4 * Copyright(c) 2010-2016 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_malloc.h>
52 #include <rte_memory.h>
53 #include <rte_memcpy.h>
54 #include <rte_memzone.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>
70 #include <rte_mempool.h>
73 #define RTE_LOGTYPE_LSI RTE_LOGTYPE_USER1
77 #define MAX_PKT_BURST 32
78 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
81 * Configurable number of RX/TX ring descriptors
83 #define RTE_TEST_RX_DESC_DEFAULT 128
84 #define RTE_TEST_TX_DESC_DEFAULT 512
85 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
86 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
88 /* ethernet addresses of ports */
89 static struct ether_addr lsi_ports_eth_addr[RTE_MAX_ETHPORTS];
91 /* mask of enabled ports */
92 static uint32_t lsi_enabled_port_mask = 0;
94 static unsigned int lsi_rx_queue_per_lcore = 1;
96 /* destination port for L2 forwarding */
97 static unsigned lsi_dst_ports[RTE_MAX_ETHPORTS] = {0};
99 #define MAX_PKT_BURST 32
101 #define MAX_RX_QUEUE_PER_LCORE 16
102 #define MAX_TX_QUEUE_PER_PORT 16
103 struct lcore_queue_conf {
105 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
106 unsigned tx_queue_id;
107 } __rte_cache_aligned;
108 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
110 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
112 static const struct rte_eth_conf port_conf = {
115 .header_split = 0, /**< Header Split disabled */
116 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
117 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
118 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
119 .hw_strip_crc = 0, /**< CRC stripped by hardware */
122 .mq_mode = ETH_MQ_TX_NONE,
125 .lsc = 1, /**< lsc interrupt feature enabled */
129 struct rte_mempool * lsi_pktmbuf_pool = NULL;
131 /* Per-port statistics struct */
132 struct lsi_port_statistics {
136 } __rte_cache_aligned;
137 struct lsi_port_statistics port_statistics[RTE_MAX_ETHPORTS];
139 /* A tsc-based timer responsible for triggering statistics printout */
140 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
141 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
142 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
144 /* Print out statistics on packets dropped */
148 struct rte_eth_link link;
149 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
152 total_packets_dropped = 0;
153 total_packets_tx = 0;
154 total_packets_rx = 0;
156 const char clr[] = { 27, '[', '2', 'J', '\0' };
157 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
159 /* Clear screen and move to top left */
160 printf("%s%s", clr, topLeft);
162 printf("\nPort statistics ====================================");
164 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
165 /* skip ports that are not enabled */
166 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
169 memset(&link, 0, sizeof(link));
170 rte_eth_link_get_nowait((uint8_t)portid, &link);
171 printf("\nStatistics for port %u ------------------------------"
172 "\nLink status: %25s"
174 "\nLink duplex: %25s"
175 "\nPackets sent: %24"PRIu64
176 "\nPackets received: %20"PRIu64
177 "\nPackets dropped: %21"PRIu64,
179 (link.link_status ? "Link up" : "Link down"),
180 (unsigned)link.link_speed,
181 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
182 "full-duplex" : "half-duplex"),
183 port_statistics[portid].tx,
184 port_statistics[portid].rx,
185 port_statistics[portid].dropped);
187 total_packets_dropped += port_statistics[portid].dropped;
188 total_packets_tx += port_statistics[portid].tx;
189 total_packets_rx += port_statistics[portid].rx;
191 printf("\nAggregate statistics ==============================="
192 "\nTotal packets sent: %18"PRIu64
193 "\nTotal packets received: %14"PRIu64
194 "\nTotal packets dropped: %15"PRIu64,
197 total_packets_dropped);
198 printf("\n====================================================\n");
202 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
204 struct ether_hdr *eth;
206 unsigned dst_port = lsi_dst_ports[portid];
208 struct rte_eth_dev_tx_buffer *buffer;
210 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
212 /* 02:00:00:00:00:xx */
213 tmp = ð->d_addr.addr_bytes[0];
214 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
217 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
219 buffer = tx_buffer[dst_port];
220 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
222 port_statistics[dst_port].tx += sent;
225 /* main processing loop */
229 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
233 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
234 unsigned i, j, portid, nb_rx;
235 struct lcore_queue_conf *qconf;
236 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S *
238 struct rte_eth_dev_tx_buffer *buffer;
243 lcore_id = rte_lcore_id();
244 qconf = &lcore_queue_conf[lcore_id];
246 if (qconf->n_rx_port == 0) {
247 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
251 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
253 for (i = 0; i < qconf->n_rx_port; i++) {
255 portid = qconf->rx_port_list[i];
256 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
262 cur_tsc = rte_rdtsc();
265 * TX burst queue drain
267 diff_tsc = cur_tsc - prev_tsc;
268 if (unlikely(diff_tsc > drain_tsc)) {
270 for (i = 0; i < qconf->n_rx_port; i++) {
272 portid = lsi_dst_ports[qconf->rx_port_list[i]];
273 buffer = tx_buffer[portid];
