4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 #include <sys/types.h>
40 #include <sys/queue.h>
45 #include <rte_common.h>
47 #include <rte_byteorder.h>
49 #include <rte_memory.h>
50 #include <rte_memcpy.h>
51 #include <rte_memzone.h>
53 #include <rte_per_lcore.h>
54 #include <rte_launch.h>
55 #include <rte_atomic.h>
56 #include <rte_cycles.h>
57 #include <rte_prefetch.h>
58 #include <rte_lcore.h>
59 #include <rte_per_lcore.h>
60 #include <rte_branch_prediction.h>
61 #include <rte_interrupts.h>
63 #include <rte_random.h>
64 #include <rte_debug.h>
65 #include <rte_ether.h>
66 #include <rte_ethdev.h>
68 #include <rte_mempool.h>
73 #include <rte_string_fns.h>
75 #define APP_LOOKUP_EXACT_MATCH 0
76 #define APP_LOOKUP_LPM 1
77 #define DO_RFC_1812_CHECKS
79 #ifndef APP_LOOKUP_METHOD
80 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
84 * When set to zero, simple forwaring path is eanbled.
85 * When set to one, optimized forwarding path is enabled.
86 * Note that LPM optimisation path uses SSE4.1 instructions.
88 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
89 #define ENABLE_MULTI_BUFFER_OPTIMIZE 0
91 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1
94 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
96 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
100 #error "APP_LOOKUP_METHOD set to incorrect value"
104 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
105 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
106 #define IPv6_BYTES(addr) \
107 addr[0], addr[1], addr[2], addr[3], \
108 addr[4], addr[5], addr[6], addr[7], \
109 addr[8], addr[9], addr[10], addr[11],\
110 addr[12], addr[13],addr[14], addr[15]
114 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
116 #define MAX_JUMBO_PKT_LEN 9600
118 #define IPV6_ADDR_LEN 16
120 #define MEMPOOL_CACHE_SIZE 256
123 * This expression is used to calculate the number of mbufs needed depending on user input, taking
124 * into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
125 * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
128 #define NB_MBUF RTE_MAX ( \
129 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
130 nb_ports*nb_lcores*MAX_PKT_BURST + \
131 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
132 nb_lcores*MEMPOOL_CACHE_SIZE), \
135 #define MAX_PKT_BURST 32
136 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
139 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
141 #define MAX_TX_BURST (MAX_PKT_BURST / 2)
145 /* Configure how many packets ahead to prefetch, when reading packets */
146 #define PREFETCH_OFFSET 3
148 /* Used to mark destination port as 'invalid'. */
149 #define BAD_PORT ((uint16_t)-1)
154 * Configurable number of RX/TX ring descriptors
156 #define RTE_TEST_RX_DESC_DEFAULT 128
157 #define RTE_TEST_TX_DESC_DEFAULT 512
158 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
159 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
161 /* ethernet addresses of ports */
162 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
164 static __m128i val_eth[RTE_MAX_ETHPORTS];
166 /* replace first 12B of the ethernet header. */
167 #define MASK_ETH 0x3f
169 /* mask of enabled ports */
170 static uint32_t enabled_port_mask = 0;
171 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
172 static int numa_on = 1; /**< NUMA is enabled by default. */
174 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
175 static int ipv6 = 0; /**< ipv6 is false by default. */
180 struct rte_mbuf *m_table[MAX_PKT_BURST];
183 struct lcore_rx_queue {
186 } __rte_cache_aligned;
188 #define MAX_RX_QUEUE_PER_LCORE 16
189 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
190 #define MAX_RX_QUEUE_PER_PORT 128
192 #define MAX_LCORE_PARAMS 1024
193 struct lcore_params {
197 } __rte_cache_aligned;
199 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
200 static struct lcore_params lcore_params_array_default[] = {
212 static struct lcore_params * lcore_params = lcore_params_array_default;
213 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
214 sizeof(lcore_params_array_default[0]);
216 static struct rte_eth_conf port_conf = {
218 .mq_mode = ETH_MQ_RX_RSS,
219 .max_rx_pkt_len = ETHER_MAX_LEN,
221 .header_split = 0, /**< Header Split disabled */
222 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
223 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
224 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
225 .hw_strip_crc = 0, /**< CRC stripped by hardware */
230 .rss_hf = ETH_RSS_IP,
234 .mq_mode = ETH_MQ_TX_NONE,
238 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
240 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
242 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
243 #include <rte_hash_crc.h>
244 #define DEFAULT_HASH_FUNC rte_hash_crc
246 #include <rte_jhash.h>
247 #define DEFAULT_HASH_FUNC rte_jhash
256 } __attribute__((__packed__));
258 union ipv4_5tuple_host {
271 #define XMM_NUM_IN_IPV6_5TUPLE 3
274 uint8_t ip_dst[IPV6_ADDR_LEN];
275 uint8_t ip_src[IPV6_ADDR_LEN];
279 } __attribute__((__packed__));
281 union ipv6_5tuple_host {
286 uint8_t ip_src[IPV6_ADDR_LEN];
287 uint8_t ip_dst[IPV6_ADDR_LEN];
292 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
295 struct ipv4_l3fwd_route {
296 struct ipv4_5tuple key;
300 struct ipv6_l3fwd_route {
301 struct ipv6_5tuple key;
305 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
306 {{IPv4(101,0,0,0), IPv4(100,10,0,1), 101, 11, IPPROTO_TCP}, 0},
307 {{IPv4(201,0,0,0), IPv4(200,20,0,1), 102, 12, IPPROTO_TCP}, 1},
308 {{IPv4(111,0,0,0), IPv4(100,30,0,1), 101, 11, IPPROTO_TCP}, 2},
309 {{IPv4(211,0,0,0), IPv4(200,40,0,1), 102, 12, IPPROTO_TCP}, 3},
312 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
314 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
315 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
316 101, 11, IPPROTO_TCP}, 0},
319 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
320 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
321 102, 12, IPPROTO_TCP}, 1},
324 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
325 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
326 101, 11, IPPROTO_TCP}, 2},
329 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
330 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
331 102, 12, IPPROTO_TCP}, 3},
334 typedef struct rte_hash lookup_struct_t;
335 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
336 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
338 #ifdef RTE_ARCH_X86_64
339 /* default to 4 million hash entries (approx) */
340 #define L3FWD_HASH_ENTRIES 1024*1024*4
342 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
343 #define L3FWD_HASH_ENTRIES 1024*1024*1
345 #define HASH_ENTRY_NUMBER_DEFAULT 4
347 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
349 static inline uint32_t
350 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
353 const union ipv4_5tuple_host *k;
359 p = (const uint32_t *)&k->port_src;
361 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
362 init_val = rte_hash_crc_4byte(t, init_val);
363 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
