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 .ignore_offload_bitfield = 1,
310 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
311 DEV_RX_OFFLOAD_CRC_STRIP),
316 .rss_hf = ETH_RSS_TCP,
320 .mq_mode = ETH_MQ_TX_NONE,
324 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
326 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
328 #include <rte_hash_crc.h>
329 #define DEFAULT_HASH_FUNC rte_hash_crc
337 } __attribute__((__packed__));
339 union ipv4_5tuple_host {
352 #define XMM_NUM_IN_IPV6_5TUPLE 3
355 uint8_t ip_dst[IPV6_ADDR_LEN];
356 uint8_t ip_src[IPV6_ADDR_LEN];
360 } __attribute__((__packed__));
362 union ipv6_5tuple_host {
367 uint8_t ip_src[IPV6_ADDR_LEN];
368 uint8_t ip_dst[IPV6_ADDR_LEN];
373 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
376 struct ipv4_l3fwd_route {
377 struct ipv4_5tuple key;
381 struct ipv6_l3fwd_route {
382 struct ipv6_5tuple key;
386 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
387 {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1), 101, 11, IPPROTO_TCP}, 0},
388 {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1), 102, 12, IPPROTO_TCP}, 1},
389 {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1), 101, 11, IPPROTO_TCP}, 2},
390 {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1), 102, 12, IPPROTO_TCP}, 3},
393 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
395 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
396 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
398 101, 11, IPPROTO_TCP}, 0},
401 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
402 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
404 102, 12, IPPROTO_TCP}, 1},
407 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
408 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
410 101, 11, IPPROTO_TCP}, 2},
413 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
414 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
416 102, 12, IPPROTO_TCP}, 3},
419 typedef struct rte_hash lookup_struct_t;
420 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
421 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
423 #ifdef RTE_ARCH_X86_64
424 /* default to 4 million hash entries (approx) */
425 #define L3FWD_HASH_ENTRIES (1024*1024*4)
427 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
428 #define L3FWD_HASH_ENTRIES (1024*1024*1)
430 #define HASH_ENTRY_NUMBER_DEFAULT 4
432 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
434 static inline uint32_t
435 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
438 const union ipv4_5tuple_host *k;
444 p = (const uint32_t *)&k->port_src;
446 init_val = rte_hash_crc_4byte(t, init_val);
447 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
448 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
449 init_val = rte_hash_crc_4byte(*p, init_val);
453 static inline uint32_t
454 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
457 const union ipv6_5tuple_host *k;
460 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
461 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
465 p = (const uint32_t *)&k->port_src;
467 ip_src0 = (const uint32_t *) k->ip_src;
468 ip_src1 = (const uint32_t *)(k->ip_src + 4);
469 ip_src2 = (const uint32_t *)(k->ip_src + 8);
470 ip_src3 = (const uint32_t *)(k->ip_src + 12);
471 ip_dst0 = (const uint32_t *) k->ip_dst;
472 ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
473 ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
474 ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
475 init_val = rte_hash_crc_4byte(t, init_val);
476 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
477 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
478 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
479 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
480 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
481 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
482 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
483 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
484 init_val = rte_hash_crc_4byte(*p, init_val);
488 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
489 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
491 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
492 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
496 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
497 struct ipv4_l3fwd_route {
503 struct ipv6_l3fwd_route {
509 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
510 {IPv4(1, 1, 1, 0), 24, 0},
511 {IPv4(2, 1, 1, 0), 24, 1},
512 {IPv4(3, 1, 1, 0), 24, 2},
513 {IPv4(4, 1, 1, 0), 24, 3},
514 {IPv4(5, 1, 1, 0), 24, 4},
515 {IPv4(6, 1, 1, 0), 24, 5},
516 {IPv4(7, 1, 1, 0), 24, 6},
517 {IPv4(8, 1, 1, 0), 24, 7},
520 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
521 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
522 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
523 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
524 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
525 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
526 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
527 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
528 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
531 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
532 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
534 #define IPV4_L3FWD_LPM_MAX_RULES 1024
535 #define IPV6_L3FWD_LPM_MAX_RULES 1024
536 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
538 typedef struct rte_lpm lookup_struct_t;
539 typedef struct rte_lpm6 lookup6_struct_t;
540 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
541 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
545 lookup_struct_t *ipv4_lookup_struct;
546 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
547 lookup6_struct_t *ipv6_lookup_struct;
549 lookup_struct_t *ipv6_lookup_struct;
552 } __rte_cache_aligned;
554 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
555 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
557 #define MAX_RX_QUEUE_PER_THREAD 16
558 #define MAX_TX_PORT_PER_THREAD RTE_MAX_ETHPORTS
559 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
560 #define MAX_RX_QUEUE_PER_PORT 128
562 #define MAX_RX_THREAD 1024
563 #define MAX_TX_THREAD 1024
564 #define MAX_THREAD (MAX_RX_THREAD + MAX_TX_THREAD)
567 * Producers and consumers threads configuration
569 static int lthreads_on = 1; /**< Use lthreads for processing*/
571 rte_atomic16_t rx_counter; /**< Number of spawned rx threads */
572 rte_atomic16_t tx_counter; /**< Number of spawned tx threads */
575 uint16_t lcore_id; /**< Initial lcore for rx thread */
576 uint16_t cpu_id; /**< Cpu id for cpu load stats counter */
577 uint16_t thread_id; /**< Thread ID */
579 #if (APP_CPU_LOAD > 0)
580 int busy[MAX_CPU_COUNTER];
584 struct thread_rx_conf {
585 struct thread_conf conf;
588 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
590 uint16_t n_ring; /**< Number of output rings */
591 struct rte_ring *ring[RTE_MAX_LCORE];
592 struct lthread_cond *ready[RTE_MAX_LCORE];
594 #if (APP_CPU_LOAD > 0)
595 int busy[MAX_CPU_COUNTER];
597 } __rte_cache_aligned;
599 uint16_t n_rx_thread;
600 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
602 struct thread_tx_conf {
603 struct thread_conf conf;
605 uint16_t tx_queue_id[RTE_MAX_LCORE];
606 struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
608 struct rte_ring *ring;
609 struct lthread_cond **ready;
611 } __rte_cache_aligned;
613 uint16_t n_tx_thread;
614 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
616 /* Send burst of packets on an output interface */
618 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
620 struct rte_mbuf **m_table;
624 queueid = qconf->tx_queue_id[port];
625 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
627 ret = rte_eth_tx_burst(port, queueid, m_table, n);
628 if (unlikely(ret < n)) {
630 rte_pktmbuf_free(m_table[ret]);
637 /* Enqueue a single packet, and send burst if queue is filled */
639 send_single_packet(struct rte_mbuf *m, uint16_t port)
642 struct thread_tx_conf *qconf;
645 qconf = (struct thread_tx_conf *)lthread_get_data();
647 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
649 len = qconf->tx_mbufs[port].len;
650 qconf->tx_mbufs[port].m_table[len] = m;
653 /* enough pkts to be sent */
654 if (unlikely(len == MAX_PKT_BURST)) {
655 send_burst(qconf, MAX_PKT_BURST, port);
659 qconf->tx_mbufs[port].len = len;
663 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
664 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
665 static __rte_always_inline void
666 send_packetsx4(uint16_t port,
667 struct rte_mbuf *m[], uint32_t num)
670 struct thread_tx_conf *qconf;
673 qconf = (struct thread_tx_conf *)lthread_get_data();
675 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
677 len = qconf->tx_mbufs[port].len;
680 * If TX buffer for that queue is empty, and we have enough packets,
681 * then send them straightway.
683 if (num >= MAX_TX_BURST && len == 0) {
684 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
685 if (unlikely(n < num)) {
687 rte_pktmbuf_free(m[n]);
694 * Put packets into TX buffer for that queue.
698 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
701 switch (n % FWDSTEP) {
704 qconf->tx_mbufs[port].m_table[len + j] = m[j];
708 qconf->tx_mbufs[port].m_table[len + j] = m[j];
712 qconf->tx_mbufs[port].m_table[len + j] = m[j];
716 qconf->tx_mbufs[port].m_table[len + j] = m[j];
723 /* enough pkts to be sent */
724 if (unlikely(len == MAX_PKT_BURST)) {
726 send_burst(qconf, MAX_PKT_BURST, port);
728 /* copy rest of the packets into the TX buffer. */
731 switch (len % FWDSTEP) {
734 qconf->tx_mbufs[port].m_table[j] = m[n + j];
738 qconf->tx_mbufs[port].m_table[j] = m[n + j];
742 qconf->tx_mbufs[port].m_table[j] = m[n + j];
746 qconf->tx_mbufs[port].m_table[j] = m[n + j];
752 qconf->tx_mbufs[port].len = len;
754 #endif /* APP_LOOKUP_LPM */
756 #ifdef DO_RFC_1812_CHECKS
758 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
760 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
762 * 1. The packet length reported by the Link Layer must be large
763 * enough to hold the minimum length legal IP datagram (20 bytes).
765 if (link_len < sizeof(struct ipv4_hdr))
768 /* 2. The IP checksum must be correct. */
769 /* this is checked in H/W */
772 * 3. The IP version number must be 4. If the version number is not 4
773 * then the packet may be another version of IP, such as IPng or
776 if (((pkt->version_ihl) >> 4) != 4)
779 * 4. The IP header length field must be large enough to hold the
780 * minimum length legal IP datagram (20 bytes = 5 words).
