1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/types.h>
13 #include <sys/queue.h>
18 #include <rte_common.h>
20 #include <rte_byteorder.h>
22 #include <rte_memory.h>
23 #include <rte_memcpy.h>
25 #include <rte_launch.h>
26 #include <rte_atomic.h>
27 #include <rte_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
43 #include <rte_string_fns.h>
44 #include <rte_pause.h>
46 #include <cmdline_parse.h>
47 #include <cmdline_parse_etheraddr.h>
49 #include <lthread_api.h>
51 #define APP_LOOKUP_EXACT_MATCH 0
52 #define APP_LOOKUP_LPM 1
53 #define DO_RFC_1812_CHECKS
55 /* Enable cpu-load stats 0-off, 1-on */
56 #define APP_CPU_LOAD 1
58 #ifndef APP_LOOKUP_METHOD
59 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
62 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
63 #define sched_getcpu() rte_lcore_id()
67 check_ptype(int portid)
70 int ipv4 = 0, ipv6 = 0;
72 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
79 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
81 for (i = 0; i < ret; ++i) {
82 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
84 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
95 parse_ptype(struct rte_mbuf *m)
97 struct ether_hdr *eth_hdr;
98 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
101 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
102 ether_type = eth_hdr->ether_type;
103 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
104 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
105 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
106 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
108 m->packet_type = packet_type;
112 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
113 struct rte_mbuf *pkts[], uint16_t nb_pkts,
114 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
118 for (i = 0; i < nb_pkts; i++)
119 parse_ptype(pkts[i]);
125 * When set to zero, simple forwaring path is eanbled.
126 * When set to one, optimized forwarding path is enabled.
127 * Note that LPM optimisation path uses SSE4.1 instructions.
129 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1
131 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
132 #include <rte_hash.h>
133 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
135 #include <rte_lpm6.h>
137 #error "APP_LOOKUP_METHOD set to incorrect value"
140 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
142 #define MAX_JUMBO_PKT_LEN 9600
144 #define IPV6_ADDR_LEN 16
146 #define MEMPOOL_CACHE_SIZE 256
149 * This expression is used to calculate the number of mbufs needed depending on
150 * user input, taking into account memory for rx and tx hardware rings, cache
151 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
152 * NB_MBUF never goes below a minimum value of 8192
155 #define NB_MBUF RTE_MAX(\
156 (nb_ports*nb_rx_queue*nb_rxd + \
157 nb_ports*nb_lcores*MAX_PKT_BURST + \
158 nb_ports*n_tx_queue*nb_txd + \
159 nb_lcores*MEMPOOL_CACHE_SIZE), \
162 #define MAX_PKT_BURST 32
163 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
166 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
168 #define MAX_TX_BURST (MAX_PKT_BURST / 2)
169 #define BURST_SIZE MAX_TX_BURST
173 /* Configure how many packets ahead to prefetch, when reading packets */
174 #define PREFETCH_OFFSET 3
176 /* Used to mark destination port as 'invalid'. */
177 #define BAD_PORT ((uint16_t)-1)
182 * Configurable number of RX/TX ring descriptors
184 #define RTE_TEST_RX_DESC_DEFAULT 128
185 #define RTE_TEST_TX_DESC_DEFAULT 128
186 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
187 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
189 /* ethernet addresses of ports */
190 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
191 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
193 static xmm_t val_eth[RTE_MAX_ETHPORTS];
195 /* replace first 12B of the ethernet header. */
196 #define MASK_ETH 0x3f
198 /* mask of enabled ports */
199 static uint32_t enabled_port_mask;
200 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
201 static int numa_on = 1; /**< NUMA is enabled by default. */
202 static int parse_ptype_on;
204 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
205 static int ipv6; /**< ipv6 is false by default. */
208 #if (APP_CPU_LOAD == 1)
210 #define MAX_CPU RTE_MAX_LCORE
211 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000) /**< Timeout for collecting 5s */
213 #define CPU_PROCESS 0
215 #define MAX_CPU_COUNTER 2
220 uint64_t hits[MAX_CPU_COUNTER][MAX_CPU];
221 } __rte_cache_aligned;
223 static struct cpu_load cpu_load;
224 static int cpu_load_lcore_id = -1;
226 #define SET_CPU_BUSY(thread, counter) \
227 thread->conf.busy[counter] = 1
229 #define SET_CPU_IDLE(thread, counter) \
230 thread->conf.busy[counter] = 0
232 #define IS_CPU_BUSY(thread, counter) \
233 (thread->conf.busy[counter] > 0)
237 #define SET_CPU_BUSY(thread, counter)
238 #define SET_CPU_IDLE(thread, counter)
239 #define IS_CPU_BUSY(thread, counter) 0
245 struct rte_mbuf *m_table[MAX_PKT_BURST];
248 struct lcore_rx_queue {
251 } __rte_cache_aligned;
253 #define MAX_RX_QUEUE_PER_LCORE 16
254 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
255 #define MAX_RX_QUEUE_PER_PORT 128
257 #define MAX_LCORE_PARAMS 1024
258 struct rx_thread_params {
263 } __rte_cache_aligned;
265 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
266 static struct rx_thread_params rx_thread_params_array_default[] = {
278 static struct rx_thread_params *rx_thread_params =
279 rx_thread_params_array_default;
280 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
282 struct tx_thread_params {
285 } __rte_cache_aligned;
287 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
288 static struct tx_thread_params tx_thread_params_array_default[] = {
300 static struct tx_thread_params *tx_thread_params =
301 tx_thread_params_array_default;
302 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
304 static struct rte_eth_conf port_conf = {
306 .mq_mode = ETH_MQ_RX_RSS,
307 .max_rx_pkt_len = ETHER_MAX_LEN,
309 .header_split = 0, /**< Header Split disabled */
310 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
311 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
312 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
313 .hw_strip_crc = 1, /**< CRC stripped by hardware */
318 .rss_hf = ETH_RSS_TCP,
322 .mq_mode = ETH_MQ_TX_NONE,
326 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
328 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
330 #include <rte_hash_crc.h>
331 #define DEFAULT_HASH_FUNC rte_hash_crc
339 } __attribute__((__packed__));
341 union ipv4_5tuple_host {
354 #define XMM_NUM_IN_IPV6_5TUPLE 3
357 uint8_t ip_dst[IPV6_ADDR_LEN];
358 uint8_t ip_src[IPV6_ADDR_LEN];
362 } __attribute__((__packed__));
364 union ipv6_5tuple_host {
369 uint8_t ip_src[IPV6_ADDR_LEN];
370 uint8_t ip_dst[IPV6_ADDR_LEN];
375 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
378 struct ipv4_l3fwd_route {
379 struct ipv4_5tuple key;
383 struct ipv6_l3fwd_route {
384 struct ipv6_5tuple key;
388 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
389 {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1), 101, 11, IPPROTO_TCP}, 0},
390 {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1), 102, 12, IPPROTO_TCP}, 1},
391 {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1), 101, 11, IPPROTO_TCP}, 2},
392 {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1), 102, 12, IPPROTO_TCP}, 3},
395 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
397 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
398 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
400 101, 11, IPPROTO_TCP}, 0},
403 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
404 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
406 102, 12, IPPROTO_TCP}, 1},
409 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
410 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
412 101, 11, IPPROTO_TCP}, 2},
415 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
416 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
418 102, 12, IPPROTO_TCP}, 3},
421 typedef struct rte_hash lookup_struct_t;
422 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
423 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
425 #ifdef RTE_ARCH_X86_64
426 /* default to 4 million hash entries (approx) */
427 #define L3FWD_HASH_ENTRIES (1024*1024*4)
429 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
430 #define L3FWD_HASH_ENTRIES (1024*1024*1)
432 #define HASH_ENTRY_NUMBER_DEFAULT 4
434 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
436 static inline uint32_t
437 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
440 const union ipv4_5tuple_host *k;
446 p = (const uint32_t *)&k->port_src;
448 init_val = rte_hash_crc_4byte(t, init_val);
449 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
450 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
451 init_val = rte_hash_crc_4byte(*p, init_val);
455 static inline uint32_t
456 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
459 const union ipv6_5tuple_host *k;
462 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
463 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
467 p = (const uint32_t *)&k->port_src;
469 ip_src0 = (const uint32_t *) k->ip_src;
470 ip_src1 = (const uint32_t *)(k->ip_src + 4);
471 ip_src2 = (const uint32_t *)(k->ip_src + 8);
472 ip_src3 = (const uint32_t *)(k->ip_src + 12);
473 ip_dst0 = (const uint32_t *) k->ip_dst;
474 ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
475 ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
476 ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
477 init_val = rte_hash_crc_4byte(t, init_val);
478 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
479 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
480 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
481 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
482 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
483 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
484 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
485 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
486 init_val = rte_hash_crc_4byte(*p, init_val);
490 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
491 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
493 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
494 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
499 struct ipv4_l3fwd_route {
505 struct ipv6_l3fwd_route {
511 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
512 {IPv4(1, 1, 1, 0), 24, 0},
513 {IPv4(2, 1, 1, 0), 24, 1},
514 {IPv4(3, 1, 1, 0), 24, 2},
515 {IPv4(4, 1, 1, 0), 24, 3},
516 {IPv4(5, 1, 1, 0), 24, 4},
517 {IPv4(6, 1, 1, 0), 24, 5},
518 {IPv4(7, 1, 1, 0), 24, 6},
519 {IPv4(8, 1, 1, 0), 24, 7},
522 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
523 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
524 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
525 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
526 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
527 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
528 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
529 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
530 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
533 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
534 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
536 #define IPV4_L3FWD_LPM_MAX_RULES 1024
537 #define IPV6_L3FWD_LPM_MAX_RULES 1024
538 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
540 typedef struct rte_lpm lookup_struct_t;
541 typedef struct rte_lpm6 lookup6_struct_t;
542 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
543 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
547 lookup_struct_t *ipv4_lookup_struct;
548 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
549 lookup6_struct_t *ipv6_lookup_struct;
551 lookup_struct_t *ipv6_lookup_struct;
