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 1024
78 #define RTE_TEST_TX_DESC_DEFAULT 1024
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 rte_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 .offloads = DEV_RX_OFFLOAD_JUMBO_FRAME,
115 .mq_mode = ETH_MQ_TX_NONE,
116 .offloads = DEV_TX_OFFLOAD_MULTI_SEGS,
120 static struct rte_mempool *packet_pool, *header_pool, *clone_pool;
124 static struct rte_fbk_hash_params mcast_hash_params = {
125 .name = "MCAST_HASH",
127 .entries_per_bucket = 4,
133 struct rte_fbk_hash_table *mcast_hash = NULL;
135 struct mcast_group_params {
140 static struct mcast_group_params mcast_group_table[] = {
141 {RTE_IPV4(224,0,0,101), 0x1},
142 {RTE_IPV4(224,0,0,102), 0x2},
143 {RTE_IPV4(224,0,0,103), 0x3},
144 {RTE_IPV4(224,0,0,104), 0x4},
145 {RTE_IPV4(224,0,0,105), 0x5},
146 {RTE_IPV4(224,0,0,106), 0x6},
147 {RTE_IPV4(224,0,0,107), 0x7},
148 {RTE_IPV4(224,0,0,108), 0x8},
149 {RTE_IPV4(224,0,0,109), 0x9},
150 {RTE_IPV4(224,0,0,110), 0xA},
151 {RTE_IPV4(224,0,0,111), 0xB},
152 {RTE_IPV4(224,0,0,112), 0xC},
153 {RTE_IPV4(224,0,0,113), 0xD},
154 {RTE_IPV4(224,0,0,114), 0xE},
155 {RTE_IPV4(224,0,0,115), 0xF},
158 /* Send burst of packets on an output interface */
160 send_burst(struct lcore_queue_conf *qconf, uint16_t port)
162 struct rte_mbuf **m_table;
166 queueid = qconf->tx_queue_id[port];
167 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
168 n = qconf->tx_mbufs[port].len;
170 ret = rte_eth_tx_burst(port, queueid, m_table, n);
171 while (unlikely (ret < n)) {
172 rte_pktmbuf_free(m_table[ret]);
176 qconf->tx_mbufs[port].len = 0;
179 /* Get number of bits set. */
180 static inline uint32_t
185 for (n = 0; v != 0; v &= v - 1, n++)
192 * Create the output multicast packet based on the given input packet.
193 * There are two approaches for creating outgoing packet, though both
194 * are based on data zero-copy idea, they differ in few details:
195 * First one creates a clone of the input packet, e.g - walk though all
196 * segments of the input packet, and for each of them create a new packet
197 * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone()
198 * for more details). Then new mbuf is allocated for the packet header
199 * and is prepended to the 'clone' mbuf.
200 * Second approach doesn't make a clone, it just increment refcnt for all
201 * input packet segments. Then it allocates new mbuf for the packet header
202 * and prepends it to the input packet.
203 * Basically first approach reuses only input packet's data, but creates
204 * it's own copy of packet's metadata. Second approach reuses both input's
205 * packet data and metadata.
206 * The advantage of first approach - is that each outgoing packet has it's
207 * own copy of metadata, so we can safely modify data pointer of the
208 * input packet. That allows us to skip creation if the output packet for
209 * the last destination port, but instead modify input packet's header inplace,
210 * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times.
211 * The advantage of second approach - less work for each outgoing packet,
212 * e.g: we skip "clone" operation completely. Though it comes with a price -
213 * input packet's metadata has to be intact. So for N destination ports we
214 * need to invoke mcast_out_pkt N times.
215 * So for small number of outgoing ports (and segments in the input packet)
216 * first approach will be faster.
217 * As number of outgoing ports (and/or input segments) will grow,
218 * second way will become more preferable.
223 * Control which of the two approaches described above should be used:
224 * - 0 - use second approach:
225 * Don't "clone" input packet.
226 * Prepend new header directly to the input packet
227 * - 1 - use first approach:
228 * Make a "clone" of input packet first.
229 * Prepend new header to the clone of the input packet
231 * - The pointer to the new outgoing packet.
232 * - NULL if operation failed.