782 if ((pkt->version_ihl & 0xf) < 5)
786 * 5. The IP total length field must be large enough to hold the IP
787 * datagram header, whose length is specified in the IP header length
790 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
797 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
799 static __m128i mask0;
800 static __m128i mask1;
801 static __m128i mask2;
802 static inline uint16_t
803 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
804 lookup_struct_t *ipv4_l3fwd_lookup_struct)
807 union ipv4_5tuple_host key;
809 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
810 __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
811 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
813 key.xmm = _mm_and_si128(data, mask0);
814 /* Find destination port */
815 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
816 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
819 static inline uint16_t
820 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
821 lookup_struct_t *ipv6_l3fwd_lookup_struct)
824 union ipv6_5tuple_host key;
826 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
827 __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
828 __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
830 __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
831 sizeof(__m128i) + sizeof(__m128i)));
832 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
833 key.xmm[0] = _mm_and_si128(data0, mask1);
834 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
837 /* Get part of 5 tuple: dst port and src port and dst IP address higher
839 key.xmm[2] = _mm_and_si128(data2, mask2);
841 /* Find destination port */
842 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
843 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
847 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
849 static inline uint16_t
850 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
851 lookup_struct_t *ipv4_l3fwd_lookup_struct)
855 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
856 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
857 &next_hop) == 0) ? next_hop : portid);
860 static inline uint16_t
861 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
862 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
866 return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
867 ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
872 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
873 __attribute__((unused));
875 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
876 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
878 #define MASK_ALL_PKTS 0xff
879 #define EXCLUDE_1ST_PKT 0xfe
880 #define EXCLUDE_2ND_PKT 0xfd
881 #define EXCLUDE_3RD_PKT 0xfb
882 #define EXCLUDE_4TH_PKT 0xf7
883 #define EXCLUDE_5TH_PKT 0xef
884 #define EXCLUDE_6TH_PKT 0xdf
885 #define EXCLUDE_7TH_PKT 0xbf
886 #define EXCLUDE_8TH_PKT 0x7f
889 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
891 struct ether_hdr *eth_hdr[8];
892 struct ipv4_hdr *ipv4_hdr[8];
893 uint16_t dst_port[8];
895 union ipv4_5tuple_host key[8];
898 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
899 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
900 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
901 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
902 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
903 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
904 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
905 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
907 /* Handle IPv4 headers.*/
908 ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
909 sizeof(struct ether_hdr));
910 ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
911 sizeof(struct ether_hdr));
912 ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
913 sizeof(struct ether_hdr));
914 ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
915 sizeof(struct ether_hdr));
916 ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
917 sizeof(struct ether_hdr));
918 ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
919 sizeof(struct ether_hdr));
920 ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
921 sizeof(struct ether_hdr));
922 ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
923 sizeof(struct ether_hdr));
925 #ifdef DO_RFC_1812_CHECKS
926 /* Check to make sure the packet is valid (RFC1812) */
927 uint8_t valid_mask = MASK_ALL_PKTS;
929 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
930 rte_pktmbuf_free(m[0]);
931 valid_mask &= EXCLUDE_1ST_PKT;
933 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
934 rte_pktmbuf_free(m[1]);
935 valid_mask &= EXCLUDE_2ND_PKT;
937 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
938 rte_pktmbuf_free(m[2]);
939 valid_mask &= EXCLUDE_3RD_PKT;
941 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
942 rte_pktmbuf_free(m[3]);
943 valid_mask &= EXCLUDE_4TH_PKT;
945 if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
946 rte_pktmbuf_free(m[4]);
947 valid_mask &= EXCLUDE_5TH_PKT;
949 if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
950 rte_pktmbuf_free(m[5]);
951 valid_mask &= EXCLUDE_6TH_PKT;
953 if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
954 rte_pktmbuf_free(m[6]);
955 valid_mask &= EXCLUDE_7TH_PKT;
957 if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
958 rte_pktmbuf_free(m[7]);
959 valid_mask &= EXCLUDE_8TH_PKT;
961 if (unlikely(valid_mask != MASK_ALL_PKTS)) {
967 for (i = 0; i < 8; i++)
968 if ((0x1 << i) & valid_mask)
969 l3fwd_simple_forward(m[i], portid);
971 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
973 data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
974 sizeof(struct ether_hdr) +
975 offsetof(struct ipv4_hdr, time_to_live)));
976 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
977 sizeof(struct ether_hdr) +
978 offsetof(struct ipv4_hdr, time_to_live)));
979 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
980 sizeof(struct ether_hdr) +
981 offsetof(struct ipv4_hdr, time_to_live)));
982 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
983 sizeof(struct ether_hdr) +
984 offsetof(struct ipv4_hdr, time_to_live)));
985 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
986 sizeof(struct ether_hdr) +
987 offsetof(struct ipv4_hdr, time_to_live)));
988 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
989 sizeof(struct ether_hdr) +
990 offsetof(struct ipv4_hdr, time_to_live)));
991 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
992 sizeof(struct ether_hdr) +
993 offsetof(struct ipv4_hdr, time_to_live)));
994 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
995 sizeof(struct ether_hdr) +
996 offsetof(struct ipv4_hdr, time_to_live)));
998 key[0].xmm = _mm_and_si128(data[0], mask0);
999 key[1].xmm = _mm_and_si128(data[1], mask0);
1000 key[2].xmm = _mm_and_si128(data[2], mask0);
1001 key[3].xmm = _mm_and_si128(data[3], mask0);
1002 key[4].xmm = _mm_and_si128(data[4], mask0);
1003 key[5].xmm = _mm_and_si128(data[5], mask0);
1004 key[6].xmm = _mm_and_si128(data[6], mask0);
1005 key[7].xmm = _mm_and_si128(data[7], mask0);
1007 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1008 &key[4], &key[5], &key[6], &key[7]};
1010 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1011 &key_array[0], 8, ret);
1012 dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1013 dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1014 dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1015 dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1016 dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1017 dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1018 dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1019 dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1021 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1022 (enabled_port_mask & 1 << dst_port[0]) == 0)
1023 dst_port[0] = portid;
1024 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1025 (enabled_port_mask & 1 << dst_port[1]) == 0)
1026 dst_port[1] = portid;
1027 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1028 (enabled_port_mask & 1 << dst_port[2]) == 0)
1029 dst_port[2] = portid;
1030 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1031 (enabled_port_mask & 1 << dst_port[3]) == 0)
1032 dst_port[3] = portid;
1033 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1034 (enabled_port_mask & 1 << dst_port[4]) == 0)
1035 dst_port[4] = portid;
1036 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1037 (enabled_port_mask & 1 << dst_port[5]) == 0)
1038 dst_port[5] = portid;
1039 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1040 (enabled_port_mask & 1 << dst_port[6]) == 0)
1041 dst_port[6] = portid;
1042 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1043 (enabled_port_mask & 1 << dst_port[7]) == 0)
1044 dst_port[7] = portid;
1046 #ifdef DO_RFC_1812_CHECKS
1047 /* Update time to live and header checksum */
1048 --(ipv4_hdr[0]->time_to_live);
1049 --(ipv4_hdr[1]->time_to_live);
1050 --(ipv4_hdr[2]->time_to_live);
1051 --(ipv4_hdr[3]->time_to_live);
1052 ++(ipv4_hdr[0]->hdr_checksum);
1053 ++(ipv4_hdr[1]->hdr_checksum);
1054 ++(ipv4_hdr[2]->hdr_checksum);
1055 ++(ipv4_hdr[3]->hdr_checksum);
1056 --(ipv4_hdr[4]->time_to_live);
1057 --(ipv4_hdr[5]->time_to_live);
1058 --(ipv4_hdr[6]->time_to_live);
1059 --(ipv4_hdr[7]->time_to_live);
1060 ++(ipv4_hdr[4]->hdr_checksum);
1061 ++(ipv4_hdr[5]->hdr_checksum);
1062 ++(ipv4_hdr[6]->hdr_checksum);
1063 ++(ipv4_hdr[7]->hdr_checksum);
1067 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1068 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1069 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1070 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1071 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1072 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1073 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1074 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1077 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1078 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1079 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1080 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1081 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1082 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1083 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1084 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1086 send_single_packet(m[0], (uint8_t)dst_port[0]);
1087 send_single_packet(m[1], (uint8_t)dst_port[1]);
1088 send_single_packet(m[2], (uint8_t)dst_port[2]);
1089 send_single_packet(m[3], (uint8_t)dst_port[3]);
1090 send_single_packet(m[4], (uint8_t)dst_port[4]);
1091 send_single_packet(m[5], (uint8_t)dst_port[5]);
1092 send_single_packet(m[6], (uint8_t)dst_port[6]);
1093 send_single_packet(m[7], (uint8_t)dst_port[7]);
1097 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1098 __m128i mask1, union ipv6_5tuple_host *key)
1100 __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1101 __m128i *, sizeof(struct ether_hdr) +
1102 offsetof(struct ipv6_hdr, payload_len)));
1103 __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1104 __m128i *, sizeof(struct ether_hdr) +
1105 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1106 __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1107 __m128i *, sizeof(struct ether_hdr) +
1108 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1110 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1111 key->xmm[1] = tmpdata1;
1112 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1116 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1119 uint16_t dst_port[8];
1120 struct ether_hdr *eth_hdr[8];
1121 union ipv6_5tuple_host key[8];
1123 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1125 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1126 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1127 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1128 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1129 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1130 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1131 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1132 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1134 /* Handle IPv6 headers.*/
1135 ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1136 sizeof(struct ether_hdr));
1137 ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1138 sizeof(struct ether_hdr));
