4 * Copyright(c) 2010-2013 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.
40 #include <sys/types.h>
42 #include <sys/queue.h>
43 #include <netinet/in.h>
50 #include <rte_common.h>
52 #include <rte_memory.h>
53 #include <rte_memcpy.h>
54 #include <rte_memzone.h>
55 #include <rte_tailq.h>
57 #include <rte_per_lcore.h>
58 #include <rte_launch.h>
59 #include <rte_atomic.h>
60 #include <rte_cycles.h>
61 #include <rte_prefetch.h>
62 #include <rte_lcore.h>
63 #include <rte_per_lcore.h>
64 #include <rte_branch_prediction.h>
65 #include <rte_interrupts.h>
67 #include <rte_random.h>
68 #include <rte_debug.h>
69 #include <rte_ether.h>
70 #include <rte_ethdev.h>
72 #include <rte_mempool.h>
77 #define RTE_LOGTYPE_LSI RTE_LOGTYPE_USER1
79 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
83 * RX and TX Prefetch, Host, and Write-back threshold values should be
84 * carefully set for optimal performance. Consult the network
85 * controller's datasheet and supporting DPDK documentation for guidance
86 * on how these parameters should be set.
88 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
89 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
90 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
93 * These default values are optimized for use with the Intel(R) 82599 10 GbE
94 * Controller and the DPDK ixgbe PMD. Consider using other values for other
95 * network controllers and/or network drivers.
97 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
98 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
99 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
101 #define MAX_PKT_BURST 32
102 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
105 * Configurable number of RX/TX ring descriptors
107 #define RTE_TEST_RX_DESC_DEFAULT 128
108 #define RTE_TEST_TX_DESC_DEFAULT 512
109 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
110 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
112 /* ethernet addresses of ports */
113 static struct ether_addr lsi_ports_eth_addr[RTE_MAX_ETHPORTS];
115 /* mask of enabled ports */
116 static uint32_t lsi_enabled_port_mask = 0;
118 static unsigned int lsi_rx_queue_per_lcore = 1;
120 /* destination port for L2 forwarding */
121 static unsigned lsi_dst_ports[RTE_MAX_ETHPORTS] = {0};
123 #define MAX_PKT_BURST 32
126 struct rte_mbuf *m_table[MAX_PKT_BURST];
129 #define MAX_RX_QUEUE_PER_LCORE 16
130 #define MAX_TX_QUEUE_PER_PORT 16
131 struct lcore_queue_conf {
133 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
134 unsigned tx_queue_id;
135 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
137 } __rte_cache_aligned;
138 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
140 static const struct rte_eth_conf port_conf = {
143 .header_split = 0, /**< Header Split disabled */
144 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
145 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
146 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
147 .hw_strip_crc = 0, /**< CRC stripped by hardware */
150 .mq_mode = ETH_MQ_TX_NONE,
153 .lsc = 1, /**< lsc interrupt feature enabled */
157 static const struct rte_eth_rxconf rx_conf = {
159 .pthresh = RX_PTHRESH,
160 .hthresh = RX_HTHRESH,
161 .wthresh = RX_WTHRESH,
165 static const struct rte_eth_txconf tx_conf = {
167 .pthresh = TX_PTHRESH,
168 .hthresh = TX_HTHRESH,
169 .wthresh = TX_WTHRESH,
171 .tx_free_thresh = 0, /* Use PMD default values */
172 .tx_rs_thresh = 0, /* Use PMD default values */
175 struct rte_mempool * lsi_pktmbuf_pool = NULL;
177 /* Per-port statistics struct */
178 struct lsi_port_statistics {
182 } __rte_cache_aligned;
183 struct lsi_port_statistics port_statistics[RTE_MAX_ETHPORTS];
185 /* A tsc-based timer responsible for triggering statistics printout */
186 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
187 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
188 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
190 /* Print out statistics on packets dropped */
194 struct rte_eth_link link;
195 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
198 total_packets_dropped = 0;
199 total_packets_tx = 0;
200 total_packets_rx = 0;
202 const char clr[] = { 27, '[', '2', 'J', '\0' };
203 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
205 /* Clear screen and move to top left */
206 printf("%s%s", clr, topLeft);
208 printf("\nPort statistics ====================================");
210 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
211 /* skip ports that are not enabled */
