1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
11 #include <sys/queue.h>
16 #include <rte_common.h>
17 #include <rte_byteorder.h>
19 #include <rte_memory.h>
20 #include <rte_memcpy.h>
22 #include <rte_launch.h>
23 #include <rte_atomic.h>
24 #include <rte_cycles.h>
25 #include <rte_prefetch.h>
26 #include <rte_lcore.h>
27 #include <rte_per_lcore.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_interrupts.h>
30 #include <rte_random.h>
31 #include <rte_debug.h>
32 #include <rte_ether.h>
33 #include <rte_ethdev.h>
34 #include <rte_mempool.h>
36 #include <rte_malloc.h>
37 #include <rte_fbk_hash.h>
40 #define RTE_LOGTYPE_IPv4_MULTICAST RTE_LOGTYPE_USER1
44 #define MCAST_CLONE_PORTS 2
45 #define MCAST_CLONE_SEGS 2
47 #define PKT_MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
48 #define NB_PKT_MBUF 8192
50 #define HDR_MBUF_DATA_SIZE (2 * RTE_PKTMBUF_HEADROOM)
51 #define NB_HDR_MBUF (NB_PKT_MBUF * MAX_PORTS)
53 #define NB_CLONE_MBUF (NB_PKT_MBUF * MCAST_CLONE_PORTS * MCAST_CLONE_SEGS * 2)
55 /* allow max jumbo frame 9.5 KB */
56 #define JUMBO_FRAME_MAX_SIZE 0x2600
58 #define MAX_PKT_BURST 32
59 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
61 /* Configure how many packets ahead to prefetch, when reading packets */
62 #define PREFETCH_OFFSET 3
65 * Construct Ethernet multicast address from IPv4 multicast address.
66 * Citing RFC 1112, section 6.4:
67 * "An IP host group address is mapped to an Ethernet multicast address
68 * by placing the low-order 23-bits of the IP address into the low-order
69 * 23 bits of the Ethernet multicast address 01-00-5E-00-00-00 (hex)."
71 #define ETHER_ADDR_FOR_IPV4_MCAST(x) \
72 (rte_cpu_to_be_64(0x01005e000000ULL | ((x) & 0x7fffff)) >> 16)
75 * Configurable number of RX/TX ring descriptors
77 #define RTE_TEST_RX_DESC_DEFAULT 128
78 #define RTE_TEST_TX_DESC_DEFAULT 512
79 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
80 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
82 /* ethernet addresses of ports */
83 static struct ether_addr ports_eth_addr[MAX_PORTS];
85 /* mask of enabled ports */
86 static uint32_t enabled_port_mask = 0;
88 static uint16_t nb_ports;
90 static int rx_queue_per_lcore = 1;
94 struct rte_mbuf *m_table[MAX_PKT_BURST];
97 #define MAX_RX_QUEUE_PER_LCORE 16
98 #define MAX_TX_QUEUE_PER_PORT 16
99 struct lcore_queue_conf {
102 uint8_t rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
103 uint16_t tx_queue_id[MAX_PORTS];
104 struct mbuf_table tx_mbufs[MAX_PORTS];
105 } __rte_cache_aligned;
106 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
108 static struct rte_eth_conf port_conf = {
110 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
112 .header_split = 0, /**< Header Split disabled */
113 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
114 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
115 .jumbo_frame = 1, /**< Jumbo Frame Support enabled */
116 .hw_strip_crc = 1, /**< CRC stripped by hardware */
119 .mq_mode = ETH_MQ_TX_NONE,
123 static struct rte_mempool *packet_pool, *header_pool, *clone_pool;
127 static struct rte_fbk_hash_params mcast_hash_params = {
128 .name = "MCAST_HASH",
130 .entries_per_bucket = 4,
136 struct rte_fbk_hash_table *mcast_hash = NULL;
138 struct mcast_group_params {
143 static struct mcast_group_params mcast_group_table[] = {
144 {IPv4(224,0,0,101), 0x1},
145 {IPv4(224,0,0,102), 0x2},
146 {IPv4(224,0,0,103), 0x3},
147 {IPv4(224,0,0,104), 0x4},
148 {IPv4(224,0,0,105), 0x5},
149 {IPv4(224,0,0,106), 0x6},
150 {IPv4(224,0,0,107), 0x7},
151 {IPv4(224,0,0,108), 0x8},
152 {IPv4(224,0,0,109), 0x9},
153 {IPv4(224,0,0,110), 0xA},
154 {IPv4(224,0,0,111), 0xB},
155 {IPv4(224,0,0,112), 0xC},
156 {IPv4(224,0,0,113), 0xD},
157 {IPv4(224,0,0,114), 0xE},
158 {IPv4(224,0,0,115), 0xF},
161 #define N_MCAST_GROUPS \
162 (sizeof (mcast_group_table) / sizeof (mcast_group_table[0]))
165 /* Send burst of packets on an output interface */
167 send_burst(struct lcore_queue_conf *qconf, uint16_t port)
169 struct rte_mbuf **m_table;
173 queueid = qconf->tx_queue_id[port];
174 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
175 n = qconf->tx_mbufs[port].len;
177 ret = rte_eth_tx_burst(port, queueid, m_table, n);
178 while (unlikely (ret < n)) {
179 rte_pktmbuf_free(m_table[ret]);
183 qconf->tx_mbufs[port].len = 0;
186 /* Get number of bits set. */
187 static inline uint32_t
192 for (n = 0; v != 0; v &= v - 1, n++)
199 * Create the output multicast packet based on the given input packet.
