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
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
102 struct rte_mbuf *m_table[MAX_PKT_BURST];
105 #define MAX_RX_QUEUE_PER_LCORE 16
106 #define MAX_TX_QUEUE_PER_PORT 16
107 struct lcore_queue_conf {
109 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
110 unsigned tx_queue_id;
111 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
113 } __rte_cache_aligned;
114 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
116 static const struct rte_eth_conf port_conf = {
119 .header_split = 0, /**< Header Split disabled */
120 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
121 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
122 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
123 .hw_strip_crc = 0, /**< CRC stripped by hardware */
126 .mq_mode = ETH_MQ_TX_NONE,
129 .lsc = 1, /**< lsc interrupt feature enabled */
133 struct rte_mempool * lsi_pktmbuf_pool = NULL;
135 /* Per-port statistics struct */
136 struct lsi_port_statistics {
140 } __rte_cache_aligned;
141 struct lsi_port_statistics port_statistics[RTE_MAX_ETHPORTS];
143 /* A tsc-based timer responsible for triggering statistics printout */
144 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
145 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
146 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
148 /* Print out statistics on packets dropped */
152 struct rte_eth_link link;
153 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
156 total_packets_dropped = 0;
157 total_packets_tx = 0;
158 total_packets_rx = 0;
160 const char clr[] = { 27, '[', '2', 'J', '\0' };
161 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
163 /* Clear screen and move to top left */
164 printf("%s%s", clr, topLeft);
166 printf("\nPort statistics ====================================");
168 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
169 /* skip ports that are not enabled */
170 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
173 memset(&link, 0, sizeof(link));
174 rte_eth_link_get_nowait((uint8_t)portid, &link);
175 printf("\nStatistics for port %u ------------------------------"
176 "\nLink status: %25s"
178 "\nLink duplex: %25s"
179 "\nPackets sent: %24"PRIu64
180 "\nPackets received: %20"PRIu64
181 "\nPackets dropped: %21"PRIu64,
183 (link.link_status ? "Link up" : "Link down"),
184 (unsigned)link.link_speed,
185 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
186 "full-duplex" : "half-duplex"),
187 port_statistics[portid].tx,
188 port_statistics[portid].rx,
189 port_statistics[portid].dropped);
191 total_packets_dropped += port_statistics[portid].dropped;
192 total_packets_tx += port_statistics[portid].tx;
193 total_packets_rx += port_statistics[portid].rx;
195 printf("\nAggregate statistics ==============================="
196 "\nTotal packets sent: %18"PRIu64
197 "\nTotal packets received: %14"PRIu64
198 "\nTotal packets dropped: %15"PRIu64,
201 total_packets_dropped);
202 printf("\n====================================================\n");
205 /* Send the packet on an output interface */
207 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
209 struct rte_mbuf **m_table;
213 queueid = (uint16_t) qconf->tx_queue_id;
214 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
216 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
217 port_statistics[port].tx += ret;
218 if (unlikely(ret < n)) {
219 port_statistics[port].dropped += (n - ret);
221 rte_pktmbuf_free(m_table[ret]);
228 /* Send the packet on an output interface */
230 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
232 unsigned lcore_id, len;
233 struct lcore_queue_conf *qconf;
235 lcore_id = rte_lcore_id();
237 qconf = &lcore_queue_conf[lcore_id];
238 len = qconf->tx_mbufs[port].len;
239 qconf->tx_mbufs[port].m_table[len] = m;
242 /* enough pkts to be sent */
243 if (unlikely(len == MAX_PKT_BURST)) {
244 lsi_send_burst(qconf, MAX_PKT_BURST, port);
248 qconf->tx_mbufs[port].len = len;
253 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
255 struct ether_hdr *eth;
257 unsigned dst_port = lsi_dst_ports[portid];
259 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
261 /* 02:00:00:00:00:xx */
262 tmp = ð->d_addr.addr_bytes[0];
263 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
266 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
268 lsi_send_packet(m, (uint8_t) dst_port);
271 /* main processing loop */
275 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
278 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
279 unsigned i, j, portid, nb_rx;
280 struct lcore_queue_conf *qconf;
281 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
286 lcore_id = rte_lcore_id();
287 qconf = &lcore_queue_conf[lcore_id];
289 if (qconf->n_rx_port == 0) {
290 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
294 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
296 for (i = 0; i < qconf->n_rx_port; i++) {
298 portid = qconf->rx_port_list[i];
299 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
305 cur_tsc = rte_rdtsc();
308 * TX burst queue drain
310 diff_tsc = cur_tsc - prev_tsc;
311 if (unlikely(diff_tsc > drain_tsc)) {
313 /* this could be optimized (use queueid instead of
314 * portid), but it is not called so often */
315 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
316 if (qconf->tx_mbufs[portid].len == 0)
318 lsi_send_burst(&lcore_queue_conf[lcore_id],
