1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/types.h>
13 #include <sys/queue.h>
18 #include <rte_common.h>
20 #include <rte_byteorder.h>
22 #include <rte_memory.h>
23 #include <rte_memcpy.h>
25 #include <rte_launch.h>
26 #include <rte_atomic.h>
27 #include <rte_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
43 #include <rte_string_fns.h>
44 #include <rte_pause.h>
46 #include <cmdline_parse.h>
47 #include <cmdline_parse_etheraddr.h>
49 #include <lthread_api.h>
51 #define APP_LOOKUP_EXACT_MATCH 0
52 #define APP_LOOKUP_LPM 1
53 #define DO_RFC_1812_CHECKS
55 /* Enable cpu-load stats 0-off, 1-on */
56 #define APP_CPU_LOAD 1
58 #ifndef APP_LOOKUP_METHOD
59 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
62 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
63 #define sched_getcpu() rte_lcore_id()
67 check_ptype(int portid)
70 int ipv4 = 0, ipv6 = 0;
72 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
79 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
81 for (i = 0; i < ret; ++i) {
82 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
84 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
95 parse_ptype(struct rte_mbuf *m)
97 struct ether_hdr *eth_hdr;
98 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
101 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
102 ether_type = eth_hdr->ether_type;
103 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
104 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
105 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
106 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
108 m->packet_type = packet_type;
112 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
113 struct rte_mbuf *pkts[], uint16_t nb_pkts,
114 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
118 for (i = 0; i < nb_pkts; i++)
119 parse_ptype(pkts[i]);
125 * When set to zero, simple forwaring path is eanbled.
126 * When set to one, optimized forwarding path is enabled.
127 * Note that LPM optimisation path uses SSE4.1 instructions.
129 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1
131 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
132 #include <rte_hash.h>
133 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
135 #include <rte_lpm6.h>
137 #error "APP_LOOKUP_METHOD set to incorrect value"
140 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
142 #define MAX_JUMBO_PKT_LEN 9600
144 #define IPV6_ADDR_LEN 16
146 #define MEMPOOL_CACHE_SIZE 256
149 * This expression is used to calculate the number of mbufs needed depending on
150 * user input, taking into account memory for rx and tx hardware rings, cache
151 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
152 * NB_MBUF never goes below a minimum value of 8192
155 #define NB_MBUF RTE_MAX(\
156 (nb_ports*nb_rx_queue*nb_rxd + \
157 nb_ports*nb_lcores*MAX_PKT_BURST + \
158 nb_ports*n_tx_queue*nb_txd + \
159 nb_lcores*MEMPOOL_CACHE_SIZE), \
162 #define MAX_PKT_BURST 32
163 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
166 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
168 #define MAX_TX_BURST (MAX_PKT_BURST / 2)
169 #define BURST_SIZE MAX_TX_BURST
173 /* Configure how many packets ahead to prefetch, when reading packets */
174 #define PREFETCH_OFFSET 3
176 /* Used to mark destination port as 'invalid'. */
177 #define BAD_PORT ((uint16_t)-1)
182 * Configurable number of RX/TX ring descriptors
184 #define RTE_TEST_RX_DESC_DEFAULT 1024
185 #define RTE_TEST_TX_DESC_DEFAULT 1024
186 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
187 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
189 /* ethernet addresses of ports */
190 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
191 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
193 static xmm_t val_eth[RTE_MAX_ETHPORTS];
195 /* replace first 12B of the ethernet header. */
196 #define MASK_ETH 0x3f
198 /* mask of enabled ports */
199 static uint32_t enabled_port_mask;
200 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
201 static int numa_on = 1; /**< NUMA is enabled by default. */
202 static int parse_ptype_on;
204 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
205 static int ipv6; /**< ipv6 is false by default. */
208 #if (APP_CPU_LOAD == 1)
210 #define MAX_CPU RTE_MAX_LCORE
211 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000) /**< Timeout for collecting 5s */
213 #define CPU_PROCESS 0
215 #define MAX_CPU_COUNTER 2
220 uint64_t hits[MAX_CPU_COUNTER][MAX_CPU];
221 } __rte_cache_aligned;
223 static struct cpu_load cpu_load;
224 static int cpu_load_lcore_id = -1;
226 #define SET_CPU_BUSY(thread, counter) \
227 thread->conf.busy[counter] = 1
229 #define SET_CPU_IDLE(thread, counter) \
230 thread->conf.busy[counter] = 0
232 #define IS_CPU_BUSY(thread, counter) \
233 (thread->conf.busy[counter] > 0)
237 #define SET_CPU_BUSY(thread, counter)
238 #define SET_CPU_IDLE(thread, counter)
239 #define IS_CPU_BUSY(thread, counter) 0
245 struct rte_mbuf *m_table[MAX_PKT_BURST];
248 struct lcore_rx_queue {
251 } __rte_cache_aligned;
253 #define MAX_RX_QUEUE_PER_LCORE 16
254 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
255 #define MAX_RX_QUEUE_PER_PORT 128
257 #define MAX_LCORE_PARAMS 1024
258 struct rx_thread_params {
263 } __rte_cache_aligned;
265 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
266 static struct rx_thread_params rx_thread_params_array_default[] = {
278 static struct rx_thread_params *rx_thread_params =
279 rx_thread_params_array_default;
280 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
282 struct tx_thread_params {
285 } __rte_cache_aligned;
287 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
288 static struct tx_thread_params tx_thread_params_array_default[] = {
300 static struct tx_thread_params *tx_thread_params =
301 tx_thread_params_array_default;
302 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
304 static struct rte_eth_conf port_conf = {
306 .mq_mode = ETH_MQ_RX_RSS,
307 .max_rx_pkt_len = ETHER_MAX_LEN,
309 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
314 .rss_hf = ETH_RSS_TCP,
318 .mq_mode = ETH_MQ_TX_NONE,
322 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
324 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
326 #include <rte_hash_crc.h>
327 #define DEFAULT_HASH_FUNC rte_hash_crc
335 } __attribute__((__packed__));
337 union ipv4_5tuple_host {
350 #define XMM_NUM_IN_IPV6_5TUPLE 3
353 uint8_t ip_dst[IPV6_ADDR_LEN];
354 uint8_t ip_src[IPV6_ADDR_LEN];
358 } __attribute__((__packed__));
360 union ipv6_5tuple_host {
365 uint8_t ip_src[IPV6_ADDR_LEN];
366 uint8_t ip_dst[IPV6_ADDR_LEN];
371 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
374 struct ipv4_l3fwd_route {
375 struct ipv4_5tuple key;
379 struct ipv6_l3fwd_route {
380 struct ipv6_5tuple key;
384 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
385 {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1), 101, 11, IPPROTO_TCP}, 0},
386 {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1), 102, 12, IPPROTO_TCP}, 1},
387 {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1), 101, 11, IPPROTO_TCP}, 2},
388 {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1), 102, 12, IPPROTO_TCP}, 3},
391 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
393 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
394 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
396 101, 11, IPPROTO_TCP}, 0},
399 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
400 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
402 102, 12, IPPROTO_TCP}, 1},
405 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
406 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
408 101, 11, IPPROTO_TCP}, 2},
411 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
412 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
414 102, 12, IPPROTO_TCP}, 3},
417 typedef struct rte_hash lookup_struct_t;
418 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
419 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
421 #ifdef RTE_ARCH_X86_64
422 /* default to 4 million hash entries (approx) */
423 #define L3FWD_HASH_ENTRIES (1024*1024*4)
425 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
426 #define L3FWD_HASH_ENTRIES (1024*1024*1)
428 #define HASH_ENTRY_NUMBER_DEFAULT 4
430 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
432 static inline uint32_t
433 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
436 const union ipv4_5tuple_host *k;
442 p = (const uint32_t *)&k->port_src;
444 init_val = rte_hash_crc_4byte(t, init_val);
445 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
446 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
447 init_val = rte_hash_crc_4byte(*p, init_val);
451 static inline uint32_t
452 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
455 const union ipv6_5tuple_host *k;
458 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
459 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
463 p = (const uint32_t *)&k->port_src;
465 ip_src0 = (const uint32_t *) k->ip_src;
466 ip_src1 = (const uint32_t *)(k->ip_src + 4);
467 ip_src2 = (const uint32_t *)(k->ip_src + 8);
468 ip_src3 = (const uint32_t *)(k->ip_src + 12);
469 ip_dst0 = (const uint32_t *) k->ip_dst;
470 ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
471 ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
472 ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
473 init_val = rte_hash_crc_4byte(t, init_val);
474 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
475 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
476 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
477 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
478 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
479 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
480 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
481 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
482 init_val = rte_hash_crc_4byte(*p, init_val);
486 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
487 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
489 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
490 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
494 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
495 struct ipv4_l3fwd_route {
501 struct ipv6_l3fwd_route {
507 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
508 {IPv4(1, 1, 1, 0), 24, 0},
509 {IPv4(2, 1, 1, 0), 24, 1},
510 {IPv4(3, 1, 1, 0), 24, 2},
511 {IPv4(4, 1, 1, 0), 24, 3},
512 {IPv4(5, 1, 1, 0), 24, 4},
513 {IPv4(6, 1, 1, 0), 24, 5},
514 {IPv4(7, 1, 1, 0), 24, 6},
515 {IPv4(8, 1, 1, 0), 24, 7},
518 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
519 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
520 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
521 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
522 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
523 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
524 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
525 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
526 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
529 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
530 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
532 #define IPV4_L3FWD_LPM_MAX_RULES 1024
533 #define IPV6_L3FWD_LPM_MAX_RULES 1024
534 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
536 typedef struct rte_lpm lookup_struct_t;
537 typedef struct rte_lpm6 lookup6_struct_t;
538 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
539 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
543 lookup_struct_t *ipv4_lookup_struct;
544 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
