4 * Copyright(c) 2010-2012 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_queue_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 */
156 .lsc = 1, /**< lsc interrupt feature enabled */
160 static const struct rte_eth_rxconf rx_conf = {
162 .pthresh = RX_PTHRESH,
163 .hthresh = RX_HTHRESH,
164 .wthresh = RX_WTHRESH,
168 static const struct rte_eth_txconf tx_conf = {
170 .pthresh = TX_PTHRESH,
171 .hthresh = TX_HTHRESH,
172 .wthresh = TX_WTHRESH,
174 .tx_free_thresh = 0, /* Use PMD default values */
175 .tx_rs_thresh = 0, /* Use PMD default values */
178 struct rte_mempool * lsi_pktmbuf_pool = NULL;
180 /* Per-port statistics struct */
181 struct lsi_port_statistics {
185 } __rte_cache_aligned;
186 struct lsi_port_statistics port_statistics[LSI_MAX_PORTS];
188 /* A tsc-based timer responsible for triggering statistics printout */
189 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
190 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
191 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
193 /* Print out statistics on packets dropped */
197 struct rte_eth_link link;
198 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
201 total_packets_dropped = 0;
202 total_packets_tx = 0;
203 total_packets_rx = 0;
205 const char clr[] = { 27, '[', '2', 'J', '\0' };
206 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
208 /* Clear screen and move to top left */
209 printf("%s%s", clr, topLeft);
211 printf("\nPort statistics ====================================");
213 for (portid = 0; portid < LSI_MAX_PORTS; portid++) {
214 /* skip ports that are not enabled */
215 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
218 memset(&link, 0, sizeof(link));
219 rte_eth_link_get_nowait((uint8_t)portid, &link);
220 printf("\nStatistics for port %u ------------------------------"
221 "\nLink status: %25s"
223 "\nLink duplex: %25s"
224 "\nPackets sent: %24"PRIu64
225 "\nPackets received: %20"PRIu64
226 "\nPackets dropped: %21"PRIu64,
228 (link.link_status ? "Link up" : "Link down"),
229 (unsigned)link.link_speed,
230 (link.link_duplex == ETH_LINK_FULL_DUPLEX ? \
231 "full-duplex" : "half-duplex"),
232 port_statistics[portid].tx,
233 port_statistics[portid].rx,
234 port_statistics[portid].dropped);
236 total_packets_dropped += port_statistics[portid].dropped;
237 total_packets_tx += port_statistics[portid].tx;
238 total_packets_rx += port_statistics[portid].rx;
240 printf("\nAggregate statistics ==============================="
241 "\nTotal packets sent: %18"PRIu64
242 "\nTotal packets received: %14"PRIu64
243 "\nTotal packets dropped: %15"PRIu64,
246 total_packets_dropped);
247 printf("\n====================================================\n");
250 /* Send the packet on an output interface */
252 lsi_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
254 struct rte_mbuf **m_table;
258 queueid = (uint16_t) qconf->tx_queue_id;
259 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
261 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
262 port_statistics[port].tx += ret;
263 if (unlikely(ret < n)) {
264 port_statistics[port].dropped += (n - ret);
266 rte_pktmbuf_free(m_table[ret]);
273 /* Send the packet on an output interface */
275 lsi_send_packet(struct rte_mbuf *m, uint8_t port)
277 unsigned lcore_id, len;
278 struct lcore_queue_conf *qconf;
280 lcore_id = rte_lcore_id();
282 qconf = &lcore_queue_conf[lcore_id];
283 len = qconf->tx_mbufs[port].len;
284 qconf->tx_mbufs[port].m_table[len] = m;
287 /* enough pkts to be sent */
288 if (unlikely(len == MAX_PKT_BURST)) {
289 lsi_send_burst(qconf, MAX_PKT_BURST, port);
293 qconf->tx_mbufs[port].len = len;
298 lsi_simple_forward(struct rte_mbuf *m, unsigned portid)
300 struct ether_hdr *eth;
302 unsigned dst_port = lsi_dst_ports[portid];
304 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
306 /* 00:09:c0:00:00:xx */
307 tmp = ð->d_addr.addr_bytes[0];