319 qconf->tx_mbufs[portid].len,
321 qconf->tx_mbufs[portid].len = 0;
324 /* if timer is enabled */
325 if (timer_period > 0) {
327 /* advance the timer */
328 timer_tsc += diff_tsc;
330 /* if timer has reached its timeout */
331 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
333 /* do this only on master core */
334 if (lcore_id == rte_get_master_lcore()) {
336 /* reset the timer */
346 * Read packet from RX queues
348 for (i = 0; i < qconf->n_rx_port; i++) {
350 portid = qconf->rx_port_list[i];
351 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
352 pkts_burst, MAX_PKT_BURST);
354 port_statistics[portid].rx += nb_rx;
356 for (j = 0; j < nb_rx; j++) {
358 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
359 lsi_simple_forward(m, portid);
366 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
374 lsi_usage(const char *prgname)
376 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
377 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
378 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
379 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
384 lsi_parse_portmask(const char *portmask)
389 /* parse hexadecimal string */
390 pm = strtoul(portmask, &end, 16);
391 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
401 lsi_parse_nqueue(const char *q_arg)
406 /* parse hexadecimal string */
407 n = strtoul(q_arg, &end, 10);
408 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
412 if (n >= MAX_RX_QUEUE_PER_LCORE)
419 lsi_parse_timer_period(const char *q_arg)
424 /* parse number string */
425 n = strtol(q_arg, &end, 10);
426 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
428 if (n >= MAX_TIMER_PERIOD)
434 /* Parse the argument given in the command line of the application */
436 lsi_parse_args(int argc, char **argv)
441 char *prgname = argv[0];
442 static struct option lgopts[] = {
448 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
449 lgopts, &option_index)) != EOF) {
454 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
455 if (lsi_enabled_port_mask == 0) {
456 printf("invalid portmask\n");
464 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
465 if (lsi_rx_queue_per_lcore == 0) {
466 printf("invalid queue number\n");
474 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
475 if (timer_period < 0) {
476 printf("invalid timer period\n");
494 argv[optind-1] = prgname;
497 optind = 0; /* reset getopt lib */
502 * It will be called as the callback for specified port after a LSI interrupt
503 * has been fully handled. This callback needs to be implemented carefully as
504 * it will be called in the interrupt host thread which is different from the
505 * application main thread.
512 * Pointer to(address of) the parameters.
518 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
520 struct rte_eth_link link;
524 printf("\n\nIn registered callback...\n");
525 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
526 rte_eth_link_get_nowait(port_id, &link);
527 if (link.link_status) {
528 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
529 port_id, (unsigned)link.link_speed,
530 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
531 ("full-duplex") : ("half-duplex"));
533 printf("Port %d Link Down\n\n", port_id);
536 /* Check the link status of all ports in up to 9s, and print them finally */
538 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
540 #define CHECK_INTERVAL 100 /* 100ms */
541 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
542 uint8_t portid, count, all_ports_up, print_flag = 0;
543 struct rte_eth_link link;
545 printf("\nChecking link status");
547 for (count = 0; count <= MAX_CHECK_TIME; count++) {
549 for (portid = 0; portid < port_num; portid++) {
550 if ((port_mask & (1 << portid)) == 0)
552 memset(&link, 0, sizeof(link));
553 rte_eth_link_get_nowait(portid, &link);
554 /* print link status if flag set */
555 if (print_flag == 1) {
556 if (link.link_status)
557 printf("Port %d Link Up - speed %u "
558 "Mbps - %s\n", (uint8_t)portid,
559 (unsigned)link.link_speed,
560 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
561 ("full-duplex") : ("half-duplex\n"));
563 printf("Port %d Link Down\n",
567 /* clear all_ports_up flag if any link down */
568 if (link.link_status == 0) {
573 /* after finally printing all link status, get out */
577 if (all_ports_up == 0) {
580 rte_delay_ms(CHECK_INTERVAL);
583 /* set the print_flag if all ports up or timeout */
584 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
592 main(int argc, char **argv)
594 struct lcore_queue_conf *qconf;
595 struct rte_eth_dev_info dev_info;
598 uint8_t portid, portid_last = 0;
599 unsigned lcore_id, rx_lcore_id;
600 unsigned nb_ports_in_mask = 0;
603 ret = rte_eal_init(argc, argv);
605 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
609 /* parse application arguments (after the EAL ones) */
610 ret = lsi_parse_args(argc, argv);
612 rte_exit(EXIT_FAILURE, "Invalid arguments");
614 /* create the mbuf pool */
616 rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32, 0,
617 RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
618 if (lsi_pktmbuf_pool == NULL)
619 rte_panic("Cannot init mbuf pool\n");
621 nb_ports = rte_eth_dev_count();
623 rte_panic("No Ethernet port - bye\n");
625 if (nb_ports > RTE_MAX_ETHPORTS)
626 nb_ports = RTE_MAX_ETHPORTS;
629 * Each logical core is assigned a dedicated TX queue on each port.