364 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
365 init_val = rte_hash_crc_4byte(*p, init_val);
366 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
367 init_val = rte_jhash_1word(t, init_val);
368 init_val = rte_jhash_1word(k->ip_src, init_val);
369 init_val = rte_jhash_1word(k->ip_dst, init_val);
370 init_val = rte_jhash_1word(*p, init_val);
371 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
375 static inline uint32_t
376 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
378 const union ipv6_5tuple_host *k;
381 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
382 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
383 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
384 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
388 p = (const uint32_t *)&k->port_src;
390 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
391 ip_src0 = (const uint32_t *) k->ip_src;
392 ip_src1 = (const uint32_t *)(k->ip_src+4);
393 ip_src2 = (const uint32_t *)(k->ip_src+8);
394 ip_src3 = (const uint32_t *)(k->ip_src+12);
395 ip_dst0 = (const uint32_t *) k->ip_dst;
396 ip_dst1 = (const uint32_t *)(k->ip_dst+4);
397 ip_dst2 = (const uint32_t *)(k->ip_dst+8);
398 ip_dst3 = (const uint32_t *)(k->ip_dst+12);
399 init_val = rte_hash_crc_4byte(t, init_val);
400 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
401 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
402 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
403 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
404 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
405 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
406 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
407 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
408 init_val = rte_hash_crc_4byte(*p, init_val);
409 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
410 init_val = rte_jhash_1word(t, init_val);
411 init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
412 init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
413 init_val = rte_jhash_1word(*p, init_val);
414 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
418 #define IPV4_L3FWD_NUM_ROUTES \
419 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
421 #define IPV6_L3FWD_NUM_ROUTES \
422 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
424 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
425 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
429 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
430 struct ipv4_l3fwd_route {
436 struct ipv6_l3fwd_route {
442 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
443 {IPv4(1,1,1,0), 24, 0},
444 {IPv4(2,1,1,0), 24, 1},
445 {IPv4(3,1,1,0), 24, 2},
446 {IPv4(4,1,1,0), 24, 3},
447 {IPv4(5,1,1,0), 24, 4},
448 {IPv4(6,1,1,0), 24, 5},
449 {IPv4(7,1,1,0), 24, 6},
450 {IPv4(8,1,1,0), 24, 7},
453 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
454 {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
455 {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
456 {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
457 {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
458 {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
459 {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
460 {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
461 {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
464 #define IPV4_L3FWD_NUM_ROUTES \
465 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
466 #define IPV6_L3FWD_NUM_ROUTES \
467 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
469 #define IPV4_L3FWD_LPM_MAX_RULES 1024
470 #define IPV6_L3FWD_LPM_MAX_RULES 1024
471 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
473 typedef struct rte_lpm lookup_struct_t;
474 typedef struct rte_lpm6 lookup6_struct_t;
475 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
476 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
481 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
482 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
483 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
484 lookup_struct_t * ipv4_lookup_struct;
485 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
486 lookup6_struct_t * ipv6_lookup_struct;
488 lookup_struct_t * ipv6_lookup_struct;
490 } __rte_cache_aligned;
492 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
494 /* Send burst of packets on an output interface */
496 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
498 struct rte_mbuf **m_table;
502 queueid = qconf->tx_queue_id[port];
503 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
505 ret = rte_eth_tx_burst(port, queueid, m_table, n);
506 if (unlikely(ret < n)) {
508 rte_pktmbuf_free(m_table[ret]);
515 /* Enqueue a single packet, and send burst if queue is filled */
517 send_single_packet(struct rte_mbuf *m, uint8_t port)
521 struct lcore_conf *qconf;
523 lcore_id = rte_lcore_id();
525 qconf = &lcore_conf[lcore_id];
526 len = qconf->tx_mbufs[port].len;
527 qconf->tx_mbufs[port].m_table[len] = m;
530 /* enough pkts to be sent */
531 if (unlikely(len == MAX_PKT_BURST)) {
532 send_burst(qconf, MAX_PKT_BURST, port);
536 qconf->tx_mbufs[port].len = len;
540 static inline __attribute__((always_inline)) void
541 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
542 struct rte_mbuf *m[], uint32_t num)
546 len = qconf->tx_mbufs[port].len;
549 * If TX buffer for that queue is empty, and we have enough packets,
550 * then send them straightway.
552 if (num >= MAX_TX_BURST && len == 0) {
553 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
554 if (unlikely(n < num)) {
556 rte_pktmbuf_free(m[n]);
563 * Put packets into TX buffer for that queue.
567 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
570 switch (n % FWDSTEP) {
573 qconf->tx_mbufs[port].m_table[len + j] = m[j];
576 qconf->tx_mbufs[port].m_table[len + j] = m[j];
579 qconf->tx_mbufs[port].m_table[len + j] = m[j];
582 qconf->tx_mbufs[port].m_table[len + j] = m[j];
589 /* enough pkts to be sent */
590 if (unlikely(len == MAX_PKT_BURST)) {
592 send_burst(qconf, MAX_PKT_BURST, port);
594 /* copy rest of the packets into the TX buffer. */
597 switch (len % FWDSTEP) {
600 qconf->tx_mbufs[port].m_table[j] = m[n + j];
603 qconf->tx_mbufs[port].m_table[j] = m[n + j];
606 qconf->tx_mbufs[port].m_table[j] = m[n + j];
609 qconf->tx_mbufs[port].m_table[j] = m[n + j];
615 qconf->tx_mbufs[port].len = len;
618 #ifdef DO_RFC_1812_CHECKS
620 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
622 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
624 * 1. The packet length reported by the Link Layer must be large
625 * enough to hold the minimum length legal IP datagram (20 bytes).
627 if (link_len < sizeof(struct ipv4_hdr))
630 /* 2. The IP checksum must be correct. */
631 /* this is checked in H/W */
634 * 3. The IP version number must be 4. If the version number is not 4
635 * then the packet may be another version of IP, such as IPng or
638 if (((pkt->version_ihl) >> 4) != 4)
641 * 4. The IP header length field must be large enough to hold the
642 * minimum length legal IP datagram (20 bytes = 5 words).