308 *((uint64_t *)tmp) = 0x000000c00900 + (dst_port << 24);
311 ether_addr_copy(&lsi_ports_eth_addr[dst_port], ð->s_addr);
313 lsi_send_packet(m, (uint8_t) dst_port);
316 /* main processing loop */
320 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
323 uint64_t prev_tsc = 0;
324 uint64_t diff_tsc, cur_tsc, timer_tsc;
325 unsigned i, j, portid, nb_rx;
326 struct lcore_queue_conf *qconf;
330 lcore_id = rte_lcore_id();
331 qconf = &lcore_queue_conf[lcore_id];
333 if (qconf->n_rx_queue == 0) {
334 RTE_LOG(INFO, LSI, "lcore %u has nothing to do\n", lcore_id);
338 RTE_LOG(INFO, LSI, "entering main loop on lcore %u\n", lcore_id);
340 for (i = 0; i < qconf->n_rx_queue; i++) {
342 portid = qconf->rx_queue_list[i];
343 RTE_LOG(INFO, LSI, " -- lcoreid=%u portid=%u\n", lcore_id,
349 cur_tsc = rte_rdtsc();
352 * TX burst queue drain
354 diff_tsc = cur_tsc - prev_tsc;
355 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
357 /* this could be optimized (use queueid instead of
358 * portid), but it is not called so often */
359 for (portid = 0; portid < LSI_MAX_PORTS; portid++) {
360 if (qconf->tx_mbufs[portid].len == 0)
362 lsi_send_burst(&lcore_queue_conf[lcore_id],
363 qconf->tx_mbufs[portid].len,
365 qconf->tx_mbufs[portid].len = 0;
368 /* if timer is enabled */
369 if (timer_period > 0) {
371 /* advance the timer */
372 timer_tsc += diff_tsc;
374 /* if timer has reached its timeout */
375 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
377 /* do this only on master core */
378 if (lcore_id == rte_get_master_lcore()) {
380 /* reset the timer */
390 * Read packet from RX queues
392 for (i = 0; i < qconf->n_rx_queue; i++) {
394 portid = qconf->rx_queue_list[i];
395 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
396 pkts_burst, MAX_PKT_BURST);
398 port_statistics[portid].rx += nb_rx;
400 for (j = 0; j < nb_rx; j++) {
402 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
403 lsi_simple_forward(m, portid);
410 lsi_launch_one_lcore(__attribute__((unused)) void *dummy)
418 lsi_usage(const char *prgname)
420 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
421 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
422 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
423 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
428 lsi_parse_portmask(const char *portmask)
433 /* parse hexadecimal string */
434 pm = strtoul(portmask, &end, 16);
435 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
445 lsi_parse_nqueue(const char *q_arg)
450 /* parse hexadecimal string */
451 n = strtoul(q_arg, &end, 10);
452 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
456 if (n >= MAX_RX_QUEUE_PER_LCORE)
463 lsi_parse_timer_period(const char *q_arg)
468 /* parse number string */
469 n = strtol(q_arg, &end, 10);
470 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
472 if (n >= MAX_TIMER_PERIOD)
478 /* Parse the argument given in the command line of the application */
480 lsi_parse_args(int argc, char **argv)
485 char *prgname = argv[0];
486 static struct option lgopts[] = {
492 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
493 lgopts, &option_index)) != EOF) {
498 lsi_enabled_port_mask = lsi_parse_portmask(optarg);
499 if (lsi_enabled_port_mask == 0) {
500 printf("invalid portmask\n");
508 lsi_rx_queue_per_lcore = lsi_parse_nqueue(optarg);
509 if (lsi_rx_queue_per_lcore == 0) {
510 printf("invalid queue number\n");
518 timer_period = lsi_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
519 if (timer_period < 0) {
520 printf("invalid timer period\n");
538 argv[optind-1] = prgname;
541 optind = 0; /* reset getopt lib */
546 * It will be called as the callback for specified port after a LSI interrupt
547 * has been fully handled. This callback needs to be implemented carefully as
548 * it will be called in the interrupt host thread which is different from the
549 * application main thread.
