1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/types.h>
13 #include <sys/queue.h>
18 #include <rte_common.h>
20 #include <rte_byteorder.h>
22 #include <rte_memory.h>
23 #include <rte_memcpy.h>
25 #include <rte_launch.h>
26 #include <rte_atomic.h>
27 #include <rte_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
43 #include <rte_string_fns.h>
44 #include <rte_pause.h>
46 #include <cmdline_parse.h>
47 #include <cmdline_parse_etheraddr.h>
49 #include <lthread_api.h>
51 #define APP_LOOKUP_EXACT_MATCH 0
52 #define APP_LOOKUP_LPM 1
53 #define DO_RFC_1812_CHECKS
55 /* Enable cpu-load stats 0-off, 1-on */
56 #define APP_CPU_LOAD 1
58 #ifndef APP_LOOKUP_METHOD
59 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
62 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
63 #define sched_getcpu() rte_lcore_id()
67 check_ptype(int portid)
70 int ipv4 = 0, ipv6 = 0;
72 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
79 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
81 for (i = 0; i < ret; ++i) {
82 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
84 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
95 parse_ptype(struct rte_mbuf *m)
97 struct ether_hdr *eth_hdr;
98 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
101 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
102 ether_type = eth_hdr->ether_type;
103 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
104 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
105 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
106 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
108 m->packet_type = packet_type;
112 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
113 struct rte_mbuf *pkts[], uint16_t nb_pkts,
114 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
118 for (i = 0; i < nb_pkts; i++)
119 parse_ptype(pkts[i]);
125 * When set to zero, simple forwaring path is eanbled.
126 * When set to one, optimized forwarding path is enabled.
127 * Note that LPM optimisation path uses SSE4.1 instructions.
129 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1
131 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
132 #include <rte_hash.h>
133 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
135 #include <rte_lpm6.h>
137 #error "APP_LOOKUP_METHOD set to incorrect value"
140 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
142 #define MAX_JUMBO_PKT_LEN 9600
144 #define IPV6_ADDR_LEN 16
146 #define MEMPOOL_CACHE_SIZE 256
149 * This expression is used to calculate the number of mbufs needed depending on
150 * user input, taking into account memory for rx and tx hardware rings, cache
151 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
152 * NB_MBUF never goes below a minimum value of 8192
155 #define NB_MBUF RTE_MAX(\
156 (nb_ports*nb_rx_queue*nb_rxd + \
157 nb_ports*nb_lcores*MAX_PKT_BURST + \
158 nb_ports*n_tx_queue*nb_txd + \
159 nb_lcores*MEMPOOL_CACHE_SIZE), \
162 #define MAX_PKT_BURST 32
163 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
166 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
168 #define MAX_TX_BURST (MAX_PKT_BURST / 2)
169 #define BURST_SIZE MAX_TX_BURST
173 /* Configure how many packets ahead to prefetch, when reading packets */
174 #define PREFETCH_OFFSET 3
176 /* Used to mark destination port as 'invalid'. */
177 #define BAD_PORT ((uint16_t)-1)
182 * Configurable number of RX/TX ring descriptors
184 #define RTE_TEST_RX_DESC_DEFAULT 1024
185 #define RTE_TEST_TX_DESC_DEFAULT 1024
186 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
187 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
189 /* ethernet addresses of ports */
190 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
191 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
193 static xmm_t val_eth[RTE_MAX_ETHPORTS];
195 /* replace first 12B of the ethernet header. */
196 #define MASK_ETH 0x3f
198 /* mask of enabled ports */
199 static uint32_t enabled_port_mask;
200 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
201 static int numa_on = 1; /**< NUMA is enabled by default. */
202 static int parse_ptype_on;
204 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
205 static int ipv6; /**< ipv6 is false by default. */
208 #if (APP_CPU_LOAD == 1)
210 #define MAX_CPU RTE_MAX_LCORE
211 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000) /**< Timeout for collecting 5s */
213 #define CPU_PROCESS 0
215 #define MAX_CPU_COUNTER 2
220 uint64_t hits[MAX_CPU_COUNTER][MAX_CPU];
221 } __rte_cache_aligned;
223 static struct cpu_load cpu_load;
224 static int cpu_load_lcore_id = -1;
226 #define SET_CPU_BUSY(thread, counter) \
227 thread->conf.busy[counter] = 1
229 #define SET_CPU_IDLE(thread, counter) \
230 thread->conf.busy[counter] = 0
232 #define IS_CPU_BUSY(thread, counter) \
233 (thread->conf.busy[counter] > 0)
237 #define SET_CPU_BUSY(thread, counter)
238 #define SET_CPU_IDLE(thread, counter)
239 #define IS_CPU_BUSY(thread, counter) 0
245 struct rte_mbuf *m_table[MAX_PKT_BURST];
248 struct lcore_rx_queue {
251 } __rte_cache_aligned;
253 #define MAX_RX_QUEUE_PER_LCORE 16
254 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
255 #define MAX_RX_QUEUE_PER_PORT 128
257 #define MAX_LCORE_PARAMS 1024
258 struct rx_thread_params {
263 } __rte_cache_aligned;
265 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
266 static struct rx_thread_params rx_thread_params_array_default[] = {
278 static struct rx_thread_params *rx_thread_params =
279 rx_thread_params_array_default;
280 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
282 struct tx_thread_params {
285 } __rte_cache_aligned;
287 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
288 static struct tx_thread_params tx_thread_params_array_default[] = {
300 static struct tx_thread_params *tx_thread_params =
301 tx_thread_params_array_default;
302 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
304 static struct rte_eth_conf port_conf = {
306 .mq_mode = ETH_MQ_RX_RSS,
307 .max_rx_pkt_len = ETHER_MAX_LEN,
309 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
310 DEV_RX_OFFLOAD_CRC_STRIP),
315 .rss_hf = ETH_RSS_TCP,
319 .mq_mode = ETH_MQ_TX_NONE,
323 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
325 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
327 #include <rte_hash_crc.h>
328 #define DEFAULT_HASH_FUNC rte_hash_crc
336 } __attribute__((__packed__));
338 union ipv4_5tuple_host {
351 #define XMM_NUM_IN_IPV6_5TUPLE 3
354 uint8_t ip_dst[IPV6_ADDR_LEN];
355 uint8_t ip_src[IPV6_ADDR_LEN];
359 } __attribute__((__packed__));
361 union ipv6_5tuple_host {
366 uint8_t ip_src[IPV6_ADDR_LEN];
367 uint8_t ip_dst[IPV6_ADDR_LEN];
372 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
375 struct ipv4_l3fwd_route {
376 struct ipv4_5tuple key;
380 struct ipv6_l3fwd_route {
381 struct ipv6_5tuple key;
385 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
386 {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1), 101, 11, IPPROTO_TCP}, 0},
387 {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1), 102, 12, IPPROTO_TCP}, 1},
388 {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1), 101, 11, IPPROTO_TCP}, 2},
389 {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1), 102, 12, IPPROTO_TCP}, 3},
392 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
394 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
395 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
397 101, 11, IPPROTO_TCP}, 0},
400 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
401 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
403 102, 12, IPPROTO_TCP}, 1},
406 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
407 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
409 101, 11, IPPROTO_TCP}, 2},
412 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
413 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
415 102, 12, IPPROTO_TCP}, 3},
418 typedef struct rte_hash lookup_struct_t;
419 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
420 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
422 #ifdef RTE_ARCH_X86_64
423 /* default to 4 million hash entries (approx) */
424 #define L3FWD_HASH_ENTRIES (1024*1024*4)
426 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
427 #define L3FWD_HASH_ENTRIES (1024*1024*1)
429 #define HASH_ENTRY_NUMBER_DEFAULT 4
431 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
433 static inline uint32_t
434 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
437 const union ipv4_5tuple_host *k;
443 p = (const uint32_t *)&k->port_src;
445 init_val = rte_hash_crc_4byte(t, init_val);
446 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
447 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
448 init_val = rte_hash_crc_4byte(*p, init_val);
452 static inline uint32_t
453 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
456 const union ipv6_5tuple_host *k;
459 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
460 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
464 p = (const uint32_t *)&k->port_src;
466 ip_src0 = (const uint32_t *) k->ip_src;
467 ip_src1 = (const uint32_t *)(k->ip_src + 4);
468 ip_src2 = (const uint32_t *)(k->ip_src + 8);
469 ip_src3 = (const uint32_t *)(k->ip_src + 12);
470 ip_dst0 = (const uint32_t *) k->ip_dst;
471 ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
472 ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
473 ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
474 init_val = rte_hash_crc_4byte(t, init_val);
475 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
476 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
477 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
478 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
479 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
480 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
481 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
482 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
483 init_val = rte_hash_crc_4byte(*p, init_val);
487 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
488 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
490 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
491 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
495 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
496 struct ipv4_l3fwd_route {
502 struct ipv6_l3fwd_route {
508 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
509 {IPv4(1, 1, 1, 0), 24, 0},
510 {IPv4(2, 1, 1, 0), 24, 1},
511 {IPv4(3, 1, 1, 0), 24, 2},
512 {IPv4(4, 1, 1, 0), 24, 3},
513 {IPv4(5, 1, 1, 0), 24, 4},
514 {IPv4(6, 1, 1, 0), 24, 5},
515 {IPv4(7, 1, 1, 0), 24, 6},
516 {IPv4(8, 1, 1, 0), 24, 7},
519 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
520 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
521 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
522 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
523 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
524 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
525 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
526 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
527 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
530 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
531 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
533 #define IPV4_L3FWD_LPM_MAX_RULES 1024
534 #define IPV6_L3FWD_LPM_MAX_RULES 1024
535 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
537 typedef struct rte_lpm lookup_struct_t;
538 typedef struct rte_lpm6 lookup6_struct_t;
539 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
540 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
544 lookup_struct_t *ipv4_lookup_struct;
545 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
546 lookup6_struct_t *ipv6_lookup_struct;
548 lookup_struct_t *ipv6_lookup_struct;
551 } __rte_cache_aligned;
553 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
554 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
556 #define MAX_RX_QUEUE_PER_THREAD 16
557 #define MAX_TX_PORT_PER_THREAD RTE_MAX_ETHPORTS
558 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
559 #define MAX_RX_QUEUE_PER_PORT 128
561 #define MAX_RX_THREAD 1024
562 #define MAX_TX_THREAD 1024
563 #define MAX_THREAD (MAX_RX_THREAD + MAX_TX_THREAD)
566 * Producers and consumers threads configuration
568 static int lthreads_on = 1; /**< Use lthreads for processing*/
570 rte_atomic16_t rx_counter; /**< Number of spawned rx threads */
571 rte_atomic16_t tx_counter; /**< Number of spawned tx threads */
574 uint16_t lcore_id; /**< Initial lcore for rx thread */