554 } __rte_cache_aligned;
556 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
557 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
559 #define MAX_RX_QUEUE_PER_THREAD 16
560 #define MAX_TX_PORT_PER_THREAD RTE_MAX_ETHPORTS
561 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
562 #define MAX_RX_QUEUE_PER_PORT 128
564 #define MAX_RX_THREAD 1024
565 #define MAX_TX_THREAD 1024
566 #define MAX_THREAD (MAX_RX_THREAD + MAX_TX_THREAD)
569 * Producers and consumers threads configuration
571 static int lthreads_on = 1; /**< Use lthreads for processing*/
573 rte_atomic16_t rx_counter; /**< Number of spawned rx threads */
574 rte_atomic16_t tx_counter; /**< Number of spawned tx threads */
577 uint16_t lcore_id; /**< Initial lcore for rx thread */
578 uint16_t cpu_id; /**< Cpu id for cpu load stats counter */
579 uint16_t thread_id; /**< Thread ID */
581 #if (APP_CPU_LOAD > 0)
582 int busy[MAX_CPU_COUNTER];
586 struct thread_rx_conf {
587 struct thread_conf conf;
590 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
592 uint16_t n_ring; /**< Number of output rings */
593 struct rte_ring *ring[RTE_MAX_LCORE];
594 struct lthread_cond *ready[RTE_MAX_LCORE];
596 #if (APP_CPU_LOAD > 0)
597 int busy[MAX_CPU_COUNTER];
599 } __rte_cache_aligned;
601 uint16_t n_rx_thread;
602 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
604 struct thread_tx_conf {
605 struct thread_conf conf;
607 uint16_t tx_queue_id[RTE_MAX_LCORE];
608 struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
610 struct rte_ring *ring;
611 struct lthread_cond **ready;
613 } __rte_cache_aligned;
615 uint16_t n_tx_thread;
616 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
618 /* Send burst of packets on an output interface */
620 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
622 struct rte_mbuf **m_table;
626 queueid = qconf->tx_queue_id[port];
627 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
629 ret = rte_eth_tx_burst(port, queueid, m_table, n);
630 if (unlikely(ret < n)) {
632 rte_pktmbuf_free(m_table[ret]);
639 /* Enqueue a single packet, and send burst if queue is filled */
641 send_single_packet(struct rte_mbuf *m, uint16_t port)
644 struct thread_tx_conf *qconf;
647 qconf = (struct thread_tx_conf *)lthread_get_data();
649 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
651 len = qconf->tx_mbufs[port].len;
652 qconf->tx_mbufs[port].m_table[len] = m;
655 /* enough pkts to be sent */
656 if (unlikely(len == MAX_PKT_BURST)) {
657 send_burst(qconf, MAX_PKT_BURST, port);
661 qconf->tx_mbufs[port].len = len;
665 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
666 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
667 static __rte_always_inline void
668 send_packetsx4(uint16_t port,
669 struct rte_mbuf *m[], uint32_t num)
672 struct thread_tx_conf *qconf;
675 qconf = (struct thread_tx_conf *)lthread_get_data();
677 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
679 len = qconf->tx_mbufs[port].len;
682 * If TX buffer for that queue is empty, and we have enough packets,
683 * then send them straightway.
685 if (num >= MAX_TX_BURST && len == 0) {
686 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
687 if (unlikely(n < num)) {
689 rte_pktmbuf_free(m[n]);
696 * Put packets into TX buffer for that queue.
700 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
703 switch (n % FWDSTEP) {
706 qconf->tx_mbufs[port].m_table[len + j] = m[j];
710 qconf->tx_mbufs[port].m_table[len + j] = m[j];
714 qconf->tx_mbufs[port].m_table[len + j] = m[j];
718 qconf->tx_mbufs[port].m_table[len + j] = m[j];
725 /* enough pkts to be sent */
726 if (unlikely(len == MAX_PKT_BURST)) {
728 send_burst(qconf, MAX_PKT_BURST, port);
730 /* copy rest of the packets into the TX buffer. */
733 switch (len % FWDSTEP) {
736 qconf->tx_mbufs[port].m_table[j] = m[n + j];
740 qconf->tx_mbufs[port].m_table[j] = m[n + j];
744 qconf->tx_mbufs[port].m_table[j] = m[n + j];
748 qconf->tx_mbufs[port].m_table[j] = m[n + j];
754 qconf->tx_mbufs[port].len = len;
756 #endif /* APP_LOOKUP_LPM */
758 #ifdef DO_RFC_1812_CHECKS
760 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
762 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
764 * 1. The packet length reported by the Link Layer must be large
765 * enough to hold the minimum length legal IP datagram (20 bytes).
767 if (link_len < sizeof(struct ipv4_hdr))
770 /* 2. The IP checksum must be correct. */
771 /* this is checked in H/W */
774 * 3. The IP version number must be 4. If the version number is not 4
775 * then the packet may be another version of IP, such as IPng or
778 if (((pkt->version_ihl) >> 4) != 4)
781 * 4. The IP header length field must be large enough to hold the
782 * minimum length legal IP datagram (20 bytes = 5 words).
784 if ((pkt->version_ihl & 0xf) < 5)
788 * 5. The IP total length field must be large enough to hold the IP
789 * datagram header, whose length is specified in the IP header length
792 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
799 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
801 static __m128i mask0;
802 static __m128i mask1;
803 static __m128i mask2;
804 static inline uint16_t
805 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
806 lookup_struct_t *ipv4_l3fwd_lookup_struct)
809 union ipv4_5tuple_host key;
811 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
812 __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
813 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
815 key.xmm = _mm_and_si128(data, mask0);
816 /* Find destination port */
817 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
818 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
821 static inline uint16_t
822 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
823 lookup_struct_t *ipv6_l3fwd_lookup_struct)
826 union ipv6_5tuple_host key;
828 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
829 __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
830 __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
832 __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
833 sizeof(__m128i) + sizeof(__m128i)));
834 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
835 key.xmm[0] = _mm_and_si128(data0, mask1);
836 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
839 /* Get part of 5 tuple: dst port and src port and dst IP address higher
841 key.xmm[2] = _mm_and_si128(data2, mask2);
843 /* Find destination port */
844 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
845 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
849 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
851 static inline uint16_t
852 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
853 lookup_struct_t *ipv4_l3fwd_lookup_struct)
857 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
858 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
859 &next_hop) == 0) ? next_hop : portid);
862 static inline uint16_t
863 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
864 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
868 return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
869 ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
874 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
875 __attribute__((unused));
877 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
878 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
880 #define MASK_ALL_PKTS 0xff
881 #define EXCLUDE_1ST_PKT 0xfe
882 #define EXCLUDE_2ND_PKT 0xfd
883 #define EXCLUDE_3RD_PKT 0xfb
884 #define EXCLUDE_4TH_PKT 0xf7
885 #define EXCLUDE_5TH_PKT 0xef
886 #define EXCLUDE_6TH_PKT 0xdf
887 #define EXCLUDE_7TH_PKT 0xbf
888 #define EXCLUDE_8TH_PKT 0x7f
891 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
893 struct ether_hdr *eth_hdr[8];
894 struct ipv4_hdr *ipv4_hdr[8];
895 uint16_t dst_port[8];
897 union ipv4_5tuple_host key[8];
900 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
901 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
902 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
903 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
904 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
905 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
906 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
907 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
909 /* Handle IPv4 headers.*/
910 ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
911 sizeof(struct ether_hdr));
912 ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
913 sizeof(struct ether_hdr));
914 ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
915 sizeof(struct ether_hdr));
916 ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
917 sizeof(struct ether_hdr));
918 ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
919 sizeof(struct ether_hdr));
920 ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
921 sizeof(struct ether_hdr));
922 ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
923 sizeof(struct ether_hdr));
924 ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
925 sizeof(struct ether_hdr));
927 #ifdef DO_RFC_1812_CHECKS
928 /* Check to make sure the packet is valid (RFC1812) */
929 uint8_t valid_mask = MASK_ALL_PKTS;
931 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
932 rte_pktmbuf_free(m[0]);
933 valid_mask &= EXCLUDE_1ST_PKT;
935 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
936 rte_pktmbuf_free(m[1]);
937 valid_mask &= EXCLUDE_2ND_PKT;
939 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
940 rte_pktmbuf_free(m[2]);
941 valid_mask &= EXCLUDE_3RD_PKT;
943 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
944 rte_pktmbuf_free(m[3]);
945 valid_mask &= EXCLUDE_4TH_PKT;
947 if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
948 rte_pktmbuf_free(m[4]);
949 valid_mask &= EXCLUDE_5TH_PKT;
951 if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
952 rte_pktmbuf_free(m[5]);
953 valid_mask &= EXCLUDE_6TH_PKT;
955 if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
956 rte_pktmbuf_free(m[6]);
957 valid_mask &= EXCLUDE_7TH_PKT;
959 if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
960 rte_pktmbuf_free(m[7]);
961 valid_mask &= EXCLUDE_8TH_PKT;
963 if (unlikely(valid_mask != MASK_ALL_PKTS)) {
969 for (i = 0; i < 8; i++)
970 if ((0x1 << i) & valid_mask)
971 l3fwd_simple_forward(m[i], portid);
973 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
975 data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
976 sizeof(struct ether_hdr) +
977 offsetof(struct ipv4_hdr, time_to_live)));
978 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
979 sizeof(struct ether_hdr) +
980 offsetof(struct ipv4_hdr, time_to_live)));
981 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
982 sizeof(struct ether_hdr) +
983 offsetof(struct ipv4_hdr, time_to_live)));
984 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
985 sizeof(struct ether_hdr) +
986 offsetof(struct ipv4_hdr, time_to_live)));
987 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
988 sizeof(struct ether_hdr) +
989 offsetof(struct ipv4_hdr, time_to_live)));
990 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
991 sizeof(struct ether_hdr) +
992 offsetof(struct ipv4_hdr, time_to_live)));
993 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
994 sizeof(struct ether_hdr) +
995 offsetof(struct ipv4_hdr, time_to_live)));
996 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
997 sizeof(struct ether_hdr) +
998 offsetof(struct ipv4_hdr, time_to_live)));
1000 key[0].xmm = _mm_and_si128(data[0], mask0);
1001 key[1].xmm = _mm_and_si128(data[1], mask0);
1002 key[2].xmm = _mm_and_si128(data[2], mask0);
1003 key[3].xmm = _mm_and_si128(data[3], mask0);
1004 key[4].xmm = _mm_and_si128(data[4], mask0);
1005 key[5].xmm = _mm_and_si128(data[5], mask0);
1006 key[6].xmm = _mm_and_si128(data[6], mask0);
1007 key[7].xmm = _mm_and_si128(data[7], mask0);
1009 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1010 &key[4], &key[5], &key[6], &key[7]};