234 static inline struct rte_mbuf *
235 mcast_out_pkt(struct rte_mbuf *pkt, int use_clone)
237 struct rte_mbuf *hdr;
239 /* Create new mbuf for the header. */
240 if (unlikely ((hdr = rte_pktmbuf_alloc(header_pool)) == NULL))
243 /* If requested, then make a new clone packet. */
244 if (use_clone != 0 &&
245 unlikely ((pkt = rte_pktmbuf_clone(pkt, clone_pool)) == NULL)) {
246 rte_pktmbuf_free(hdr);
250 /* prepend new header */
253 /* update header's fields */
254 hdr->pkt_len = (uint16_t)(hdr->data_len + pkt->pkt_len);
255 hdr->nb_segs = pkt->nb_segs + 1;
257 __rte_mbuf_sanity_check(hdr, 1);
262 * Write new Ethernet header to the outgoing packet,
263 * and put it into the outgoing queue for the given port.
266 mcast_send_pkt(struct rte_mbuf *pkt, struct rte_ether_addr *dest_addr,
267 struct lcore_queue_conf *qconf, uint16_t port)
269 struct rte_ether_hdr *ethdr;
272 /* Construct Ethernet header. */
273 ethdr = (struct rte_ether_hdr *)
274 rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr));
275 RTE_ASSERT(ethdr != NULL);
277 rte_ether_addr_copy(dest_addr, ðdr->d_addr);
278 rte_ether_addr_copy(&ports_eth_addr[port], ðdr->s_addr);
279 ethdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
281 /* Put new packet into the output queue */
282 len = qconf->tx_mbufs[port].len;
283 qconf->tx_mbufs[port].m_table[len] = pkt;
284 qconf->tx_mbufs[port].len = ++len;
286 /* Transmit packets */
287 if (unlikely(MAX_PKT_BURST == len))
288 send_burst(qconf, port);
291 /* Multicast forward of the input packet */
293 mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf)
296 struct rte_ipv4_hdr *iphdr;
297 uint32_t dest_addr, port_mask, port_num, use_clone;
302 struct rte_ether_addr as_addr;
305 /* Remove the Ethernet header from the input packet */
306 iphdr = (struct rte_ipv4_hdr *)
307 rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
308 RTE_ASSERT(iphdr != NULL);
310 dest_addr = rte_be_to_cpu_32(iphdr->dst_addr);
313 * Check that it is a valid multicast address and
314 * we have some active ports assigned to it.
316 if (!RTE_IS_IPV4_MCAST(dest_addr) ||
317 (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 ||
318 (port_mask = hash & enabled_port_mask) == 0) {
323 /* Calculate number of destination ports. */
324 port_num = bitcnt(port_mask);
326 /* Should we use rte_pktmbuf_clone() or not. */
327 use_clone = (port_num <= MCAST_CLONE_PORTS &&
328 m->nb_segs <= MCAST_CLONE_SEGS);
330 /* Mark all packet's segments as referenced port_num times */
332 rte_pktmbuf_refcnt_update(m, (uint16_t)port_num);
334 /* construct destination ethernet address */
335 dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr);
337 for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) {
339 /* Prepare output packet and send it out. */
340 if ((port_mask & 1) != 0) {
341 if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL))
342 mcast_send_pkt(mc, &dst_eth_addr.as_addr,
344 else if (use_clone == 0)
350 * If we making clone packets, then, for the last destination port,
351 * we can overwrite input packet's metadata.
