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.
39 #include <sys/types.h>
41 #include <sys/queue.h>
46 #include <rte_common.h>
47 #include <rte_byteorder.h>
49 #include <rte_memory.h>
50 #include <rte_memcpy.h>
51 #include <rte_memzone.h>
52 #include <rte_tailq.h>
54 #include <rte_per_lcore.h>
55 #include <rte_launch.h>
56 #include <rte_atomic.h>
57 #include <rte_cycles.h>
58 #include <rte_prefetch.h>
59 #include <rte_lcore.h>
60 #include <rte_per_lcore.h>
61 #include <rte_branch_prediction.h>
62 #include <rte_interrupts.h>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
74 #include <rte_string_fns.h>
78 #define APP_LOOKUP_EXACT_MATCH 0
79 #define APP_LOOKUP_LPM 1
80 #define DO_RFC_1812_CHECKS
82 //#define APP_LOOKUP_METHOD APP_LOOKUP_EXACT_MATCH
83 #ifndef APP_LOOKUP_METHOD
84 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
87 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
89 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
92 #error "APP_LOOKUP_METHOD set to incorrect value"
96 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
97 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
98 #define IPv6_BYTES(addr) \
99 addr[0], addr[1], addr[2], addr[3], \
100 addr[4], addr[5], addr[6], addr[7], \
101 addr[8], addr[9], addr[10], addr[11],\
102 addr[12], addr[13],addr[14], addr[15]
106 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
110 #define MAX_JUMBO_PKT_LEN 9600
112 #define IPV6_ADDR_LEN 16
114 #define MEMPOOL_CACHE_SIZE 256
116 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
119 * This expression is used to calculate the number of mbufs needed depending on user input, taking
120 * into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
121 * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
124 #define NB_MBUF RTE_MAX ( \
125 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
126 nb_ports*nb_lcores*MAX_PKT_BURST + \
127 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
128 nb_lcores*MEMPOOL_CACHE_SIZE), \
132 * RX and TX Prefetch, Host, and Write-back threshold values should be
133 * carefully set for optimal performance. Consult the network
134 * controller's datasheet and supporting DPDK documentation for guidance
135 * on how these parameters should be set.
137 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
138 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
139 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
142 * These default values are optimized for use with the Intel(R) 82599 10 GbE
143 * Controller and the DPDK ixgbe PMD. Consider using other values for other
144 * network controllers and/or network drivers.
146 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
147 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
148 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
150 #define MAX_PKT_BURST 32
151 #define BURST_TX_DRAIN 200000ULL /* around 100us at 2 Ghz */
157 /* Configure how many packets ahead to prefetch, when reading packets */
158 #define PREFETCH_OFFSET 3
161 * Configurable number of RX/TX ring descriptors
163 #define RTE_TEST_RX_DESC_DEFAULT 128
164 #define RTE_TEST_TX_DESC_DEFAULT 512
165 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
166 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
168 /* ethernet addresses of ports */
169 static struct ether_addr ports_eth_addr[MAX_PORTS];
171 /* mask of enabled ports */
172 static uint32_t enabled_port_mask = 0;
173 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
174 static int numa_on = 1; /**< NUMA is enabled by default. */
178 struct rte_mbuf *m_table[MAX_PKT_BURST];
181 struct lcore_rx_queue {
184 } __rte_cache_aligned;
186 #define MAX_RX_QUEUE_PER_LCORE 16
187 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
188 #define MAX_RX_QUEUE_PER_PORT 128
190 #define MAX_LCORE_PARAMS 1024
191 struct lcore_params {
195 } __rte_cache_aligned;
197 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
198 static struct lcore_params lcore_params_array_default[] = {
210 static struct lcore_params * lcore_params = lcore_params_array_default;
211 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
212 sizeof(lcore_params_array_default[0]);
214 static struct rte_eth_conf port_conf = {
216 .max_rx_pkt_len = ETHER_MAX_LEN,
218 .header_split = 0, /**< Header Split disabled */
219 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
220 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
221 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
222 .hw_strip_crc = 0, /**< CRC stripped by hardware */
227 .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