1012 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1013 &key_array[0], 8, ret);
1014 dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1015 dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1016 dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1017 dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1018 dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1019 dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1020 dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1021 dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1023 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1024 (enabled_port_mask & 1 << dst_port[0]) == 0)
1025 dst_port[0] = portid;
1026 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1027 (enabled_port_mask & 1 << dst_port[1]) == 0)
1028 dst_port[1] = portid;
1029 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1030 (enabled_port_mask & 1 << dst_port[2]) == 0)
1031 dst_port[2] = portid;
1032 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1033 (enabled_port_mask & 1 << dst_port[3]) == 0)
1034 dst_port[3] = portid;
1035 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1036 (enabled_port_mask & 1 << dst_port[4]) == 0)
1037 dst_port[4] = portid;
1038 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1039 (enabled_port_mask & 1 << dst_port[5]) == 0)
1040 dst_port[5] = portid;
1041 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1042 (enabled_port_mask & 1 << dst_port[6]) == 0)
1043 dst_port[6] = portid;
1044 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1045 (enabled_port_mask & 1 << dst_port[7]) == 0)
1046 dst_port[7] = portid;
1048 #ifdef DO_RFC_1812_CHECKS
1049 /* Update time to live and header checksum */
1050 --(ipv4_hdr[0]->time_to_live);
1051 --(ipv4_hdr[1]->time_to_live);
1052 --(ipv4_hdr[2]->time_to_live);
1053 --(ipv4_hdr[3]->time_to_live);
1054 ++(ipv4_hdr[0]->hdr_checksum);
1055 ++(ipv4_hdr[1]->hdr_checksum);
1056 ++(ipv4_hdr[2]->hdr_checksum);
1057 ++(ipv4_hdr[3]->hdr_checksum);
1058 --(ipv4_hdr[4]->time_to_live);
1059 --(ipv4_hdr[5]->time_to_live);
1060 --(ipv4_hdr[6]->time_to_live);
1061 --(ipv4_hdr[7]->time_to_live);
1062 ++(ipv4_hdr[4]->hdr_checksum);
1063 ++(ipv4_hdr[5]->hdr_checksum);
1064 ++(ipv4_hdr[6]->hdr_checksum);
1065 ++(ipv4_hdr[7]->hdr_checksum);
1069 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1070 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1071 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1072 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1073 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1074 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1075 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1076 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1079 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1080 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1081 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1082 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1083 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1084 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1085 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1086 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1088 send_single_packet(m[0], (uint8_t)dst_port[0]);
1089 send_single_packet(m[1], (uint8_t)dst_port[1]);
1090 send_single_packet(m[2], (uint8_t)dst_port[2]);
1091 send_single_packet(m[3], (uint8_t)dst_port[3]);
1092 send_single_packet(m[4], (uint8_t)dst_port[4]);
1093 send_single_packet(m[5], (uint8_t)dst_port[5]);
1094 send_single_packet(m[6], (uint8_t)dst_port[6]);
1095 send_single_packet(m[7], (uint8_t)dst_port[7]);
1099 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1100 __m128i mask1, union ipv6_5tuple_host *key)
1102 __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1103 __m128i *, sizeof(struct ether_hdr) +
1104 offsetof(struct ipv6_hdr, payload_len)));
1105 __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1106 __m128i *, sizeof(struct ether_hdr) +
1107 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1108 __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1109 __m128i *, sizeof(struct ether_hdr) +
1110 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1112 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1113 key->xmm[1] = tmpdata1;
1114 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1118 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1121 uint16_t dst_port[8];
1122 struct ether_hdr *eth_hdr[8];
1123 union ipv6_5tuple_host key[8];
1125 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1127 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1128 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1129 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1130 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1131 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1132 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1133 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1134 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1136 /* Handle IPv6 headers.*/
1137 ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1138 sizeof(struct ether_hdr));
1139 ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1140 sizeof(struct ether_hdr));
1141 ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1142 sizeof(struct ether_hdr));
1143 ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1144 sizeof(struct ether_hdr));
1145 ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1146 sizeof(struct ether_hdr));
1147 ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1148 sizeof(struct ether_hdr));
1149 ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1150 sizeof(struct ether_hdr));
1151 ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1152 sizeof(struct ether_hdr));
1154 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1155 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1156 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1157 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1158 get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1159 get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1160 get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1161 get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1163 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1164 &key[4], &key[5], &key[6], &key[7]};
1166 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1167 &key_array[0], 4, ret);
1168 dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1169 dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1170 dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1171 dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1172 dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1173 dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1174 dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1175 dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1177 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1178 (enabled_port_mask & 1 << dst_port[0]) == 0)
1179 dst_port[0] = portid;
1180 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1181 (enabled_port_mask & 1 << dst_port[1]) == 0)
1182 dst_port[1] = portid;
1183 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1184 (enabled_port_mask & 1 << dst_port[2]) == 0)
1185 dst_port[2] = portid;
1186 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1187 (enabled_port_mask & 1 << dst_port[3]) == 0)
1188 dst_port[3] = portid;
1189 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1190 (enabled_port_mask & 1 << dst_port[4]) == 0)
1191 dst_port[4] = portid;
1192 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1193 (enabled_port_mask & 1 << dst_port[5]) == 0)
1194 dst_port[5] = portid;
1195 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1196 (enabled_port_mask & 1 << dst_port[6]) == 0)
1197 dst_port[6] = portid;
1198 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1199 (enabled_port_mask & 1 << dst_port[7]) == 0)
1200 dst_port[7] = portid;
1203 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1204 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1205 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1206 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1207 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1208 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1209 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1210 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1213 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1214 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1215 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1216 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1217 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1218 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1219 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1220 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1222 send_single_packet(m[0], dst_port[0]);
1223 send_single_packet(m[1], dst_port[1]);
1224 send_single_packet(m[2], dst_port[2]);
1225 send_single_packet(m[3], dst_port[3]);
1226 send_single_packet(m[4], dst_port[4]);
1227 send_single_packet(m[5], dst_port[5]);
1228 send_single_packet(m[6], dst_port[6]);
1229 send_single_packet(m[7], dst_port[7]);
1232 #endif /* APP_LOOKUP_METHOD */
1234 static __rte_always_inline void
1235 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1237 struct ether_hdr *eth_hdr;
1238 struct ipv4_hdr *ipv4_hdr;
1241 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1243 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1244 /* Handle IPv4 headers.*/
1245 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1246 sizeof(struct ether_hdr));
1248 #ifdef DO_RFC_1812_CHECKS
1249 /* Check to make sure the packet is valid (RFC1812) */
1250 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1251 rte_pktmbuf_free(m);
1256 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1257 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1258 if (dst_port >= RTE_MAX_ETHPORTS ||
1259 (enabled_port_mask & 1 << dst_port) == 0)
1262 #ifdef DO_RFC_1812_CHECKS
1263 /* Update time to live and header checksum */
1264 --(ipv4_hdr->time_to_live);
1265 ++(ipv4_hdr->hdr_checksum);
1268 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1271 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1273 send_single_packet(m, dst_port);
1274 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1275 /* Handle IPv6 headers.*/
1276 struct ipv6_hdr *ipv6_hdr;
1278 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1279 sizeof(struct ether_hdr));
1281 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1282 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1284 if (dst_port >= RTE_MAX_ETHPORTS ||
1285 (enabled_port_mask & 1 << dst_port) == 0)
1289 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1292 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1294 send_single_packet(m, dst_port);
1296 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1297 rte_pktmbuf_free(m);
1300 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1301 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1302 #ifdef DO_RFC_1812_CHECKS
1304 #define IPV4_MIN_VER_IHL 0x45
1305 #define IPV4_MAX_VER_IHL 0x4f
1306 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1308 /* Minimum value of IPV4 total length (20B) in network byte order. */
1309 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1312 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1313 * - The IP version number must be 4.
1314 * - The IP header length field must be large enough to hold the
1315 * minimum length legal IP datagram (20 bytes = 5 words).
1316 * - The IP total length field must be large enough to hold the IP
1317 * datagram header, whose length is specified in the IP header length
1319 * If we encounter invalid IPV4 packet, then set destination port for it
1320 * to BAD_PORT value.