275 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
277 port_statistics[portid].tx += sent;
281 /* if timer is enabled */
282 if (timer_period > 0) {
284 /* advance the timer */
285 timer_tsc += diff_tsc;
287 /* if timer has reached its timeout */
288 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
290 /* do this only on master core */
291 if (lcore_id == rte_get_master_lcore()) {
293 /* reset the timer */
303 * Read packet from RX queues
305 for (i = 0; i < qconf->n_rx_port; i++) {
307 portid = qconf->rx_port_list[i];
308 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
309 pkts_burst, MAX_PKT_BURST);
311 port_statistics[portid].rx += nb_rx;
313 for (j = 0; j < nb_rx; j++) {
315 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
316 lsi_simple_forward(m, portid);
323 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
331 lsi_usage(const char *prgname)
333 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
334 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
335 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
336 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
341 lsi_parse_portmask(const char *portmask)
346 /* parse hexadecimal string */
347 pm = strtoul(portmask, &end, 16);
348 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
358 lsi_parse_nqueue(const char *q_arg)
363 /* parse hexadecimal string */
364 n = strtoul(q_arg, &end, 10);
365 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
369 if (n >= MAX_RX_QUEUE_PER_LCORE)
376 lsi_parse_timer_period(const char *q_arg)
381 /* parse number string */
382 n = strtol(q_arg, &end, 10);
383 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
385 if (n >= MAX_TIMER_PERIOD)
391 /* Parse the argument given in the command line of the application */
393 lsi_parse_args(int argc, char **argv)
398 char *prgname = argv[0];
399 static struct option lgopts[] = {
405 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
406 lgopts, &option_index)) != EOF) {
411 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
412 if (lsi_enabled_port_mask == 0) {
413 printf("invalid portmask\n");
421 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
422 if (lsi_rx_queue_per_lcore == 0) {
423 printf("invalid queue number\n");
431 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
432 if (timer_period < 0) {
433 printf("invalid timer period\n");
451 argv[optind-1] = prgname;
454 optind = 0; /* reset getopt lib */
459 * It will be called as the callback for specified port after a LSI interrupt
460 * has been fully handled. This callback needs to be implemented carefully as
461 * it will be called in the interrupt host thread which is different from the
462 * application main thread.
469 * Pointer to(address of) the parameters.
475 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
477 struct rte_eth_link link;
481 printf("\n\nIn registered callback...\n");
482 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
483 rte_eth_link_get_nowait(port_id, &link);
484 if (link.link_status) {
485 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
486 port_id, (unsigned)link.link_speed,
487 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
488 ("full-duplex") : ("half-duplex"));
490 printf("Port %d Link Down\n\n", port_id);
493 /* Check the link status of all ports in up to 9s, and print them finally */
495 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
497 #define CHECK_INTERVAL 100 /* 100ms */
498 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
499 uint8_t portid, count, all_ports_up, print_flag = 0;
500 struct rte_eth_link link;
502 printf("\nChecking link status");
504 for (count = 0; count <= MAX_CHECK_TIME; count++) {
506 for (portid = 0; portid < port_num; portid++) {
507 if ((port_mask & (1 << portid)) == 0)
509 memset(&link, 0, sizeof(link));
510 rte_eth_link_get_nowait(portid, &link);
511 /* print link status if flag set */
512 if (print_flag == 1) {
513 if (link.link_status)
514 printf("Port %d Link Up - speed %u "
515 "Mbps - %s\n", (uint8_t)portid,
516 (unsigned)link.link_speed,
517 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
518 ("full-duplex") : ("half-duplex\n"));
520 printf("Port %d Link Down\n",
524 /* clear all_ports_up flag if any link down */
525 if (link.link_status == ETH_LINK_DOWN) {
530 /* after finally printing all link status, get out */
534 if (all_ports_up == 0) {
537 rte_delay_ms(CHECK_INTERVAL);
540 /* set the print_flag if all ports up or timeout */
541 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
549 main(int argc, char **argv)
551 struct lcore_queue_conf *qconf;
552 struct rte_eth_dev_info dev_info;
555 uint8_t portid, portid_last = 0;
556 unsigned lcore_id, rx_lcore_id;
557 unsigned nb_ports_in_mask = 0;
560 ret = rte_eal_init(argc, argv);
562 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
566 /* parse application arguments (after the EAL ones) */
567 ret = lsi_parse_args(argc, argv);
569 rte_exit(EXIT_FAILURE, "Invalid arguments");
571 /* create the mbuf pool */
573 rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32, 0,
574 RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
575 if (lsi_pktmbuf_pool == NULL)
576 rte_panic("Cannot init mbuf pool\n");
578 nb_ports = rte_eth_dev_count();
580 rte_panic("No Ethernet port - bye\n");
583 * Each logical core is assigned a dedicated TX queue on each port.