1139 ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1140 sizeof(struct ether_hdr));
1141 ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1142 sizeof(struct ether_hdr));
1143 ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1144 sizeof(struct ether_hdr));
1145 ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1146 sizeof(struct ether_hdr));
1147 ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1148 sizeof(struct ether_hdr));
1149 ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1150 sizeof(struct ether_hdr));
1152 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1153 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1154 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1155 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1156 get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1157 get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1158 get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1159 get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1161 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1162 &key[4], &key[5], &key[6], &key[7]};
1164 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1165 &key_array[0], 4, ret);
1166 dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1167 dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1168 dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1169 dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1170 dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1171 dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1172 dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1173 dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1175 if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1176 (enabled_port_mask & 1 << dst_port[0]) == 0)
1177 dst_port[0] = portid;
1178 if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1179 (enabled_port_mask & 1 << dst_port[1]) == 0)
1180 dst_port[1] = portid;
1181 if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1182 (enabled_port_mask & 1 << dst_port[2]) == 0)
1183 dst_port[2] = portid;
1184 if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1185 (enabled_port_mask & 1 << dst_port[3]) == 0)
1186 dst_port[3] = portid;
1187 if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1188 (enabled_port_mask & 1 << dst_port[4]) == 0)
1189 dst_port[4] = portid;
1190 if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1191 (enabled_port_mask & 1 << dst_port[5]) == 0)
1192 dst_port[5] = portid;
1193 if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1194 (enabled_port_mask & 1 << dst_port[6]) == 0)
1195 dst_port[6] = portid;
1196 if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1197 (enabled_port_mask & 1 << dst_port[7]) == 0)
1198 dst_port[7] = portid;
1201 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1202 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1203 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1204 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1205 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1206 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1207 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1208 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1211 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
1212 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
1213 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
1214 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
1215 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr);
1216 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr);
1217 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr);
1218 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr);
1220 send_single_packet(m[0], dst_port[0]);
1221 send_single_packet(m[1], dst_port[1]);
1222 send_single_packet(m[2], dst_port[2]);
1223 send_single_packet(m[3], dst_port[3]);
1224 send_single_packet(m[4], dst_port[4]);
1225 send_single_packet(m[5], dst_port[5]);
1226 send_single_packet(m[6], dst_port[6]);
1227 send_single_packet(m[7], dst_port[7]);
1230 #endif /* APP_LOOKUP_METHOD */
1232 static __rte_always_inline void
1233 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1235 struct ether_hdr *eth_hdr;
1236 struct ipv4_hdr *ipv4_hdr;
1239 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1241 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1242 /* Handle IPv4 headers.*/
1243 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1244 sizeof(struct ether_hdr));
1246 #ifdef DO_RFC_1812_CHECKS
1247 /* Check to make sure the packet is valid (RFC1812) */
1248 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1249 rte_pktmbuf_free(m);
1254 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1255 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1256 if (dst_port >= RTE_MAX_ETHPORTS ||
1257 (enabled_port_mask & 1 << dst_port) == 0)
1260 #ifdef DO_RFC_1812_CHECKS
1261 /* Update time to live and header checksum */
1262 --(ipv4_hdr->time_to_live);
1263 ++(ipv4_hdr->hdr_checksum);
1266 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1269 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1271 send_single_packet(m, dst_port);
1272 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1273 /* Handle IPv6 headers.*/
1274 struct ipv6_hdr *ipv6_hdr;
1276 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1277 sizeof(struct ether_hdr));
1279 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1280 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1282 if (dst_port >= RTE_MAX_ETHPORTS ||
1283 (enabled_port_mask & 1 << dst_port) == 0)
1287 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port];
1290 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1292 send_single_packet(m, dst_port);
1294 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1295 rte_pktmbuf_free(m);
1298 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1299 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1300 #ifdef DO_RFC_1812_CHECKS
1302 #define IPV4_MIN_VER_IHL 0x45
1303 #define IPV4_MAX_VER_IHL 0x4f
1304 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1306 /* Minimum value of IPV4 total length (20B) in network byte order. */
1307 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1310 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1311 * - The IP version number must be 4.
1312 * - The IP header length field must be large enough to hold the
1313 * minimum length legal IP datagram (20 bytes = 5 words).
1314 * - The IP total length field must be large enough to hold the IP
1315 * datagram header, whose length is specified in the IP header length
1317 * If we encounter invalid IPV4 packet, then set destination port for it
1318 * to BAD_PORT value.
1320 static __rte_always_inline void
1321 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1325 if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1326 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1328 ipv4_hdr->time_to_live--;
1329 ipv4_hdr->hdr_checksum++;
1331 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1332 ((uint8_t)ipv4_hdr->total_length == 0 &&
1333 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1340 #define rfc1812_process(mb, dp, ptype) do { } while (0)
1341 #endif /* DO_RFC_1812_CHECKS */
1342 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1345 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1346 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1348 static __rte_always_inline uint16_t
1349 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1352 struct ipv6_hdr *ipv6_hdr;
1353 struct ether_hdr *eth_hdr;
1355 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1356 return (uint16_t) ((rte_lpm_lookup(
1357 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1358 &next_hop) == 0) ? next_hop : portid);
1360 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1362 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1363 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1365 return (uint16_t) ((rte_lpm6_lookup(
1366 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1367 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1376 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1378 struct ether_hdr *eth_hdr;
1379 struct ipv4_hdr *ipv4_hdr;
1384 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1385 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1387 dst_ipv4 = ipv4_hdr->dst_addr;
1388 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1389 dp = get_dst_port(pkt, dst_ipv4, portid);
1391 te = _mm_load_si128((__m128i *)eth_hdr);
1395 rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1397 te = _mm_blend_epi16(te, ve, MASK_ETH);
1398 _mm_store_si128((__m128i *)eth_hdr, te);
1402 * Read packet_type and destination IPV4 addresses from 4 mbufs.
1405 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1407 uint32_t *ipv4_flag)
1409 struct ipv4_hdr *ipv4_hdr;
1410 struct ether_hdr *eth_hdr;
1411 uint32_t x0, x1, x2, x3;
1413 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1414 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1415 x0 = ipv4_hdr->dst_addr;
1416 ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1418 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1419 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1420 x1 = ipv4_hdr->dst_addr;
1421 ipv4_flag[0] &= pkt[1]->packet_type;
1423 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1424 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1425 x2 = ipv4_hdr->dst_addr;
1426 ipv4_flag[0] &= pkt[2]->packet_type;
1428 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1429 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1430 x3 = ipv4_hdr->dst_addr;
1431 ipv4_flag[0] &= pkt[3]->packet_type;
1433 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1437 * Lookup into LPM for destination port.
1438 * If lookup fails, use incoming port (portid) as destination port.
1441 processx4_step2(__m128i dip,
1444 struct rte_mbuf *pkt[FWDSTEP],
1445 uint16_t dprt[FWDSTEP])
1448 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1449 4, 5, 6, 7, 0, 1, 2, 3);
1451 /* Byte swap 4 IPV4 addresses. */
1452 dip = _mm_shuffle_epi8(dip, bswap_mask);
1454 /* if all 4 packets are IPV4. */
1455 if (likely(ipv4_flag)) {
1456 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1459 /* get rid of unused upper 16 bit for each dport. */
1460 dst.x = _mm_packs_epi32(dst.x, dst.x);
1461 *(uint64_t *)dprt = dst.u64[0];
1464 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1465 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1466 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1467 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1472 * Update source and destination MAC addresses in the ethernet header.
1473 * Perform RFC1812 checks and updates for IPV4 packets.
1476 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1478 __m128i te[FWDSTEP];
1479 __m128i ve[FWDSTEP];
1480 __m128i *p[FWDSTEP];
1482 p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1483 p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1484 p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1485 p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1487 ve[0] = val_eth[dst_port[0]];
1488 te[0] = _mm_load_si128(p[0]);
1490 ve[1] = val_eth[dst_port[1]];
1491 te[1] = _mm_load_si128(p[1]);
1493 ve[2] = val_eth[dst_port[2]];
1494 te[2] = _mm_load_si128(p[2]);
1496 ve[3] = val_eth[dst_port[3]];
1497 te[3] = _mm_load_si128(p[3]);
1499 /* Update first 12 bytes, keep rest bytes intact. */
1500 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1501 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1502 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1503 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1505 _mm_store_si128(p[0], te[0]);
1506 _mm_store_si128(p[1], te[1]);
1507 _mm_store_si128(p[2], te[2]);
1508 _mm_store_si128(p[3], te[3]);
1510 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1511 &dst_port[0], pkt[0]->packet_type);
1512 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1513 &dst_port[1], pkt[1]->packet_type);
1514 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1515 &dst_port[2], pkt[2]->packet_type);
1516 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1517 &dst_port[3], pkt[3]->packet_type);
1521 * We group consecutive packets with the same destionation port into one burst.
