4 * Copyright(c) 2010-2017 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.
38 #include <rte_mempool.h>
39 #include <rte_cycles.h>
40 #include <rte_common.h>
42 #include <rte_distributor.h>
43 #include <rte_pause.h>
45 #define ITER_POWER_CL 25 /* log 2 of how many iterations for Cache Line test */
46 #define ITER_POWER 21 /* log 2 of how many iterations we do when timing. */
48 #define BIG_BATCH 1024
50 /* static vars - zero initialized by default */
51 static volatile int quit;
52 static volatile unsigned worker_idx;
55 volatile unsigned handled_packets;
56 } __rte_cache_aligned;
57 struct worker_stats worker_stats[RTE_MAX_LCORE];
60 * worker thread used for testing the time to do a round-trip of a cache
61 * line between two cores and back again
64 flip_bit(volatile uint64_t *arg)
67 while (old_val != 2) {
76 * test case to time the number of cycles to round-trip a cache line between
77 * two cores and back again.
80 time_cache_line_switch(void)
82 /* allocate a full cache line for data, we use only first byte of it */
83 uint64_t data[RTE_CACHE_LINE_SIZE*3 / sizeof(uint64_t)];
85 unsigned i, slaveid = rte_get_next_lcore(rte_lcore_id(), 0, 0);
86 volatile uint64_t *pdata = &data[0];
88 rte_eal_remote_launch((lcore_function_t *)flip_bit, &data[0], slaveid);
92 const uint64_t start_time = rte_rdtsc();
93 for (i = 0; i < (1 << ITER_POWER_CL); i++) {
98 const uint64_t end_time = rte_rdtsc();
103 rte_eal_wait_lcore(slaveid);
104 printf("==== Cache line switch test ===\n");
105 printf("Time for %u iterations = %"PRIu64" ticks\n", (1<<ITER_POWER_CL),
106 end_time-start_time);
107 printf("Ticks per iteration = %"PRIu64"\n\n",
108 (end_time-start_time) >> ITER_POWER_CL);
112 * returns the total count of the number of packets handled by the worker
113 * functions given below.
116 total_packet_count(void)
118 unsigned i, count = 0;
119 for (i = 0; i < worker_idx; i++)
120 count += worker_stats[i].handled_packets;
124 /* resets the packet counts for a new test */
126 clear_packet_count(void)
128 memset(&worker_stats, 0, sizeof(worker_stats));
132 * This is the basic worker function for performance tests.
133 * it does nothing but return packets and count them.
136 handle_work(void *arg)
138 struct rte_distributor *d = arg;
139 unsigned int count = 0;
140 unsigned int num = 0;
142 unsigned int id = __sync_fetch_and_add(&worker_idx, 1);
143 struct rte_mbuf *buf[8] __rte_cache_aligned;
145 for (i = 0; i < 8; i++)
148 num = rte_distributor_get_pkt(d, id, buf, buf, num);
150 worker_stats[id].handled_packets += num;
152 num = rte_distributor_get_pkt(d, id, buf, buf, num);
154 worker_stats[id].handled_packets += num;
156 rte_distributor_return_pkt(d, id, buf, num);
161 * This basic performance test just repeatedly sends in 32 packets at a time
162 * to the distributor and verifies at the end that we got them all in the worker
163 * threads and finally how long per packet the processing took.
166 perf_test(struct rte_distributor *d, struct rte_mempool *p)
170 struct rte_mbuf *bufs[BURST];
172 clear_packet_count();
173 if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
174 printf("Error getting mbufs from pool\n");
177 /* ensure we have different hash value for each pkt */
178 for (i = 0; i < BURST; i++)
179 bufs[i]->hash.usr = i;
182 for (i = 0; i < (1<<ITER_POWER); i++)
183 rte_distributor_process(d, bufs, BURST);
188 rte_distributor_process(d, NULL, 0);
189 } while (total_packet_count() < (BURST << ITER_POWER));
191 rte_distributor_clear_returns(d);
193 printf("Time per burst: %"PRIu64"\n", (end - start) >> ITER_POWER);
194 printf("Time per packet: %"PRIu64"\n\n",
195 ((end - start) >> ITER_POWER)/BURST);
196 rte_mempool_put_bulk(p, (void *)bufs, BURST);
198 for (i = 0; i < rte_lcore_count() - 1; i++)
199 printf("Worker %u handled %u packets\n", i,
200 worker_stats[i].handled_packets);
201 printf("Total packets: %u (%x)\n", total_packet_count(),
202 total_packet_count());
203 printf("=== Perf test done ===\n\n");
208 /* Useful function which ensures that all worker functions terminate */
210 quit_workers(struct rte_distributor *d, struct rte_mempool *p)
212 const unsigned int num_workers = rte_lcore_count() - 1;
214 struct rte_mbuf *bufs[RTE_MAX_LCORE];
216 rte_mempool_get_bulk(p, (void *)bufs, num_workers);
219 for (i = 0; i < num_workers; i++)
220 bufs[i]->hash.usr = i << 1;
221 rte_distributor_process(d, bufs, num_workers);
223 rte_mempool_put_bulk(p, (void *)bufs, num_workers);
225 rte_distributor_process(d, NULL, 0);
226 rte_eal_mp_wait_lcore();
232 test_distributor_perf(void)
234 static struct rte_distributor *ds;
235 static struct rte_distributor *db;
236 static struct rte_mempool *p;
238 if (rte_lcore_count() < 2) {
239 printf("ERROR: not enough cores to test distributor\n");
243 /* first time how long it takes to round-trip a cache line */
244 time_cache_line_switch();
247 ds = rte_distributor_create("Test_perf", rte_socket_id(),
248 rte_lcore_count() - 1,
249 RTE_DIST_ALG_SINGLE);
251 printf("Error creating distributor\n");
255 rte_distributor_clear_returns(ds);
259 db = rte_distributor_create("Test_burst", rte_socket_id(),
260 rte_lcore_count() - 1,
263 printf("Error creating burst distributor\n");
267 rte_distributor_clear_returns(db);
270 const unsigned nb_bufs = (511 * rte_lcore_count()) < BIG_BATCH ?
271 (BIG_BATCH * 2) - 1 : (511 * rte_lcore_count());
273 p = rte_pktmbuf_pool_create("DPT_MBUF_POOL", nb_bufs, BURST,
274 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
276 printf("Error creating mempool\n");
281 printf("=== Performance test of distributor (single mode) ===\n");
282 rte_eal_mp_remote_launch(handle_work, ds, SKIP_MASTER);
283 if (perf_test(ds, p) < 0)
287 printf("=== Performance test of distributor (burst mode) ===\n");
288 rte_eal_mp_remote_launch(handle_work, db, SKIP_MASTER);
289 if (perf_test(db, p) < 0)
296 REGISTER_TEST_COMMAND(distributor_perf_autotest, test_distributor_perf);