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.
38 #include <rte_mempool.h>
39 #include <rte_cycles.h>
40 #include <rte_common.h>
42 #include <rte_distributor.h>
44 #define ITER_POWER 20 /* log 2 of how many iterations we do when timing. */
46 #define BIG_BATCH 1024
48 /* static vars - zero initialized by default */
49 static volatile int quit;
50 static volatile unsigned worker_idx;
53 volatile unsigned handled_packets;
54 } __rte_cache_aligned;
55 struct worker_stats worker_stats[RTE_MAX_LCORE];
57 /* worker thread used for testing the time to do a round-trip of a cache
58 * line between two cores and back again
61 flip_bit(volatile uint64_t *arg)
64 while (old_val != 2) {
72 /* test case to time the number of cycles to round-trip a cache line between
73 * two cores and back again.
76 time_cache_line_switch(void)
78 /* allocate a full cache line for data, we use only first byte of it */
79 uint64_t data[RTE_CACHE_LINE_SIZE*3 / sizeof(uint64_t)];
81 unsigned i, slaveid = rte_get_next_lcore(rte_lcore_id(), 0, 0);
82 volatile uint64_t *pdata = &data[0];
84 rte_eal_remote_launch((lcore_function_t *)flip_bit, &data[0], slaveid);
88 const uint64_t start_time = rte_rdtsc();
89 for (i = 0; i < (1 << ITER_POWER); i++) {
94 const uint64_t end_time = rte_rdtsc();
99 rte_eal_wait_lcore(slaveid);
100 printf("==== Cache line switch test ===\n");
101 printf("Time for %u iterations = %"PRIu64" ticks\n", (1<<ITER_POWER),
102 end_time-start_time);
103 printf("Ticks per iteration = %"PRIu64"\n\n",
104 (end_time-start_time) >> ITER_POWER);
107 /* returns the total count of the number of packets handled by the worker
108 * functions given below.
111 total_packet_count(void)
113 unsigned i, count = 0;
114 for (i = 0; i < worker_idx; i++)
115 count += worker_stats[i].handled_packets;
119 /* resets the packet counts for a new test */
121 clear_packet_count(void)
123 memset(&worker_stats, 0, sizeof(worker_stats));
126 /* this is the basic worker function for performance tests.
127 * it does nothing but return packets and count them.
130 handle_work(void *arg)
132 struct rte_mbuf *pkt = NULL;
133 struct rte_distributor *d = arg;
135 unsigned id = __sync_fetch_and_add(&worker_idx, 1);
137 pkt = rte_distributor_get_pkt(d, id, NULL);
139 worker_stats[id].handled_packets++, count++;
140 pkt = rte_distributor_get_pkt(d, id, pkt);
142 worker_stats[id].handled_packets++, count++;
143 rte_distributor_return_pkt(d, id, pkt);
147 /* this basic performance test just repeatedly sends in 32 packets at a time
148 * to the distributor and verifies at the end that we got them all in the worker
149 * threads and finally how long per packet the processing took.
152 perf_test(struct rte_distributor *d, struct rte_mempool *p)
156 struct rte_mbuf *bufs[BURST];
158 clear_packet_count();
159 if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
160 printf("Error getting mbufs from pool\n");
163 /* ensure we have different hash value for each pkt */
164 for (i = 0; i < BURST; i++)
165 bufs[i]->hash.usr = i;
168 for (i = 0; i < (1<<ITER_POWER); i++)
169 rte_distributor_process(d, bufs, BURST);
174 rte_distributor_process(d, NULL, 0);
175 } while (total_packet_count() < (BURST << ITER_POWER));
177 printf("=== Performance test of distributor ===\n");
178 printf("Time per burst: %"PRIu64"\n", (end - start) >> ITER_POWER);
179 printf("Time per packet: %"PRIu64"\n\n",
180 ((end - start) >> ITER_POWER)/BURST);
181 rte_mempool_put_bulk(p, (void *)bufs, BURST);
183 for (i = 0; i < rte_lcore_count() - 1; i++)
184 printf("Worker %u handled %u packets\n", i,
185 worker_stats[i].handled_packets);
186 printf("Total packets: %u (%x)\n", total_packet_count(),
187 total_packet_count());
188 printf("=== Perf test done ===\n\n");
193 /* Useful function which ensures that all worker functions terminate */
195 quit_workers(struct rte_distributor *d, struct rte_mempool *p)
197 const unsigned num_workers = rte_lcore_count() - 1;
199 struct rte_mbuf *bufs[RTE_MAX_LCORE];
200 rte_mempool_get_bulk(p, (void *)bufs, num_workers);
203 for (i = 0; i < num_workers; i++)
204 bufs[i]->hash.usr = i << 1;
205 rte_distributor_process(d, bufs, num_workers);
207 rte_mempool_put_bulk(p, (void *)bufs, num_workers);
209 rte_distributor_process(d, NULL, 0);
210 rte_eal_mp_wait_lcore();
216 test_distributor_perf(void)
218 static struct rte_distributor *d;
219 static struct rte_mempool *p;
221 if (rte_lcore_count() < 2) {
222 printf("ERROR: not enough cores to test distributor\n");
226 /* first time how long it takes to round-trip a cache line */
227 time_cache_line_switch();
230 d = rte_distributor_create("Test_perf", rte_socket_id(),
231 rte_lcore_count() - 1);
233 printf("Error creating distributor\n");
237 rte_distributor_flush(d);
238 rte_distributor_clear_returns(d);
241 const unsigned nb_bufs = (511 * rte_lcore_count()) < BIG_BATCH ?
242 (BIG_BATCH * 2) - 1 : (511 * rte_lcore_count());
244 p = rte_pktmbuf_pool_create("DPT_MBUF_POOL", nb_bufs, BURST,
245 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
247 printf("Error creating mempool\n");
252 rte_eal_mp_remote_launch(handle_work, d, SKIP_MASTER);
253 if (perf_test(d, p) < 0)
260 static struct test_command distributor_perf_cmd = {
261 .command = "distributor_perf_autotest",
262 .callback = test_distributor_perf,
264 REGISTER_TEST_COMMAND(distributor_perf_cmd);