644 if ((pkt->version_ihl & 0xf) < 5)
648 * 5. The IP total length field must be large enough to hold the IP
649 * datagram header, whose length is specified in the IP header length
652 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
659 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
661 static __m128i mask0;
662 static __m128i mask1;
663 static __m128i mask2;
664 static inline uint8_t
665 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
668 union ipv4_5tuple_host key;
670 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
671 __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
672 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
673 key.xmm = _mm_and_si128(data, mask0);
674 /* Find destination port */
675 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
676 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
679 static inline uint8_t
680 get_ipv6_dst_port(void *ipv6_hdr, uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
683 union ipv6_5tuple_host key;
685 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
686 __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
687 __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
688 __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
689 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
690 key.xmm[0] = _mm_and_si128(data0, mask1);
691 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
693 /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
694 key.xmm[2] = _mm_and_si128(data2, mask2);
696 /* Find destination port */
697 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
698 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
702 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
704 static inline uint8_t
705 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
709 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
710 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
711 &next_hop) == 0) ? next_hop : portid);
714 static inline uint8_t
715 get_ipv6_dst_port(void *ipv6_hdr, uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
718 return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
719 ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
724 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
725 struct lcore_conf *qconf) __attribute__((unused));
727 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
728 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
730 #define MASK_ALL_PKTS 0xf
731 #define EXECLUDE_1ST_PKT 0xe
732 #define EXECLUDE_2ND_PKT 0xd
733 #define EXECLUDE_3RD_PKT 0xb
734 #define EXECLUDE_4TH_PKT 0x7
737 simple_ipv4_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
739 struct ether_hdr *eth_hdr[4];
740 struct ipv4_hdr *ipv4_hdr[4];
741 void *d_addr_bytes[4];
744 union ipv4_5tuple_host key[4];
747 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
748 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
749 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
750 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
752 /* Handle IPv4 headers.*/
753 ipv4_hdr[0] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
754 sizeof(struct ether_hdr));
755 ipv4_hdr[1] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
756 sizeof(struct ether_hdr));
757 ipv4_hdr[2] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
758 sizeof(struct ether_hdr));
759 ipv4_hdr[3] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
760 sizeof(struct ether_hdr));
762 #ifdef DO_RFC_1812_CHECKS
763 /* Check to make sure the packet is valid (RFC1812) */
764 uint8_t valid_mask = MASK_ALL_PKTS;
765 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
766 rte_pktmbuf_free(m[0]);
767 valid_mask &= EXECLUDE_1ST_PKT;
769 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
770 rte_pktmbuf_free(m[1]);
771 valid_mask &= EXECLUDE_2ND_PKT;
773 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
774 rte_pktmbuf_free(m[2]);
775 valid_mask &= EXECLUDE_3RD_PKT;
777 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
778 rte_pktmbuf_free(m[3]);
779 valid_mask &= EXECLUDE_4TH_PKT;
781 if (unlikely(valid_mask != MASK_ALL_PKTS)) {
782 if (valid_mask == 0){
786 for (i = 0; i < 4; i++) {
787 if ((0x1 << i) & valid_mask) {
788 l3fwd_simple_forward(m[i], portid, qconf);
794 #endif // End of #ifdef DO_RFC_1812_CHECKS
796 data[0] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[0], unsigned char *) +
797 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
798 data[1] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[1], unsigned char *) +
799 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
800 data[2] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[2], unsigned char *) +
801 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
802 data[3] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[3], unsigned char *) +
803 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
805 key[0].xmm = _mm_and_si128(data[0], mask0);
806 key[1].xmm = _mm_and_si128(data[1], mask0);
807 key[2].xmm = _mm_and_si128(data[2], mask0);
808 key[3].xmm = _mm_and_si128(data[3], mask0);
810 const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
811 rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 4, ret);
812 dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
813 dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
814 dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
815 dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
817 if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
818 dst_port[0] = portid;
819 if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
820 dst_port[1] = portid;
821 if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
822 dst_port[2] = portid;
823 if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
824 dst_port[3] = portid;
826 /* 02:00:00:00:00:xx */
827 d_addr_bytes[0] = ð_hdr[0]->d_addr.addr_bytes[0];
828 d_addr_bytes[1] = ð_hdr[1]->d_addr.addr_bytes[0];
829 d_addr_bytes[2] = ð_hdr[2]->d_addr.addr_bytes[0];
830 d_addr_bytes[3] = ð_hdr[3]->d_addr.addr_bytes[0];
831 *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
832 *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
833 *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
834 *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
836 #ifdef DO_RFC_1812_CHECKS
837 /* Update time to live and header checksum */
838 --(ipv4_hdr[0]->time_to_live);
839 --(ipv4_hdr[1]->time_to_live);
840 --(ipv4_hdr[2]->time_to_live);
841 --(ipv4_hdr[3]->time_to_live);
842 ++(ipv4_hdr[0]->hdr_checksum);
843 ++(ipv4_hdr[1]->hdr_checksum);
844 ++(ipv4_hdr[2]->hdr_checksum);
845 ++(ipv4_hdr[3]->hdr_checksum);
849 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
850 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
851 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
852 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
854 send_single_packet(m[0], (uint8_t)dst_port[0]);
855 send_single_packet(m[1], (uint8_t)dst_port[1]);
856 send_single_packet(m[2], (uint8_t)dst_port[2]);
857 send_single_packet(m[3], (uint8_t)dst_port[3]);
861 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
862 union ipv6_5tuple_host * key)
864 __m128i tmpdata0 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
865 + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
866 __m128i tmpdata1 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
867 + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
869 __m128i tmpdata2 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
870 + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
871 + sizeof(__m128i) + sizeof(__m128i)));
872 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
873 key->xmm[1] = tmpdata1;
874 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
879 simple_ipv6_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
881 struct ether_hdr *eth_hdr[4];
882 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[4];
883 void *d_addr_bytes[4];
886 union ipv6_5tuple_host key[4];
888 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
889 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
890 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
891 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
893 /* Handle IPv6 headers.*/
894 ipv6_hdr[0] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
895 sizeof(struct ether_hdr));
896 ipv6_hdr[1] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
897 sizeof(struct ether_hdr));
898 ipv6_hdr[2] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
899 sizeof(struct ether_hdr));
900 ipv6_hdr[3] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
901 sizeof(struct ether_hdr));
903 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
904 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
905 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
906 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
908 const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
909 rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
910 dst_port[0] = (uint8_t) ((ret[0] < 0)? portid:ipv6_l3fwd_out_if[ret[0]]);
911 dst_port[1] = (uint8_t) ((ret[1] < 0)? portid:ipv6_l3fwd_out_if[ret[1]]);
912 dst_port[2] = (uint8_t) ((ret[2] < 0)? portid:ipv6_l3fwd_out_if[ret[2]]);
913 dst_port[3] = (uint8_t) ((ret[3] < 0)? portid:ipv6_l3fwd_out_if[ret[3]]);
915 if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
916 dst_port[0] = portid;
917 if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
918 dst_port[1] = portid;
919 if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
920 dst_port[2] = portid;
921 if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
922 dst_port[3] = portid;
924 /* 02:00:00:00:00:xx */
925 d_addr_bytes[0] = ð_hdr[0]->d_addr.addr_bytes[0];
926 d_addr_bytes[1] = ð_hdr[1]->d_addr.addr_bytes[0];
927 d_addr_bytes[2] = ð_hdr[2]->d_addr.addr_bytes[0];
928 d_addr_bytes[3] = ð_hdr[3]->d_addr.addr_bytes[0];
929 *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
930 *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
931 *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
932 *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
935 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
936 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
937 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
938 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
940 send_single_packet(m[0], (uint8_t)dst_port[0]);
941 send_single_packet(m[1], (uint8_t)dst_port[1]);
942 send_single_packet(m[2], (uint8_t)dst_port[2]);
943 send_single_packet(m[3], (uint8_t)dst_port[3]);
946 #endif /* APP_LOOKUP_METHOD */
948 static inline __attribute__((always_inline)) void
949 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
951 struct ether_hdr *eth_hdr;
952 struct ipv4_hdr *ipv4_hdr;
956 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
958 if (m->ol_flags & PKT_RX_IPV4_HDR) {
959 /* Handle IPv4 headers.*/
960 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
961 sizeof(struct ether_hdr));
963 #ifdef DO_RFC_1812_CHECKS
964 /* Check to make sure the packet is valid (RFC1812) */
965 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
971 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
972 qconf->ipv4_lookup_struct);
973 if (dst_port >= RTE_MAX_ETHPORTS ||
974 (enabled_port_mask & 1 << dst_port) == 0)
977 /* 02:00:00:00:00:xx */
978 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
979 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
980 ((uint64_t)dst_port << 40);
982 #ifdef DO_RFC_1812_CHECKS
983 /* Update time to live and header checksum */
984 --(ipv4_hdr->time_to_live);
985 ++(ipv4_hdr->hdr_checksum);
989 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
991 send_single_packet(m, dst_port);
994 /* Handle IPv6 headers.*/
995 struct ipv6_hdr *ipv6_hdr;
997 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
998 sizeof(struct ether_hdr));
1000 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1002 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1005 /* 02:00:00:00:00:xx */
1006 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
1007 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
1008 ((uint64_t)dst_port << 40);
1011 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1013 send_single_packet(m, dst_port);
1018 #ifdef DO_RFC_1812_CHECKS
1020 #define IPV4_MIN_VER_IHL 0x45
1021 #define IPV4_MAX_VER_IHL 0x4f
1022 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1024 /* Minimum value of IPV4 total length (20B) in network byte order. */
1025 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1028 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1029 * - The IP version number must be 4.