1522 * To avoid extra latency this is done together with some other packet
1523 * processing, but after we made a final decision about packet's destination.
1524 * To do this we maintain:
1525 * pnum - array of number of consecutive packets with the same dest port for
1526 * each packet in the input burst.
1527 * lp - pointer to the last updated element in the pnum.
1528 * dlp - dest port value lp corresponds to.
1531 #define GRPSZ (1 << FWDSTEP)
1532 #define GRPMSK (GRPSZ - 1)
1534 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1535 if (likely((dlp) == (dcp)[(idx)])) { \
1538 (dlp) = (dcp)[idx]; \
1539 (lp) = (pn) + (idx); \
1545 * Group consecutive packets with the same destination port in bursts of 4.
1546 * Suppose we have array of destionation ports:
1547 * dst_port[] = {a, b, c, d,, e, ... }
1548 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1549 * We doing 4 comparisons at once and the result is 4 bit mask.
1550 * This mask is used as an index into prebuild array of pnum values.
1552 static inline uint16_t *
1553 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1555 static const struct {
1556 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1557 int32_t idx; /* index for new last updated elemnet. */
1558 uint16_t lpv; /* add value to the last updated element. */
1561 /* 0: a != b, b != c, c != d, d != e */
1562 .pnum = UINT64_C(0x0001000100010001),
1567 /* 1: a == b, b != c, c != d, d != e */
1568 .pnum = UINT64_C(0x0001000100010002),
1573 /* 2: a != b, b == c, c != d, d != e */
1574 .pnum = UINT64_C(0x0001000100020001),
1579 /* 3: a == b, b == c, c != d, d != e */
1580 .pnum = UINT64_C(0x0001000100020003),
1585 /* 4: a != b, b != c, c == d, d != e */
1586 .pnum = UINT64_C(0x0001000200010001),
1591 /* 5: a == b, b != c, c == d, d != e */
1592 .pnum = UINT64_C(0x0001000200010002),
1597 /* 6: a != b, b == c, c == d, d != e */
1598 .pnum = UINT64_C(0x0001000200030001),
1603 /* 7: a == b, b == c, c == d, d != e */
1604 .pnum = UINT64_C(0x0001000200030004),
1609 /* 8: a != b, b != c, c != d, d == e */
1610 .pnum = UINT64_C(0x0002000100010001),
1615 /* 9: a == b, b != c, c != d, d == e */
1616 .pnum = UINT64_C(0x0002000100010002),
1621 /* 0xa: a != b, b == c, c != d, d == e */
1622 .pnum = UINT64_C(0x0002000100020001),
1627 /* 0xb: a == b, b == c, c != d, d == e */
1628 .pnum = UINT64_C(0x0002000100020003),
1633 /* 0xc: a != b, b != c, c == d, d == e */
1634 .pnum = UINT64_C(0x0002000300010001),
1639 /* 0xd: a == b, b != c, c == d, d == e */
1640 .pnum = UINT64_C(0x0002000300010002),
1645 /* 0xe: a != b, b == c, c == d, d == e */
1646 .pnum = UINT64_C(0x0002000300040001),
1651 /* 0xf: a == b, b == c, c == d, d == e */
1652 .pnum = UINT64_C(0x0002000300040005),
1659 uint16_t u16[FWDSTEP + 1];
1661 } *pnum = (void *)pn;
1665 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1666 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1667 v = _mm_movemask_ps((__m128)dp1);
1669 /* update last port counter. */
1670 lp[0] += gptbl[v].lpv;
1672 /* if dest port value has changed. */
1674 pnum->u64 = gptbl[v].pnum;
1675 pnum->u16[FWDSTEP] = 1;
1676 lp = pnum->u16 + gptbl[v].idx;
1682 #endif /* APP_LOOKUP_METHOD */
1685 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1691 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1692 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1696 uint16_t dst_port[MAX_PKT_BURST];
1697 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1698 uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1699 uint16_t pnum[MAX_PKT_BURST + 1];
1703 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1704 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1707 * Send nb_rx - nb_rx%8 packets
1710 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1712 for (j = 0; j < n; j += 8) {
1714 pkts_burst[j]->packet_type &
1715 pkts_burst[j+1]->packet_type &
1716 pkts_burst[j+2]->packet_type &
1717 pkts_burst[j+3]->packet_type &
1718 pkts_burst[j+4]->packet_type &
1719 pkts_burst[j+5]->packet_type &
1720 pkts_burst[j+6]->packet_type &
1721 pkts_burst[j+7]->packet_type;
1722 if (pkt_type & RTE_PTYPE_L3_IPV4) {
1723 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1724 } else if (pkt_type &
1725 RTE_PTYPE_L3_IPV6) {
1726 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1728 l3fwd_simple_forward(pkts_burst[j], portid);
1729 l3fwd_simple_forward(pkts_burst[j+1], portid);
1730 l3fwd_simple_forward(pkts_burst[j+2], portid);
1731 l3fwd_simple_forward(pkts_burst[j+3], portid);
1732 l3fwd_simple_forward(pkts_burst[j+4], portid);
1733 l3fwd_simple_forward(pkts_burst[j+5], portid);
1734 l3fwd_simple_forward(pkts_burst[j+6], portid);
1735 l3fwd_simple_forward(pkts_burst[j+7], portid);
1738 for (; j < nb_rx ; j++)
1739 l3fwd_simple_forward(pkts_burst[j], portid);
1741 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1743 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1744 for (j = 0; j != k; j += FWDSTEP)
1745 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1746 &ipv4_flag[j / FWDSTEP]);
1748 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1749 for (j = 0; j != k; j += FWDSTEP)
1750 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1751 portid, &pkts_burst[j], &dst_port[j]);
1754 * Finish packet processing and group consecutive
1755 * packets with the same destination port.
1757 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1764 processx4_step3(pkts_burst, dst_port);
1766 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1767 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1769 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1770 processx4_step3(&pkts_burst[j], &dst_port[j]);
1774 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1776 dp2 = _mm_loadu_si128(
1777 (__m128i *)&dst_port[j - FWDSTEP + 1]);
1778 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1782 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1784 dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1785 sizeof(dst_port[0]));
1789 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1791 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1792 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1795 * remove values added by the last repeated
1799 dlp = dst_port[j - 1];
1801 /* set dlp and lp to the never used values. */
1803 lp = pnum + MAX_PKT_BURST;
1806 /* Process up to last 3 packets one by one. */
1807 switch (nb_rx % FWDSTEP) {
1809 process_packet(pkts_burst[j], dst_port + j, portid);
1810 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1814 process_packet(pkts_burst[j], dst_port + j, portid);
1815 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1819 process_packet(pkts_burst[j], dst_port + j, portid);
1820 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1825 * Send packets out, through destination port.
1826 * Consecuteve pacekts with the same destination port
1827 * are already grouped together.
1828 * If destination port for the packet equals BAD_PORT,
1829 * then free the packet without sending it out.
1831 for (j = 0; j < nb_rx; j += k) {
1839 if (likely(pn != BAD_PORT))
1840 send_packetsx4(pn, pkts_burst + j, k);
1842 for (m = j; m != j + k; m++)
1843 rte_pktmbuf_free(pkts_burst[m]);
1847 #endif /* APP_LOOKUP_METHOD */
1848 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1850 /* Prefetch first packets */
1851 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1852 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1854 /* Prefetch and forward already prefetched packets */
1855 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1856 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1857 j + PREFETCH_OFFSET], void *));
1858 l3fwd_simple_forward(pkts_burst[j], portid);
1861 /* Forward remaining prefetched packets */
1862 for (; j < nb_rx; j++)
1863 l3fwd_simple_forward(pkts_burst[j], portid);
1865 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1869 #if (APP_CPU_LOAD > 0)
1872 * CPU-load stats collector
1875 cpu_load_collector(__rte_unused void *arg) {
1878 uint64_t prev_tsc, diff_tsc, cur_tsc;
1879 uint64_t total[MAX_CPU] = { 0 };
1880 unsigned min_cpu = MAX_CPU;
1881 unsigned max_cpu = 0;
1886 unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1887 struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1889 struct thread_conf *thread_conf;
1891 const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1892 US_PER_S * CPU_LOAD_TIMEOUT_US;
1896 * Wait for all threads
1899 printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1902 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1905 while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1908 for (i = 0; i < n_rx_thread; i++) {
1910 thread_conf = &rx_thread[i].conf;
1911 cpu_id = thread_conf->cpu_id;
1912 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1914 if (cpu_id > max_cpu)
1916 if (cpu_id < min_cpu)
1919 for (i = 0; i < n_tx_thread; i++) {
1921 thread_conf = &tx_thread[i].conf;
1922 cpu_id = thread_conf->cpu_id;
1923 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1925 if (thread_conf->cpu_id > max_cpu)
1926 max_cpu = thread_conf->cpu_id;
1927 if (thread_conf->cpu_id < min_cpu)
1928 min_cpu = thread_conf->cpu_id;
1934 for (i = min_cpu; i <= max_cpu; i++) {
1935 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1936 for (k = 0; k < n_thread_per_cpu[i]; k++)
1937 if (thread_per_cpu[i][k]->busy[j]) {
1942 cpu_load.hits[j][i]++;
1954 cur_tsc = rte_rdtsc();
1956 diff_tsc = cur_tsc - prev_tsc;
1957 if (unlikely(diff_tsc > interval_tsc)) {
1961 printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1962 n_rx_thread, n_tx_thread);
1964 printf("cpu# proc%% poll%% overhead%%\n\n");
1966 for (i = min_cpu; i <= max_cpu; i++) {
1968 printf("CPU %d:", i);
1969 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1970 printf("%7" PRIu64 "",
1971 cpu_load.hits[j][i] * 100 / cpu_load.counter);
1972 hits += cpu_load.hits[j][i];
1973 cpu_load.hits[j][i] = 0;
1975 printf("%7" PRIu64 "\n",
1976 100 - total[i] * 100 / cpu_load.counter);
1979 cpu_load.counter = 0;
1986 #endif /* APP_CPU_LOAD */
1989 * Null processing lthread loop
1991 * This loop is used to start empty scheduler on lcore.