354 mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port);
359 /* Send burst of outgoing packet, if timeout expires. */
361 send_timeout_burst(struct lcore_queue_conf *qconf)
365 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
367 cur_tsc = rte_rdtsc();
368 if (likely (cur_tsc < qconf->tx_tsc + drain_tsc))
371 for (portid = 0; portid < MAX_PORTS; portid++) {
372 if (qconf->tx_mbufs[portid].len != 0)
373 send_burst(qconf, portid);
375 qconf->tx_tsc = cur_tsc;
378 /* main processing loop */
380 main_loop(__rte_unused void *dummy)
382 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
386 struct lcore_queue_conf *qconf;
388 lcore_id = rte_lcore_id();
389 qconf = &lcore_queue_conf[lcore_id];
392 if (qconf->n_rx_queue == 0) {
393 RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n",
398 RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n",
401 for (i = 0; i < qconf->n_rx_queue; i++) {
403 portid = qconf->rx_queue_list[i];
404 RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n",
411 * Read packet from RX queues
413 for (i = 0; i < qconf->n_rx_queue; i++) {
415 portid = qconf->rx_queue_list[i];
416 nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
419 /* Prefetch first packets */
420 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
421 rte_prefetch0(rte_pktmbuf_mtod(
422 pkts_burst[j], void *));
425 /* Prefetch and forward already prefetched packets */
426 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
427 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
428 j + PREFETCH_OFFSET], void *));
429 mcast_forward(pkts_burst[j], qconf);
432 /* Forward remaining prefetched packets */
433 for (; j < nb_rx; j++) {
434 mcast_forward(pkts_burst[j], qconf);
438 /* Send out packets from TX queues */
439 send_timeout_burst(qconf);
445 print_usage(const char *prgname)
447 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
448 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
449 " -q NQ: number of queue (=ports) per lcore (default is 1)\n",
454 parse_portmask(const char *portmask)
459 /* parse hexadecimal string */
460 pm = strtoul(portmask, &end, 16);
461 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
468 parse_nqueue(const char *q_arg)
473 /* parse numerical string */
475 n = strtoul(q_arg, &end, 0);
476 if (errno != 0 || end == NULL || *end != '\0' ||
477 n == 0 || n >= MAX_RX_QUEUE_PER_LCORE)
483 /* Parse the argument given in the command line of the application */
485 parse_args(int argc, char **argv)
490 char *prgname = argv[0];
491 static struct option lgopts[] = {
497 while ((opt = getopt_long(argc, argvopt, "p:q:",
498 lgopts, &option_index)) != EOF) {
503 enabled_port_mask = parse_portmask(optarg);
504 if (enabled_port_mask == 0) {
505 printf("invalid portmask\n");
506 print_usage(prgname);
513 rx_queue_per_lcore = parse_nqueue(optarg);
514 if (rx_queue_per_lcore < 0) {
515 printf("invalid queue number\n");
516 print_usage(prgname);
522 print_usage(prgname);
528 argv[optind-1] = prgname;
531 optind = 1; /* reset getopt lib */
536 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
538 char buf[RTE_ETHER_ADDR_FMT_SIZE];
539 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
540 printf("%s%s", name, buf);
544 init_mcast_hash(void)
548 mcast_hash_params.socket_id = rte_socket_id();
549 mcast_hash = rte_fbk_hash_create(&mcast_hash_params);
550 if (mcast_hash == NULL){
554 for (i = 0; i < RTE_DIM(mcast_group_table); i++) {
555 if (rte_fbk_hash_add_key(mcast_hash,
556 mcast_group_table[i].ip,
557 mcast_group_table[i].port_mask) < 0) {
565 /* Check the link status of all ports in up to 9s, and print them finally */
567 check_all_ports_link_status(uint32_t port_mask)
569 #define CHECK_INTERVAL 100 /* 100ms */
570 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
572 uint8_t count, all_ports_up, print_flag = 0;
573 struct rte_eth_link link;
575 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
577 printf("\nChecking link status");
579 for (count = 0; count <= MAX_CHECK_TIME; count++) {
581 RTE_ETH_FOREACH_DEV(portid) {
582 if ((port_mask & (1 << portid)) == 0)
584 memset(&link, 0, sizeof(link));
585 ret = rte_eth_link_get_nowait(portid, &link);
589 printf("Port %u link get failed: %s\n",
590 portid, rte_strerror(-ret));
593 /* print link status if flag set */
594 if (print_flag == 1) {
595 rte_eth_link_to_str(link_status_text,