575 uint16_t cpu_id; /**< Cpu id for cpu load stats counter */
576 uint16_t thread_id; /**< Thread ID */
578 #if (APP_CPU_LOAD > 0)
579 int busy[MAX_CPU_COUNTER];
583 struct thread_rx_conf {
584 struct thread_conf conf;
587 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
589 uint16_t n_ring; /**< Number of output rings */
590 struct rte_ring *ring[RTE_MAX_LCORE];
591 struct lthread_cond *ready[RTE_MAX_LCORE];
593 #if (APP_CPU_LOAD > 0)
594 int busy[MAX_CPU_COUNTER];
596 } __rte_cache_aligned;
598 uint16_t n_rx_thread;
599 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
601 struct thread_tx_conf {
602 struct thread_conf conf;
604 uint16_t tx_queue_id[RTE_MAX_LCORE];
605 struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
607 struct rte_ring *ring;
608 struct lthread_cond **ready;
610 } __rte_cache_aligned;
612 uint16_t n_tx_thread;
613 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
615 /* Send burst of packets on an output interface */
617 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
619 struct rte_mbuf **m_table;
623 queueid = qconf->tx_queue_id[port];
624 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
626 ret = rte_eth_tx_burst(port, queueid, m_table, n);
627 if (unlikely(ret < n)) {
629 rte_pktmbuf_free(m_table[ret]);
636 /* Enqueue a single packet, and send burst if queue is filled */
638 send_single_packet(struct rte_mbuf *m, uint16_t port)
641 struct thread_tx_conf *qconf;
644 qconf = (struct thread_tx_conf *)lthread_get_data();
646 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
648 len = qconf->tx_mbufs[port].len;
649 qconf->tx_mbufs[port].m_table[len] = m;
652 /* enough pkts to be sent */
653 if (unlikely(len == MAX_PKT_BURST)) {
654 send_burst(qconf, MAX_PKT_BURST, port);
658 qconf->tx_mbufs[port].len = len;
662 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
663 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
664 static __rte_always_inline void
665 send_packetsx4(uint16_t port,
666 struct rte_mbuf *m[], uint32_t num)
669 struct thread_tx_conf *qconf;
672 qconf = (struct thread_tx_conf *)lthread_get_data();
674 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
676 len = qconf->tx_mbufs[port].len;
679 * If TX buffer for that queue is empty, and we have enough packets,
680 * then send them straightway.
682 if (num >= MAX_TX_BURST && len == 0) {
683 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
684 if (unlikely(n < num)) {
686 rte_pktmbuf_free(m[n]);
693 * Put packets into TX buffer for that queue.
697 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
700 switch (n % FWDSTEP) {
703 qconf->tx_mbufs[port].m_table[len + j] = m[j];
707 qconf->tx_mbufs[port].m_table[len + j] = m[j];
711 qconf->tx_mbufs[port].m_table[len + j] = m[j];
715 qconf->tx_mbufs[port].m_table[len + j] = m[j];
722 /* enough pkts to be sent */
723 if (unlikely(len == MAX_PKT_BURST)) {
725 send_burst(qconf, MAX_PKT_BURST, port);
727 /* copy rest of the packets into the TX buffer. */
730 switch (len % FWDSTEP) {
733 qconf->tx_mbufs[port].m_table[j] = m[n + j];
737 qconf->tx_mbufs[port].m_table[j] = m[n + j];
741 qconf->tx_mbufs[port].m_table[j] = m[n + j];
745 qconf->tx_mbufs[port].m_table[j] = m[n + j];
751 qconf->tx_mbufs[port].len = len;
753 #endif /* APP_LOOKUP_LPM */
755 #ifdef DO_RFC_1812_CHECKS
757 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
759 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
761 * 1. The packet length reported by the Link Layer must be large
762 * enough to hold the minimum length legal IP datagram (20 bytes).
764 if (link_len < sizeof(struct ipv4_hdr))
767 /* 2. The IP checksum must be correct. */
768 /* this is checked in H/W */
771 * 3. The IP version number must be 4. If the version number is not 4
772 * then the packet may be another version of IP, such as IPng or
775 if (((pkt->version_ihl) >> 4) != 4)
778 * 4. The IP header length field must be large enough to hold the
779 * minimum length legal IP datagram (20 bytes = 5 words).
781 if ((pkt->version_ihl & 0xf) < 5)
785 * 5. The IP total length field must be large enough to hold the IP
786 * datagram header, whose length is specified in the IP header length
789 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
796 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
798 static __m128i mask0;
799 static __m128i mask1;
800 static __m128i mask2;
801 static inline uint16_t
802 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
803 lookup_struct_t *ipv4_l3fwd_lookup_struct)
806 union ipv4_5tuple_host key;
808 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
809 __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
810 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
812 key.xmm = _mm_and_si128(data, mask0);
813 /* Find destination port */
814 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
815 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
818 static inline uint16_t
819 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
820 lookup_struct_t *ipv6_l3fwd_lookup_struct)
823 union ipv6_5tuple_host key;
825 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
826 __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
827 __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
829 __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
830 sizeof(__m128i) + sizeof(__m128i)));
831 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
832 key.xmm[0] = _mm_and_si128(data0, mask1);
833 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
836 /* Get part of 5 tuple: dst port and src port and dst IP address higher
838 key.xmm[2] = _mm_and_si128(data2, mask2);
840 /* Find destination port */
841 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
842 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
846 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
848 static inline uint16_t
849 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
850 lookup_struct_t *ipv4_l3fwd_lookup_struct)
854 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
855 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
856 &next_hop) == 0) ? next_hop : portid);
859 static inline uint16_t
860 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
861 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
865 return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
866 ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
871 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
872 __attribute__((unused));
874 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
875 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
877 #define MASK_ALL_PKTS 0xff
878 #define EXCLUDE_1ST_PKT 0xfe
879 #define EXCLUDE_2ND_PKT 0xfd
880 #define EXCLUDE_3RD_PKT 0xfb
881 #define EXCLUDE_4TH_PKT 0xf7
882 #define EXCLUDE_5TH_PKT 0xef
883 #define EXCLUDE_6TH_PKT 0xdf
884 #define EXCLUDE_7TH_PKT 0xbf
885 #define EXCLUDE_8TH_PKT 0x7f
888 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
890 struct ether_hdr *eth_hdr[8];
891 struct ipv4_hdr *ipv4_hdr[8];
892 uint16_t dst_port[8];
894 union ipv4_5tuple_host key[8];
897 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
898 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
899 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
900 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
901 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
902 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
903 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
904 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
906 /* Handle IPv4 headers.*/
907 ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
908 sizeof(struct ether_hdr));
909 ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
910 sizeof(struct ether_hdr));
911 ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
912 sizeof(struct ether_hdr));
913 ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
914 sizeof(struct ether_hdr));
915 ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
916 sizeof(struct ether_hdr));
917 ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
918 sizeof(struct ether_hdr));
919 ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
920 sizeof(struct ether_hdr));
921 ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
922 sizeof(struct ether_hdr));
924 #ifdef DO_RFC_1812_CHECKS
925 /* Check to make sure the packet is valid (RFC1812) */
926 uint8_t valid_mask = MASK_ALL_PKTS;
928 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
929 rte_pktmbuf_free(m[0]);
930 valid_mask &= EXCLUDE_1ST_PKT;
932 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
933 rte_pktmbuf_free(m[1]);
934 valid_mask &= EXCLUDE_2ND_PKT;
936 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
937 rte_pktmbuf_free(m[2]);
938 valid_mask &= EXCLUDE_3RD_PKT;
940 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
941 rte_pktmbuf_free(m[3]);
942 valid_mask &= EXCLUDE_4TH_PKT;
944 if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
945 rte_pktmbuf_free(m[4]);
946 valid_mask &= EXCLUDE_5TH_PKT;
948 if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
949 rte_pktmbuf_free(m[5]);
950 valid_mask &= EXCLUDE_6TH_PKT;
952 if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
953 rte_pktmbuf_free(m[6]);
954 valid_mask &= EXCLUDE_7TH_PKT;
956 if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
957 rte_pktmbuf_free(m[7]);
958 valid_mask &= EXCLUDE_8TH_PKT;
960 if (unlikely(valid_mask != MASK_ALL_PKTS)) {
966 for (i = 0; i < 8; i++)
967 if ((0x1 << i) & valid_mask)
968 l3fwd_simple_forward(m[i], portid);
970 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
972 data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
973 sizeof(struct ether_hdr) +
974 offsetof(struct ipv4_hdr, time_to_live)));
975 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
976 sizeof(struct ether_hdr) +
977 offsetof(struct ipv4_hdr, time_to_live)));
978 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
979 sizeof(struct ether_hdr) +
980 offsetof(struct ipv4_hdr, time_to_live)));
981 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
982 sizeof(struct ether_hdr) +
983 offsetof(struct ipv4_hdr, time_to_live)));
984 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
985 sizeof(struct ether_hdr) +
986 offsetof(struct ipv4_hdr, time_to_live)));
987 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
988 sizeof(struct ether_hdr) +
989 offsetof(struct ipv4_hdr, time_to_live)));
990 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
991 sizeof(struct ether_hdr) +
992 offsetof(struct ipv4_hdr, time_to_live)));
993 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
994 sizeof(struct ether_hdr) +
995 offsetof(struct ipv4_hdr, time_to_live)));
997 key[0].xmm = _mm_and_si128(data[0], mask0);
998 key[1].xmm = _mm_and_si128(data[1], mask0);
999 key[2].xmm = _mm_and_si128(data[2], mask0);
1000 key[3].xmm = _mm_and_si128(data[3], mask0);
1001 key[4].xmm = _mm_and_si128(data[4], mask0);
1002 key[5].xmm = _mm_and_si128(data[5], mask0);
1003 key[6].xmm = _mm_and_si128(data[6], mask0);
1004 key[7].xmm = _mm_and_si128(data[7], mask0);
1006 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1007 &key[4], &key[5], &key[6], &key[7]};
1009 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1010 &key_array[0], 8, ret);
1011 dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1012 dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1013 dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1014 dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1015 dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1016 dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1017 dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1018 dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1020 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1021 (enabled_port_mask & 1 << dst_port[0]) == 0)
1022 dst_port[0] = portid;
1023 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1024 (enabled_port_mask & 1 << dst_port[1]) == 0)
1025 dst_port[1] = portid;
1026 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1027 (enabled_port_mask & 1 << dst_port[2]) == 0)
1028 dst_port[2] = portid;
1029 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1030 (enabled_port_mask & 1 << dst_port[3]) == 0)
1031 dst_port[3] = portid;
1032 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1033 (enabled_port_mask & 1 << dst_port[4]) == 0)
1034 dst_port[4] = portid;
1035 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1036 (enabled_port_mask & 1 << dst_port[5]) == 0)
1037 dst_port[5] = portid;
