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)
82 * RX and TX Prefetch, Host, and Write-back threshold values should be
83 * carefully set for optimal performance. Consult the network
84 * controller's datasheet and supporting DPDK documentation for guidance
85 * on how these parameters should be set.
87 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
88 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
89 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
92 * These default values are optimized for use with the Intel(R) 82599 10 GbE
93 * Controller and the DPDK ixgbe PMD. Consider using other values for other
94 * network controllers and/or network drivers.
96 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
97 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
98 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
100 #define MAX_PKT_BURST 32
101 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
104 * Configurable number of RX/TX ring descriptors
106 #define RTE_TEST_RX_DESC_DEFAULT 128
107 #define RTE_TEST_TX_DESC_DEFAULT 512
108 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
109 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
111 /* ethernet addresses of ports */
112 static struct ether_addr lsi_ports_eth_addr[RTE_MAX_ETHPORTS];
114 /* mask of enabled ports */
115 static uint32_t lsi_enabled_port_mask = 0;
117 static unsigned int lsi_rx_queue_per_lcore = 1;
119 /* destination port for L2 forwarding */
120 static unsigned lsi_dst_ports[RTE_MAX_ETHPORTS] = {0};
122 #define MAX_PKT_BURST 32
125 struct rte_mbuf *m_table[MAX_PKT_BURST];
128 #define MAX_RX_QUEUE_PER_LCORE 16
129 #define MAX_TX_QUEUE_PER_PORT 16
130 struct lcore_queue_conf {
132 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
133 unsigned tx_queue_id;
134 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
136 } __rte_cache_aligned;
137 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
139 static const struct rte_eth_conf port_conf = {
142 .header_split = 0, /**< Header Split disabled */
143 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
144 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
145 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
146 .hw_strip_crc = 0, /**< CRC stripped by hardware */
149 .mq_mode = ETH_MQ_TX_NONE,
152 .lsc = 1, /**< lsc interrupt feature enabled */
156 static const struct rte_eth_rxconf rx_conf = {
158 .pthresh = RX_PTHRESH,
159 .hthresh = RX_HTHRESH,
160 .wthresh = RX_WTHRESH,
164 static const struct rte_eth_txconf tx_conf = {
166 .pthresh = TX_PTHRESH,
167 .hthresh = TX_HTHRESH,
168 .wthresh = TX_WTHRESH,
170 .tx_free_thresh = 0, /* Use PMD default values */
171 .tx_rs_thresh = 0, /* Use PMD default values */
174 struct rte_mempool * lsi_pktmbuf_pool = NULL;
176 /* Per-port statistics struct */
177 struct lsi_port_statistics {
181 } __rte_cache_aligned;
182 struct lsi_port_statistics port_statistics[RTE_MAX_ETHPORTS];
184 /* A tsc-based timer responsible for triggering statistics printout */
185 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
186 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
187 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
189 /* Print out statistics on packets dropped */
193 struct rte_eth_link link;
194 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
197 total_packets_dropped = 0;
198 total_packets_tx = 0;
199 total_packets_rx = 0;
201 const char clr[] = { 27, '[', '2', 'J', '\0' };
202 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
204 /* Clear screen and move to top left */
205 printf("%s%s", clr, topLeft);
207 printf("\nPort statistics ====================================");
209 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
210 /* skip ports that are not enabled */
211 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
214 memset(&link, 0, sizeof(link));
215 rte_eth_link_get_nowait((uint8_t)portid, &link);
216 printf("\nStatistics for port %u ------------------------------"
217 "\nLink status: %25s"
219 "\nLink duplex: %25s"
220 "\nPackets sent: %24"PRIu64
221 "\nPackets received: %20"PRIu64
222 "\nPackets dropped: %21"PRIu64,
224 (link.link_status ? "Link up" : "Link down"),
225 (unsigned)link.link_speed,
226 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
227 "full-duplex" : "half-duplex"),
228 port_statistics[portid].tx,
229 port_statistics[portid].rx,
230 port_statistics[portid].dropped);
232 total_packets_dropped += port_statistics[portid].dropped;
233 total_packets_tx += port_statistics[portid].tx;
234 total_packets_rx += port_statistics[portid].rx;
236 printf("\nAggregate statistics ==============================="
237 "\nTotal packets sent: %18"PRIu64
238 "\nTotal packets received: %14"PRIu64
239 "\nTotal packets dropped: %15"PRIu64,
242 total_packets_dropped);
243 printf("\n====================================================\n");
246 /* Send the packet on an output interface */
248 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