545 lookup6_struct_t *ipv6_lookup_struct;
547 lookup_struct_t *ipv6_lookup_struct;
550 } __rte_cache_aligned;
552 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
553 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
555 #define MAX_RX_QUEUE_PER_THREAD 16
556 #define MAX_TX_PORT_PER_THREAD RTE_MAX_ETHPORTS
557 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
558 #define MAX_RX_QUEUE_PER_PORT 128
560 #define MAX_RX_THREAD 1024
561 #define MAX_TX_THREAD 1024
562 #define MAX_THREAD (MAX_RX_THREAD + MAX_TX_THREAD)
565 * Producers and consumers threads configuration
567 static int lthreads_on = 1; /**< Use lthreads for processing*/
569 rte_atomic16_t rx_counter; /**< Number of spawned rx threads */
570 rte_atomic16_t tx_counter; /**< Number of spawned tx threads */
573 uint16_t lcore_id; /**< Initial lcore for rx thread */
574 uint16_t cpu_id; /**< Cpu id for cpu load stats counter */
575 uint16_t thread_id; /**< Thread ID */
577 #if (APP_CPU_LOAD > 0)
578 int busy[MAX_CPU_COUNTER];
582 struct thread_rx_conf {
583 struct thread_conf conf;
586 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
588 uint16_t n_ring; /**< Number of output rings */
589 struct rte_ring *ring[RTE_MAX_LCORE];
590 struct lthread_cond *ready[RTE_MAX_LCORE];
592 #if (APP_CPU_LOAD > 0)
593 int busy[MAX_CPU_COUNTER];
595 } __rte_cache_aligned;
597 uint16_t n_rx_thread;
598 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
600 struct thread_tx_conf {
601 struct thread_conf conf;
603 uint16_t tx_queue_id[RTE_MAX_LCORE];
604 struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
606 struct rte_ring *ring;
607 struct lthread_cond **ready;
609 } __rte_cache_aligned;
611 uint16_t n_tx_thread;
612 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
614 /* Send burst of packets on an output interface */
616 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
618 struct rte_mbuf **m_table;
622 queueid = qconf->tx_queue_id[port];
623 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
625 ret = rte_eth_tx_burst(port, queueid, m_table, n);
626 if (unlikely(ret < n)) {
628 rte_pktmbuf_free(m_table[ret]);
635 /* Enqueue a single packet, and send burst if queue is filled */
637 send_single_packet(struct rte_mbuf *m, uint16_t port)
640 struct thread_tx_conf *qconf;
643 qconf = (struct thread_tx_conf *)lthread_get_data();
645 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
647 len = qconf->tx_mbufs[port].len;
648 qconf->tx_mbufs[port].m_table[len] = m;
651 /* enough pkts to be sent */
652 if (unlikely(len == MAX_PKT_BURST)) {
653 send_burst(qconf, MAX_PKT_BURST, port);
657 qconf->tx_mbufs[port].len = len;
661 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
662 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
663 static __rte_always_inline void
664 send_packetsx4(uint16_t port,
665 struct rte_mbuf *m[], uint32_t num)
668 struct thread_tx_conf *qconf;
671 qconf = (struct thread_tx_conf *)lthread_get_data();
673 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
675 len = qconf->tx_mbufs[port].len;
678 * If TX buffer for that queue is empty, and we have enough packets,
679 * then send them straightway.
681 if (num >= MAX_TX_BURST && len == 0) {
682 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
683 if (unlikely(n < num)) {
685 rte_pktmbuf_free(m[n]);
692 * Put packets into TX buffer for that queue.
696 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
699 switch (n % FWDSTEP) {
702 qconf->tx_mbufs[port].m_table[len + j] = m[j];
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];
721 /* enough pkts to be sent */
722 if (unlikely(len == MAX_PKT_BURST)) {
724 send_burst(qconf, MAX_PKT_BURST, port);
726 /* copy rest of the packets into the TX buffer. */
729 switch (len % FWDSTEP) {
732 qconf->tx_mbufs[port].m_table[j] = m[n + j];
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];
750 qconf->tx_mbufs[port].len = len;
752 #endif /* APP_LOOKUP_LPM */
754 #ifdef DO_RFC_1812_CHECKS
756 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
758 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
760 * 1. The packet length reported by the Link Layer must be large
761 * enough to hold the minimum length legal IP datagram (20 bytes).
763 if (link_len < sizeof(struct ipv4_hdr))
766 /* 2. The IP checksum must be correct. */
767 /* this is checked in H/W */
770 * 3. The IP version number must be 4. If the version number is not 4
771 * then the packet may be another version of IP, such as IPng or
774 if (((pkt->version_ihl) >> 4) != 4)
777 * 4. The IP header length field must be large enough to hold the
778 * minimum length legal IP datagram (20 bytes = 5 words).
780 if ((pkt->version_ihl & 0xf) < 5)
784 * 5. The IP total length field must be large enough to hold the IP
785 * datagram header, whose length is specified in the IP header length
788 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
795 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
797 static __m128i mask0;
798 static __m128i mask1;
799 static __m128i mask2;
800 static inline uint16_t
801 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
802 lookup_struct_t *ipv4_l3fwd_lookup_struct)
805 union ipv4_5tuple_host key;
807 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
808 __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
809 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
811 key.xmm = _mm_and_si128(data, mask0);
812 /* Find destination port */
813 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
814 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
817 static inline uint16_t
818 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
819 lookup_struct_t *ipv6_l3fwd_lookup_struct)
822 union ipv6_5tuple_host key;
824 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct 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 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 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 ether_hdr *eth_hdr[8];
890 struct 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 ether_hdr *);
897 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
898 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
899 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
900 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
901 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
902 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
903 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
905 /* Handle IPv4 headers.*/
906 ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
907 sizeof(struct ether_hdr));
908 ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
909 sizeof(struct ether_hdr));
910 ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
911 sizeof(struct ether_hdr));
912 ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
913 sizeof(struct ether_hdr));
914 ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
915 sizeof(struct ether_hdr));
916 ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
917 sizeof(struct ether_hdr));
918 ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
919 sizeof(struct ether_hdr));
920 ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
921 sizeof(struct 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 ether_hdr) +
973 offsetof(struct ipv4_hdr, time_to_live)));
974 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
975 sizeof(struct ether_hdr) +
976 offsetof(struct ipv4_hdr, time_to_live)));
977 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
978 sizeof(struct ether_hdr) +
979 offsetof(struct ipv4_hdr, time_to_live)));
980 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
981 sizeof(struct ether_hdr) +
982 offsetof(struct ipv4_hdr, time_to_live)));
983 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
984 sizeof(struct ether_hdr) +
985 offsetof(struct ipv4_hdr, time_to_live)));
986 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
987 sizeof(struct ether_hdr) +
988 offsetof(struct ipv4_hdr, time_to_live)));
989 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
990 sizeof(struct ether_hdr) +
991 offsetof(struct ipv4_hdr, time_to_live)));
992 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
993 sizeof(struct ether_hdr) +
994 offsetof(struct 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 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1076 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1077 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1078 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1079 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1080 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1081 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1082 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 ether_hdr) +
1100 offsetof(struct ipv6_hdr, payload_len)));
1101 __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1102 __m128i *, sizeof(struct ether_hdr) +
1103 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1104 __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1105 __m128i *, sizeof(struct ether_hdr) +
1106 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1108 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1109 key->xmm[1] = tmpdata1;
1110 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1114 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1117 uint16_t dst_port[8];
1118 struct ether_hdr *eth_hdr[8];
1119 union ipv6_5tuple_host key[8];
1121 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1123 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1124 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1125 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1126 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1127 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1128 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1129 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1130 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1132 /* Handle IPv6 headers.*/
1133 ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1134 sizeof(struct ether_hdr));
1135 ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1136 sizeof(struct ether_hdr));
1137 ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1138 sizeof(struct ether_hdr));
1139 ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1140 sizeof(struct ether_hdr));
1141 ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1142 sizeof(struct ether_hdr));
1143 ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1144 sizeof(struct ether_hdr));
1145 ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1146 sizeof(struct ether_hdr));
1147 ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1148 sizeof(struct ether_hdr));
1150 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1151 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1152 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1153 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1154 get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1155 get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1156 get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1157 get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1159 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1160 &key[4], &key[5], &key[6], &key[7]};
1162 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1163 &key_array[0], 4, ret);
1164 dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1165 dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1166 dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1167 dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1168 dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1169 dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1170 dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1171 dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1173 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1174 (enabled_port_mask & 1 << dst_port[0]) == 0)
