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 <rte_ethdev.h>
42 #include <rte_cycles.h>
43 #include <rte_malloc.h>
44 #include <rte_debug.h>
45 #include <rte_distributor.h>
47 #define RX_RING_SIZE 256
48 #define TX_RING_SIZE 512
49 #define NUM_MBUFS ((64*1024)-1)
50 #define MBUF_CACHE_SIZE 250
52 #define RTE_RING_SZ 1024
54 /* uncommnet below line to enable debug logs */
58 #define LOG_LEVEL RTE_LOG_DEBUG
59 #define LOG_DEBUG(log_type, fmt, args...) do { \
60 RTE_LOG(DEBUG, log_type, fmt, ##args); \
63 #define LOG_LEVEL RTE_LOG_INFO
64 #define LOG_DEBUG(log_type, fmt, args...) do {} while (0)
67 #define RTE_LOGTYPE_DISTRAPP RTE_LOGTYPE_USER1
69 /* mask of enabled ports */
70 static uint32_t enabled_port_mask;
71 volatile uint8_t quit_signal;
72 volatile uint8_t quit_signal_rx;
74 static volatile struct app_stats {
77 uint64_t returned_pkts;
78 uint64_t enqueued_pkts;
79 } rx __rte_cache_aligned;
82 uint64_t dequeue_pkts;
84 } tx __rte_cache_aligned;
87 static const struct rte_eth_conf port_conf_default = {
89 .mq_mode = ETH_MQ_RX_RSS,
90 .max_rx_pkt_len = ETHER_MAX_LEN,
93 .mq_mode = ETH_MQ_TX_NONE,
97 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
98 ETH_RSS_TCP | ETH_RSS_SCTP,
103 struct output_buffer {
105 struct rte_mbuf *mbufs[BURST_SIZE];
109 * Initialises a given port using global settings and with the rx buffers
110 * coming from the mbuf_pool passed as parameter
113 port_init(uint8_t port, struct rte_mempool *mbuf_pool)
115 struct rte_eth_conf port_conf = port_conf_default;
116 const uint16_t rxRings = 1, txRings = rte_lcore_count() - 1;
120 if (port >= rte_eth_dev_count())
123 retval = rte_eth_dev_configure(port, rxRings, txRings, &port_conf);
127 for (q = 0; q < rxRings; q++) {
128 retval = rte_eth_rx_queue_setup(port, q, RX_RING_SIZE,
129 rte_eth_dev_socket_id(port),
135 for (q = 0; q < txRings; q++) {
136 retval = rte_eth_tx_queue_setup(port, q, TX_RING_SIZE,
137 rte_eth_dev_socket_id(port),
143 retval = rte_eth_dev_start(port);
147 struct rte_eth_link link;
148 rte_eth_link_get_nowait(port, &link);
149 if (!link.link_status) {
151 rte_eth_link_get_nowait(port, &link);
154 if (!link.link_status) {
155 printf("Link down on port %"PRIu8"\n", port);
159 struct ether_addr addr;
160 rte_eth_macaddr_get(port, &addr);
161 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
162 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
164 addr.addr_bytes[0], addr.addr_bytes[1],
165 addr.addr_bytes[2], addr.addr_bytes[3],
166 addr.addr_bytes[4], addr.addr_bytes[5]);
168 rte_eth_promiscuous_enable(port);
173 struct lcore_params {
175 struct rte_distributor *d;
177 struct rte_mempool *mem_pool;
181 quit_workers(struct rte_distributor *d, struct rte_mempool *p)
183 const unsigned num_workers = rte_lcore_count() - 2;
185 struct rte_mbuf *bufs[num_workers];
186 rte_mempool_get_bulk(p, (void *)bufs, num_workers);
188 for (i = 0; i < num_workers; i++)
189 bufs[i]->hash.rss = i << 1;
191 rte_distributor_process(d, bufs, num_workers);
192 rte_mempool_put_bulk(p, (void *)bufs, num_workers);
196 lcore_rx(struct lcore_params *p)
198 struct rte_distributor *d = p->d;