200 * There are two approaches for creating outgoing packet, though both
201 * are based on data zero-copy idea, they differ in few details:
202 * First one creates a clone of the input packet, e.g - walk though all
203 * segments of the input packet, and for each of them create a new packet
204 * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone()
205 * for more details). Then new mbuf is allocated for the packet header
206 * and is prepended to the 'clone' mbuf.
207 * Second approach doesn't make a clone, it just increment refcnt for all
208 * input packet segments. Then it allocates new mbuf for the packet header
209 * and prepends it to the input packet.
210 * Basically first approach reuses only input packet's data, but creates
211 * it's own copy of packet's metadata. Second approach reuses both input's
212 * packet data and metadata.
213 * The advantage of first approach - is that each outgoing packet has it's
214 * own copy of metadata, so we can safely modify data pointer of the
215 * input packet. That allows us to skip creation if the output packet for
216 * the last destination port, but instead modify input packet's header inplace,
217 * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times.
218 * The advantage of second approach - less work for each outgoing packet,
219 * e.g: we skip "clone" operation completely. Though it comes with a price -
220 * input packet's metadata has to be intact. So for N destination ports we
221 * need to invoke mcast_out_pkt N times.
222 * So for small number of outgoing ports (and segments in the input packet)
223 * first approach will be faster.
224 * As number of outgoing ports (and/or input segments) will grow,
225 * second way will become more preferable.
230 * Control which of the two approaches described above should be used:
231 * - 0 - use second approach:
232 * Don't "clone" input packet.
233 * Prepend new header directly to the input packet
234 * - 1 - use first approach:
235 * Make a "clone" of input packet first.
236 * Prepend new header to the clone of the input packet
238 * - The pointer to the new outgoing packet.
239 * - NULL if operation failed.
241 static inline struct rte_mbuf *
242 mcast_out_pkt(struct rte_mbuf *pkt, int use_clone)
244 struct rte_mbuf *hdr;
246 /* Create new mbuf for the header. */
247 if (unlikely ((hdr = rte_pktmbuf_alloc(header_pool)) == NULL))
250 /* If requested, then make a new clone packet. */
251 if (use_clone != 0 &&
252 unlikely ((pkt = rte_pktmbuf_clone(pkt, clone_pool)) == NULL)) {
253 rte_pktmbuf_free(hdr);
257 /* prepend new header */
261 /* update header's fields */
262 hdr->pkt_len = (uint16_t)(hdr->data_len + pkt->pkt_len);
263 hdr->nb_segs = pkt->nb_segs + 1;
265 /* copy metadata from source packet*/
266 hdr->port = pkt->port;
267 hdr->vlan_tci = pkt->vlan_tci;
268 hdr->vlan_tci_outer = pkt->vlan_tci_outer;
269 hdr->tx_offload = pkt->tx_offload;
270 hdr->hash = pkt->hash;
272 hdr->ol_flags = pkt->ol_flags;
274 __rte_mbuf_sanity_check(hdr, 1);
279 * Write new Ethernet header to the outgoing packet,
280 * and put it into the outgoing queue for the given port.
283 mcast_send_pkt(struct rte_mbuf *pkt, struct ether_addr *dest_addr,
284 struct lcore_queue_conf *qconf, uint16_t port)
286 struct ether_hdr *ethdr;
289 /* Construct Ethernet header. */
290 ethdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr));
291 RTE_ASSERT(ethdr != NULL);
293 ether_addr_copy(dest_addr, ðdr->d_addr);
294 ether_addr_copy(&ports_eth_addr[port], ðdr->s_addr);
295 ethdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
297 /* Put new packet into the output queue */
298 len = qconf->tx_mbufs[port].len;
299 qconf->tx_mbufs[port].m_table[len] = pkt;
300 qconf->tx_mbufs[port].len = ++len;
302 /* Transmit packets */
303 if (unlikely(MAX_PKT_BURST == len))
304 send_burst(qconf, port);
307 /* Multicast forward of the input packet */
309 mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf)
312 struct ipv4_hdr *iphdr;
313 uint32_t dest_addr, port_mask, port_num, use_clone;
318 struct ether_addr as_addr;
321 /* Remove the Ethernet header from the input packet */
322 iphdr = (struct ipv4_hdr *)rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
323 RTE_ASSERT(iphdr != NULL);
325 dest_addr = rte_be_to_cpu_32(iphdr->dst_addr);
328 * Check that it is a valid multicast address and
329 * we have some active ports assigned to it.