1030 * - The IP header length field must be large enough to hold the
1031 * minimum length legal IP datagram (20 bytes = 5 words).
1032 * - The IP total length field must be large enough to hold the IP
1033 * datagram header, whose length is specified in the IP header length
1035 * If we encounter invalid IPV4 packet, then set destination port for it
1036 * to BAD_PORT value.
1038 static inline __attribute__((always_inline)) void
1039 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t flags)
1043 if ((flags & PKT_RX_IPV4_HDR) != 0) {
1045 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1047 ipv4_hdr->time_to_live--;
1048 ipv4_hdr->hdr_checksum++;
1050 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1051 ((uint8_t)ipv4_hdr->total_length == 0 &&
1052 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1059 #define rfc1812_process(mb, dp) do { } while (0)
1060 #endif /* DO_RFC_1812_CHECKS */
1063 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1064 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1066 static inline __attribute__((always_inline)) uint16_t
1067 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1068 uint32_t dst_ipv4, uint8_t portid)
1071 struct ipv6_hdr *ipv6_hdr;
1072 struct ether_hdr *eth_hdr;
1074 if (pkt->ol_flags & PKT_RX_IPV4_HDR) {
1075 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1078 } else if (pkt->ol_flags & PKT_RX_IPV6_HDR) {
1079 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1080 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1081 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1082 ipv6_hdr->dst_addr, &next_hop) != 0)
1092 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1093 uint16_t *dst_port, uint8_t portid)
1095 struct ether_hdr *eth_hdr;
1096 struct ipv4_hdr *ipv4_hdr;
1101 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1102 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1104 dst_ipv4 = ipv4_hdr->dst_addr;
1105 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1106 dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1108 te = _mm_load_si128((__m128i *)eth_hdr);
1112 rfc1812_process(ipv4_hdr, dst_port, pkt->ol_flags);
1114 te = _mm_blend_epi16(te, ve, MASK_ETH);
1115 _mm_store_si128((__m128i *)eth_hdr, te);
1119 * Read ol_flags and destination IPV4 addresses from 4 mbufs.
1122 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], __m128i *dip, uint32_t *flag)
1124 struct ipv4_hdr *ipv4_hdr;
1125 struct ether_hdr *eth_hdr;
1126 uint32_t x0, x1, x2, x3;
1128 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1129 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1130 x0 = ipv4_hdr->dst_addr;
1131 flag[0] = pkt[0]->ol_flags & PKT_RX_IPV4_HDR;
1133 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1134 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1135 x1 = ipv4_hdr->dst_addr;
1136 flag[0] &= pkt[1]->ol_flags;
1138 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1139 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1140 x2 = ipv4_hdr->dst_addr;
1141 flag[0] &= pkt[2]->ol_flags;
1143 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1144 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1145 x3 = ipv4_hdr->dst_addr;
1146 flag[0] &= pkt[3]->ol_flags;
1148 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1152 * Lookup into LPM for destination port.
1153 * If lookup fails, use incoming port (portid) as destination port.
1156 processx4_step2(const struct lcore_conf *qconf, __m128i dip, uint32_t flag,
1157 uint8_t portid, struct rte_mbuf *pkt[FWDSTEP], uint16_t dprt[FWDSTEP])
1160 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1161 4, 5, 6, 7, 0, 1, 2, 3);
1163 /* Byte swap 4 IPV4 addresses. */
1164 dip = _mm_shuffle_epi8(dip, bswap_mask);
1166 /* if all 4 packets are IPV4. */
1167 if (likely(flag != 0)) {
1168 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1171 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1172 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1173 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1174 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1179 * Update source and destination MAC addresses in the ethernet header.
1180 * Perform RFC1812 checks and updates for IPV4 packets.
1183 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1185 __m128i te[FWDSTEP];
1186 __m128i ve[FWDSTEP];
1187 __m128i *p[FWDSTEP];
1189 p[0] = (rte_pktmbuf_mtod(pkt[0], __m128i *));
1190 p[1] = (rte_pktmbuf_mtod(pkt[1], __m128i *));
1191 p[2] = (rte_pktmbuf_mtod(pkt[2], __m128i *));
1192 p[3] = (rte_pktmbuf_mtod(pkt[3], __m128i *));
1194 ve[0] = val_eth[dst_port[0]];
1195 te[0] = _mm_load_si128(p[0]);
1197 ve[1] = val_eth[dst_port[1]];
1198 te[1] = _mm_load_si128(p[1]);
1200 ve[2] = val_eth[dst_port[2]];
1201 te[2] = _mm_load_si128(p[2]);
1203 ve[3] = val_eth[dst_port[3]];
1204 te[3] = _mm_load_si128(p[3]);
1206 /* Update first 12 bytes, keep rest bytes intact. */
1207 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1208 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1209 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1210 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1212 _mm_store_si128(p[0], te[0]);
1213 _mm_store_si128(p[1], te[1]);
1214 _mm_store_si128(p[2], te[2]);
1215 _mm_store_si128(p[3], te[3]);
1217 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1218 &dst_port[0], pkt[0]->ol_flags);
1219 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1220 &dst_port[1], pkt[1]->ol_flags);
1221 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1222 &dst_port[2], pkt[2]->ol_flags);
1223 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1224 &dst_port[3], pkt[3]->ol_flags);
1228 * We group consecutive packets with the same destionation port into one burst.
1229 * To avoid extra latency this is done together with some other packet
1230 * processing, but after we made a final decision about packet's destination.