1175 dst_port[0] = portid;
1176 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1177 (enabled_port_mask & 1 << dst_port[1]) == 0)
1178 dst_port[1] = portid;
1179 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1180 (enabled_port_mask & 1 << dst_port[2]) == 0)
1181 dst_port[2] = portid;
1182 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1183 (enabled_port_mask & 1 << dst_port[3]) == 0)
1184 dst_port[3] = portid;
1185 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1186 (enabled_port_mask & 1 << dst_port[4]) == 0)
1187 dst_port[4] = portid;
1188 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1189 (enabled_port_mask & 1 << dst_port[5]) == 0)
1190 dst_port[5] = portid;
1191 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1192 (enabled_port_mask & 1 << dst_port[6]) == 0)
1193 dst_port[6] = portid;
1194 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1195 (enabled_port_mask & 1 << dst_port[7]) == 0)
1196 dst_port[7] = portid;
1199 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1200 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1201 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1202 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1203 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1204 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1205 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1206 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1209 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1210 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1211 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1212 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1213 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1214 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1215 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1216 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1218 send_single_packet(m[0], dst_port[0]);
1219 send_single_packet(m[1], dst_port[1]);
1220 send_single_packet(m[2], dst_port[2]);
1221 send_single_packet(m[3], dst_port[3]);
1222 send_single_packet(m[4], dst_port[4]);
1223 send_single_packet(m[5], dst_port[5]);
1224 send_single_packet(m[6], dst_port[6]);
1225 send_single_packet(m[7], dst_port[7]);
1228 #endif /* APP_LOOKUP_METHOD */
1230 static __rte_always_inline void
1231 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1233 struct ether_hdr *eth_hdr;
1234 struct ipv4_hdr *ipv4_hdr;
1237 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1239 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1240 /* Handle IPv4 headers.*/
1241 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1242 sizeof(struct ether_hdr));
1244 #ifdef DO_RFC_1812_CHECKS
1245 /* Check to make sure the packet is valid (RFC1812) */
1246 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1247 rte_pktmbuf_free(m);
1252 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1253 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1254 if (dst_port >= RTE_MAX_ETHPORTS ||
1255 (enabled_port_mask & 1 << dst_port) == 0)
1258 #ifdef DO_RFC_1812_CHECKS
1259 /* Update time to live and header checksum */
1260 --(ipv4_hdr->time_to_live);
1261 ++(ipv4_hdr->hdr_checksum);
1264 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1267 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1269 send_single_packet(m, dst_port);
1270 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1271 /* Handle IPv6 headers.*/
1272 struct ipv6_hdr *ipv6_hdr;
1274 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1275 sizeof(struct ether_hdr));
1277 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1278 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1280 if (dst_port >= RTE_MAX_ETHPORTS ||
1281 (enabled_port_mask & 1 << dst_port) == 0)
1285 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1288 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1290 send_single_packet(m, dst_port);
1292 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1293 rte_pktmbuf_free(m);
1296 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1297 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1298 #ifdef DO_RFC_1812_CHECKS
1300 #define IPV4_MIN_VER_IHL 0x45
1301 #define IPV4_MAX_VER_IHL 0x4f
1302 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1304 /* Minimum value of IPV4 total length (20B) in network byte order. */
1305 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1308 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1309 * - The IP version number must be 4.
1310 * - The IP header length field must be large enough to hold the
1311 * minimum length legal IP datagram (20 bytes = 5 words).
1312 * - The IP total length field must be large enough to hold the IP
1313 * datagram header, whose length is specified in the IP header length
1315 * If we encounter invalid IPV4 packet, then set destination port for it
1316 * to BAD_PORT value.
1318 static __rte_always_inline void
1319 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1323 if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1324 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1326 ipv4_hdr->time_to_live--;
1327 ipv4_hdr->hdr_checksum++;
1329 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1330 ((uint8_t)ipv4_hdr->total_length == 0 &&
1331 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1338 #define rfc1812_process(mb, dp, ptype) do { } while (0)
1339 #endif /* DO_RFC_1812_CHECKS */
1340 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1343 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1344 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1346 static __rte_always_inline uint16_t
1347 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1350 struct ipv6_hdr *ipv6_hdr;
1351 struct ether_hdr *eth_hdr;
1353 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1354 return (uint16_t) ((rte_lpm_lookup(
1355 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1356 &next_hop) == 0) ? next_hop : portid);
1358 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1360 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1361 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1363 return (uint16_t) ((rte_lpm6_lookup(
1364 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1365 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1374 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1376 struct ether_hdr *eth_hdr;
1377 struct ipv4_hdr *ipv4_hdr;
1382 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1383 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1385 dst_ipv4 = ipv4_hdr->dst_addr;
1386 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1387 dp = get_dst_port(pkt, dst_ipv4, portid);
1389 te = _mm_load_si128((__m128i *)eth_hdr);
1393 rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1395 te = _mm_blend_epi16(te, ve, MASK_ETH);
1396 _mm_store_si128((__m128i *)eth_hdr, te);
1400 * Read packet_type and destination IPV4 addresses from 4 mbufs.
1403 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1405 uint32_t *ipv4_flag)
1407 struct ipv4_hdr *ipv4_hdr;
1408 struct ether_hdr *eth_hdr;
1409 uint32_t x0, x1, x2, x3;
1411 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1412 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1413 x0 = ipv4_hdr->dst_addr;
1414 ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1416 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1417 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1418 x1 = ipv4_hdr->dst_addr;
1419 ipv4_flag[0] &= pkt[1]->packet_type;
1421 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1422 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1423 x2 = ipv4_hdr->dst_addr;
1424 ipv4_flag[0] &= pkt[2]->packet_type;
1426 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1427 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1428 x3 = ipv4_hdr->dst_addr;
1429 ipv4_flag[0] &= pkt[3]->packet_type;
1431 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1435 * Lookup into LPM for destination port.
1436 * If lookup fails, use incoming port (portid) as destination port.
1439 processx4_step2(__m128i dip,
1442 struct rte_mbuf *pkt[FWDSTEP],
1443 uint16_t dprt[FWDSTEP])
1446 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1447 4, 5, 6, 7, 0, 1, 2, 3);
1449 /* Byte swap 4 IPV4 addresses. */
1450 dip = _mm_shuffle_epi8(dip, bswap_mask);
1452 /* if all 4 packets are IPV4. */
1453 if (likely(ipv4_flag)) {
1454 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1457 /* get rid of unused upper 16 bit for each dport. */
1458 dst.x = _mm_packs_epi32(dst.x, dst.x);
1459 *(uint64_t *)dprt = dst.u64[0];
1462 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1463 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1464 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1465 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1470 * Update source and destination MAC addresses in the ethernet header.
1471 * Perform RFC1812 checks and updates for IPV4 packets.
1474 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1476 __m128i te[FWDSTEP];
1477 __m128i ve[FWDSTEP];
1478 __m128i *p[FWDSTEP];
1480 p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1481 p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1482 p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1483 p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1485 ve[0] = val_eth[dst_port[0]];
1486 te[0] = _mm_load_si128(p[0]);
1488 ve[1] = val_eth[dst_port[1]];
1489 te[1] = _mm_load_si128(p[1]);
1491 ve[2] = val_eth[dst_port[2]];
1492 te[2] = _mm_load_si128(p[2]);
1494 ve[3] = val_eth[dst_port[3]];
1495 te[3] = _mm_load_si128(p[3]);
1497 /* Update first 12 bytes, keep rest bytes intact. */
1498 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1499 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1500 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1501 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1503 _mm_store_si128(p[0], te[0]);
1504 _mm_store_si128(p[1], te[1]);
1505 _mm_store_si128(p[2], te[2]);
1506 _mm_store_si128(p[3], te[3]);
1508 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1509 &dst_port[0], pkt[0]->packet_type);
1510 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1511 &dst_port[1], pkt[1]->packet_type);
1512 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1513 &dst_port[2], pkt[2]->packet_type);
1514 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1515 &dst_port[3], pkt[3]->packet_type);
1519 * We group consecutive packets with the same destionation port into one burst.
1520 * To avoid extra latency this is done together with some other packet
1521 * processing, but after we made a final decision about packet's destination.
1522 * To do this we maintain:
1523 * pnum - array of number of consecutive packets with the same dest port for
1524 * each packet in the input burst.
1525 * lp - pointer to the last updated element in the pnum.
1526 * dlp - dest port value lp corresponds to.
1529 #define GRPSZ (1 << FWDSTEP)
1530 #define GRPMSK (GRPSZ - 1)
1532 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1533 if (likely((dlp) == (dcp)[(idx)])) { \
1536 (dlp) = (dcp)[idx]; \
1537 (lp) = (pn) + (idx); \
1543 * Group consecutive packets with the same destination port in bursts of 4.
1544 * Suppose we have array of destionation ports:
1545 * dst_port[] = {a, b, c, d,, e, ... }
1546 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1547 * We doing 4 comparisons at once and the result is 4 bit mask.