596 sizeof(link_status_text),
598 printf("Port %d %s\n", portid,
602 /* clear all_ports_up flag if any link down */
603 if (link.link_status == ETH_LINK_DOWN) {
608 /* after finally printing all link status, get out */
612 if (all_ports_up == 0) {
615 rte_delay_ms(CHECK_INTERVAL);
618 /* set the print_flag if all ports up or timeout */
619 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
627 main(int argc, char **argv)
629 struct lcore_queue_conf *qconf;
630 struct rte_eth_dev_info dev_info;
631 struct rte_eth_txconf *txconf;
634 unsigned lcore_id = 0, rx_lcore_id = 0;
635 uint32_t n_tx_queue, nb_lcores;
639 ret = rte_eal_init(argc, argv);
641 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
645 /* parse application arguments (after the EAL ones) */
646 ret = parse_args(argc, argv);
648 rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n");
650 /* create the mbuf pools */
651 packet_pool = rte_pktmbuf_pool_create("packet_pool", NB_PKT_MBUF, 32,
652 0, PKT_MBUF_DATA_SIZE, rte_socket_id());
654 if (packet_pool == NULL)
655 rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n");
657 header_pool = rte_pktmbuf_pool_create("header_pool", NB_HDR_MBUF, 32,
658 0, HDR_MBUF_DATA_SIZE, rte_socket_id());
660 if (header_pool == NULL)
661 rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n");
663 clone_pool = rte_pktmbuf_pool_create("clone_pool", NB_CLONE_MBUF, 32,
664 0, 0, rte_socket_id());
666 if (clone_pool == NULL)
667 rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");
669 nb_ports = rte_eth_dev_count_avail();
671 rte_exit(EXIT_FAILURE, "No physical ports!\n");
672 if (nb_ports > MAX_PORTS)
673 nb_ports = MAX_PORTS;
675 nb_lcores = rte_lcore_count();
677 /* initialize all ports */
678 RTE_ETH_FOREACH_DEV(portid) {
679 struct rte_eth_rxconf rxq_conf;
680 struct rte_eth_conf local_port_conf = port_conf;
682 /* skip ports that are not enabled */
683 if ((enabled_port_mask & (1 << portid)) == 0) {
684 printf("Skipping disabled port %d\n", portid);
688 qconf = &lcore_queue_conf[rx_lcore_id];
690 /* limit the frame size to the maximum supported by NIC */
691 ret = rte_eth_dev_info_get(portid, &dev_info);
693 rte_exit(EXIT_FAILURE,
694 "Error during getting device (port %u) info: %s\n",
695 portid, strerror(-ret));
697 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
698 dev_info.max_rx_pktlen,
699 local_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;
723 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
726 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
729 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
732 rte_exit(EXIT_FAILURE,
733 "Cannot adjust number of descriptors: err=%d, port=%d\n",
736 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
738 rte_exit(EXIT_FAILURE,
739 "Cannot get MAC address: err=%d, port=%d\n",
742 print_ethaddr(" Address:", &ports_eth_addr[portid]);
745 /* init one RX queue */
747 printf("rxq=%hu ", queueid);
749 rxq_conf = dev_info.default_rxconf;
750 rxq_conf.offloads = local_port_conf.rxmode.offloads;
751 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
752 rte_eth_dev_socket_id(portid),
756 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%d\n",
759 /* init one TX queue per couple (lcore,port) */
762 RTE_LCORE_FOREACH(lcore_id) {
763 if (rte_lcore_is_enabled(lcore_id) == 0)
765 printf("txq=%u,%hu ", lcore_id, queueid);
768 txconf = &dev_info.default_txconf;
769 txconf->offloads = local_port_conf.txmode.offloads;
770 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
771 rte_lcore_to_socket_id(lcore_id), txconf);
773 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
774 "port=%d\n", ret, portid);
776 qconf = &lcore_queue_conf[lcore_id];
777 qconf->tx_queue_id[portid] = queueid;
780 ret = rte_eth_allmulticast_enable(portid);
782 rte_exit(EXIT_FAILURE,
783 "rte_eth_allmulticast_enable: err=%d, port=%d\n",
786 ret = rte_eth_dev_start(portid);
788 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
794 check_all_ports_link_status(enabled_port_mask);
796 /* initialize the multicast hash */
797 int retval = init_mcast_hash();
799 rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n");
801 /* launch per-lcore init on every lcore */
802 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
803 RTE_LCORE_FOREACH_WORKER(lcore_id) {
804 if (rte_eal_wait_lcore(lcore_id) < 0)
808 /* clean up the EAL */