1038 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1039 (enabled_port_mask & 1 << dst_port[6]) == 0)
1040 dst_port[6] = portid;
1041 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1042 (enabled_port_mask & 1 << dst_port[7]) == 0)
1043 dst_port[7] = portid;
1045 #ifdef DO_RFC_1812_CHECKS
1046 /* Update time to live and header checksum */
1047 --(ipv4_hdr[0]->time_to_live);
1048 --(ipv4_hdr[1]->time_to_live);
1049 --(ipv4_hdr[2]->time_to_live);
1050 --(ipv4_hdr[3]->time_to_live);
1051 ++(ipv4_hdr[0]->hdr_checksum);
1052 ++(ipv4_hdr[1]->hdr_checksum);
1053 ++(ipv4_hdr[2]->hdr_checksum);
1054 ++(ipv4_hdr[3]->hdr_checksum);
1055 --(ipv4_hdr[4]->time_to_live);
1056 --(ipv4_hdr[5]->time_to_live);
1057 --(ipv4_hdr[6]->time_to_live);
1058 --(ipv4_hdr[7]->time_to_live);
1059 ++(ipv4_hdr[4]->hdr_checksum);
1060 ++(ipv4_hdr[5]->hdr_checksum);
1061 ++(ipv4_hdr[6]->hdr_checksum);
1062 ++(ipv4_hdr[7]->hdr_checksum);
1066 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1067 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1068 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1069 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1070 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1071 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1072 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1073 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1076 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1077 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1078 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1079 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1080 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1081 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1082 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1083 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1085 send_single_packet(m[0], (uint8_t)dst_port[0]);
1086 send_single_packet(m[1], (uint8_t)dst_port[1]);
1087 send_single_packet(m[2], (uint8_t)dst_port[2]);
1088 send_single_packet(m[3], (uint8_t)dst_port[3]);
1089 send_single_packet(m[4], (uint8_t)dst_port[4]);
1090 send_single_packet(m[5], (uint8_t)dst_port[5]);
1091 send_single_packet(m[6], (uint8_t)dst_port[6]);
1092 send_single_packet(m[7], (uint8_t)dst_port[7]);
1096 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1097 __m128i mask1, union ipv6_5tuple_host *key)
1099 __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1100 __m128i *, sizeof(struct ether_hdr) +
1101 offsetof(struct ipv6_hdr, payload_len)));
1102 __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1103 __m128i *, sizeof(struct ether_hdr) +
1104 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1105 __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1106 __m128i *, sizeof(struct ether_hdr) +
1107 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1109 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1110 key->xmm[1] = tmpdata1;
1111 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1115 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1118 uint16_t dst_port[8];
1119 struct ether_hdr *eth_hdr[8];
1120 union ipv6_5tuple_host key[8];
1122 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1124 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1125 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1126 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1127 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1128 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1129 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1130 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1131 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1133 /* Handle IPv6 headers.*/
1134 ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1135 sizeof(struct ether_hdr));
1136 ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1137 sizeof(struct ether_hdr));
1138 ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1139 sizeof(struct ether_hdr));
1140 ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1141 sizeof(struct ether_hdr));
1142 ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1143 sizeof(struct ether_hdr));
1144 ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1145 sizeof(struct ether_hdr));
1146 ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1147 sizeof(struct ether_hdr));
1148 ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1149 sizeof(struct ether_hdr));
1151 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1152 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1153 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1154 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1155 get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1156 get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1157 get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1158 get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1160 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1161 &key[4], &key[5], &key[6], &key[7]};
1163 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1164 &key_array[0], 4, ret);
1165 dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1166 dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1167 dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1168 dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1169 dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1170 dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1171 dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1172 dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1174 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1175 (enabled_port_mask & 1 << dst_port[0]) == 0)
1176 dst_port[0] = portid;
1177 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1178 (enabled_port_mask & 1 << dst_port[1]) == 0)
1179 dst_port[1] = portid;
1180 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1181 (enabled_port_mask & 1 << dst_port[2]) == 0)
1182 dst_port[2] = portid;
1183 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1184 (enabled_port_mask & 1 << dst_port[3]) == 0)
1185 dst_port[3] = portid;
1186 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1187 (enabled_port_mask & 1 << dst_port[4]) == 0)
1188 dst_port[4] = portid;
1189 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1190 (enabled_port_mask & 1 << dst_port[5]) == 0)
1191 dst_port[5] = portid;
1192 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1193 (enabled_port_mask & 1 << dst_port[6]) == 0)
1194 dst_port[6] = portid;
1195 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1196 (enabled_port_mask & 1 << dst_port[7]) == 0)
1197 dst_port[7] = portid;
1200 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1201 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1202 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1203 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1204 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1205 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1206 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1207 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1210 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1211 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1212 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1213 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1214 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1215 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1216 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1217 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1219 send_single_packet(m[0], dst_port[0]);
1220 send_single_packet(m[1], dst_port[1]);
1221 send_single_packet(m[2], dst_port[2]);
1222 send_single_packet(m[3], dst_port[3]);
1223 send_single_packet(m[4], dst_port[4]);
1224 send_single_packet(m[5], dst_port[5]);
1225 send_single_packet(m[6], dst_port[6]);
1226 send_single_packet(m[7], dst_port[7]);
1229 #endif /* APP_LOOKUP_METHOD */
1231 static __rte_always_inline void
1232 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1234 struct ether_hdr *eth_hdr;
1235 struct ipv4_hdr *ipv4_hdr;
1238 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1240 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1241 /* Handle IPv4 headers.*/
1242 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1243 sizeof(struct ether_hdr));
1245 #ifdef DO_RFC_1812_CHECKS
1246 /* Check to make sure the packet is valid (RFC1812) */
1247 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1248 rte_pktmbuf_free(m);
1253 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1254 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1255 if (dst_port >= RTE_MAX_ETHPORTS ||
1256 (enabled_port_mask & 1 << dst_port) == 0)
1259 #ifdef DO_RFC_1812_CHECKS
1260 /* Update time to live and header checksum */
1261 --(ipv4_hdr->time_to_live);
1262 ++(ipv4_hdr->hdr_checksum);
1265 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1268 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1270 send_single_packet(m, dst_port);
1271 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1272 /* Handle IPv6 headers.*/
1273 struct ipv6_hdr *ipv6_hdr;
1275 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1276 sizeof(struct ether_hdr));
1278 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1279 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1281 if (dst_port >= RTE_MAX_ETHPORTS ||
1282 (enabled_port_mask & 1 << dst_port) == 0)
1286 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1289 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1291 send_single_packet(m, dst_port);
1293 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1294 rte_pktmbuf_free(m);
1297 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1298 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1299 #ifdef DO_RFC_1812_CHECKS
1301 #define IPV4_MIN_VER_IHL 0x45
1302 #define IPV4_MAX_VER_IHL 0x4f
1303 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1305 /* Minimum value of IPV4 total length (20B) in network byte order. */
1306 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1309 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1310 * - The IP version number must be 4.
1311 * - The IP header length field must be large enough to hold the
1312 * minimum length legal IP datagram (20 bytes = 5 words).
1313 * - The IP total length field must be large enough to hold the IP
1314 * datagram header, whose length is specified in the IP header length
1316 * If we encounter invalid IPV4 packet, then set destination port for it
1317 * to BAD_PORT value.
1319 static __rte_always_inline void
1320 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1324 if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1325 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1327 ipv4_hdr->time_to_live--;
1328 ipv4_hdr->hdr_checksum++;
1330 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1331 ((uint8_t)ipv4_hdr->total_length == 0 &&
1332 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1339 #define rfc1812_process(mb, dp, ptype) do { } while (0)
1340 #endif /* DO_RFC_1812_CHECKS */
1341 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1344 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1345 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1347 static __rte_always_inline uint16_t
1348 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1351 struct ipv6_hdr *ipv6_hdr;
1352 struct ether_hdr *eth_hdr;
1354 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1355 return (uint16_t) ((rte_lpm_lookup(
1356 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1357 &next_hop) == 0) ? next_hop : portid);
1359 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1361 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1362 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1364 return (uint16_t) ((rte_lpm6_lookup(
1365 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1366 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1375 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1377 struct ether_hdr *eth_hdr;
1378 struct ipv4_hdr *ipv4_hdr;
1383 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1384 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1386 dst_ipv4 = ipv4_hdr->dst_addr;
1387 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1388 dp = get_dst_port(pkt, dst_ipv4, portid);
1390 te = _mm_load_si128((__m128i *)eth_hdr);
1394 rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1396 te = _mm_blend_epi16(te, ve, MASK_ETH);
1397 _mm_store_si128((__m128i *)eth_hdr, te);
1401 * Read packet_type and destination IPV4 addresses from 4 mbufs.