631 for (portid = 0; portid < nb_ports; portid++) {
632 /* skip ports that are not enabled */
633 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
636 /* save the destination port id */
637 if (nb_ports_in_mask % 2) {
638 lsi_dst_ports[portid] = portid_last;
639 lsi_dst_ports[portid_last] = portid;
642 portid_last = portid;
646 rte_eth_dev_info_get(portid, &dev_info);
648 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
649 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
650 "but it should be even and at least 2\n",
654 qconf = &lcore_queue_conf[rx_lcore_id];
656 /* Initialize the port/queue configuration of each logical core */
657 for (portid = 0; portid < nb_ports; portid++) {
658 /* skip ports that are not enabled */
659 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
662 /* get the lcore_id for this port */
663 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
664 lcore_queue_conf[rx_lcore_id].n_rx_port ==
665 lsi_rx_queue_per_lcore) {
668 if (rx_lcore_id >= RTE_MAX_LCORE)
669 rte_exit(EXIT_FAILURE, "Not enough cores\n");
671 if (qconf != &lcore_queue_conf[rx_lcore_id])
672 /* Assigned a new logical core in the loop above. */
673 qconf = &lcore_queue_conf[rx_lcore_id];
675 qconf->rx_port_list[qconf->n_rx_port] = portid;
677 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
680 /* Initialise each port */
681 for (portid = 0; portid < nb_ports; portid++) {
682 /* skip ports that are not enabled */
683 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
684 printf("Skipping disabled port %u\n", (unsigned) portid);
688 printf("Initializing port %u... ", (unsigned) portid);
690 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
692 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
693 ret, (unsigned) portid);
695 /* register lsi interrupt callback, need to be after
696 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
697 * lsc interrupt will be present, and below callback to
698 * be registered will never be called.
700 rte_eth_dev_callback_register(portid,
701 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
703 rte_eth_macaddr_get(portid,
704 &lsi_ports_eth_addr[portid]);
706 /* init one RX queue */
708 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
709 rte_eth_dev_socket_id(portid),
713 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
714 ret, (unsigned) portid);
716 /* init one TX queue logical core on each port */
718 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
719 rte_eth_dev_socket_id(portid),
722 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
723 ret, (unsigned) portid);
726 ret = rte_eth_dev_start(portid);
728 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
729 ret, (unsigned) portid);
732 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
734 lsi_ports_eth_addr[portid].addr_bytes[0],
735 lsi_ports_eth_addr[portid].addr_bytes[1],
736 lsi_ports_eth_addr[portid].addr_bytes[2],
737 lsi_ports_eth_addr[portid].addr_bytes[3],
738 lsi_ports_eth_addr[portid].addr_bytes[4],
739 lsi_ports_eth_addr[portid].addr_bytes[5]);
741 /* initialize port stats */
742 memset(&port_statistics, 0, sizeof(port_statistics));
745 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
747 /* launch per-lcore init on every lcore */
748 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
749 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
750 if (rte_eal_wait_lcore(lcore_id) < 0)