1994 lthread_null(__rte_unused void *args)
1996 int lcore_id = rte_lcore_id();
1998 RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2003 /* main processing loop */
2005 lthread_tx_per_ring(void *dummy)
2009 struct rte_ring *ring;
2010 struct thread_tx_conf *tx_conf;
2011 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2012 struct lthread_cond *ready;
2014 tx_conf = (struct thread_tx_conf *)dummy;
2015 ring = tx_conf->ring;
2016 ready = *tx_conf->ready;
2018 lthread_set_data((void *)tx_conf);
2021 * Move this lthread to lcore
2023 lthread_set_affinity(tx_conf->conf.lcore_id);
2025 RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2028 rte_atomic16_inc(&tx_counter);
2032 * Read packet from ring
2034 SET_CPU_BUSY(tx_conf, CPU_POLL);
2035 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2036 MAX_PKT_BURST, NULL);
2037 SET_CPU_IDLE(tx_conf, CPU_POLL);
2040 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2041 portid = pkts_burst[0]->port;
2042 process_burst(pkts_burst, nb_rx, portid);
2043 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2046 lthread_cond_wait(ready, 0);
2053 * Main tx-lthreads spawner lthread.
2055 * This lthread is used to spawn one new lthread per ring from producers.
2059 lthread_tx(void *args)
2065 struct thread_tx_conf *tx_conf;
2067 tx_conf = (struct thread_tx_conf *)args;
2068 lthread_set_data((void *)tx_conf);
2071 * Move this lthread to the selected lcore
2073 lthread_set_affinity(tx_conf->conf.lcore_id);
2076 * Spawn tx readers (one per input ring)
2078 lthread_create(<, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2081 lcore_id = rte_lcore_id();
2083 RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2085 tx_conf->conf.cpu_id = sched_getcpu();
2088 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2091 * TX burst queue drain
2093 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2094 if (tx_conf->tx_mbufs[portid].len == 0)
2096 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2097 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2098 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2099 tx_conf->tx_mbufs[portid].len = 0;
2107 lthread_rx(void *dummy)
2115 int len[RTE_MAX_LCORE] = { 0 };
2116 int old_len, new_len;
2117 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2118 struct thread_rx_conf *rx_conf;
2120 rx_conf = (struct thread_rx_conf *)dummy;
2121 lthread_set_data((void *)rx_conf);
2124 * Move this lthread to lcore
2126 lthread_set_affinity(rx_conf->conf.lcore_id);
2128 if (rx_conf->n_rx_queue == 0) {
2129 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2133 RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2135 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2137 portid = rx_conf->rx_queue_list[i].port_id;
2138 queueid = rx_conf->rx_queue_list[i].queue_id;
2139 RTE_LOG(INFO, L3FWD,
2140 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2141 rte_lcore_id(), portid, queueid);
2145 * Init all condition variables (one per rx thread)
2147 for (i = 0; i < rx_conf->n_rx_queue; i++)
2148 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2152 rx_conf->conf.cpu_id = sched_getcpu();
2153 rte_atomic16_inc(&rx_counter);
2157 * Read packet from RX queues
2159 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2160 portid = rx_conf->rx_queue_list[i].port_id;
2161 queueid = rx_conf->rx_queue_list[i].queue_id;
2163 SET_CPU_BUSY(rx_conf, CPU_POLL);
2164 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2166 SET_CPU_IDLE(rx_conf, CPU_POLL);
2169 worker_id = (worker_id + 1) % rx_conf->n_ring;
2170 old_len = len[worker_id];
2172 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2173 ret = rte_ring_sp_enqueue_burst(
2174 rx_conf->ring[worker_id],
2175 (void **) pkts_burst,
2178 new_len = old_len + ret;
2180 if (new_len >= BURST_SIZE) {
2181 lthread_cond_signal(rx_conf->ready[worker_id]);
2185 len[worker_id] = new_len;
2187 if (unlikely(ret < nb_rx)) {
2190 for (k = ret; k < nb_rx; k++) {
2191 struct rte_mbuf *m = pkts_burst[k];
2193 rte_pktmbuf_free(m);
2196 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2206 * Start scheduler with initial lthread on lcore
2208 * This lthread loop spawns all rx and tx lthreads on master lcore
2212 lthread_spawner(__rte_unused void *arg)
2214 struct lthread *lt[MAX_THREAD];
2218 printf("Entering lthread_spawner\n");
2221 * Create producers (rx threads) on default lcore
2223 for (i = 0; i < n_rx_thread; i++) {
2224 rx_thread[i].conf.thread_id = i;
2225 lthread_create(<[n_thread], -1, lthread_rx,
2226 (void *)&rx_thread[i]);
2231 * Wait for all producers. Until some producers can be started on the same
2232 * scheduler as this lthread, yielding is required to let them to run and
2233 * prevent deadlock here.
2235 while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2236 lthread_sleep(100000);
2239 * Create consumers (tx threads) on default lcore_id
2241 for (i = 0; i < n_tx_thread; i++) {
2242 tx_thread[i].conf.thread_id = i;
2243 lthread_create(<[n_thread], -1, lthread_tx,
2244 (void *)&tx_thread[i]);
2249 * Wait for all threads finished
2251 for (i = 0; i < n_thread; i++)
2252 lthread_join(lt[i], NULL);
2258 * Start master scheduler with initial lthread spawning rx and tx lthreads
2259 * (main_lthread_master).
2262 lthread_master_spawner(__rte_unused void *arg) {
2264 int lcore_id = rte_lcore_id();
2266 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2267 lthread_create(<, -1, lthread_spawner, NULL);
2274 * Start scheduler on lcore.
2277 sched_spawner(__rte_unused void *arg) {
2279 int lcore_id = rte_lcore_id();
2282 if (lcore_id == cpu_load_lcore_id) {
2283 cpu_load_collector(arg);
2286 #endif /* APP_CPU_LOAD */
2288 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2289 lthread_create(<, -1, lthread_null, NULL);
2295 /* main processing loop */
2297 pthread_tx(void *dummy)
2299 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2300 uint64_t prev_tsc, diff_tsc, cur_tsc;
2303 struct thread_tx_conf *tx_conf;
2305 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2306 US_PER_S * BURST_TX_DRAIN_US;
2310 tx_conf = (struct thread_tx_conf *)dummy;
2312 RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2314 tx_conf->conf.cpu_id = sched_getcpu();
2315 rte_atomic16_inc(&tx_counter);
2318 cur_tsc = rte_rdtsc();
2321 * TX burst queue drain
2323 diff_tsc = cur_tsc - prev_tsc;
2324 if (unlikely(diff_tsc > drain_tsc)) {
2327 * This could be optimized (use queueid instead of
2328 * portid), but it is not called so often
2330 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2331 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2332 if (tx_conf->tx_mbufs[portid].len == 0)
2334 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2335 tx_conf->tx_mbufs[portid].len = 0;
2337 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2343 * Read packet from ring
2345 SET_CPU_BUSY(tx_conf, CPU_POLL);
2346 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2347 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2348 SET_CPU_IDLE(tx_conf, CPU_POLL);
2350 if (unlikely(nb_rx == 0)) {
2355 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2356 portid = pkts_burst[0]->port;
2357 process_burst(pkts_burst, nb_rx, portid);
2358 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2364 pthread_rx(void *dummy)
2373 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2375 struct thread_rx_conf *rx_conf;
2377 lcore_id = rte_lcore_id();
2378 rx_conf = (struct thread_rx_conf *)dummy;
2380 if (rx_conf->n_rx_queue == 0) {
2381 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2385 RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2387 for (i = 0; i < rx_conf->n_rx_queue; i++) {
2389 portid = rx_conf->rx_queue_list[i].port_id;
2390 queueid = rx_conf->rx_queue_list[i].queue_id;
2391 RTE_LOG(INFO, L3FWD,
2392 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2393 lcore_id, portid, queueid);
2397 rx_conf->conf.cpu_id = sched_getcpu();
2398 rte_atomic16_inc(&rx_counter);
2402 * Read packet from RX queues
2404 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2405 portid = rx_conf->rx_queue_list[i].port_id;
2406 queueid = rx_conf->rx_queue_list[i].queue_id;
2408 SET_CPU_BUSY(rx_conf, CPU_POLL);
2409 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2411 SET_CPU_IDLE(rx_conf, CPU_POLL);
2418 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2419 worker_id = (worker_id + 1) % rx_conf->n_ring;
2420 n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2421 (void **)pkts_burst, nb_rx, NULL);
2423 if (unlikely(n != nb_rx)) {
2426 for (k = n; k < nb_rx; k++) {
2427 struct rte_mbuf *m = pkts_burst[k];
2429 rte_pktmbuf_free(m);
2433 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2443 pthread_run(__rte_unused void *arg) {
2444 int lcore_id = rte_lcore_id();
2447 for (i = 0; i < n_rx_thread; i++)
2448 if (rx_thread[i].conf.lcore_id == lcore_id) {
2449 printf("Start rx thread on %d...\n", lcore_id);
2450 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2451 RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2452 pthread_rx((void *)&rx_thread[i]);
2456 for (i = 0; i < n_tx_thread; i++)
2457 if (tx_thread[i].conf.lcore_id == lcore_id) {
2458 printf("Start tx thread on %d...\n", lcore_id);
2459 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2460 RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2461 pthread_tx((void *)&tx_thread[i]);
2466 if (lcore_id == cpu_load_lcore_id)
2467 cpu_load_collector(arg);
2468 #endif /* APP_CPU_LOAD */
2474 check_lcore_params(void)
2476 uint8_t queue, lcore;
2480 for (i = 0; i < nb_rx_thread_params; ++i) {
2481 queue = rx_thread_params[i].queue_id;
2482 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2483 printf("invalid queue number: %hhu\n", queue);
2486 lcore = rx_thread_params[i].lcore_id;
2487 if (!rte_lcore_is_enabled(lcore)) {