199 struct rte_mempool *mem_pool = p->mem_pool;
200 struct rte_ring *r = p->r;
201 const uint8_t nb_ports = rte_eth_dev_count();
202 const int socket_id = rte_socket_id();
205 for (port = 0; port < nb_ports; port++) {
206 /* skip ports that are not enabled */
207 if ((enabled_port_mask & (1 << port)) == 0)
210 if (rte_eth_dev_socket_id(port) > 0 &&
211 rte_eth_dev_socket_id(port) != socket_id)
212 printf("WARNING, port %u is on remote NUMA node to "
213 "RX thread.\n\tPerformance will not "
214 "be optimal.\n", port);
217 printf("\nCore %u doing packet RX.\n", rte_lcore_id());
219 while (!quit_signal_rx) {
221 /* skip ports that are not enabled */
222 if ((enabled_port_mask & (1 << port)) == 0) {
223 if (++port == nb_ports)
227 struct rte_mbuf *bufs[BURST_SIZE*2];
228 const uint16_t nb_rx = rte_eth_rx_burst(port, 0, bufs,
230 app_stats.rx.rx_pkts += nb_rx;
232 rte_distributor_process(d, bufs, nb_rx);
233 const uint16_t nb_ret = rte_distributor_returned_pkts(d,
235 app_stats.rx.returned_pkts += nb_ret;
236 if (unlikely(nb_ret == 0))
239 uint16_t sent = rte_ring_enqueue_burst(r, (void *)bufs, nb_ret);
240 app_stats.rx.enqueued_pkts += sent;
241 if (unlikely(sent < nb_ret)) {
242 LOG_DEBUG(DISTRAPP, "%s:Packet loss due to full ring\n", __func__);
243 while (sent < nb_ret)
244 rte_pktmbuf_free(bufs[sent++]);
246 if (++port == nb_ports)
249 rte_distributor_process(d, NULL, 0);
250 /* flush distributor to bring to known state */
251 rte_distributor_flush(d);
252 /* set worker & tx threads quit flag */
255 * worker threads may hang in get packet as
256 * distributor process is not running, just make sure workers
257 * get packets till quit_signal is actually been
258 * received and they gracefully shutdown
260 quit_workers(d, mem_pool);
261 /* rx thread should quit at last */
266 flush_one_port(struct output_buffer *outbuf, uint8_t outp)
268 unsigned nb_tx = rte_eth_tx_burst(outp, 0, outbuf->mbufs,
270 app_stats.tx.tx_pkts += nb_tx;
272 if (unlikely(nb_tx < outbuf->count)) {
273 LOG_DEBUG(DISTRAPP, "%s:Packet loss with tx_burst\n", __func__);
275 rte_pktmbuf_free(outbuf->mbufs[nb_tx]);
276 } while (++nb_tx < outbuf->count);
282 flush_all_ports(struct output_buffer *tx_buffers, uint8_t nb_ports)
285 for (outp = 0; outp < nb_ports; outp++) {
286 /* skip ports that are not enabled */
287 if ((enabled_port_mask & (1 << outp)) == 0)
290 if (tx_buffers[outp].count == 0)
293 flush_one_port(&tx_buffers[outp], outp);
298 lcore_tx(struct rte_ring *in_r)
300 static struct output_buffer tx_buffers[RTE_MAX_ETHPORTS];
301 const uint8_t nb_ports = rte_eth_dev_count();
302 const int socket_id = rte_socket_id();
305 for (port = 0; port < nb_ports; port++) {
306 /* skip ports that are not enabled */
307 if ((enabled_port_mask & (1 << port)) == 0)
310 if (rte_eth_dev_socket_id(port) > 0 &&
311 rte_eth_dev_socket_id(port) != socket_id)
312 printf("WARNING, port %u is on remote NUMA node to "
313 "TX thread.\n\tPerformance will not "
314 "be optimal.\n", port);
317 printf("\nCore %u doing packet TX.\n", rte_lcore_id());
318 while (!quit_signal) {
320 for (port = 0; port < nb_ports; port++) {
321 /* skip ports that are not enabled */