1548 * This mask is used as an index into prebuild array of pnum values.
1550 static inline uint16_t *
1551 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1553 static const struct {
1554 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1555 int32_t idx; /* index for new last updated elemnet. */
1556 uint16_t lpv; /* add value to the last updated element. */
1559 /* 0: a != b, b != c, c != d, d != e */
1560 .pnum = UINT64_C(0x0001000100010001),
1565 /* 1: a == b, b != c, c != d, d != e */
1566 .pnum = UINT64_C(0x0001000100010002),
1571 /* 2: a != b, b == c, c != d, d != e */
1572 .pnum = UINT64_C(0x0001000100020001),
1577 /* 3: a == b, b == c, c != d, d != e */
1578 .pnum = UINT64_C(0x0001000100020003),
1583 /* 4: a != b, b != c, c == d, d != e */
1584 .pnum = UINT64_C(0x0001000200010001),
1589 /* 5: a == b, b != c, c == d, d != e */
1590 .pnum = UINT64_C(0x0001000200010002),
1595 /* 6: a != b, b == c, c == d, d != e */
1596 .pnum = UINT64_C(0x0001000200030001),
1601 /* 7: a == b, b == c, c == d, d != e */
1602 .pnum = UINT64_C(0x0001000200030004),
1607 /* 8: a != b, b != c, c != d, d == e */
1608 .pnum = UINT64_C(0x0002000100010001),
1613 /* 9: a == b, b != c, c != d, d == e */
1614 .pnum = UINT64_C(0x0002000100010002),
1619 /* 0xa: a != b, b == c, c != d, d == e */
1620 .pnum = UINT64_C(0x0002000100020001),
1625 /* 0xb: a == b, b == c, c != d, d == e */
1626 .pnum = UINT64_C(0x0002000100020003),
1631 /* 0xc: a != b, b != c, c == d, d == e */
1632 .pnum = UINT64_C(0x0002000300010001),
1637 /* 0xd: a == b, b != c, c == d, d == e */
1638 .pnum = UINT64_C(0x0002000300010002),
1643 /* 0xe: a != b, b == c, c == d, d == e */
1644 .pnum = UINT64_C(0x0002000300040001),
1649 /* 0xf: a == b, b == c, c == d, d == e */
1650 .pnum = UINT64_C(0x0002000300040005),
1657 uint16_t u16[FWDSTEP + 1];
1659 } *pnum = (void *)pn;
1663 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1664 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1665 v = _mm_movemask_ps((__m128)dp1);
1667 /* update last port counter. */
1668 lp[0] += gptbl[v].lpv;
1670 /* if dest port value has changed. */
1672 pnum->u64 = gptbl[v].pnum;
1673 pnum->u16[FWDSTEP] = 1;
1674 lp = pnum->u16 + gptbl[v].idx;
1680 #endif /* APP_LOOKUP_METHOD */
1683 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1689 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1690 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1694 uint16_t dst_port[MAX_PKT_BURST];
1695 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1696 uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1697 uint16_t pnum[MAX_PKT_BURST + 1];
1701 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1702 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1705 * Send nb_rx - nb_rx%8 packets
1708 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1710 for (j = 0; j < n; j += 8) {
1712 pkts_burst[j]->packet_type &
1713 pkts_burst[j+1]->packet_type &
1714 pkts_burst[j+2]->packet_type &
1715 pkts_burst[j+3]->packet_type &
1716 pkts_burst[j+4]->packet_type &
1717 pkts_burst[j+5]->packet_type &
1718 pkts_burst[j+6]->packet_type &
1719 pkts_burst[j+7]->packet_type;
1720 if (pkt_type & RTE_PTYPE_L3_IPV4) {
1721 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1722 } else if (pkt_type &
1723 RTE_PTYPE_L3_IPV6) {
1724 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1726 l3fwd_simple_forward(pkts_burst[j], portid);
1727 l3fwd_simple_forward(pkts_burst[j+1], portid);
1728 l3fwd_simple_forward(pkts_burst[j+2], portid);
1729 l3fwd_simple_forward(pkts_burst[j+3], portid);
1730 l3fwd_simple_forward(pkts_burst[j+4], portid);
1731 l3fwd_simple_forward(pkts_burst[j+5], portid);
1732 l3fwd_simple_forward(pkts_burst[j+6], portid);
1733 l3fwd_simple_forward(pkts_burst[j+7], portid);
1736 for (; j < nb_rx ; j++)
1737 l3fwd_simple_forward(pkts_burst[j], portid);
1739 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1741 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1742 for (j = 0; j != k; j += FWDSTEP)
1743 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1744 &ipv4_flag[j / FWDSTEP]);
1746 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1747 for (j = 0; j != k; j += FWDSTEP)
1748 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1749 portid, &pkts_burst[j], &dst_port[j]);
1752 * Finish packet processing and group consecutive
1753 * packets with the same destination port.
1755 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1762 processx4_step3(pkts_burst, dst_port);
1764 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1765 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1767 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1768 processx4_step3(&pkts_burst[j], &dst_port[j]);
1772 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1774 dp2 = _mm_loadu_si128(
1775 (__m128i *)&dst_port[j - FWDSTEP + 1]);
1776 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1780 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1782 dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1783 sizeof(dst_port[0]));
1787 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1789 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1790 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1793 * remove values added by the last repeated
1797 dlp = dst_port[j - 1];
1799 /* set dlp and lp to the never used values. */
1801 lp = pnum + MAX_PKT_BURST;
1804 /* Process up to last 3 packets one by one. */
1805 switch (nb_rx % FWDSTEP) {
1807 process_packet(pkts_burst[j], dst_port + j, portid);
1808 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1812 process_packet(pkts_burst[j], dst_port + j, portid);
1813 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1817 process_packet(pkts_burst[j], dst_port + j, portid);
1818 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1823 * Send packets out, through destination port.
1824 * Consecuteve pacekts with the same destination port
1825 * are already grouped together.
1826 * If destination port for the packet equals BAD_PORT,
1827 * then free the packet without sending it out.
1829 for (j = 0; j < nb_rx; j += k) {
1837 if (likely(pn != BAD_PORT))
1838 send_packetsx4(pn, pkts_burst + j, k);
1840 for (m = j; m != j + k; m++)
1841 rte_pktmbuf_free(pkts_burst[m]);
1845 #endif /* APP_LOOKUP_METHOD */
1846 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1848 /* Prefetch first packets */
1849 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1850 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1852 /* Prefetch and forward already prefetched packets */
1853 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1854 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1855 j + PREFETCH_OFFSET], void *));
1856 l3fwd_simple_forward(pkts_burst[j], portid);
1859 /* Forward remaining prefetched packets */
1860 for (; j < nb_rx; j++)
1861 l3fwd_simple_forward(pkts_burst[j], portid);
1863 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1867 #if (APP_CPU_LOAD > 0)
1870 * CPU-load stats collector
1873 cpu_load_collector(__rte_unused void *arg) {
1876 uint64_t prev_tsc, diff_tsc, cur_tsc;
1877 uint64_t total[MAX_CPU] = { 0 };
1878 unsigned min_cpu = MAX_CPU;
1879 unsigned max_cpu = 0;
1884 unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1885 struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1887 struct thread_conf *thread_conf;
1889 const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1890 US_PER_S * CPU_LOAD_TIMEOUT_US;
1894 * Wait for all threads
1897 printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1900 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1903 while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1906 for (i = 0; i < n_rx_thread; i++) {
1908 thread_conf = &rx_thread[i].conf;
1909 cpu_id = thread_conf->cpu_id;
1910 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1912 if (cpu_id > max_cpu)
1914 if (cpu_id < min_cpu)
1917 for (i = 0; i < n_tx_thread; i++) {
1919 thread_conf = &tx_thread[i].conf;
1920 cpu_id = thread_conf->cpu_id;
1921 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1923 if (thread_conf->cpu_id > max_cpu)
1924 max_cpu = thread_conf->cpu_id;
1925 if (thread_conf->cpu_id < min_cpu)
1926 min_cpu = thread_conf->cpu_id;
1932 for (i = min_cpu; i <= max_cpu; i++) {
1933 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1934 for (k = 0; k < n_thread_per_cpu[i]; k++)
1935 if (thread_per_cpu[i][k]->busy[j]) {
1940 cpu_load.hits[j][i]++;
1952 cur_tsc = rte_rdtsc();
1954 diff_tsc = cur_tsc - prev_tsc;
1955 if (unlikely(diff_tsc > interval_tsc)) {
1959 printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1960 n_rx_thread, n_tx_thread);
1962 printf("cpu# proc%% poll%% overhead%%\n\n");
1964 for (i = min_cpu; i <= max_cpu; i++) {
1966 printf("CPU %d:", i);
1967 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1968 printf("%7" PRIu64 "",
1969 cpu_load.hits[j][i] * 100 / cpu_load.counter);
1970 hits += cpu_load.hits[j][i];
1971 cpu_load.hits[j][i] = 0;
1973 printf("%7" PRIu64 "\n",
1974 100 - total[i] * 100 / cpu_load.counter);
1977 cpu_load.counter = 0;
1984 #endif /* APP_CPU_LOAD */
1987 * Null processing lthread loop
1989 * This loop is used to start empty scheduler on lcore.