2488 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2491 socketid = rte_lcore_to_socket_id(lcore);
2492 if ((socketid != 0) && (numa_on == 0))
2493 printf("warning: lcore %hhu is on socket %d with numa off\n",
2500 check_port_config(void)
2505 for (i = 0; i < nb_rx_thread_params; ++i) {
2506 portid = rx_thread_params[i].port_id;
2507 if ((enabled_port_mask & (1 << portid)) == 0) {
2508 printf("port %u is not enabled in port mask\n", portid);
2511 if (!rte_eth_dev_is_valid_port(portid)) {
2512 printf("port %u is not present on the board\n", portid);
2520 get_port_n_rx_queues(const uint16_t port)
2525 for (i = 0; i < nb_rx_thread_params; ++i)
2526 if (rx_thread_params[i].port_id == port &&
2527 rx_thread_params[i].queue_id > queue)
2528 queue = rx_thread_params[i].queue_id;
2530 return (uint8_t)(++queue);
2537 struct thread_rx_conf *rx_conf;
2538 struct thread_tx_conf *tx_conf;
2539 unsigned rx_thread_id, tx_thread_id;
2541 struct rte_ring *ring = NULL;
2543 for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2545 tx_conf = &tx_thread[tx_thread_id];
2547 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2548 tx_conf->conf.thread_id);
2550 rx_thread_id = tx_conf->conf.thread_id;
2551 if (rx_thread_id > n_tx_thread) {
2552 printf("connection from tx-thread %u to rx-thread %u fails "
2553 "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2557 rx_conf = &rx_thread[rx_thread_id];
2558 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2560 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2561 socket_io, rx_thread_id, tx_thread_id);
2563 ring = rte_ring_create(name, 1024 * 4, socket_io,
2564 RING_F_SP_ENQ | RING_F_SC_DEQ);
2567 rte_panic("Cannot create ring to connect rx-thread %u "
2568 "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2571 rx_conf->ring[rx_conf->n_ring] = ring;
2573 tx_conf->ring = ring;
2574 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2582 init_rx_queues(void)
2584 uint16_t i, nb_rx_queue;
2589 for (i = 0; i < nb_rx_thread_params; ++i) {
2590 thread = rx_thread_params[i].thread_id;
2591 nb_rx_queue = rx_thread[thread].n_rx_queue;
2593 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2594 printf("error: too many queues (%u) for thread: %u\n",
2595 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2599 rx_thread[thread].conf.thread_id = thread;
2600 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2601 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2602 rx_thread_params[i].port_id;
2603 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2604 rx_thread_params[i].queue_id;
2605 rx_thread[thread].n_rx_queue++;
2607 if (thread >= n_rx_thread)
2608 n_rx_thread = thread + 1;
2615 init_tx_threads(void)
2620 for (i = 0; i < nb_tx_thread_params; ++i) {
2621 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2622 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2630 print_usage(const char *prgname)
2632 printf("%s [EAL options] -- -p PORTMASK -P"
2633 " [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2634 " [--tx (lcore,thread)[,(lcore,thread]]"
2635 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2636 " [--parse-ptype]\n\n"
2637 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2638 " -P : enable promiscuous mode\n"
2639 " --rx (port,queue,lcore,thread): rx queues configuration\n"
2640 " --tx (lcore,thread): tx threads configuration\n"
2641 " --stat-lcore LCORE: use lcore for stat collector\n"
2642 " --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2643 " --no-numa: optional, disable numa awareness\n"
2644 " --ipv6: optional, specify it if running ipv6 packets\n"
2645 " --enable-jumbo: enable jumbo frame"
2646 " which max packet len is PKTLEN in decimal (64-9600)\n"
2647 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2648 " --no-lthreads: turn off lthread model\n"
2649 " --parse-ptype: set to use software to analyze packet type\n\n",
2653 static int parse_max_pkt_len(const char *pktlen)
2658 /* parse decimal string */
2659 len = strtoul(pktlen, &end, 10);
2660 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2670 parse_portmask(const char *portmask)
2675 /* parse hexadecimal string */
2676 pm = strtoul(portmask, &end, 16);
2677 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2686 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2688 parse_hash_entry_number(const char *hash_entry_num)
2691 unsigned long hash_en;
2693 /* parse hexadecimal string */
2694 hash_en = strtoul(hash_entry_num, &end, 16);
2695 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2706 parse_rx_config(const char *q_arg)
2709 const char *p, *p0 = q_arg;
2718 unsigned long int_fld[_NUM_FLD];
2719 char *str_fld[_NUM_FLD];
2723 nb_rx_thread_params = 0;
2725 while ((p = strchr(p0, '(')) != NULL) {
2727 p0 = strchr(p, ')');
2732 if (size >= sizeof(s))
2735 snprintf(s, sizeof(s), "%.*s", size, p);
2736 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2738 for (i = 0; i < _NUM_FLD; i++) {
2740 int_fld[i] = strtoul(str_fld[i], &end, 0);
2741 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2744 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2745 printf("exceeded max number of rx params: %hu\n",
2746 nb_rx_thread_params);
2749 rx_thread_params_array[nb_rx_thread_params].port_id =
2751 rx_thread_params_array[nb_rx_thread_params].queue_id =
2752 (uint8_t)int_fld[FLD_QUEUE];
2753 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2754 (uint8_t)int_fld[FLD_LCORE];
2755 rx_thread_params_array[nb_rx_thread_params].thread_id =
2756 (uint8_t)int_fld[FLD_THREAD];
2757 ++nb_rx_thread_params;
2759 rx_thread_params = rx_thread_params_array;
2764 parse_tx_config(const char *q_arg)
2767 const char *p, *p0 = q_arg;
2774 unsigned long int_fld[_NUM_FLD];
2775 char *str_fld[_NUM_FLD];
2779 nb_tx_thread_params = 0;
2781 while ((p = strchr(p0, '(')) != NULL) {
2783 p0 = strchr(p, ')');
2788 if (size >= sizeof(s))
2791 snprintf(s, sizeof(s), "%.*s", size, p);
2792 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2794 for (i = 0; i < _NUM_FLD; i++) {
2796 int_fld[i] = strtoul(str_fld[i], &end, 0);
2797 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2800 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2801 printf("exceeded max number of tx params: %hu\n",
2802 nb_tx_thread_params);
2805 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2806 (uint8_t)int_fld[FLD_LCORE];
2807 tx_thread_params_array[nb_tx_thread_params].thread_id =
2808 (uint8_t)int_fld[FLD_THREAD];
2809 ++nb_tx_thread_params;
2811 tx_thread_params = tx_thread_params_array;
2816 #if (APP_CPU_LOAD > 0)
2818 parse_stat_lcore(const char *stat_lcore)
2821 unsigned long lcore_id;
2823 lcore_id = strtoul(stat_lcore, &end, 10);
2824 if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2832 parse_eth_dest(const char *optarg)
2836 uint8_t c, *dest, peer_addr[6];
2839 portid = strtoul(optarg, &port_end, 10);
2840 if (errno != 0 || port_end == optarg || *port_end++ != ',')
2841 rte_exit(EXIT_FAILURE,
2842 "Invalid eth-dest: %s", optarg);
2843 if (portid >= RTE_MAX_ETHPORTS)
2844 rte_exit(EXIT_FAILURE,
2845 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2846 portid, RTE_MAX_ETHPORTS);
2848 if (cmdline_parse_etheraddr(NULL, port_end,
2849 &peer_addr, sizeof(peer_addr)) < 0)
2850 rte_exit(EXIT_FAILURE,
2851 "Invalid ethernet address: %s\n",
2853 dest = (uint8_t *)&dest_eth_addr[portid];
2854 for (c = 0; c < 6; c++)
2855 dest[c] = peer_addr[c];
2856 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2859 #define CMD_LINE_OPT_RX_CONFIG "rx"
2860 #define CMD_LINE_OPT_TX_CONFIG "tx"
2861 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2862 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2863 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2864 #define CMD_LINE_OPT_IPV6 "ipv6"
2865 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2866 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2867 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2868 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2870 /* Parse the argument given in the command line of the application */
2872 parse_args(int argc, char **argv)
2877 char *prgname = argv[0];
2878 static struct option lgopts[] = {
2879 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2880 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2881 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2882 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2883 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2884 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2885 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2886 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2887 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2888 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2894 while ((opt = getopt_long(argc, argvopt, "p:P",
2895 lgopts, &option_index)) != EOF) {
2900 enabled_port_mask = parse_portmask(optarg);
2901 if (enabled_port_mask == 0) {
2902 printf("invalid portmask\n");
2903 print_usage(prgname);
2908 printf("Promiscuous mode selected\n");
2914 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2915 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2916 ret = parse_rx_config(optarg);
2918 printf("invalid rx-config\n");
2919 print_usage(prgname);
2924 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2925 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2926 ret = parse_tx_config(optarg);
2928 printf("invalid tx-config\n");
2929 print_usage(prgname);
2934 #if (APP_CPU_LOAD > 0)
2935 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2936 sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2937 cpu_load_lcore_id = parse_stat_lcore(optarg);
2941 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2942 sizeof(CMD_LINE_OPT_ETH_DEST)))
2943 parse_eth_dest(optarg);