1231 * To do this we maintain:
1232 * pnum - array of number of consecutive packets with the same dest port for
1233 * each packet in the input burst.
1234 * lp - pointer to the last updated element in the pnum.
1235 * dlp - dest port value lp corresponds to.
1238 #define GRPSZ (1 << FWDSTEP)
1239 #define GRPMSK (GRPSZ - 1)
1241 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1242 if (likely((dlp) == (dcp)[(idx)])) { \
1245 (dlp) = (dcp)[idx]; \
1246 (lp) = (pn) + (idx); \
1252 * Group consecutive packets with the same destination port in bursts of 4.
1253 * Suppose we have array of destionation ports:
1254 * dst_port[] = {a, b, c, d,, e, ... }
1255 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1256 * We doing 4 comparisions at once and the result is 4 bit mask.
1257 * This mask is used as an index into prebuild array of pnum values.
1259 static inline uint16_t *
1260 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1262 static const struct {
1263 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1264 int32_t idx; /* index for new last updated elemnet. */
1265 uint16_t lpv; /* add value to the last updated element. */
1268 /* 0: a != b, b != c, c != d, d != e */
1269 .pnum = UINT64_C(0x0001000100010001),
1274 /* 1: a == b, b != c, c != d, d != e */
1275 .pnum = UINT64_C(0x0001000100010002),
1280 /* 2: a != b, b == c, c != d, d != e */
1281 .pnum = UINT64_C(0x0001000100020001),
1286 /* 3: a == b, b == c, c != d, d != e */
1287 .pnum = UINT64_C(0x0001000100020003),
1292 /* 4: a != b, b != c, c == d, d != e */
1293 .pnum = UINT64_C(0x0001000200010001),
1298 /* 5: a == b, b != c, c == d, d != e */
1299 .pnum = UINT64_C(0x0001000200010002),
1304 /* 6: a != b, b == c, c == d, d != e */
1305 .pnum = UINT64_C(0x0001000200030001),
1310 /* 7: a == b, b == c, c == d, d != e */
1311 .pnum = UINT64_C(0x0001000200030004),
1316 /* 8: a != b, b != c, c != d, d == e */
1317 .pnum = UINT64_C(0x0002000100010001),
1322 /* 9: a == b, b != c, c != d, d == e */
1323 .pnum = UINT64_C(0x0002000100010002),
1328 /* 0xa: a != b, b == c, c != d, d == e */
1329 .pnum = UINT64_C(0x0002000100020001),
1334 /* 0xb: a == b, b == c, c != d, d == e */
1335 .pnum = UINT64_C(0x0002000100020003),
1340 /* 0xc: a != b, b != c, c == d, d == e */
1341 .pnum = UINT64_C(0x0002000300010001),
1346 /* 0xd: a == b, b != c, c == d, d == e */
1347 .pnum = UINT64_C(0x0002000300010002),
1352 /* 0xe: a != b, b == c, c == d, d == e */
1353 .pnum = UINT64_C(0x0002000300040001),
1358 /* 0xf: a == b, b == c, c == d, d == e */
1359 .pnum = UINT64_C(0x0002000300040005),
1366 uint16_t u16[FWDSTEP + 1];
1368 } *pnum = (void *)pn;
1372 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1373 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1374 v = _mm_movemask_ps((__m128)dp1);
1376 /* update last port counter. */
1377 lp[0] += gptbl[v].lpv;
1379 /* if dest port value has changed. */
1381 lp = pnum->u16 + gptbl[v].idx;
1383 pnum->u64 = gptbl[v].pnum;
1389 #endif /* APP_LOOKUP_METHOD */
1391 /* main processing loop */
1393 main_loop(__attribute__((unused)) void *dummy)
1395 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1397 uint64_t prev_tsc, diff_tsc, cur_tsc;
1399 uint8_t portid, queueid;
1400 struct lcore_conf *qconf;
1401 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1402 US_PER_S * BURST_TX_DRAIN_US;
1404 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1405 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1409 uint16_t dst_port[MAX_PKT_BURST];
1410 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1411 uint32_t flag[MAX_PKT_BURST / FWDSTEP];
1412 uint16_t pnum[MAX_PKT_BURST + 1];
1417 lcore_id = rte_lcore_id();
1418 qconf = &lcore_conf[lcore_id];
1420 if (qconf->n_rx_queue == 0) {
1421 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1425 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1427 for (i = 0; i < qconf->n_rx_queue; i++) {
1429 portid = qconf->rx_queue_list[i].port_id;
1430 queueid = qconf->rx_queue_list[i].queue_id;
1431 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1437 cur_tsc = rte_rdtsc();
1440 * TX burst queue drain
1442 diff_tsc = cur_tsc - prev_tsc;
1443 if (unlikely(diff_tsc > drain_tsc)) {
1446 * This could be optimized (use queueid instead of
1447 * portid), but it is not called so often
1449 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1450 if (qconf->tx_mbufs[portid].len == 0)
1453 qconf->tx_mbufs[portid].len,
1455 qconf->tx_mbufs[portid].len = 0;
1462 * Read packet from RX queues
1464 for (i = 0; i < qconf->n_rx_queue; ++i) {
1465 portid = qconf->rx_queue_list[i].port_id;
1466 queueid = qconf->rx_queue_list[i].queue_id;
1467 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1472 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1473 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1476 * Send nb_rx - nb_rx%4 packets
1479 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 4);
1480 for (j = 0; j < n ; j+=4) {
1481 uint32_t ol_flag = pkts_burst[j]->ol_flags
1482 & pkts_burst[j+1]->ol_flags
1483 & pkts_burst[j+2]->ol_flags
1484 & pkts_burst[j+3]->ol_flags;
1485 if (ol_flag & PKT_RX_IPV4_HDR ) {
1486 simple_ipv4_fwd_4pkts(&pkts_burst[j],
1488 } else if (ol_flag & PKT_RX_IPV6_HDR) {
1489 simple_ipv6_fwd_4pkts(&pkts_burst[j],
1492 l3fwd_simple_forward(pkts_burst[j],
1494 l3fwd_simple_forward(pkts_burst[j+1],
1496 l3fwd_simple_forward(pkts_burst[j+2],
1498 l3fwd_simple_forward(pkts_burst[j+3],
1502 for (; j < nb_rx ; j++) {
1503 l3fwd_simple_forward(pkts_burst[j],
1507 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1509 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1510 for (j = 0; j != k; j += FWDSTEP) {
1511 processx4_step1(&pkts_burst[j],
1513 &flag[j / FWDSTEP]);
1516 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1517 for (j = 0; j != k; j += FWDSTEP) {
1518 processx4_step2(qconf, dip[j / FWDSTEP],
1519 flag[j / FWDSTEP], portid,
1520 &pkts_burst[j], &dst_port[j]);
1524 * Finish packet processing and group consecutive
1525 * packets with the same destination port.
