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>
46 #include <rte_common_vect.h>
47 #include <rte_byteorder.h>
49 #include <rte_memory.h>
50 #include <rte_memcpy.h>
51 #include <rte_memzone.h>
52 #include <rte_tailq.h>
54 #include <rte_per_lcore.h>
55 #include <rte_launch.h>
56 #include <rte_atomic.h>
57 #include <rte_cycles.h>
58 #include <rte_prefetch.h>
59 #include <rte_lcore.h>
60 #include <rte_per_lcore.h>
61 #include <rte_branch_prediction.h>
62 #include <rte_interrupts.h>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
74 #include <rte_string_fns.h>
78 #define APP_LOOKUP_EXACT_MATCH 0
79 #define APP_LOOKUP_LPM 1
80 #define DO_RFC_1812_CHECKS
82 #ifndef APP_LOOKUP_METHOD
83 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
87 * When set to zero, simple forwaring path is eanbled.
88 * When set to one, optimized forwarding path is enabled.
89 * Note that LPM optimisation path uses SSE4.1 instructions.
91 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
92 #define ENABLE_MULTI_BUFFER_OPTIMIZE 0
94 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1
97 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
99 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
101 #include <rte_lpm6.h>
103 #error "APP_LOOKUP_METHOD set to incorrect value"
107 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
108 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
109 #define IPv6_BYTES(addr) \
110 addr[0], addr[1], addr[2], addr[3], \
111 addr[4], addr[5], addr[6], addr[7], \
112 addr[8], addr[9], addr[10], addr[11],\
113 addr[12], addr[13],addr[14], addr[15]
117 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
119 #define MAX_JUMBO_PKT_LEN 9600
121 #define IPV6_ADDR_LEN 16
123 #define MEMPOOL_CACHE_SIZE 256
125 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
128 * This expression is used to calculate the number of mbufs needed depending on user input, taking
129 * into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
130 * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
133 #define NB_MBUF RTE_MAX ( \
134 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
135 nb_ports*nb_lcores*MAX_PKT_BURST + \
136 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
137 nb_lcores*MEMPOOL_CACHE_SIZE), \
141 * RX and TX Prefetch, Host, and Write-back threshold values should be
142 * carefully set for optimal performance. Consult the network
143 * controller's datasheet and supporting DPDK documentation for guidance
144 * on how these parameters should be set.
146 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
147 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
148 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
151 * These default values are optimized for use with the Intel(R) 82599 10 GbE
152 * Controller and the DPDK ixgbe PMD. Consider using other values for other
153 * network controllers and/or network drivers.
155 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
156 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
157 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
159 #define MAX_PKT_BURST 32
160 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
163 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
165 #define MAX_TX_BURST (MAX_PKT_BURST / 2)
169 /* Configure how many packets ahead to prefetch, when reading packets */
170 #define PREFETCH_OFFSET 3
172 /* Used to mark destination port as 'invalid'. */
173 #define BAD_PORT ((uint16_t)-1)
178 * Configurable number of RX/TX ring descriptors
180 #define RTE_TEST_RX_DESC_DEFAULT 128
181 #define RTE_TEST_TX_DESC_DEFAULT 512
182 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
183 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
185 /* ethernet addresses of ports */
186 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
188 static __m128i val_eth[RTE_MAX_ETHPORTS];
190 /* replace first 12B of the ethernet header. */
191 #define MASK_ETH 0x3f
193 /* mask of enabled ports */
194 static uint32_t enabled_port_mask = 0;
195 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
196 static int numa_on = 1; /**< NUMA is enabled by default. */
198 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
199 static int ipv6 = 0; /**< ipv6 is false by default. */
204 struct rte_mbuf *m_table[MAX_PKT_BURST];
207 struct lcore_rx_queue {
210 } __rte_cache_aligned;
212 #define MAX_RX_QUEUE_PER_LCORE 16
213 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
214 #define MAX_RX_QUEUE_PER_PORT 128
216 #define MAX_LCORE_PARAMS 1024
217 struct lcore_params {
221 } __rte_cache_aligned;
223 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
224 static struct lcore_params lcore_params_array_default[] = {
236 static struct lcore_params * lcore_params = lcore_params_array_default;
237 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
238 sizeof(lcore_params_array_default[0]);
240 static struct rte_eth_conf port_conf = {
242 .mq_mode = ETH_MQ_RX_RSS,
243 .max_rx_pkt_len = ETHER_MAX_LEN,
245 .header_split = 0, /**< Header Split disabled */
246 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
247 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
248 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
249 .hw_strip_crc = 0, /**< CRC stripped by hardware */
254 .rss_hf = ETH_RSS_IP,
258 .mq_mode = ETH_MQ_TX_NONE,
262 static const struct rte_eth_rxconf rx_conf = {
264 .pthresh = RX_PTHRESH,
265 .hthresh = RX_HTHRESH,
266 .wthresh = RX_WTHRESH,
268 .rx_free_thresh = 32,
271 static struct rte_eth_txconf tx_conf = {
273 .pthresh = TX_PTHRESH,
274 .hthresh = TX_HTHRESH,
275 .wthresh = TX_WTHRESH,
277 .tx_free_thresh = 0, /* Use PMD default values */
278 .tx_rs_thresh = 0, /* Use PMD default values */
279 .txq_flags = (ETH_TXQ_FLAGS_NOMULTSEGS |
280 ETH_TXQ_FLAGS_NOVLANOFFL |
281 ETH_TXQ_FLAGS_NOXSUMSCTP |
282 ETH_TXQ_FLAGS_NOXSUMUDP |
283 ETH_TXQ_FLAGS_NOXSUMTCP)
287 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
289 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
291 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
292 #include <rte_hash_crc.h>
293 #define DEFAULT_HASH_FUNC rte_hash_crc
295 #include <rte_jhash.h>
296 #define DEFAULT_HASH_FUNC rte_jhash
305 } __attribute__((__packed__));
307 union ipv4_5tuple_host {
320 #define XMM_NUM_IN_IPV6_5TUPLE 3
323 uint8_t ip_dst[IPV6_ADDR_LEN];
324 uint8_t ip_src[IPV6_ADDR_LEN];
328 } __attribute__((__packed__));
330 union ipv6_5tuple_host {
335 uint8_t ip_src[IPV6_ADDR_LEN];
336 uint8_t ip_dst[IPV6_ADDR_LEN];
341 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
344 struct ipv4_l3fwd_route {
345 struct ipv4_5tuple key;
349 struct ipv6_l3fwd_route {
350 struct ipv6_5tuple key;
354 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
355 {{IPv4(101,0,0,0), IPv4(100,10,0,1), 101, 11, IPPROTO_TCP}, 0},
356 {{IPv4(201,0,0,0), IPv4(200,20,0,1), 102, 12, IPPROTO_TCP}, 1},
357 {{IPv4(111,0,0,0), IPv4(100,30,0,1), 101, 11, IPPROTO_TCP}, 2},
358 {{IPv4(211,0,0,0), IPv4(200,40,0,1), 102, 12, IPPROTO_TCP}, 3},
361 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
363 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
364 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
365 101, 11, IPPROTO_TCP}, 0},
368 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
369 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
370 102, 12, IPPROTO_TCP}, 1},
373 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
374 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
375 101, 11, IPPROTO_TCP}, 2},
378 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
379 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
380 102, 12, IPPROTO_TCP}, 3},
383 typedef struct rte_hash lookup_struct_t;
384 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
385 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