2945 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2946 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2947 printf("numa is disabled\n");
2951 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2952 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2953 sizeof(CMD_LINE_OPT_IPV6))) {
2954 printf("ipv6 is specified\n");
2959 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2960 sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2961 printf("l-threads model is disabled\n");
2965 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2966 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2967 printf("software packet type parsing enabled\n");
2971 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2972 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2973 struct option lenopts = {"max-pkt-len", required_argument, 0,
2976 printf("jumbo frame is enabled - disabling simple TX path\n");
2977 port_conf.rxmode.offloads |=
2978 DEV_RX_OFFLOAD_JUMBO_FRAME;
2979 port_conf.txmode.offloads |=
2980 DEV_TX_OFFLOAD_MULTI_SEGS;
2982 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2983 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2986 ret = parse_max_pkt_len(optarg);
2987 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2988 printf("invalid packet length\n");
2989 print_usage(prgname);
2992 port_conf.rxmode.max_rx_pkt_len = ret;
2994 printf("set jumbo frame max packet length to %u\n",
2995 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2997 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2998 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2999 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
3000 ret = parse_hash_entry_number(optarg);
3001 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3002 hash_entry_number = ret;
3004 printf("invalid hash entry number\n");
3005 print_usage(prgname);
3013 print_usage(prgname);
3019 argv[optind-1] = prgname;
3022 optind = 1; /* reset getopt lib */
3027 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
3029 char buf[ETHER_ADDR_FMT_SIZE];
3031 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
3032 printf("%s%s", name, buf);
3035 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3037 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3038 union ipv4_5tuple_host *key2)
3040 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3041 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3042 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3043 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3044 key2->proto = key1->proto;
3049 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3050 union ipv6_5tuple_host *key2)
3054 for (i = 0; i < 16; i++) {
3055 key2->ip_dst[i] = key1->ip_dst[i];
3056 key2->ip_src[i] = key1->ip_src[i];
3058 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3059 key2->port_src = rte_cpu_to_be_16(key1->port_src);
3060 key2->proto = key1->proto;
3066 #define BYTE_VALUE_MAX 256
3067 #define ALL_32_BITS 0xffffffff
3068 #define BIT_8_TO_15 0x0000ff00
3070 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3074 uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3076 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3077 for (i = 0; i < array_len; i++) {
3078 struct ipv4_l3fwd_route entry;
3079 union ipv4_5tuple_host newkey;
3081 entry = ipv4_l3fwd_route_array[i];
3082 convert_ipv4_5tuple(&entry.key, &newkey);
3083 ret = rte_hash_add_key(h, (void *)&newkey);
3085 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3086 " to the l3fwd hash.\n", i);
3088 ipv4_l3fwd_out_if[ret] = entry.if_out;
3090 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3093 #define BIT_16_TO_23 0x00ff0000
3095 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3099 uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3101 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3102 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3103 for (i = 0; i < array_len; i++) {
3104 struct ipv6_l3fwd_route entry;
3105 union ipv6_5tuple_host newkey;
3107 entry = ipv6_l3fwd_route_array[i];
3108 convert_ipv6_5tuple(&entry.key, &newkey);
3109 ret = rte_hash_add_key(h, (void *)&newkey);
3111 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3112 " to the l3fwd hash.\n", i);
3114 ipv6_l3fwd_out_if[ret] = entry.if_out;
3116 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3119 #define NUMBER_PORT_USED 4
3121 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3122 unsigned int nr_flow)
3126 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3128 for (i = 0; i < nr_flow; i++) {
3129 struct ipv4_l3fwd_route entry;
3130 union ipv4_5tuple_host newkey;
3131 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3132 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3134 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3136 /* Create the ipv4 exact match flow */
3137 memset(&entry, 0, sizeof(entry));
3138 switch (i & (NUMBER_PORT_USED - 1)) {
3140 entry = ipv4_l3fwd_route_array[0];
3141 entry.key.ip_dst = IPv4(101, c, b, a);
3144 entry = ipv4_l3fwd_route_array[1];
3145 entry.key.ip_dst = IPv4(201, c, b, a);
3148 entry = ipv4_l3fwd_route_array[2];
3149 entry.key.ip_dst = IPv4(111, c, b, a);
3152 entry = ipv4_l3fwd_route_array[3];
3153 entry.key.ip_dst = IPv4(211, c, b, a);
3156 convert_ipv4_5tuple(&entry.key, &newkey);
3157 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3160 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3162 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3165 printf("Hash: Adding 0x%x keys\n", nr_flow);
3169 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3170 unsigned int nr_flow)
3174 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3175 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3176 for (i = 0; i < nr_flow; i++) {
3177 struct ipv6_l3fwd_route entry;
3178 union ipv6_5tuple_host newkey;
3180 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3181 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3183 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3186 /* Create the ipv6 exact match flow */
3187 memset(&entry, 0, sizeof(entry));
3188 switch (i & (NUMBER_PORT_USED - 1)) {
3190 entry = ipv6_l3fwd_route_array[0];
3193 entry = ipv6_l3fwd_route_array[1];
3196 entry = ipv6_l3fwd_route_array[2];
3199 entry = ipv6_l3fwd_route_array[3];
3202 entry.key.ip_dst[13] = c;
3203 entry.key.ip_dst[14] = b;
3204 entry.key.ip_dst[15] = a;
3205 convert_ipv6_5tuple(&entry.key, &newkey);
3206 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3209 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3211 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3214 printf("Hash: Adding 0x%x keys\n", nr_flow);
3218 setup_hash(int socketid)
3220 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3222 .entries = L3FWD_HASH_ENTRIES,
3223 .key_len = sizeof(union ipv4_5tuple_host),
3224 .hash_func = ipv4_hash_crc,
3225 .hash_func_init_val = 0,
3228 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3230 .entries = L3FWD_HASH_ENTRIES,
3231 .key_len = sizeof(union ipv6_5tuple_host),
3232 .hash_func = ipv6_hash_crc,
3233 .hash_func_init_val = 0,
3238 /* create ipv4 hash */
3239 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3240 ipv4_l3fwd_hash_params.name = s;
3241 ipv4_l3fwd_hash_params.socket_id = socketid;
3242 ipv4_l3fwd_lookup_struct[socketid] =
3243 rte_hash_create(&ipv4_l3fwd_hash_params);
3244 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3245 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3246 "socket %d\n", socketid);
3248 /* create ipv6 hash */
3249 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3250 ipv6_l3fwd_hash_params.name = s;
3251 ipv6_l3fwd_hash_params.socket_id = socketid;
3252 ipv6_l3fwd_lookup_struct[socketid] =
3253 rte_hash_create(&ipv6_l3fwd_hash_params);
3254 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3255 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3256 "socket %d\n", socketid);
3258 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3259 /* For testing hash matching with a large number of flows we
3260 * generate millions of IP 5-tuples with an incremented dst
3261 * address to initialize the hash table. */
3263 /* populate the ipv4 hash */
3264 populate_ipv4_many_flow_into_table(
3265 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3267 /* populate the ipv6 hash */
3268 populate_ipv6_many_flow_into_table(
3269 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3272 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3275 /* populate the ipv4 hash */
3276 populate_ipv4_few_flow_into_table(
3277 ipv4_l3fwd_lookup_struct[socketid]);
3279 /* populate the ipv6 hash */
3280 populate_ipv6_few_flow_into_table(
3281 ipv6_l3fwd_lookup_struct[socketid]);
3287 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3289 setup_lpm(int socketid)
3291 struct rte_lpm6_config config;
3292 struct rte_lpm_config lpm_ipv4_config;
3297 /* create the LPM table */
3298 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3299 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3300 lpm_ipv4_config.number_tbl8s = 256;
3301 lpm_ipv4_config.flags = 0;
3302 ipv4_l3fwd_lookup_struct[socketid] =
3303 rte_lpm_create(s, socketid, &lpm_ipv4_config);
3304 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3305 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3306 " on socket %d\n", socketid);
3308 /* populate the LPM table */
3309 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3311 /* skip unused ports */
3312 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3313 enabled_port_mask) == 0)
3316 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3317 ipv4_l3fwd_route_array[i].ip,
3318 ipv4_l3fwd_route_array[i].depth,
3319 ipv4_l3fwd_route_array[i].if_out);
3322 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3323 "l3fwd LPM table on socket %d\n",
