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>
54 #include <rte_tailq.h>
56 #include <rte_per_lcore.h>
57 #include <rte_launch.h>
58 #include <rte_atomic.h>
59 #include <rte_cycles.h>
60 #include <rte_prefetch.h>
61 #include <rte_lcore.h>
62 #include <rte_per_lcore.h>
63 #include <rte_branch_prediction.h>
64 #include <rte_interrupts.h>
66 #include <rte_random.h>
67 #include <rte_debug.h>
68 #include <rte_ether.h>
69 #include <rte_ethdev.h>
71 #include <rte_mempool.h>
76 #define RTE_LOGTYPE_LSI RTE_LOGTYPE_USER1
78 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
81 #define MAX_PKT_BURST 32
82 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
85 * Configurable number of RX/TX ring descriptors
87 #define RTE_TEST_RX_DESC_DEFAULT 128
88 #define RTE_TEST_TX_DESC_DEFAULT 512
89 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
90 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
92 /* ethernet addresses of ports */
93 static struct ether_addr lsi_ports_eth_addr[RTE_MAX_ETHPORTS];
95 /* mask of enabled ports */
96 static uint32_t lsi_enabled_port_mask = 0;
98 static unsigned int lsi_rx_queue_per_lcore = 1;
100 /* destination port for L2 forwarding */
101 static unsigned lsi_dst_ports[RTE_MAX_ETHPORTS] = {0};
103 #define MAX_PKT_BURST 32
106 struct rte_mbuf *m_table[MAX_PKT_BURST];
109 #define MAX_RX_QUEUE_PER_LCORE 16
110 #define MAX_TX_QUEUE_PER_PORT 16
111 struct lcore_queue_conf {
113 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
114 unsigned tx_queue_id;
115 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
117 } __rte_cache_aligned;
118 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
120 static const struct rte_eth_conf port_conf = {
123 .header_split = 0, /**< Header Split disabled */
124 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
125 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
126 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
127 .hw_strip_crc = 0, /**< CRC stripped by hardware */
130 .mq_mode = ETH_MQ_TX_NONE,
133 .lsc = 1, /**< lsc interrupt feature enabled */
137 struct rte_mempool * lsi_pktmbuf_pool = NULL;
139 /* Per-port statistics struct */
140 struct lsi_port_statistics {
144 } __rte_cache_aligned;
145 struct lsi_port_statistics port_statistics[RTE_MAX_ETHPORTS];
147 /* A tsc-based timer responsible for triggering statistics printout */
148 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
149 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
150 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
152 /* Print out statistics on packets dropped */
156 struct rte_eth_link link;
157 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
160 total_packets_dropped = 0;
161 total_packets_tx = 0;
162 total_packets_rx = 0;
164 const char clr[] = { 27, '[', '2', 'J', '\0' };
165 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
167 /* Clear screen and move to top left */
168 printf("%s%s", clr, topLeft);
170 printf("\nPort statistics ====================================");
172 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
173 /* skip ports that are not enabled */
174 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
177 memset(&link, 0, sizeof(link));
178 rte_eth_link_get_nowait((uint8_t)portid, &link);
179 printf("\nStatistics for port %u ------------------------------"
180 "\nLink status: %25s"
182 "\nLink duplex: %25s"
183 "\nPackets sent: %24"PRIu64
184 "\nPackets received: %20"PRIu64
185 "\nPackets dropped: %21"PRIu64,
187 (link.link_status ? "Link up" : "Link down"),
188 (unsigned)link.link_speed,
189 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
190 "full-duplex" : "half-duplex"),
191 port_statistics[portid].tx,
192 port_statistics[portid].rx,
193 port_statistics[portid].dropped);
195 total_packets_dropped += port_statistics[portid].dropped;
196 total_packets_tx += port_statistics[portid].tx;
197 total_packets_rx += port_statistics[portid].rx;
199 printf("\nAggregate statistics ==============================="
200 "\nTotal packets sent: %18"PRIu64
201 "\nTotal packets received: %14"PRIu64
202 "\nTotal packets dropped: %15"PRIu64,
205 total_packets_dropped);
206 printf("\n====================================================\n");
209 /* Send the packet on an output interface */
211 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
213 struct rte_mbuf **m_table;
217 queueid = (uint16_t) qconf->tx_queue_id;
218 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
220 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
221 port_statistics[port].tx += ret;
222 if (unlikely(ret < n)) {
223 port_statistics[port].dropped += (n - ret);
225 rte_pktmbuf_free(m_table[ret]);