387 #ifdef RTE_ARCH_X86_64
388 /* default to 4 million hash entries (approx) */
389 #define L3FWD_HASH_ENTRIES 1024*1024*4
391 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
392 #define L3FWD_HASH_ENTRIES 1024*1024*1
394 #define HASH_ENTRY_NUMBER_DEFAULT 4
396 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
398 static inline uint32_t
399 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
402 const union ipv4_5tuple_host *k;
408 p = (const uint32_t *)&k->port_src;
410 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
411 init_val = rte_hash_crc_4byte(t, init_val);
412 init_val = rte_hash_crc_4byte(k->ip_src, init_val);
413 init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
414 init_val = rte_hash_crc_4byte(*p, init_val);
415 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
416 init_val = rte_jhash_1word(t, init_val);
417 init_val = rte_jhash_1word(k->ip_src, init_val);
418 init_val = rte_jhash_1word(k->ip_dst, init_val);
419 init_val = rte_jhash_1word(*p, init_val);
420 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
424 static inline uint32_t
425 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
427 const union ipv6_5tuple_host *k;
430 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
431 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
432 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
433 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
437 p = (const uint32_t *)&k->port_src;
439 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
440 ip_src0 = (const uint32_t *) k->ip_src;
441 ip_src1 = (const uint32_t *)(k->ip_src+4);
442 ip_src2 = (const uint32_t *)(k->ip_src+8);
443 ip_src3 = (const uint32_t *)(k->ip_src+12);
444 ip_dst0 = (const uint32_t *) k->ip_dst;
445 ip_dst1 = (const uint32_t *)(k->ip_dst+4);
446 ip_dst2 = (const uint32_t *)(k->ip_dst+8);
447 ip_dst3 = (const uint32_t *)(k->ip_dst+12);
448 init_val = rte_hash_crc_4byte(t, init_val);
449 init_val = rte_hash_crc_4byte(*ip_src0, init_val);
450 init_val = rte_hash_crc_4byte(*ip_src1, init_val);
451 init_val = rte_hash_crc_4byte(*ip_src2, init_val);
452 init_val = rte_hash_crc_4byte(*ip_src3, init_val);
453 init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
454 init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
455 init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
456 init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
457 init_val = rte_hash_crc_4byte(*p, init_val);
458 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
459 init_val = rte_jhash_1word(t, init_val);
460 init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
461 init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
462 init_val = rte_jhash_1word(*p, init_val);
463 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
467 #define IPV4_L3FWD_NUM_ROUTES \
468 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
470 #define IPV6_L3FWD_NUM_ROUTES \
471 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
473 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
474 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
478 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
479 struct ipv4_l3fwd_route {
485 struct ipv6_l3fwd_route {
491 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
492 {IPv4(1,1,1,0), 24, 0},
493 {IPv4(2,1,1,0), 24, 1},
494 {IPv4(3,1,1,0), 24, 2},
495 {IPv4(4,1,1,0), 24, 3},
496 {IPv4(5,1,1,0), 24, 4},
497 {IPv4(6,1,1,0), 24, 5},
498 {IPv4(7,1,1,0), 24, 6},
499 {IPv4(8,1,1,0), 24, 7},
502 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
503 {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
504 {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
505 {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
506 {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
507 {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
508 {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
509 {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
510 {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
513 #define IPV4_L3FWD_NUM_ROUTES \
514 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
515 #define IPV6_L3FWD_NUM_ROUTES \
516 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
518 #define IPV4_L3FWD_LPM_MAX_RULES 1024
519 #define IPV6_L3FWD_LPM_MAX_RULES 1024
520 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
522 typedef struct rte_lpm lookup_struct_t;
523 typedef struct rte_lpm6 lookup6_struct_t;
524 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
525 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
530 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
531 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
532 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
533 lookup_struct_t * ipv4_lookup_struct;
534 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
535 lookup6_struct_t * ipv6_lookup_struct;
537 lookup_struct_t * ipv6_lookup_struct;
539 } __rte_cache_aligned;
541 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
543 /* Send burst of packets on an output interface */
545 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
547 struct rte_mbuf **m_table;
551 queueid = qconf->tx_queue_id[port];
552 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
554 ret = rte_eth_tx_burst(port, queueid, m_table, n);
555 if (unlikely(ret < n)) {
557 rte_pktmbuf_free(m_table[ret]);
564 /* Enqueue a single packet, and send burst if queue is filled */
566 send_single_packet(struct rte_mbuf *m, uint8_t port)
570 struct lcore_conf *qconf;
572 lcore_id = rte_lcore_id();
574 qconf = &lcore_conf[lcore_id];
575 len = qconf->tx_mbufs[port].len;
576 qconf->tx_mbufs[port].m_table[len] = m;
579 /* enough pkts to be sent */
580 if (unlikely(len == MAX_PKT_BURST)) {
581 send_burst(qconf, MAX_PKT_BURST, port);
585 qconf->tx_mbufs[port].len = len;
589 static inline __attribute__((always_inline)) void
590 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
591 struct rte_mbuf *m[], uint32_t num)
595 len = qconf->tx_mbufs[port].len;
598 * If TX buffer for that queue is empty, and we have enough packets,
599 * then send them straightway.
601 if (num >= MAX_TX_BURST && len == 0) {
602 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
603 if (unlikely(n < num)) {
605 rte_pktmbuf_free(m[n]);
612 * Put packets into TX buffer for that queue.
616 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
619 switch (n % FWDSTEP) {
622 qconf->tx_mbufs[port].m_table[len + j] = m[j];
625 qconf->tx_mbufs[port].m_table[len + j] = m[j];
628 qconf->tx_mbufs[port].m_table[len + j] = m[j];
631 qconf->tx_mbufs[port].m_table[len + j] = m[j];
638 /* enough pkts to be sent */
639 if (unlikely(len == MAX_PKT_BURST)) {
641 send_burst(qconf, MAX_PKT_BURST, port);
643 /* copy rest of the packets into the TX buffer. */
646 switch (len % FWDSTEP) {
649 qconf->tx_mbufs[port].m_table[j] = m[n + j];
652 qconf->tx_mbufs[port].m_table[j] = m[n + j];
655 qconf->tx_mbufs[port].m_table[j] = m[n + j];
658 qconf->tx_mbufs[port].m_table[j] = m[n + j];
664 qconf->tx_mbufs[port].len = len;
667 #ifdef DO_RFC_1812_CHECKS
669 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
671 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
673 * 1. The packet length reported by the Link Layer must be large
674 * enough to hold the minimum length legal IP datagram (20 bytes).
676 if (link_len < sizeof(struct ipv4_hdr))
679 /* 2. The IP checksum must be correct. */
680 /* this is checked in H/W */
683 * 3. The IP version number must be 4. If the version number is not 4
684 * then the packet may be another version of IP, such as IPng or
687 if (((pkt->version_ihl) >> 4) != 4)
690 * 4. The IP header length field must be large enough to hold the
691 * minimum length legal IP datagram (20 bytes = 5 words).
