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 LSI_MAX_PORTS 32
81 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
85 * RX and TX Prefetch, Host, and Write-back threshold values should be
86 * carefully set for optimal performance. Consult the network
87 * controller's datasheet and supporting DPDK documentation for guidance
88 * on how these parameters should be set.
90 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
91 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
92 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
95 * These default values are optimized for use with the Intel(R) 82599 10 GbE
96 * Controller and the DPDK ixgbe PMD. Consider using other values for other
97 * network controllers and/or network drivers.
99 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
100 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
101 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
103 #define MAX_PKT_BURST 32
104 #define BURST_TX_DRAIN 200000ULL /* around 100us at 2 Ghz */
107 * Configurable number of RX/TX ring descriptors
109 #define RTE_TEST_RX_DESC_DEFAULT 128
110 #define RTE_TEST_TX_DESC_DEFAULT 512
111 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
112 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
114 /* ethernet addresses of ports */
115 static struct ether_addr lsi_ports_eth_addr[LSI_MAX_PORTS];
117 /* mask of enabled ports */
118 static uint32_t lsi_enabled_port_mask = 0;
120 static unsigned int lsi_rx_queue_per_lcore = 1;
122 /* destination port for L2 forwarding */
123 static unsigned lsi_dst_ports[LSI_MAX_PORTS] = {0};
125 #define MAX_PKT_BURST 32
128 struct rte_mbuf *m_table[MAX_PKT_BURST];
131 #define MAX_RX_QUEUE_PER_LCORE 16
132 #define MAX_TX_QUEUE_PER_PORT 16
133 struct lcore_queue_conf {
135 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
136 unsigned tx_queue_id;
137 struct mbuf_table tx_mbufs[LSI_MAX_PORTS];
139 } __rte_cache_aligned;
140 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
142 static const struct rte_eth_conf port_conf = {
145 .header_split = 0, /**< Header Split disabled */
146 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
147 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
148 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
149 .hw_strip_crc = 0, /**< CRC stripped by hardware */
152 .mq_mode = ETH_MQ_TX_NONE,
155 .lsc = 1, /**< lsc interrupt feature enabled */
159 static const struct rte_eth_rxconf rx_conf = {
161 .pthresh = RX_PTHRESH,
162 .hthresh = RX_HTHRESH,
163 .wthresh = RX_WTHRESH,
167 static const struct rte_eth_txconf tx_conf = {
169 .pthresh = TX_PTHRESH,
170 .hthresh = TX_HTHRESH,
171 .wthresh = TX_WTHRESH,
173 .tx_free_thresh = 0, /* Use PMD default values */
174 .tx_rs_thresh = 0, /* Use PMD default values */
177 struct rte_mempool * lsi_pktmbuf_pool = NULL;
179 /* Per-port statistics struct */
180 struct lsi_port_statistics {
184 } __rte_cache_aligned;
185 struct lsi_port_statistics port_statistics[LSI_MAX_PORTS];
187 /* A tsc-based timer responsible for triggering statistics printout */
188 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
189 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
190 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
192 /* Print out statistics on packets dropped */
196 struct rte_eth_link link;
197 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
200 total_packets_dropped = 0;
201 total_packets_tx = 0;
202 total_packets_rx = 0;
204 const char clr[] = { 27, '[', '2', 'J', '\0' };
205 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
207 /* Clear screen and move to top left */
208 printf("%s%s", clr, topLeft);
210 printf("\nPort statistics ====================================");
212 for (portid = 0; portid < LSI_MAX_PORTS; portid++) {
213 /* skip ports that are not enabled */
214 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
217 memset(&link, 0, sizeof(link));
218 rte_eth_link_get_nowait((uint8_t)portid, &link);
219 printf("\nStatistics for port %u ------------------------------"
220 "\nLink status: %25s"
222 "\nLink duplex: %25s"
223 "\nPackets sent: %24"PRIu64
224 "\nPackets received: %20"PRIu64
225 "\nPackets dropped: %21"PRIu64,
227 (link.link_status ? "Link up" : "Link down"),
228 (unsigned)link.link_speed,
229 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
230 "full-duplex" : "half-duplex"),
231 port_statistics[portid].tx,
232 port_statistics[portid].rx,
233 port_statistics[portid].dropped);
235 total_packets_dropped += port_statistics[portid].dropped;
236 total_packets_tx += port_statistics[portid].tx;
237 total_packets_rx += port_statistics[portid].rx;
239 printf("\nAggregate statistics ==============================="
240 "\nTotal packets sent: %18"PRIu64
241 "\nTotal packets received: %14"PRIu64
242 "\nTotal packets dropped: %15"PRIu64,
245 total_packets_dropped);
246 printf("\n====================================================\n");
249 /* Send the packet on an output interface */