1322 static __rte_always_inline void
1323 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1327 if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1328 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1330 ipv4_hdr->time_to_live--;
1331 ipv4_hdr->hdr_checksum++;
1333 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1334 ((uint8_t)ipv4_hdr->total_length == 0 &&
1335 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1342 #define rfc1812_process(mb, dp, ptype) do { } while (0)
1343 #endif /* DO_RFC_1812_CHECKS */
1344 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1347 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1348 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1350 static __rte_always_inline uint16_t
1351 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1354 struct ipv6_hdr *ipv6_hdr;
1355 struct ether_hdr *eth_hdr;
1357 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1358 return (uint16_t) ((rte_lpm_lookup(
1359 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1360 &next_hop) == 0) ? next_hop : portid);
1362 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1364 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1365 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1367 return (uint16_t) ((rte_lpm6_lookup(
1368 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1369 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1378 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1380 struct ether_hdr *eth_hdr;
1381 struct ipv4_hdr *ipv4_hdr;
1386 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1387 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1389 dst_ipv4 = ipv4_hdr->dst_addr;
1390 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1391 dp = get_dst_port(pkt, dst_ipv4, portid);
1393 te = _mm_load_si128((__m128i *)eth_hdr);
1397 rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1399 te = _mm_blend_epi16(te, ve, MASK_ETH);
1400 _mm_store_si128((__m128i *)eth_hdr, te);
1404 * Read packet_type and destination IPV4 addresses from 4 mbufs.
1407 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1409 uint32_t *ipv4_flag)
1411 struct ipv4_hdr *ipv4_hdr;
1412 struct ether_hdr *eth_hdr;
1413 uint32_t x0, x1, x2, x3;
1415 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1416 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1417 x0 = ipv4_hdr->dst_addr;
1418 ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1420 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1421 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1422 x1 = ipv4_hdr->dst_addr;
1423 ipv4_flag[0] &= pkt[1]->packet_type;
1425 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1426 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1427 x2 = ipv4_hdr->dst_addr;
1428 ipv4_flag[0] &= pkt[2]->packet_type;
1430 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1431 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1432 x3 = ipv4_hdr->dst_addr;
1433 ipv4_flag[0] &= pkt[3]->packet_type;
1435 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1439 * Lookup into LPM for destination port.
1440 * If lookup fails, use incoming port (portid) as destination port.
1443 processx4_step2(__m128i dip,
1446 struct rte_mbuf *pkt[FWDSTEP],
1447 uint16_t dprt[FWDSTEP])
1450 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1451 4, 5, 6, 7, 0, 1, 2, 3);
1453 /* Byte swap 4 IPV4 addresses. */
1454 dip = _mm_shuffle_epi8(dip, bswap_mask);
1456 /* if all 4 packets are IPV4. */
1457 if (likely(ipv4_flag)) {
1458 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1461 /* get rid of unused upper 16 bit for each dport. */
1462 dst.x = _mm_packs_epi32(dst.x, dst.x);
1463 *(uint64_t *)dprt = dst.u64[0];
1466 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1467 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1468 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1469 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1474 * Update source and destination MAC addresses in the ethernet header.
1475 * Perform RFC1812 checks and updates for IPV4 packets.
1478 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1480 __m128i te[FWDSTEP];
1481 __m128i ve[FWDSTEP];
1482 __m128i *p[FWDSTEP];
1484 p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1485 p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1486 p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1487 p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1489 ve[0] = val_eth[dst_port[0]];
1490 te[0] = _mm_load_si128(p[0]);
1492 ve[1] = val_eth[dst_port[1]];
1493 te[1] = _mm_load_si128(p[1]);
1495 ve[2] = val_eth[dst_port[2]];
1496 te[2] = _mm_load_si128(p[2]);
1498 ve[3] = val_eth[dst_port[3]];
1499 te[3] = _mm_load_si128(p[3]);
1501 /* Update first 12 bytes, keep rest bytes intact. */
1502 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1503 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1504 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1505 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1507 _mm_store_si128(p[0], te[0]);
1508 _mm_store_si128(p[1], te[1]);
1509 _mm_store_si128(p[2], te[2]);
1510 _mm_store_si128(p[3], te[3]);
1512 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1513 &dst_port[0], pkt[0]->packet_type);
1514 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1515 &dst_port[1], pkt[1]->packet_type);
1516 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1517 &dst_port[2], pkt[2]->packet_type);
1518 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1519 &dst_port[3], pkt[3]->packet_type);
1523 * We group consecutive packets with the same destionation port into one burst.
1524 * To avoid extra latency this is done together with some other packet
1525 * processing, but after we made a final decision about packet's destination.
1526 * To do this we maintain:
1527 * pnum - array of number of consecutive packets with the same dest port for
1528 * each packet in the input burst.
1529 * lp - pointer to the last updated element in the pnum.
1530 * dlp - dest port value lp corresponds to.
1533 #define GRPSZ (1 << FWDSTEP)
1534 #define GRPMSK (GRPSZ - 1)
1536 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1537 if (likely((dlp) == (dcp)[(idx)])) { \
1540 (dlp) = (dcp)[idx]; \
1541 (lp) = (pn) + (idx); \
1547 * Group consecutive packets with the same destination port in bursts of 4.
1548 * Suppose we have array of destionation ports:
1549 * dst_port[] = {a, b, c, d,, e, ... }
1550 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1551 * We doing 4 comparisons at once and the result is 4 bit mask.
1552 * This mask is used as an index into prebuild array of pnum values.
1554 static inline uint16_t *
1555 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1557 static const struct {
1558 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1559 int32_t idx; /* index for new last updated elemnet. */
1560 uint16_t lpv; /* add value to the last updated element. */
1563 /* 0: a != b, b != c, c != d, d != e */
1564 .pnum = UINT64_C(0x0001000100010001),
1569 /* 1: a == b, b != c, c != d, d != e */
1570 .pnum = UINT64_C(0x0001000100010002),
1575 /* 2: a != b, b == c, c != d, d != e */
1576 .pnum = UINT64_C(0x0001000100020001),
1581 /* 3: a == b, b == c, c != d, d != e */
1582 .pnum = UINT64_C(0x0001000100020003),
1587 /* 4: a != b, b != c, c == d, d != e */
1588 .pnum = UINT64_C(0x0001000200010001),
1593 /* 5: a == b, b != c, c == d, d != e */
1594 .pnum = UINT64_C(0x0001000200010002),
1599 /* 6: a != b, b == c, c == d, d != e */
1600 .pnum = UINT64_C(0x0001000200030001),
1605 /* 7: a == b, b == c, c == d, d != e */
1606 .pnum = UINT64_C(0x0001000200030004),
1611 /* 8: a != b, b != c, c != d, d == e */
1612 .pnum = UINT64_C(0x0002000100010001),
1617 /* 9: a == b, b != c, c != d, d == e */
1618 .pnum = UINT64_C(0x0002000100010002),
1623 /* 0xa: a != b, b == c, c != d, d == e */
1624 .pnum = UINT64_C(0x0002000100020001),
1629 /* 0xb: a == b, b == c, c != d, d == e */
1630 .pnum = UINT64_C(0x0002000100020003),
1635 /* 0xc: a != b, b != c, c == d, d == e */
1636 .pnum = UINT64_C(0x0002000300010001),
1641 /* 0xd: a == b, b != c, c == d, d == e */
1642 .pnum = UINT64_C(0x0002000300010002),
1647 /* 0xe: a != b, b == c, c == d, d == e */
1648 .pnum = UINT64_C(0x0002000300040001),
1653 /* 0xf: a == b, b == c, c == d, d == e */
1654 .pnum = UINT64_C(0x0002000300040005),
1661 uint16_t u16[FWDSTEP + 1];
1663 } *pnum = (void *)pn;
1667 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1668 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1669 v = _mm_movemask_ps((__m128)dp1);
1671 /* update last port counter. */
1672 lp[0] += gptbl[v].lpv;
1674 /* if dest port value has changed. */
1676 pnum->u64 = gptbl[v].pnum;
1677 pnum->u16[FWDSTEP] = 1;
1678 lp = pnum->u16 + gptbl[v].idx;
1684 #endif /* APP_LOOKUP_METHOD */
1687 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1693 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1694 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1698 uint16_t dst_port[MAX_PKT_BURST];
1699 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1700 uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1701 uint16_t pnum[MAX_PKT_BURST + 1];
1705 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1706 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1709 * Send nb_rx - nb_rx%8 packets
1712 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1714 for (j = 0; j < n; j += 8) {
1716 pkts_burst[j]->packet_type &
1717 pkts_burst[j+1]->packet_type &
1718 pkts_burst[j+2]->packet_type &
1719 pkts_burst[j+3]->packet_type &
1720 pkts_burst[j+4]->packet_type &
1721 pkts_burst[j+5]->packet_type &
1722 pkts_burst[j+6]->packet_type &
1723 pkts_burst[j+7]->packet_type;
1724 if (pkt_type & RTE_PTYPE_L3_IPV4) {
1725 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1726 } else if (pkt_type &
1727 RTE_PTYPE_L3_IPV6) {
1728 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1730 l3fwd_simple_forward(pkts_burst[j], portid);
1731 l3fwd_simple_forward(pkts_burst[j+1], portid);
1732 l3fwd_simple_forward(pkts_burst[j+2], portid);
1733 l3fwd_simple_forward(pkts_burst[j+3], portid);
1734 l3fwd_simple_forward(pkts_burst[j+4], portid);
1735 l3fwd_simple_forward(pkts_burst[j+5], portid);
1736 l3fwd_simple_forward(pkts_burst[j+6], portid);
1737 l3fwd_simple_forward(pkts_burst[j+7], portid);
1740 for (; j < nb_rx ; j++)
1741 l3fwd_simple_forward(pkts_burst[j], portid);
1743 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1745 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1746 for (j = 0; j != k; j += FWDSTEP)
1747 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1748 &ipv4_flag[j / FWDSTEP]);
1750 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1751 for (j = 0; j != k; j += FWDSTEP)
1752 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1753 portid, &pkts_burst[j], &dst_port[j]);
1756 * Finish packet processing and group consecutive
1757 * packets with the same destination port.
