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.
39 #include <sys/types.h>
41 #include <sys/queue.h>
42 #include <netinet/in.h>
49 #include <rte_common.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
70 #include <rte_mempool.h>
73 #define RTE_LOGTYPE_LSI RTE_LOGTYPE_USER1
75 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
78 #define MAX_PKT_BURST 32
79 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
82 * Configurable number of RX/TX ring descriptors
84 #define RTE_TEST_RX_DESC_DEFAULT 128
85 #define RTE_TEST_TX_DESC_DEFAULT 512
86 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
87 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
89 /* ethernet addresses of ports */
90 static struct ether_addr lsi_ports_eth_addr[RTE_MAX_ETHPORTS];
92 /* mask of enabled ports */
93 static uint32_t lsi_enabled_port_mask = 0;
95 static unsigned int lsi_rx_queue_per_lcore = 1;
97 /* destination port for L2 forwarding */
98 static unsigned lsi_dst_ports[RTE_MAX_ETHPORTS] = {0};
100 #define MAX_PKT_BURST 32
103 struct rte_mbuf *m_table[MAX_PKT_BURST];
106 #define MAX_RX_QUEUE_PER_LCORE 16
107 #define MAX_TX_QUEUE_PER_PORT 16
108 struct lcore_queue_conf {
110 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
111 unsigned tx_queue_id;
112 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
114 } __rte_cache_aligned;
115 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
117 static const struct rte_eth_conf port_conf = {
120 .header_split = 0, /**< Header Split disabled */
121 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
122 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
123 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
124 .hw_strip_crc = 0, /**< CRC stripped by hardware */
127 .mq_mode = ETH_MQ_TX_NONE,
130 .lsc = 1, /**< lsc interrupt feature enabled */
134 struct rte_mempool * lsi_pktmbuf_pool = NULL;
136 /* Per-port statistics struct */
137 struct lsi_port_statistics {
141 } __rte_cache_aligned;
142 struct lsi_port_statistics port_statistics[RTE_MAX_ETHPORTS];
144 /* A tsc-based timer responsible for triggering statistics printout */
145 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
146 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
147 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
149 /* Print out statistics on packets dropped */
153 struct rte_eth_link link;
154 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
157 total_packets_dropped = 0;
158 total_packets_tx = 0;
159 total_packets_rx = 0;
161 const char clr[] = { 27, '[', '2', 'J', '\0' };
162 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
164 /* Clear screen and move to top left */
165 printf("%s%s", clr, topLeft);
167 printf("\nPort statistics ====================================");
169 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
170 /* skip ports that are not enabled */
171 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
174 memset(&link, 0, sizeof(link));
175 rte_eth_link_get_nowait((uint8_t)portid, &link);
176 printf("\nStatistics for port %u ------------------------------"
177 "\nLink status: %25s"
179 "\nLink duplex: %25s"
180 "\nPackets sent: %24"PRIu64
181 "\nPackets received: %20"PRIu64
182 "\nPackets dropped: %21"PRIu64,
184 (link.link_status ? "Link up" : "Link down"),
185 (unsigned)link.link_speed,
186 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
187 "full-duplex" : "half-duplex"),
188 port_statistics[portid].tx,
189 port_statistics[portid].rx,
190 port_statistics[portid].dropped);
192 total_packets_dropped += port_statistics[portid].dropped;
193 total_packets_tx += port_statistics[portid].tx;
194 total_packets_rx += port_statistics[portid].rx;
196 printf("\nAggregate statistics ==============================="
197 "\nTotal packets sent: %18"PRIu64
198 "\nTotal packets received: %14"PRIu64
199 "\nTotal packets dropped: %15"PRIu64,
202 total_packets_dropped);
203 printf("\n====================================================\n");
206 /* Send the packet on an output interface */
208 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
210 struct rte_mbuf **m_table;
214 queueid = (uint16_t) qconf->tx_queue_id;
215 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
217 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
218 port_statistics[port].tx += ret;
219 if (unlikely(ret < n)) {
220 port_statistics[port].dropped += (n - ret);
222 rte_pktmbuf_free(m_table[ret]);
229 /* Send the packet on an output interface */
231 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
