic32_set(&synchro, 1);
258 ret = per_lcore_mempool_test(NULL);
261 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
265 if (rte_eal_wait_lcore(lcore_id) < 0)
270 printf("per-lcore test returned -1\n");
275 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
276 rate += (stats[lcore_id].enq_count / TIME_S);
278 printf("rate_persec=%" PRIu64 "\n", rate);
283 /* for a given number of core, launch all test cases */
285 do_one_mempool_test(unsigned cores)
287 unsigned bulk_tab_get[] = { 1, 4, 32, 0 };
288 unsigned bulk_tab_put[] = { 1, 4, 32, 0 };
289 unsigned keep_tab[] = { 32, 128, 0 };
290 unsigned *get_bulk_ptr;
291 unsigned *put_bulk_ptr;
295 for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
296 for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr; put_bulk_ptr++) {
297 for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
299 n_get_bulk = *get_bulk_ptr;
300 n_put_bulk = *put_bulk_ptr;
302 ret = launch_cores(cores);
313 test_mempool_perf(void)
315 rte_atomic32_init(&synchro);
317 /* create a mempool (without cache) */
318 if (mp_nocache == NULL)
319 mp_nocache = rte_mempool_create("perf_test_nocache", MEMPOOL_SIZE,
320 MEMPOOL_ELT_SIZE, 0, 0,
324 if (mp_nocache == NULL)
327 /* create a mempool (with cache) */
328 if (mp_cache == NULL)
329 mp_cache = rte_mempool_create("perf_test_cache", MEMPOOL_SIZE,
331 RTE_MEMPOOL_CACHE_MAX_SIZE, 0,
335 if (mp_cache == NULL)
338 /* performance test with 1, 2 and max cores */
339 printf("start performance test (without cache)\n");
342 if (do_one_mempool_test(1) < 0)
345 if (do_one_mempool_test(2) < 0)
348 if (do_one_mempool_test(rte_lcore_count()) < 0)
351 /* performance test with 1, 2 and max cores */
352 printf("start performance test (with cache)\n");
355 if (do_one_mempool_test(1) < 0)
358 if (do_one_mempool_test(2) < 0)
361 if (do_one_mempool_test(rte_lcore_count()) < 0)
364 /* performance test with 1, 2 and max cores */
365 printf("start performance test (with user-owned cache)\n");
367 use_external_cache = 1;
369 if (do_one_mempool_test(1) < 0)
372 if (do_one_mempool_test(2) < 0)
375 if (do_one_mempool_test(rte_lcore_count()) < 0)
378 rte_mempool_list_dump(stdout);
383 REGISTER_TEST_COMMAND(mempool_perf_autotest, test_mempool_perf);