585 for (portid = 0; portid < nb_ports; portid++) {
586 /* skip ports that are not enabled */
587 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
590 /* save the destination port id */
591 if (nb_ports_in_mask % 2) {
592 lsi_dst_ports[portid] = portid_last;
593 lsi_dst_ports[portid_last] = portid;
596 portid_last = portid;
600 rte_eth_dev_info_get(portid, &dev_info);
602 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
603 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
604 "but it should be even and at least 2\n",
608 qconf = &lcore_queue_conf[rx_lcore_id];
610 /* Initialize the port/queue configuration of each logical core */
611 for (portid = 0; portid < nb_ports; portid++) {
612 /* skip ports that are not enabled */
613 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
616 /* get the lcore_id for this port */
617 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
618 lcore_queue_conf[rx_lcore_id].n_rx_port ==
619 lsi_rx_queue_per_lcore) {
622 if (rx_lcore_id >= RTE_MAX_LCORE)
623 rte_exit(EXIT_FAILURE, "Not enough cores\n");
625 if (qconf != &lcore_queue_conf[rx_lcore_id])
626 /* Assigned a new logical core in the loop above. */
627 qconf = &lcore_queue_conf[rx_lcore_id];
629 qconf->rx_port_list[qconf->n_rx_port] = portid;
631 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
634 /* Initialise each port */
635 for (portid = 0; portid < nb_ports; portid++) {
636 /* skip ports that are not enabled */
637 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
638 printf("Skipping disabled port %u\n", (unsigned) portid);
642 printf("Initializing port %u... ", (unsigned) portid);
644 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
646 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
647 ret, (unsigned) portid);
649 /* register lsi interrupt callback, need to be after
650 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
651 * lsc interrupt will be present, and below callback to
652 * be registered will never be called.
654 rte_eth_dev_callback_register(portid,
655 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
657 rte_eth_macaddr_get(portid,
658 &lsi_ports_eth_addr[portid]);
660 /* init one RX queue */
662 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
663 rte_eth_dev_socket_id(portid),
667 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
668 ret, (unsigned) portid);
670 /* init one TX queue logical core on each port */
672 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
673 rte_eth_dev_socket_id(portid),
676 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
677 ret, (unsigned) portid);
679 /* Initialize TX buffers */
680 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
681 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
682 rte_eth_dev_socket_id(portid));
683 if (tx_buffer[portid] == NULL)
684 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
687 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
689 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
690 rte_eth_tx_buffer_count_callback,
691 &port_statistics[portid].dropped);
693 rte_exit(EXIT_FAILURE, "Cannot set error callback for "
694 "tx buffer on port %u\n", (unsigned) portid);
697 ret = rte_eth_dev_start(portid);
699 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
700 ret, (unsigned) portid);
703 rte_eth_promiscuous_enable(portid);
705 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
707 lsi_ports_eth_addr[portid].addr_bytes[0],
708 lsi_ports_eth_addr[portid].addr_bytes[1],
709 lsi_ports_eth_addr[portid].addr_bytes[2],
710 lsi_ports_eth_addr[portid].addr_bytes[3],
711 lsi_ports_eth_addr[portid].addr_bytes[4],
712 lsi_ports_eth_addr[portid].addr_bytes[5]);
714 /* initialize port stats */
715 memset(&port_statistics, 0, sizeof(port_statistics));
718 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
720 /* launch per-lcore init on every lcore */
721 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
722 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
723 if (rte_eal_wait_lcore(lcore_id) < 0)