1527 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1534 processx4_step3(pkts_burst, dst_port);
1536 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1537 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1539 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1540 processx4_step3(&pkts_burst[j],
1545 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1547 dp2 = _mm_loadu_si128((__m128i *)
1548 &dst_port[j - FWDSTEP + 1]);
1549 lp = port_groupx4(&pnum[j - FWDSTEP],
1554 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1556 dp1 = _mm_srli_si128(dp2,
1558 sizeof(dst_port[0]));
1562 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1564 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1565 lp = port_groupx4(&pnum[j - FWDSTEP], lp,
1569 * remove values added by the last repeated
1573 dlp = dst_port[j - 1];
1575 /* set dlp and lp to the never used values. */
1577 lp = pnum + MAX_PKT_BURST;
1580 /* Process up to last 3 packets one by one. */
1581 switch (nb_rx % FWDSTEP) {
1583 process_packet(qconf, pkts_burst[j],
1584 dst_port + j, portid);
1585 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1588 process_packet(qconf, pkts_burst[j],
1589 dst_port + j, portid);
1590 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1593 process_packet(qconf, pkts_burst[j],
1594 dst_port + j, portid);
1595 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1600 * Send packets out, through destination port.
1601 * Consecuteve pacekts with the same destination port
1602 * are already grouped together.
1603 * If destination port for the packet equals BAD_PORT,
1604 * then free the packet without sending it out.
1606 for (j = 0; j < nb_rx; j += k) {
1614 if (likely(pn != BAD_PORT)) {
1615 send_packetsx4(qconf, pn,
1618 for (m = j; m != j + k; m++)
1619 rte_pktmbuf_free(pkts_burst[m]);
1623 #endif /* APP_LOOKUP_METHOD */
1624 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1626 /* Prefetch first packets */
1627 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1628 rte_prefetch0(rte_pktmbuf_mtod(
1629 pkts_burst[j], void *));
1632 /* Prefetch and forward already prefetched packets */
1633 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1634 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1635 j + PREFETCH_OFFSET], void *));
1636 l3fwd_simple_forward(pkts_burst[j], portid,
1640 /* Forward remaining prefetched packets */
1641 for (; j < nb_rx; j++) {
1642 l3fwd_simple_forward(pkts_burst[j], portid,
1645 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1652 check_lcore_params(void)
1654 uint8_t queue, lcore;
1658 for (i = 0; i < nb_lcore_params; ++i) {
1659 queue = lcore_params[i].queue_id;
1660 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1661 printf("invalid queue number: %hhu\n", queue);
1664 lcore = lcore_params[i].lcore_id;
1665 if (!rte_lcore_is_enabled(lcore)) {
1666 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1669 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1671 printf("warning: lcore %hhu is on socket %d with numa off \n",
1679 check_port_config(const unsigned nb_ports)
1684 for (i = 0; i < nb_lcore_params; ++i) {
1685 portid = lcore_params[i].port_id;
1686 if ((enabled_port_mask & (1 << portid)) == 0) {
1687 printf("port %u is not enabled in port mask\n", portid);
1690 if (portid >= nb_ports) {
1691 printf("port %u is not present on the board\n", portid);
1699 get_port_n_rx_queues(const uint8_t port)
1704 for (i = 0; i < nb_lcore_params; ++i) {
1705 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1706 queue = lcore_params[i].queue_id;
1708 return (uint8_t)(++queue);
1712 init_lcore_rx_queues(void)
1714 uint16_t i, nb_rx_queue;
1717 for (i = 0; i < nb_lcore_params; ++i) {
1718 lcore = lcore_params[i].lcore_id;
1719 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1720 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1721 printf("error: too many queues (%u) for lcore: %u\n",
1722 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1725 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1726 lcore_params[i].port_id;
1727 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1728 lcore_params[i].queue_id;
1729 lcore_conf[lcore].n_rx_queue++;
1737 print_usage(const char *prgname)
1739 printf ("%s [EAL options] -- -p PORTMASK -P"
1740 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1741 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1742 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1743 " -P : enable promiscuous mode\n"
1744 " --config (port,queue,lcore): rx queues configuration\n"
1745 " --no-numa: optional, disable numa awareness\n"
1746 " --ipv6: optional, specify it if running ipv6 packets\n"
1747 " --enable-jumbo: enable jumbo frame"
1748 " which max packet len is PKTLEN in decimal (64-9600)\n"
1749 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1753 static int parse_max_pkt_len(const char *pktlen)
1758 /* parse decimal string */
1759 len = strtoul(pktlen, &end, 10);
1760 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1770 parse_portmask(const char *portmask)
1775 /* parse hexadecimal string */
1776 pm = strtoul(portmask, &end, 16);
1777 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1786 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1788 parse_hash_entry_number(const char *hash_entry_num)
1791 unsigned long hash_en;
1792 /* parse hexadecimal string */
1793 hash_en = strtoul(hash_entry_num, &end, 16);
1794 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
1805 parse_config(const char *q_arg)
1808 const char *p, *p0 = q_arg;
1816 unsigned long int_fld[_NUM_FLD];
1817 char *str_fld[_NUM_FLD];
1821 nb_lcore_params = 0;
1823 while ((p = strchr(p0,'(')) != NULL) {
1825 if((p0 = strchr(p,')')) == NULL)
1829 if(size >= sizeof(s))
1832 snprintf(s, sizeof(s), "%.*s", size, p);
1833 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1835 for (i = 0; i < _NUM_FLD; i++){
1837 int_fld[i] = strtoul(str_fld[i], &end, 0);
1838 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1841 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1842 printf("exceeded max number of lcore params: %hu\n",
1846 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1847 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1848 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1851 lcore_params = lcore_params_array;
1855 #define CMD_LINE_OPT_CONFIG "config"
1856 #define CMD_LINE_OPT_NO_NUMA "no-numa"
1857 #define CMD_LINE_OPT_IPV6 "ipv6"
1858 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
1859 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
1861 /* Parse the argument given in the command line of the application */
1863 parse_args(int argc, char **argv)
1868 char *prgname = argv[0];
1869 static struct option lgopts[] = {
1870 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
1871 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
1872 {CMD_LINE_OPT_IPV6, 0, 0, 0},
1873 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
1874 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
1880 while ((opt = getopt_long(argc, argvopt, "p:P",
1881 lgopts, &option_index)) != EOF) {
1886 enabled_port_mask = parse_portmask(optarg);
1887 if (enabled_port_mask == 0) {
1888 printf("invalid portmask\n");
1889 print_usage(prgname);
1894 printf("Promiscuous mode selected\n");
1900 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
1901 sizeof (CMD_LINE_OPT_CONFIG))) {
1902 ret = parse_config(optarg);
1904 printf("invalid config\n");
1905 print_usage(prgname);
1910 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
1911 sizeof(CMD_LINE_OPT_NO_NUMA))) {
1912 printf("numa is disabled \n");
1916 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1917 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
1918 sizeof(CMD_LINE_OPT_IPV6))) {
1919 printf("ipv6 is specified \n");
1924 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
1925 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
1926 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1928 printf("jumbo frame is enabled - disabling simple TX path\n");
1929 port_conf.rxmode.jumbo_frame = 1;