693 if ((pkt->version_ihl & 0xf) < 5)
697 * 5. The IP total length field must be large enough to hold the IP
698 * datagram header, whose length is specified in the IP header length
701 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
708 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
710 static __m128i mask0;
711 static __m128i mask1;
712 static __m128i mask2;
713 static inline uint8_t
714 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
717 union ipv4_5tuple_host key;
719 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
720 __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
721 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
722 key.xmm = _mm_and_si128(data, mask0);
723 /* Find destination port */
724 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
725 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
728 static inline uint8_t
729 get_ipv6_dst_port(void *ipv6_hdr, uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
732 union ipv6_5tuple_host key;
734 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
735 __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
736 __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
737 __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
738 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
739 key.xmm[0] = _mm_and_si128(data0, mask1);
740 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
742 /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
743 key.xmm[2] = _mm_and_si128(data2, mask2);
745 /* Find destination port */
746 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
747 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
751 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
753 static inline uint8_t
754 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
758 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
759 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
760 &next_hop) == 0) ? next_hop : portid);
763 static inline uint8_t
764 get_ipv6_dst_port(void *ipv6_hdr, uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
767 return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
768 ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
773 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
774 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
775 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf);
777 #define MASK_ALL_PKTS 0xf
778 #define EXECLUDE_1ST_PKT 0xe
779 #define EXECLUDE_2ND_PKT 0xd
780 #define EXECLUDE_3RD_PKT 0xb
781 #define EXECLUDE_4TH_PKT 0x7
784 simple_ipv4_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
786 struct ether_hdr *eth_hdr[4];
787 struct ipv4_hdr *ipv4_hdr[4];
788 void *d_addr_bytes[4];
791 union ipv4_5tuple_host key[4];
794 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
795 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
796 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
797 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
799 /* Handle IPv4 headers.*/
800 ipv4_hdr[0] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
801 sizeof(struct ether_hdr));
802 ipv4_hdr[1] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
803 sizeof(struct ether_hdr));
804 ipv4_hdr[2] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
805 sizeof(struct ether_hdr));
806 ipv4_hdr[3] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
807 sizeof(struct ether_hdr));
809 #ifdef DO_RFC_1812_CHECKS
810 /* Check to make sure the packet is valid (RFC1812) */
811 uint8_t valid_mask = MASK_ALL_PKTS;
812 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
813 rte_pktmbuf_free(m[0]);
814 valid_mask &= EXECLUDE_1ST_PKT;
816 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
817 rte_pktmbuf_free(m[1]);
818 valid_mask &= EXECLUDE_2ND_PKT;
820 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
821 rte_pktmbuf_free(m[2]);
822 valid_mask &= EXECLUDE_3RD_PKT;
824 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
825 rte_pktmbuf_free(m[3]);
826 valid_mask &= EXECLUDE_4TH_PKT;
828 if (unlikely(valid_mask != MASK_ALL_PKTS)) {
829 if (valid_mask == 0){
833 for (i = 0; i < 4; i++) {
834 if ((0x1 << i) & valid_mask) {
835 l3fwd_simple_forward(m[i], portid, qconf);
841 #endif // End of #ifdef DO_RFC_1812_CHECKS
843 data[0] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[0], unsigned char *) +
844 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
845 data[1] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[1], unsigned char *) +
846 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
847 data[2] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[2], unsigned char *) +
848 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
849 data[3] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[3], unsigned char *) +
850 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
852 key[0].xmm = _mm_and_si128(data[0], mask0);
853 key[1].xmm = _mm_and_si128(data[1], mask0);
854 key[2].xmm = _mm_and_si128(data[2], mask0);
855 key[3].xmm = _mm_and_si128(data[3], mask0);
857 const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
858 rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 4, ret);
859 dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
860 dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
861 dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
862 dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
864 if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
865 dst_port[0] = portid;
866 if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
867 dst_port[1] = portid;
868 if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
869 dst_port[2] = portid;
870 if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
871 dst_port[3] = portid;
873 /* 02:00:00:00:00:xx */
874 d_addr_bytes[0] = ð_hdr[0]->d_addr.addr_bytes[0];
875 d_addr_bytes[1] = ð_hdr[1]->d_addr.addr_bytes[0];
876 d_addr_bytes[2] = ð_hdr[2]->d_addr.addr_bytes[0];
877 d_addr_bytes[3] = ð_hdr[3]->d_addr.addr_bytes[0];
878 *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
879 *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
880 *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
881 *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
883 #ifdef DO_RFC_1812_CHECKS
884 /* Update time to live and header checksum */
885 --(ipv4_hdr[0]->time_to_live);
886 --(ipv4_hdr[1]->time_to_live);
887 --(ipv4_hdr[2]->time_to_live);
888 --(ipv4_hdr[3]->time_to_live);
889 ++(ipv4_hdr[0]->hdr_checksum);
890 ++(ipv4_hdr[1]->hdr_checksum);
891 ++(ipv4_hdr[2]->hdr_checksum);
892 ++(ipv4_hdr[3]->hdr_checksum);
896 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
897 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
898 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
899 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
901 send_single_packet(m[0], (uint8_t)dst_port[0]);
902 send_single_packet(m[1], (uint8_t)dst_port[1]);
903 send_single_packet(m[2], (uint8_t)dst_port[2]);
904 send_single_packet(m[3], (uint8_t)dst_port[3]);
908 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
909 union ipv6_5tuple_host * key)
911 __m128i tmpdata0 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
912 + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
913 __m128i tmpdata1 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
914 + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
916 __m128i tmpdata2 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
917 + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
918 + sizeof(__m128i) + sizeof(__m128i)));
919 key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
920 key->xmm[1] = tmpdata1;
921 key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
926 simple_ipv6_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
928 struct ether_hdr *eth_hdr[4];
929 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[4];
930 void *d_addr_bytes[4];
933 union ipv6_5tuple_host key[4];
935 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
936 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
937 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
938 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
940 /* Handle IPv6 headers.*/
941 ipv6_hdr[0] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
942 sizeof(struct ether_hdr));
943 ipv6_hdr[1] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
944 sizeof(struct ether_hdr));
945 ipv6_hdr[2] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
946 sizeof(struct ether_hdr));
947 ipv6_hdr[3] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
948 sizeof(struct ether_hdr));
950 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
951 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
952 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
953 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
955 const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
956 rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
957 dst_port[0] = (uint8_t) ((ret[0] < 0)? portid:ipv6_l3fwd_out_if[ret[0]]);
958 dst_port[1] = (uint8_t) ((ret[1] < 0)? portid:ipv6_l3fwd_out_if[ret[1]]);
959 dst_port[2] = (uint8_t) ((ret[2] < 0)? portid:ipv6_l3fwd_out_if[ret[2]]);