231 .mq_mode = ETH_MQ_TX_NONE,
235 static const struct rte_eth_rxconf rx_conf = {
237 .pthresh = RX_PTHRESH,
238 .hthresh = RX_HTHRESH,
239 .wthresh = RX_WTHRESH,
241 .rx_free_thresh = 32,
244 static const struct rte_eth_txconf tx_conf = {
246 .pthresh = TX_PTHRESH,
247 .hthresh = TX_HTHRESH,
248 .wthresh = TX_WTHRESH,
250 .tx_free_thresh = 0, /* Use PMD default values */
251 .tx_rs_thresh = 0, /* Use PMD default values */
255 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
258 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
260 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
261 #include <rte_hash_crc.h>
262 #define DEFAULT_HASH_FUNC rte_hash_crc
264 #include <rte_jhash.h>
265 #define DEFAULT_HASH_FUNC rte_jhash
274 } __attribute__((__packed__));
277 uint8_t ip_dst[IPV6_ADDR_LEN];
278 uint8_t ip_src[IPV6_ADDR_LEN];
282 } __attribute__((__packed__));
284 struct ipv4_l3fwd_route {
285 struct ipv4_5tuple key;
289 struct ipv6_l3fwd_route {
290 struct ipv6_5tuple key;
294 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
295 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
296 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
297 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
298 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
301 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
304 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
305 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
306 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
307 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
313 typedef struct rte_hash lookup_struct_t;
314 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
315 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
317 #define L3FWD_HASH_ENTRIES 1024
319 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
320 .name = "ipv4_l3fwd_hash_0",
321 .entries = L3FWD_HASH_ENTRIES,
323 .key_len = sizeof(struct ipv4_5tuple),
324 .hash_func = DEFAULT_HASH_FUNC,
325 .hash_func_init_val = 0,
326 .socket_id = SOCKET0,
329 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
330 .name = "ipv6_l3fwd_hash_0",
331 .entries = L3FWD_HASH_ENTRIES,
333 .key_len = sizeof(struct ipv6_5tuple),
334 .hash_func = DEFAULT_HASH_FUNC,
335 .hash_func_init_val = 0,
336 .socket_id = SOCKET0,
339 #define IPV4_L3FWD_NUM_ROUTES \
340 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
342 #define IPV6_L3FWD_NUM_ROUTES \
343 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
345 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
346 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
349 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
350 struct ipv4_l3fwd_route {
356 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
357 {IPv4(1,1,1,0), 24, 0},
358 {IPv4(2,1,1,0), 24, 1},
359 {IPv4(3,1,1,0), 24, 2},
360 {IPv4(4,1,1,0), 24, 3},
361 {IPv4(5,1,1,0), 24, 4},
362 {IPv4(6,1,1,0), 24, 5},
363 {IPv4(7,1,1,0), 24, 6},
364 {IPv4(8,1,1,0), 24, 7},
367 #define IPV4_L3FWD_NUM_ROUTES \
368 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
370 #define IPV4_L3FWD_LPM_MAX_RULES 1024
372 typedef struct rte_lpm lookup_struct_t;
373 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
378 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
379 uint16_t tx_queue_id[MAX_PORTS];
380 struct mbuf_table tx_mbufs[MAX_PORTS];
381 lookup_struct_t * ipv4_lookup_struct;
382 lookup_struct_t * ipv6_lookup_struct;
383 } __rte_cache_aligned;
385 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
387 /* Send burst of packets on an output interface */
389 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
391 struct rte_mbuf **m_table;
395 queueid = qconf->tx_queue_id[port];
396 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
398 ret = rte_eth_tx_burst(port, queueid, m_table, n);
399 if (unlikely(ret < n)) {
401 rte_pktmbuf_free(m_table[ret]);
408 /* Enqueue a single packet, and send burst if queue is filled */
410 send_single_packet(struct rte_mbuf *m, uint8_t port)
414 struct lcore_conf *qconf;
416 lcore_id = rte_lcore_id();
418 qconf = &lcore_conf[lcore_id];
419 len = qconf->tx_mbufs[port].len;
420 qconf->tx_mbufs[port].m_table[len] = m;
423 /* enough pkts to be sent */
424 if (unlikely(len == MAX_PKT_BURST)) {
425 send_burst(qconf, MAX_PKT_BURST, port);
429 qconf->tx_mbufs[port].len = len;
433 #ifdef DO_RFC_1812_CHECKS
435 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
437 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
439 * 1. The packet length reported by the Link Layer must be large
440 * enough to hold the minimum length legal IP datagram (20 bytes).