212 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
215 memset(&link, 0, sizeof(link));
216 rte_eth_link_get_nowait((uint8_t)portid, &link);
217 printf("\nStatistics for port %u ------------------------------"
218 "\nLink status: %25s"
220 "\nLink duplex: %25s"
221 "\nPackets sent: %24"PRIu64
222 "\nPackets received: %20"PRIu64
223 "\nPackets dropped: %21"PRIu64,
225 (link.link_status ? "Link up" : "Link down"),
226 (unsigned)link.link_speed,
227 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
228 "full-duplex" : "half-duplex"),
229 port_statistics[portid].tx,
230 port_statistics[portid].rx,
231 port_statistics[portid].dropped);
233 total_packets_dropped += port_statistics[portid].dropped;
234 total_packets_tx += port_statistics[portid].tx;
235 total_packets_rx += port_statistics[portid].rx;
237 printf("\nAggregate statistics ==============================="
238 "\nTotal packets sent: %18"PRIu64
239 "\nTotal packets received: %14"PRIu64
240 "\nTotal packets dropped: %15"PRIu64,
243 total_packets_dropped);
244 printf("\n====================================================\n");
247 /* Send the packet on an output interface */
249 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
251 struct rte_mbuf **m_table;
255 queueid = (uint16_t) qconf->tx_queue_id;
256 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
258 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
259 port_statistics[port].tx += ret;
260 if (unlikely(ret < n)) {
261 port_statistics[port].dropped += (n - ret);
263 rte_pktmbuf_free(m_table[ret]);
270 /* Send the packet on an output interface */
272 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
274 unsigned lcore_id, len;
275 struct lcore_queue_conf *qconf;
277 lcore_id = rte_lcore_id();
279 qconf = &lcore_queue_conf[lcore_id];
280 len = qconf->tx_mbufs[port].len;
281 qconf->tx_mbufs[port].m_table[len] = m;
284 /* enough pkts to be sent */
285 if (unlikely(len == MAX_PKT_BURST)) {
286 lsi_send_burst(qconf, MAX_PKT_BURST, port);
290 qconf->tx_mbufs[port].len = len;
295 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
297 struct ether_hdr *eth;
299 unsigned dst_port = lsi_dst_ports[portid];
301 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
303 /* 02:00:00:00:00:xx */
304 tmp = ð->d_addr.addr_bytes[0];
305 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
308 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
310 lsi_send_packet(m, (uint8_t) dst_port);
313 /* main processing loop */
317 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
320 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
321 unsigned i, j, portid, nb_rx;
322 struct lcore_queue_conf *qconf;
323 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
328 lcore_id = rte_lcore_id();
329 qconf = &lcore_queue_conf[lcore_id];
331 if (qconf->n_rx_port == 0) {
332 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
336 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
338 for (i = 0; i < qconf->n_rx_port; i++) {
340 portid = qconf->rx_port_list[i];
341 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
347 cur_tsc = rte_rdtsc();
350 * TX burst queue drain
352 diff_tsc = cur_tsc - prev_tsc;
353 if (unlikely(diff_tsc > drain_tsc)) {
355 /* this could be optimized (use queueid instead of
356 * portid), but it is not called so often */
357 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
358 if (qconf->tx_mbufs[portid].len == 0)
360 lsi_send_burst(&lcore_queue_conf[lcore_id],
361 qconf->tx_mbufs[portid].len,
363 qconf->tx_mbufs[portid].len = 0;
366 /* if timer is enabled */
367 if (timer_period > 0) {
369 /* advance the timer */
370 timer_tsc += diff_tsc;
372 /* if timer has reached its timeout */
373 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
375 /* do this only on master core */
376 if (lcore_id == rte_get_master_lcore()) {
378 /* reset the timer */
388 * Read packet from RX queues
390 for (i = 0; i < qconf->n_rx_port; i++) {
392 portid = qconf->rx_port_list[i];
393 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
394 pkts_burst, MAX_PKT_BURST);
396 port_statistics[portid].rx += nb_rx;
398 for (j = 0; j < nb_rx; j++) {
400 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
401 lsi_simple_forward(m, portid);
408 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
416 lsi_usage(const char *prgname)
418 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
419 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
420 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