1992 lthread_null(__rte_unused void *args)
1994 int lcore_id = rte_lcore_id();
1996 RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2001 /* main processing loop */
2003 lthread_tx_per_ring(void *dummy)
2007 struct rte_ring *ring;
2008 struct thread_tx_conf *tx_conf;
2009 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2010 struct lthread_cond *ready;
2012 tx_conf = (struct thread_tx_conf *)dummy;
2013 ring = tx_conf->ring;
2014 ready = *tx_conf->ready;
2016 lthread_set_data((void *)tx_conf);
2019 * Move this lthread to lcore
2021 lthread_set_affinity(tx_conf->conf.lcore_id);
2023 RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2026 rte_atomic16_inc(&tx_counter);
2030 * Read packet from ring
2032 SET_CPU_BUSY(tx_conf, CPU_POLL);
2033 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2034 MAX_PKT_BURST, NULL);
2035 SET_CPU_IDLE(tx_conf, CPU_POLL);
2038 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2039 portid = pkts_burst[0]->port;
2040 process_burst(pkts_burst, nb_rx, portid);
2041 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2044 lthread_cond_wait(ready, 0);
2051 * Main tx-lthreads spawner lthread.
2053 * This lthread is used to spawn one new lthread per ring from producers.
2057 lthread_tx(void *args)
2063 struct thread_tx_conf *tx_conf;
2065 tx_conf = (struct thread_tx_conf *)args;
2066 lthread_set_data((void *)tx_conf);
2069 * Move this lthread to the selected lcore
2071 lthread_set_affinity(tx_conf->conf.lcore_id);
2074 * Spawn tx readers (one per input ring)
2076 lthread_create(<, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2079 lcore_id = rte_lcore_id();
2081 RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2083 tx_conf->conf.cpu_id = sched_getcpu();
2086 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2089 * TX burst queue drain
2091 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2092 if (tx_conf->tx_mbufs[portid].len == 0)
2094 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2095 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2096 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2097 tx_conf->tx_mbufs[portid].len = 0;
2105 lthread_rx(void *dummy)
2113 int len[RTE_MAX_LCORE] = { 0 };
2114 int old_len, new_len;
2115 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2116 struct thread_rx_conf *rx_conf;
2118 rx_conf = (struct thread_rx_conf *)dummy;
2119 lthread_set_data((void *)rx_conf);
2122 * Move this lthread to lcore
2124 lthread_set_affinity(rx_conf->conf.lcore_id);
2126 if (rx_conf->n_rx_queue == 0) {
2127 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2131 RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2133 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2135 portid = rx_conf->rx_queue_list[i].port_id;
2136 queueid = rx_conf->rx_queue_list[i].queue_id;
2137 RTE_LOG(INFO, L3FWD,
2138 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2139 rte_lcore_id(), portid, queueid);
2143 * Init all condition variables (one per rx thread)
2145 for (i = 0; i < rx_conf->n_rx_queue; i++)
2146 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2150 rx_conf->conf.cpu_id = sched_getcpu();
2151 rte_atomic16_inc(&rx_counter);
2155 * Read packet from RX queues
2157 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2158 portid = rx_conf->rx_queue_list[i].port_id;
2159 queueid = rx_conf->rx_queue_list[i].queue_id;
2161 SET_CPU_BUSY(rx_conf, CPU_POLL);
2162 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2164 SET_CPU_IDLE(rx_conf, CPU_POLL);
2167 worker_id = (worker_id + 1) % rx_conf->n_ring;
2168 old_len = len[worker_id];
2170 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2171 ret = rte_ring_sp_enqueue_burst(
2172 rx_conf->ring[worker_id],
2173 (void **) pkts_burst,
2176 new_len = old_len + ret;
2178 if (new_len >= BURST_SIZE) {
2179 lthread_cond_signal(rx_conf->ready[worker_id]);
2183 len[worker_id] = new_len;
2185 if (unlikely(ret < nb_rx)) {
2188 for (k = ret; k < nb_rx; k++) {
2189 struct rte_mbuf *m = pkts_burst[k];
2191 rte_pktmbuf_free(m);
2194 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)
2212 struct lthread *lt[MAX_THREAD];
2216 printf("Entering lthread_spawner\n");
2219 * Create producers (rx threads) on default lcore
2221 for (i = 0; i < n_rx_thread; i++) {
2222 rx_thread[i].conf.thread_id = i;
2223 lthread_create(<[n_thread], -1, lthread_rx,
2224 (void *)&rx_thread[i]);
2229 * Wait for all producers. Until some producers can be started on the same
2230 * scheduler as this lthread, yielding is required to let them to run and
2231 * prevent deadlock here.
2233 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2234 lthread_sleep(100000);
2237 * Create consumers (tx threads) on default lcore_id
2239 for (i = 0; i < n_tx_thread; i++) {
2240 tx_thread[i].conf.thread_id = i;
2241 lthread_create(<[n_thread], -1, lthread_tx,
2242 (void *)&tx_thread[i]);
2247 * Wait for all threads finished
2249 for (i = 0; i < n_thread; i++)
2250 lthread_join(lt[i], NULL);
2256 * Start master scheduler with initial lthread spawning rx and tx lthreads
2257 * (main_lthread_master).
2260 lthread_master_spawner(__rte_unused void *arg) {
2262 int lcore_id = rte_lcore_id();
2264 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2265 lthread_create(<, -1, lthread_spawner, NULL);
2272 * Start scheduler on lcore.
2275 sched_spawner(__rte_unused void *arg) {
2277 int lcore_id = rte_lcore_id();
2280 if (lcore_id == cpu_load_lcore_id) {
2281 cpu_load_collector(arg);
2284 #endif /* APP_CPU_LOAD */
2286 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2287 lthread_create(<, -1, lthread_null, NULL);
2293 /* main processing loop */
2295 pthread_tx(void *dummy)
2297 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2298 uint64_t prev_tsc, diff_tsc, cur_tsc;
2301 struct thread_tx_conf *tx_conf;
2303 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2304 US_PER_S * BURST_TX_DRAIN_US;
2308 tx_conf = (struct thread_tx_conf *)dummy;
2310 RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2312 tx_conf->conf.cpu_id = sched_getcpu();
2313 rte_atomic16_inc(&tx_counter);
2316 cur_tsc = rte_rdtsc();
2319 * TX burst queue drain
2321 diff_tsc = cur_tsc - prev_tsc;
2322 if (unlikely(diff_tsc > drain_tsc)) {
2325 * This could be optimized (use queueid instead of
2326 * portid), but it is not called so often
2328 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2329 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2330 if (tx_conf->tx_mbufs[portid].len == 0)
2332 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2333 tx_conf->tx_mbufs[portid].len = 0;
2335 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2341 * Read packet from ring
2343 SET_CPU_BUSY(tx_conf, CPU_POLL);
2344 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2345 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2346 SET_CPU_IDLE(tx_conf, CPU_POLL);
2348 if (unlikely(nb_rx == 0)) {
2353 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2354 portid = pkts_burst[0]->port;
2355 process_burst(pkts_burst, nb_rx, portid);
2356 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2362 pthread_rx(void *dummy)
2371 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2373 struct thread_rx_conf *rx_conf;
2375 lcore_id = rte_lcore_id();
2376 rx_conf = (struct thread_rx_conf *)dummy;
2378 if (rx_conf->n_rx_queue == 0) {
2379 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2383 RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2385 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2387 portid = rx_conf->rx_queue_list[i].port_id;
2388 queueid = rx_conf->rx_queue_list[i].queue_id;
2389 RTE_LOG(INFO, L3FWD,
2390 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2391 lcore_id, portid, queueid);
2395 rx_conf->conf.cpu_id = sched_getcpu();
2396 rte_atomic16_inc(&rx_counter);
2400 * Read packet from RX queues
2402 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2403 portid = rx_conf->rx_queue_list[i].port_id;
2404 queueid = rx_conf->rx_queue_list[i].queue_id;
2406 SET_CPU_BUSY(rx_conf, CPU_POLL);
2407 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2409 SET_CPU_IDLE(rx_conf, CPU_POLL);
2416 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2417 worker_id = (worker_id + 1) % rx_conf->n_ring;
2418 n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2419 (void **)pkts_burst, nb_rx, NULL);
2421 if (unlikely(n != nb_rx)) {
2424 for (k = n; k < nb_rx; k++) {
2425 struct rte_mbuf *m = pkts_burst[k];
2427 rte_pktmbuf_free(m);
2431 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2441 pthread_run(__rte_unused void *arg) {
2442 int lcore_id = rte_lcore_id();
2445 for (i = 0; i < n_rx_thread; i++)
2446 if (rx_thread[i].conf.lcore_id == lcore_id) {
2447 printf("Start rx thread on %d...\n", lcore_id);
2448 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2449 RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2450 pthread_rx((void *)&rx_thread[i]);
2454 for (i = 0; i < n_tx_thread; i++)
2455 if (tx_thread[i].conf.lcore_id == lcore_id) {
2456 printf("Start tx thread on %d...\n", lcore_id);
2457 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2458 RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2459 pthread_tx((void *)&tx_thread[i]);
2464 if (lcore_id == cpu_load_lcore_id)
2465 cpu_load_collector(arg);
2466 #endif /* APP_CPU_LOAD */
2472 check_lcore_params(void)
2474 uint8_t queue, lcore;
2478 for (i = 0; i < nb_rx_thread_params; ++i) {
2479 queue = rx_thread_params[i].queue_id;
2480 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2481 printf("invalid queue number: %hhu\n", queue);
2484 lcore = rx_thread_params[i].lcore_id;
2485 if (!rte_lcore_is_enabled(lcore)) {
2486 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2489 socketid = rte_lcore_to_socket_id(lcore);
2490 if ((socketid != 0) && (numa_on == 0))
2491 printf("warning: lcore %hhu is on socket %d with numa off\n",
2498 check_port_config(void)