1404 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1406 uint32_t *ipv4_flag)
1408 struct ipv4_hdr *ipv4_hdr;
1409 struct ether_hdr *eth_hdr;
1410 uint32_t x0, x1, x2, x3;
1412 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1413 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1414 x0 = ipv4_hdr->dst_addr;
1415 ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1417 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1418 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1419 x1 = ipv4_hdr->dst_addr;
1420 ipv4_flag[0] &= pkt[1]->packet_type;
1422 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1423 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1424 x2 = ipv4_hdr->dst_addr;
1425 ipv4_flag[0] &= pkt[2]->packet_type;
1427 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1428 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1429 x3 = ipv4_hdr->dst_addr;
1430 ipv4_flag[0] &= pkt[3]->packet_type;
1432 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1436 * Lookup into LPM for destination port.
1437 * If lookup fails, use incoming port (portid) as destination port.
1440 processx4_step2(__m128i dip,
1443 struct rte_mbuf *pkt[FWDSTEP],
1444 uint16_t dprt[FWDSTEP])
1447 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1448 4, 5, 6, 7, 0, 1, 2, 3);
1450 /* Byte swap 4 IPV4 addresses. */
1451 dip = _mm_shuffle_epi8(dip, bswap_mask);
1453 /* if all 4 packets are IPV4. */
1454 if (likely(ipv4_flag)) {
1455 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1458 /* get rid of unused upper 16 bit for each dport. */
1459 dst.x = _mm_packs_epi32(dst.x, dst.x);
1460 *(uint64_t *)dprt = dst.u64[0];
1463 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1464 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1465 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1466 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1471 * Update source and destination MAC addresses in the ethernet header.
1472 * Perform RFC1812 checks and updates for IPV4 packets.
1475 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1477 __m128i te[FWDSTEP];
1478 __m128i ve[FWDSTEP];
1479 __m128i *p[FWDSTEP];
1481 p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1482 p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1483 p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1484 p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1486 ve[0] = val_eth[dst_port[0]];
1487 te[0] = _mm_load_si128(p[0]);
1489 ve[1] = val_eth[dst_port[1]];
1490 te[1] = _mm_load_si128(p[1]);
1492 ve[2] = val_eth[dst_port[2]];
1493 te[2] = _mm_load_si128(p[2]);
1495 ve[3] = val_eth[dst_port[3]];
1496 te[3] = _mm_load_si128(p[3]);
1498 /* Update first 12 bytes, keep rest bytes intact. */
1499 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1500 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1501 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1502 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1504 _mm_store_si128(p[0], te[0]);
1505 _mm_store_si128(p[1], te[1]);
1506 _mm_store_si128(p[2], te[2]);
1507 _mm_store_si128(p[3], te[3]);
1509 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1510 &dst_port[0], pkt[0]->packet_type);
1511 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1512 &dst_port[1], pkt[1]->packet_type);
1513 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1514 &dst_port[2], pkt[2]->packet_type);
1515 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1516 &dst_port[3], pkt[3]->packet_type);
1520 * We group consecutive packets with the same destionation port into one burst.
1521 * To avoid extra latency this is done together with some other packet
1522 * processing, but after we made a final decision about packet's destination.
1523 * To do this we maintain:
1524 * pnum - array of number of consecutive packets with the same dest port for
1525 * each packet in the input burst.
1526 * lp - pointer to the last updated element in the pnum.
1527 * dlp - dest port value lp corresponds to.
1530 #define GRPSZ (1 << FWDSTEP)
1531 #define GRPMSK (GRPSZ - 1)
1533 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1534 if (likely((dlp) == (dcp)[(idx)])) { \
1537 (dlp) = (dcp)[idx]; \
1538 (lp) = (pn) + (idx); \
1544 * Group consecutive packets with the same destination port in bursts of 4.
1545 * Suppose we have array of destionation ports:
1546 * dst_port[] = {a, b, c, d,, e, ... }
1547 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1548 * We doing 4 comparisons at once and the result is 4 bit mask.
1549 * This mask is used as an index into prebuild array of pnum values.
1551 static inline uint16_t *
1552 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1554 static const struct {
1555 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1556 int32_t idx; /* index for new last updated elemnet. */
1557 uint16_t lpv; /* add value to the last updated element. */
1560 /* 0: a != b, b != c, c != d, d != e */
1561 .pnum = UINT64_C(0x0001000100010001),
1566 /* 1: a == b, b != c, c != d, d != e */
1567 .pnum = UINT64_C(0x0001000100010002),
1572 /* 2: a != b, b == c, c != d, d != e */
1573 .pnum = UINT64_C(0x0001000100020001),
1578 /* 3: a == b, b == c, c != d, d != e */
1579 .pnum = UINT64_C(0x0001000100020003),
1584 /* 4: a != b, b != c, c == d, d != e */
1585 .pnum = UINT64_C(0x0001000200010001),
1590 /* 5: a == b, b != c, c == d, d != e */
1591 .pnum = UINT64_C(0x0001000200010002),
1596 /* 6: a != b, b == c, c == d, d != e */
1597 .pnum = UINT64_C(0x0001000200030001),
1602 /* 7: a == b, b == c, c == d, d != e */
1603 .pnum = UINT64_C(0x0001000200030004),
1608 /* 8: a != b, b != c, c != d, d == e */
1609 .pnum = UINT64_C(0x0002000100010001),
1614 /* 9: a == b, b != c, c != d, d == e */
1615 .pnum = UINT64_C(0x0002000100010002),
1620 /* 0xa: a != b, b == c, c != d, d == e */
1621 .pnum = UINT64_C(0x0002000100020001),
1626 /* 0xb: a == b, b == c, c != d, d == e */
1627 .pnum = UINT64_C(0x0002000100020003),
1632 /* 0xc: a != b, b != c, c == d, d == e */
1633 .pnum = UINT64_C(0x0002000300010001),
1638 /* 0xd: a == b, b != c, c == d, d == e */
1639 .pnum = UINT64_C(0x0002000300010002),
1644 /* 0xe: a != b, b == c, c == d, d == e */
1645 .pnum = UINT64_C(0x0002000300040001),
1650 /* 0xf: a == b, b == c, c == d, d == e */
1651 .pnum = UINT64_C(0x0002000300040005),
1658 uint16_t u16[FWDSTEP + 1];
1660 } *pnum = (void *)pn;
1664 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1665 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1666 v = _mm_movemask_ps((__m128)dp1);
1668 /* update last port counter. */
1669 lp[0] += gptbl[v].lpv;
1671 /* if dest port value has changed. */
1673 pnum->u64 = gptbl[v].pnum;
1674 pnum->u16[FWDSTEP] = 1;
1675 lp = pnum->u16 + gptbl[v].idx;
1681 #endif /* APP_LOOKUP_METHOD */
1684 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1690 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1691 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1695 uint16_t dst_port[MAX_PKT_BURST];
1696 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1697 uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1698 uint16_t pnum[MAX_PKT_BURST + 1];
1702 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1703 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1706 * Send nb_rx - nb_rx%8 packets
1709 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1711 for (j = 0; j < n; j += 8) {
1713 pkts_burst[j]->packet_type &
1714 pkts_burst[j+1]->packet_type &
1715 pkts_burst[j+2]->packet_type &
1716 pkts_burst[j+3]->packet_type &
1717 pkts_burst[j+4]->packet_type &
1718 pkts_burst[j+5]->packet_type &
1719 pkts_burst[j+6]->packet_type &
1720 pkts_burst[j+7]->packet_type;
1721 if (pkt_type & RTE_PTYPE_L3_IPV4) {
1722 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1723 } else if (pkt_type &
1724 RTE_PTYPE_L3_IPV6) {
1725 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1727 l3fwd_simple_forward(pkts_burst[j], portid);
1728 l3fwd_simple_forward(pkts_burst[j+1], portid);
1729 l3fwd_simple_forward(pkts_burst[j+2], portid);
1730 l3fwd_simple_forward(pkts_burst[j+3], portid);
1731 l3fwd_simple_forward(pkts_burst[j+4], portid);
1732 l3fwd_simple_forward(pkts_burst[j+5], portid);
1733 l3fwd_simple_forward(pkts_burst[j+6], portid);
1734 l3fwd_simple_forward(pkts_burst[j+7], portid);
1737 for (; j < nb_rx ; j++)
1738 l3fwd_simple_forward(pkts_burst[j], portid);
1740 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1742 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1743 for (j = 0; j != k; j += FWDSTEP)
1744 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1745 &ipv4_flag[j / FWDSTEP]);
1747 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1748 for (j = 0; j != k; j += FWDSTEP)
1749 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1750 portid, &pkts_burst[j], &dst_port[j]);
1753 * Finish packet processing and group consecutive
1754 * packets with the same destination port.
1756 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1763 processx4_step3(pkts_burst, dst_port);
1765 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1766 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1768 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1769 processx4_step3(&pkts_burst[j], &dst_port[j]);
1773 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1775 dp2 = _mm_loadu_si128(
1776 (__m128i *)&dst_port[j - FWDSTEP + 1]);
1777 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1781 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1783 dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1784 sizeof(dst_port[0]));
1788 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1790 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1791 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1794 * remove values added by the last repeated
1798 dlp = dst_port[j - 1];
1800 /* set dlp and lp to the never used values. */
1802 lp = pnum + MAX_PKT_BURST;
1805 /* Process up to last 3 packets one by one. */
1806 switch (nb_rx % FWDSTEP) {
1808 process_packet(pkts_burst[j], dst_port + j, portid);
1809 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1813 process_packet(pkts_burst[j], dst_port + j, portid);
1814 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1818 process_packet(pkts_burst[j], dst_port + j, portid);
1819 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1824 * Send packets out, through destination port.
