1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/queue.h>
16 #include <rte_common.h>
18 #include <rte_byteorder.h>
20 #include <rte_memory.h>
21 #include <rte_memcpy.h>
23 #include <rte_launch.h>
24 #include <rte_atomic.h>
25 #include <rte_cycles.h>
26 #include <rte_prefetch.h>
27 #include <rte_lcore.h>
28 #include <rte_per_lcore.h>
29 #include <rte_branch_prediction.h>
30 #include <rte_interrupts.h>
31 #include <rte_random.h>
32 #include <rte_debug.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
36 #include <rte_mempool.h>
41 #include <rte_string_fns.h>
42 #include <rte_pause.h>
44 #include <cmdline_parse.h>
45 #include <cmdline_parse_etheraddr.h>
47 #include <lthread_api.h>
49 #define APP_LOOKUP_EXACT_MATCH 0
50 #define APP_LOOKUP_LPM 1
51 #define DO_RFC_1812_CHECKS
53 /* Enable cpu-load stats 0-off, 1-on */
54 #define APP_CPU_LOAD 1
56 #ifndef APP_LOOKUP_METHOD
57 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
60 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
61 #define sched_getcpu() rte_lcore_id()
65 check_ptype(int portid)
68 int ipv4 = 0, ipv6 = 0;
70 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
77 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
79 for (i = 0; i < ret; ++i) {
80 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
82 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
93 parse_ptype(struct rte_mbuf *m)
95 struct rte_ether_hdr *eth_hdr;
96 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
99 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
100 ether_type = eth_hdr->ether_type;
101 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv4))
102 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
103 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv6))
104 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
106 m->packet_type = packet_type;
110 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
111 struct rte_mbuf *pkts[], uint16_t nb_pkts,
112 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
116 for (i = 0; i < nb_pkts; i++)
117 parse_ptype(pkts[i]);
123 * When set to zero, simple forwaring path is eanbled.
124 * When set to one, optimized forwarding path is enabled.
125 * Note that LPM optimisation path uses SSE4.1 instructions.
127 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1
129 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
130 #include <rte_hash.h>
131 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
133 #include <rte_lpm6.h>
135 #error "APP_LOOKUP_METHOD set to incorrect value"
138 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
140 #define MAX_JUMBO_PKT_LEN 9600
142 #define IPV6_ADDR_LEN 16
144 #define MEMPOOL_CACHE_SIZE 256
147 * This expression is used to calculate the number of mbufs needed depending on
148 * user input, taking into account memory for rx and tx hardware rings, cache
149 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
150 * NB_MBUF never goes below a minimum value of 8192
153 #define NB_MBUF RTE_MAX(\
154 (nb_ports*nb_rx_queue*nb_rxd + \
155 nb_ports*nb_lcores*MAX_PKT_BURST + \
156 nb_ports*n_tx_queue*nb_txd + \
157 nb_lcores*MEMPOOL_CACHE_SIZE), \
160 #define MAX_PKT_BURST 32
161 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
164 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
166 #define MAX_TX_BURST (MAX_PKT_BURST / 2)
167 #define BURST_SIZE MAX_TX_BURST
171 /* Configure how many packets ahead to prefetch, when reading packets */
172 #define PREFETCH_OFFSET 3
174 /* Used to mark destination port as 'invalid'. */
175 #define BAD_PORT ((uint16_t)-1)
180 * Configurable number of RX/TX ring descriptors
182 #define RTE_TEST_RX_DESC_DEFAULT 1024
183 #define RTE_TEST_TX_DESC_DEFAULT 1024
184 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
185 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
187 /* ethernet addresses of ports */
188 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
189 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
191 static xmm_t val_eth[RTE_MAX_ETHPORTS];
193 /* replace first 12B of the ethernet header. */
194 #define MASK_ETH 0x3f
196 /* mask of enabled ports */
197 static uint32_t enabled_port_mask;
198 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
199 static int numa_on = 1; /**< NUMA is enabled by default. */
200 static int parse_ptype_on;
202 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
203 static int ipv6; /**< ipv6 is false by default. */
206 #if (APP_CPU_LOAD == 1)
208 #define MAX_CPU RTE_MAX_LCORE
209 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000) /**< Timeout for collecting 5s */
211 #define CPU_PROCESS 0
213 #define MAX_CPU_COUNTER 2
218 uint64_t hits[MAX_CPU_COUNTER][MAX_CPU];
219 } __rte_cache_aligned;
221 static struct cpu_load cpu_load;
222 static int cpu_load_lcore_id = -1;
224 #define SET_CPU_BUSY(thread, counter) \
225 thread->conf.busy[counter] = 1
227 #define SET_CPU_IDLE(thread, counter) \
228 thread->conf.busy[counter] = 0
230 #define IS_CPU_BUSY(thread, counter) \
231 (thread->conf.busy[counter] > 0)
235 #define SET_CPU_BUSY(thread, counter)
236 #define SET_CPU_IDLE(thread, counter)
237 #define IS_CPU_BUSY(thread, counter) 0
243 struct rte_mbuf *m_table[MAX_PKT_BURST];
246 struct lcore_rx_queue {
249 } __rte_cache_aligned;
251 #define MAX_RX_QUEUE_PER_LCORE 16
252 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
253 #define MAX_RX_QUEUE_PER_PORT 128
255 #define MAX_LCORE_PARAMS 1024
256 struct rx_thread_params {
261 } __rte_cache_aligned;
263 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
264 static struct rx_thread_params rx_thread_params_array_default[] = {
276 static struct rx_thread_params *rx_thread_params =
277 rx_thread_params_array_default;
278 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
280 struct tx_thread_params {
283 } __rte_cache_aligned;
285 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
286 static struct tx_thread_params tx_thread_params_array_default[] = {
298 static struct tx_thread_params *tx_thread_params =
299 tx_thread_params_array_default;
300 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
302 static struct rte_eth_conf port_conf = {
304 .mq_mode = ETH_MQ_RX_RSS,
305 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
307 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
312 .rss_hf = ETH_RSS_TCP,
316 .mq_mode = ETH_MQ_TX_NONE,
320 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
322 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
324 #include <rte_hash_crc.h>
325 #define DEFAULT_HASH_FUNC rte_hash_crc
333 } __attribute__((__packed__));
335 union ipv4_5tuple_host {
348 #define XMM_NUM_IN_IPV6_5TUPLE 3
351 uint8_t ip_dst[IPV6_ADDR_LEN];
352 uint8_t ip_src[IPV6_ADDR_LEN];
356 } __attribute__((__packed__));
358 union ipv6_5tuple_host {
363 uint8_t ip_src[IPV6_ADDR_LEN];
364 uint8_t ip_dst[IPV6_ADDR_LEN];
369 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
372 struct ipv4_l3fwd_route {
373 struct ipv4_5tuple key;
377 struct ipv6_l3fwd_route {
378 struct ipv6_5tuple key;
382 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
383 {{RTE_IPv4(101, 0, 0, 0), RTE_IPv4(100, 10, 0, 1), 101, 11, IPPROTO_TCP}, 0},
384 {{RTE_IPv4(201, 0, 0, 0), RTE_IPv4(200, 20, 0, 1), 102, 12, IPPROTO_TCP}, 1},
385 {{RTE_IPv4(111, 0, 0, 0), RTE_IPv4(100, 30, 0, 1), 101, 11, IPPROTO_TCP}, 2},
386 {{RTE_IPv4(211, 0, 0, 0), RTE_IPv4(200, 40, 0, 1), 102, 12, IPPROTO_TCP}, 3},
389 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
391 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
392 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
394 101, 11, IPPROTO_TCP}, 0},
397 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
398 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
400 102, 12, IPPROTO_TCP}, 1},
403 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
404 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
406 101, 11, IPPROTO_TCP}, 2},
409 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
410 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
412 102, 12, IPPROTO_TCP}, 3},
415 typedef struct rte_hash lookup_struct_t;
416 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
417 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
419 #ifdef RTE_ARCH_X86_64
420 /* default to 4 million hash entries (approx) */
421 #define L3FWD_HASH_ENTRIES (1024*1024*4)
423 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
424 #define L3FWD_HASH_ENTRIES (1024*1024*1)
426 #define HASH_ENTRY_NUMBER_DEFAULT 4
428 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
430 static inline uint32_t
431 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
434 const union ipv4_5tuple_host *k;
440 p = (const uint32_t *)&k->port_src;
442 init_val = rte_hash_crc_4byte(t, init_val);
443 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
444 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
445 init_val = rte_hash_crc_4byte(*p, init_val);
449 static inline uint32_t
450 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
453 const union ipv6_5tuple_host *k;
456 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
457 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
461 p = (const uint32_t *)&k->port_src;
463 ip_src0 = (const uint32_t *) k->ip_src;
464 ip_src1 = (const uint32_t *)(k->ip_src + 4);
465 ip_src2 = (const uint32_t *)(k->ip_src + 8);
466 ip_src3 = (const uint32_t *)(k->ip_src + 12);
467 ip_dst0 = (const uint32_t *) k->ip_dst;
468 ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
469 ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
470 ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
471 init_val = rte_hash_crc_4byte(t, init_val);
472 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
473 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
474 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
475 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
476 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
477 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
478 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
479 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
480 init_val = rte_hash_crc_4byte(*p, init_val);
484 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
485 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
487 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
488 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
492 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
493 struct ipv4_l3fwd_route {
499 struct ipv6_l3fwd_route {
505 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
506 {RTE_IPv4(1, 1, 1, 0), 24, 0},
507 {RTE_IPv4(2, 1, 1, 0), 24, 1},
508 {RTE_IPv4(3, 1, 1, 0), 24, 2},
509 {RTE_IPv4(4, 1, 1, 0), 24, 3},
510 {RTE_IPv4(5, 1, 1, 0), 24, 4},
511 {RTE_IPv4(6, 1, 1, 0), 24, 5},
512 {RTE_IPv4(7, 1, 1, 0), 24, 6},
513 {RTE_IPv4(8, 1, 1, 0), 24, 7},
516 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
517 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
518 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
519 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
520 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
521 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
522 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
523 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
524 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
527 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
528 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