1759 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1766 processx4_step3(pkts_burst, dst_port);
1768 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1769 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1771 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1772 processx4_step3(&pkts_burst[j], &dst_port[j]);
1776 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1778 dp2 = _mm_loadu_si128(
1779 (__m128i *)&dst_port[j - FWDSTEP + 1]);
1780 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1784 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1786 dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1787 sizeof(dst_port[0]));
1791 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1793 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1794 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1797 * remove values added by the last repeated
1801 dlp = dst_port[j - 1];
1803 /* set dlp and lp to the never used values. */
1805 lp = pnum + MAX_PKT_BURST;
1808 /* Process up to last 3 packets one by one. */
1809 switch (nb_rx % FWDSTEP) {
1811 process_packet(pkts_burst[j], dst_port + j, portid);
1812 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1816 process_packet(pkts_burst[j], dst_port + j, portid);
1817 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1821 process_packet(pkts_burst[j], dst_port + j, portid);
1822 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1827 * Send packets out, through destination port.
1828 * Consecuteve pacekts with the same destination port
1829 * are already grouped together.
1830 * If destination port for the packet equals BAD_PORT,
1831 * then free the packet without sending it out.
1833 for (j = 0; j < nb_rx; j += k) {
1841 if (likely(pn != BAD_PORT))
1842 send_packetsx4(pn, pkts_burst + j, k);
1844 for (m = j; m != j + k; m++)
1845 rte_pktmbuf_free(pkts_burst[m]);
1849 #endif /* APP_LOOKUP_METHOD */
1850 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1852 /* Prefetch first packets */
1853 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1854 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1856 /* Prefetch and forward already prefetched packets */
1857 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1858 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1859 j + PREFETCH_OFFSET], void *));
1860 l3fwd_simple_forward(pkts_burst[j], portid);
1863 /* Forward remaining prefetched packets */
1864 for (; j < nb_rx; j++)
1865 l3fwd_simple_forward(pkts_burst[j], portid);
1867 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1871 #if (APP_CPU_LOAD > 0)
1874 * CPU-load stats collector
1877 cpu_load_collector(__rte_unused void *arg) {
1880 uint64_t prev_tsc, diff_tsc, cur_tsc;
1881 uint64_t total[MAX_CPU] = { 0 };
1882 unsigned min_cpu = MAX_CPU;
1883 unsigned max_cpu = 0;
1888 unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1889 struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1891 struct thread_conf *thread_conf;
1893 const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1894 US_PER_S * CPU_LOAD_TIMEOUT_US;
1898 * Wait for all threads
1901 printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1904 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1907 while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1910 for (i = 0; i < n_rx_thread; i++) {
1912 thread_conf = &rx_thread[i].conf;
1913 cpu_id = thread_conf->cpu_id;
1914 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1916 if (cpu_id > max_cpu)
1918 if (cpu_id < min_cpu)
1921 for (i = 0; i < n_tx_thread; i++) {
1923 thread_conf = &tx_thread[i].conf;
1924 cpu_id = thread_conf->cpu_id;
1925 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1927 if (thread_conf->cpu_id > max_cpu)
1928 max_cpu = thread_conf->cpu_id;
1929 if (thread_conf->cpu_id < min_cpu)
1930 min_cpu = thread_conf->cpu_id;
1936 for (i = min_cpu; i <= max_cpu; i++) {
1937 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1938 for (k = 0; k < n_thread_per_cpu[i]; k++)
1939 if (thread_per_cpu[i][k]->busy[j]) {
1944 cpu_load.hits[j][i]++;
1956 cur_tsc = rte_rdtsc();
1958 diff_tsc = cur_tsc - prev_tsc;
1959 if (unlikely(diff_tsc > interval_tsc)) {
1963 printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1964 n_rx_thread, n_tx_thread);
1966 printf("cpu# proc%% poll%% overhead%%\n\n");
1968 for (i = min_cpu; i <= max_cpu; i++) {
1970 printf("CPU %d:", i);
1971 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1972 printf("%7" PRIu64 "",
1973 cpu_load.hits[j][i] * 100 / cpu_load.counter);
1974 hits += cpu_load.hits[j][i];
1975 cpu_load.hits[j][i] = 0;
1977 printf("%7" PRIu64 "\n",
1978 100 - total[i] * 100 / cpu_load.counter);
1981 cpu_load.counter = 0;
1988 #endif /* APP_CPU_LOAD */
1991 * Null processing lthread loop
1993 * This loop is used to start empty scheduler on lcore.
1996 lthread_null(__rte_unused void *args)
1998 int lcore_id = rte_lcore_id();
2000 RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2004 /* main processing loop */
2006 lthread_tx_per_ring(void *dummy)
2010 struct rte_ring *ring;
2011 struct thread_tx_conf *tx_conf;
2012 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2013 struct lthread_cond *ready;
2015 tx_conf = (struct thread_tx_conf *)dummy;
2016 ring = tx_conf->ring;
2017 ready = *tx_conf->ready;
2019 lthread_set_data((void *)tx_conf);
2022 * Move this lthread to lcore
2024 lthread_set_affinity(tx_conf->conf.lcore_id);
2026 RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2029 rte_atomic16_inc(&tx_counter);
2033 * Read packet from ring
2035 SET_CPU_BUSY(tx_conf, CPU_POLL);
2036 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2037 MAX_PKT_BURST, NULL);
2038 SET_CPU_IDLE(tx_conf, CPU_POLL);
2041 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2042 portid = pkts_burst[0]->port;
2043 process_burst(pkts_burst, nb_rx, portid);
2044 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2047 lthread_cond_wait(ready, 0);
2053 * Main tx-lthreads spawner lthread.
2055 * This lthread is used to spawn one new lthread per ring from producers.
2059 lthread_tx(void *args)
2065 struct thread_tx_conf *tx_conf;
2067 tx_conf = (struct thread_tx_conf *)args;
2068 lthread_set_data((void *)tx_conf);
2071 * Move this lthread to the selected lcore
2073 lthread_set_affinity(tx_conf->conf.lcore_id);
2076 * Spawn tx readers (one per input ring)
2078 lthread_create(<, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2081 lcore_id = rte_lcore_id();
2083 RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2085 tx_conf->conf.cpu_id = sched_getcpu();
2088 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2091 * TX burst queue drain
2093 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2094 if (tx_conf->tx_mbufs[portid].len == 0)
2096 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2097 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2098 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2099 tx_conf->tx_mbufs[portid].len = 0;
2106 lthread_rx(void *dummy)
2114 int len[RTE_MAX_LCORE] = { 0 };
2115 int old_len, new_len;
2116 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2117 struct thread_rx_conf *rx_conf;
2119 rx_conf = (struct thread_rx_conf *)dummy;
2120 lthread_set_data((void *)rx_conf);
2123 * Move this lthread to lcore
2125 lthread_set_affinity(rx_conf->conf.lcore_id);
2127 if (rx_conf->n_rx_queue == 0) {
2128 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2132 RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2134 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2136 portid = rx_conf->rx_queue_list[i].port_id;
2137 queueid = rx_conf->rx_queue_list[i].queue_id;
2138 RTE_LOG(INFO, L3FWD,
2139 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2140 rte_lcore_id(), portid, queueid);
2144 * Init all condition variables (one per rx thread)
2146 for (i = 0; i < rx_conf->n_rx_queue; i++)
2147 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2151 rx_conf->conf.cpu_id = sched_getcpu();
2152 rte_atomic16_inc(&rx_counter);
2156 * Read packet from RX queues
2158 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2159 portid = rx_conf->rx_queue_list[i].port_id;
2160 queueid = rx_conf->rx_queue_list[i].queue_id;
2162 SET_CPU_BUSY(rx_conf, CPU_POLL);
2163 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2165 SET_CPU_IDLE(rx_conf, CPU_POLL);
2168 worker_id = (worker_id + 1) % rx_conf->n_ring;
2169 old_len = len[worker_id];
2171 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2172 ret = rte_ring_sp_enqueue_burst(
2173 rx_conf->ring[worker_id],
2174 (void **) pkts_burst,
2177 new_len = old_len + ret;
2179 if (new_len >= BURST_SIZE) {
2180 lthread_cond_signal(rx_conf->ready[worker_id]);
2184 len[worker_id] = new_len;
2186 if (unlikely(ret < nb_rx)) {
2189 for (k = ret; k < nb_rx; k++) {
2190 struct rte_mbuf *m = pkts_burst[k];
2192 rte_pktmbuf_free(m);
2195 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2204 * Start scheduler with initial lthread on lcore
2206 * This lthread loop spawns all rx and tx lthreads on master lcore
2210 lthread_spawner(__rte_unused void *arg) {
2211 struct lthread *lt[MAX_THREAD];
2215 printf("Entering lthread_spawner\n");
2218 * Create producers (rx threads) on default lcore
2220 for (i = 0; i < n_rx_thread; i++) {
2221 rx_thread[i].conf.thread_id = i;
2222 lthread_create(<[n_thread], -1, lthread_rx,
2223 (void *)&rx_thread[i]);
2228 * Wait for all producers. Until some producers can be started on the same
2229 * scheduler as this lthread, yielding is required to let them to run and
2230 * prevent deadlock here.
2232 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2233 lthread_sleep(100000);
2236 * Create consumers (tx threads) on default lcore_id
2238 for (i = 0; i < n_tx_thread; i++) {
2239 tx_thread[i].conf.thread_id = i;
2240 lthread_create(<[n_thread], -1, lthread_tx,
2241 (void *)&tx_thread[i]);
2246 * Wait for all threads finished
2248 for (i = 0; i < n_thread; i++)
2249 lthread_join(lt[i], NULL);
2254 * Start master scheduler with initial lthread spawning rx and tx lthreads
2255 * (main_lthread_master).
2258 lthread_master_spawner(__rte_unused void *arg) {
2260 int lcore_id = rte_lcore_id();
2262 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2263 lthread_create(<, -1, lthread_spawner, NULL);
2270 * Start scheduler on lcore.