442 if (link_len < sizeof(struct ipv4_hdr))
445 /* 2. The IP checksum must be correct. */
446 /* this is checked in H/W */
449 * 3. The IP version number must be 4. If the version number is not 4
450 * then the packet may be another version of IP, such as IPng or
453 if (((pkt->version_ihl) >> 4) != 4)
456 * 4. The IP header length field must be large enough to hold the
457 * minimum length legal IP datagram (20 bytes = 5 words).
459 if ((pkt->version_ihl & 0xf) < 5)
463 * 5. The IP total length field must be large enough to hold the IP
464 * datagram header, whose length is specified in the IP header length
467 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
474 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
476 print_ipv4_key(struct ipv4_5tuple key)
478 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
479 (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
482 print_ipv6_key(struct ipv6_5tuple key)
484 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
485 "port dst = %d, port src = %d, proto = %d\n",
486 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
487 key.port_dst, key.port_src, key.proto);
490 static inline uint8_t
491 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
493 struct ipv4_5tuple key;
498 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
499 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
500 key.proto = ipv4_hdr->next_proto_id;
502 switch (ipv4_hdr->next_proto_id) {
504 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
505 sizeof(struct ipv4_hdr));
506 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
507 key.port_src = rte_be_to_cpu_16(tcp->src_port);
511 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
512 sizeof(struct ipv4_hdr));
513 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
514 key.port_src = rte_be_to_cpu_16(udp->src_port);
523 /* Find destination port */
524 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
525 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
528 static inline uint8_t
529 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
531 struct ipv6_5tuple key;
536 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
537 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
539 key.proto = ipv6_hdr->proto;
541 switch (ipv6_hdr->proto) {
543 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
544 sizeof(struct ipv6_hdr));
545 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
546 key.port_src = rte_be_to_cpu_16(tcp->src_port);
550 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
551 sizeof(struct ipv6_hdr));
552 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
553 key.port_src = rte_be_to_cpu_16(udp->src_port);
562 /* Find destination port */
563 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
564 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
568 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
569 static inline uint8_t
570 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
574 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
575 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
581 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
583 struct ether_hdr *eth_hdr;
584 struct ipv4_hdr *ipv4_hdr;
588 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
590 if (m->ol_flags & PKT_RX_IPV4_HDR) {
591 /* Handle IPv4 headers.*/
592 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
593 sizeof(struct ether_hdr));
595 #ifdef DO_RFC_1812_CHECKS
596 /* Check to make sure the packet is valid (RFC1812) */
597 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
603 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, qconf->ipv4_lookup_struct);
604 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
607 /* 02:00:00:00:00:xx */
608 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
609 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
611 #ifdef DO_RFC_1812_CHECKS
612 /* Update time to live and header checksum */
613 --(ipv4_hdr->time_to_live);
614 ++(ipv4_hdr->hdr_checksum);
618 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
620 send_single_packet(m, dst_port);
623 /* Handle IPv6 headers.*/
624 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
625 struct ipv6_hdr *ipv6_hdr;
627 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
628 sizeof(struct ether_hdr));
630 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
632 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
635 /* 02:00:00:00:00:xx */
636 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
637 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
640 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
642 send_single_packet(m, dst_port);
644 /* We don't currently handle IPv6 packets in LPM mode. */
651 /* main processing loop */
652 static __attribute__((noreturn)) int
653 main_loop(__attribute__((unused)) void *dummy)
655 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
657 uint64_t prev_tsc = 0;
658 uint64_t diff_tsc, cur_tsc;
660 uint8_t portid, queueid;
661 struct lcore_conf *qconf;
663 lcore_id = rte_lcore_id();
664 qconf = &lcore_conf[lcore_id];
666 if (qconf->n_rx_queue == 0) {
667 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