530 #define IPV4_L3FWD_LPM_MAX_RULES 1024
531 #define IPV6_L3FWD_LPM_MAX_RULES 1024
532 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
534 typedef struct rte_lpm lookup_struct_t;
535 typedef struct rte_lpm6 lookup6_struct_t;
536 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
537 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
541 lookup_struct_t *ipv4_lookup_struct;
542 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
543 lookup6_struct_t *ipv6_lookup_struct;
545 lookup_struct_t *ipv6_lookup_struct;
548 } __rte_cache_aligned;
550 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
551 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
553 #define MAX_RX_QUEUE_PER_THREAD 16
554 #define MAX_TX_PORT_PER_THREAD RTE_MAX_ETHPORTS
555 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
556 #define MAX_RX_QUEUE_PER_PORT 128
558 #define MAX_RX_THREAD 1024
559 #define MAX_TX_THREAD 1024
560 #define MAX_THREAD (MAX_RX_THREAD + MAX_TX_THREAD)
563 * Producers and consumers threads configuration
565 static int lthreads_on = 1; /**< Use lthreads for processing*/
567 rte_atomic16_t rx_counter; /**< Number of spawned rx threads */
568 rte_atomic16_t tx_counter; /**< Number of spawned tx threads */
571 uint16_t lcore_id; /**< Initial lcore for rx thread */
572 uint16_t cpu_id; /**< Cpu id for cpu load stats counter */
573 uint16_t thread_id; /**< Thread ID */
575 #if (APP_CPU_LOAD > 0)
576 int busy[MAX_CPU_COUNTER];
580 struct thread_rx_conf {
581 struct thread_conf conf;
584 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
586 uint16_t n_ring; /**< Number of output rings */
587 struct rte_ring *ring[RTE_MAX_LCORE];
588 struct lthread_cond *ready[RTE_MAX_LCORE];
590 #if (APP_CPU_LOAD > 0)
591 int busy[MAX_CPU_COUNTER];
593 } __rte_cache_aligned;
595 uint16_t n_rx_thread;
596 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
598 struct thread_tx_conf {
599 struct thread_conf conf;
601 uint16_t tx_queue_id[RTE_MAX_LCORE];
602 struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
604 struct rte_ring *ring;
605 struct lthread_cond **ready;
607 } __rte_cache_aligned;
609 uint16_t n_tx_thread;
610 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
612 /* Send burst of packets on an output interface */
614 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
616 struct rte_mbuf **m_table;
620 queueid = qconf->tx_queue_id[port];
621 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
623 ret = rte_eth_tx_burst(port, queueid, m_table, n);
624 if (unlikely(ret < n)) {
626 rte_pktmbuf_free(m_table[ret]);
633 /* Enqueue a single packet, and send burst if queue is filled */
635 send_single_packet(struct rte_mbuf *m, uint16_t port)
638 struct thread_tx_conf *qconf;
641 qconf = (struct thread_tx_conf *)lthread_get_data();
643 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
645 len = qconf->tx_mbufs[port].len;
646 qconf->tx_mbufs[port].m_table[len] = m;
649 /* enough pkts to be sent */
650 if (unlikely(len == MAX_PKT_BURST)) {
651 send_burst(qconf, MAX_PKT_BURST, port);
655 qconf->tx_mbufs[port].len = len;
659 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
660 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
661 static __rte_always_inline void
662 send_packetsx4(uint16_t port,
663 struct rte_mbuf *m[], uint32_t num)
666 struct thread_tx_conf *qconf;
669 qconf = (struct thread_tx_conf *)lthread_get_data();
671 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
673 len = qconf->tx_mbufs[port].len;
676 * If TX buffer for that queue is empty, and we have enough packets,
677 * then send them straightway.
679 if (num >= MAX_TX_BURST && len == 0) {
680 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
681 if (unlikely(n < num)) {
683 rte_pktmbuf_free(m[n]);
690 * Put packets into TX buffer for that queue.
694 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
697 switch (n % FWDSTEP) {
700 qconf->tx_mbufs[port].m_table[len + j] = m[j];
704 qconf->tx_mbufs[port].m_table[len + j] = m[j];
708 qconf->tx_mbufs[port].m_table[len + j] = m[j];
712 qconf->tx_mbufs[port].m_table[len + j] = m[j];
719 /* enough pkts to be sent */
720 if (unlikely(len == MAX_PKT_BURST)) {
722 send_burst(qconf, MAX_PKT_BURST, port);
724 /* copy rest of the packets into the TX buffer. */
727 switch (len % FWDSTEP) {
730 qconf->tx_mbufs[port].m_table[j] = m[n + j];
734 qconf->tx_mbufs[port].m_table[j] = m[n + j];
738 qconf->tx_mbufs[port].m_table[j] = m[n + j];
742 qconf->tx_mbufs[port].m_table[j] = m[n + j];
748 qconf->tx_mbufs[port].len = len;
750 #endif /* APP_LOOKUP_LPM */
752 #ifdef DO_RFC_1812_CHECKS
754 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
756 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
758 * 1. The packet length reported by the Link Layer must be large
759 * enough to hold the minimum length legal IP datagram (20 bytes).
761 if (link_len < sizeof(struct rte_ipv4_hdr))
764 /* 2. The IP checksum must be correct. */
765 /* this is checked in H/W */
768 * 3. The IP version number must be 4. If the version number is not 4
769 * then the packet may be another version of IP, such as IPng or
772 if (((pkt->version_ihl) >> 4) != 4)
775 * 4. The IP header length field must be large enough to hold the
776 * minimum length legal IP datagram (20 bytes = 5 words).
778 if ((pkt->version_ihl & 0xf) < 5)
782 * 5. The IP total length field must be large enough to hold the IP
783 * datagram header, whose length is specified in the IP header length
786 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
793 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
795 static __m128i mask0;
796 static __m128i mask1;
797 static __m128i mask2;
798 static inline uint16_t
799 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
800 lookup_struct_t *ipv4_l3fwd_lookup_struct)
803 union ipv4_5tuple_host key;
805 ipv4_hdr = (uint8_t *)ipv4_hdr +
806 offsetof(struct rte_ipv4_hdr, time_to_live);
807 __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
808 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
810 key.xmm = _mm_and_si128(data, mask0);
811 /* Find destination port */
812 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
813 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
816 static inline uint16_t
817 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
818 lookup_struct_t *ipv6_l3fwd_lookup_struct)
821 union ipv6_5tuple_host key;
823 ipv6_hdr = (uint8_t *)ipv6_hdr +
824 offsetof(struct rte_ipv6_hdr, payload_len);
825 __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
826 __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
828 __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
829 sizeof(__m128i) + sizeof(__m128i)));
830 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
831 key.xmm[0] = _mm_and_si128(data0, mask1);
832 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
835 /* Get part of 5 tuple: dst port and src port and dst IP address higher
837 key.xmm[2] = _mm_and_si128(data2, mask2);
839 /* Find destination port */
840 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
841 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
845 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
847 static inline uint16_t
848 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
849 lookup_struct_t *ipv4_l3fwd_lookup_struct)
853 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
854 rte_be_to_cpu_32(((struct rte_ipv4_hdr *)ipv4_hdr)->dst_addr),
855 &next_hop) == 0) ? next_hop : portid);
858 static inline uint16_t
859 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
860 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
864 return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
865 ((struct rte_ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
870 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
871 __attribute__((unused));
873 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
874 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
876 #define MASK_ALL_PKTS 0xff
877 #define EXCLUDE_1ST_PKT 0xfe
878 #define EXCLUDE_2ND_PKT 0xfd
879 #define EXCLUDE_3RD_PKT 0xfb
880 #define EXCLUDE_4TH_PKT 0xf7
881 #define EXCLUDE_5TH_PKT 0xef
882 #define EXCLUDE_6TH_PKT 0xdf
883 #define EXCLUDE_7TH_PKT 0xbf
884 #define EXCLUDE_8TH_PKT 0x7f
887 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
889 struct rte_ether_hdr *eth_hdr[8];
890 struct rte_ipv4_hdr *ipv4_hdr[8];
891 uint16_t dst_port[8];
893 union ipv4_5tuple_host key[8];
896 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
897 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
898 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
899 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
900 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
901 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
902 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
903 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
905 /* Handle IPv4 headers.*/
906 ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv4_hdr *,
907 sizeof(struct rte_ether_hdr));
908 ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv4_hdr *,
909 sizeof(struct rte_ether_hdr));
910 ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv4_hdr *,
911 sizeof(struct rte_ether_hdr));
912 ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv4_hdr *,
913 sizeof(struct rte_ether_hdr));
914 ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv4_hdr *,
915 sizeof(struct rte_ether_hdr));
916 ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv4_hdr *,
917 sizeof(struct rte_ether_hdr));
918 ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv4_hdr *,
919 sizeof(struct rte_ether_hdr));
920 ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv4_hdr *,
921 sizeof(struct rte_ether_hdr));
923 #ifdef DO_RFC_1812_CHECKS
924 /* Check to make sure the packet is valid (RFC1812) */
925 uint8_t valid_mask = MASK_ALL_PKTS;
927 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
928 rte_pktmbuf_free(m[0]);
929 valid_mask &= EXCLUDE_1ST_PKT;
931 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
932 rte_pktmbuf_free(m[1]);
933 valid_mask &= EXCLUDE_2ND_PKT;
935 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
936 rte_pktmbuf_free(m[2]);
937 valid_mask &= EXCLUDE_3RD_PKT;
939 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
940 rte_pktmbuf_free(m[3]);
941 valid_mask &= EXCLUDE_4TH_PKT;
943 if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
944 rte_pktmbuf_free(m[4]);
945 valid_mask &= EXCLUDE_5TH_PKT;
947 if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
948 rte_pktmbuf_free(m[5]);
949 valid_mask &= EXCLUDE_6TH_PKT;
951 if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
952 rte_pktmbuf_free(m[6]);
953 valid_mask &= EXCLUDE_7TH_PKT;
955 if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
956 rte_pktmbuf_free(m[7]);
957 valid_mask &= EXCLUDE_8TH_PKT;
959 if (unlikely(valid_mask != MASK_ALL_PKTS)) {
965 for (i = 0; i < 8; i++)
966 if ((0x1 << i) & valid_mask)
967 l3fwd_simple_forward(m[i], portid);
969 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
971 data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
972 sizeof(struct rte_ether_hdr) +
973 offsetof(struct rte_ipv4_hdr, time_to_live)));
974 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
975 sizeof(struct rte_ether_hdr) +
976 offsetof(struct rte_ipv4_hdr, time_to_live)));
977 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
978 sizeof(struct rte_ether_hdr) +
979 offsetof(struct rte_ipv4_hdr, time_to_live)));
