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.
36 #ifdef RTE_LIBRTE_DISTRIBUTOR
39 #include <rte_cycles.h>
40 #include <rte_distributor.h>
42 #define ITER_POWER 20 /* log 2 of how many iterations we do when timing. */
44 #define BIG_BATCH 1024
46 /* static vars - zero initialized by default */
47 static volatile int quit;
48 static volatile unsigned worker_idx;
51 volatile unsigned handled_packets;
52 } __rte_cache_aligned;
53 struct worker_stats worker_stats[RTE_MAX_LCORE];
55 /* worker thread used for testing the time to do a round-trip of a cache
56 * line between two cores and back again
59 flip_bit(volatile uint64_t *arg)
62 while (old_val != 2) {
70 /* test case to time the number of cycles to round-trip a cache line between
71 * two cores and back again.
74 time_cache_line_switch(void)
76 /* allocate a full cache line for data, we use only first byte of it */
77 uint64_t data[CACHE_LINE_SIZE*3 / sizeof(uint64_t)];
79 unsigned i, slaveid = rte_get_next_lcore(rte_lcore_id(), 0, 0);
80 volatile uint64_t *pdata = &data[0];
82 rte_eal_remote_launch((lcore_function_t *)flip_bit, &data[0], slaveid);
86 const uint64_t start_time = rte_rdtsc();
87 for (i = 0; i < (1 << ITER_POWER); i++) {
92 const uint64_t end_time = rte_rdtsc();
97 rte_eal_wait_lcore(slaveid);
98 printf("==== Cache line switch test ===\n");
99 printf("Time for %u iterations = %"PRIu64" ticks\n", (1<<ITER_POWER),
100 end_time-start_time);
101 printf("Ticks per iteration = %"PRIu64"\n\n",
102 (end_time-start_time) >> ITER_POWER);
105 /* returns the total count of the number of packets handled by the worker
106 * functions given below.
109 total_packet_count(void)
111 unsigned i, count = 0;
112 for (i = 0; i < worker_idx; i++)
113 count += worker_stats[i].handled_packets;
117 /* resets the packet counts for a new test */
119 clear_packet_count(void)
121 memset(&worker_stats, 0, sizeof(worker_stats));
124 /* this is the basic worker function for performance tests.
125 * it does nothing but return packets and count them.
128 handle_work(void *arg)
130 struct rte_mbuf *pkt = NULL;
131 struct rte_distributor *d = arg;
133 unsigned id = __sync_fetch_and_add(&worker_idx, 1);
135 pkt = rte_distributor_get_pkt(d, id, NULL);
137 worker_stats[id].handled_packets++, count++;
138 pkt = rte_distributor_get_pkt(d, id, pkt);
140 worker_stats[id].handled_packets++, count++;
141 rte_distributor_return_pkt(d, id, pkt);
145 /* this basic performance test just repeatedly sends in 32 packets at a time
146 * to the distributor and verifies at the end that we got them all in the worker
147 * threads and finally how long per packet the processing took.
150 perf_test(struct rte_distributor *d, struct rte_mempool *p)
154 struct rte_mbuf *bufs[BURST];
156 clear_packet_count();
157 if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
158 printf("Error getting mbufs from pool\n");
161 /* ensure we have different hash value for each pkt */
162 for (i = 0; i < BURST; i++)
163 bufs[i]->pkt.hash.rss = i;
166 for (i = 0; i < (1<<ITER_POWER); i++)
167 rte_distributor_process(d, bufs, BURST);
172 rte_distributor_process(d, NULL, 0);
173 } while (total_packet_count() < (BURST << ITER_POWER));
175 printf("=== Performance test of distributor ===\n");
176 printf("Time per burst: %"PRIu64"\n", (end - start) >> ITER_POWER);
177 printf("Time per packet: %"PRIu64"\n\n",
178 ((end - start) >> ITER_POWER)/BURST);
179 rte_mempool_put_bulk(p, (void *)bufs, BURST);
181 for (i = 0; i < rte_lcore_count() - 1; i++)
182 printf("Worker %u handled %u packets\n", i,
183 worker_stats[i].handled_packets);
184 printf("Total packets: %u (%x)\n", total_packet_count(),
185 total_packet_count());
186 printf("=== Perf test done ===\n\n");
191 /* Useful function which ensures that all worker functions terminate */
193 quit_workers(struct rte_distributor *d, struct rte_mempool *p)
195 const unsigned num_workers = rte_lcore_count() - 1;
197 struct rte_mbuf *bufs[RTE_MAX_LCORE];
198 rte_mempool_get_bulk(p, (void *)bufs, num_workers);
201 for (i = 0; i < num_workers; i++)
202 bufs[i]->pkt.hash.rss = i << 1;
203 rte_distributor_process(d, bufs, num_workers);
205 rte_mempool_put_bulk(p, (void *)bufs, num_workers);
207 rte_distributor_process(d, NULL, 0);
208 rte_eal_mp_wait_lcore();
213 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
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_mempool_create("DPT_MBUF_POOL", nb_bufs,
246 sizeof(struct rte_pktmbuf_pool_private),
247 rte_pktmbuf_pool_init, NULL,
248 rte_pktmbuf_init, NULL,
251 printf("Error creating mempool\n");
256 rte_eal_mp_remote_launch(handle_work, d, SKIP_MASTER);
257 if (perf_test(d, p) < 0)
269 test_distributor_perf(void)
271 printf("Distributor is not enabled in configuration\n");