671 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
673 for (i = 0; i < qconf->n_rx_queue; i++) {
675 portid = qconf->rx_queue_list[i].port_id;
676 queueid = qconf->rx_queue_list[i].queue_id;
677 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
683 cur_tsc = rte_rdtsc();
686 * TX burst queue drain
688 diff_tsc = cur_tsc - prev_tsc;
689 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
692 * This could be optimized (use queueid instead of
693 * portid), but it is not called so often
695 for (portid = 0; portid < MAX_PORTS; portid++) {
696 if (qconf->tx_mbufs[portid].len == 0)
698 send_burst(&lcore_conf[lcore_id],
699 qconf->tx_mbufs[portid].len,
701 qconf->tx_mbufs[portid].len = 0;
708 * Read packet from RX queues
710 for (i = 0; i < qconf->n_rx_queue; ++i) {
712 portid = qconf->rx_queue_list[i].port_id;
713 queueid = qconf->rx_queue_list[i].queue_id;
714 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
716 /* Prefetch first packets */
717 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
718 rte_prefetch0(rte_pktmbuf_mtod(
719 pkts_burst[j], void *));
722 /* Prefetch and forward already prefetched packets */
723 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
724 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
725 j + PREFETCH_OFFSET], void *));
726 l3fwd_simple_forward(pkts_burst[j], portid, qconf);
729 /* Forward remaining prefetched packets */
730 for (; j < nb_rx; j++) {
731 l3fwd_simple_forward(pkts_burst[j], portid, qconf);
738 check_lcore_params(void)
740 uint8_t queue, lcore;
744 for (i = 0; i < nb_lcore_params; ++i) {
745 queue = lcore_params[i].queue_id;
746 if (queue >= MAX_RX_QUEUE_PER_PORT) {
747 printf("invalid queue number: %hhu\n", queue);
750 lcore = lcore_params[i].lcore_id;
751 if (!rte_lcore_is_enabled(lcore)) {
752 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
755 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
757 printf("warning: lcore %hhu is on socket %d with numa off \n",
765 check_port_config(const unsigned nb_ports)
770 for (i = 0; i < nb_lcore_params; ++i) {
771 portid = lcore_params[i].port_id;
772 if ((enabled_port_mask & (1 << portid)) == 0) {
773 printf("port %u is not enabled in port mask\n", portid);
776 if (portid >= nb_ports) {
777 printf("port %u is not present on the board\n", portid);
785 get_port_n_rx_queues(const uint8_t port)
790 for (i = 0; i < nb_lcore_params; ++i) {
791 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
792 queue = lcore_params[i].queue_id;
794 return (uint8_t)(++queue);
798 init_lcore_rx_queues(void)
800 uint16_t i, nb_rx_queue;
803 for (i = 0; i < nb_lcore_params; ++i) {
804 lcore = lcore_params[i].lcore_id;
805 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
806 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
807 printf("error: too many queues (%u) for lcore: %u\n",
808 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
811 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
812 lcore_params[i].port_id;
813 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
814 lcore_params[i].queue_id;
815 lcore_conf[lcore].n_rx_queue++;
823 print_usage(const char *prgname)
825 printf ("%s [EAL options] -- -p PORTMASK -P"
826 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
827 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
828 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
829 " -P : enable promiscuous mode\n"
830 " --config (port,queue,lcore): rx queues configuration\n"
831 " --no-numa: optional, disable numa awareness\n"
832 " --enable-jumbo: enable jumbo frame"
833 " which max packet len is PKTLEN in decimal (64-9600)\n",
837 static int parse_max_pkt_len(const char *pktlen)
842 /* parse decimal string */
843 len = strtoul(pktlen, &end, 10);
844 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
854 parse_portmask(const char *portmask)
859 /* parse hexadecimal string */
860 pm = strtoul(portmask, &end, 16);
861 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
871 parse_config(const char *q_arg)
874 const char *p, *p0 = q_arg;
882 unsigned long int_fld[_NUM_FLD];
883 char *str_fld[_NUM_FLD];
889 while ((p = strchr(p0,'(')) != NULL) {
891 if((p0 = strchr(p,')')) == NULL)
895 if(size >= sizeof(s))
898 rte_snprintf(s, sizeof(s), "%.*s", size, p);
899 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
901 for (i = 0; i < _NUM_FLD; i++){
903 int_fld[i] = strtoul(str_fld[i], &end, 0);
904 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
907 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
908 printf("exceeded max number of lcore params: %hu\n",
912 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
913 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
914 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
917 lcore_params = lcore_params_array;
921 /* Parse the argument given in the command line of the application */
923 parse_args(int argc, char **argv)
928 char *prgname = argv[0];
929 static struct option lgopts[] = {
931 {"no-numa", 0, 0, 0},
932 {"enable-jumbo", 0, 0, 0},
938 while ((opt = getopt_long(argc, argvopt, "p:P",
939 lgopts, &option_index)) != EOF) {
944 enabled_port_mask = parse_portmask(optarg);
945 if (enabled_port_mask == 0) {
946 printf("invalid portmask\n");