960 dst_port[3] = (uint8_t) ((ret[3] < 0)? portid:ipv6_l3fwd_out_if[ret[3]]);
962 if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
963 dst_port[0] = portid;
964 if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
965 dst_port[1] = portid;
966 if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
967 dst_port[2] = portid;
968 if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
969 dst_port[3] = portid;
971 /* 02:00:00:00:00:xx */
972 d_addr_bytes[0] = ð_hdr[0]->d_addr.addr_bytes[0];
973 d_addr_bytes[1] = ð_hdr[1]->d_addr.addr_bytes[0];
974 d_addr_bytes[2] = ð_hdr[2]->d_addr.addr_bytes[0];
975 d_addr_bytes[3] = ð_hdr[3]->d_addr.addr_bytes[0];
976 *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
977 *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
978 *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
979 *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
982 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr);
983 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr);
984 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr);
985 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr);
987 send_single_packet(m[0], (uint8_t)dst_port[0]);
988 send_single_packet(m[1], (uint8_t)dst_port[1]);
989 send_single_packet(m[2], (uint8_t)dst_port[2]);
990 send_single_packet(m[3], (uint8_t)dst_port[3]);
993 #endif /* APP_LOOKUP_METHOD */
995 static inline __attribute__((always_inline)) void
996 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
998 struct ether_hdr *eth_hdr;
999 struct ipv4_hdr *ipv4_hdr;
1003 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1005 if (m->ol_flags & PKT_RX_IPV4_HDR) {
1006 /* Handle IPv4 headers.*/
1007 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
1008 sizeof(struct ether_hdr));
1010 #ifdef DO_RFC_1812_CHECKS
1011 /* Check to make sure the packet is valid (RFC1812) */
1012 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1013 rte_pktmbuf_free(m);
1018 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1019 qconf->ipv4_lookup_struct);
1020 if (dst_port >= RTE_MAX_ETHPORTS ||
1021 (enabled_port_mask & 1 << dst_port) == 0)
1024 /* 02:00:00:00:00:xx */
1025 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
1026 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
1027 ((uint64_t)dst_port << 40);
1029 #ifdef DO_RFC_1812_CHECKS
1030 /* Update time to live and header checksum */
1031 --(ipv4_hdr->time_to_live);
1032 ++(ipv4_hdr->hdr_checksum);
1036 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1038 send_single_packet(m, dst_port);
1041 /* Handle IPv6 headers.*/
1042 struct ipv6_hdr *ipv6_hdr;
1044 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
1045 sizeof(struct ether_hdr));
1047 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1049 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1052 /* 02:00:00:00:00:xx */
1053 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
1054 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
1055 ((uint64_t)dst_port << 40);
1058 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
1060 send_single_packet(m, dst_port);
1065 #ifdef DO_RFC_1812_CHECKS
1067 #define IPV4_MIN_VER_IHL 0x45
1068 #define IPV4_MAX_VER_IHL 0x4f
1069 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1071 /* Minimum value of IPV4 total length (20B) in network byte order. */
1072 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1075 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1076 * - The IP version number must be 4.
1077 * - The IP header length field must be large enough to hold the
1078 * minimum length legal IP datagram (20 bytes = 5 words).
1079 * - The IP total length field must be large enough to hold the IP
1080 * datagram header, whose length is specified in the IP header length
1082 * If we encounter invalid IPV4 packet, then set destination port for it
1083 * to BAD_PORT value.
1085 static inline __attribute__((always_inline)) void
1086 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t flags)
1090 if ((flags & PKT_RX_IPV4_HDR) != 0) {
1092 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1094 ipv4_hdr->time_to_live--;
1095 ipv4_hdr->hdr_checksum++;
1097 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1098 ((uint8_t)ipv4_hdr->total_length == 0 &&
1099 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1106 #define rfc1812_process(mb, dp) do { } while (0)
1107 #endif /* DO_RFC_1812_CHECKS */
1110 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1111 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1113 static inline __attribute__((always_inline)) uint16_t
1114 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1115 uint32_t dst_ipv4, uint8_t portid)
1118 struct ipv6_hdr *ipv6_hdr;
1119 struct ether_hdr *eth_hdr;
1121 if (pkt->ol_flags & PKT_RX_IPV4_HDR) {
1122 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1125 } else if (pkt->ol_flags & PKT_RX_IPV6_HDR) {
1126 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1127 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1128 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1129 ipv6_hdr->dst_addr, &next_hop) != 0)
1139 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1140 uint16_t *dst_port, uint8_t portid)
1142 struct ether_hdr *eth_hdr;
1143 struct ipv4_hdr *ipv4_hdr;
1148 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1149 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1151 dst_ipv4 = ipv4_hdr->dst_addr;
1152 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1153 dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1155 te = _mm_load_si128((__m128i *)eth_hdr);
1159 rfc1812_process(ipv4_hdr, dst_port, pkt->ol_flags);
1161 te = _mm_blend_epi16(te, ve, MASK_ETH);
1162 _mm_store_si128((__m128i *)eth_hdr, te);
1166 * Read ol_flags and destination IPV4 addresses from 4 mbufs.
1169 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], __m128i *dip, uint32_t *flag)
1171 struct ipv4_hdr *ipv4_hdr;
1172 struct ether_hdr *eth_hdr;
1173 uint32_t x0, x1, x2, x3;
1175 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1176 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1177 x0 = ipv4_hdr->dst_addr;
1178 flag[0] = pkt[0]->ol_flags & PKT_RX_IPV4_HDR;
1180 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1181 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1182 x1 = ipv4_hdr->dst_addr;
1183 flag[0] &= pkt[1]->ol_flags;
1185 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1186 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1187 x2 = ipv4_hdr->dst_addr;
1188 flag[0] &= pkt[2]->ol_flags;
1190 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1191 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1192 x3 = ipv4_hdr->dst_addr;
1193 flag[0] &= pkt[3]->ol_flags;
1195 dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1199 * Lookup into LPM for destination port.
1200 * If lookup fails, use incoming port (portid) as destination port.
1203 processx4_step2(const struct lcore_conf *qconf, __m128i dip, uint32_t flag,
1204 uint8_t portid, struct rte_mbuf *pkt[FWDSTEP], uint16_t dprt[FWDSTEP])
1207 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1208 4, 5, 6, 7, 0, 1, 2, 3);
1210 /* Byte swap 4 IPV4 addresses. */
1211 dip = _mm_shuffle_epi8(dip, bswap_mask);
1213 /* if all 4 packets are IPV4. */
1214 if (likely(flag != 0)) {
1215 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1218 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1219 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1220 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1221 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1226 * Update source and destination MAC addresses in the ethernet header.
1227 * Perform RFC1812 checks and updates for IPV4 packets.
1230 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1232 __m128i te[FWDSTEP];
1233 __m128i ve[FWDSTEP];
1234 __m128i *p[FWDSTEP];
1236 p[0] = (rte_pktmbuf_mtod(pkt[0], __m128i *));
1237 p[1] = (rte_pktmbuf_mtod(pkt[1], __m128i *));
1238 p[2] = (rte_pktmbuf_mtod(pkt[2], __m128i *));
1239 p[3] = (rte_pktmbuf_mtod(pkt[3], __m128i *));
1241 ve[0] = val_eth[dst_port[0]];
1242 te[0] = _mm_load_si128(p[0]);
1244 ve[1] = val_eth[dst_port[1]];
1245 te[1] = _mm_load_si128(p[1]);
1247 ve[2] = val_eth[dst_port[2]];
1248 te[2] = _mm_load_si128(p[2]);
1250 ve[3] = val_eth[dst_port[3]];
1251 te[3] = _mm_load_si128(p[3]);
1253 /* Update first 12 bytes, keep rest bytes intact. */
1254 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1255 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1256 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1257 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1259 _mm_store_si128(p[0], te[0]);
1260 _mm_store_si128(p[1], te[1]);
1261 _mm_store_si128(p[2], te[2]);
1262 _mm_store_si128(p[3], te[3]);
1264 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1265 &dst_port[0], pkt[0]->ol_flags);
1266 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1267 &dst_port[1], pkt[1]->ol_flags);
1268 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1269 &dst_port[2], pkt[2]->ol_flags);
1270 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1271 &dst_port[3], pkt[3]->ol_flags);
1275 * We group consecutive packets with the same destionation port into one burst.
