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 */
109 * Configurable number of RX/TX ring descriptors
111 #define RTE_TEST_RX_DESC_DEFAULT 128
112 #define RTE_TEST_TX_DESC_DEFAULT 512
113 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
114 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
116 /* ethernet addresses of ports */
117 static struct ether_addr lsi_ports_eth_addr[LSI_MAX_PORTS];
119 /* mask of enabled ports */
120 static uint32_t lsi_enabled_port_mask = 0;
122 static unsigned int lsi_rx_queue_per_lcore = 1;
124 /* destination port for L2 forwarding */
125 static unsigned lsi_dst_ports[LSI_MAX_PORTS] = {0};
127 #define MAX_PKT_BURST 32
130 struct rte_mbuf *m_table[MAX_PKT_BURST];
133 #define MAX_RX_QUEUE_PER_LCORE 16
134 #define MAX_TX_QUEUE_PER_PORT 16
135 struct lcore_queue_conf {
137 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
138 unsigned tx_queue_id;
139 struct mbuf_table tx_mbufs[LSI_MAX_PORTS];
141 } __rte_cache_aligned;
142 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
144 static const struct rte_eth_conf port_conf = {
147 .header_split = 0, /**< Header Split disabled */
148 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
149 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
150 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
151 .hw_strip_crc = 0, /**< CRC stripped by hardware */
154 .mq_mode = ETH_MQ_TX_NONE,
157 .lsc = 1, /**< lsc interrupt feature enabled */
161 static const struct rte_eth_rxconf rx_conf = {
163 .pthresh = RX_PTHRESH,
164 .hthresh = RX_HTHRESH,
165 .wthresh = RX_WTHRESH,
169 static const struct rte_eth_txconf tx_conf = {
171 .pthresh = TX_PTHRESH,
172 .hthresh = TX_HTHRESH,
173 .wthresh = TX_WTHRESH,
175 .tx_free_thresh = 0, /* Use PMD default values */
176 .tx_rs_thresh = 0, /* Use PMD default values */
179 struct rte_mempool * lsi_pktmbuf_pool = NULL;
181 /* Per-port statistics struct */
182 struct lsi_port_statistics {
186 } __rte_cache_aligned;
187 struct lsi_port_statistics port_statistics[LSI_MAX_PORTS];
189 /* A tsc-based timer responsible for triggering statistics printout */
190 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
191 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
192 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
194 /* Print out statistics on packets dropped */
198 struct rte_eth_link link;
199 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
202 total_packets_dropped = 0;
203 total_packets_tx = 0;
204 total_packets_rx = 0;
206 const char clr[] = { 27, '[', '2', 'J', '\0' };
207 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
209 /* Clear screen and move to top left */
210 printf("%s%s", clr, topLeft);
212 printf("\nPort statistics ====================================");
214 for (portid = 0; portid < LSI_MAX_PORTS; portid++) {
215 /* skip ports that are not enabled */
216 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
219 memset(&link, 0, sizeof(link));
220 rte_eth_link_get_nowait((uint8_t)portid, &link);
221 printf("\nStatistics for port %u ------------------------------"
222 "\nLink status: %25s"
224 "\nLink duplex: %25s"
225 "\nPackets sent: %24"PRIu64
226 "\nPackets received: %20"PRIu64
227 "\nPackets dropped: %21"PRIu64,
229 (link.link_status ? "Link up" : "Link down"),
230 (unsigned)link.link_speed,
231 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
232 "full-duplex" : "half-duplex"),
233 port_statistics[portid].tx,
234 port_statistics[portid].rx,
235 port_statistics[portid].dropped);
237 total_packets_dropped += port_statistics[portid].dropped;
238 total_packets_tx += port_statistics[portid].tx;
239 total_packets_rx += port_statistics[portid].rx;
241 printf("\nAggregate statistics ==============================="
242 "\nTotal packets sent: %18"PRIu64
243 "\nTotal packets received: %14"PRIu64
244 "\nTotal packets dropped: %15"PRIu64,
247 total_packets_dropped);
248 printf("\n====================================================\n");
251 /* Send the packet on an output interface */
253 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
255 struct rte_mbuf **m_table;
259 queueid = (uint16_t) qconf->tx_queue_id;
260 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
262 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
263 port_statistics[port].tx += ret;
264 if (unlikely(ret < n)) {
265 port_statistics[port].dropped += (n - ret);
267 rte_pktmbuf_free(m_table[ret]);
274 /* Send the packet on an output interface */
276 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