980 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
981 sizeof(struct rte_ether_hdr) +
982 offsetof(struct rte_ipv4_hdr, time_to_live)));
983 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
984 sizeof(struct rte_ether_hdr) +
985 offsetof(struct rte_ipv4_hdr, time_to_live)));
986 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
987 sizeof(struct rte_ether_hdr) +
988 offsetof(struct rte_ipv4_hdr, time_to_live)));
989 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
990 sizeof(struct rte_ether_hdr) +
991 offsetof(struct rte_ipv4_hdr, time_to_live)));
992 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
993 sizeof(struct rte_ether_hdr) +
994 offsetof(struct rte_ipv4_hdr, time_to_live)));
996 key[0].xmm = _mm_and_si128(data[0], mask0);
997 key[1].xmm = _mm_and_si128(data[1], mask0);
998 key[2].xmm = _mm_and_si128(data[2], mask0);
999 key[3].xmm = _mm_and_si128(data[3], mask0);
1000 key[4].xmm = _mm_and_si128(data[4], mask0);
1001 key[5].xmm = _mm_and_si128(data[5], mask0);
1002 key[6].xmm = _mm_and_si128(data[6], mask0);
1003 key[7].xmm = _mm_and_si128(data[7], mask0);
1005 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1006 &key[4], &key[5], &key[6], &key[7]};
1008 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1009 &key_array[0], 8, ret);
1010 dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1011 dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1012 dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1013 dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1014 dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1015 dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1016 dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1017 dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1019 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1020 (enabled_port_mask & 1 << dst_port[0]) == 0)
1021 dst_port[0] = portid;
1022 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1023 (enabled_port_mask & 1 << dst_port[1]) == 0)
1024 dst_port[1] = portid;
1025 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1026 (enabled_port_mask & 1 << dst_port[2]) == 0)
1027 dst_port[2] = portid;
1028 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1029 (enabled_port_mask & 1 << dst_port[3]) == 0)
1030 dst_port[3] = portid;
1031 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1032 (enabled_port_mask & 1 << dst_port[4]) == 0)
1033 dst_port[4] = portid;
1034 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1035 (enabled_port_mask & 1 << dst_port[5]) == 0)
1036 dst_port[5] = portid;
1037 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1038 (enabled_port_mask & 1 << dst_port[6]) == 0)
1039 dst_port[6] = portid;
1040 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1041 (enabled_port_mask & 1 << dst_port[7]) == 0)
1042 dst_port[7] = portid;
1044 #ifdef DO_RFC_1812_CHECKS
1045 /* Update time to live and header checksum */
1046 --(ipv4_hdr[0]->time_to_live);
1047 --(ipv4_hdr[1]->time_to_live);
1048 --(ipv4_hdr[2]->time_to_live);
1049 --(ipv4_hdr[3]->time_to_live);
1050 ++(ipv4_hdr[0]->hdr_checksum);
1051 ++(ipv4_hdr[1]->hdr_checksum);
1052 ++(ipv4_hdr[2]->hdr_checksum);
1053 ++(ipv4_hdr[3]->hdr_checksum);
1054 --(ipv4_hdr[4]->time_to_live);
1055 --(ipv4_hdr[5]->time_to_live);
1056 --(ipv4_hdr[6]->time_to_live);
1057 --(ipv4_hdr[7]->time_to_live);
1058 ++(ipv4_hdr[4]->hdr_checksum);
1059 ++(ipv4_hdr[5]->hdr_checksum);
1060 ++(ipv4_hdr[6]->hdr_checksum);
1061 ++(ipv4_hdr[7]->hdr_checksum);
1065 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1066 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1067 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1068 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1069 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1070 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1071 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1072 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1075 rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1076 rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1077 rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1078 rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1079 rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1080 rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1081 rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1082 rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1084 send_single_packet(m[0], (uint8_t)dst_port[0]);
1085 send_single_packet(m[1], (uint8_t)dst_port[1]);
1086 send_single_packet(m[2], (uint8_t)dst_port[2]);
1087 send_single_packet(m[3], (uint8_t)dst_port[3]);
1088 send_single_packet(m[4], (uint8_t)dst_port[4]);
1089 send_single_packet(m[5], (uint8_t)dst_port[5]);
1090 send_single_packet(m[6], (uint8_t)dst_port[6]);
1091 send_single_packet(m[7], (uint8_t)dst_port[7]);
1095 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1096 __m128i mask1, union ipv6_5tuple_host *key)
1098 __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1099 __m128i *, sizeof(struct rte_ether_hdr) +
1100 offsetof(struct rte_ipv6_hdr, payload_len)));
1101 __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1102 __m128i *, sizeof(struct rte_ether_hdr) +
1103 offsetof(struct rte_ipv6_hdr, payload_len) +
1105 __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1106 __m128i *, sizeof(struct rte_ether_hdr) +
1107 offsetof(struct rte_ipv6_hdr, payload_len) +
1108 sizeof(__m128i) + 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 rte_ether_hdr *eth_hdr[8];
1120 union ipv6_5tuple_host key[8];
1122 __attribute__((unused)) struct rte_ipv6_hdr *ipv6_hdr[8];
1124 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
1125 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
1126 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
1127 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
1128 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
1129 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
1130 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
1131 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
1133 /* Handle IPv6 headers.*/
1134 ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv6_hdr *,
1135 sizeof(struct rte_ether_hdr));
1136 ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv6_hdr *,
1137 sizeof(struct rte_ether_hdr));
1138 ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv6_hdr *,
1139 sizeof(struct rte_ether_hdr));
1140 ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv6_hdr *,
1141 sizeof(struct rte_ether_hdr));
1142 ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv6_hdr *,
1143 sizeof(struct rte_ether_hdr));
1144 ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv6_hdr *,
1145 sizeof(struct rte_ether_hdr));
1146 ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv6_hdr *,
1147 sizeof(struct rte_ether_hdr));
1148 ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv6_hdr *,
1149 sizeof(struct rte_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 rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1211 rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1212 rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1213 rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1214 rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1215 rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1216 rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1217 rte_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 rte_ether_hdr *eth_hdr;
1235 struct rte_ipv4_hdr *ipv4_hdr;
1238 eth_hdr = rte_pktmbuf_mtod(m, struct rte_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 rte_ipv4_hdr *,
1243 sizeof(struct rte_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 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1271 send_single_packet(m, dst_port);
1272 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1273 /* Handle IPv6 headers.*/
1274 struct rte_ipv6_hdr *ipv6_hdr;
1276 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
1277 sizeof(struct rte_ether_hdr));
1279 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1280 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1282 if (dst_port >= RTE_MAX_ETHPORTS ||
1283 (enabled_port_mask & 1 << dst_port) == 0)
1287 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1290 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1293 send_single_packet(m, dst_port);
1295 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1296 rte_pktmbuf_free(m);
1299 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1300 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1301 #ifdef DO_RFC_1812_CHECKS
1303 #define IPV4_MIN_VER_IHL 0x45
1304 #define IPV4_MAX_VER_IHL 0x4f
1305 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1307 /* Minimum value of IPV4 total length (20B) in network byte order. */
1308 #define IPV4_MIN_LEN_BE (sizeof(struct rte_ipv4_hdr) << 8)
1311 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1312 * - The IP version number must be 4.
1313 * - The IP header length field must be large enough to hold the
1314 * minimum length legal IP datagram (20 bytes = 5 words).
1315 * - The IP total length field must be large enough to hold the IP
1316 * datagram header, whose length is specified in the IP header length
1318 * If we encounter invalid IPV4 packet, then set destination port for it
1319 * to BAD_PORT value.
1321 static __rte_always_inline void
1322 rfc1812_process(struct rte_ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1326 if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1327 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1329 ipv4_hdr->time_to_live--;
1330 ipv4_hdr->hdr_checksum++;
1332 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1333 ((uint8_t)ipv4_hdr->total_length == 0 &&
1334 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1341 #define rfc1812_process(mb, dp, ptype) do { } while (0)
1342 #endif /* DO_RFC_1812_CHECKS */
1343 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1346 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1347 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1349 static __rte_always_inline uint16_t
1350 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1353 struct rte_ipv6_hdr *ipv6_hdr;
1354 struct rte_ether_hdr *eth_hdr;
1356 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1357 return (uint16_t) ((rte_lpm_lookup(
1358 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1359 &next_hop) == 0) ? next_hop : portid);
1361 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1363 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1364 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
1366 return (uint16_t) ((rte_lpm6_lookup(
1367 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1368 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1377 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1379 struct rte_ether_hdr *eth_hdr;
1380 struct rte_ipv4_hdr *ipv4_hdr;
1385 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1386 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1388 dst_ipv4 = ipv4_hdr->dst_addr;
1389 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1390 dp = get_dst_port(pkt, dst_ipv4, portid);
1392 te = _mm_load_si128((__m128i *)eth_hdr);
1396 rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1398 te = _mm_blend_epi16(te, ve, MASK_ETH);
1399 _mm_store_si128((__m128i *)eth_hdr, te);
1403 * Read packet_type and destination IPV4 addresses from 4 mbufs.