250 struct rte_mbuf **m_table;
254 queueid = (uint16_t) qconf->tx_queue_id;
255 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
257 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
258 port_statistics[port].tx += ret;
259 if (unlikely(ret < n)) {
260 port_statistics[port].dropped += (n - ret);
262 rte_pktmbuf_free(m_table[ret]);
269 /* Send the packet on an output interface */
271 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
273 unsigned lcore_id, len;
274 struct lcore_queue_conf *qconf;
276 lcore_id = rte_lcore_id();
278 qconf = &lcore_queue_conf[lcore_id];
279 len = qconf->tx_mbufs[port].len;
280 qconf->tx_mbufs[port].m_table[len] = m;
283 /* enough pkts to be sent */
284 if (unlikely(len == MAX_PKT_BURST)) {
285 lsi_send_burst(qconf, MAX_PKT_BURST, port);
289 qconf->tx_mbufs[port].len = len;
294 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
296 struct ether_hdr *eth;
298 unsigned dst_port = lsi_dst_ports[portid];
300 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
302 /* 02:00:00:00:00:xx */
303 tmp = ð->d_addr.addr_bytes[0];
304 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
307 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
309 lsi_send_packet(m, (uint8_t) dst_port);
312 /* main processing loop */
316 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
319 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
320 unsigned i, j, portid, nb_rx;
321 struct lcore_queue_conf *qconf;
322 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
327 lcore_id = rte_lcore_id();
328 qconf = &lcore_queue_conf[lcore_id];
330 if (qconf->n_rx_port == 0) {
331 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
335 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
337 for (i = 0; i < qconf->n_rx_port; i++) {
339 portid = qconf->rx_port_list[i];
340 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
346 cur_tsc = rte_rdtsc();
349 * TX burst queue drain
351 diff_tsc = cur_tsc - prev_tsc;
352 if (unlikely(diff_tsc > drain_tsc)) {
354 /* this could be optimized (use queueid instead of
355 * portid), but it is not called so often */
356 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
357 if (qconf->tx_mbufs[portid].len == 0)
359 lsi_send_burst(&lcore_queue_conf[lcore_id],
360 qconf->tx_mbufs[portid].len,
362 qconf->tx_mbufs[portid].len = 0;
365 /* if timer is enabled */
366 if (timer_period > 0) {
368 /* advance the timer */
369 timer_tsc += diff_tsc;
371 /* if timer has reached its timeout */
372 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
374 /* do this only on master core */
375 if (lcore_id == rte_get_master_lcore()) {
377 /* reset the timer */
387 * Read packet from RX queues
389 for (i = 0; i < qconf->n_rx_port; i++) {
391 portid = qconf->rx_port_list[i];
392 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
393 pkts_burst, MAX_PKT_BURST);
395 port_statistics[portid].rx += nb_rx;
397 for (j = 0; j < nb_rx; j++) {
399 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
400 lsi_simple_forward(m, portid);
407 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
415 lsi_usage(const char *prgname)
417 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
418 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
419 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
420 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
425 lsi_parse_portmask(const char *portmask)
430 /* parse hexadecimal string */
431 pm = strtoul(portmask, &end, 16);
432 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
442 lsi_parse_nqueue(const char *q_arg)
447 /* parse hexadecimal string */
448 n = strtoul(q_arg, &end, 10);
449 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
453 if (n >= MAX_RX_QUEUE_PER_LCORE)
460 lsi_parse_timer_period(const char *q_arg)
465 /* parse number string */
466 n = strtol(q_arg, &end, 10);
467 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
469 if (n >= MAX_TIMER_PERIOD)
475 /* Parse the argument given in the command line of the application */
477 lsi_parse_args(int argc, char **argv)
482 char *prgname = argv[0];
483 static struct option lgopts[] = {
489 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
490 lgopts, &option_index)) != EOF) {
495 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
496 if (lsi_enabled_port_mask == 0) {
497 printf("invalid portmask\n");
505 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
506 if (lsi_rx_queue_per_lcore == 0) {
507 printf("invalid queue number\n");
515 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
516 if (timer_period < 0) {
517 printf("invalid timer period\n");
535 argv[optind-1] = prgname;
538 optind = 0; /* reset getopt lib */
543 * It will be called as the callback for specified port after a LSI interrupt
544 * has been fully handled. This callback needs to be implemented carefully as
545 * it will be called in the interrupt host thread which is different from the
546 * application main thread.
553 * Pointer to(address of) the parameters.