331 if(!IS_IPV4_MCAST(dest_addr) ||
332 (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 ||
333 (port_mask = hash & enabled_port_mask) == 0) {
338 /* Calculate number of destination ports. */
339 port_num = bitcnt(port_mask);
341 /* Should we use rte_pktmbuf_clone() or not. */
342 use_clone = (port_num <= MCAST_CLONE_PORTS &&
343 m->nb_segs <= MCAST_CLONE_SEGS);
345 /* Mark all packet's segments as referenced port_num times */
347 rte_pktmbuf_refcnt_update(m, (uint16_t)port_num);
349 /* construct destination ethernet address */
350 dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr);
352 for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) {
354 /* Prepare output packet and send it out. */
355 if ((port_mask & 1) != 0) {
356 if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL))
357 mcast_send_pkt(mc, &dst_eth_addr.as_addr,
359 else if (use_clone == 0)
365 * If we making clone packets, then, for the last destination port,
366 * we can overwrite input packet's metadata.
369 mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port);
374 /* Send burst of outgoing packet, if timeout expires. */
376 send_timeout_burst(struct lcore_queue_conf *qconf)
380 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
382 cur_tsc = rte_rdtsc();
383 if (likely (cur_tsc < qconf->tx_tsc + drain_tsc))
386 for (portid = 0; portid < MAX_PORTS; portid++) {
387 if (qconf->tx_mbufs[portid].len != 0)
388 send_burst(qconf, portid);
390 qconf->tx_tsc = cur_tsc;
393 /* main processing loop */
395 main_loop(__rte_unused void *dummy)
397 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
401 struct lcore_queue_conf *qconf;
403 lcore_id = rte_lcore_id();
404 qconf = &lcore_queue_conf[lcore_id];
407 if (qconf->n_rx_queue == 0) {
408 RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n",
413 RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n",
416 for (i = 0; i < qconf->n_rx_queue; i++) {
418 portid = qconf->rx_queue_list[i];
419 RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n",
426 * Read packet from RX queues
428 for (i = 0; i < qconf->n_rx_queue; i++) {
430 portid = qconf->rx_queue_list[i];
431 nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
434 /* Prefetch first packets */
435 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
436 rte_prefetch0(rte_pktmbuf_mtod(
437 pkts_burst[j], void *));
440 /* Prefetch and forward already prefetched packets */
441 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
442 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
443 j + PREFETCH_OFFSET], void *));
444 mcast_forward(pkts_burst[j], qconf);
447 /* Forward remaining prefetched packets */
448 for (; j < nb_rx; j++) {
449 mcast_forward(pkts_burst[j], qconf);
453 /* Send out packets from TX queues */
454 send_timeout_burst(qconf);
460 print_usage(const char *prgname)
462 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
463 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
464 " -q NQ: number of queue (=ports) per lcore (default is 1)\n",
469 parse_portmask(const char *portmask)
474 /* parse hexadecimal string */
475 pm = strtoul(portmask, &end, 16);
476 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
483 parse_nqueue(const char *q_arg)
488 /* parse numerical string */
490 n = strtoul(q_arg, &end, 0);
491 if (errno != 0 || end == NULL || *end != '\0' ||
492 n == 0 || n >= MAX_RX_QUEUE_PER_LCORE)
498 /* Parse the argument given in the command line of the application */
500 parse_args(int argc, char **argv)
505 char *prgname = argv[0];
506 static struct option lgopts[] = {
512 while ((opt = getopt_long(argc, argvopt, "p:q:",
513 lgopts, &option_index)) != EOF) {
518 enabled_port_mask = parse_portmask(optarg);
519 if (enabled_port_mask == 0) {
520 printf("invalid portmask\n");
521 print_usage(prgname);
528 rx_queue_per_lcore = parse_nqueue(optarg);
529 if (rx_queue_per_lcore < 0) {
530 printf("invalid queue number\n");
531 print_usage(prgname);
537 print_usage(prgname);
543 argv[optind-1] = prgname;
546 optind = 1; /* reset getopt lib */
551 print_ethaddr(const char *name, struct ether_addr *eth_addr)
553 char buf[ETHER_ADDR_FMT_SIZE];
554 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
555 printf("%s%s", name, buf);
559 init_mcast_hash(void)
563 mcast_hash_params.socket_id = rte_socket_id();
564 mcast_hash = rte_fbk_hash_create(&mcast_hash_params);
565 if (mcast_hash == NULL){
569 for (i = 0; i < N_MCAST_GROUPS; i ++){
570 if (rte_fbk_hash_add_key(mcast_hash,
571 mcast_group_table[i].ip,
572 mcast_group_table[i].port_mask) < 0) {
580 /* Check the link status of all ports in up to 9s, and print them finally */
582 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