2503 for (i = 0; i < nb_rx_thread_params; ++i) {
2504 portid = rx_thread_params[i].port_id;
2505 if ((enabled_port_mask & (1 << portid)) == 0) {
2506 printf("port %u is not enabled in port mask\n", portid);
2509 if (!rte_eth_dev_is_valid_port(portid)) {
2510 printf("port %u is not present on the board\n", portid);
2518 get_port_n_rx_queues(const uint16_t port)
2523 for (i = 0; i < nb_rx_thread_params; ++i)
2524 if (rx_thread_params[i].port_id == port &&
2525 rx_thread_params[i].queue_id > queue)
2526 queue = rx_thread_params[i].queue_id;
2528 return (uint8_t)(++queue);
2535 struct thread_rx_conf *rx_conf;
2536 struct thread_tx_conf *tx_conf;
2537 unsigned rx_thread_id, tx_thread_id;
2539 struct rte_ring *ring = NULL;
2541 for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2543 tx_conf = &tx_thread[tx_thread_id];
2545 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2546 tx_conf->conf.thread_id);
2548 rx_thread_id = tx_conf->conf.thread_id;
2549 if (rx_thread_id > n_tx_thread) {
2550 printf("connection from tx-thread %u to rx-thread %u fails "
2551 "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2555 rx_conf = &rx_thread[rx_thread_id];
2556 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2558 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2559 socket_io, rx_thread_id, tx_thread_id);
2561 ring = rte_ring_create(name, 1024 * 4, socket_io,
2562 RING_F_SP_ENQ | RING_F_SC_DEQ);
2565 rte_panic("Cannot create ring to connect rx-thread %u "
2566 "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2569 rx_conf->ring[rx_conf->n_ring] = ring;
2571 tx_conf->ring = ring;
2572 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2580 init_rx_queues(void)
2582 uint16_t i, nb_rx_queue;
2587 for (i = 0; i < nb_rx_thread_params; ++i) {
2588 thread = rx_thread_params[i].thread_id;
2589 nb_rx_queue = rx_thread[thread].n_rx_queue;
2591 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2592 printf("error: too many queues (%u) for thread: %u\n",
2593 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2597 rx_thread[thread].conf.thread_id = thread;
2598 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2599 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2600 rx_thread_params[i].port_id;
2601 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2602 rx_thread_params[i].queue_id;
2603 rx_thread[thread].n_rx_queue++;
2605 if (thread >= n_rx_thread)
2606 n_rx_thread = thread + 1;
2613 init_tx_threads(void)
2618 for (i = 0; i < nb_tx_thread_params; ++i) {
2619 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2620 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2628 print_usage(const char *prgname)
2630 printf("%s [EAL options] -- -p PORTMASK -P"
2631 " [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2632 " [--tx (lcore,thread)[,(lcore,thread]]"
2633 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2634 " [--parse-ptype]\n\n"
2635 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2636 " -P : enable promiscuous mode\n"
2637 " --rx (port,queue,lcore,thread): rx queues configuration\n"
2638 " --tx (lcore,thread): tx threads configuration\n"
2639 " --stat-lcore LCORE: use lcore for stat collector\n"
2640 " --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2641 " --no-numa: optional, disable numa awareness\n"
2642 " --ipv6: optional, specify it if running ipv6 packets\n"
2643 " --enable-jumbo: enable jumbo frame"
2644 " which max packet len is PKTLEN in decimal (64-9600)\n"
2645 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2646 " --no-lthreads: turn off lthread model\n"
2647 " --parse-ptype: set to use software to analyze packet type\n\n",
2651 static int parse_max_pkt_len(const char *pktlen)
2656 /* parse decimal string */
2657 len = strtoul(pktlen, &end, 10);
2658 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2668 parse_portmask(const char *portmask)
2673 /* parse hexadecimal string */
2674 pm = strtoul(portmask, &end, 16);
2675 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2684 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2686 parse_hash_entry_number(const char *hash_entry_num)
2689 unsigned long hash_en;
2691 /* parse hexadecimal string */
2692 hash_en = strtoul(hash_entry_num, &end, 16);
2693 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2704 parse_rx_config(const char *q_arg)
2707 const char *p, *p0 = q_arg;
2716 unsigned long int_fld[_NUM_FLD];
2717 char *str_fld[_NUM_FLD];
2721 nb_rx_thread_params = 0;
2723 while ((p = strchr(p0, '(')) != NULL) {
2725 p0 = strchr(p, ')');
2730 if (size >= sizeof(s))
2733 snprintf(s, sizeof(s), "%.*s", size, p);
2734 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2736 for (i = 0; i < _NUM_FLD; i++) {
2738 int_fld[i] = strtoul(str_fld[i], &end, 0);
2739 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2742 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2743 printf("exceeded max number of rx params: %hu\n",
2744 nb_rx_thread_params);
2747 rx_thread_params_array[nb_rx_thread_params].port_id =
2749 rx_thread_params_array[nb_rx_thread_params].queue_id =
2750 (uint8_t)int_fld[FLD_QUEUE];
2751 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2752 (uint8_t)int_fld[FLD_LCORE];
2753 rx_thread_params_array[nb_rx_thread_params].thread_id =
2754 (uint8_t)int_fld[FLD_THREAD];
2755 ++nb_rx_thread_params;
2757 rx_thread_params = rx_thread_params_array;
2762 parse_tx_config(const char *q_arg)
2765 const char *p, *p0 = q_arg;
2772 unsigned long int_fld[_NUM_FLD];
2773 char *str_fld[_NUM_FLD];
2777 nb_tx_thread_params = 0;
2779 while ((p = strchr(p0, '(')) != NULL) {
2781 p0 = strchr(p, ')');
2786 if (size >= sizeof(s))
2789 snprintf(s, sizeof(s), "%.*s", size, p);
2790 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2792 for (i = 0; i < _NUM_FLD; i++) {
2794 int_fld[i] = strtoul(str_fld[i], &end, 0);
2795 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2798 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2799 printf("exceeded max number of tx params: %hu\n",
2800 nb_tx_thread_params);
2803 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2804 (uint8_t)int_fld[FLD_LCORE];
2805 tx_thread_params_array[nb_tx_thread_params].thread_id =
2806 (uint8_t)int_fld[FLD_THREAD];
2807 ++nb_tx_thread_params;
2809 tx_thread_params = tx_thread_params_array;
2814 #if (APP_CPU_LOAD > 0)
2816 parse_stat_lcore(const char *stat_lcore)
2819 unsigned long lcore_id;
2821 lcore_id = strtoul(stat_lcore, &end, 10);
2822 if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2830 parse_eth_dest(const char *optarg)
2834 uint8_t c, *dest, peer_addr[6];
2837 portid = strtoul(optarg, &port_end, 10);
2838 if (errno != 0 || port_end == optarg || *port_end++ != ',')
2839 rte_exit(EXIT_FAILURE,
2840 "Invalid eth-dest: %s", optarg);
2841 if (portid >= RTE_MAX_ETHPORTS)
2842 rte_exit(EXIT_FAILURE,
2843 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2844 portid, RTE_MAX_ETHPORTS);
2846 if (cmdline_parse_etheraddr(NULL, port_end,
2847 &peer_addr, sizeof(peer_addr)) < 0)
2848 rte_exit(EXIT_FAILURE,
2849 "Invalid ethernet address: %s\n",
2851 dest = (uint8_t *)&dest_eth_addr[portid];
2852 for (c = 0; c < 6; c++)
2853 dest[c] = peer_addr[c];
2854 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2857 #define CMD_LINE_OPT_RX_CONFIG "rx"
2858 #define CMD_LINE_OPT_TX_CONFIG "tx"
2859 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2860 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2861 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2862 #define CMD_LINE_OPT_IPV6 "ipv6"
2863 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2864 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2865 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2866 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2868 /* Parse the argument given in the command line of the application */
2870 parse_args(int argc, char **argv)
2875 char *prgname = argv[0];
2876 static struct option lgopts[] = {
2877 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2878 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2879 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2880 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2881 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2882 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2883 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2884 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2885 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2886 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2892 while ((opt = getopt_long(argc, argvopt, "p:P",
2893 lgopts, &option_index)) != EOF) {
2898 enabled_port_mask = parse_portmask(optarg);
2899 if (enabled_port_mask == 0) {
2900 printf("invalid portmask\n");
2901 print_usage(prgname);
2906 printf("Promiscuous mode selected\n");
2912 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2913 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2914 ret = parse_rx_config(optarg);
2916 printf("invalid rx-config\n");
2917 print_usage(prgname);
2922 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2923 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2924 ret = parse_tx_config(optarg);
2926 printf("invalid tx-config\n");
2927 print_usage(prgname);
2932 #if (APP_CPU_LOAD > 0)
2933 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2934 sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2935 cpu_load_lcore_id = parse_stat_lcore(optarg);
2939 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2940 sizeof(CMD_LINE_OPT_ETH_DEST)))
2941 parse_eth_dest(optarg);
2943 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2944 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2945 printf("numa is disabled\n");
2949 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2950 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2951 sizeof(CMD_LINE_OPT_IPV6))) {
2952 printf("ipv6 is specified\n");