1931 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1932 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1933 ret = parse_max_pkt_len(optarg);
1934 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1935 printf("invalid packet length\n");
1936 print_usage(prgname);
1939 port_conf.rxmode.max_rx_pkt_len = ret;
1941 printf("set jumbo frame max packet length to %u\n",
1942 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1944 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1945 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
1946 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
1947 ret = parse_hash_entry_number(optarg);
1948 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
1949 hash_entry_number = ret;
1951 printf("invalid hash entry number\n");
1952 print_usage(prgname);
1960 print_usage(prgname);
1966 argv[optind-1] = prgname;
1969 optind = 0; /* reset getopt lib */
1974 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1976 char buf[ETHER_ADDR_FMT_SIZE];
1977 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1978 printf("%s%s", name, buf);
1981 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1983 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
1984 union ipv4_5tuple_host* key2)
1986 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
1987 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
1988 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
1989 key2->port_src = rte_cpu_to_be_16(key1->port_src);
1990 key2->proto = key1->proto;
1996 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
1997 union ipv6_5tuple_host* key2)
2000 for (i = 0; i < 16; i++)
2002 key2->ip_dst[i] = key1->ip_dst[i];
2003 key2->ip_src[i] = key1->ip_src[i];
2005 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2006 key2->port_src = rte_cpu_to_be_16(key1->port_src);
2007 key2->proto = key1->proto;
2014 #define BYTE_VALUE_MAX 256
2015 #define ALL_32_BITS 0xffffffff
2016 #define BIT_8_TO_15 0x0000ff00
2018 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
2022 uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
2024 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2025 for (i = 0; i < array_len; i++) {
2026 struct ipv4_l3fwd_route entry;
2027 union ipv4_5tuple_host newkey;
2028 entry = ipv4_l3fwd_route_array[i];
2029 convert_ipv4_5tuple(&entry.key, &newkey);
2030 ret = rte_hash_add_key (h,(void *) &newkey);
2032 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2033 " to the l3fwd hash.\n", i);
2035 ipv4_l3fwd_out_if[ret] = entry.if_out;
2037 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
2040 #define BIT_16_TO_23 0x00ff0000
2042 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
2046 uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
2048 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2049 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2050 for (i = 0; i < array_len; i++) {
2051 struct ipv6_l3fwd_route entry;
2052 union ipv6_5tuple_host newkey;
2053 entry = ipv6_l3fwd_route_array[i];
2054 convert_ipv6_5tuple(&entry.key, &newkey);
2055 ret = rte_hash_add_key (h, (void *) &newkey);
2057 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2058 " to the l3fwd hash.\n", i);
2060 ipv6_l3fwd_out_if[ret] = entry.if_out;
2062 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
2065 #define NUMBER_PORT_USED 4
2067 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
2068 unsigned int nr_flow)
2071 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2072 for (i = 0; i < nr_flow; i++) {
2073 struct ipv4_l3fwd_route entry;
2074 union ipv4_5tuple_host newkey;
2075 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2076 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2077 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2078 /* Create the ipv4 exact match flow */
2079 memset(&entry, 0, sizeof(entry));
2080 switch (i & (NUMBER_PORT_USED -1)) {
2082 entry = ipv4_l3fwd_route_array[0];
2083 entry.key.ip_dst = IPv4(101,c,b,a);
2086 entry = ipv4_l3fwd_route_array[1];
2087 entry.key.ip_dst = IPv4(201,c,b,a);
2090 entry = ipv4_l3fwd_route_array[2];
2091 entry.key.ip_dst = IPv4(111,c,b,a);
2094 entry = ipv4_l3fwd_route_array[3];
2095 entry.key.ip_dst = IPv4(211,c,b,a);
2098 convert_ipv4_5tuple(&entry.key, &newkey);
2099 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2101 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2103 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2106 printf("Hash: Adding 0x%x keys\n", nr_flow);
2110 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
2111 unsigned int nr_flow)
2114 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2115 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2116 for (i = 0; i < nr_flow; i++) {
2117 struct ipv6_l3fwd_route entry;
2118 union ipv6_5tuple_host newkey;
2119 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2120 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2121 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2122 /* Create the ipv6 exact match flow */
2123 memset(&entry, 0, sizeof(entry));
2124 switch (i & (NUMBER_PORT_USED - 1)) {
2125 case 0: entry = ipv6_l3fwd_route_array[0]; break;
2126 case 1: entry = ipv6_l3fwd_route_array[1]; break;
2127 case 2: entry = ipv6_l3fwd_route_array[2]; break;
2128 case 3: entry = ipv6_l3fwd_route_array[3]; break;
2130 entry.key.ip_dst[13] = c;
2131 entry.key.ip_dst[14] = b;
2132 entry.key.ip_dst[15] = a;
2133 convert_ipv6_5tuple(&entry.key, &newkey);
2134 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2136 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2138 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2141 printf("Hash: Adding 0x%x keys\n", nr_flow);
2145 setup_hash(int socketid)
2147 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2149 .entries = L3FWD_HASH_ENTRIES,
2150 .bucket_entries = 4,
2151 .key_len = sizeof(union ipv4_5tuple_host),
2152 .hash_func = ipv4_hash_crc,
2153 .hash_func_init_val = 0,
2156 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2158 .entries = L3FWD_HASH_ENTRIES,
2159 .bucket_entries = 4,
2160 .key_len = sizeof(union ipv6_5tuple_host),
2161 .hash_func = ipv6_hash_crc,
2162 .hash_func_init_val = 0,
2167 /* create ipv4 hash */
2168 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2169 ipv4_l3fwd_hash_params.name = s;
2170 ipv4_l3fwd_hash_params.socket_id = socketid;
2171 ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2172 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2173 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2174 "socket %d\n", socketid);
2176 /* create ipv6 hash */
2177 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2178 ipv6_l3fwd_hash_params.name = s;
2179 ipv6_l3fwd_hash_params.socket_id = socketid;
2180 ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2181 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2182 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2183 "socket %d\n", socketid);
2185 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2186 /* For testing hash matching with a large number of flows we
2187 * generate millions of IP 5-tuples with an incremented dst
2188 * address to initialize the hash table. */
2190 /* populate the ipv4 hash */
2191 populate_ipv4_many_flow_into_table(
2192 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2194 /* populate the ipv6 hash */
2195 populate_ipv6_many_flow_into_table(
2196 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2199 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2201 /* populate the ipv4 hash */
2202 populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2204 /* populate the ipv6 hash */
2205 populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2211 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2213 setup_lpm(int socketid)
2215 struct rte_lpm6_config config;
2220 /* create the LPM table */
2221 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2222 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2223 IPV4_L3FWD_LPM_MAX_RULES, 0);
2224 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2225 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2226 " on socket %d\n", socketid);