2273 sched_spawner(__rte_unused void *arg) {
2275 int lcore_id = rte_lcore_id();
2278 if (lcore_id == cpu_load_lcore_id) {
2279 cpu_load_collector(arg);
2282 #endif /* APP_CPU_LOAD */
2284 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2285 lthread_create(<, -1, lthread_null, NULL);
2291 /* main processing loop */
2293 pthread_tx(void *dummy)
2295 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2296 uint64_t prev_tsc, diff_tsc, cur_tsc;
2299 struct thread_tx_conf *tx_conf;
2301 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2302 US_PER_S * BURST_TX_DRAIN_US;
2306 tx_conf = (struct thread_tx_conf *)dummy;
2308 RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2310 tx_conf->conf.cpu_id = sched_getcpu();
2311 rte_atomic16_inc(&tx_counter);
2314 cur_tsc = rte_rdtsc();
2317 * TX burst queue drain
2319 diff_tsc = cur_tsc - prev_tsc;
2320 if (unlikely(diff_tsc > drain_tsc)) {
2323 * This could be optimized (use queueid instead of
2324 * portid), but it is not called so often
2326 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2327 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2328 if (tx_conf->tx_mbufs[portid].len == 0)
2330 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2331 tx_conf->tx_mbufs[portid].len = 0;
2333 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2339 * Read packet from ring
2341 SET_CPU_BUSY(tx_conf, CPU_POLL);
2342 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2343 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2344 SET_CPU_IDLE(tx_conf, CPU_POLL);
2346 if (unlikely(nb_rx == 0)) {
2351 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2352 portid = pkts_burst[0]->port;
2353 process_burst(pkts_burst, nb_rx, portid);
2354 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2360 pthread_rx(void *dummy)
2369 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2371 struct thread_rx_conf *rx_conf;
2373 lcore_id = rte_lcore_id();
2374 rx_conf = (struct thread_rx_conf *)dummy;
2376 if (rx_conf->n_rx_queue == 0) {
2377 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2381 RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2383 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2385 portid = rx_conf->rx_queue_list[i].port_id;
2386 queueid = rx_conf->rx_queue_list[i].queue_id;
2387 RTE_LOG(INFO, L3FWD,
2388 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2389 lcore_id, portid, queueid);
2393 rx_conf->conf.cpu_id = sched_getcpu();
2394 rte_atomic16_inc(&rx_counter);
2398 * Read packet from RX queues
2400 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2401 portid = rx_conf->rx_queue_list[i].port_id;
2402 queueid = rx_conf->rx_queue_list[i].queue_id;
2404 SET_CPU_BUSY(rx_conf, CPU_POLL);
2405 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2407 SET_CPU_IDLE(rx_conf, CPU_POLL);
2414 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2415 worker_id = (worker_id + 1) % rx_conf->n_ring;
2416 n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2417 (void **)pkts_burst, nb_rx, NULL);
2419 if (unlikely(n != nb_rx)) {
2422 for (k = n; k < nb_rx; k++) {
2423 struct rte_mbuf *m = pkts_burst[k];
2425 rte_pktmbuf_free(m);
2429 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2439 pthread_run(__rte_unused void *arg) {
2440 int lcore_id = rte_lcore_id();
2443 for (i = 0; i < n_rx_thread; i++)
2444 if (rx_thread[i].conf.lcore_id == lcore_id) {
2445 printf("Start rx thread on %d...\n", lcore_id);
2446 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2447 RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2448 pthread_rx((void *)&rx_thread[i]);
2452 for (i = 0; i < n_tx_thread; i++)
2453 if (tx_thread[i].conf.lcore_id == lcore_id) {
2454 printf("Start tx thread on %d...\n", lcore_id);
2455 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2456 RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2457 pthread_tx((void *)&tx_thread[i]);
2462 if (lcore_id == cpu_load_lcore_id)
2463 cpu_load_collector(arg);
2464 #endif /* APP_CPU_LOAD */
2470 check_lcore_params(void)
2472 uint8_t queue, lcore;
2476 for (i = 0; i < nb_rx_thread_params; ++i) {
2477 queue = rx_thread_params[i].queue_id;
2478 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2479 printf("invalid queue number: %hhu\n", queue);
2482 lcore = rx_thread_params[i].lcore_id;
2483 if (!rte_lcore_is_enabled(lcore)) {
2484 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2487 socketid = rte_lcore_to_socket_id(lcore);
2488 if ((socketid != 0) && (numa_on == 0))
2489 printf("warning: lcore %hhu is on socket %d with numa off\n",
2496 check_port_config(const unsigned nb_ports)
2501 for (i = 0; i < nb_rx_thread_params; ++i) {
2502 portid = rx_thread_params[i].port_id;
2503 if ((enabled_port_mask & (1 << portid)) == 0) {
2504 printf("port %u is not enabled in port mask\n", portid);
2507 if (portid >= nb_ports) {
2508 printf("port %u is not present on the board\n", portid);
2516 get_port_n_rx_queues(const uint16_t port)
2521 for (i = 0; i < nb_rx_thread_params; ++i)
2522 if (rx_thread_params[i].port_id == port &&
2523 rx_thread_params[i].queue_id > queue)
2524 queue = rx_thread_params[i].queue_id;
2526 return (uint8_t)(++queue);
2533 struct thread_rx_conf *rx_conf;
2534 struct thread_tx_conf *tx_conf;
2535 unsigned rx_thread_id, tx_thread_id;
2537 struct rte_ring *ring = NULL;
2539 for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2541 tx_conf = &tx_thread[tx_thread_id];
2543 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2544 tx_conf->conf.thread_id);
2546 rx_thread_id = tx_conf->conf.thread_id;
2547 if (rx_thread_id > n_tx_thread) {
2548 printf("connection from tx-thread %u to rx-thread %u fails "
2549 "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2553 rx_conf = &rx_thread[rx_thread_id];
2554 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2556 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2557 socket_io, rx_thread_id, tx_thread_id);
2559 ring = rte_ring_create(name, 1024 * 4, socket_io,
2560 RING_F_SP_ENQ | RING_F_SC_DEQ);
2563 rte_panic("Cannot create ring to connect rx-thread %u "
2564 "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2567 rx_conf->ring[rx_conf->n_ring] = ring;
2569 tx_conf->ring = ring;
2570 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2578 init_rx_queues(void)
2580 uint16_t i, nb_rx_queue;
2585 for (i = 0; i < nb_rx_thread_params; ++i) {
2586 thread = rx_thread_params[i].thread_id;
2587 nb_rx_queue = rx_thread[thread].n_rx_queue;
2589 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2590 printf("error: too many queues (%u) for thread: %u\n",
2591 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2595 rx_thread[thread].conf.thread_id = thread;
2596 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2597 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2598 rx_thread_params[i].port_id;
2599 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2600 rx_thread_params[i].queue_id;
2601 rx_thread[thread].n_rx_queue++;
2603 if (thread >= n_rx_thread)
2604 n_rx_thread = thread + 1;
2611 init_tx_threads(void)
2616 for (i = 0; i < nb_tx_thread_params; ++i) {
2617 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2618 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2626 print_usage(const char *prgname)
2628 printf("%s [EAL options] -- -p PORTMASK -P"
2629 " [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2630 " [--tx (lcore,thread)[,(lcore,thread]]"
2631 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2632 " [--parse-ptype]\n\n"
2633 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2634 " -P : enable promiscuous mode\n"
2635 " --rx (port,queue,lcore,thread): rx queues configuration\n"
2636 " --tx (lcore,thread): tx threads configuration\n"
2637 " --stat-lcore LCORE: use lcore for stat collector\n"
2638 " --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2639 " --no-numa: optional, disable numa awareness\n"
2640 " --ipv6: optional, specify it if running ipv6 packets\n"
2641 " --enable-jumbo: enable jumbo frame"
2642 " which max packet len is PKTLEN in decimal (64-9600)\n"
2643 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2644 " --no-lthreads: turn off lthread model\n"
2645 " --parse-ptype: set to use software to analyze packet type\n\n",
2649 static int parse_max_pkt_len(const char *pktlen)
2654 /* parse decimal string */
2655 len = strtoul(pktlen, &end, 10);
2656 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2666 parse_portmask(const char *portmask)
2671 /* parse hexadecimal string */
2672 pm = strtoul(portmask, &end, 16);
2673 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2682 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2684 parse_hash_entry_number(const char *hash_entry_num)
2687 unsigned long hash_en;
2689 /* parse hexadecimal string */
2690 hash_en = strtoul(hash_entry_num, &end, 16);
2691 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2702 parse_rx_config(const char *q_arg)
2705 const char *p, *p0 = q_arg;
2714 unsigned long int_fld[_NUM_FLD];
2715 char *str_fld[_NUM_FLD];
2719 nb_rx_thread_params = 0;
2721 while ((p = strchr(p0, '(')) != NULL) {
2723 p0 = strchr(p, ')');
2728 if (size >= sizeof(s))
2731 snprintf(s, sizeof(s), "%.*s", size, p);
2732 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2734 for (i = 0; i < _NUM_FLD; i++) {
2736 int_fld[i] = strtoul(str_fld[i], &end, 0);
2737 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2740 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2741 printf("exceeded max number of rx params: %hu\n",
2742 nb_rx_thread_params);
2745 rx_thread_params_array[nb_rx_thread_params].port_id =
2747 rx_thread_params_array[nb_rx_thread_params].queue_id =
2748 (uint8_t)int_fld[FLD_QUEUE];
2749 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2750 (uint8_t)int_fld[FLD_LCORE];
2751 rx_thread_params_array[nb_rx_thread_params].thread_id =
2752 (uint8_t)int_fld[FLD_THREAD];
2753 ++nb_rx_thread_params;
2755 rx_thread_params = rx_thread_params_array;
2760 parse_tx_config(const char *q_arg)
2763 const char *p, *p0 = q_arg;
2770 unsigned long int_fld[_NUM_FLD];
2771 char *str_fld[_NUM_FLD];
2775 nb_tx_thread_params = 0;
2777 while ((p = strchr(p0, '(')) != NULL) {
2779 p0 = strchr(p, ')');
2784 if (size >= sizeof(s))
2787 snprintf(s, sizeof(s), "%.*s", size, p);
2788 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2790 for (i = 0; i < _NUM_FLD; i++) {
2792 int_fld[i] = strtoul(str_fld[i], &end, 0);
2793 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2796 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2797 printf("exceeded max number of tx params: %hu\n",
2798 nb_tx_thread_params);
2801 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2802 (uint8_t)int_fld[FLD_LCORE];
2803 tx_thread_params_array[nb_tx_thread_params].thread_id =
2804 (uint8_t)int_fld[FLD_THREAD];
2805 ++nb_tx_thread_params;
2807 tx_thread_params = tx_thread_params_array;
2812 #if (APP_CPU_LOAD > 0)
2814 parse_stat_lcore(const char *stat_lcore)
2817 unsigned long lcore_id;
2819 lcore_id = strtoul(stat_lcore, &end, 10);
2820 if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2828 parse_eth_dest(const char *optarg)