559 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
561 struct rte_eth_link link;
565 printf("\n\nIn registered callback...\n");
566 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
567 rte_eth_link_get_nowait(port_id, &link);
568 if (link.link_status) {
569 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
570 port_id, (unsigned)link.link_speed,
571 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
572 ("full-duplex") : ("half-duplex"));
574 printf("Port %d Link Down\n\n", port_id);
577 /* Check the link status of all ports in up to 9s, and print them finally */
579 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
581 #define CHECK_INTERVAL 100 /* 100ms */
582 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
583 uint8_t portid, count, all_ports_up, print_flag = 0;
584 struct rte_eth_link link;
586 printf("\nChecking link status");
588 for (count = 0; count <= MAX_CHECK_TIME; count++) {
590 for (portid = 0; portid < port_num; portid++) {
591 if ((port_mask & (1 << portid)) == 0)
593 memset(&link, 0, sizeof(link));
594 rte_eth_link_get_nowait(portid, &link);
595 /* print link status if flag set */
596 if (print_flag == 1) {
597 if (link.link_status)
598 printf("Port %d Link Up - speed %u "
599 "Mbps - %s\n", (uint8_t)portid,
600 (unsigned)link.link_speed,
601 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
602 ("full-duplex") : ("half-duplex\n"));
604 printf("Port %d Link Down\n",
608 /* clear all_ports_up flag if any link down */
609 if (link.link_status == 0) {
614 /* after finally printing all link status, get out */
618 if (all_ports_up == 0) {
621 rte_delay_ms(CHECK_INTERVAL);
624 /* set the print_flag if all ports up or timeout */
625 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
633 MAIN(int argc, char **argv)
635 struct lcore_queue_conf *qconf;
636 struct rte_eth_dev_info dev_info;
639 uint8_t portid, portid_last = 0;
640 unsigned lcore_id, rx_lcore_id;
641 unsigned nb_ports_in_mask = 0;
644 ret = rte_eal_init(argc, argv);
646 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
650 /* parse application arguments (after the EAL ones) */
651 ret = lsi_parse_args(argc, argv);
653 rte_exit(EXIT_FAILURE, "Invalid arguments");
655 /* create the mbuf pool */
657 rte_mempool_create("mbuf_pool", NB_MBUF,
659 sizeof(struct rte_pktmbuf_pool_private),
660 rte_pktmbuf_pool_init, NULL,
661 rte_pktmbuf_init, NULL,
663 if (lsi_pktmbuf_pool == NULL)
664 rte_panic("Cannot init mbuf pool\n");
666 if (rte_eal_pci_probe() < 0)
667 rte_panic("Cannot probe PCI\n");
669 nb_ports = rte_eth_dev_count();
671 rte_panic("No Ethernet port - bye\n");
673 if (nb_ports > RTE_MAX_ETHPORTS)
674 nb_ports = RTE_MAX_ETHPORTS;
677 * Each logical core is assigned a dedicated TX queue on each port.
679 for (portid = 0; portid < nb_ports; portid++) {
680 /* skip ports that are not enabled */
681 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
684 /* save the destination port id */
685 if (nb_ports_in_mask % 2) {
686 lsi_dst_ports[portid] = portid_last;
687 lsi_dst_ports[portid_last] = portid;
690 portid_last = portid;
694 rte_eth_dev_info_get(portid, &dev_info);
696 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
697 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
698 "but it should be even and at least 2\n",
702 qconf = &lcore_queue_conf[rx_lcore_id];
704 /* Initialize the port/queue configuration of each logical core */
705 for (portid = 0; portid < nb_ports; portid++) {
706 /* skip ports that are not enabled */
707 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
710 /* get the lcore_id for this port */
711 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
712 lcore_queue_conf[rx_lcore_id].n_rx_port ==
713 lsi_rx_queue_per_lcore) {
716 if (rx_lcore_id >= RTE_MAX_LCORE)
717 rte_exit(EXIT_FAILURE, "Not enough cores\n");
719 if (qconf != &lcore_queue_conf[rx_lcore_id])
720 /* Assigned a new logical core in the loop above. */
721 qconf = &lcore_queue_conf[rx_lcore_id];
723 qconf->rx_port_list[qconf->n_rx_port] = portid;
725 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
728 /* Initialise each port */
729 for (portid = 0; portid < nb_ports; portid++) {
730 /* skip ports that are not enabled */
731 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
732 printf("Skipping disabled port %u\n", (unsigned) portid);
736 printf("Initializing port %u... ", (unsigned) portid);
738 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
740 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
741 ret, (unsigned) portid);
743 /* register lsi interrupt callback, need to be after
744 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
745 * lsc interrupt will be present, and below callback to
746 * be registered will never be called.
748 rte_eth_dev_callback_register(portid,
749 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
751 rte_eth_macaddr_get(portid,
752 &lsi_ports_eth_addr[portid]);
754 /* init one RX queue */
756 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
757 rte_eth_dev_socket_id(portid), &rx_conf,
760 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
761 ret, (unsigned) portid);
763 /* init one TX queue logical core on each port */
765 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
766 rte_eth_dev_socket_id(portid), &tx_conf);
768 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
769 ret, (unsigned) portid);
772 ret = rte_eth_dev_start(portid);
774 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
775 ret, (unsigned) portid);
778 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
780 lsi_ports_eth_addr[portid].addr_bytes[0],
781 lsi_ports_eth_addr[portid].addr_bytes[1],
782 lsi_ports_eth_addr[portid].addr_bytes[2],
783 lsi_ports_eth_addr[portid].addr_bytes[3],
784 lsi_ports_eth_addr[portid].addr_bytes[4],
785 lsi_ports_eth_addr[portid].addr_bytes[5]);
787 /* initialize port stats */
788 memset(&port_statistics, 0, sizeof(port_statistics));
791 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
793 /* launch per-lcore init on every lcore */
794 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
795 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
796 if (rte_eal_wait_lcore(lcore_id) < 0)