1406 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1408 uint32_t *ipv4_flag)
1410 struct rte_ipv4_hdr *ipv4_hdr;
1411 struct rte_ether_hdr *eth_hdr;
1412 uint32_t x0, x1, x2, x3;
1414 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct rte_ether_hdr *);
1415 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1416 x0 = ipv4_hdr->dst_addr;
1417 ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1419 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct rte_ether_hdr *);
1420 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1421 x1 = ipv4_hdr->dst_addr;
1422 ipv4_flag[0] &= pkt[1]->packet_type;
1424 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct rte_ether_hdr *);
1425 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1426 x2 = ipv4_hdr->dst_addr;
1427 ipv4_flag[0] &= pkt[2]->packet_type;
1429 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct rte_ether_hdr *);
1430 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1431 x3 = ipv4_hdr->dst_addr;
1432 ipv4_flag[0] &= pkt[3]->packet_type;
1434 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1438 * Lookup into LPM for destination port.
1439 * If lookup fails, use incoming port (portid) as destination port.
1442 processx4_step2(__m128i dip,
1445 struct rte_mbuf *pkt[FWDSTEP],
1446 uint16_t dprt[FWDSTEP])
1449 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1450 4, 5, 6, 7, 0, 1, 2, 3);
1452 /* Byte swap 4 IPV4 addresses. */
1453 dip = _mm_shuffle_epi8(dip, bswap_mask);
1455 /* if all 4 packets are IPV4. */
1456 if (likely(ipv4_flag)) {
1457 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1460 /* get rid of unused upper 16 bit for each dport. */
1461 dst.x = _mm_packs_epi32(dst.x, dst.x);
1462 *(uint64_t *)dprt = dst.u64[0];
1465 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1466 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1467 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1468 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1473 * Update source and destination MAC addresses in the ethernet header.
1474 * Perform RFC1812 checks and updates for IPV4 packets.
1477 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1479 __m128i te[FWDSTEP];
1480 __m128i ve[FWDSTEP];
1481 __m128i *p[FWDSTEP];
1483 p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1484 p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1485 p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1486 p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1488 ve[0] = val_eth[dst_port[0]];
1489 te[0] = _mm_load_si128(p[0]);
1491 ve[1] = val_eth[dst_port[1]];
1492 te[1] = _mm_load_si128(p[1]);
1494 ve[2] = val_eth[dst_port[2]];
1495 te[2] = _mm_load_si128(p[2]);
1497 ve[3] = val_eth[dst_port[3]];
1498 te[3] = _mm_load_si128(p[3]);
1500 /* Update first 12 bytes, keep rest bytes intact. */
1501 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1502 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1503 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1504 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1506 _mm_store_si128(p[0], te[0]);
1507 _mm_store_si128(p[1], te[1]);
1508 _mm_store_si128(p[2], te[2]);
1509 _mm_store_si128(p[3], te[3]);
1511 rfc1812_process((struct rte_ipv4_hdr *)
1512 ((struct rte_ether_hdr *)p[0] + 1),
1513 &dst_port[0], pkt[0]->packet_type);
1514 rfc1812_process((struct rte_ipv4_hdr *)
1515 ((struct rte_ether_hdr *)p[1] + 1),
1516 &dst_port[1], pkt[1]->packet_type);
1517 rfc1812_process((struct rte_ipv4_hdr *)
1518 ((struct rte_ether_hdr *)p[2] + 1),
1519 &dst_port[2], pkt[2]->packet_type);
1520 rfc1812_process((struct rte_ipv4_hdr *)
1521 ((struct rte_ether_hdr *)p[3] + 1),
1522 &dst_port[3], pkt[3]->packet_type);
1526 * We group consecutive packets with the same destionation port into one burst.
1527 * To avoid extra latency this is done together with some other packet
1528 * processing, but after we made a final decision about packet's destination.
1529 * To do this we maintain:
1530 * pnum - array of number of consecutive packets with the same dest port for
1531 * each packet in the input burst.
1532 * lp - pointer to the last updated element in the pnum.
1533 * dlp - dest port value lp corresponds to.
1536 #define GRPSZ (1 << FWDSTEP)
1537 #define GRPMSK (GRPSZ - 1)
1539 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1540 if (likely((dlp) == (dcp)[(idx)])) { \
1543 (dlp) = (dcp)[idx]; \
1544 (lp) = (pn) + (idx); \
1550 * Group consecutive packets with the same destination port in bursts of 4.
1551 * Suppose we have array of destionation ports:
1552 * dst_port[] = {a, b, c, d,, e, ... }
1553 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1554 * We doing 4 comparisons at once and the result is 4 bit mask.
1555 * This mask is used as an index into prebuild array of pnum values.
1557 static inline uint16_t *
1558 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1560 static const struct {
1561 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1562 int32_t idx; /* index for new last updated elemnet. */
1563 uint16_t lpv; /* add value to the last updated element. */
1566 /* 0: a != b, b != c, c != d, d != e */
1567 .pnum = UINT64_C(0x0001000100010001),
1572 /* 1: a == b, b != c, c != d, d != e */
1573 .pnum = UINT64_C(0x0001000100010002),
1578 /* 2: a != b, b == c, c != d, d != e */
1579 .pnum = UINT64_C(0x0001000100020001),
1584 /* 3: a == b, b == c, c != d, d != e */
1585 .pnum = UINT64_C(0x0001000100020003),
1590 /* 4: a != b, b != c, c == d, d != e */
1591 .pnum = UINT64_C(0x0001000200010001),
1596 /* 5: a == b, b != c, c == d, d != e */
1597 .pnum = UINT64_C(0x0001000200010002),
1602 /* 6: a != b, b == c, c == d, d != e */
1603 .pnum = UINT64_C(0x0001000200030001),
1608 /* 7: a == b, b == c, c == d, d != e */
1609 .pnum = UINT64_C(0x0001000200030004),
1614 /* 8: a != b, b != c, c != d, d == e */
1615 .pnum = UINT64_C(0x0002000100010001),
1620 /* 9: a == b, b != c, c != d, d == e */
1621 .pnum = UINT64_C(0x0002000100010002),
1626 /* 0xa: a != b, b == c, c != d, d == e */
1627 .pnum = UINT64_C(0x0002000100020001),
1632 /* 0xb: a == b, b == c, c != d, d == e */
1633 .pnum = UINT64_C(0x0002000100020003),
1638 /* 0xc: a != b, b != c, c == d, d == e */
1639 .pnum = UINT64_C(0x0002000300010001),
1644 /* 0xd: a == b, b != c, c == d, d == e */
1645 .pnum = UINT64_C(0x0002000300010002),
1650 /* 0xe: a != b, b == c, c == d, d == e */
1651 .pnum = UINT64_C(0x0002000300040001),
1656 /* 0xf: a == b, b == c, c == d, d == e */
1657 .pnum = UINT64_C(0x0002000300040005),
1664 uint16_t u16[FWDSTEP + 1];
1666 } *pnum = (void *)pn;
1670 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1671 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1672 v = _mm_movemask_ps((__m128)dp1);
1674 /* update last port counter. */
1675 lp[0] += gptbl[v].lpv;
1677 /* if dest port value has changed. */
1679 pnum->u64 = gptbl[v].pnum;
1680 pnum->u16[FWDSTEP] = 1;
1681 lp = pnum->u16 + gptbl[v].idx;
1687 #endif /* APP_LOOKUP_METHOD */
1690 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1696 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1697 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1701 uint16_t dst_port[MAX_PKT_BURST];
1702 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1703 uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1704 uint16_t pnum[MAX_PKT_BURST + 1];
1708 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1709 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1712 * Send nb_rx - nb_rx%8 packets
1715 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1717 for (j = 0; j < n; j += 8) {
1719 pkts_burst[j]->packet_type &
1720 pkts_burst[j+1]->packet_type &
1721 pkts_burst[j+2]->packet_type &
1722 pkts_burst[j+3]->packet_type &
1723 pkts_burst[j+4]->packet_type &
1724 pkts_burst[j+5]->packet_type &
1725 pkts_burst[j+6]->packet_type &
1726 pkts_burst[j+7]->packet_type;
1727 if (pkt_type & RTE_PTYPE_L3_IPV4) {
1728 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1729 } else if (pkt_type &
1730 RTE_PTYPE_L3_IPV6) {
1731 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1733 l3fwd_simple_forward(pkts_burst[j], portid);
1734 l3fwd_simple_forward(pkts_burst[j+1], portid);
1735 l3fwd_simple_forward(pkts_burst[j+2], portid);
1736 l3fwd_simple_forward(pkts_burst[j+3], portid);
1737 l3fwd_simple_forward(pkts_burst[j+4], portid);
1738 l3fwd_simple_forward(pkts_burst[j+5], portid);
1739 l3fwd_simple_forward(pkts_burst[j+6], portid);
1740 l3fwd_simple_forward(pkts_burst[j+7], portid);
1743 for (; j < nb_rx ; j++)
1744 l3fwd_simple_forward(pkts_burst[j], portid);
1746 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1748 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1749 for (j = 0; j != k; j += FWDSTEP)
1750 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1751 &ipv4_flag[j / FWDSTEP]);
1753 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1754 for (j = 0; j != k; j += FWDSTEP)
1755 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1756 portid, &pkts_burst[j], &dst_port[j]);
1759 * Finish packet processing and group consecutive
1760 * packets with the same destination port.
1762 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1769 processx4_step3(pkts_burst, dst_port);
1771 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1772 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1774 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1775 processx4_step3(&pkts_burst[j], &dst_port[j]);
1779 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1781 dp2 = _mm_loadu_si128(
1782 (__m128i *)&dst_port[j - FWDSTEP + 1]);
1783 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1787 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1789 dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1790 sizeof(dst_port[0]));
1794 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1796 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1797 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1800 * remove values added by the last repeated
1804 dlp = dst_port[j - 1];
1806 /* set dlp and lp to the never used values. */
1808 lp = pnum + MAX_PKT_BURST;
1811 /* Process up to last 3 packets one by one. */
1812 switch (nb_rx % FWDSTEP) {
1814 process_packet(pkts_burst[j], dst_port + j, portid);
1815 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1819 process_packet(pkts_burst[j], dst_port + j, portid);
1820 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1824 process_packet(pkts_burst[j], dst_port + j, portid);
1825 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1830 * Send packets out, through destination port.