947 print_usage(prgname);
952 printf("Promiscuous mode selected\n");
958 if (!strncmp(lgopts[option_index].name, "config", 6)) {
959 ret = parse_config(optarg);
961 printf("invalid config\n");
962 print_usage(prgname);
967 if (!strncmp(lgopts[option_index].name, "no-numa", 7)) {
968 printf("numa is disabled \n");
972 if (!strncmp(lgopts[option_index].name, "enable-jumbo", 12)) {
973 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
975 printf("jumbo frame is enabled \n");
976 port_conf.rxmode.jumbo_frame = 1;
978 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
979 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
980 ret = parse_max_pkt_len(optarg);
981 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
982 printf("invalid packet length\n");
983 print_usage(prgname);
986 port_conf.rxmode.max_rx_pkt_len = ret;
988 printf("set jumbo frame max packet length to %u\n",
989 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
995 print_usage(prgname);
1001 argv[optind-1] = prgname;
1004 optind = 0; /* reset getopt lib */
1009 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1011 printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1012 eth_addr->addr_bytes[0],
1013 eth_addr->addr_bytes[1],
1014 eth_addr->addr_bytes[2],
1015 eth_addr->addr_bytes[3],
1016 eth_addr->addr_bytes[4],
1017 eth_addr->addr_bytes[5]);
1020 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1022 setup_hash(int socketid)
1028 /* create ipv4 hash */
1029 rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1030 ipv4_l3fwd_hash_params.name = s;
1031 ipv4_l3fwd_hash_params.socket_id = socketid;
1032 ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1033 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1034 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1035 "socket %d\n", socketid);
1037 /* create ipv6 hash */
1038 rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1039 ipv6_l3fwd_hash_params.name = s;
1040 ipv6_l3fwd_hash_params.socket_id = socketid;
1041 ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1042 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1043 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1044 "socket %d\n", socketid);
1047 /* populate the ipv4 hash */
1048 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1049 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1050 (void *) &ipv4_l3fwd_route_array[i].key);
1052 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1053 "l3fwd hash on socket %d\n", i, socketid);
1055 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1056 printf("Hash: Adding key\n");
1057 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1060 /* populate the ipv6 hash */
1061 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1062 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1063 (void *) &ipv6_l3fwd_route_array[i].key);
1065 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1066 "l3fwd hash on socket %d\n", i, socketid);
1068 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1069 printf("Hash: Adding key\n");
1070 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1075 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1077 setup_lpm(int socketid)
1083 /* create the LPM table */
1084 rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1085 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1086 IPV4_L3FWD_LPM_MAX_RULES, 0);
1087 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1088 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1089 " on socket %d\n", socketid);
1091 /* populate the LPM table */
1092 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1093 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1094 ipv4_l3fwd_route_array[i].ip,
1095 ipv4_l3fwd_route_array[i].depth,
1096 ipv4_l3fwd_route_array[i].if_out);
1099 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1100 "l3fwd LPM table on socket %d\n",
1104 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1105 (unsigned)ipv4_l3fwd_route_array[i].ip,
1106 ipv4_l3fwd_route_array[i].depth,
1107 ipv4_l3fwd_route_array[i].if_out);
1113 init_mem(unsigned nb_mbuf)
1115 struct lcore_conf *qconf;
1120 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1121 if (rte_lcore_is_enabled(lcore_id) == 0)
1125 socketid = rte_lcore_to_socket_id(lcore_id);
1129 if (socketid >= NB_SOCKETS) {
1130 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
1131 socketid, lcore_id, NB_SOCKETS);
1133 if (pktmbuf_pool[socketid] == NULL) {
1134 rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1135 pktmbuf_pool[socketid] =
1136 rte_mempool_create(s, nb_mbuf, MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1137 sizeof(struct rte_pktmbuf_pool_private),
1138 rte_pktmbuf_pool_init, NULL,
1139 rte_pktmbuf_init, NULL,
1141 if (pktmbuf_pool[socketid] == NULL)
1142 rte_exit(EXIT_FAILURE,
1143 "Cannot init mbuf pool on socket %d\n", socketid);
1145 printf("Allocated mbuf pool on socket %d\n", socketid);
1147 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1148 setup_lpm(socketid);
1150 setup_hash(socketid);
1153 qconf = &lcore_conf[lcore_id];