2832 uint8_t c, *dest, peer_addr[6];
2835 portid = strtoul(optarg, &port_end, 10);
2836 if (errno != 0 || port_end == optarg || *port_end++ != ',')
2837 rte_exit(EXIT_FAILURE,
2838 "Invalid eth-dest: %s", optarg);
2839 if (portid >= RTE_MAX_ETHPORTS)
2840 rte_exit(EXIT_FAILURE,
2841 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2842 portid, RTE_MAX_ETHPORTS);
2844 if (cmdline_parse_etheraddr(NULL, port_end,
2845 &peer_addr, sizeof(peer_addr)) < 0)
2846 rte_exit(EXIT_FAILURE,
2847 "Invalid ethernet address: %s\n",
2849 dest = (uint8_t *)&dest_eth_addr[portid];
2850 for (c = 0; c < 6; c++)
2851 dest[c] = peer_addr[c];
2852 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2855 #define CMD_LINE_OPT_RX_CONFIG "rx"
2856 #define CMD_LINE_OPT_TX_CONFIG "tx"
2857 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2858 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2859 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2860 #define CMD_LINE_OPT_IPV6 "ipv6"
2861 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2862 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2863 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2864 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2866 /* Parse the argument given in the command line of the application */
2868 parse_args(int argc, char **argv)
2873 char *prgname = argv[0];
2874 static struct option lgopts[] = {
2875 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2876 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2877 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2878 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2879 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2880 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2881 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2882 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2883 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2884 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2890 while ((opt = getopt_long(argc, argvopt, "p:P",
2891 lgopts, &option_index)) != EOF) {
2896 enabled_port_mask = parse_portmask(optarg);
2897 if (enabled_port_mask == 0) {
2898 printf("invalid portmask\n");
2899 print_usage(prgname);
2904 printf("Promiscuous mode selected\n");
2910 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2911 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2912 ret = parse_rx_config(optarg);
2914 printf("invalid rx-config\n");
2915 print_usage(prgname);
2920 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2921 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2922 ret = parse_tx_config(optarg);
2924 printf("invalid tx-config\n");
2925 print_usage(prgname);
2930 #if (APP_CPU_LOAD > 0)
2931 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2932 sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2933 cpu_load_lcore_id = parse_stat_lcore(optarg);
2937 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2938 sizeof(CMD_LINE_OPT_ETH_DEST)))
2939 parse_eth_dest(optarg);
2941 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2942 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2943 printf("numa is disabled\n");
2947 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2948 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2949 sizeof(CMD_LINE_OPT_IPV6))) {
2950 printf("ipv6 is specified\n");
2955 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2956 sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2957 printf("l-threads model is disabled\n");
2961 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2962 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2963 printf("software packet type parsing enabled\n");
2967 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2968 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2969 struct option lenopts = {"max-pkt-len", required_argument, 0,
2972 printf("jumbo frame is enabled - disabling simple TX path\n");
2973 port_conf.rxmode.jumbo_frame = 1;
2975 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2976 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2979 ret = parse_max_pkt_len(optarg);
2980 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2981 printf("invalid packet length\n");
2982 print_usage(prgname);
2985 port_conf.rxmode.max_rx_pkt_len = ret;
2987 printf("set jumbo frame max packet length to %u\n",
2988 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2990 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2991 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2992 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2993 ret = parse_hash_entry_number(optarg);
2994 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2995 hash_entry_number = ret;
2997 printf("invalid hash entry number\n");
2998 print_usage(prgname);
3006 print_usage(prgname);
3012 argv[optind-1] = prgname;
3015 optind = 1; /* reset getopt lib */
3020 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
3022 char buf[ETHER_ADDR_FMT_SIZE];
3024 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
3025 printf("%s%s", name, buf);
3028 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3030 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3031 union ipv4_5tuple_host *key2)
3033 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3034 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3035 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3036 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3037 key2->proto = key1->proto;
3042 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3043 union ipv6_5tuple_host *key2)
3047 for (i = 0; i < 16; i++) {
3048 key2->ip_dst[i] = key1->ip_dst[i];
3049 key2->ip_src[i] = key1->ip_src[i];
3051 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3052 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3053 key2->proto = key1->proto;
3059 #define BYTE_VALUE_MAX 256
3060 #define ALL_32_BITS 0xffffffff
3061 #define BIT_8_TO_15 0x0000ff00
3063 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3067 uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3069 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3070 for (i = 0; i < array_len; i++) {
3071 struct ipv4_l3fwd_route entry;
3072 union ipv4_5tuple_host newkey;
3074 entry = ipv4_l3fwd_route_array[i];
3075 convert_ipv4_5tuple(&entry.key, &newkey);
3076 ret = rte_hash_add_key(h, (void *)&newkey);
3078 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3079 " to the l3fwd hash.\n", i);
3081 ipv4_l3fwd_out_if[ret] = entry.if_out;
3083 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3086 #define BIT_16_TO_23 0x00ff0000
3088 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3092 uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3094 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3095 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3096 for (i = 0; i < array_len; i++) {
3097 struct ipv6_l3fwd_route entry;
3098 union ipv6_5tuple_host newkey;
3100 entry = ipv6_l3fwd_route_array[i];
3101 convert_ipv6_5tuple(&entry.key, &newkey);
3102 ret = rte_hash_add_key(h, (void *)&newkey);
3104 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3105 " to the l3fwd hash.\n", i);
3107 ipv6_l3fwd_out_if[ret] = entry.if_out;
3109 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3112 #define NUMBER_PORT_USED 4
3114 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3115 unsigned int nr_flow)
3119 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3121 for (i = 0; i < nr_flow; i++) {
3122 struct ipv4_l3fwd_route entry;
3123 union ipv4_5tuple_host newkey;
3124 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3125 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3127 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3129 /* Create the ipv4 exact match flow */
3130 memset(&entry, 0, sizeof(entry));
3131 switch (i & (NUMBER_PORT_USED - 1)) {
3133 entry = ipv4_l3fwd_route_array[0];
3134 entry.key.ip_dst = IPv4(101, c, b, a);
3137 entry = ipv4_l3fwd_route_array[1];
3138 entry.key.ip_dst = IPv4(201, c, b, a);
3141 entry = ipv4_l3fwd_route_array[2];
3142 entry.key.ip_dst = IPv4(111, c, b, a);
3145 entry = ipv4_l3fwd_route_array[3];
3146 entry.key.ip_dst = IPv4(211, c, b, a);
3149 convert_ipv4_5tuple(&entry.key, &newkey);
3150 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3153 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3155 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3158 printf("Hash: Adding 0x%x keys\n", nr_flow);
3162 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3163 unsigned int nr_flow)
3167 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3168 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3169 for (i = 0; i < nr_flow; i++) {
3170 struct ipv6_l3fwd_route entry;
3171 union ipv6_5tuple_host newkey;
3173 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3174 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3176 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3179 /* Create the ipv6 exact match flow */
3180 memset(&entry, 0, sizeof(entry));
3181 switch (i & (NUMBER_PORT_USED - 1)) {
3183 entry = ipv6_l3fwd_route_array[0];
3186 entry = ipv6_l3fwd_route_array[1];
3189 entry = ipv6_l3fwd_route_array[2];
3192 entry = ipv6_l3fwd_route_array[3];
3195 entry.key.ip_dst[13] = c;
3196 entry.key.ip_dst[14] = b;
3197 entry.key.ip_dst[15] = a;
3198 convert_ipv6_5tuple(&entry.key, &newkey);
3199 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3202 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3204 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3207 printf("Hash: Adding 0x%x keys\n", nr_flow);
3211 setup_hash(int socketid)
3213 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3215 .entries = L3FWD_HASH_ENTRIES,
3216 .key_len = sizeof(union ipv4_5tuple_host),
3217 .hash_func = ipv4_hash_crc,
3218 .hash_func_init_val = 0,
3221 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3223 .entries = L3FWD_HASH_ENTRIES,
3224 .key_len = sizeof(union ipv6_5tuple_host),
3225 .hash_func = ipv6_hash_crc,
3226 .hash_func_init_val = 0,
3231 /* create ipv4 hash */
3232 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3233 ipv4_l3fwd_hash_params.name = s;
3234 ipv4_l3fwd_hash_params.socket_id = socketid;
3235 ipv4_l3fwd_lookup_struct[socketid] =
3236 rte_hash_create(&ipv4_l3fwd_hash_params);
3237 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3238 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3239 "socket %d\n", socketid);
3241 /* create ipv6 hash */
3242 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3243 ipv6_l3fwd_hash_params.name = s;
3244 ipv6_l3fwd_hash_params.socket_id = socketid;
3245 ipv6_l3fwd_lookup_struct[socketid] =
3246 rte_hash_create(&ipv6_l3fwd_hash_params);
3247 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3248 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3249 "socket %d\n", socketid);
3251 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3252 /* For testing hash matching with a large number of flows we
3253 * generate millions of IP 5-tuples with an incremented dst