556 * Pointer to(address of) the parameters.
562 lsi_event_callback(uint8_t port_id, enum rte_eth_event_type type, void *param)
564 struct rte_eth_link link;
568 printf("\n\nIn registered callback...\n");
569 printf("Event type: %s\n", type == RTE_ETH_EVENT_INTR_LSC ? "LSC interrupt" : "unknown event");
570 rte_eth_link_get(port_id, &link);
571 if (link.link_status) {
572 printf("Port %d Link Up - speed %u Mbps - %s\n\n",
573 port_id, (unsigned)link.link_speed,
574 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
575 ("full-duplex") : ("half-duplex"));
577 printf("Port %d Link Down\n\n", port_id);
581 MAIN(int argc, char **argv)
583 struct lcore_queue_conf *qconf;
584 struct rte_eth_dev_info dev_info;
585 struct rte_eth_link link;
587 unsigned int nb_ports, nb_lcores;
588 unsigned portid, portid_last = 0, queueid = 0;
589 unsigned lcore_id, rx_lcore_id;
590 unsigned n_tx_queue, max_tx_queues;
591 unsigned nb_ports_in_mask = 0;
594 ret = rte_eal_init(argc, argv);
596 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
600 /* parse application arguments (after the EAL ones) */
601 ret = lsi_parse_args(argc, argv);
603 rte_exit(EXIT_FAILURE, "Invalid arguments");
605 /* create the mbuf pool */
607 rte_mempool_create("mbuf_pool", NB_MBUF,
609 sizeof(struct rte_pktmbuf_pool_private),
610 rte_pktmbuf_pool_init, NULL,
611 rte_pktmbuf_init, NULL,
613 if (lsi_pktmbuf_pool == NULL)
614 rte_panic("Cannot init mbuf pool\n");
617 #ifdef RTE_LIBRTE_IGB_PMD
618 if (rte_igb_pmd_init() < 0)
619 rte_panic("Cannot init igb pmd\n");
621 #ifdef RTE_LIBRTE_IXGBE_PMD
622 if (rte_ixgbe_pmd_init() < 0)
623 rte_panic("Cannot init ixgbe pmd\n");
626 if (rte_eal_pci_probe() < 0)
627 rte_panic("Cannot probe PCI\n");
629 nb_ports = rte_eth_dev_count();
631 rte_panic("No Ethernet port - bye\n");
633 if (nb_ports > LSI_MAX_PORTS)
634 nb_ports = LSI_MAX_PORTS;
636 nb_lcores = rte_lcore_count();
639 * Each logical core is assigned a dedicated TX queue on each port.
640 * Compute the maximum number of TX queues that can be used.