1154 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1155 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1156 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1162 /* Check the link status of all ports in up to 9s, and print them finally */
1164 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1166 #define CHECK_INTERVAL 100 /* 100ms */
1167 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1168 uint8_t portid, count, all_ports_up, print_flag = 0;
1169 struct rte_eth_link link;
1171 printf("\nChecking link status");
1173 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1175 for (portid = 0; portid < port_num; portid++) {
1176 if ((port_mask & (1 << portid)) == 0)
1178 memset(&link, 0, sizeof(link));
1179 rte_eth_link_get_nowait(portid, &link);
1180 /* print link status if flag set */
1181 if (print_flag == 1) {
1182 if (link.link_status)
1183 printf("Port %d Link Up - speed %u "
1184 "Mbps - %s\n", (uint8_t)portid,
1185 (unsigned)link.link_speed,
1186 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1187 ("full-duplex") : ("half-duplex\n"));
1189 printf("Port %d Link Down\n",
1193 /* clear all_ports_up flag if any link down */
1194 if (link.link_status == 0) {
1199 /* after finally printing all link status, get out */
1200 if (print_flag == 1)
1203 if (all_ports_up == 0) {
1206 rte_delay_ms(CHECK_INTERVAL);
1209 /* set the print_flag if all ports up or timeout */
1210 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1218 MAIN(int argc, char **argv)
1220 struct lcore_conf *qconf;
1225 uint32_t n_tx_queue, nb_lcores;
1226 uint8_t portid, nb_rx_queue, queue, socketid;
1229 ret = rte_eal_init(argc, argv);
1231 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1235 /* parse application arguments (after the EAL ones) */
1236 ret = parse_args(argc, argv);
1238 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1240 if (check_lcore_params() < 0)
1241 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1243 ret = init_lcore_rx_queues();
1245 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1248 /* init driver(s) */
1249 if (rte_pmd_init_all() < 0)
1250 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1252 if (rte_eal_pci_probe() < 0)
1253 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1255 nb_ports = rte_eth_dev_count();
1256 if (nb_ports > MAX_PORTS)
1257 nb_ports = MAX_PORTS;
1259 if (check_port_config(nb_ports) < 0)
1260 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1262 nb_lcores = rte_lcore_count();
1264 /* initialize all ports */
1265 for (portid = 0; portid < nb_ports; portid++) {
1266 /* skip ports that are not enabled */
1267 if ((enabled_port_mask & (1 << portid)) == 0) {
1268 printf("\nSkipping disabled port %d\n", portid);
1273 printf("Initializing port %d ... ", portid );
1276 nb_rx_queue = get_port_n_rx_queues(portid);
1277 n_tx_queue = nb_lcores;
1278 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1279 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1280 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1281 nb_rx_queue, (unsigned)n_tx_queue );
1282 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1283 (uint16_t)n_tx_queue, &port_conf);
1285 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1288 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1289 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1293 ret = init_mem(NB_MBUF);
1295 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1297 /* init one TX queue per couple (lcore,port) */
1299 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1300 if (rte_lcore_is_enabled(lcore_id) == 0)
1304 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1308 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1310 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1311 socketid, &tx_conf);
1313 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1314 "port=%d\n", ret, portid);
1316 qconf = &lcore_conf[lcore_id];
1317 qconf->tx_queue_id[portid] = queueid;
1323 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1324 if (rte_lcore_is_enabled(lcore_id) == 0)
1326 qconf = &lcore_conf[lcore_id];
1327 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1329 /* init RX queues */
1330 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1331 portid = qconf->rx_queue_list[queue].port_id;
1332 queueid = qconf->rx_queue_list[queue].queue_id;
1335 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1339 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1342 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1343 socketid, &rx_conf, pktmbuf_pool[socketid]);
1345 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1346 "port=%d\n", ret, portid);
1353 for (portid = 0; portid < nb_ports; portid++) {
1354 if ((enabled_port_mask & (1 << portid)) == 0) {
1358 ret = rte_eth_dev_start(portid);
1360 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1364 * If enabled, put device in promiscuous mode.
1365 * This allows IO forwarding mode to forward packets
1366 * to itself through 2 cross-connected ports of the
1370 rte_eth_promiscuous_enable(portid);
1373 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1375 /* launch per-lcore init on every lcore */
1376 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1377 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1378 if (rte_eal_wait_lcore(lcore_id) < 0)