1831 * Consecuteve pacekts with the same destination port
1832 * are already grouped together.
1833 * If destination port for the packet equals BAD_PORT,
1834 * then free the packet without sending it out.
1836 for (j = 0; j < nb_rx; j += k) {
1844 if (likely(pn != BAD_PORT))
1845 send_packetsx4(pn, pkts_burst + j, k);
1847 for (m = j; m != j + k; m++)
1848 rte_pktmbuf_free(pkts_burst[m]);
1852 #endif /* APP_LOOKUP_METHOD */
1853 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1855 /* Prefetch first packets */
1856 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1857 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1859 /* Prefetch and forward already prefetched packets */
1860 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1861 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1862 j + PREFETCH_OFFSET], void *));
1863 l3fwd_simple_forward(pkts_burst[j], portid);
1866 /* Forward remaining prefetched packets */
1867 for (; j < nb_rx; j++)
1868 l3fwd_simple_forward(pkts_burst[j], portid);
1870 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1874 #if (APP_CPU_LOAD > 0)
1877 * CPU-load stats collector
1880 cpu_load_collector(__rte_unused void *arg) {
1883 uint64_t prev_tsc, diff_tsc, cur_tsc;
1884 uint64_t total[MAX_CPU] = { 0 };
1885 unsigned min_cpu = MAX_CPU;
1886 unsigned max_cpu = 0;
1891 unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1892 struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1894 struct thread_conf *thread_conf;
1896 const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1897 US_PER_S * CPU_LOAD_TIMEOUT_US;
1901 * Wait for all threads
1904 printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1907 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1910 while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1913 for (i = 0; i < n_rx_thread; i++) {
1915 thread_conf = &rx_thread[i].conf;
1916 cpu_id = thread_conf->cpu_id;
1917 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1919 if (cpu_id > max_cpu)
1921 if (cpu_id < min_cpu)
1924 for (i = 0; i < n_tx_thread; i++) {
1926 thread_conf = &tx_thread[i].conf;
1927 cpu_id = thread_conf->cpu_id;
1928 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1930 if (thread_conf->cpu_id > max_cpu)
1931 max_cpu = thread_conf->cpu_id;
1932 if (thread_conf->cpu_id < min_cpu)
1933 min_cpu = thread_conf->cpu_id;
1939 for (i = min_cpu; i <= max_cpu; i++) {
1940 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1941 for (k = 0; k < n_thread_per_cpu[i]; k++)
1942 if (thread_per_cpu[i][k]->busy[j]) {
1947 cpu_load.hits[j][i]++;
1959 cur_tsc = rte_rdtsc();
1961 diff_tsc = cur_tsc - prev_tsc;
1962 if (unlikely(diff_tsc > interval_tsc)) {
1966 printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1967 n_rx_thread, n_tx_thread);
1969 printf("cpu# proc%% poll%% overhead%%\n\n");
1971 for (i = min_cpu; i <= max_cpu; i++) {
1973 printf("CPU %d:", i);
1974 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1975 printf("%7" PRIu64 "",
1976 cpu_load.hits[j][i] * 100 / cpu_load.counter);
1977 hits += cpu_load.hits[j][i];
1978 cpu_load.hits[j][i] = 0;
1980 printf("%7" PRIu64 "\n",
1981 100 - total[i] * 100 / cpu_load.counter);
1984 cpu_load.counter = 0;
1991 #endif /* APP_CPU_LOAD */
1994 * Null processing lthread loop
1996 * This loop is used to start empty scheduler on lcore.
1999 lthread_null(__rte_unused void *args)
2001 int lcore_id = rte_lcore_id();
2003 RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2008 /* main processing loop */
2010 lthread_tx_per_ring(void *dummy)
2014 struct rte_ring *ring;
2015 struct thread_tx_conf *tx_conf;
2016 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2017 struct lthread_cond *ready;
2019 tx_conf = (struct thread_tx_conf *)dummy;
2020 ring = tx_conf->ring;
2021 ready = *tx_conf->ready;
2023 lthread_set_data((void *)tx_conf);
2026 * Move this lthread to lcore
2028 lthread_set_affinity(tx_conf->conf.lcore_id);
2030 RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2033 rte_atomic16_inc(&tx_counter);
2037 * Read packet from ring
2039 SET_CPU_BUSY(tx_conf, CPU_POLL);
2040 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2041 MAX_PKT_BURST, NULL);
2042 SET_CPU_IDLE(tx_conf, CPU_POLL);
2045 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2046 portid = pkts_burst[0]->port;
2047 process_burst(pkts_burst, nb_rx, portid);
2048 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2051 lthread_cond_wait(ready, 0);
2058 * Main tx-lthreads spawner lthread.
2060 * This lthread is used to spawn one new lthread per ring from producers.
2064 lthread_tx(void *args)
2070 struct thread_tx_conf *tx_conf;
2072 tx_conf = (struct thread_tx_conf *)args;
2073 lthread_set_data((void *)tx_conf);
2076 * Move this lthread to the selected lcore
2078 lthread_set_affinity(tx_conf->conf.lcore_id);
2081 * Spawn tx readers (one per input ring)
2083 lthread_create(<, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2086 lcore_id = rte_lcore_id();
2088 RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2090 tx_conf->conf.cpu_id = sched_getcpu();
2093 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2096 * TX burst queue drain
2098 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2099 if (tx_conf->tx_mbufs[portid].len == 0)
2101 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2102 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2103 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2104 tx_conf->tx_mbufs[portid].len = 0;
2112 lthread_rx(void *dummy)
2120 int len[RTE_MAX_LCORE] = { 0 };
2121 int old_len, new_len;
2122 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2123 struct thread_rx_conf *rx_conf;
2125 rx_conf = (struct thread_rx_conf *)dummy;
2126 lthread_set_data((void *)rx_conf);
2129 * Move this lthread to lcore
2131 lthread_set_affinity(rx_conf->conf.lcore_id);
2133 if (rx_conf->n_rx_queue == 0) {
2134 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2138 RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2140 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2142 portid = rx_conf->rx_queue_list[i].port_id;
2143 queueid = rx_conf->rx_queue_list[i].queue_id;
2144 RTE_LOG(INFO, L3FWD,
2145 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2146 rte_lcore_id(), portid, queueid);
2150 * Init all condition variables (one per rx thread)
2152 for (i = 0; i < rx_conf->n_rx_queue; i++)
2153 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2157 rx_conf->conf.cpu_id = sched_getcpu();
2158 rte_atomic16_inc(&rx_counter);
2162 * Read packet from RX queues
2164 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2165 portid = rx_conf->rx_queue_list[i].port_id;
2166 queueid = rx_conf->rx_queue_list[i].queue_id;
2168 SET_CPU_BUSY(rx_conf, CPU_POLL);
2169 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2171 SET_CPU_IDLE(rx_conf, CPU_POLL);
2174 worker_id = (worker_id + 1) % rx_conf->n_ring;
2175 old_len = len[worker_id];
2177 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2178 ret = rte_ring_sp_enqueue_burst(
2179 rx_conf->ring[worker_id],
2180 (void **) pkts_burst,
2183 new_len = old_len + ret;
2185 if (new_len >= BURST_SIZE) {
2186 lthread_cond_signal(rx_conf->ready[worker_id]);
2190 len[worker_id] = new_len;
2192 if (unlikely(ret < nb_rx)) {
2195 for (k = ret; k < nb_rx; k++) {
2196 struct rte_mbuf *m = pkts_burst[k];
2198 rte_pktmbuf_free(m);
2201 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2211 * Start scheduler with initial lthread on lcore
2213 * This lthread loop spawns all rx and tx lthreads on master lcore
2217 lthread_spawner(__rte_unused void *arg)
2219 struct lthread *lt[MAX_THREAD];
2223 printf("Entering lthread_spawner\n");
2226 * Create producers (rx threads) on default lcore
2228 for (i = 0; i < n_rx_thread; i++) {
2229 rx_thread[i].conf.thread_id = i;
2230 lthread_create(<[n_thread], -1, lthread_rx,
2231 (void *)&rx_thread[i]);
2236 * Wait for all producers. Until some producers can be started on the same
2237 * scheduler as this lthread, yielding is required to let them to run and
2238 * prevent deadlock here.
2240 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2241 lthread_sleep(100000);
2244 * Create consumers (tx threads) on default lcore_id
2246 for (i = 0; i < n_tx_thread; i++) {
2247 tx_thread[i].conf.thread_id = i;
2248 lthread_create(<[n_thread], -1, lthread_tx,
2249 (void *)&tx_thread[i]);
2254 * Wait for all threads finished
2256 for (i = 0; i < n_thread; i++)
2257 lthread_join(lt[i], NULL);
2263 * Start master scheduler with initial lthread spawning rx and tx lthreads
2264 * (main_lthread_master).
2267 lthread_master_spawner(__rte_unused void *arg) {
2269 int lcore_id = rte_lcore_id();
2271 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2272 lthread_create(<, -1, lthread_spawner, NULL);
2279 * Start scheduler on lcore.