1276 * To avoid extra latency this is done together with some other packet
1277 * processing, but after we made a final decision about packet's destination.
1278 * To do this we maintain:
1279 * pnum - array of number of consecutive packets with the same dest port for
1280 * each packet in the input burst.
1281 * lp - pointer to the last updated element in the pnum.
1282 * dlp - dest port value lp corresponds to.
1285 #define GRPSZ (1 << FWDSTEP)
1286 #define GRPMSK (GRPSZ - 1)
1288 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \
1289 if (likely((dlp) == (dcp)[(idx)])) { \
1292 (dlp) = (dcp)[idx]; \
1293 (lp) = (pn) + (idx); \
1299 * Group consecutive packets with the same destination port in bursts of 4.
1300 * Suppose we have array of destionation ports:
1301 * dst_port[] = {a, b, c, d,, e, ... }
1302 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1303 * We doing 4 comparisions at once and the result is 4 bit mask.
1304 * This mask is used as an index into prebuild array of pnum values.
1306 static inline uint16_t *
1307 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1309 static const struct {
1310 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1311 int32_t idx; /* index for new last updated elemnet. */
1312 uint16_t lpv; /* add value to the last updated element. */
1315 /* 0: a != b, b != c, c != d, d != e */
1316 .pnum = UINT64_C(0x0001000100010001),
1321 /* 1: a == b, b != c, c != d, d != e */
1322 .pnum = UINT64_C(0x0001000100010002),
1327 /* 2: a != b, b == c, c != d, d != e */
1328 .pnum = UINT64_C(0x0001000100020001),
1333 /* 3: a == b, b == c, c != d, d != e */
1334 .pnum = UINT64_C(0x0001000100020003),
1339 /* 4: a != b, b != c, c == d, d != e */
1340 .pnum = UINT64_C(0x0001000200010001),
1345 /* 5: a == b, b != c, c == d, d != e */
1346 .pnum = UINT64_C(0x0001000200010002),
1351 /* 6: a != b, b == c, c == d, d != e */
1352 .pnum = UINT64_C(0x0001000200030001),
1357 /* 7: a == b, b == c, c == d, d != e */
1358 .pnum = UINT64_C(0x0001000200030004),
1363 /* 8: a != b, b != c, c != d, d == e */
1364 .pnum = UINT64_C(0x0002000100010001),
1369 /* 9: a == b, b != c, c != d, d == e */
1370 .pnum = UINT64_C(0x0002000100010002),
1375 /* 0xa: a != b, b == c, c != d, d == e */
1376 .pnum = UINT64_C(0x0002000100020001),
1381 /* 0xb: a == b, b == c, c != d, d == e */
1382 .pnum = UINT64_C(0x0002000100020003),
1387 /* 0xc: a != b, b != c, c == d, d == e */
1388 .pnum = UINT64_C(0x0002000300010001),
1393 /* 0xd: a == b, b != c, c == d, d == e */
1394 .pnum = UINT64_C(0x0002000300010002),
1399 /* 0xe: a != b, b == c, c == d, d == e */
1400 .pnum = UINT64_C(0x0002000300040001),
1405 /* 0xf: a == b, b == c, c == d, d == e */
1406 .pnum = UINT64_C(0x0002000300040005),
1413 uint16_t u16[FWDSTEP + 1];
1415 } *pnum = (void *)pn;
1419 dp1 = _mm_cmpeq_epi16(dp1, dp2);
1420 dp1 = _mm_unpacklo_epi16(dp1, dp1);
1421 v = _mm_movemask_ps((__m128)dp1);
1423 /* update last port counter. */
1424 lp[0] += gptbl[v].lpv;
1426 /* if dest port value has changed. */
1428 lp = pnum->u16 + gptbl[v].idx;
1430 pnum->u64 = gptbl[v].pnum;
1436 #endif /* APP_LOOKUP_METHOD */
1438 /* main processing loop */
1440 main_loop(__attribute__((unused)) void *dummy)
1442 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1444 uint64_t prev_tsc, diff_tsc, cur_tsc;
1446 uint8_t portid, queueid;
1447 struct lcore_conf *qconf;
1448 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1449 US_PER_S * BURST_TX_DRAIN_US;
1451 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1452 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1456 uint16_t dst_port[MAX_PKT_BURST];
1457 __m128i dip[MAX_PKT_BURST / FWDSTEP];
1458 uint32_t flag[MAX_PKT_BURST / FWDSTEP];
1459 uint16_t pnum[MAX_PKT_BURST + 1];
1464 lcore_id = rte_lcore_id();
1465 qconf = &lcore_conf[lcore_id];
1467 if (qconf->n_rx_queue == 0) {
1468 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1472 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1474 for (i = 0; i < qconf->n_rx_queue; i++) {
1476 portid = qconf->rx_queue_list[i].port_id;
1477 queueid = qconf->rx_queue_list[i].queue_id;
1478 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1484 cur_tsc = rte_rdtsc();
1487 * TX burst queue drain
1489 diff_tsc = cur_tsc - prev_tsc;
1490 if (unlikely(diff_tsc > drain_tsc)) {
1493 * This could be optimized (use queueid instead of
1494 * portid), but it is not called so often
1496 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1497 if (qconf->tx_mbufs[portid].len == 0)
1500 qconf->tx_mbufs[portid].len,
1502 qconf->tx_mbufs[portid].len = 0;
1509 * Read packet from RX queues
1511 for (i = 0; i < qconf->n_rx_queue; ++i) {
1512 portid = qconf->rx_queue_list[i].port_id;
1513 queueid = qconf->rx_queue_list[i].queue_id;
1514 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1519 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1520 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1523 * Send nb_rx - nb_rx%4 packets
1526 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 4);
1527 for (j = 0; j < n ; j+=4) {
1528 uint32_t ol_flag = pkts_burst[j]->ol_flags
1529 & pkts_burst[j+1]->ol_flags
1530 & pkts_burst[j+2]->ol_flags
1531 & pkts_burst[j+3]->ol_flags;
1532 if (ol_flag & PKT_RX_IPV4_HDR ) {
1533 simple_ipv4_fwd_4pkts(&pkts_burst[j],
1535 } else if (ol_flag & PKT_RX_IPV6_HDR) {
1536 simple_ipv6_fwd_4pkts(&pkts_burst[j],
1539 l3fwd_simple_forward(pkts_burst[j],
1541 l3fwd_simple_forward(pkts_burst[j+1],
1543 l3fwd_simple_forward(pkts_burst[j+2],
1545 l3fwd_simple_forward(pkts_burst[j+3],
1549 for (; j < nb_rx ; j++) {
1550 l3fwd_simple_forward(pkts_burst[j],
1554 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1556 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1557 for (j = 0; j != k; j += FWDSTEP) {
1558 processx4_step1(&pkts_burst[j],
1560 &flag[j / FWDSTEP]);
1563 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1564 for (j = 0; j != k; j += FWDSTEP) {
1565 processx4_step2(qconf, dip[j / FWDSTEP],
1566 flag[j / FWDSTEP], portid,
1567 &pkts_burst[j], &dst_port[j]);
1571 * Finish packet processing and group consecutive
1572 * packets with the same destination port.