642 max_tx_queues = nb_lcores;
643 for (portid = 0; portid < nb_ports; portid++) {
644 /* skip ports that are not enabled */
645 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
648 /* save the destination port id */
649 if (nb_ports_in_mask % 2) {
650 lsi_dst_ports[portid] = portid_last;
651 lsi_dst_ports[portid_last] = portid;
654 portid_last = portid;
658 rte_eth_dev_info_get((uint8_t) portid, &dev_info);
659 if (max_tx_queues > dev_info.max_tx_queues)
660 max_tx_queues = dev_info.max_tx_queues;
663 if (nb_ports_in_mask < 2 || nb_ports_in_mask % 2)
664 rte_exit(EXIT_FAILURE, "Current enabled port number is %u, "
665 "but it should be even and at least 2\n",
669 qconf = &lcore_queue_conf[rx_lcore_id];
670 qconf->tx_queue_id = 0;
673 /* Initialize the port/queue configuration of each logical core */
674 for (portid = 0; portid < nb_ports; portid++) {
675 /* skip ports that are not enabled */
676 if ((lsi_enabled_port_mask & (1 << portid)) == 0)
679 /* get the lcore_id for this port */
680 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
681 lcore_queue_conf[rx_lcore_id].n_rx_queue ==
682 lsi_rx_queue_per_lcore) {
685 if (rx_lcore_id >= RTE_MAX_LCORE)
686 rte_exit(EXIT_FAILURE, "Not enough cores\n");
687 if (n_tx_queue == max_tx_queues)
688 rte_exit(EXIT_FAILURE, "Not enough TX queues\n");
690 if (qconf != &lcore_queue_conf[rx_lcore_id]) {
691 /* Assigned a new logical core in the loop above. */
692 qconf = &lcore_queue_conf[rx_lcore_id];
693 qconf->tx_queue_id = n_tx_queue;
696 qconf->rx_queue_list[qconf->n_rx_queue] = portid;
698 printf("Lcore %u: RX port %u TX queue %u\n",
699 rx_lcore_id, portid, qconf->tx_queue_id);
702 /* Initialise each port */
703 for (portid = 0; portid < nb_ports; portid++) {
705 /* skip ports that are not enabled */
706 if ((lsi_enabled_port_mask & (1 << portid)) == 0) {
707 printf("Skipping disabled port %u\n", portid);
711 printf("Initializing port %u... ", portid);
713 ret = rte_eth_dev_configure((uint8_t) portid, 1,
714 (uint16_t) n_tx_queue, &port_conf);
716 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
719 /* register lsi interrupt callback, need to be after
720 * rte_eth_dev_configure(). if (intr_conf.lsc == 0), no
721 * lsc interrupt will be present, and below callback to
722 * be registered will never be called.
724 rte_eth_dev_callback_register((uint8_t)portid,
725 RTE_ETH_EVENT_INTR_LSC, lsi_event_callback, NULL);
727 rte_eth_macaddr_get((uint8_t) portid,
728 &lsi_ports_eth_addr[portid]);
730 /* init one RX queue */
732 ret = rte_eth_rx_queue_setup((uint8_t) portid, 0, nb_rxd,
736 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%u\n",
739 /* init one TX queue logical core on each port */
740 for (queueid = 0; queueid < n_tx_queue; queueid++) {
742 ret = rte_eth_tx_queue_setup((uint8_t) portid,
743 (uint16_t) queueid, nb_txd,
746 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
747 "port=%u queue=%u\n",
748 ret, portid, queueid);
752 ret = rte_eth_dev_start((uint8_t) portid);
754 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%u\n",
759 /* get link status */
760 rte_eth_link_get((uint8_t) portid, &link);
761 if (link.link_status) {
762 printf(" Link Up - speed %u Mbps - %s\n",
763 (unsigned) link.link_speed,
764 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
765 ("full-duplex") : ("half-duplex\n"));
767 printf(" Link Down\n");
770 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
772 lsi_ports_eth_addr[portid].addr_bytes[0],
773 lsi_ports_eth_addr[portid].addr_bytes[1],
774 lsi_ports_eth_addr[portid].addr_bytes[2],
775 lsi_ports_eth_addr[portid].addr_bytes[3],
776 lsi_ports_eth_addr[portid].addr_bytes[4],
777 lsi_ports_eth_addr[portid].addr_bytes[5]);
779 /* initialize port stats */
780 memset(&port_statistics, 0, sizeof(port_statistics));
783 /* launch per-lcore init on every lcore */
784 rte_eal_mp_remote_launch(lsi_launch_one_lcore, NULL, CALL_MASTER);
785 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
786 if (rte_eal_wait_lcore(lcore_id) < 0)