322 if ((enabled_port_mask & (1 << port)) == 0)
325 struct rte_mbuf *bufs[BURST_SIZE];
326 const uint16_t nb_rx = rte_ring_dequeue_burst(in_r,
327 (void *)bufs, BURST_SIZE);
328 app_stats.tx.dequeue_pkts += nb_rx;
330 /* if we get no traffic, flush anything we have */
331 if (unlikely(nb_rx == 0)) {
332 flush_all_ports(tx_buffers, nb_ports);
336 /* for traffic we receive, queue it up for transmit */
338 _mm_prefetch((void *)bufs[0], 0);
339 _mm_prefetch((void *)bufs[1], 0);
340 _mm_prefetch((void *)bufs[2], 0);
341 for (i = 0; i < nb_rx; i++) {
342 struct output_buffer *outbuf;
344 _mm_prefetch((void *)bufs[i + 3], 0);
346 * workers should update in_port to hold the
349 outp = bufs[i]->port;
350 /* skip ports that are not enabled */
351 if ((enabled_port_mask & (1 << outp)) == 0)
354 outbuf = &tx_buffers[outp];
355 outbuf->mbufs[outbuf->count++] = bufs[i];
356 if (outbuf->count == BURST_SIZE)
357 flush_one_port(outbuf, outp);
365 int_handler(int sig_num)
367 printf("Exiting on signal %d\n", sig_num);
368 /* set quit flag for rx thread to exit */
375 struct rte_eth_stats eth_stats;
378 printf("\nRX thread stats:\n");
379 printf(" - Received: %"PRIu64"\n", app_stats.rx.rx_pkts);
380 printf(" - Processed: %"PRIu64"\n", app_stats.rx.returned_pkts);
381 printf(" - Enqueued: %"PRIu64"\n", app_stats.rx.enqueued_pkts);
383 printf("\nTX thread stats:\n");
384 printf(" - Dequeued: %"PRIu64"\n", app_stats.tx.dequeue_pkts);
385 printf(" - Transmitted: %"PRIu64"\n", app_stats.tx.tx_pkts);
387 for (i = 0; i < rte_eth_dev_count(); i++) {
388 rte_eth_stats_get(i, ð_stats);
389 printf("\nPort %u stats:\n", i);
390 printf(" - Pkts in: %"PRIu64"\n", eth_stats.ipackets);
391 printf(" - Pkts out: %"PRIu64"\n", eth_stats.opackets);
392 printf(" - In Errs: %"PRIu64"\n", eth_stats.ierrors);
393 printf(" - Out Errs: %"PRIu64"\n", eth_stats.oerrors);
394 printf(" - Mbuf Errs: %"PRIu64"\n", eth_stats.rx_nombuf);
399 lcore_worker(struct lcore_params *p)
401 struct rte_distributor *d = p->d;
402 const unsigned id = p->worker_id;
404 * for single port, xor_val will be zero so we won't modify the output
405 * port, otherwise we send traffic from 0 to 1, 2 to 3, and vice versa
407 const unsigned xor_val = (rte_eth_dev_count() > 1);
408 struct rte_mbuf *buf = NULL;
410 printf("\nCore %u acting as worker core.\n", rte_lcore_id());
411 while (!quit_signal) {
412 buf = rte_distributor_get_pkt(d, id, buf);
413 buf->port ^= xor_val;
420 print_usage(const char *prgname)
422 printf("%s [EAL options] -- -p PORTMASK\n"
423 " -p PORTMASK: hexadecimal bitmask of ports to configure\n",
428 parse_portmask(const char *portmask)
433 /* parse hexadecimal string */
434 pm = strtoul(portmask, &end, 16);
435 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
444 /* Parse the argument given in the command line of the application */
446 parse_args(int argc, char **argv)
451 char *prgname = argv[0];
452 static struct option lgopts[] = {
458 while ((opt = getopt_long(argc, argvopt, "p:",
459 lgopts, &option_index)) != EOF) {
464 enabled_port_mask = parse_portmask(optarg);
465 if (enabled_port_mask == 0) {
466 printf("invalid portmask\n");
467 print_usage(prgname);