2957 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2958 sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2959 printf("l-threads model is disabled\n");
2963 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2964 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2965 printf("software packet type parsing enabled\n");
2969 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2970 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2971 struct option lenopts = {"max-pkt-len", required_argument, 0,
2974 printf("jumbo frame is enabled - disabling simple TX path\n");
2975 port_conf.rxmode.offloads |=
2976 DEV_RX_OFFLOAD_JUMBO_FRAME;
2977 port_conf.txmode.offloads |=
2978 DEV_TX_OFFLOAD_MULTI_SEGS;
2980 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2981 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2984 ret = parse_max_pkt_len(optarg);
2985 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2986 printf("invalid packet length\n");
2987 print_usage(prgname);
2990 port_conf.rxmode.max_rx_pkt_len = ret;
2992 printf("set jumbo frame max packet length to %u\n",
2993 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2995 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2996 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2997 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2998 ret = parse_hash_entry_number(optarg);
2999 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3000 hash_entry_number = ret;
3002 printf("invalid hash entry number\n");
3003 print_usage(prgname);
3011 print_usage(prgname);
3017 argv[optind-1] = prgname;
3020 optind = 1; /* reset getopt lib */
3025 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
3027 char buf[ETHER_ADDR_FMT_SIZE];
3029 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
3030 printf("%s%s", name, buf);
3033 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3035 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3036 union ipv4_5tuple_host *key2)
3038 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3039 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3040 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3041 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3042 key2->proto = key1->proto;
3047 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3048 union ipv6_5tuple_host *key2)
3052 for (i = 0; i < 16; i++) {
3053 key2->ip_dst[i] = key1->ip_dst[i];
3054 key2->ip_src[i] = key1->ip_src[i];
3056 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3057 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3058 key2->proto = key1->proto;
3064 #define BYTE_VALUE_MAX 256
3065 #define ALL_32_BITS 0xffffffff
3066 #define BIT_8_TO_15 0x0000ff00
3068 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3072 uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3074 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3075 for (i = 0; i < array_len; i++) {
3076 struct ipv4_l3fwd_route entry;
3077 union ipv4_5tuple_host newkey;
3079 entry = ipv4_l3fwd_route_array[i];
3080 convert_ipv4_5tuple(&entry.key, &newkey);
3081 ret = rte_hash_add_key(h, (void *)&newkey);
3083 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3084 " to the l3fwd hash.\n", i);
3086 ipv4_l3fwd_out_if[ret] = entry.if_out;
3088 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3091 #define BIT_16_TO_23 0x00ff0000
3093 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3097 uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3099 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3100 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3101 for (i = 0; i < array_len; i++) {
3102 struct ipv6_l3fwd_route entry;
3103 union ipv6_5tuple_host newkey;
3105 entry = ipv6_l3fwd_route_array[i];
3106 convert_ipv6_5tuple(&entry.key, &newkey);
3107 ret = rte_hash_add_key(h, (void *)&newkey);
3109 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3110 " to the l3fwd hash.\n", i);
3112 ipv6_l3fwd_out_if[ret] = entry.if_out;
3114 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3117 #define NUMBER_PORT_USED 4
3119 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3120 unsigned int nr_flow)
3124 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3126 for (i = 0; i < nr_flow; i++) {
3127 struct ipv4_l3fwd_route entry;
3128 union ipv4_5tuple_host newkey;
3129 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3130 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3132 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3134 /* Create the ipv4 exact match flow */
3135 memset(&entry, 0, sizeof(entry));
3136 switch (i & (NUMBER_PORT_USED - 1)) {
3138 entry = ipv4_l3fwd_route_array[0];
3139 entry.key.ip_dst = IPv4(101, c, b, a);
3142 entry = ipv4_l3fwd_route_array[1];
3143 entry.key.ip_dst = IPv4(201, c, b, a);
3146 entry = ipv4_l3fwd_route_array[2];
3147 entry.key.ip_dst = IPv4(111, c, b, a);
3150 entry = ipv4_l3fwd_route_array[3];
3151 entry.key.ip_dst = IPv4(211, c, b, a);
3154 convert_ipv4_5tuple(&entry.key, &newkey);
3155 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3158 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3160 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3163 printf("Hash: Adding 0x%x keys\n", nr_flow);
3167 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3168 unsigned int nr_flow)
3172 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3173 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3174 for (i = 0; i < nr_flow; i++) {
3175 struct ipv6_l3fwd_route entry;
3176 union ipv6_5tuple_host newkey;
3178 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3179 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3181 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3184 /* Create the ipv6 exact match flow */
3185 memset(&entry, 0, sizeof(entry));
3186 switch (i & (NUMBER_PORT_USED - 1)) {
3188 entry = ipv6_l3fwd_route_array[0];
3191 entry = ipv6_l3fwd_route_array[1];
3194 entry = ipv6_l3fwd_route_array[2];
3197 entry = ipv6_l3fwd_route_array[3];
3200 entry.key.ip_dst[13] = c;
3201 entry.key.ip_dst[14] = b;
3202 entry.key.ip_dst[15] = a;
3203 convert_ipv6_5tuple(&entry.key, &newkey);
3204 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3207 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3209 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3212 printf("Hash: Adding 0x%x keys\n", nr_flow);
3216 setup_hash(int socketid)
3218 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3220 .entries = L3FWD_HASH_ENTRIES,
3221 .key_len = sizeof(union ipv4_5tuple_host),
3222 .hash_func = ipv4_hash_crc,
3223 .hash_func_init_val = 0,
3226 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3228 .entries = L3FWD_HASH_ENTRIES,
3229 .key_len = sizeof(union ipv6_5tuple_host),
3230 .hash_func = ipv6_hash_crc,
3231 .hash_func_init_val = 0,
3236 /* create ipv4 hash */
3237 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3238 ipv4_l3fwd_hash_params.name = s;
3239 ipv4_l3fwd_hash_params.socket_id = socketid;
3240 ipv4_l3fwd_lookup_struct[socketid] =
3241 rte_hash_create(&ipv4_l3fwd_hash_params);
3242 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3243 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3244 "socket %d\n", socketid);
3246 /* create ipv6 hash */
3247 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3248 ipv6_l3fwd_hash_params.name = s;
3249 ipv6_l3fwd_hash_params.socket_id = socketid;
3250 ipv6_l3fwd_lookup_struct[socketid] =
3251 rte_hash_create(&ipv6_l3fwd_hash_params);
3252 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3253 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3254 "socket %d\n", socketid);
3256 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3257 /* For testing hash matching with a large number of flows we
3258 * generate millions of IP 5-tuples with an incremented dst
3259 * address to initialize the hash table. */
3261 /* populate the ipv4 hash */
3262 populate_ipv4_many_flow_into_table(
3263 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3265 /* populate the ipv6 hash */
3266 populate_ipv6_many_flow_into_table(
3267 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3270 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3273 /* populate the ipv4 hash */
3274 populate_ipv4_few_flow_into_table(
3275 ipv4_l3fwd_lookup_struct[socketid]);
3277 /* populate the ipv6 hash */
3278 populate_ipv6_few_flow_into_table(
3279 ipv6_l3fwd_lookup_struct[socketid]);
3285 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3287 setup_lpm(int socketid)
3289 struct rte_lpm6_config config;
3290 struct rte_lpm_config lpm_ipv4_config;
3295 /* create the LPM table */
3296 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3297 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3298 lpm_ipv4_config.number_tbl8s = 256;
3299 lpm_ipv4_config.flags = 0;
3300 ipv4_l3fwd_lookup_struct[socketid] =
3301 rte_lpm_create(s, socketid, &lpm_ipv4_config);
3302 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3303 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3304 " on socket %d\n", socketid);
3306 /* populate the LPM table */
3307 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3309 /* skip unused ports */
3310 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3311 enabled_port_mask) == 0)
3314 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3315 ipv4_l3fwd_route_array[i].ip,
3316 ipv4_l3fwd_route_array[i].depth,
3317 ipv4_l3fwd_route_array[i].if_out);
3320 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3321 "l3fwd LPM table on socket %d\n",
3325 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3326 (unsigned)ipv4_l3fwd_route_array[i].ip,
3327 ipv4_l3fwd_route_array[i].depth,
3328 ipv4_l3fwd_route_array[i].if_out);
3331 /* create the LPM6 table */