278 unsigned lcore_id, len;
279 struct lcore_queue_conf *qconf;
281 lcore_id = rte_lcore_id();
283 qconf = &lcore_queue_conf[lcore_id];
284 len = qconf->tx_mbufs[port].len;
285 qconf->tx_mbufs[port].m_table[len] = m;
288 /* enough pkts to be sent */
289 if (unlikely(len == MAX_PKT_BURST)) {
290 lsi_send_burst(qconf, MAX_PKT_BURST, port);
294 qconf->tx_mbufs[port].len = len;
299 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
301 struct ether_hdr *eth;
303 unsigned dst_port = lsi_dst_ports[portid];
305 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
307 /* 02:00:00:00:00:xx */
308 tmp = ð->d_addr.addr_bytes[0];
309 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
312 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
314 lsi_send_packet(m, (uint8_t) dst_port);
317 /* main processing loop */
321 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
324 uint64_t prev_tsc = 0;
325 uint64_t diff_tsc, cur_tsc, timer_tsc;
326 unsigned i, j, portid, nb_rx;
327 struct lcore_queue_conf *qconf;
331 lcore_id = rte_lcore_id();
332 qconf = &lcore_queue_conf[lcore_id];
334 if (qconf->n_rx_port == 0) {
335 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
339 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
341 for (i = 0; i < qconf->n_rx_port; i++) {
343 portid = qconf->rx_port_list[i];
344 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
350 cur_tsc = rte_rdtsc();
353 * TX burst queue drain
355 diff_tsc = cur_tsc - prev_tsc;
356 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
358 /* this could be optimized (use queueid instead of
359 * portid), but it is not called so often */
360 for (portid = 0; portid < LSI_MAX_PORTS; portid++) {
361 if (qconf->tx_mbufs[portid].len == 0)
363 lsi_send_burst(&lcore_queue_conf[lcore_id],
364 qconf->tx_mbufs[portid].len,
366 qconf->tx_mbufs[portid].len = 0;
369 /* if timer is enabled */
370 if (timer_period > 0) {
372 /* advance the timer */
373 timer_tsc += diff_tsc;
375 /* if timer has reached its timeout */
376 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
378 /* do this only on master core */
379 if (lcore_id == rte_get_master_lcore()) {
381 /* reset the timer */
391 * Read packet from RX queues
393 for (i = 0; i < qconf->n_rx_port; i++) {
395 portid = qconf->rx_port_list[i];
396 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
397 pkts_burst, MAX_PKT_BURST);
399 port_statistics[portid].rx += nb_rx;
401 for (j = 0; j < nb_rx; j++) {
403 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
404 lsi_simple_forward(m, portid);
411 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
419 lsi_usage(const char *prgname)
421 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
422 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
423 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
424 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
429 lsi_parse_portmask(const char *portmask)
434 /* parse hexadecimal string */
435 pm = strtoul(portmask, &end, 16);
436 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
446 lsi_parse_nqueue(const char *q_arg)
451 /* parse hexadecimal string */
452 n = strtoul(q_arg, &end, 10);
453 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
457 if (n >= MAX_RX_QUEUE_PER_LCORE)
464 lsi_parse_timer_period(const char *q_arg)
469 /* parse number string */
470 n = strtol(q_arg, &end, 10);
471 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
473 if (n >= MAX_TIMER_PERIOD)
479 /* Parse the argument given in the command line of the application */
481 lsi_parse_args(int argc, char **argv)
486 char *prgname = argv[0];
487 static struct option lgopts[] = {
493 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
494 lgopts, &option_index)) != EOF) {
499 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
500 if (lsi_enabled_port_mask == 0) {
501 printf("invalid portmask\n");
509 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
510 if (lsi_rx_queue_per_lcore == 0) {
511 printf("invalid queue number\n");
519 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
520 if (timer_period < 0) {
521 printf("invalid timer period\n");
539 argv[optind-1] = prgname;
542 optind = 0; /* reset getopt lib */
547 * It will be called as the callback for specified port after a LSI interrupt
548 * has been fully handled. This callback needs to be implemented carefully as
549 * it will be called in the interrupt host thread which is different from the
550 * application main thread.
557 * Pointer to(address of) the parameters.