584 #define CHECK_INTERVAL 100 /* 100ms */
585 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
587 uint8_t 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)
603 "Port%d Link Up. Speed %u Mbps - %s\n",
604 portid, link.link_speed,
605 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
606 ("full-duplex") : ("half-duplex\n"));
608 printf("Port %d Link Down\n", portid);
611 /* clear all_ports_up flag if any link down */
612 if (link.link_status == ETH_LINK_DOWN) {
617 /* after finally printing all link status, get out */
621 if (all_ports_up == 0) {
624 rte_delay_ms(CHECK_INTERVAL);
627 /* set the print_flag if all ports up or timeout */
628 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
636 main(int argc, char **argv)
638 struct lcore_queue_conf *qconf;
639 struct rte_eth_dev_info dev_info;
640 struct rte_eth_txconf *txconf;
643 unsigned lcore_id = 0, rx_lcore_id = 0;
644 uint32_t n_tx_queue, nb_lcores;
648 ret = rte_eal_init(argc, argv);
650 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
654 /* parse application arguments (after the EAL ones) */
655 ret = parse_args(argc, argv);
657 rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n");
659 /* create the mbuf pools */
660 packet_pool = rte_pktmbuf_pool_create("packet_pool", NB_PKT_MBUF, 32,
661 0, PKT_MBUF_DATA_SIZE, rte_socket_id());
663 if (packet_pool == NULL)
664 rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n");
666 header_pool = rte_pktmbuf_pool_create("header_pool", NB_HDR_MBUF, 32,
667 0, HDR_MBUF_DATA_SIZE, rte_socket_id());
669 if (header_pool == NULL)
670 rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n");
672 clone_pool = rte_pktmbuf_pool_create("clone_pool", NB_CLONE_MBUF, 32,
673 0, 0, rte_socket_id());
675 if (clone_pool == NULL)
676 rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");
678 nb_ports = rte_eth_dev_count();
680 rte_exit(EXIT_FAILURE, "No physical ports!\n");
681 if (nb_ports > MAX_PORTS)
682 nb_ports = MAX_PORTS;
684 nb_lcores = rte_lcore_count();
686 /* initialize all ports */
687 for (portid = 0; portid < nb_ports; portid++) {
688 /* skip ports that are not enabled */
689 if ((enabled_port_mask & (1 << portid)) == 0) {
690 printf("Skipping disabled port %d\n", portid);
694 qconf = &lcore_queue_conf[rx_lcore_id];
696 /* limit the frame size to the maximum supported by NIC */
697 rte_eth_dev_info_get(portid, &dev_info);
698 port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
699 dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
701 /* get the lcore_id for this port */
702 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
703 qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
706 qconf = &lcore_queue_conf[rx_lcore_id];
708 if (rx_lcore_id >= RTE_MAX_LCORE)
709 rte_exit(EXIT_FAILURE, "Not enough cores\n");
711 qconf->rx_queue_list[qconf->n_rx_queue] = portid;
715 printf("Initializing port %d on lcore %u... ", portid,
719 n_tx_queue = nb_lcores;
720 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
721 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
722 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
725 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
728 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
731 rte_exit(EXIT_FAILURE,
732 "Cannot adjust number of descriptors: err=%d, port=%d\n",
735 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
736 print_ethaddr(" Address:", &ports_eth_addr[portid]);
739 /* init one RX queue */
741 printf("rxq=%hu ", queueid);
743 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
744 rte_eth_dev_socket_id(portid),
748 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%d\n",
751 /* init one TX queue per couple (lcore,port) */
754 RTE_LCORE_FOREACH(lcore_id) {
755 if (rte_lcore_is_enabled(lcore_id) == 0)
757 printf("txq=%u,%hu ", lcore_id, queueid);
760 txconf = &dev_info.default_txconf;
761 txconf->txq_flags = 0;
762 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
763 rte_lcore_to_socket_id(lcore_id), txconf);
765 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
766 "port=%d\n", ret, portid);
768 qconf = &lcore_queue_conf[lcore_id];
769 qconf->tx_queue_id[portid] = queueid;
774 ret = rte_eth_dev_start(portid);
776 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
782 check_all_ports_link_status(nb_ports, enabled_port_mask);
784 /* initialize the multicast hash */
785 int retval = init_mcast_hash();
787 rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n");
789 /* launch per-lcore init on every lcore */
790 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
791 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
792 if (rte_eal_wait_lcore(lcore_id) < 0)