232 /* Send the packet on an output interface */
234 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
236 unsigned lcore_id, len;
237 struct lcore_queue_conf *qconf;
239 lcore_id = rte_lcore_id();
241 qconf = &lcore_queue_conf[lcore_id];
242 len = qconf->tx_mbufs[port].len;
243 qconf->tx_mbufs[port].m_table[len] = m;
246 /* enough pkts to be sent */
247 if (unlikely(len == MAX_PKT_BURST)) {
248 lsi_send_burst(qconf, MAX_PKT_BURST, port);
252 qconf->tx_mbufs[port].len = len;
257 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
259 struct ether_hdr *eth;
261 unsigned dst_port = lsi_dst_ports[portid];
263 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
265 /* 02:00:00:00:00:xx */
266 tmp = ð->d_addr.addr_bytes[0];
267 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
270 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
272 lsi_send_packet(m, (uint8_t) dst_port);
275 /* main processing loop */
279 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
282 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
283 unsigned i, j, portid, nb_rx;
284 struct lcore_queue_conf *qconf;
285 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
290 lcore_id = rte_lcore_id();
291 qconf = &lcore_queue_conf[lcore_id];
293 if (qconf->n_rx_port == 0) {
294 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
298 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
300 for (i = 0; i < qconf->n_rx_port; i++) {
302 portid = qconf->rx_port_list[i];
303 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
309 cur_tsc = rte_rdtsc();
312 * TX burst queue drain
314 diff_tsc = cur_tsc - prev_tsc;
315 if (unlikely(diff_tsc > drain_tsc)) {
317 /* this could be optimized (use queueid instead of
318 * portid), but it is not called so often */
319 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
320 if (qconf->tx_mbufs[portid].len == 0)
322 lsi_send_burst(&lcore_queue_conf[lcore_id],
323 qconf->tx_mbufs[portid].len,
325 qconf->tx_mbufs[portid].len = 0;
328 /* if timer is enabled */
329 if (timer_period > 0) {
331 /* advance the timer */
332 timer_tsc += diff_tsc;
334 /* if timer has reached its timeout */
335 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
337 /* do this only on master core */
338 if (lcore_id == rte_get_master_lcore()) {
340 /* reset the timer */
350 * Read packet from RX queues
352 for (i = 0; i < qconf->n_rx_port; i++) {
354 portid = qconf->rx_port_list[i];
355 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
356 pkts_burst, MAX_PKT_BURST);
358 port_statistics[portid].rx += nb_rx;
360 for (j = 0; j < nb_rx; j++) {
362 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
363 lsi_simple_forward(m, portid);
370 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
378 lsi_usage(const char *prgname)
380 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
381 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
382 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
383 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
388 lsi_parse_portmask(const char *portmask)
393 /* parse hexadecimal string */
394 pm = strtoul(portmask, &end, 16);
395 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
405 lsi_parse_nqueue(const char *q_arg)
410 /* parse hexadecimal string */
411 n = strtoul(q_arg, &end, 10);
412 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
416 if (n >= MAX_RX_QUEUE_PER_LCORE)
423 lsi_parse_timer_period(const char *q_arg)
428 /* parse number string */
429 n = strtol(q_arg, &end, 10);
430 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
432 if (n >= MAX_TIMER_PERIOD)
438 /* Parse the argument given in the command line of the application */
440 lsi_parse_args(int argc, char **argv)
445 char *prgname = argv[0];
446 static struct option lgopts[] = {
452 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
453 lgopts, &option_index)) != EOF) {
458 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
459 if (lsi_enabled_port_mask == 0) {
460 printf("invalid portmask\n");
468 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
469 if (lsi_rx_queue_per_lcore == 0) {
470 printf("invalid queue number\n");
478 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
479 if (timer_period < 0) {
480 printf("invalid timer period\n");
498 argv[optind-1] = prgname;
501 optind = 0; /* reset getopt lib */
506 * It will be called as the callback for specified port after a LSI interrupt
507 * has been fully handled. This callback needs to be implemented carefully as
508 * it will be called in the interrupt host thread which is different from the
509 * application main thread.
516 * Pointer to(address of) the parameters.