1574 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1581 processx4_step3(pkts_burst, dst_port);
1583 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1584 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1586 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1587 processx4_step3(&pkts_burst[j],
1592 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1594 dp2 = _mm_loadu_si128((__m128i *)
1595 &dst_port[j - FWDSTEP + 1]);
1596 lp = port_groupx4(&pnum[j - FWDSTEP],
1601 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1603 dp1 = _mm_srli_si128(dp2,
1605 sizeof(dst_port[0]));
1609 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1611 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1612 lp = port_groupx4(&pnum[j - FWDSTEP], lp,
1616 * remove values added by the last repeated
1620 dlp = dst_port[j - 1];
1622 /* set dlp and lp to the never used values. */
1624 lp = pnum + MAX_PKT_BURST;
1627 /* Process up to last 3 packets one by one. */
1628 switch (nb_rx % FWDSTEP) {
1630 process_packet(qconf, pkts_burst[j],
1631 dst_port + j, portid);
1632 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1635 process_packet(qconf, pkts_burst[j],
1636 dst_port + j, portid);
1637 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1640 process_packet(qconf, pkts_burst[j],
1641 dst_port + j, portid);
1642 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1647 * Send packets out, through destination port.
1648 * Consecuteve pacekts with the same destination port
1649 * are already grouped together.
1650 * If destination port for the packet equals BAD_PORT,
1651 * then free the packet without sending it out.
1653 for (j = 0; j < nb_rx; j += k) {
1661 if (likely(pn != BAD_PORT)) {
1662 send_packetsx4(qconf, pn,
1665 for (m = j; m != j + k; m++)
1666 rte_pktmbuf_free(pkts_burst[m]);
1670 #endif /* APP_LOOKUP_METHOD */
1671 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1673 /* Prefetch first packets */
1674 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1675 rte_prefetch0(rte_pktmbuf_mtod(
1676 pkts_burst[j], void *));
1679 /* Prefetch and forward already prefetched packets */
1680 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1681 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1682 j + PREFETCH_OFFSET], void *));
1683 l3fwd_simple_forward(pkts_burst[j], portid,
1687 /* Forward remaining prefetched packets */
1688 for (; j < nb_rx; j++) {
1689 l3fwd_simple_forward(pkts_burst[j], portid,
1692 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1699 check_lcore_params(void)
1701 uint8_t queue, lcore;
1705 for (i = 0; i < nb_lcore_params; ++i) {
1706 queue = lcore_params[i].queue_id;
1707 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1708 printf("invalid queue number: %hhu\n", queue);
1711 lcore = lcore_params[i].lcore_id;
1712 if (!rte_lcore_is_enabled(lcore)) {
1713 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1716 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1718 printf("warning: lcore %hhu is on socket %d with numa off \n",
1726 check_port_config(const unsigned nb_ports)
1731 for (i = 0; i < nb_lcore_params; ++i) {
1732 portid = lcore_params[i].port_id;
1733 if ((enabled_port_mask & (1 << portid)) == 0) {
1734 printf("port %u is not enabled in port mask\n", portid);
1737 if (portid >= nb_ports) {
1738 printf("port %u is not present on the board\n", portid);
1746 get_port_n_rx_queues(const uint8_t port)
1751 for (i = 0; i < nb_lcore_params; ++i) {
1752 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1753 queue = lcore_params[i].queue_id;
1755 return (uint8_t)(++queue);
1759 init_lcore_rx_queues(void)
1761 uint16_t i, nb_rx_queue;
1764 for (i = 0; i < nb_lcore_params; ++i) {
1765 lcore = lcore_params[i].lcore_id;
1766 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1767 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1768 printf("error: too many queues (%u) for lcore: %u\n",
1769 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1772 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1773 lcore_params[i].port_id;
1774 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1775 lcore_params[i].queue_id;
1776 lcore_conf[lcore].n_rx_queue++;
1784 print_usage(const char *prgname)
1786 printf ("%s [EAL options] -- -p PORTMASK -P"
1787 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1788 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1789 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1790 " -P : enable promiscuous mode\n"
1791 " --config (port,queue,lcore): rx queues configuration\n"
1792 " --no-numa: optional, disable numa awareness\n"
1793 " --ipv6: optional, specify it if running ipv6 packets\n"
1794 " --enable-jumbo: enable jumbo frame"
1795 " which max packet len is PKTLEN in decimal (64-9600)\n"
1796 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1800 static int parse_max_pkt_len(const char *pktlen)
1805 /* parse decimal string */
1806 len = strtoul(pktlen, &end, 10);
1807 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1817 parse_portmask(const char *portmask)
1822 /* parse hexadecimal string */
1823 pm = strtoul(portmask, &end, 16);
1824 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1833 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1835 parse_hash_entry_number(const char *hash_entry_num)
1838 unsigned long hash_en;
1839 /* parse hexadecimal string */
1840 hash_en = strtoul(hash_entry_num, &end, 16);
1841 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
1852 parse_config(const char *q_arg)
1855 const char *p, *p0 = q_arg;
1863 unsigned long int_fld[_NUM_FLD];
1864 char *str_fld[_NUM_FLD];
1868 nb_lcore_params = 0;
1870 while ((p = strchr(p0,'(')) != NULL) {
1872 if((p0 = strchr(p,')')) == NULL)
1876 if(size >= sizeof(s))
1879 snprintf(s, sizeof(s), "%.*s", size, p);
1880 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1882 for (i = 0; i < _NUM_FLD; i++){
1884 int_fld[i] = strtoul(str_fld[i], &end, 0);
1885 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1888 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1889 printf("exceeded max number of lcore params: %hu\n",
1893 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1894 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1895 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1898 lcore_params = lcore_params_array;
1902 #define CMD_LINE_OPT_CONFIG "config"
1903 #define CMD_LINE_OPT_NO_NUMA "no-numa"
1904 #define CMD_LINE_OPT_IPV6 "ipv6"
1905 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
1906 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
1908 /* Parse the argument given in the command line of the application */
1910 parse_args(int argc, char **argv)
1915 char *prgname = argv[0];
1916 static struct option lgopts[] = {
1917 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
1918 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
1919 {CMD_LINE_OPT_IPV6, 0, 0, 0},
1920 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
1921 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
1927 while ((opt = getopt_long(argc, argvopt, "p:P",
1928 lgopts, &option_index)) != EOF) {
1933 enabled_port_mask = parse_portmask(optarg);
1934 if (enabled_port_mask == 0) {
1935 printf("invalid portmask\n");
1936 print_usage(prgname);
1941 printf("Promiscuous mode selected\n");
1947 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
1948 sizeof (CMD_LINE_OPT_CONFIG))) {
1949 ret = parse_config(optarg);
1951 printf("invalid config\n");
1952 print_usage(prgname);
1957 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
1958 sizeof(CMD_LINE_OPT_NO_NUMA))) {
1959 printf("numa is disabled \n");
1963 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1964 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
1965 sizeof(CMD_LINE_OPT_IPV6))) {
1966 printf("ipv6 is specified \n");
1971 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
1972 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
1973 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1975 printf("jumbo frame is enabled - disabling simple TX path\n");
1976 port_conf.rxmode.jumbo_frame = 1;
1977 tx_conf.txq_flags = 0;