1825 * Consecuteve pacekts with the same destination port
1826 * are already grouped together.
1827 * If destination port for the packet equals BAD_PORT,
1828 * then free the packet without sending it out.
1830 for (j = 0; j < nb_rx; j += k) {
1838 if (likely(pn != BAD_PORT))
1839 send_packetsx4(pn, pkts_burst + j, k);
1841 for (m = j; m != j + k; m++)
1842 rte_pktmbuf_free(pkts_burst[m]);
1846 #endif /* APP_LOOKUP_METHOD */
1847 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1849 /* Prefetch first packets */
1850 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1851 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1853 /* Prefetch and forward already prefetched packets */
1854 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1855 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1856 j + PREFETCH_OFFSET], void *));
1857 l3fwd_simple_forward(pkts_burst[j], portid);
1860 /* Forward remaining prefetched packets */
1861 for (; j < nb_rx; j++)
1862 l3fwd_simple_forward(pkts_burst[j], portid);
1864 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1868 #if (APP_CPU_LOAD > 0)
1871 * CPU-load stats collector
1874 cpu_load_collector(__rte_unused void *arg) {
1877 uint64_t prev_tsc, diff_tsc, cur_tsc;
1878 uint64_t total[MAX_CPU] = { 0 };
1879 unsigned min_cpu = MAX_CPU;
1880 unsigned max_cpu = 0;
1885 unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1886 struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1888 struct thread_conf *thread_conf;
1890 const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1891 US_PER_S * CPU_LOAD_TIMEOUT_US;
1895 * Wait for all threads
1898 printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1901 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1904 while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1907 for (i = 0; i < n_rx_thread; i++) {
1909 thread_conf = &rx_thread[i].conf;
1910 cpu_id = thread_conf->cpu_id;
1911 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1913 if (cpu_id > max_cpu)
1915 if (cpu_id < min_cpu)
1918 for (i = 0; i < n_tx_thread; i++) {
1920 thread_conf = &tx_thread[i].conf;
1921 cpu_id = thread_conf->cpu_id;
1922 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1924 if (thread_conf->cpu_id > max_cpu)
1925 max_cpu = thread_conf->cpu_id;
1926 if (thread_conf->cpu_id < min_cpu)
1927 min_cpu = thread_conf->cpu_id;
1933 for (i = min_cpu; i <= max_cpu; i++) {
1934 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1935 for (k = 0; k < n_thread_per_cpu[i]; k++)
1936 if (thread_per_cpu[i][k]->busy[j]) {
1941 cpu_load.hits[j][i]++;
1953 cur_tsc = rte_rdtsc();
1955 diff_tsc = cur_tsc - prev_tsc;
1956 if (unlikely(diff_tsc > interval_tsc)) {
1960 printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1961 n_rx_thread, n_tx_thread);
1963 printf("cpu# proc%% poll%% overhead%%\n\n");
1965 for (i = min_cpu; i <= max_cpu; i++) {
1967 printf("CPU %d:", i);
1968 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1969 printf("%7" PRIu64 "",
1970 cpu_load.hits[j][i] * 100 / cpu_load.counter);
1971 hits += cpu_load.hits[j][i];
1972 cpu_load.hits[j][i] = 0;
1974 printf("%7" PRIu64 "\n",
1975 100 - total[i] * 100 / cpu_load.counter);
1978 cpu_load.counter = 0;
1985 #endif /* APP_CPU_LOAD */
1988 * Null processing lthread loop
1990 * This loop is used to start empty scheduler on lcore.
1993 lthread_null(__rte_unused void *args)
1995 int lcore_id = rte_lcore_id();
1997 RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2002 /* main processing loop */
2004 lthread_tx_per_ring(void *dummy)
2008 struct rte_ring *ring;
2009 struct thread_tx_conf *tx_conf;
2010 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2011 struct lthread_cond *ready;
2013 tx_conf = (struct thread_tx_conf *)dummy;
2014 ring = tx_conf->ring;
2015 ready = *tx_conf->ready;
2017 lthread_set_data((void *)tx_conf);
2020 * Move this lthread to lcore
2022 lthread_set_affinity(tx_conf->conf.lcore_id);
2024 RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2027 rte_atomic16_inc(&tx_counter);
2031 * Read packet from ring
2033 SET_CPU_BUSY(tx_conf, CPU_POLL);
2034 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2035 MAX_PKT_BURST, NULL);
2036 SET_CPU_IDLE(tx_conf, CPU_POLL);
2039 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2040 portid = pkts_burst[0]->port;
2041 process_burst(pkts_burst, nb_rx, portid);
2042 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2045 lthread_cond_wait(ready, 0);
2052 * Main tx-lthreads spawner lthread.
2054 * This lthread is used to spawn one new lthread per ring from producers.
2058 lthread_tx(void *args)
2064 struct thread_tx_conf *tx_conf;
2066 tx_conf = (struct thread_tx_conf *)args;
2067 lthread_set_data((void *)tx_conf);
2070 * Move this lthread to the selected lcore
2072 lthread_set_affinity(tx_conf->conf.lcore_id);
2075 * Spawn tx readers (one per input ring)
2077 lthread_create(<, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2080 lcore_id = rte_lcore_id();
2082 RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2084 tx_conf->conf.cpu_id = sched_getcpu();
2087 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2090 * TX burst queue drain
2092 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2093 if (tx_conf->tx_mbufs[portid].len == 0)
2095 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2096 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2097 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2098 tx_conf->tx_mbufs[portid].len = 0;
2106 lthread_rx(void *dummy)
2114 int len[RTE_MAX_LCORE] = { 0 };
2115 int old_len, new_len;
2116 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2117 struct thread_rx_conf *rx_conf;
2119 rx_conf = (struct thread_rx_conf *)dummy;
2120 lthread_set_data((void *)rx_conf);
2123 * Move this lthread to lcore
2125 lthread_set_affinity(rx_conf->conf.lcore_id);
2127 if (rx_conf->n_rx_queue == 0) {
2128 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2132 RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2134 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2136 portid = rx_conf->rx_queue_list[i].port_id;
2137 queueid = rx_conf->rx_queue_list[i].queue_id;
2138 RTE_LOG(INFO, L3FWD,
2139 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2140 rte_lcore_id(), portid, queueid);
2144 * Init all condition variables (one per rx thread)
2146 for (i = 0; i < rx_conf->n_rx_queue; i++)
2147 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2151 rx_conf->conf.cpu_id = sched_getcpu();
2152 rte_atomic16_inc(&rx_counter);
2156 * Read packet from RX queues
2158 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2159 portid = rx_conf->rx_queue_list[i].port_id;
2160 queueid = rx_conf->rx_queue_list[i].queue_id;
2162 SET_CPU_BUSY(rx_conf, CPU_POLL);
2163 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2165 SET_CPU_IDLE(rx_conf, CPU_POLL);
2168 worker_id = (worker_id + 1) % rx_conf->n_ring;
2169 old_len = len[worker_id];
2171 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2172 ret = rte_ring_sp_enqueue_burst(
2173 rx_conf->ring[worker_id],
2174 (void **) pkts_burst,
2177 new_len = old_len + ret;
2179 if (new_len >= BURST_SIZE) {
2180 lthread_cond_signal(rx_conf->ready[worker_id]);
2184 len[worker_id] = new_len;
2186 if (unlikely(ret < nb_rx)) {
2189 for (k = ret; k < nb_rx; k++) {
2190 struct rte_mbuf *m = pkts_burst[k];
2192 rte_pktmbuf_free(m);
2195 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2205 * Start scheduler with initial lthread on lcore
2207 * This lthread loop spawns all rx and tx lthreads on master lcore
2211 lthread_spawner(__rte_unused void *arg)
2213 struct lthread *lt[MAX_THREAD];
2217 printf("Entering lthread_spawner\n");
2220 * Create producers (rx threads) on default lcore
2222 for (i = 0; i < n_rx_thread; i++) {
2223 rx_thread[i].conf.thread_id = i;
2224 lthread_create(<[n_thread], -1, lthread_rx,
2225 (void *)&rx_thread[i]);
2230 * Wait for all producers. Until some producers can be started on the same
2231 * scheduler as this lthread, yielding is required to let them to run and
2232 * prevent deadlock here.
2234 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2235 lthread_sleep(100000);
2238 * Create consumers (tx threads) on default lcore_id
2240 for (i = 0; i < n_tx_thread; i++) {
2241 tx_thread[i].conf.thread_id = i;
2242 lthread_create(<[n_thread], -1, lthread_tx,
2243 (void *)&tx_thread[i]);
2248 * Wait for all threads finished
2250 for (i = 0; i < n_thread; i++)
2251 lthread_join(lt[i], NULL);
2257 * Start master scheduler with initial lthread spawning rx and tx lthreads
2258 * (main_lthread_master).
2261 lthread_master_spawner(__rte_unused void *arg) {
2263 int lcore_id = rte_lcore_id();
2265 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2266 lthread_create(<, -1, lthread_spawner, NULL);
2273 * Start scheduler on lcore.