522 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
524 struct rte_eth_link link;
528 printf("\n\nIn registered callback...\n");
529 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
530 rte_eth_link_get_nowait(port_id, &link);
531 if (link.link_status) {
532 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
533 port_id, (unsigned)link.link_speed,
534 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
535 ("full-duplex") : ("half-duplex"));
537 printf("Port %d Link Down\n\n", port_id);
540 /* Check the link status of all ports in up to 9s, and print them finally */
542 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
544 #define CHECK_INTERVAL 100 /* 100ms */
545 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
546 uint8_t portid, count, all_ports_up, print_flag = 0;
547 struct rte_eth_link link;
549 printf("\nChecking link status");
551 for (count = 0; count <= MAX_CHECK_TIME; count++) {
553 for (portid = 0; portid < port_num; portid++) {
554 if ((port_mask & (1 << portid)) == 0)
556 memset(&link, 0, sizeof(link));
557 rte_eth_link_get_nowait(portid, &link);
558 /* print link status if flag set */
559 if (print_flag == 1) {
560 if (link.link_status)
561 printf("Port %d Link Up - speed %u "
562 "Mbps - %s\n", (uint8_t)portid,
563 (unsigned)link.link_speed,
564 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
565 ("full-duplex") : ("half-duplex\n"));
567 printf("Port %d Link Down\n",
571 /* clear all_ports_up flag if any link down */
572 if (link.link_status == 0) {
577 /* after finally printing all link status, get out */
581 if (all_ports_up == 0) {
584 rte_delay_ms(CHECK_INTERVAL);
587 /* set the print_flag if all ports up or timeout */
588 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
596 MAIN(int argc, char **argv)
598 struct lcore_queue_conf *qconf;
599 struct rte_eth_dev_info dev_info;
602 uint8_t portid, portid_last = 0;
603 unsigned lcore_id, rx_lcore_id;
604 unsigned nb_ports_in_mask = 0;
607 ret = rte_eal_init(argc, argv);
609 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
613 /* parse application arguments (after the EAL ones) */
614 ret = lsi_parse_args(argc, argv);
616 rte_exit(EXIT_FAILURE, "Invalid arguments");
618 /* create the mbuf pool */
620 rte_mempool_create("mbuf_pool", NB_MBUF,
622 sizeof(struct rte_pktmbuf_pool_private),
623 rte_pktmbuf_pool_init, NULL,
624 rte_pktmbuf_init, NULL,
626 if (lsi_pktmbuf_pool == NULL)
627 rte_panic("Cannot init mbuf pool\n");
629 nb_ports = rte_eth_dev_count();
631 rte_panic("No Ethernet port - bye\n");
633 if (nb_ports > RTE_MAX_ETHPORTS)
634 nb_ports = RTE_MAX_ETHPORTS;
637 * Each logical core is assigned a dedicated TX queue on each port.
639 for (portid = 0; portid < nb_ports; portid++) {
640 /* skip ports that are not enabled */
641 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
644 /* save the destination port id */
645 if (nb_ports_in_mask % 2) {
646 lsi_dst_ports[portid] = portid_last;
647 lsi_dst_ports[portid_last] = portid;
650 portid_last = portid;
654 rte_eth_dev_info_get(portid, &dev_info);
656 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
657 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
658 "but it should be even and at least 2\n",
662 qconf = &lcore_queue_conf[rx_lcore_id];
664 /* Initialize the port/queue configuration of each logical core */
665 for (portid = 0; portid < nb_ports; portid++) {
666 /* skip ports that are not enabled */
667 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
670 /* get the lcore_id for this port */
671 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
672 lcore_queue_conf[rx_lcore_id].n_rx_port ==
673 lsi_rx_queue_per_lcore) {
676 if (rx_lcore_id >= RTE_MAX_LCORE)
677 rte_exit(EXIT_FAILURE, "Not enough cores\n");
679 if (qconf != &lcore_queue_conf[rx_lcore_id])
680 /* Assigned a new logical core in the loop above. */
681 qconf = &lcore_queue_conf[rx_lcore_id];
683 qconf->rx_port_list[qconf->n_rx_port] = portid;
685 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
688 /* Initialise each port */
689 for (portid = 0; portid < nb_ports; portid++) {
690 /* skip ports that are not enabled */
691 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
692 printf("Skipping disabled port %u\n", (unsigned) portid);
696 printf("Initializing port %u... ", (unsigned) portid);
698 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
700 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
701 ret, (unsigned) portid);
703 /* register lsi interrupt callback, need to be after
704 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
705 * lsc interrupt will be present, and below callback to
706 * be registered will never be called.
708 rte_eth_dev_callback_register(portid,
709 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
711 rte_eth_macaddr_get(portid,
712 &lsi_ports_eth_addr[portid]);
714 /* init one RX queue */
716 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
717 rte_eth_dev_socket_id(portid),
721 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
722 ret, (unsigned) portid);
724 /* init one TX queue logical core on each port */
726 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
727 rte_eth_dev_socket_id(portid),
730 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
731 ret, (unsigned) portid);
734 ret = rte_eth_dev_start(portid);
736 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
737 ret, (unsigned) portid);
740 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
742 lsi_ports_eth_addr[portid].addr_bytes[0],
743 lsi_ports_eth_addr[portid].addr_bytes[1],
744 lsi_ports_eth_addr[portid].addr_bytes[2],
745 lsi_ports_eth_addr[portid].addr_bytes[3],
746 lsi_ports_eth_addr[portid].addr_bytes[4],
747 lsi_ports_eth_addr[portid].addr_bytes[5]);
749 /* initialize port stats */
750 memset(&port_statistics, 0, sizeof(port_statistics));
753 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
755 /* launch per-lcore init on every lcore */
756 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
757 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
758 if (rte_eal_wait_lcore(lcore_id) < 0)