1979 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1980 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1981 ret = parse_max_pkt_len(optarg);
1982 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1983 printf("invalid packet length\n");
1984 print_usage(prgname);
1987 port_conf.rxmode.max_rx_pkt_len = ret;
1989 printf("set jumbo frame max packet length to %u\n",
1990 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1992 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1993 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
1994 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
1995 ret = parse_hash_entry_number(optarg);
1996 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
1997 hash_entry_number = ret;
1999 printf("invalid hash entry number\n");
2000 print_usage(prgname);
2008 print_usage(prgname);
2014 argv[optind-1] = prgname;
2017 optind = 0; /* reset getopt lib */
2022 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
2024 printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
2025 eth_addr->addr_bytes[0],
2026 eth_addr->addr_bytes[1],
2027 eth_addr->addr_bytes[2],
2028 eth_addr->addr_bytes[3],
2029 eth_addr->addr_bytes[4],
2030 eth_addr->addr_bytes[5]);
2033 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2035 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
2036 union ipv4_5tuple_host* key2)
2038 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
2039 key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
2040 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2041 key2->port_src = rte_cpu_to_be_16(key1->port_src);
2042 key2->proto = key1->proto;
2048 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
2049 union ipv6_5tuple_host* key2)
2052 for (i = 0; i < 16; i++)
2054 key2->ip_dst[i] = key1->ip_dst[i];
2055 key2->ip_src[i] = key1->ip_src[i];
2057 key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2058 key2->port_src = rte_cpu_to_be_16(key1->port_src);
2059 key2->proto = key1->proto;
2066 #define BYTE_VALUE_MAX 256
2067 #define ALL_32_BITS 0xffffffff
2068 #define BIT_8_TO_15 0x0000ff00
2070 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
2074 uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
2076 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2077 for (i = 0; i < array_len; i++) {
2078 struct ipv4_l3fwd_route entry;
2079 union ipv4_5tuple_host newkey;
2080 entry = ipv4_l3fwd_route_array[i];
2081 convert_ipv4_5tuple(&entry.key, &newkey);
2082 ret = rte_hash_add_key (h,(void *) &newkey);
2084 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2085 " to the l3fwd hash.\n", i);
2087 ipv4_l3fwd_out_if[ret] = entry.if_out;
2089 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
2092 #define BIT_16_TO_23 0x00ff0000
2094 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
2098 uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
2100 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2101 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2102 for (i = 0; i < array_len; i++) {
2103 struct ipv6_l3fwd_route entry;
2104 union ipv6_5tuple_host newkey;
2105 entry = ipv6_l3fwd_route_array[i];
2106 convert_ipv6_5tuple(&entry.key, &newkey);
2107 ret = rte_hash_add_key (h, (void *) &newkey);
2109 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2110 " to the l3fwd hash.\n", i);
2112 ipv6_l3fwd_out_if[ret] = entry.if_out;
2114 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
2117 #define NUMBER_PORT_USED 4
2119 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
2120 unsigned int nr_flow)
2123 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2124 for (i = 0; i < nr_flow; i++) {
2125 struct ipv4_l3fwd_route entry;
2126 union ipv4_5tuple_host newkey;
2127 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2128 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2129 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2130 /* Create the ipv4 exact match flow */
2131 memset(&entry, 0, sizeof(entry));
2132 switch (i & (NUMBER_PORT_USED -1)) {
2134 entry = ipv4_l3fwd_route_array[0];
2135 entry.key.ip_dst = IPv4(101,c,b,a);
2138 entry = ipv4_l3fwd_route_array[1];
2139 entry.key.ip_dst = IPv4(201,c,b,a);
2142 entry = ipv4_l3fwd_route_array[2];
2143 entry.key.ip_dst = IPv4(111,c,b,a);
2146 entry = ipv4_l3fwd_route_array[3];
2147 entry.key.ip_dst = IPv4(211,c,b,a);
2150 convert_ipv4_5tuple(&entry.key, &newkey);
2151 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2153 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2155 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2158 printf("Hash: Adding 0x%x keys\n", nr_flow);
2162 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
2163 unsigned int nr_flow)
2166 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2167 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2168 for (i = 0; i < nr_flow; i++) {
2169 struct ipv6_l3fwd_route entry;
2170 union ipv6_5tuple_host newkey;
2171 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2172 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2173 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2174 /* Create the ipv6 exact match flow */
2175 memset(&entry, 0, sizeof(entry));
2176 switch (i & (NUMBER_PORT_USED - 1)) {
2177 case 0: entry = ipv6_l3fwd_route_array[0]; break;
2178 case 1: entry = ipv6_l3fwd_route_array[1]; break;
2179 case 2: entry = ipv6_l3fwd_route_array[2]; break;
2180 case 3: entry = ipv6_l3fwd_route_array[3]; break;
2182 entry.key.ip_dst[13] = c;
2183 entry.key.ip_dst[14] = b;
2184 entry.key.ip_dst[15] = a;
2185 convert_ipv6_5tuple(&entry.key, &newkey);
2186 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2188 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2190 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2193 printf("Hash: Adding 0x%x keys\n", nr_flow);
2197 setup_hash(int socketid)
2199 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2201 .entries = L3FWD_HASH_ENTRIES,
2202 .bucket_entries = 4,
2203 .key_len = sizeof(union ipv4_5tuple_host),
2204 .hash_func = ipv4_hash_crc,
2205 .hash_func_init_val = 0,
2208 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2210 .entries = L3FWD_HASH_ENTRIES,
2211 .bucket_entries = 4,
2212 .key_len = sizeof(union ipv6_5tuple_host),
2213 .hash_func = ipv6_hash_crc,
2214 .hash_func_init_val = 0,
2219 /* create ipv4 hash */
2220 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2221 ipv4_l3fwd_hash_params.name = s;
2222 ipv4_l3fwd_hash_params.socket_id = socketid;
2223 ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2224 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2225 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2226 "socket %d\n", socketid);
2228 /* create ipv6 hash */
2229 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2230 ipv6_l3fwd_hash_params.name = s;
2231 ipv6_l3fwd_hash_params.socket_id = socketid;
2232 ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2233 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2234 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2235 "socket %d\n", socketid);
2237 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2238 /* For testing hash matching with a large number of flows we
2239 * generate millions of IP 5-tuples with an incremented dst
2240 * address to initialize the hash table. */
2242 /* populate the ipv4 hash */
2243 populate_ipv4_many_flow_into_table(
2244 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2246 /* populate the ipv6 hash */
2247 populate_ipv6_many_flow_into_table(
2248 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2251 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2253 /* populate the ipv4 hash */
2254 populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2256 /* populate the ipv6 hash */
2257 populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2263 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2265 setup_lpm(int socketid)
2267 struct rte_lpm6_config config;
2272 /* create the LPM table */
2273 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2274 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2275 IPV4_L3FWD_LPM_MAX_RULES, 0);