3327 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3328 (unsigned)ipv4_l3fwd_route_array[i].ip,
3329 ipv4_l3fwd_route_array[i].depth,
3330 ipv4_l3fwd_route_array[i].if_out);
3333 /* create the LPM6 table */
3334 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3336 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3337 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3339 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3341 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3342 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3343 " on socket %d\n", socketid);
3345 /* populate the LPM table */
3346 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3348 /* skip unused ports */
3349 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3350 enabled_port_mask) == 0)
3353 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3354 ipv6_l3fwd_route_array[i].ip,
3355 ipv6_l3fwd_route_array[i].depth,
3356 ipv6_l3fwd_route_array[i].if_out);
3359 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3360 "l3fwd LPM table on socket %d\n",
3364 printf("LPM: Adding route %s / %d (%d)\n",
3366 ipv6_l3fwd_route_array[i].depth,
3367 ipv6_l3fwd_route_array[i].if_out);
3373 init_mem(unsigned nb_mbuf)
3375 struct lcore_conf *qconf;
3380 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3381 if (rte_lcore_is_enabled(lcore_id) == 0)
3385 socketid = rte_lcore_to_socket_id(lcore_id);
3389 if (socketid >= NB_SOCKETS) {
3390 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3391 socketid, lcore_id, NB_SOCKETS);
3393 if (pktmbuf_pool[socketid] == NULL) {
3394 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3395 pktmbuf_pool[socketid] =
3396 rte_pktmbuf_pool_create(s, nb_mbuf,
3397 MEMPOOL_CACHE_SIZE, 0,
3398 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3399 if (pktmbuf_pool[socketid] == NULL)
3400 rte_exit(EXIT_FAILURE,
3401 "Cannot init mbuf pool on socket %d\n", socketid);
3403 printf("Allocated mbuf pool on socket %d\n", socketid);
3405 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3406 setup_lpm(socketid);
3408 setup_hash(socketid);
3411 qconf = &lcore_conf[lcore_id];
3412 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3413 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3418 /* Check the link status of all ports in up to 9s, and print them finally */
3420 check_all_ports_link_status(uint32_t port_mask)
3422 #define CHECK_INTERVAL 100 /* 100ms */
3423 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3425 uint8_t count, all_ports_up, print_flag = 0;
3426 struct rte_eth_link link;
3428 printf("\nChecking link status");
3430 for (count = 0; count <= MAX_CHECK_TIME; count++) {
3432 RTE_ETH_FOREACH_DEV(portid) {
3433 if ((port_mask & (1 << portid)) == 0)
3435 memset(&link, 0, sizeof(link));
3436 rte_eth_link_get_nowait(portid, &link);
3437 /* print link status if flag set */
3438 if (print_flag == 1) {
3439 if (link.link_status)
3441 "Port%d Link Up. Speed %u Mbps - %s\n",
3442 portid, link.link_speed,
3443 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3444 ("full-duplex") : ("half-duplex\n"));
3446 printf("Port %d Link Down\n", portid);
3449 /* clear all_ports_up flag if any link down */
3450 if (link.link_status == ETH_LINK_DOWN) {
3455 /* after finally printing all link status, get out */
3456 if (print_flag == 1)
3459 if (all_ports_up == 0) {
3462 rte_delay_ms(CHECK_INTERVAL);
3465 /* set the print_flag if all ports up or timeout */
3466 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3474 main(int argc, char **argv)
3476 struct rte_eth_dev_info dev_info;
3477 struct rte_eth_txconf *txconf;
3481 uint16_t queueid, portid;
3483 uint32_t n_tx_queue, nb_lcores;
3484 uint8_t nb_rx_queue, queue, socketid;
3487 ret = rte_eal_init(argc, argv);
3489 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3493 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3494 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3495 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3496 ((uint64_t)portid << 40);
3497 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3500 /* parse application arguments (after the EAL ones) */
3501 ret = parse_args(argc, argv);
3503 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3505 if (check_lcore_params() < 0)
3506 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3508 printf("Initializing rx-queues...\n");
3509 ret = init_rx_queues();
3511 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3513 printf("Initializing tx-threads...\n");
3514 ret = init_tx_threads();
3516 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3518 printf("Initializing rings...\n");
3519 ret = init_rx_rings();
3521 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3523 nb_ports = rte_eth_dev_count_avail();
3525 if (check_port_config() < 0)
3526 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3528 nb_lcores = rte_lcore_count();
3530 /* initialize all ports */
3531 RTE_ETH_FOREACH_DEV(portid) {
3532 struct rte_eth_conf local_port_conf = port_conf;
3534 /* skip ports that are not enabled */
3535 if ((enabled_port_mask & (1 << portid)) == 0) {
3536 printf("\nSkipping disabled port %d\n", portid);
3541 printf("Initializing port %d ... ", portid);
3544 nb_rx_queue = get_port_n_rx_queues(portid);
3545 n_tx_queue = nb_lcores;
3546 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3547 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3548 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3549 nb_rx_queue, (unsigned)n_tx_queue);
3550 rte_eth_dev_info_get(portid, &dev_info);
3551 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3552 local_port_conf.txmode.offloads |=
3553 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3554 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3555 (uint16_t)n_tx_queue, &local_port_conf);
3557 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3560 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3563 rte_exit(EXIT_FAILURE,
3564 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3567 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3568 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3570 print_ethaddr("Destination:",
3571 (const struct ether_addr *)&dest_eth_addr[portid]);
3575 * prepare src MACs for each port.
3577 ether_addr_copy(&ports_eth_addr[portid],
3578 (struct ether_addr *)(val_eth + portid) + 1);
3581 ret = init_mem(NB_MBUF);
3583 rte_exit(EXIT_FAILURE, "init_mem failed\n");
3585 /* init one TX queue per couple (lcore,port) */
3587 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3588 if (rte_lcore_is_enabled(lcore_id) == 0)
3592 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3596 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3599 txconf = &dev_info.default_txconf;
3600 txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
3601 txconf->offloads = local_port_conf.txmode.offloads;
3602 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3605 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3606 "port=%d\n", ret, portid);
3608 tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3614 for (i = 0; i < n_rx_thread; i++) {
3615 lcore_id = rx_thread[i].conf.lcore_id;
3617 if (rte_lcore_is_enabled(lcore_id) == 0) {
3618 rte_exit(EXIT_FAILURE,
3619 "Cannot start Rx thread on lcore %u: lcore disabled\n",
3624 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3628 /* init RX queues */
3629 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3630 struct rte_eth_dev *dev;
3631 struct rte_eth_conf *conf;
3632 struct rte_eth_rxconf rxq_conf;
3634 portid = rx_thread[i].rx_queue_list[queue].port_id;
3635 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3636 dev = &rte_eth_devices[portid];
3637 conf = &dev->data->dev_conf;
3640 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3644 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3647 rte_eth_dev_info_get(portid, &dev_info);
3648 rxq_conf = dev_info.default_rxconf;
3649 rxq_conf.offloads = conf->rxmode.offloads;
3650 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3653 pktmbuf_pool[socketid]);
3655 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3656 "port=%d\n", ret, portid);
3663 RTE_ETH_FOREACH_DEV(portid) {
3664 if ((enabled_port_mask & (1 << portid)) == 0)
3668 ret = rte_eth_dev_start(portid);
3670 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3674 * If enabled, put device in promiscuous mode.
3675 * This allows IO forwarding mode to forward packets
3676 * to itself through 2 cross-connected ports of the
3680 rte_eth_promiscuous_enable(portid);
3683 for (i = 0; i < n_rx_thread; i++) {
3684 lcore_id = rx_thread[i].conf.lcore_id;
3685 if (rte_lcore_is_enabled(lcore_id) == 0)
3688 /* check if hw packet type is supported */
3689 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3690 portid = rx_thread[i].rx_queue_list[queue].port_id;
3691 queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3693 if (parse_ptype_on) {
3694 if (!rte_eth_add_rx_callback(portid, queueid,
3695 cb_parse_ptype, NULL))
3696 rte_exit(EXIT_FAILURE,
3697 "Failed to add rx callback: "
3698 "port=%d\n", portid);
3699 } else if (!check_ptype(portid))
3700 rte_exit(EXIT_FAILURE,
3701 "Port %d cannot parse packet type.\n\n"
3702 "Please add --parse-ptype to use sw "
3703 "packet type analyzer.\n\n",
3708 check_all_ports_link_status(enabled_port_mask);
3711 printf("Starting L-Threading Model\n");
3713 #if (APP_CPU_LOAD > 0)
3714 if (cpu_load_lcore_id > 0)
3715 /* Use one lcore for cpu load collector */
3719 lthread_num_schedulers_set(nb_lcores);
3720 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3721 lthread_master_spawner(NULL);
3724 printf("Starting P-Threading Model\n");
3725 /* launch per-lcore init on every lcore */
3726 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3727 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3728 if (rte_eal_wait_lcore(lcore_id) < 0)