251 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
253 struct rte_mbuf **m_table;
257 queueid = (uint16_t) qconf->tx_queue_id;
258 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
260 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
261 port_statistics[port].tx += ret;
262 if (unlikely(ret < n)) {
263 port_statistics[port].dropped += (n - ret);
265 rte_pktmbuf_free(m_table[ret]);
272 /* Send the packet on an output interface */
274 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
276 unsigned lcore_id, len;
277 struct lcore_queue_conf *qconf;
279 lcore_id = rte_lcore_id();
281 qconf = &lcore_queue_conf[lcore_id];
282 len = qconf->tx_mbufs[port].len;
283 qconf->tx_mbufs[port].m_table[len] = m;
286 /* enough pkts to be sent */
287 if (unlikely(len == MAX_PKT_BURST)) {
288 lsi_send_burst(qconf, MAX_PKT_BURST, port);
292 qconf->tx_mbufs[port].len = len;
297 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
299 struct ether_hdr *eth;
301 unsigned dst_port = lsi_dst_ports[portid];
303 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
305 /* 02:00:00:00:00:xx */
306 tmp = ð->d_addr.addr_bytes[0];
307 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
310 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
312 lsi_send_packet(m, (uint8_t) dst_port);
315 /* main processing loop */
319 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
322 uint64_t prev_tsc = 0;
323 uint64_t diff_tsc, cur_tsc, timer_tsc;
324 unsigned i, j, portid, nb_rx;
325 struct lcore_queue_conf *qconf;
329 lcore_id = rte_lcore_id();
330 qconf = &lcore_queue_conf[lcore_id];
332 if (qconf->n_rx_port == 0) {
333 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
337 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
339 for (i = 0; i < qconf->n_rx_port; i++) {
341 portid = qconf->rx_port_list[i];
342 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
348 cur_tsc = rte_rdtsc();
351 * TX burst queue drain
353 diff_tsc = cur_tsc - prev_tsc;
354 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
356 /* this could be optimized (use queueid instead of
357 * portid), but it is not called so often */
358 for (portid = 0; portid < LSI_MAX_PORTS; portid++) {
359 if (qconf->tx_mbufs[portid].len == 0)
361 lsi_send_burst(&lcore_queue_conf[lcore_id],
362 qconf->tx_mbufs[portid].len,
364 qconf->tx_mbufs[portid].len = 0;
367 /* if timer is enabled */
368 if (timer_period > 0) {
370 /* advance the timer */
371 timer_tsc += diff_tsc;
373 /* if timer has reached its timeout */
374 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
376 /* do this only on master core */
377 if (lcore_id == rte_get_master_lcore()) {
379 /* reset the timer */
389 * Read packet from RX queues
391 for (i = 0; i < qconf->n_rx_port; i++) {
393 portid = qconf->rx_port_list[i];
394 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
395 pkts_burst, MAX_PKT_BURST);
397 port_statistics[portid].rx += nb_rx;
399 for (j = 0; j < nb_rx; j++) {
401 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
402 lsi_simple_forward(m, portid);
409 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
417 lsi_usage(const char *prgname)
419 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
420 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
421 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
422 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
427 lsi_parse_portmask(const char *portmask)
432 /* parse hexadecimal string */
433 pm = strtoul(portmask, &end, 16);
434 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
444 lsi_parse_nqueue(const char *q_arg)
449 /* parse hexadecimal string */
450 n = strtoul(q_arg, &end, 10);
451 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
455 if (n >= MAX_RX_QUEUE_PER_LCORE)
462 lsi_parse_timer_period(const char *q_arg)
467 /* parse number string */
468 n = strtol(q_arg, &end, 10);
469 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
471 if (n >= MAX_TIMER_PERIOD)
477 /* Parse the argument given in the command line of the application */
479 lsi_parse_args(int argc, char **argv)
484 char *prgname = argv[0];
485 static struct option lgopts[] = {
491 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
492 lgopts, &option_index)) != EOF) {
497 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
498 if (lsi_enabled_port_mask == 0) {
499 printf("invalid portmask\n");
507 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
508 if (lsi_rx_queue_per_lcore == 0) {
509 printf("invalid queue number\n");
517 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
518 if (timer_period < 0) {
519 printf("invalid timer period\n");
537 argv[optind-1] = prgname;
540 optind = 0; /* reset getopt lib */
545 * It will be called as the callback for specified port after a LSI interrupt
546 * has been fully handled. This callback needs to be implemented carefully as
547 * it will be called in the interrupt host thread which is different from the
548 * application main thread.
555 * Pointer to(address of) the parameters.