2276 sched_spawner(__rte_unused void *arg) {
2278 int lcore_id = rte_lcore_id();
2281 if (lcore_id == cpu_load_lcore_id) {
2282 cpu_load_collector(arg);
2285 #endif /* APP_CPU_LOAD */
2287 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2288 lthread_create(<, -1, lthread_null, NULL);
2294 /* main processing loop */
2296 pthread_tx(void *dummy)
2298 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2299 uint64_t prev_tsc, diff_tsc, cur_tsc;
2302 struct thread_tx_conf *tx_conf;
2304 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2305 US_PER_S * BURST_TX_DRAIN_US;
2309 tx_conf = (struct thread_tx_conf *)dummy;
2311 RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2313 tx_conf->conf.cpu_id = sched_getcpu();
2314 rte_atomic16_inc(&tx_counter);
2317 cur_tsc = rte_rdtsc();
2320 * TX burst queue drain
2322 diff_tsc = cur_tsc - prev_tsc;
2323 if (unlikely(diff_tsc > drain_tsc)) {
2326 * This could be optimized (use queueid instead of
2327 * portid), but it is not called so often
2329 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2330 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2331 if (tx_conf->tx_mbufs[portid].len == 0)
2333 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2334 tx_conf->tx_mbufs[portid].len = 0;
2336 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2342 * Read packet from ring
2344 SET_CPU_BUSY(tx_conf, CPU_POLL);
2345 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2346 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2347 SET_CPU_IDLE(tx_conf, CPU_POLL);
2349 if (unlikely(nb_rx == 0)) {
2354 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2355 portid = pkts_burst[0]->port;
2356 process_burst(pkts_burst, nb_rx, portid);
2357 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2363 pthread_rx(void *dummy)
2372 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2374 struct thread_rx_conf *rx_conf;
2376 lcore_id = rte_lcore_id();
2377 rx_conf = (struct thread_rx_conf *)dummy;
2379 if (rx_conf->n_rx_queue == 0) {
2380 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2384 RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2386 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2388 portid = rx_conf->rx_queue_list[i].port_id;
2389 queueid = rx_conf->rx_queue_list[i].queue_id;
2390 RTE_LOG(INFO, L3FWD,
2391 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2392 lcore_id, portid, queueid);
2396 rx_conf->conf.cpu_id = sched_getcpu();
2397 rte_atomic16_inc(&rx_counter);
2401 * Read packet from RX queues
2403 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2404 portid = rx_conf->rx_queue_list[i].port_id;
2405 queueid = rx_conf->rx_queue_list[i].queue_id;
2407 SET_CPU_BUSY(rx_conf, CPU_POLL);
2408 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2410 SET_CPU_IDLE(rx_conf, CPU_POLL);
2417 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2418 worker_id = (worker_id + 1) % rx_conf->n_ring;
2419 n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2420 (void **)pkts_burst, nb_rx, NULL);
2422 if (unlikely(n != nb_rx)) {
2425 for (k = n; k < nb_rx; k++) {
2426 struct rte_mbuf *m = pkts_burst[k];
2428 rte_pktmbuf_free(m);
2432 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2442 pthread_run(__rte_unused void *arg) {
2443 int lcore_id = rte_lcore_id();
2446 for (i = 0; i < n_rx_thread; i++)
2447 if (rx_thread[i].conf.lcore_id == lcore_id) {
2448 printf("Start rx thread on %d...\n", lcore_id);
2449 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2450 RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2451 pthread_rx((void *)&rx_thread[i]);
2455 for (i = 0; i < n_tx_thread; i++)
2456 if (tx_thread[i].conf.lcore_id == lcore_id) {
2457 printf("Start tx thread on %d...\n", lcore_id);
2458 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2459 RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2460 pthread_tx((void *)&tx_thread[i]);
2465 if (lcore_id == cpu_load_lcore_id)
2466 cpu_load_collector(arg);
2467 #endif /* APP_CPU_LOAD */
2473 check_lcore_params(void)
2475 uint8_t queue, lcore;
2479 for (i = 0; i < nb_rx_thread_params; ++i) {
2480 queue = rx_thread_params[i].queue_id;
2481 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2482 printf("invalid queue number: %hhu\n", queue);
2485 lcore = rx_thread_params[i].lcore_id;
2486 if (!rte_lcore_is_enabled(lcore)) {
2487 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2490 socketid = rte_lcore_to_socket_id(lcore);
2491 if ((socketid != 0) && (numa_on == 0))
2492 printf("warning: lcore %hhu is on socket %d with numa off\n",
2499 check_port_config(void)
2504 for (i = 0; i < nb_rx_thread_params; ++i) {
2505 portid = rx_thread_params[i].port_id;
2506 if ((enabled_port_mask & (1 << portid)) == 0) {
2507 printf("port %u is not enabled in port mask\n", portid);
2510 if (!rte_eth_dev_is_valid_port(portid)) {
2511 printf("port %u is not present on the board\n", portid);
2519 get_port_n_rx_queues(const uint16_t port)
2524 for (i = 0; i < nb_rx_thread_params; ++i)
2525 if (rx_thread_params[i].port_id == port &&
2526 rx_thread_params[i].queue_id > queue)
2527 queue = rx_thread_params[i].queue_id;
2529 return (uint8_t)(++queue);
2536 struct thread_rx_conf *rx_conf;
2537 struct thread_tx_conf *tx_conf;
2538 unsigned rx_thread_id, tx_thread_id;
2540 struct rte_ring *ring = NULL;
2542 for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2544 tx_conf = &tx_thread[tx_thread_id];
2546 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2547 tx_conf->conf.thread_id);
2549 rx_thread_id = tx_conf->conf.thread_id;
2550 if (rx_thread_id > n_tx_thread) {
2551 printf("connection from tx-thread %u to rx-thread %u fails "
2552 "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2556 rx_conf = &rx_thread[rx_thread_id];
2557 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2559 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2560 socket_io, rx_thread_id, tx_thread_id);
2562 ring = rte_ring_create(name, 1024 * 4, socket_io,
2563 RING_F_SP_ENQ | RING_F_SC_DEQ);
2566 rte_panic("Cannot create ring to connect rx-thread %u "
2567 "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2570 rx_conf->ring[rx_conf->n_ring] = ring;
2572 tx_conf->ring = ring;
2573 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2581 init_rx_queues(void)
2583 uint16_t i, nb_rx_queue;
2588 for (i = 0; i < nb_rx_thread_params; ++i) {
2589 thread = rx_thread_params[i].thread_id;
2590 nb_rx_queue = rx_thread[thread].n_rx_queue;
2592 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2593 printf("error: too many queues (%u) for thread: %u\n",
2594 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2598 rx_thread[thread].conf.thread_id = thread;
2599 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2600 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2601 rx_thread_params[i].port_id;
2602 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2603 rx_thread_params[i].queue_id;
2604 rx_thread[thread].n_rx_queue++;
2606 if (thread >= n_rx_thread)
2607 n_rx_thread = thread + 1;
2614 init_tx_threads(void)
2619 for (i = 0; i < nb_tx_thread_params; ++i) {
2620 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2621 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2629 print_usage(const char *prgname)
2631 printf("%s [EAL options] -- -p PORTMASK -P"
2632 " [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2633 " [--tx (lcore,thread)[,(lcore,thread]]"
2634 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2635 " [--parse-ptype]\n\n"
2636 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2637 " -P : enable promiscuous mode\n"
2638 " --rx (port,queue,lcore,thread): rx queues configuration\n"
2639 " --tx (lcore,thread): tx threads configuration\n"
2640 " --stat-lcore LCORE: use lcore for stat collector\n"
2641 " --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2642 " --no-numa: optional, disable numa awareness\n"
2643 " --ipv6: optional, specify it if running ipv6 packets\n"
2644 " --enable-jumbo: enable jumbo frame"
2645 " which max packet len is PKTLEN in decimal (64-9600)\n"
2646 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2647 " --no-lthreads: turn off lthread model\n"
2648 " --parse-ptype: set to use software to analyze packet type\n\n",
2652 static int parse_max_pkt_len(const char *pktlen)
2657 /* parse decimal string */
2658 len = strtoul(pktlen, &end, 10);
2659 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2669 parse_portmask(const char *portmask)
2674 /* parse hexadecimal string */
2675 pm = strtoul(portmask, &end, 16);
2676 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2685 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2687 parse_hash_entry_number(const char *hash_entry_num)
2690 unsigned long hash_en;
2692 /* parse hexadecimal string */
2693 hash_en = strtoul(hash_entry_num, &end, 16);
2694 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2705 parse_rx_config(const char *q_arg)
2708 const char *p, *p0 = q_arg;
2717 unsigned long int_fld[_NUM_FLD];
2718 char *str_fld[_NUM_FLD];
2722 nb_rx_thread_params = 0;
2724 while ((p = strchr(p0, '(')) != NULL) {
2726 p0 = strchr(p, ')');
2731 if (size >= sizeof(s))
2734 snprintf(s, sizeof(s), "%.*s", size, p);
2735 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2737 for (i = 0; i < _NUM_FLD; i++) {
2739 int_fld[i] = strtoul(str_fld[i], &end, 0);
2740 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2743 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2744 printf("exceeded max number of rx params: %hu\n",
2745 nb_rx_thread_params);
2748 rx_thread_params_array[nb_rx_thread_params].port_id =
2750 rx_thread_params_array[nb_rx_thread_params].queue_id =
2751 (uint8_t)int_fld[FLD_QUEUE];
2752 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2753 (uint8_t)int_fld[FLD_LCORE];
2754 rx_thread_params_array[nb_rx_thread_params].thread_id =
2755 (uint8_t)int_fld[FLD_THREAD];
2756 ++nb_rx_thread_params;
2758 rx_thread_params = rx_thread_params_array;
2763 parse_tx_config(const char *q_arg)
2766 const char *p, *p0 = q_arg;
2773 unsigned long int_fld[_NUM_FLD];
2774 char *str_fld[_NUM_FLD];
2778 nb_tx_thread_params = 0;
2780 while ((p = strchr(p0, '(')) != NULL) {
2782 p0 = strchr(p, ')');
2787 if (size >= sizeof(s))
2790 snprintf(s, sizeof(s), "%.*s", size, p);
2791 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2793 for (i = 0; i < _NUM_FLD; i++) {
2795 int_fld[i] = strtoul(str_fld[i], &end, 0);
2796 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2799 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2800 printf("exceeded max number of tx params: %hu\n",
2801 nb_tx_thread_params);
2804 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2805 (uint8_t)int_fld[FLD_LCORE];
2806 tx_thread_params_array[nb_tx_thread_params].thread_id =
2807 (uint8_t)int_fld[FLD_THREAD];
2808 ++nb_tx_thread_params;
2810 tx_thread_params = tx_thread_params_array;
2815 #if (APP_CPU_LOAD > 0)
2817 parse_stat_lcore(const char *stat_lcore)
2820 unsigned long lcore_id;
2822 lcore_id = strtoul(stat_lcore, &end, 10);
2823 if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2831 parse_eth_dest(const char *optarg)
2835 uint8_t c, *dest, peer_addr[6];
2838 portid = strtoul(optarg, &port_end, 10);
2839 if (errno != 0 || port_end == optarg || *port_end++ != ',')
2840 rte_exit(EXIT_FAILURE,
2841 "Invalid eth-dest: %s", optarg);
2842 if (portid >= RTE_MAX_ETHPORTS)
2843 rte_exit(EXIT_FAILURE,