473 print_usage(prgname);
479 print_usage(prgname);
483 argv[optind-1] = prgname;
485 optind = 0; /* reset getopt lib */
489 /* Main function, does initialization and calls the per-lcore functions */
491 main(int argc, char *argv[])
493 struct rte_mempool *mbuf_pool;
494 struct rte_distributor *d;
495 struct rte_ring *output_ring;
496 unsigned lcore_id, worker_id = 0;
499 uint8_t nb_ports_available;
501 /* catch ctrl-c so we can print on exit */
502 signal(SIGINT, int_handler);
505 int ret = rte_eal_init(argc, argv);
507 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
511 /* parse application arguments (after the EAL ones) */
512 ret = parse_args(argc, argv);
514 rte_exit(EXIT_FAILURE, "Invalid distributor parameters\n");
516 if (rte_lcore_count() < 3)
517 rte_exit(EXIT_FAILURE, "Error, This application needs at "
518 "least 3 logical cores to run:\n"
519 "1 lcore for packet RX and distribution\n"
520 "1 lcore for packet TX\n"
521 "and at least 1 lcore for worker threads\n");
523 nb_ports = rte_eth_dev_count();
525 rte_exit(EXIT_FAILURE, "Error: no ethernet ports detected\n");
526 if (nb_ports != 1 && (nb_ports & 1))
527 rte_exit(EXIT_FAILURE, "Error: number of ports must be even, except "
528 "when using a single port\n");
530 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
531 NUM_MBUFS * nb_ports, MBUF_CACHE_SIZE, 0,
532 RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
533 if (mbuf_pool == NULL)
534 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
535 nb_ports_available = nb_ports;
537 /* initialize all ports */
538 for (portid = 0; portid < nb_ports; portid++) {
539 /* skip ports that are not enabled */
540 if ((enabled_port_mask & (1 << portid)) == 0) {
541 printf("\nSkipping disabled port %d\n", portid);
542 nb_ports_available--;
546 printf("Initializing port %u... done\n", (unsigned) portid);
548 if (port_init(portid, mbuf_pool) != 0)
549 rte_exit(EXIT_FAILURE, "Cannot initialize port %"PRIu8"\n",
553 if (!nb_ports_available) {
554 rte_exit(EXIT_FAILURE,
555 "All available ports are disabled. Please set portmask.\n");
558 d = rte_distributor_create("PKT_DIST", rte_socket_id(),
559 rte_lcore_count() - 2);
561 rte_exit(EXIT_FAILURE, "Cannot create distributor\n");
564 * scheduler ring is read only by the transmitter core, but written to
565 * by multiple threads
567 output_ring = rte_ring_create("Output_ring", RTE_RING_SZ,
568 rte_socket_id(), RING_F_SC_DEQ);
569 if (output_ring == NULL)
570 rte_exit(EXIT_FAILURE, "Cannot create output ring\n");
572 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
573 if (worker_id == rte_lcore_count() - 2)
574 rte_eal_remote_launch((lcore_function_t *)lcore_tx,
575 output_ring, lcore_id);
577 struct lcore_params *p =
578 rte_malloc(NULL, sizeof(*p), 0);
580 rte_panic("malloc failure\n");
581 *p = (struct lcore_params){worker_id, d, output_ring, mbuf_pool};
583 rte_eal_remote_launch((lcore_function_t *)lcore_worker,
588 /* call lcore_main on master core only */
589 struct lcore_params p = { 0, d, output_ring, mbuf_pool};
592 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
593 if (rte_eal_wait_lcore(lcore_id) < 0)