3254 * address to initialize the hash table. */
3256 /* populate the ipv4 hash */
3257 populate_ipv4_many_flow_into_table(
3258 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3260 /* populate the ipv6 hash */
3261 populate_ipv6_many_flow_into_table(
3262 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3265 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3268 /* populate the ipv4 hash */
3269 populate_ipv4_few_flow_into_table(
3270 ipv4_l3fwd_lookup_struct[socketid]);
3272 /* populate the ipv6 hash */
3273 populate_ipv6_few_flow_into_table(
3274 ipv6_l3fwd_lookup_struct[socketid]);
3280 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3282 setup_lpm(int socketid)
3284 struct rte_lpm6_config config;
3285 struct rte_lpm_config lpm_ipv4_config;
3290 /* create the LPM table */
3291 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3292 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3293 lpm_ipv4_config.number_tbl8s = 256;
3294 lpm_ipv4_config.flags = 0;
3295 ipv4_l3fwd_lookup_struct[socketid] =
3296 rte_lpm_create(s, socketid, &lpm_ipv4_config);
3297 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3298 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3299 " on socket %d\n", socketid);
3301 /* populate the LPM table */
3302 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3304 /* skip unused ports */
3305 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3306 enabled_port_mask) == 0)
3309 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3310 ipv4_l3fwd_route_array[i].ip,
3311 ipv4_l3fwd_route_array[i].depth,
3312 ipv4_l3fwd_route_array[i].if_out);
3315 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3316 "l3fwd LPM table on socket %d\n",
3320 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3321 (unsigned)ipv4_l3fwd_route_array[i].ip,
3322 ipv4_l3fwd_route_array[i].depth,
3323 ipv4_l3fwd_route_array[i].if_out);
3326 /* create the LPM6 table */
3327 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3329 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3330 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3332 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3334 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3335 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3336 " on socket %d\n", socketid);
3338 /* populate the LPM table */
3339 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3341 /* skip unused ports */
3342 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3343 enabled_port_mask) == 0)
3346 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3347 ipv6_l3fwd_route_array[i].ip,
3348 ipv6_l3fwd_route_array[i].depth,
3349 ipv6_l3fwd_route_array[i].if_out);
3352 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3353 "l3fwd LPM table on socket %d\n",
3357 printf("LPM: Adding route %s / %d (%d)\n",
3359 ipv6_l3fwd_route_array[i].depth,
3360 ipv6_l3fwd_route_array[i].if_out);
3366 init_mem(unsigned nb_mbuf)
3368 struct lcore_conf *qconf;
3373 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3374 if (rte_lcore_is_enabled(lcore_id) == 0)
3378 socketid = rte_lcore_to_socket_id(lcore_id);
3382 if (socketid >= NB_SOCKETS) {
3383 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3384 socketid, lcore_id, NB_SOCKETS);
3386 if (pktmbuf_pool[socketid] == NULL) {
3387 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3388 pktmbuf_pool[socketid] =
3389 rte_pktmbuf_pool_create(s, nb_mbuf,
3390 MEMPOOL_CACHE_SIZE, 0,
3391 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3392 if (pktmbuf_pool[socketid] == NULL)
3393 rte_exit(EXIT_FAILURE,
3394 "Cannot init mbuf pool on socket %d\n", socketid);
3396 printf("Allocated mbuf pool on socket %d\n", socketid);
3398 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3399 setup_lpm(socketid);
3401 setup_hash(socketid);
3404 qconf = &lcore_conf[lcore_id];
3405 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3406 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3411 /* Check the link status of all ports in up to 9s, and print them finally */
3413 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
3415 #define CHECK_INTERVAL 100 /* 100ms */
3416 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3418 uint8_t count, all_ports_up, print_flag = 0;
3419 struct rte_eth_link link;
3421 printf("\nChecking link status");
3423 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3425 for (portid = 0; portid < port_num; portid++) {
3426 if ((port_mask & (1 << portid)) == 0)
3428 memset(&link, 0, sizeof(link));
3429 rte_eth_link_get_nowait(portid, &link);
3430 /* print link status if flag set */
3431 if (print_flag == 1) {
3432 if (link.link_status)
3434 "Port%d Link Up. Speed %u Mbps - %s\n",
3435 portid, link.link_speed,
3436 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3437 ("full-duplex") : ("half-duplex\n"));
3439 printf("Port %d Link Down\n", portid);
3442 /* clear all_ports_up flag if any link down */
3443 if (link.link_status == ETH_LINK_DOWN) {
3448 /* after finally printing all link status, get out */
3449 if (print_flag == 1)
3452 if (all_ports_up == 0) {
3455 rte_delay_ms(CHECK_INTERVAL);
3458 /* set the print_flag if all ports up or timeout */
3459 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3467 main(int argc, char **argv)
3469 struct rte_eth_dev_info dev_info;
3470 struct rte_eth_txconf *txconf;
3474 uint16_t queueid, portid;
3476 uint32_t n_tx_queue, nb_lcores;
3477 uint8_t nb_rx_queue, queue, socketid;
3480 ret = rte_eal_init(argc, argv);
3482 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3486 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3487 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3488 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3489 ((uint64_t)portid << 40);
3490 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3493 /* parse application arguments (after the EAL ones) */
3494 ret = parse_args(argc, argv);
3496 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3498 if (check_lcore_params() < 0)
3499 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3501 printf("Initializing rx-queues...\n");
3502 ret = init_rx_queues();
3504 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3506 printf("Initializing tx-threads...\n");
3507 ret = init_tx_threads();
3509 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3511 printf("Initializing rings...\n");
3512 ret = init_rx_rings();
3514 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3516 nb_ports = rte_eth_dev_count();
3518 if (check_port_config(nb_ports) < 0)
3519 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3521 nb_lcores = rte_lcore_count();
3523 /* initialize all ports */
3524 for (portid = 0; portid < nb_ports; portid++) {
3525 /* skip ports that are not enabled */
3526 if ((enabled_port_mask & (1 << portid)) == 0) {
3527 printf("\nSkipping disabled port %d\n", portid);
3532 printf("Initializing port %d ... ", portid);
3535 nb_rx_queue = get_port_n_rx_queues(portid);
3536 n_tx_queue = nb_lcores;
3537 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3538 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3539 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3540 nb_rx_queue, (unsigned)n_tx_queue);
3541 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3542 (uint16_t)n_tx_queue, &port_conf);
3544 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3547 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3550 rte_exit(EXIT_FAILURE,
3551 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3554 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3555 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3557 print_ethaddr("Destination:",
3558 (const struct ether_addr *)&dest_eth_addr[portid]);
3562 * prepare src MACs for each port.
3564 ether_addr_copy(&ports_eth_addr[portid],
3565 (struct ether_addr *)(val_eth + portid) + 1);
3568 ret = init_mem(NB_MBUF);
3570 rte_exit(EXIT_FAILURE, "init_mem failed\n");
3572 /* init one TX queue per couple (lcore,port) */
3574 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3575 if (rte_lcore_is_enabled(lcore_id) == 0)
3579 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3583 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3586 rte_eth_dev_info_get(portid, &dev_info);
3587 txconf = &dev_info.default_txconf;
3588 if (port_conf.rxmode.jumbo_frame)
3589 txconf->txq_flags = 0;
3590 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3593 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3594 "port=%d\n", ret, portid);
3596 tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3602 for (i = 0; i < n_rx_thread; i++) {
3603 lcore_id = rx_thread[i].conf.lcore_id;
3605 if (rte_lcore_is_enabled(lcore_id) == 0) {
3606 rte_exit(EXIT_FAILURE,
3607 "Cannot start Rx thread on lcore %u: lcore disabled\n",
3612 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3616 /* init RX queues */
3617 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3618 portid = rx_thread[i].rx_queue_list[queue].port_id;
3619 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3622 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3626 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3629 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3632 pktmbuf_pool[socketid]);
3634 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3635 "port=%d\n", ret, portid);
3642 for (portid = 0; portid < nb_ports; portid++) {
3643 if ((enabled_port_mask & (1 << portid)) == 0)
3647 ret = rte_eth_dev_start(portid);
3649 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3653 * If enabled, put device in promiscuous mode.
3654 * This allows IO forwarding mode to forward packets
3655 * to itself through 2 cross-connected ports of the
3659 rte_eth_promiscuous_enable(portid);
3662 for (i = 0; i < n_rx_thread; i++) {
3663 lcore_id = rx_thread[i].conf.lcore_id;
3664 if (rte_lcore_is_enabled(lcore_id) == 0)
3667 /* check if hw packet type is supported */
3668 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3669 portid = rx_thread[i].rx_queue_list[queue].port_id;
3670 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3672 if (parse_ptype_on) {
3673 if (!rte_eth_add_rx_callback(portid, queueid,
3674 cb_parse_ptype, NULL))
3675 rte_exit(EXIT_FAILURE,
3676 "Failed to add rx callback: "
3677 "port=%d\n", portid);
3678 } else if (!check_ptype(portid))
3679 rte_exit(EXIT_FAILURE,
3680 "Port %d cannot parse packet type.\n\n"
3681 "Please add --parse-ptype to use sw "
3682 "packet type analyzer.\n\n",
3687 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
3690 printf("Starting L-Threading Model\n");
3692 #if (APP_CPU_LOAD > 0)
3693 if (cpu_load_lcore_id > 0)
3694 /* Use one lcore for cpu load collector */
3698 lthread_num_schedulers_set(nb_lcores);
3699 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3700 lthread_master_spawner(NULL);
3703 printf("Starting P-Threading Model\n");
3704 /* launch per-lcore init on every lcore */
3705 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3706 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3707 if (rte_eal_wait_lcore(lcore_id) < 0)