561 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
563 struct rte_eth_link link;
567 printf("\n\nIn registered callback...\n");
568 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
569 rte_eth_link_get_nowait(port_id, &link);
570 if (link.link_status) {
571 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
572 port_id, (unsigned)link.link_speed,
573 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
574 ("full-duplex") : ("half-duplex"));
576 printf("Port %d Link Down\n\n", port_id);
579 /* Check the link status of all ports in up to 9s, and print them finally */
581 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
583 #define CHECK_INTERVAL 100 /* 100ms */
584 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
585 uint8_t portid, count, all_ports_up, print_flag = 0;
586 struct rte_eth_link link;
588 printf("\nChecking link status");
590 for (count = 0; count <= MAX_CHECK_TIME; count++) {
592 for (portid = 0; portid < port_num; portid++) {
593 if ((port_mask & (1 << portid)) == 0)
595 memset(&link, 0, sizeof(link));
596 rte_eth_link_get_nowait(portid, &link);
597 /* print link status if flag set */
598 if (print_flag == 1) {
599 if (link.link_status)
600 printf("Port %d Link Up - speed %u "
601 "Mbps - %s\n", (uint8_t)portid,
602 (unsigned)link.link_speed,
603 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
604 ("full-duplex") : ("half-duplex\n"));
606 printf("Port %d Link Down\n",
610 /* clear all_ports_up flag if any link down */
611 if (link.link_status == 0) {
616 /* after finally printing all link status, get out */
620 if (all_ports_up == 0) {
623 rte_delay_ms(CHECK_INTERVAL);
626 /* set the print_flag if all ports up or timeout */
627 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
635 MAIN(int argc, char **argv)
637 struct lcore_queue_conf *qconf;
638 struct rte_eth_dev_info dev_info;
641 uint8_t portid, portid_last = 0;
642 unsigned lcore_id, rx_lcore_id;
643 unsigned nb_ports_in_mask = 0;
646 ret = rte_eal_init(argc, argv);
648 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
652 /* parse application arguments (after the EAL ones) */
653 ret = lsi_parse_args(argc, argv);
655 rte_exit(EXIT_FAILURE, "Invalid arguments");
657 /* create the mbuf pool */
659 rte_mempool_create("mbuf_pool", NB_MBUF,
661 sizeof(struct rte_pktmbuf_pool_private),
662 rte_pktmbuf_pool_init, NULL,
663 rte_pktmbuf_init, NULL,
665 if (lsi_pktmbuf_pool == NULL)
666 rte_panic("Cannot init mbuf pool\n");
669 if (rte_pmd_init_all() < 0)
670 rte_panic("Cannot init pmd\n");
672 if (rte_eal_pci_probe() < 0)
673 rte_panic("Cannot probe PCI\n");
675 nb_ports = rte_eth_dev_count();
677 rte_panic("No Ethernet port - bye\n");
679 if (nb_ports > LSI_MAX_PORTS)
680 nb_ports = LSI_MAX_PORTS;
683 * Each logical core is assigned a dedicated TX queue on each port.
685 for (portid = 0; portid < nb_ports; portid++) {
686 /* skip ports that are not enabled */
687 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
690 /* save the destination port id */
691 if (nb_ports_in_mask % 2) {
692 lsi_dst_ports[portid] = portid_last;
693 lsi_dst_ports[portid_last] = portid;
696 portid_last = portid;
700 rte_eth_dev_info_get(portid, &dev_info);
702 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
703 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
704 "but it should be even and at least 2\n",
708 qconf = &lcore_queue_conf[rx_lcore_id];
710 /* Initialize the port/queue configuration of each logical core */
711 for (portid = 0; portid < nb_ports; portid++) {
712 /* skip ports that are not enabled */
713 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
716 /* get the lcore_id for this port */
717 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
718 lcore_queue_conf[rx_lcore_id].n_rx_port ==
719 lsi_rx_queue_per_lcore) {
722 if (rx_lcore_id >= RTE_MAX_LCORE)
723 rte_exit(EXIT_FAILURE, "Not enough cores\n");
725 if (qconf != &lcore_queue_conf[rx_lcore_id])
726 /* Assigned a new logical core in the loop above. */
727 qconf = &lcore_queue_conf[rx_lcore_id];
729 qconf->rx_port_list[qconf->n_rx_port] = portid;
731 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
734 /* Initialise each port */
735 for (portid = 0; portid < nb_ports; portid++) {
736 /* skip ports that are not enabled */
737 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
738 printf("Skipping disabled port %u\n", (unsigned) portid);
742 printf("Initializing port %u... ", (unsigned) portid);
744 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
746 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
747 ret, (unsigned) portid);
749 /* register lsi interrupt callback, need to be after
750 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
751 * lsc interrupt will be present, and below callback to
752 * be registered will never be called.
754 rte_eth_dev_callback_register(portid,
755 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
757 rte_eth_macaddr_get(portid,
758 &lsi_ports_eth_addr[portid]);
760 /* init one RX queue */
762 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
763 rte_eth_dev_socket_id(portid), &rx_conf,
766 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
767 ret, (unsigned) portid);
769 /* init one TX queue logical core on each port */
771 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
772 rte_eth_dev_socket_id(portid), &tx_conf);
774 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
775 ret, (unsigned) portid);
778 ret = rte_eth_dev_start(portid);
780 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
781 ret, (unsigned) portid);
784 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
786 lsi_ports_eth_addr[portid].addr_bytes[0],
787 lsi_ports_eth_addr[portid].addr_bytes[1],
788 lsi_ports_eth_addr[portid].addr_bytes[2],
789 lsi_ports_eth_addr[portid].addr_bytes[3],
790 lsi_ports_eth_addr[portid].addr_bytes[4],
791 lsi_ports_eth_addr[portid].addr_bytes[5]);
793 /* initialize port stats */
794 memset(&port_statistics, 0, sizeof(port_statistics));
797 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
799 /* launch per-lcore init on every lcore */
800 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
801 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
802 if (rte_eal_wait_lcore(lcore_id) < 0)