2276 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2277 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2278 " on socket %d\n", socketid);
2280 /* populate the LPM table */
2281 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2283 /* skip unused ports */
2284 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2285 enabled_port_mask) == 0)
2288 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2289 ipv4_l3fwd_route_array[i].ip,
2290 ipv4_l3fwd_route_array[i].depth,
2291 ipv4_l3fwd_route_array[i].if_out);
2294 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2295 "l3fwd LPM table on socket %d\n",
2299 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2300 (unsigned)ipv4_l3fwd_route_array[i].ip,
2301 ipv4_l3fwd_route_array[i].depth,
2302 ipv4_l3fwd_route_array[i].if_out);
2305 /* create the LPM6 table */
2306 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2308 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2309 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2311 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2313 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2314 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2315 " on socket %d\n", socketid);
2317 /* populate the LPM table */
2318 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2320 /* skip unused ports */
2321 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2322 enabled_port_mask) == 0)
2325 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2326 ipv6_l3fwd_route_array[i].ip,
2327 ipv6_l3fwd_route_array[i].depth,
2328 ipv6_l3fwd_route_array[i].if_out);
2331 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2332 "l3fwd LPM table on socket %d\n",
2336 printf("LPM: Adding route %s / %d (%d)\n",
2338 ipv6_l3fwd_route_array[i].depth,
2339 ipv6_l3fwd_route_array[i].if_out);
2345 init_mem(unsigned nb_mbuf)
2347 struct lcore_conf *qconf;
2352 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2353 if (rte_lcore_is_enabled(lcore_id) == 0)
2357 socketid = rte_lcore_to_socket_id(lcore_id);
2361 if (socketid >= NB_SOCKETS) {
2362 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2363 socketid, lcore_id, NB_SOCKETS);
2365 if (pktmbuf_pool[socketid] == NULL) {
2366 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2367 pktmbuf_pool[socketid] =
2368 rte_mempool_create(s, nb_mbuf, MBUF_SIZE, MEMPOOL_CACHE_SIZE,
2369 sizeof(struct rte_pktmbuf_pool_private),
2370 rte_pktmbuf_pool_init, NULL,
2371 rte_pktmbuf_init, NULL,
2373 if (pktmbuf_pool[socketid] == NULL)
2374 rte_exit(EXIT_FAILURE,
2375 "Cannot init mbuf pool on socket %d\n", socketid);
2377 printf("Allocated mbuf pool on socket %d\n", socketid);
2379 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2380 setup_lpm(socketid);
2382 setup_hash(socketid);
2385 qconf = &lcore_conf[lcore_id];
2386 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2387 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2392 /* Check the link status of all ports in up to 9s, and print them finally */
2394 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2396 #define CHECK_INTERVAL 100 /* 100ms */
2397 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2398 uint8_t portid, count, all_ports_up, print_flag = 0;
2399 struct rte_eth_link link;
2401 printf("\nChecking link status");
2403 for (count = 0; count <= MAX_CHECK_TIME; count++) {
2405 for (portid = 0; portid < port_num; portid++) {
2406 if ((port_mask & (1 << portid)) == 0)
2408 memset(&link, 0, sizeof(link));
2409 rte_eth_link_get_nowait(portid, &link);
2410 /* print link status if flag set */
2411 if (print_flag == 1) {
2412 if (link.link_status)
2413 printf("Port %d Link Up - speed %u "
2414 "Mbps - %s\n", (uint8_t)portid,
2415 (unsigned)link.link_speed,
2416 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2417 ("full-duplex") : ("half-duplex\n"));
2419 printf("Port %d Link Down\n",
2423 /* clear all_ports_up flag if any link down */
2424 if (link.link_status == 0) {
2429 /* after finally printing all link status, get out */
2430 if (print_flag == 1)
2433 if (all_ports_up == 0) {
2436 rte_delay_ms(CHECK_INTERVAL);
2439 /* set the print_flag if all ports up or timeout */
2440 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2448 MAIN(int argc, char **argv)
2450 struct lcore_conf *qconf;
2455 uint32_t n_tx_queue, nb_lcores;
2456 uint8_t portid, nb_rx_queue, queue, socketid;
2459 ret = rte_eal_init(argc, argv);
2461 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2465 /* parse application arguments (after the EAL ones) */
2466 ret = parse_args(argc, argv);
2468 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2470 if (check_lcore_params() < 0)
2471 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2473 ret = init_lcore_rx_queues();
2475 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2478 if (rte_eal_pci_probe() < 0)
2479 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
2481 nb_ports = rte_eth_dev_count();
2482 if (nb_ports > RTE_MAX_ETHPORTS)
2483 nb_ports = RTE_MAX_ETHPORTS;
2485 if (check_port_config(nb_ports) < 0)
2486 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2488 nb_lcores = rte_lcore_count();
2490 /* initialize all ports */
2491 for (portid = 0; portid < nb_ports; portid++) {
2492 /* skip ports that are not enabled */
2493 if ((enabled_port_mask & (1 << portid)) == 0) {
2494 printf("\nSkipping disabled port %d\n", portid);
2499 printf("Initializing port %d ... ", portid );
2502 nb_rx_queue = get_port_n_rx_queues(portid);
2503 n_tx_queue = nb_lcores;
2504 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2505 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2506 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2507 nb_rx_queue, (unsigned)n_tx_queue );
2508 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2509 (uint16_t)n_tx_queue, &port_conf);
2511 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2514 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2515 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2519 * prepare dst and src MACs for each port.
2521 *(uint64_t *)(val_eth + portid) =
2522 ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2523 ether_addr_copy(&ports_eth_addr[portid],
2524 (struct ether_addr *)(val_eth + portid) + 1);
2527 ret = init_mem(NB_MBUF);
2529 rte_exit(EXIT_FAILURE, "init_mem failed\n");
2531 /* init one TX queue per couple (lcore,port) */
2533 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2534 if (rte_lcore_is_enabled(lcore_id) == 0)
2538 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2542 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2544 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2545 socketid, &tx_conf);
2547 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2548 "port=%d\n", ret, portid);
2550 qconf = &lcore_conf[lcore_id];
2551 qconf->tx_queue_id[portid] = queueid;
2557 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2558 if (rte_lcore_is_enabled(lcore_id) == 0)
2560 qconf = &lcore_conf[lcore_id];
2561 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2563 /* init RX queues */
2564 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2565 portid = qconf->rx_queue_list[queue].port_id;
2566 queueid = qconf->rx_queue_list[queue].queue_id;
2569 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2573 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2576 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2577 socketid, &rx_conf, pktmbuf_pool[socketid]);
2579 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2580 "port=%d\n", ret, portid);
2587 for (portid = 0; portid < nb_ports; portid++) {
2588 if ((enabled_port_mask & (1 << portid)) == 0) {
2592 ret = rte_eth_dev_start(portid);
2594 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2598 * If enabled, put device in promiscuous mode.
2599 * This allows IO forwarding mode to forward packets
2600 * to itself through 2 cross-connected ports of the
2604 rte_eth_promiscuous_enable(portid);
2607 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2609 /* launch per-lcore init on every lcore */
2610 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2611 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2612 if (rte_eal_wait_lcore(lcore_id) < 0)