2282 sched_spawner(__rte_unused void *arg) {
2284 int lcore_id = rte_lcore_id();
2287 if (lcore_id == cpu_load_lcore_id) {
2288 cpu_load_collector(arg);
2291 #endif /* APP_CPU_LOAD */
2293 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2294 lthread_create(<, -1, lthread_null, NULL);
2300 /* main processing loop */
2302 pthread_tx(void *dummy)
2304 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2305 uint64_t prev_tsc, diff_tsc, cur_tsc;
2308 struct thread_tx_conf *tx_conf;
2310 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2311 US_PER_S * BURST_TX_DRAIN_US;
2315 tx_conf = (struct thread_tx_conf *)dummy;
2317 RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2319 tx_conf->conf.cpu_id = sched_getcpu();
2320 rte_atomic16_inc(&tx_counter);
2323 cur_tsc = rte_rdtsc();
2326 * TX burst queue drain
2328 diff_tsc = cur_tsc - prev_tsc;
2329 if (unlikely(diff_tsc > drain_tsc)) {
2332 * This could be optimized (use queueid instead of
2333 * portid), but it is not called so often
2335 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2336 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2337 if (tx_conf->tx_mbufs[portid].len == 0)
2339 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2340 tx_conf->tx_mbufs[portid].len = 0;
2342 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2348 * Read packet from ring
2350 SET_CPU_BUSY(tx_conf, CPU_POLL);
2351 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2352 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2353 SET_CPU_IDLE(tx_conf, CPU_POLL);
2355 if (unlikely(nb_rx == 0)) {
2360 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2361 portid = pkts_burst[0]->port;
2362 process_burst(pkts_burst, nb_rx, portid);
2363 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2369 pthread_rx(void *dummy)
2378 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2380 struct thread_rx_conf *rx_conf;
2382 lcore_id = rte_lcore_id();
2383 rx_conf = (struct thread_rx_conf *)dummy;
2385 if (rx_conf->n_rx_queue == 0) {
2386 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2390 RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2392 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2394 portid = rx_conf->rx_queue_list[i].port_id;
2395 queueid = rx_conf->rx_queue_list[i].queue_id;
2396 RTE_LOG(INFO, L3FWD,
2397 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2398 lcore_id, portid, queueid);
2402 rx_conf->conf.cpu_id = sched_getcpu();
2403 rte_atomic16_inc(&rx_counter);
2407 * Read packet from RX queues
2409 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2410 portid = rx_conf->rx_queue_list[i].port_id;
2411 queueid = rx_conf->rx_queue_list[i].queue_id;
2413 SET_CPU_BUSY(rx_conf, CPU_POLL);
2414 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2416 SET_CPU_IDLE(rx_conf, CPU_POLL);
2423 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2424 worker_id = (worker_id + 1) % rx_conf->n_ring;
2425 n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2426 (void **)pkts_burst, nb_rx, NULL);
2428 if (unlikely(n != nb_rx)) {
2431 for (k = n; k < nb_rx; k++) {
2432 struct rte_mbuf *m = pkts_burst[k];
2434 rte_pktmbuf_free(m);
2438 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2448 pthread_run(__rte_unused void *arg) {
2449 int lcore_id = rte_lcore_id();
2452 for (i = 0; i < n_rx_thread; i++)
2453 if (rx_thread[i].conf.lcore_id == lcore_id) {
2454 printf("Start rx thread on %d...\n", lcore_id);
2455 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2456 RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2457 pthread_rx((void *)&rx_thread[i]);
2461 for (i = 0; i < n_tx_thread; i++)
2462 if (tx_thread[i].conf.lcore_id == lcore_id) {
2463 printf("Start tx thread on %d...\n", lcore_id);
2464 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2465 RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2466 pthread_tx((void *)&tx_thread[i]);
2471 if (lcore_id == cpu_load_lcore_id)
2472 cpu_load_collector(arg);
2473 #endif /* APP_CPU_LOAD */
2479 check_lcore_params(void)
2481 uint8_t queue, lcore;
2485 for (i = 0; i < nb_rx_thread_params; ++i) {
2486 queue = rx_thread_params[i].queue_id;
2487 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2488 printf("invalid queue number: %hhu\n", queue);
2491 lcore = rx_thread_params[i].lcore_id;
2492 if (!rte_lcore_is_enabled(lcore)) {
2493 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2496 socketid = rte_lcore_to_socket_id(lcore);
2497 if ((socketid != 0) && (numa_on == 0))
2498 printf("warning: lcore %hhu is on socket %d with numa off\n",
2505 check_port_config(void)
2510 for (i = 0; i < nb_rx_thread_params; ++i) {
2511 portid = rx_thread_params[i].port_id;
2512 if ((enabled_port_mask & (1 << portid)) == 0) {
2513 printf("port %u is not enabled in port mask\n", portid);
2516 if (!rte_eth_dev_is_valid_port(portid)) {
2517 printf("port %u is not present on the board\n", portid);
2525 get_port_n_rx_queues(const uint16_t port)
2530 for (i = 0; i < nb_rx_thread_params; ++i)
2531 if (rx_thread_params[i].port_id == port &&
2532 rx_thread_params[i].queue_id > queue)
2533 queue = rx_thread_params[i].queue_id;
2535 return (uint8_t)(++queue);
2542 struct thread_rx_conf *rx_conf;
2543 struct thread_tx_conf *tx_conf;
2544 unsigned rx_thread_id, tx_thread_id;
2546 struct rte_ring *ring = NULL;
2548 for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2550 tx_conf = &tx_thread[tx_thread_id];
2552 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2553 tx_conf->conf.thread_id);
2555 rx_thread_id = tx_conf->conf.thread_id;
2556 if (rx_thread_id > n_tx_thread) {
2557 printf("connection from tx-thread %u to rx-thread %u fails "
2558 "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2562 rx_conf = &rx_thread[rx_thread_id];
2563 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2565 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2566 socket_io, rx_thread_id, tx_thread_id);
2568 ring = rte_ring_create(name, 1024 * 4, socket_io,
2569 RING_F_SP_ENQ | RING_F_SC_DEQ);
2572 rte_panic("Cannot create ring to connect rx-thread %u "
2573 "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2576 rx_conf->ring[rx_conf->n_ring] = ring;
2578 tx_conf->ring = ring;
2579 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2587 init_rx_queues(void)
2589 uint16_t i, nb_rx_queue;
2594 for (i = 0; i < nb_rx_thread_params; ++i) {
2595 thread = rx_thread_params[i].thread_id;
2596 nb_rx_queue = rx_thread[thread].n_rx_queue;
2598 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2599 printf("error: too many queues (%u) for thread: %u\n",
2600 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2604 rx_thread[thread].conf.thread_id = thread;
2605 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2606 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2607 rx_thread_params[i].port_id;
2608 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2609 rx_thread_params[i].queue_id;
2610 rx_thread[thread].n_rx_queue++;
2612 if (thread >= n_rx_thread)
2613 n_rx_thread = thread + 1;
2620 init_tx_threads(void)
2625 for (i = 0; i < nb_tx_thread_params; ++i) {
2626 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2627 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2635 print_usage(const char *prgname)
2637 printf("%s [EAL options] -- -p PORTMASK -P"
2638 " [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2639 " [--tx (lcore,thread)[,(lcore,thread]]"
2640 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2641 " [--parse-ptype]\n\n"
2642 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2643 " -P : enable promiscuous mode\n"
2644 " --rx (port,queue,lcore,thread): rx queues configuration\n"
2645 " --tx (lcore,thread): tx threads configuration\n"
2646 " --stat-lcore LCORE: use lcore for stat collector\n"
2647 " --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2648 " --no-numa: optional, disable numa awareness\n"
2649 " --ipv6: optional, specify it if running ipv6 packets\n"
2650 " --enable-jumbo: enable jumbo frame"
2651 " which max packet len is PKTLEN in decimal (64-9600)\n"
2652 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2653 " --no-lthreads: turn off lthread model\n"
2654 " --parse-ptype: set to use software to analyze packet type\n\n",
2658 static int parse_max_pkt_len(const char *pktlen)
2663 /* parse decimal string */
2664 len = strtoul(pktlen, &end, 10);
2665 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2675 parse_portmask(const char *portmask)
2680 /* parse hexadecimal string */
2681 pm = strtoul(portmask, &end, 16);
2682 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2691 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2693 parse_hash_entry_number(const char *hash_entry_num)
2696 unsigned long hash_en;
2698 /* parse hexadecimal string */
2699 hash_en = strtoul(hash_entry_num, &end, 16);
2700 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2711 parse_rx_config(const char *q_arg)
2714 const char *p, *p0 = q_arg;
2723 unsigned long int_fld[_NUM_FLD];
2724 char *str_fld[_NUM_FLD];
2728 nb_rx_thread_params = 0;
2730 while ((p = strchr(p0, '(')) != NULL) {
2732 p0 = strchr(p, ')');
2737 if (size >= sizeof(s))
2740 snprintf(s, sizeof(s), "%.*s", size, p);
2741 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2743 for (i = 0; i < _NUM_FLD; i++) {
2745 int_fld[i] = strtoul(str_fld[i], &end, 0);
2746 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2749 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2750 printf("exceeded max number of rx params: %hu\n",
2751 nb_rx_thread_params);
2754 rx_thread_params_array[nb_rx_thread_params].port_id =
2756 rx_thread_params_array[nb_rx_thread_params].queue_id =
2757 (uint8_t)int_fld[FLD_QUEUE];
2758 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2759 (uint8_t)int_fld[FLD_LCORE];
2760 rx_thread_params_array[nb_rx_thread_params].thread_id =
2761 (uint8_t)int_fld[FLD_THREAD];
2762 ++nb_rx_thread_params;
2764 rx_thread_params = rx_thread_params_array;
2769 parse_tx_config(const char *q_arg)
2772 const char *p, *p0 = q_arg;
2779 unsigned long int_fld[_NUM_FLD];
2780 char *str_fld[_NUM_FLD];
2784 nb_tx_thread_params = 0;
2786 while ((p = strchr(p0, '(')) != NULL) {
2788 p0 = strchr(p, ')');
2793 if (size >= sizeof(s))
2796 snprintf(s, sizeof(s), "%.*s", size, p);
2797 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2799 for (i = 0; i < _NUM_FLD; i++) {
2801 int_fld[i] = strtoul(str_fld[i], &end, 0);
2802 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2805 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2806 printf("exceeded max number of tx params: %hu\n",
2807 nb_tx_thread_params);
2810 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2811 (uint8_t)int_fld[FLD_LCORE];
2812 tx_thread_params_array[nb_tx_thread_params].thread_id =
2813 (uint8_t)int_fld[FLD_THREAD];
2814 ++nb_tx_thread_params;
2816 tx_thread_params = tx_thread_params_array;
2821 #if (APP_CPU_LOAD > 0)
2823 parse_stat_lcore(const char *stat_lcore)
2826 unsigned long lcore_id;
2828 lcore_id = strtoul(stat_lcore, &end, 10);
2829 if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2837 parse_eth_dest(const char *optarg)
2841 uint8_t c, *dest, peer_addr[6];
2844 portid = strtoul(optarg, &port_end, 10);
2845 if (errno != 0 || port_end == optarg || *port_end++ != ',')
2846 rte_exit(EXIT_FAILURE,
2847 "Invalid eth-dest: %s", optarg);
2848 if (portid >= RTE_MAX_ETHPORTS)
2849 rte_exit(EXIT_FAILURE,