3332 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3334 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3335 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3337 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3339 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3340 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3341 " on socket %d\n", socketid);
3343 /* populate the LPM table */
3344 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3346 /* skip unused ports */
3347 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3348 enabled_port_mask) == 0)
3351 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3352 ipv6_l3fwd_route_array[i].ip,
3353 ipv6_l3fwd_route_array[i].depth,
3354 ipv6_l3fwd_route_array[i].if_out);
3357 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3358 "l3fwd LPM table on socket %d\n",
3362 printf("LPM: Adding route %s / %d (%d)\n",
3364 ipv6_l3fwd_route_array[i].depth,
3365 ipv6_l3fwd_route_array[i].if_out);
3371 init_mem(unsigned nb_mbuf)
3373 struct lcore_conf *qconf;
3378 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3379 if (rte_lcore_is_enabled(lcore_id) == 0)
3383 socketid = rte_lcore_to_socket_id(lcore_id);
3387 if (socketid >= NB_SOCKETS) {
3388 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3389 socketid, lcore_id, NB_SOCKETS);
3391 if (pktmbuf_pool[socketid] == NULL) {
3392 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3393 pktmbuf_pool[socketid] =
3394 rte_pktmbuf_pool_create(s, nb_mbuf,
3395 MEMPOOL_CACHE_SIZE, 0,
3396 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3397 if (pktmbuf_pool[socketid] == NULL)
3398 rte_exit(EXIT_FAILURE,
3399 "Cannot init mbuf pool on socket %d\n", socketid);
3401 printf("Allocated mbuf pool on socket %d\n", socketid);
3403 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3404 setup_lpm(socketid);
3406 setup_hash(socketid);
3409 qconf = &lcore_conf[lcore_id];
3410 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3411 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3416 /* Check the link status of all ports in up to 9s, and print them finally */
3418 check_all_ports_link_status(uint32_t port_mask)
3420 #define CHECK_INTERVAL 100 /* 100ms */
3421 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3423 uint8_t count, all_ports_up, print_flag = 0;
3424 struct rte_eth_link link;
3426 printf("\nChecking link status");
3428 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3430 RTE_ETH_FOREACH_DEV(portid) {
3431 if ((port_mask & (1 << portid)) == 0)
3433 memset(&link, 0, sizeof(link));
3434 rte_eth_link_get_nowait(portid, &link);
3435 /* print link status if flag set */
3436 if (print_flag == 1) {
3437 if (link.link_status)
3439 "Port%d Link Up. Speed %u Mbps - %s\n",
3440 portid, link.link_speed,
3441 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3442 ("full-duplex") : ("half-duplex\n"));
3444 printf("Port %d Link Down\n", portid);
3447 /* clear all_ports_up flag if any link down */
3448 if (link.link_status == ETH_LINK_DOWN) {
3453 /* after finally printing all link status, get out */
3454 if (print_flag == 1)
3457 if (all_ports_up == 0) {
3460 rte_delay_ms(CHECK_INTERVAL);
3463 /* set the print_flag if all ports up or timeout */
3464 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3472 main(int argc, char **argv)
3474 struct rte_eth_dev_info dev_info;
3475 struct rte_eth_txconf *txconf;
3479 uint16_t queueid, portid;
3481 uint32_t n_tx_queue, nb_lcores;
3482 uint8_t nb_rx_queue, queue, socketid;
3485 ret = rte_eal_init(argc, argv);
3487 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3491 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3492 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3493 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3494 ((uint64_t)portid << 40);
3495 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3498 /* parse application arguments (after the EAL ones) */
3499 ret = parse_args(argc, argv);
3501 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3503 if (check_lcore_params() < 0)
3504 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3506 printf("Initializing rx-queues...\n");
3507 ret = init_rx_queues();
3509 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3511 printf("Initializing tx-threads...\n");
3512 ret = init_tx_threads();
3514 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3516 printf("Initializing rings...\n");
3517 ret = init_rx_rings();
3519 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3521 nb_ports = rte_eth_dev_count_avail();
3523 if (check_port_config() < 0)
3524 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3526 nb_lcores = rte_lcore_count();
3528 /* initialize all ports */
3529 RTE_ETH_FOREACH_DEV(portid) {
3530 struct rte_eth_conf local_port_conf = port_conf;
3532 /* skip ports that are not enabled */
3533 if ((enabled_port_mask & (1 << portid)) == 0) {
3534 printf("\nSkipping disabled port %d\n", portid);
3539 printf("Initializing port %d ... ", portid);
3542 nb_rx_queue = get_port_n_rx_queues(portid);
3543 n_tx_queue = nb_lcores;
3544 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3545 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3546 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3547 nb_rx_queue, (unsigned)n_tx_queue);
3548 rte_eth_dev_info_get(portid, &dev_info);
3549 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3550 local_port_conf.txmode.offloads |=
3551 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3553 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3554 dev_info.flow_type_rss_offloads;
3555 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3556 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3557 printf("Port %u modified RSS hash function based on hardware support,"
3558 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3560 port_conf.rx_adv_conf.rss_conf.rss_hf,
3561 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3564 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3565 (uint16_t)n_tx_queue, &local_port_conf);
3567 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3570 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3573 rte_exit(EXIT_FAILURE,
3574 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3577 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3578 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3580 print_ethaddr("Destination:",
3581 (const struct ether_addr *)&dest_eth_addr[portid]);
3585 * prepare src MACs for each port.
3587 ether_addr_copy(&ports_eth_addr[portid],
3588 (struct ether_addr *)(val_eth + portid) + 1);
3591 ret = init_mem(NB_MBUF);
3593 rte_exit(EXIT_FAILURE, "init_mem failed\n");
3595 /* init one TX queue per couple (lcore,port) */
3597 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3598 if (rte_lcore_is_enabled(lcore_id) == 0)
3602 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3606 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3609 txconf = &dev_info.default_txconf;
3610 txconf->offloads = local_port_conf.txmode.offloads;
3611 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3614 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3615 "port=%d\n", ret, portid);
3617 tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3623 for (i = 0; i < n_rx_thread; i++) {
3624 lcore_id = rx_thread[i].conf.lcore_id;
3626 if (rte_lcore_is_enabled(lcore_id) == 0) {
3627 rte_exit(EXIT_FAILURE,
3628 "Cannot start Rx thread on lcore %u: lcore disabled\n",
3633 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3637 /* init RX queues */
3638 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3639 struct rte_eth_dev *dev;
3640 struct rte_eth_conf *conf;
3641 struct rte_eth_rxconf rxq_conf;
3643 portid = rx_thread[i].rx_queue_list[queue].port_id;
3644 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3645 dev = &rte_eth_devices[portid];
3646 conf = &dev->data->dev_conf;
3649 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3653 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3656 rte_eth_dev_info_get(portid, &dev_info);
3657 rxq_conf = dev_info.default_rxconf;
3658 rxq_conf.offloads = conf->rxmode.offloads;
3659 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3662 pktmbuf_pool[socketid]);
3664 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3665 "port=%d\n", ret, portid);
3672 RTE_ETH_FOREACH_DEV(portid) {
3673 if ((enabled_port_mask & (1 << portid)) == 0)
3677 ret = rte_eth_dev_start(portid);
3679 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3683 * If enabled, put device in promiscuous mode.
3684 * This allows IO forwarding mode to forward packets
3685 * to itself through 2 cross-connected ports of the
3689 rte_eth_promiscuous_enable(portid);
3692 for (i = 0; i < n_rx_thread; i++) {
3693 lcore_id = rx_thread[i].conf.lcore_id;
3694 if (rte_lcore_is_enabled(lcore_id) == 0)
3697 /* check if hw packet type is supported */
3698 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3699 portid = rx_thread[i].rx_queue_list[queue].port_id;
3700 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3702 if (parse_ptype_on) {
3703 if (!rte_eth_add_rx_callback(portid, queueid,
3704 cb_parse_ptype, NULL))
3705 rte_exit(EXIT_FAILURE,
3706 "Failed to add rx callback: "
3707 "port=%d\n", portid);
3708 } else if (!check_ptype(portid))
3709 rte_exit(EXIT_FAILURE,
3710 "Port %d cannot parse packet type.\n\n"
3711 "Please add --parse-ptype to use sw "
3712 "packet type analyzer.\n\n",
3717 check_all_ports_link_status(enabled_port_mask);
3720 printf("Starting L-Threading Model\n");
3722 #if (APP_CPU_LOAD > 0)
3723 if (cpu_load_lcore_id > 0)
3724 /* Use one lcore for cpu load collector */
3728 lthread_num_schedulers_set(nb_lcores);
3729 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3730 lthread_master_spawner(NULL);
3733 printf("Starting P-Threading Model\n");
3734 /* launch per-lcore init on every lcore */
3735 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3736 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3737 if (rte_eal_wait_lcore(lcore_id) < 0)