233 unsigned lcore_id, len;
234 struct lcore_queue_conf *qconf;
236 lcore_id = rte_lcore_id();
238 qconf = &lcore_queue_conf[lcore_id];
239 len = qconf->tx_mbufs[port].len;
240 qconf->tx_mbufs[port].m_table[len] = m;
243 /* enough pkts to be sent */
244 if (unlikely(len == MAX_PKT_BURST)) {
245 lsi_send_burst(qconf, MAX_PKT_BURST, port);
249 qconf->tx_mbufs[port].len = len;
254 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
256 struct ether_hdr *eth;
258 unsigned dst_port = lsi_dst_ports[portid];
260 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
262 /* 02:00:00:00:00:xx */
263 tmp = ð->d_addr.addr_bytes[0];
264 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
267 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
269 lsi_send_packet(m, (uint8_t) dst_port);
272 /* main processing loop */
276 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
279 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
280 unsigned i, j, portid, nb_rx;
281 struct lcore_queue_conf *qconf;
282 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
287 lcore_id = rte_lcore_id();
288 qconf = &lcore_queue_conf[lcore_id];
290 if (qconf->n_rx_port == 0) {
291 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
295 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
297 for (i = 0; i < qconf->n_rx_port; i++) {
299 portid = qconf->rx_port_list[i];
300 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
306 cur_tsc = rte_rdtsc();
309 * TX burst queue drain
311 diff_tsc = cur_tsc - prev_tsc;
312 if (unlikely(diff_tsc > drain_tsc)) {
314 /* this could be optimized (use queueid instead of
315 * portid), but it is not called so often */
316 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
317 if (qconf->tx_mbufs[portid].len == 0)
319 lsi_send_burst(&lcore_queue_conf[lcore_id],
320 qconf->tx_mbufs[portid].len,
322 qconf->tx_mbufs[portid].len = 0;
325 /* if timer is enabled */
326 if (timer_period > 0) {
328 /* advance the timer */
329 timer_tsc += diff_tsc;
331 /* if timer has reached its timeout */
332 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
334 /* do this only on master core */
335 if (lcore_id == rte_get_master_lcore()) {
337 /* reset the timer */
347 * Read packet from RX queues
349 for (i = 0; i < qconf->n_rx_port; i++) {
351 portid = qconf->rx_port_list[i];
352 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
353 pkts_burst, MAX_PKT_BURST);
355 port_statistics[portid].rx += nb_rx;
357 for (j = 0; j < nb_rx; j++) {
359 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
360 lsi_simple_forward(m, portid);
367 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
375 lsi_usage(const char *prgname)
377 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
378 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
379 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
380 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
385 lsi_parse_portmask(const char *portmask)
390 /* parse hexadecimal string */
391 pm = strtoul(portmask, &end, 16);
392 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
402 lsi_parse_nqueue(const char *q_arg)
407 /* parse hexadecimal string */
408 n = strtoul(q_arg, &end, 10);
409 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
413 if (n >= MAX_RX_QUEUE_PER_LCORE)
420 lsi_parse_timer_period(const char *q_arg)
425 /* parse number string */
426 n = strtol(q_arg, &end, 10);
427 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
429 if (n >= MAX_TIMER_PERIOD)
435 /* Parse the argument given in the command line of the application */
437 lsi_parse_args(int argc, char **argv)
442 char *prgname = argv[0];
443 static struct option lgopts[] = {
449 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
450 lgopts, &option_index)) != EOF) {
455 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
456 if (lsi_enabled_port_mask == 0) {
457 printf("invalid portmask\n");
465 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
466 if (lsi_rx_queue_per_lcore == 0) {
467 printf("invalid queue number\n");
475 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
476 if (timer_period < 0) {
477 printf("invalid timer period\n");
495 argv[optind-1] = prgname;
498 optind = 0; /* reset getopt lib */
503 * It will be called as the callback for specified port after a LSI interrupt
504 * has been fully handled. This callback needs to be implemented carefully as
505 * it will be called in the interrupt host thread which is different from the
506 * application main thread.
513 * Pointer to(address of) the parameters.