563 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
565 struct rte_eth_link link;
569 printf("\n\nIn registered callback...\n");
570 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
571 rte_eth_link_get_nowait(port_id, &link);
572 if (link.link_status) {
573 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
574 port_id, (unsigned)link.link_speed,
575 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
576 ("full-duplex") : ("half-duplex"));
578 printf("Port %d Link Down\n\n", port_id);
581 /* Check the link status of all ports in up to 9s, and print them finally */
583 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
585 #define CHECK_INTERVAL 100 /* 100ms */
586 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
587 uint8_t portid, count, all_ports_up, print_flag = 0;
588 struct rte_eth_link link;
590 printf("\nChecking link status");
592 for (count = 0; count <= MAX_CHECK_TIME; count++) {
594 for (portid = 0; portid < port_num; portid++) {
595 if ((port_mask & (1 << portid)) == 0)
597 memset(&link, 0, sizeof(link));
598 rte_eth_link_get_nowait(portid, &link);
599 /* print link status if flag set */
600 if (print_flag == 1) {
601 if (link.link_status)
602 printf("Port %d Link Up - speed %u "
603 "Mbps - %s\n", (uint8_t)portid,
604 (unsigned)link.link_speed,
605 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
606 ("full-duplex") : ("half-duplex\n"));
608 printf("Port %d Link Down\n",
612 /* clear all_ports_up flag if any link down */
613 if (link.link_status == 0) {
618 /* after finally printing all link status, get out */
622 if (all_ports_up == 0) {
625 rte_delay_ms(CHECK_INTERVAL);
628 /* set the print_flag if all ports up or timeout */
629 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
637 MAIN(int argc, char **argv)
639 struct lcore_queue_conf *qconf;
640 struct rte_eth_dev_info dev_info;
643 uint8_t portid, portid_last = 0;
644 unsigned lcore_id, rx_lcore_id;
645 unsigned nb_ports_in_mask = 0;
648 ret = rte_eal_init(argc, argv);
650 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
654 /* parse application arguments (after the EAL ones) */
655 ret = lsi_parse_args(argc, argv);
657 rte_exit(EXIT_FAILURE, "Invalid arguments");
659 /* create the mbuf pool */
661 rte_mempool_create("mbuf_pool", NB_MBUF,
663 sizeof(struct rte_pktmbuf_pool_private),
664 rte_pktmbuf_pool_init, NULL,
665 rte_pktmbuf_init, NULL,
667 if (lsi_pktmbuf_pool == NULL)
668 rte_panic("Cannot init mbuf pool\n");
671 if (rte_pmd_init_all() < 0)
672 rte_panic("Cannot init pmd\n");
674 if (rte_eal_pci_probe() < 0)
675 rte_panic("Cannot probe PCI\n");
677 nb_ports = rte_eth_dev_count();
679 rte_panic("No Ethernet port - bye\n");
681 if (nb_ports > LSI_MAX_PORTS)
682 nb_ports = LSI_MAX_PORTS;
685 * Each logical core is assigned a dedicated TX queue on each port.
687 for (portid = 0; portid < nb_ports; portid++) {
688 /* skip ports that are not enabled */
689 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
692 /* save the destination port id */
693 if (nb_ports_in_mask % 2) {
694 lsi_dst_ports[portid] = portid_last;
695 lsi_dst_ports[portid_last] = portid;
698 portid_last = portid;
702 rte_eth_dev_info_get(portid, &dev_info);
704 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
705 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
706 "but it should be even and at least 2\n",
710 qconf = &lcore_queue_conf[rx_lcore_id];
712 /* Initialize the port/queue configuration of each logical core */
713 for (portid = 0; portid < nb_ports; portid++) {
714 /* skip ports that are not enabled */
715 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
718 /* get the lcore_id for this port */
719 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
720 lcore_queue_conf[rx_lcore_id].n_rx_port ==
721 lsi_rx_queue_per_lcore) {
724 if (rx_lcore_id >= RTE_MAX_LCORE)
725 rte_exit(EXIT_FAILURE, "Not enough cores\n");
727 if (qconf != &lcore_queue_conf[rx_lcore_id])
728 /* Assigned a new logical core in the loop above. */
729 qconf = &lcore_queue_conf[rx_lcore_id];
731 qconf->rx_port_list[qconf->n_rx_port] = portid;
733 printf("Lcore %u: RX port %u\n",rx_lcore_id, (unsigned) portid);
736 /* Initialise each port */
737 for (portid = 0; portid < nb_ports; portid++) {
738 /* skip ports that are not enabled */
739 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
740 printf("Skipping disabled port %u\n", (unsigned) portid);
744 printf("Initializing port %u... ", (unsigned) portid);
746 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
748 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
749 ret, (unsigned) portid);
751 /* register lsi interrupt callback, need to be after
752 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
753 * lsc interrupt will be present, and below callback to
754 * be registered will never be called.
756 rte_eth_dev_callback_register(portid,
757 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
759 rte_eth_macaddr_get(portid,
760 &lsi_ports_eth_addr[portid]);
762 /* init one RX queue */
764 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
768 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, port=%u\n",
769 ret, (unsigned) portid);
771 /* init one TX queue logical core on each port */
773 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
776 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d,port=%u\n",
777 ret, (unsigned) portid);
780 ret = rte_eth_dev_start(portid);
782 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
783 ret, (unsigned) portid);
786 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
788 lsi_ports_eth_addr[portid].addr_bytes[0],
789 lsi_ports_eth_addr[portid].addr_bytes[1],
790 lsi_ports_eth_addr[portid].addr_bytes[2],
791 lsi_ports_eth_addr[portid].addr_bytes[3],
792 lsi_ports_eth_addr[portid].addr_bytes[4],
793 lsi_ports_eth_addr[portid].addr_bytes[5]);
795 /* initialize port stats */
796 memset(&port_statistics, 0, sizeof(port_statistics));
799 check_all_ports_link_status(nb_ports, lsi_enabled_port_mask);
801 /* launch per-lcore init on every lcore */
802 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
803 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
804 if (rte_eal_wait_lcore(lcore_id) < 0)