2850 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2851 portid, RTE_MAX_ETHPORTS);
2853 if (cmdline_parse_etheraddr(NULL, port_end,
2854 &peer_addr, sizeof(peer_addr)) < 0)
2855 rte_exit(EXIT_FAILURE,
2856 "Invalid ethernet address: %s\n",
2858 dest = (uint8_t *)&dest_eth_addr[portid];
2859 for (c = 0; c < 6; c++)
2860 dest[c] = peer_addr[c];
2861 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2864 #define CMD_LINE_OPT_RX_CONFIG "rx"
2865 #define CMD_LINE_OPT_TX_CONFIG "tx"
2866 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2867 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2868 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2869 #define CMD_LINE_OPT_IPV6 "ipv6"
2870 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2871 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2872 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2873 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2875 /* Parse the argument given in the command line of the application */
2877 parse_args(int argc, char **argv)
2882 char *prgname = argv[0];
2883 static struct option lgopts[] = {
2884 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2885 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2886 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2887 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2888 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2889 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2890 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2891 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2892 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2893 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2899 while ((opt = getopt_long(argc, argvopt, "p:P",
2900 lgopts, &option_index)) != EOF) {
2905 enabled_port_mask = parse_portmask(optarg);
2906 if (enabled_port_mask == 0) {
2907 printf("invalid portmask\n");
2908 print_usage(prgname);
2913 printf("Promiscuous mode selected\n");
2919 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2920 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2921 ret = parse_rx_config(optarg);
2923 printf("invalid rx-config\n");
2924 print_usage(prgname);
2929 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2930 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2931 ret = parse_tx_config(optarg);
2933 printf("invalid tx-config\n");
2934 print_usage(prgname);
2939 #if (APP_CPU_LOAD > 0)
2940 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2941 sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2942 cpu_load_lcore_id = parse_stat_lcore(optarg);
2946 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2947 sizeof(CMD_LINE_OPT_ETH_DEST)))
2948 parse_eth_dest(optarg);
2950 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2951 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2952 printf("numa is disabled\n");
2956 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2957 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2958 sizeof(CMD_LINE_OPT_IPV6))) {
2959 printf("ipv6 is specified\n");
2964 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2965 sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2966 printf("l-threads model is disabled\n");
2970 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2971 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2972 printf("software packet type parsing enabled\n");
2976 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2977 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2978 struct option lenopts = {"max-pkt-len", required_argument, 0,
2981 printf("jumbo frame is enabled - disabling simple TX path\n");
2982 port_conf.rxmode.offloads |=
2983 DEV_RX_OFFLOAD_JUMBO_FRAME;
2984 port_conf.txmode.offloads |=
2985 DEV_TX_OFFLOAD_MULTI_SEGS;
2987 /* if no max-pkt-len set, use the default value
2990 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2993 ret = parse_max_pkt_len(optarg);
2994 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2995 printf("invalid packet length\n");
2996 print_usage(prgname);
2999 port_conf.rxmode.max_rx_pkt_len = ret;
3001 printf("set jumbo frame max packet length to %u\n",
3002 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
3004 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3005 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
3006 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
3007 ret = parse_hash_entry_number(optarg);
3008 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3009 hash_entry_number = ret;
3011 printf("invalid hash entry number\n");
3012 print_usage(prgname);
3020 print_usage(prgname);
3026 argv[optind-1] = prgname;
3029 optind = 1; /* reset getopt lib */
3034 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
3036 char buf[RTE_ETHER_ADDR_FMT_SIZE];
3038 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
3039 printf("%s%s", name, buf);
3042 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3044 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3045 union ipv4_5tuple_host *key2)
3047 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3048 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3049 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3050 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3051 key2->proto = key1->proto;
3056 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3057 union ipv6_5tuple_host *key2)
3061 for (i = 0; i < 16; i++) {
3062 key2->ip_dst[i] = key1->ip_dst[i];
3063 key2->ip_src[i] = key1->ip_src[i];
3065 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3066 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3067 key2->proto = key1->proto;
3073 #define BYTE_VALUE_MAX 256
3074 #define ALL_32_BITS 0xffffffff
3075 #define BIT_8_TO_15 0x0000ff00
3077 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3081 uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3083 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3084 for (i = 0; i < array_len; i++) {
3085 struct ipv4_l3fwd_route entry;
3086 union ipv4_5tuple_host newkey;
3088 entry = ipv4_l3fwd_route_array[i];
3089 convert_ipv4_5tuple(&entry.key, &newkey);
3090 ret = rte_hash_add_key(h, (void *)&newkey);
3092 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3093 " to the l3fwd hash.\n", i);
3095 ipv4_l3fwd_out_if[ret] = entry.if_out;
3097 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3100 #define BIT_16_TO_23 0x00ff0000
3102 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3106 uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3108 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3109 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3110 for (i = 0; i < array_len; i++) {
3111 struct ipv6_l3fwd_route entry;
3112 union ipv6_5tuple_host newkey;
3114 entry = ipv6_l3fwd_route_array[i];
3115 convert_ipv6_5tuple(&entry.key, &newkey);
3116 ret = rte_hash_add_key(h, (void *)&newkey);
3118 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3119 " to the l3fwd hash.\n", i);
3121 ipv6_l3fwd_out_if[ret] = entry.if_out;
3123 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3126 #define NUMBER_PORT_USED 4
3128 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3129 unsigned int nr_flow)
3133 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3135 for (i = 0; i < nr_flow; i++) {
3136 struct ipv4_l3fwd_route entry;
3137 union ipv4_5tuple_host newkey;
3138 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3139 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3141 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3143 /* Create the ipv4 exact match flow */
3144 memset(&entry, 0, sizeof(entry));
3145 switch (i & (NUMBER_PORT_USED - 1)) {
3147 entry = ipv4_l3fwd_route_array[0];
3148 entry.key.ip_dst = RTE_IPv4(101, c, b, a);
3151 entry = ipv4_l3fwd_route_array[1];
3152 entry.key.ip_dst = RTE_IPv4(201, c, b, a);
3155 entry = ipv4_l3fwd_route_array[2];
3156 entry.key.ip_dst = RTE_IPv4(111, c, b, a);
3159 entry = ipv4_l3fwd_route_array[3];
3160 entry.key.ip_dst = RTE_IPv4(211, c, b, a);
3163 convert_ipv4_5tuple(&entry.key, &newkey);
3164 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3167 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3169 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3172 printf("Hash: Adding 0x%x keys\n", nr_flow);
3176 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3177 unsigned int nr_flow)
3181 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3182 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3183 for (i = 0; i < nr_flow; i++) {
3184 struct ipv6_l3fwd_route entry;
3185 union ipv6_5tuple_host newkey;
3187 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3188 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3190 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3193 /* Create the ipv6 exact match flow */
3194 memset(&entry, 0, sizeof(entry));
3195 switch (i & (NUMBER_PORT_USED - 1)) {
3197 entry = ipv6_l3fwd_route_array[0];
3200 entry = ipv6_l3fwd_route_array[1];
3203 entry = ipv6_l3fwd_route_array[2];
3206 entry = ipv6_l3fwd_route_array[3];
3209 entry.key.ip_dst[13] = c;
3210 entry.key.ip_dst[14] = b;
3211 entry.key.ip_dst[15] = a;
3212 convert_ipv6_5tuple(&entry.key, &newkey);
3213 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3216 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3218 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3221 printf("Hash: Adding 0x%x keys\n", nr_flow);
3225 setup_hash(int socketid)
3227 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3229 .entries = L3FWD_HASH_ENTRIES,
3230 .key_len = sizeof(union ipv4_5tuple_host),
3231 .hash_func = ipv4_hash_crc,
3232 .hash_func_init_val = 0,
3235 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3237 .entries = L3FWD_HASH_ENTRIES,
3238 .key_len = sizeof(union ipv6_5tuple_host),
3239 .hash_func = ipv6_hash_crc,
3240 .hash_func_init_val = 0,
3245 /* create ipv4 hash */
3246 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3247 ipv4_l3fwd_hash_params.name = s;
3248 ipv4_l3fwd_hash_params.socket_id = socketid;
3249 ipv4_l3fwd_lookup_struct[socketid] =
3250 rte_hash_create(&ipv4_l3fwd_hash_params);
3251 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3252 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3253 "socket %d\n", socketid);
3255 /* create ipv6 hash */
3256 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3257 ipv6_l3fwd_hash_params.name = s;
3258 ipv6_l3fwd_hash_params.socket_id = socketid;
3259 ipv6_l3fwd_lookup_struct[socketid] =
3260 rte_hash_create(&ipv6_l3fwd_hash_params);
3261 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3262 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3263 "socket %d\n", socketid);
3265 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3266 /* For testing hash matching with a large number of flows we
3267 * generate millions of IP 5-tuples with an incremented dst
3268 * address to initialize the hash table. */
3270 /* populate the ipv4 hash */
3271 populate_ipv4_many_flow_into_table(
3272 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3274 /* populate the ipv6 hash */
3275 populate_ipv6_many_flow_into_table(
3276 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3279 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3282 /* populate the ipv4 hash */
3283 populate_ipv4_few_flow_into_table(
3284 ipv4_l3fwd_lookup_struct[socketid]);
3286 /* populate the ipv6 hash */
3287 populate_ipv6_few_flow_into_table(