2228 /* populate the LPM table */
2229 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2231 /* skip unused ports */
2232 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2233 enabled_port_mask) == 0)
2236 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2237 ipv4_l3fwd_route_array[i].ip,
2238 ipv4_l3fwd_route_array[i].depth,
2239 ipv4_l3fwd_route_array[i].if_out);
2242 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2243 "l3fwd LPM table on socket %d\n",
2247 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2248 (unsigned)ipv4_l3fwd_route_array[i].ip,
2249 ipv4_l3fwd_route_array[i].depth,
2250 ipv4_l3fwd_route_array[i].if_out);
2253 /* create the LPM6 table */
2254 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2256 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2257 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2259 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2261 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2262 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2263 " on socket %d\n", socketid);
2265 /* populate the LPM table */
2266 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2268 /* skip unused ports */
2269 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2270 enabled_port_mask) == 0)
2273 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2274 ipv6_l3fwd_route_array[i].ip,
2275 ipv6_l3fwd_route_array[i].depth,
2276 ipv6_l3fwd_route_array[i].if_out);
2279 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2280 "l3fwd LPM table on socket %d\n",
2284 printf("LPM: Adding route %s / %d (%d)\n",
2286 ipv6_l3fwd_route_array[i].depth,
2287 ipv6_l3fwd_route_array[i].if_out);
2293 init_mem(unsigned nb_mbuf)
2295 struct lcore_conf *qconf;
2300 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2301 if (rte_lcore_is_enabled(lcore_id) == 0)
2305 socketid = rte_lcore_to_socket_id(lcore_id);
2309 if (socketid >= NB_SOCKETS) {
2310 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2311 socketid, lcore_id, NB_SOCKETS);
2313 if (pktmbuf_pool[socketid] == NULL) {
2314 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2315 pktmbuf_pool[socketid] =
2316 rte_pktmbuf_pool_create(s, nb_mbuf,
2317 MEMPOOL_CACHE_SIZE, 0,
2318 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
2319 if (pktmbuf_pool[socketid] == NULL)
2320 rte_exit(EXIT_FAILURE,
2321 "Cannot init mbuf pool on socket %d\n", socketid);
2323 printf("Allocated mbuf pool on socket %d\n", socketid);
2325 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2326 setup_lpm(socketid);
2328 setup_hash(socketid);
2331 qconf = &lcore_conf[lcore_id];
2332 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2333 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2338 /* Check the link status of all ports in up to 9s, and print them finally */
2340 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2342 #define CHECK_INTERVAL 100 /* 100ms */
2343 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2344 uint8_t portid, count, all_ports_up, print_flag = 0;
2345 struct rte_eth_link link;
2347 printf("\nChecking link status");
2349 for (count = 0; count <= MAX_CHECK_TIME; count++) {
2351 for (portid = 0; portid < port_num; portid++) {
2352 if ((port_mask & (1 << portid)) == 0)
2354 memset(&link, 0, sizeof(link));
2355 rte_eth_link_get_nowait(portid, &link);
2356 /* print link status if flag set */
2357 if (print_flag == 1) {
2358 if (link.link_status)
2359 printf("Port %d Link Up - speed %u "
2360 "Mbps - %s\n", (uint8_t)portid,
2361 (unsigned)link.link_speed,
2362 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2363 ("full-duplex") : ("half-duplex\n"));
2365 printf("Port %d Link Down\n",
2369 /* clear all_ports_up flag if any link down */
2370 if (link.link_status == 0) {
2375 /* after finally printing all link status, get out */
2376 if (print_flag == 1)
2379 if (all_ports_up == 0) {
2382 rte_delay_ms(CHECK_INTERVAL);
2385 /* set the print_flag if all ports up or timeout */
2386 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2394 main(int argc, char **argv)
2396 struct lcore_conf *qconf;
2397 struct rte_eth_dev_info dev_info;
2398 struct rte_eth_txconf *txconf;
2403 uint32_t n_tx_queue, nb_lcores;
2404 uint8_t portid, nb_rx_queue, queue, socketid;
2407 ret = rte_eal_init(argc, argv);
2409 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2413 /* parse application arguments (after the EAL ones) */
2414 ret = parse_args(argc, argv);
2416 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2418 if (check_lcore_params() < 0)
2419 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2421 ret = init_lcore_rx_queues();
2423 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2425 nb_ports = rte_eth_dev_count();
2426 if (nb_ports > RTE_MAX_ETHPORTS)
2427 nb_ports = RTE_MAX_ETHPORTS;
2429 if (check_port_config(nb_ports) < 0)
2430 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2432 nb_lcores = rte_lcore_count();
2434 /* initialize all ports */
2435 for (portid = 0; portid < nb_ports; portid++) {
2436 /* skip ports that are not enabled */
2437 if ((enabled_port_mask & (1 << portid)) == 0) {
2438 printf("\nSkipping disabled port %d\n", portid);
2443 printf("Initializing port %d ... ", portid );
2446 nb_rx_queue = get_port_n_rx_queues(portid);
2447 n_tx_queue = nb_lcores;
2448 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2449 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2450 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2451 nb_rx_queue, (unsigned)n_tx_queue );
2452 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2453 (uint16_t)n_tx_queue, &port_conf);
2455 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2458 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2459 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2463 * prepare dst and src MACs for each port.
2465 *(uint64_t *)(val_eth + portid) =
2466 ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2467 ether_addr_copy(&ports_eth_addr[portid],
2468 (struct ether_addr *)(val_eth + portid) + 1);
2471 ret = init_mem(NB_MBUF);
2473 rte_exit(EXIT_FAILURE, "init_mem failed\n");
2475 /* init one TX queue per couple (lcore,port) */
2477 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2478 if (rte_lcore_is_enabled(lcore_id) == 0)
2482 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2486 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2489 rte_eth_dev_info_get(portid, &dev_info);
2490 txconf = &dev_info.default_txconf;
2491 if (port_conf.rxmode.jumbo_frame)
2492 txconf->txq_flags = 0;
2493 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2496 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2497 "port=%d\n", ret, portid);
2499 qconf = &lcore_conf[lcore_id];
2500 qconf->tx_queue_id[portid] = queueid;
2506 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2507 if (rte_lcore_is_enabled(lcore_id) == 0)
2509 qconf = &lcore_conf[lcore_id];
2510 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2512 /* init RX queues */
2513 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2514 portid = qconf->rx_queue_list[queue].port_id;
2515 queueid = qconf->rx_queue_list[queue].queue_id;
2518 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2522 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2525 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2528 pktmbuf_pool[socketid]);
2530 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2531 "port=%d\n", ret, portid);
2538 for (portid = 0; portid < nb_ports; portid++) {
2539 if ((enabled_port_mask & (1 << portid)) == 0) {
2543 ret = rte_eth_dev_start(portid);
2545 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2549 * If enabled, put device in promiscuous mode.
2550 * This allows IO forwarding mode to forward packets
2551 * to itself through 2 cross-connected ports of the
2555 rte_eth_promiscuous_enable(portid);
2558 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2560 /* launch per-lcore init on every lcore */
2561 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2562 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2563 if (rte_eal_wait_lcore(lcore_id) < 0)