2844 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2845 portid, RTE_MAX_ETHPORTS);
2847 if (cmdline_parse_etheraddr(NULL, port_end,
2848 &peer_addr, sizeof(peer_addr)) < 0)
2849 rte_exit(EXIT_FAILURE,
2850 "Invalid ethernet address: %s\n",
2852 dest = (uint8_t *)&dest_eth_addr[portid];
2853 for (c = 0; c < 6; c++)
2854 dest[c] = peer_addr[c];
2855 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2858 #define CMD_LINE_OPT_RX_CONFIG "rx"
2859 #define CMD_LINE_OPT_TX_CONFIG "tx"
2860 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2861 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2862 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2863 #define CMD_LINE_OPT_IPV6 "ipv6"
2864 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2865 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2866 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2867 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2869 /* Parse the argument given in the command line of the application */
2871 parse_args(int argc, char **argv)
2876 char *prgname = argv[0];
2877 static struct option lgopts[] = {
2878 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2879 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2880 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2881 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2882 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2883 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2884 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2885 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2886 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2887 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2893 while ((opt = getopt_long(argc, argvopt, "p:P",
2894 lgopts, &option_index)) != EOF) {
2899 enabled_port_mask = parse_portmask(optarg);
2900 if (enabled_port_mask == 0) {
2901 printf("invalid portmask\n");
2902 print_usage(prgname);
2907 printf("Promiscuous mode selected\n");
2913 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2914 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2915 ret = parse_rx_config(optarg);
2917 printf("invalid rx-config\n");
2918 print_usage(prgname);
2923 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2924 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2925 ret = parse_tx_config(optarg);
2927 printf("invalid tx-config\n");
2928 print_usage(prgname);
2933 #if (APP_CPU_LOAD > 0)
2934 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2935 sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2936 cpu_load_lcore_id = parse_stat_lcore(optarg);
2940 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2941 sizeof(CMD_LINE_OPT_ETH_DEST)))
2942 parse_eth_dest(optarg);
2944 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2945 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2946 printf("numa is disabled\n");
2950 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2951 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2952 sizeof(CMD_LINE_OPT_IPV6))) {
2953 printf("ipv6 is specified\n");
2958 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2959 sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2960 printf("l-threads model is disabled\n");
2964 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2965 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2966 printf("software packet type parsing enabled\n");
2970 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2971 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2972 struct option lenopts = {"max-pkt-len", required_argument, 0,
2975 printf("jumbo frame is enabled - disabling simple TX path\n");
2976 port_conf.rxmode.offloads |=
2977 DEV_RX_OFFLOAD_JUMBO_FRAME;
2978 port_conf.txmode.offloads |=
2979 DEV_TX_OFFLOAD_MULTI_SEGS;
2981 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2982 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2985 ret = parse_max_pkt_len(optarg);
2986 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2987 printf("invalid packet length\n");
2988 print_usage(prgname);
2991 port_conf.rxmode.max_rx_pkt_len = ret;
2993 printf("set jumbo frame max packet length to %u\n",
2994 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2996 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2997 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2998 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2999 ret = parse_hash_entry_number(optarg);
3000 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3001 hash_entry_number = ret;
3003 printf("invalid hash entry number\n");
3004 print_usage(prgname);
3012 print_usage(prgname);
3018 argv[optind-1] = prgname;
3021 optind = 1; /* reset getopt lib */
3026 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
3028 char buf[ETHER_ADDR_FMT_SIZE];
3030 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
3031 printf("%s%s", name, buf);
3034 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3036 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3037 union ipv4_5tuple_host *key2)
3039 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3040 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3041 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3042 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3043 key2->proto = key1->proto;
3048 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3049 union ipv6_5tuple_host *key2)
3053 for (i = 0; i < 16; i++) {
3054 key2->ip_dst[i] = key1->ip_dst[i];
3055 key2->ip_src[i] = key1->ip_src[i];
3057 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3058 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3059 key2->proto = key1->proto;
3065 #define BYTE_VALUE_MAX 256
3066 #define ALL_32_BITS 0xffffffff
3067 #define BIT_8_TO_15 0x0000ff00
3069 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3073 uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3075 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3076 for (i = 0; i < array_len; i++) {
3077 struct ipv4_l3fwd_route entry;
3078 union ipv4_5tuple_host newkey;
3080 entry = ipv4_l3fwd_route_array[i];
3081 convert_ipv4_5tuple(&entry.key, &newkey);
3082 ret = rte_hash_add_key(h, (void *)&newkey);
3084 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3085 " to the l3fwd hash.\n", i);
3087 ipv4_l3fwd_out_if[ret] = entry.if_out;
3089 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3092 #define BIT_16_TO_23 0x00ff0000
3094 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3098 uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3100 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3101 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3102 for (i = 0; i < array_len; i++) {
3103 struct ipv6_l3fwd_route entry;
3104 union ipv6_5tuple_host newkey;
3106 entry = ipv6_l3fwd_route_array[i];
3107 convert_ipv6_5tuple(&entry.key, &newkey);
3108 ret = rte_hash_add_key(h, (void *)&newkey);
3110 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3111 " to the l3fwd hash.\n", i);
3113 ipv6_l3fwd_out_if[ret] = entry.if_out;
3115 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3118 #define NUMBER_PORT_USED 4
3120 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3121 unsigned int nr_flow)
3125 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3127 for (i = 0; i < nr_flow; i++) {
3128 struct ipv4_l3fwd_route entry;
3129 union ipv4_5tuple_host newkey;
3130 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3131 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3133 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3135 /* Create the ipv4 exact match flow */
3136 memset(&entry, 0, sizeof(entry));
3137 switch (i & (NUMBER_PORT_USED - 1)) {
3139 entry = ipv4_l3fwd_route_array[0];
3140 entry.key.ip_dst = IPv4(101, c, b, a);
3143 entry = ipv4_l3fwd_route_array[1];
3144 entry.key.ip_dst = IPv4(201, c, b, a);
3147 entry = ipv4_l3fwd_route_array[2];
3148 entry.key.ip_dst = IPv4(111, c, b, a);
3151 entry = ipv4_l3fwd_route_array[3];
3152 entry.key.ip_dst = IPv4(211, c, b, a);
3155 convert_ipv4_5tuple(&entry.key, &newkey);
3156 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3159 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3161 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3164 printf("Hash: Adding 0x%x keys\n", nr_flow);
3168 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3169 unsigned int nr_flow)
3173 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3174 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3175 for (i = 0; i < nr_flow; i++) {
3176 struct ipv6_l3fwd_route entry;
3177 union ipv6_5tuple_host newkey;
3179 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3180 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3182 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3185 /* Create the ipv6 exact match flow */
3186 memset(&entry, 0, sizeof(entry));
3187 switch (i & (NUMBER_PORT_USED - 1)) {
3189 entry = ipv6_l3fwd_route_array[0];
3192 entry = ipv6_l3fwd_route_array[1];
3195 entry = ipv6_l3fwd_route_array[2];
3198 entry = ipv6_l3fwd_route_array[3];
3201 entry.key.ip_dst[13] = c;
3202 entry.key.ip_dst[14] = b;
3203 entry.key.ip_dst[15] = a;
3204 convert_ipv6_5tuple(&entry.key, &newkey);
3205 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3208 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3210 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3213 printf("Hash: Adding 0x%x keys\n", nr_flow);
3217 setup_hash(int socketid)
3219 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3221 .entries = L3FWD_HASH_ENTRIES,
3222 .key_len = sizeof(union ipv4_5tuple_host),
3223 .hash_func = ipv4_hash_crc,
3224 .hash_func_init_val = 0,
3227 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3229 .entries = L3FWD_HASH_ENTRIES,
3230 .key_len = sizeof(union ipv6_5tuple_host),
3231 .hash_func = ipv6_hash_crc,
3232 .hash_func_init_val = 0,
3237 /* create ipv4 hash */
3238 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3239 ipv4_l3fwd_hash_params.name = s;
3240 ipv4_l3fwd_hash_params.socket_id = socketid;
3241 ipv4_l3fwd_lookup_struct[socketid] =
3242 rte_hash_create(&ipv4_l3fwd_hash_params);
3243 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3244 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3245 "socket %d\n", socketid);
3247 /* create ipv6 hash */
3248 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3249 ipv6_l3fwd_hash_params.name = s;
3250 ipv6_l3fwd_hash_params.socket_id = socketid;
3251 ipv6_l3fwd_lookup_struct[socketid] =
3252 rte_hash_create(&ipv6_l3fwd_hash_params);
3253 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3254 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3255 "socket %d\n", socketid);
3257 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3258 /* For testing hash matching with a large number of flows we
3259 * generate millions of IP 5-tuples with an incremented dst
3260 * address to initialize the hash table. */
3262 /* populate the ipv4 hash */
3263 populate_ipv4_many_flow_into_table(
3264 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3266 /* populate the ipv6 hash */
3267 populate_ipv6_many_flow_into_table(
3268 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3271 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3274 /* populate the ipv4 hash */
3275 populate_ipv4_few_flow_into_table(
3276 ipv4_l3fwd_lookup_struct[socketid]);
3278 /* populate the ipv6 hash */
3279 populate_ipv6_few_flow_into_table(