3288 ipv6_l3fwd_lookup_struct[socketid]);
3294 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3296 setup_lpm(int socketid)
3298 struct rte_lpm6_config config;
3299 struct rte_lpm_config lpm_ipv4_config;
3304 /* create the LPM table */
3305 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3306 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3307 lpm_ipv4_config.number_tbl8s = 256;
3308 lpm_ipv4_config.flags = 0;
3309 ipv4_l3fwd_lookup_struct[socketid] =
3310 rte_lpm_create(s, socketid, &lpm_ipv4_config);
3311 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3312 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3313 " on socket %d\n", socketid);
3315 /* populate the LPM table */
3316 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3318 /* skip unused ports */
3319 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3320 enabled_port_mask) == 0)
3323 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3324 ipv4_l3fwd_route_array[i].ip,
3325 ipv4_l3fwd_route_array[i].depth,
3326 ipv4_l3fwd_route_array[i].if_out);
3329 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3330 "l3fwd LPM table on socket %d\n",
3334 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3335 (unsigned)ipv4_l3fwd_route_array[i].ip,
3336 ipv4_l3fwd_route_array[i].depth,
3337 ipv4_l3fwd_route_array[i].if_out);
3340 /* create the LPM6 table */
3341 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3343 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3344 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3346 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3348 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3349 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3350 " on socket %d\n", socketid);
3352 /* populate the LPM table */
3353 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3355 /* skip unused ports */
3356 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3357 enabled_port_mask) == 0)
3360 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3361 ipv6_l3fwd_route_array[i].ip,
3362 ipv6_l3fwd_route_array[i].depth,
3363 ipv6_l3fwd_route_array[i].if_out);
3366 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3367 "l3fwd LPM table on socket %d\n",
3371 printf("LPM: Adding route %s / %d (%d)\n",
3373 ipv6_l3fwd_route_array[i].depth,
3374 ipv6_l3fwd_route_array[i].if_out);
3380 init_mem(unsigned nb_mbuf)
3382 struct lcore_conf *qconf;
3387 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3388 if (rte_lcore_is_enabled(lcore_id) == 0)
3392 socketid = rte_lcore_to_socket_id(lcore_id);
3396 if (socketid >= NB_SOCKETS) {
3397 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3398 socketid, lcore_id, NB_SOCKETS);
3400 if (pktmbuf_pool[socketid] == NULL) {
3401 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3402 pktmbuf_pool[socketid] =
3403 rte_pktmbuf_pool_create(s, nb_mbuf,
3404 MEMPOOL_CACHE_SIZE, 0,
3405 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3406 if (pktmbuf_pool[socketid] == NULL)
3407 rte_exit(EXIT_FAILURE,
3408 "Cannot init mbuf pool on socket %d\n", socketid);
3410 printf("Allocated mbuf pool on socket %d\n", socketid);
3412 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3413 setup_lpm(socketid);
3415 setup_hash(socketid);
3418 qconf = &lcore_conf[lcore_id];
3419 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3420 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3425 /* Check the link status of all ports in up to 9s, and print them finally */
3427 check_all_ports_link_status(uint32_t port_mask)
3429 #define CHECK_INTERVAL 100 /* 100ms */
3430 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3432 uint8_t count, all_ports_up, print_flag = 0;
3433 struct rte_eth_link link;
3435 printf("\nChecking link status");
3437 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3439 RTE_ETH_FOREACH_DEV(portid) {
3440 if ((port_mask & (1 << portid)) == 0)
3442 memset(&link, 0, sizeof(link));
3443 rte_eth_link_get_nowait(portid, &link);
3444 /* print link status if flag set */
3445 if (print_flag == 1) {
3446 if (link.link_status)
3448 "Port%d Link Up. Speed %u Mbps - %s\n",
3449 portid, link.link_speed,
3450 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3451 ("full-duplex") : ("half-duplex\n"));
3453 printf("Port %d Link Down\n", portid);
3456 /* clear all_ports_up flag if any link down */
3457 if (link.link_status == ETH_LINK_DOWN) {
3462 /* after finally printing all link status, get out */
3463 if (print_flag == 1)
3466 if (all_ports_up == 0) {
3469 rte_delay_ms(CHECK_INTERVAL);
3472 /* set the print_flag if all ports up or timeout */
3473 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3481 main(int argc, char **argv)
3483 struct rte_eth_dev_info dev_info;
3484 struct rte_eth_txconf *txconf;
3488 uint16_t queueid, portid;
3490 uint32_t n_tx_queue, nb_lcores;
3491 uint8_t nb_rx_queue, queue, socketid;
3494 ret = rte_eal_init(argc, argv);
3496 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3500 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3501 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3502 dest_eth_addr[portid] = RTE_ETHER_LOCAL_ADMIN_ADDR +
3503 ((uint64_t)portid << 40);
3504 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3507 /* parse application arguments (after the EAL ones) */
3508 ret = parse_args(argc, argv);
3510 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3512 if (check_lcore_params() < 0)
3513 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3515 printf("Initializing rx-queues...\n");
3516 ret = init_rx_queues();
3518 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3520 printf("Initializing tx-threads...\n");
3521 ret = init_tx_threads();
3523 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3525 printf("Initializing rings...\n");
3526 ret = init_rx_rings();
3528 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3530 nb_ports = rte_eth_dev_count_avail();
3532 if (check_port_config() < 0)
3533 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3535 nb_lcores = rte_lcore_count();
3537 /* initialize all ports */
3538 RTE_ETH_FOREACH_DEV(portid) {
3539 struct rte_eth_conf local_port_conf = port_conf;
3541 /* skip ports that are not enabled */
3542 if ((enabled_port_mask & (1 << portid)) == 0) {
3543 printf("\nSkipping disabled port %d\n", portid);
3548 printf("Initializing port %d ... ", portid);
3551 nb_rx_queue = get_port_n_rx_queues(portid);
3552 n_tx_queue = nb_lcores;
3553 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3554 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3555 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3556 nb_rx_queue, (unsigned)n_tx_queue);
3557 rte_eth_dev_info_get(portid, &dev_info);
3558 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3559 local_port_conf.txmode.offloads |=
3560 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3562 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3563 dev_info.flow_type_rss_offloads;
3564 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3565 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3566 printf("Port %u modified RSS hash function based on hardware support,"
3567 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3569 port_conf.rx_adv_conf.rss_conf.rss_hf,
3570 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3573 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3574 (uint16_t)n_tx_queue, &local_port_conf);
3576 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3579 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3582 rte_exit(EXIT_FAILURE,
3583 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3586 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3587 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3589 print_ethaddr("Destination:",
3590 (const struct rte_ether_addr *)&dest_eth_addr[portid]);
3594 * prepare src MACs for each port.
3596 rte_ether_addr_copy(&ports_eth_addr[portid],
3597 (struct rte_ether_addr *)(val_eth + portid) + 1);
3600 ret = init_mem(NB_MBUF);
3602 rte_exit(EXIT_FAILURE, "init_mem failed\n");
3604 /* init one TX queue per couple (lcore,port) */
3606 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3607 if (rte_lcore_is_enabled(lcore_id) == 0)
3611 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3615 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3618 txconf = &dev_info.default_txconf;
3619 txconf->offloads = local_port_conf.txmode.offloads;
3620 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3623 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3624 "port=%d\n", ret, portid);
3626 tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3632 for (i = 0; i < n_rx_thread; i++) {
3633 lcore_id = rx_thread[i].conf.lcore_id;
3635 if (rte_lcore_is_enabled(lcore_id) == 0) {
3636 rte_exit(EXIT_FAILURE,
3637 "Cannot start Rx thread on lcore %u: lcore disabled\n",
3642 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3646 /* init RX queues */
3647 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3648 struct rte_eth_dev *dev;
3649 struct rte_eth_conf *conf;
3650 struct rte_eth_rxconf rxq_conf;
3652 portid = rx_thread[i].rx_queue_list[queue].port_id;
3653 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3654 dev = &rte_eth_devices[portid];
3655 conf = &dev->data->dev_conf;
3658 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3662 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3665 rte_eth_dev_info_get(portid, &dev_info);
3666 rxq_conf = dev_info.default_rxconf;
3667 rxq_conf.offloads = conf->rxmode.offloads;
3668 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3671 pktmbuf_pool[socketid]);
3673 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3674 "port=%d\n", ret, portid);
3681 RTE_ETH_FOREACH_DEV(portid) {
3682 if ((enabled_port_mask & (1 << portid)) == 0)
3686 ret = rte_eth_dev_start(portid);
3688 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3692 * If enabled, put device in promiscuous mode.
3693 * This allows IO forwarding mode to forward packets
3694 * to itself through 2 cross-connected ports of the
3698 rte_eth_promiscuous_enable(portid);
3701 for (i = 0; i < n_rx_thread; i++) {
3702 lcore_id = rx_thread[i].conf.lcore_id;
3703 if (rte_lcore_is_enabled(lcore_id) == 0)
3706 /* check if hw packet type is supported */
3707 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3708 portid = rx_thread[i].rx_queue_list[queue].port_id;
3709 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3711 if (parse_ptype_on) {
3712 if (!rte_eth_add_rx_callback(portid, queueid,
3713 cb_parse_ptype, NULL))
3714 rte_exit(EXIT_FAILURE,
3715 "Failed to add rx callback: "
3716 "port=%d\n", portid);
3717 } else if (!check_ptype(portid))
3718 rte_exit(EXIT_FAILURE,
3719 "Port %d cannot parse packet type.\n\n"
3720 "Please add --parse-ptype to use sw "
3721 "packet type analyzer.\n\n",
3726 check_all_ports_link_status(enabled_port_mask);
3729 printf("Starting L-Threading Model\n");
3731 #if (APP_CPU_LOAD > 0)
3732 if (cpu_load_lcore_id > 0)
3733 /* Use one lcore for cpu load collector */
3737 lthread_num_schedulers_set(nb_lcores);
3738 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3739 lthread_master_spawner(NULL);
3742 printf("Starting P-Threading Model\n");
3743 /* launch per-lcore init on every lcore */
3744 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3745 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3746 if (rte_eal_wait_lcore(lcore_id) < 0)