421 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
426 lsi_parse_portmask(const char *portmask)
431 /* parse hexadecimal string */
432 pm = strtoul(portmask, &end, 16);
433 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
443 lsi_parse_nqueue(const char *q_arg)
448 /* parse hexadecimal string */
449 n = strtoul(q_arg, &end, 10);
450 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
454 if (n >= MAX_RX_QUEUE_PER_LCORE)
461 lsi_parse_timer_period(const char *q_arg)
466 /* parse number string */
467 n = strtol(q_arg, &end, 10);
468 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
470 if (n >= MAX_TIMER_PERIOD)
476 /* Parse the argument given in the command line of the application */
478 lsi_parse_args(int argc, char **argv)
483 char *prgname = argv[0];
484 static struct option lgopts[] = {
490 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
491 lgopts, &option_index)) != EOF) {
496 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
497 if (lsi_enabled_port_mask == 0) {
498 printf("invalid portmask\n");
506 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
507 if (lsi_rx_queue_per_lcore == 0) {
508 printf("invalid queue number\n");
516 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
517 if (timer_period < 0) {
518 printf("invalid timer period\n");
536 argv[optind-1] = prgname;
539 optind = 0; /* reset getopt lib */
544 * It will be called as the callback for specified port after a LSI interrupt
545 * has been fully handled. This callback needs to be implemented carefully as
546 * it will be called in the interrupt host thread which is different from the
547 * application main thread.
554 * Pointer to(address of) the parameters.
560 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
562 struct rte_eth_link link;
566 printf("\n\nIn registered callback...\n");
567 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
568 rte_eth_link_get_nowait(port_id, &link);
569 if (link.link_status) {
570 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
571 port_id, (unsigned)link.link_speed,
572 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
573 ("full-duplex") : ("half-duplex"));
575 printf("Port %d Link Down\n\n", port_id);
578 /* Check the link status of all ports in up to 9s, and print them finally */
580 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
582 #define CHECK_INTERVAL 100 /* 100ms */
583 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
584 uint8_t portid, count, all_ports_up, print_flag = 0;
585 struct rte_eth_link link;
587 printf("\nChecking link status");
589 for (count = 0; count <= MAX_CHECK_TIME; count++) {
591 for (portid = 0; portid < port_num; portid++) {
592 if ((port_mask & (1 << portid)) == 0)
594 memset(&link, 0, sizeof(link));
595 rte_eth_link_get_nowait(portid, &link);
596 /* print link status if flag set */
597 if (print_flag == 1) {
598 if (link.link_status)
599 printf("Port %d Link Up - speed %u "
600 "Mbps - %s\n", (uint8_t)portid,
601 (unsigned)link.link_speed,
602 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
603 ("full-duplex") : ("half-duplex\n"));
605 printf("Port %d Link Down\n",
609 /* clear all_ports_up flag if any link down */
610 if (link.link_status == 0) {
615 /* after finally printing all link status, get out */
619 if (all_ports_up == 0) {
622 rte_delay_ms(CHECK_INTERVAL);
625 /* set the print_flag if all ports up or timeout */
626 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
634 MAIN(int argc, char **argv)
636 struct lcore_queue_conf *qconf;
637 struct rte_eth_dev_info dev_info;
640 uint8_t portid, portid_last = 0;
641 unsigned lcore_id, rx_lcore_id;
642 unsigned nb_ports_in_mask = 0;
645 ret = rte_eal_init(argc, argv);
647 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
651 /* parse application arguments (after the EAL ones) */
652 ret = lsi_parse_args(argc, argv);
654 rte_exit(EXIT_FAILURE, "Invalid arguments");
656 /* create the mbuf pool */
658 rte_mempool_create("mbuf_pool", NB_MBUF,
660 sizeof(struct rte_pktmbuf_pool_private),
661 rte_pktmbuf_pool_init, NULL,
662 rte_pktmbuf_init, NULL,
664 if (lsi_pktmbuf_pool == NULL)
665 rte_panic("Cannot init mbuf pool\n");
668 if (rte_pmd_init_all() < 0)
669 rte_panic("Cannot init pmd\n");
671 if (rte_eal_pci_probe() < 0)
672 rte_panic("Cannot probe PCI\n");
674 nb_ports = rte_eth_dev_count();
676 rte_panic("No Ethernet port - bye\n");
678 if (nb_ports > RTE_MAX_ETHPORTS)
679 nb_ports = RTE_MAX_ETHPORTS;
682 * Each logical core is assigned a dedicated TX queue on each port.