3280 ipv6_l3fwd_lookup_struct[socketid]);
3286 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3288 setup_lpm(int socketid)
3290 struct rte_lpm6_config config;
3291 struct rte_lpm_config lpm_ipv4_config;
3296 /* create the LPM table */
3297 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3298 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3299 lpm_ipv4_config.number_tbl8s = 256;
3300 lpm_ipv4_config.flags = 0;
3301 ipv4_l3fwd_lookup_struct[socketid] =
3302 rte_lpm_create(s, socketid, &lpm_ipv4_config);
3303 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3304 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3305 " on socket %d\n", socketid);
3307 /* populate the LPM table */
3308 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3310 /* skip unused ports */
3311 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3312 enabled_port_mask) == 0)
3315 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3316 ipv4_l3fwd_route_array[i].ip,
3317 ipv4_l3fwd_route_array[i].depth,
3318 ipv4_l3fwd_route_array[i].if_out);
3321 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3322 "l3fwd LPM table on socket %d\n",
3326 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3327 (unsigned)ipv4_l3fwd_route_array[i].ip,
3328 ipv4_l3fwd_route_array[i].depth,
3329 ipv4_l3fwd_route_array[i].if_out);
3332 /* create the LPM6 table */
3333 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3335 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3336 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3338 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3340 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3341 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3342 " on socket %d\n", socketid);
3344 /* populate the LPM table */
3345 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3347 /* skip unused ports */
3348 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3349 enabled_port_mask) == 0)
3352 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3353 ipv6_l3fwd_route_array[i].ip,
3354 ipv6_l3fwd_route_array[i].depth,
3355 ipv6_l3fwd_route_array[i].if_out);
3358 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3359 "l3fwd LPM table on socket %d\n",
3363 printf("LPM: Adding route %s / %d (%d)\n",
3365 ipv6_l3fwd_route_array[i].depth,
3366 ipv6_l3fwd_route_array[i].if_out);
3372 init_mem(unsigned nb_mbuf)
3374 struct lcore_conf *qconf;
3379 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3380 if (rte_lcore_is_enabled(lcore_id) == 0)
3384 socketid = rte_lcore_to_socket_id(lcore_id);
3388 if (socketid >= NB_SOCKETS) {
3389 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3390 socketid, lcore_id, NB_SOCKETS);
3392 if (pktmbuf_pool[socketid] == NULL) {
3393 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3394 pktmbuf_pool[socketid] =
3395 rte_pktmbuf_pool_create(s, nb_mbuf,
3396 MEMPOOL_CACHE_SIZE, 0,
3397 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3398 if (pktmbuf_pool[socketid] == NULL)
3399 rte_exit(EXIT_FAILURE,
3400 "Cannot init mbuf pool on socket %d\n", socketid);
3402 printf("Allocated mbuf pool on socket %d\n", socketid);
3404 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3405 setup_lpm(socketid);
3407 setup_hash(socketid);
3410 qconf = &lcore_conf[lcore_id];
3411 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3412 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3417 /* Check the link status of all ports in up to 9s, and print them finally */
3419 check_all_ports_link_status(uint32_t port_mask)
3421 #define CHECK_INTERVAL 100 /* 100ms */
3422 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3424 uint8_t count, all_ports_up, print_flag = 0;
3425 struct rte_eth_link link;
3427 printf("\nChecking link status");
3429 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3431 RTE_ETH_FOREACH_DEV(portid) {
3432 if ((port_mask & (1 << portid)) == 0)
3434 memset(&link, 0, sizeof(link));
3435 rte_eth_link_get_nowait(portid, &link);
3436 /* print link status if flag set */
3437 if (print_flag == 1) {
3438 if (link.link_status)
3440 "Port%d Link Up. Speed %u Mbps - %s\n",
3441 portid, link.link_speed,
3442 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3443 ("full-duplex") : ("half-duplex\n"));
3445 printf("Port %d Link Down\n", portid);
3448 /* clear all_ports_up flag if any link down */
3449 if (link.link_status == ETH_LINK_DOWN) {
3454 /* after finally printing all link status, get out */
3455 if (print_flag == 1)
3458 if (all_ports_up == 0) {
3461 rte_delay_ms(CHECK_INTERVAL);
3464 /* set the print_flag if all ports up or timeout */
3465 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3473 main(int argc, char **argv)
3475 struct rte_eth_dev_info dev_info;
3476 struct rte_eth_txconf *txconf;
3480 uint16_t queueid, portid;
3482 uint32_t n_tx_queue, nb_lcores;
3483 uint8_t nb_rx_queue, queue, socketid;
3486 ret = rte_eal_init(argc, argv);
3488 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3492 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3493 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3494 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3495 ((uint64_t)portid << 40);
3496 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3499 /* parse application arguments (after the EAL ones) */
3500 ret = parse_args(argc, argv);
3502 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3504 if (check_lcore_params() < 0)
3505 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3507 printf("Initializing rx-queues...\n");
3508 ret = init_rx_queues();
3510 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3512 printf("Initializing tx-threads...\n");
3513 ret = init_tx_threads();
3515 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3517 printf("Initializing rings...\n");
3518 ret = init_rx_rings();
3520 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3522 nb_ports = rte_eth_dev_count_avail();
3524 if (check_port_config() < 0)
3525 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3527 nb_lcores = rte_lcore_count();
3529 /* initialize all ports */
3530 RTE_ETH_FOREACH_DEV(portid) {
3531 struct rte_eth_conf local_port_conf = port_conf;
3533 /* skip ports that are not enabled */
3534 if ((enabled_port_mask & (1 << portid)) == 0) {
3535 printf("\nSkipping disabled port %d\n", portid);
3540 printf("Initializing port %d ... ", portid);
3543 nb_rx_queue = get_port_n_rx_queues(portid);
3544 n_tx_queue = nb_lcores;
3545 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3546 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3547 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3548 nb_rx_queue, (unsigned)n_tx_queue);
3549 rte_eth_dev_info_get(portid, &dev_info);
3550 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3551 local_port_conf.txmode.offloads |=
3552 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3554 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3555 dev_info.flow_type_rss_offloads;
3556 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3557 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3558 printf("Port %u modified RSS hash function based on hardware support,"
3559 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3561 port_conf.rx_adv_conf.rss_conf.rss_hf,
3562 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3565 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3566 (uint16_t)n_tx_queue, &local_port_conf);
3568 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3571 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3574 rte_exit(EXIT_FAILURE,
3575 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3578 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3579 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3581 print_ethaddr("Destination:",
3582 (const struct ether_addr *)&dest_eth_addr[portid]);
3586 * prepare src MACs for each port.
3588 ether_addr_copy(&ports_eth_addr[portid],
3589 (struct ether_addr *)(val_eth + portid) + 1);
3592 ret = init_mem(NB_MBUF);
3594 rte_exit(EXIT_FAILURE, "init_mem failed\n");
3596 /* init one TX queue per couple (lcore,port) */
3598 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3599 if (rte_lcore_is_enabled(lcore_id) == 0)
3603 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3607 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3610 txconf = &dev_info.default_txconf;
3611 txconf->offloads = local_port_conf.txmode.offloads;
3612 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3615 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3616 "port=%d\n", ret, portid);
3618 tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3624 for (i = 0; i < n_rx_thread; i++) {
3625 lcore_id = rx_thread[i].conf.lcore_id;
3627 if (rte_lcore_is_enabled(lcore_id) == 0) {
3628 rte_exit(EXIT_FAILURE,
3629 "Cannot start Rx thread on lcore %u: lcore disabled\n",
3634 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3638 /* init RX queues */
3639 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3640 struct rte_eth_dev *dev;
3641 struct rte_eth_conf *conf;
3642 struct rte_eth_rxconf rxq_conf;
3644 portid = rx_thread[i].rx_queue_list[queue].port_id;
3645 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3646 dev = &rte_eth_devices[portid];
3647 conf = &dev->data->dev_conf;
3650 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3654 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3657 rte_eth_dev_info_get(portid, &dev_info);
3658 rxq_conf = dev_info.default_rxconf;
3659 rxq_conf.offloads = conf->rxmode.offloads;
3660 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3663 pktmbuf_pool[socketid]);
3665 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3666 "port=%d\n", ret, portid);
3673 RTE_ETH_FOREACH_DEV(portid) {
3674 if ((enabled_port_mask & (1 << portid)) == 0)
3678 ret = rte_eth_dev_start(portid);
3680 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3684 * If enabled, put device in promiscuous mode.
3685 * This allows IO forwarding mode to forward packets
3686 * to itself through 2 cross-connected ports of the
3690 rte_eth_promiscuous_enable(portid);
3693 for (i = 0; i < n_rx_thread; i++) {
3694 lcore_id = rx_thread[i].conf.lcore_id;
3695 if (rte_lcore_is_enabled(lcore_id) == 0)
3698 /* check if hw packet type is supported */
3699 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3700 portid = rx_thread[i].rx_queue_list[queue].port_id;
3701 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3703 if (parse_ptype_on) {
3704 if (!rte_eth_add_rx_callback(portid, queueid,
3705 cb_parse_ptype, NULL))
3706 rte_exit(EXIT_FAILURE,
3707 "Failed to add rx callback: "
3708 "port=%d\n", portid);
3709 } else if (!check_ptype(portid))
3710 rte_exit(EXIT_FAILURE,
3711 "Port %d cannot parse packet type.\n\n"
3712 "Please add --parse-ptype to use sw "
3713 "packet type analyzer.\n\n",
3718 check_all_ports_link_status(enabled_port_mask);
3721 printf("Starting L-Threading Model\n");
3723 #if (APP_CPU_LOAD > 0)
3724 if (cpu_load_lcore_id > 0)
3725 /* Use one lcore for cpu load collector */
3729 lthread_num_schedulers_set(nb_lcores);
3730 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3731 lthread_master_spawner(NULL);
3734 printf("Starting P-Threading Model\n");
3735 /* launch per-lcore init on every lcore */
3736 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3737 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3738 if (rte_eal_wait_lcore(lcore_id) < 0)