519 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
521 struct rte_eth_link link;
525 printf("\n\nIn registered callback...\n");
526 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
527 rte_eth_link_get_nowait(port_id, &link);
528 if (link.link_status) {
529 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
530 port_id, (unsigned)link.link_speed,
531 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
532 ("full-duplex") : ("half-duplex"));
534 printf("Port %d Link Down\n\n", port_id);
537 /* Check the link status of all ports in up to 9s, and print them finally */
539 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
541 #define CHECK_INTERVAL 100 /* 100ms */
542 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
543 uint8_t portid, count, all_ports_up, print_flag = 0;
544 struct rte_eth_link link;
546 printf("\nChecking link status");
548 for (count = 0; count <= MAX_CHECK_TIME; count++) {
550 for (portid = 0; portid < port_num; portid++) {
551 if ((port_mask & (1 << portid)) == 0)
553 memset(&link, 0, sizeof(link));
554 rte_eth_link_get_nowait(portid, &link);
555 /* print link status if flag set */
556 if (print_flag == 1) {
557 if (link.link_status)
558 printf("Port %d Link Up - speed %u "
559 "Mbps - %s\n", (uint8_t)portid,
560 (unsigned)link.link_speed,
561 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
562 ("full-duplex") : ("half-duplex\n"));
564 printf("Port %d Link Down\n",
568 /* clear all_ports_up flag if any link down */
569 if (link.link_status == 0) {
574 /* after finally printing all link status, get out */
578 if (all_ports_up == 0) {
581 rte_delay_ms(CHECK_INTERVAL);
584 /* set the print_flag if all ports up or timeout */
585 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
593 main(int argc, char **argv)
595 struct lcore_queue_conf *qconf;
596 struct rte_eth_dev_info dev_info;
599 uint8_t portid, portid_last = 0;
600 unsigned lcore_id, rx_lcore_id;
601 unsigned nb_ports_in_mask = 0;
604 ret = rte_eal_init(argc, argv);
606 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
610 /* parse application arguments (after the EAL ones) */
611 ret = lsi_parse_args(argc, argv);
613 rte_exit(EXIT_FAILURE, "Invalid arguments");
615 /* create the mbuf pool */
617 rte_mempool_create("mbuf_pool", NB_MBUF,
619 sizeof(struct rte_pktmbuf_pool_private),
620 rte_pktmbuf_pool_init, NULL,
621 rte_pktmbuf_init, NULL,
623 if (lsi_pktmbuf_pool == NULL)
624 rte_panic("Cannot init mbuf pool\n");
626 nb_ports = rte_eth_dev_count();
628 rte_panic("No Ethernet port - bye\n");
630 if (nb_ports > RTE_MAX_ETHPORTS)
631 nb_ports = RTE_MAX_ETHPORTS;
634 * Each logical core is assigned a dedicated TX queue on each port.
636 for (portid = 0; portid < nb_ports; portid++) {
637 /* skip ports that are not enabled */
638 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
641 /* save the destination port id */
642 if (nb_ports_in_mask % 2) {
643 lsi_dst_ports[portid] = portid_last;
644 lsi_dst_ports[portid_last] = portid;
647 portid_last = portid;
651 rte_eth_dev_info_get(portid, &dev_info);
653 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
654 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
655 "but it should be even and at least 2\n",
659 qconf = &lcore_queue_conf[rx_lcore_id];
661 /* Initialize the port/queue configuration of each logical core */
662 for (portid = 0; portid < nb_ports; portid++) {
663 /* skip ports that are not enabled */
664 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
667 /* get the lcore_id for this port */
668 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
669 lcore_queue_conf[rx_lcore_id].n_rx_port ==
670 lsi_rx_queue_per_lcore) {
673 if (rx_lcore_id >= RTE_MAX_LCORE)
674 rte_exit(EXIT_FAILURE, "Not enough cores\n");
676 if (qconf != &lcore_queue_conf[rx_lcore_id])
677 /* Assigned a new logical core in the loop above. */
678 qconf = &lcore_queue_conf[rx_lcore_id];
680 qconf->rx_port_list[qconf->n_rx_port] = portid;
682 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
685 /* Initialise each port */
686 for (portid = 0; portid < nb_ports; portid++) {
687 /* skip ports that are not enabled */
688 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
689 printf("Skipping disabled port %u\n", (unsigned) portid);
693 printf("Initializing port %u... ", (unsigned) portid);
695 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
697 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
698 ret, (unsigned) portid);
700 /* register lsi interrupt callback, need to be after
701 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
702 * lsc interrupt will be present, and below callback to
703 * be registered will never be called.
705 rte_eth_dev_callback_register(portid,
706 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
708 rte_eth_macaddr_get(portid,
709 &lsi_ports_eth_addr[portid]);
711 /* init one RX queue */
713 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
714 rte_eth_dev_socket_id(portid),
718 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
719 ret, (unsigned) portid);
721 /* init one TX queue logical core on each port */
723 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
724 rte_eth_dev_socket_id(portid),
727 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
728 ret, (unsigned) portid);
731 ret = rte_eth_dev_start(portid);
733 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
734 ret, (unsigned) portid);
737 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
739 lsi_ports_eth_addr[portid].addr_bytes[0],
740 lsi_ports_eth_addr[portid].addr_bytes[1],
741 lsi_ports_eth_addr[portid].addr_bytes[2],
742 lsi_ports_eth_addr[portid].addr_bytes[3],
743 lsi_ports_eth_addr[portid].addr_bytes[4],
744 lsi_ports_eth_addr[portid].addr_bytes[5]);
746 /* initialize port stats */
747 memset(&port_statistics, 0, sizeof(port_statistics));
750 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
752 /* launch per-lcore init on every lcore */
753 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
754 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
755 if (rte_eal_wait_lcore(lcore_id) < 0)