684 for (portid = 0; portid < nb_ports; portid++) {
685 /* skip ports that are not enabled */
686 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
689 /* save the destination port id */
690 if (nb_ports_in_mask % 2) {
691 lsi_dst_ports[portid] = portid_last;
692 lsi_dst_ports[portid_last] = portid;
695 portid_last = portid;
699 rte_eth_dev_info_get(portid, &dev_info);
701 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
702 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
703 "but it should be even and at least 2\n",
707 qconf = &lcore_queue_conf[rx_lcore_id];
709 /* Initialize the port/queue configuration of each logical core */
710 for (portid = 0; portid < nb_ports; portid++) {
711 /* skip ports that are not enabled */
712 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
715 /* get the lcore_id for this port */
716 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
717 lcore_queue_conf[rx_lcore_id].n_rx_port ==
718 lsi_rx_queue_per_lcore) {
721 if (rx_lcore_id >= RTE_MAX_LCORE)
722 rte_exit(EXIT_FAILURE, "Not enough cores\n");
724 if (qconf != &lcore_queue_conf[rx_lcore_id])
725 /* Assigned a new logical core in the loop above. */
726 qconf = &lcore_queue_conf[rx_lcore_id];
728 qconf->rx_port_list[qconf->n_rx_port] = portid;
730 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
733 /* Initialise each port */
734 for (portid = 0; portid < nb_ports; portid++) {
735 /* skip ports that are not enabled */
736 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
737 printf("Skipping disabled port %u\n", (unsigned) portid);
741 printf("Initializing port %u... ", (unsigned) portid);
743 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
745 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
746 ret, (unsigned) portid);
748 /* register lsi interrupt callback, need to be after
749 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
750 * lsc interrupt will be present, and below callback to
751 * be registered will never be called.
753 rte_eth_dev_callback_register(portid,
754 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
756 rte_eth_macaddr_get(portid,
757 &lsi_ports_eth_addr[portid]);
759 /* init one RX queue */
761 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
762 rte_eth_dev_socket_id(portid), &rx_conf,
765 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
766 ret, (unsigned) portid);
768 /* init one TX queue logical core on each port */
770 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
771 rte_eth_dev_socket_id(portid), &tx_conf);
773 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
774 ret, (unsigned) portid);
777 ret = rte_eth_dev_start(portid);
779 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
780 ret, (unsigned) portid);
783 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
785 lsi_ports_eth_addr[portid].addr_bytes[0],
786 lsi_ports_eth_addr[portid].addr_bytes[1],
787 lsi_ports_eth_addr[portid].addr_bytes[2],
788 lsi_ports_eth_addr[portid].addr_bytes[3],
789 lsi_ports_eth_addr[portid].addr_bytes[4],
790 lsi_ports_eth_addr[portid].addr_bytes[5]);
792 /* initialize port stats */
793 memset(&port_statistics, 0, sizeof(port_statistics));
796 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
798 /* launch per-lcore init on every lcore */
799 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
800 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
801 if (rte_eal_wait_lcore(lcore_id) < 0)