4 * Copyright(c) 2010-2013 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.
39 #include <sys/types.h>
41 #include <sys/queue.h>
48 #include <rte_common.h>
49 #include <rte_byteorder.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
54 #include <rte_tailq.h>
56 #include <rte_per_lcore.h>
57 #include <rte_launch.h>
58 #include <rte_atomic.h>
59 #include <rte_cycles.h>
60 #include <rte_prefetch.h>
61 #include <rte_lcore.h>
62 #include <rte_per_lcore.h>
63 #include <rte_branch_prediction.h>
64 #include <rte_interrupts.h>
66 #include <rte_random.h>
67 #include <rte_debug.h>
68 #include <rte_ether.h>
69 #include <rte_ethdev.h>
71 #include <rte_mempool.h>
76 #include <rte_string_fns.h>
77 #include <rte_timer.h>
78 #include <rte_power.h>
82 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
84 #define MAX_PKT_BURST 32
86 #define MIN_ZERO_POLL_COUNT 5
88 /* around 100ms at 2 Ghz */
89 #define TIMER_RESOLUTION_CYCLES 200000000ULL
91 #define TIMER_NUMBER_PER_SECOND 10
93 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
94 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
96 #define APP_LOOKUP_EXACT_MATCH 0
97 #define APP_LOOKUP_LPM 1
98 #define DO_RFC_1812_CHECKS
100 #ifndef APP_LOOKUP_METHOD
101 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
104 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
105 #include <rte_hash.h>
106 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
109 #error "APP_LOOKUP_METHOD set to incorrect value"
113 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
114 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
115 #define IPv6_BYTES(addr) \
116 addr[0], addr[1], addr[2], addr[3], \
117 addr[4], addr[5], addr[6], addr[7], \
118 addr[8], addr[9], addr[10], addr[11],\
119 addr[12], addr[13],addr[14], addr[15]
122 #define MAX_JUMBO_PKT_LEN 9600
124 #define IPV6_ADDR_LEN 16
126 #define MEMPOOL_CACHE_SIZE 256
128 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
131 * This expression is used to calculate the number of mbufs needed depending on
132 * user input, taking into account memory for rx and tx hardware rings, cache
133 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
134 * NB_MBUF never goes below a minimum value of 8192.
137 #define NB_MBUF RTE_MAX ( \
138 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
139 nb_ports*nb_lcores*MAX_PKT_BURST + \
140 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
141 nb_lcores*MEMPOOL_CACHE_SIZE), \
145 * RX and TX Prefetch, Host, and Write-back threshold values should be
146 * carefully set for optimal performance. Consult the network
147 * controller's datasheet and supporting DPDK documentation for guidance
148 * on how these parameters should be set.
150 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
151 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
152 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
155 * These default values are optimized for use with the Intel(R) 82599 10 GbE
156 * Controller and the DPDK ixgbe PMD. Consider using other values for other
157 * network controllers and/or network drivers.
159 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
160 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
161 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
163 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
167 /* Configure how many packets ahead to prefetch, when reading packets */
168 #define PREFETCH_OFFSET 3
171 * Configurable number of RX/TX ring descriptors
173 #define RTE_TEST_RX_DESC_DEFAULT 128
174 #define RTE_TEST_TX_DESC_DEFAULT 512
175 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
176 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
178 /* ethernet addresses of ports */
179 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
181 /* mask of enabled ports */
182 static uint32_t enabled_port_mask = 0;
183 /* Ports set in promiscuous mode off by default. */
184 static int promiscuous_on = 0;
185 /* NUMA is enabled by default. */
186 static int numa_on = 1;
188 enum freq_scale_hint_t
198 struct rte_mbuf *m_table[MAX_PKT_BURST];
201 struct lcore_rx_queue {
204 enum freq_scale_hint_t freq_up_hint;
205 uint32_t zero_rx_packet_count;
207 } __rte_cache_aligned;
209 #define MAX_RX_QUEUE_PER_LCORE 16
210 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
211 #define MAX_RX_QUEUE_PER_PORT 128
213 #define MAX_LCORE_PARAMS 1024
214 struct lcore_params {
218 } __rte_cache_aligned;
220 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
221 static struct lcore_params lcore_params_array_default[] = {
233 static struct lcore_params * lcore_params = lcore_params_array_default;
234 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
235 sizeof(lcore_params_array_default[0]);
237 static struct rte_eth_conf port_conf = {
239 .max_rx_pkt_len = ETHER_MAX_LEN,
241 .header_split = 0, /**< Header Split disabled */
242 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
243 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
244 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
245 .hw_strip_crc = 0, /**< CRC stripped by hardware */
250 .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
254 .mq_mode = ETH_DCB_NONE,
258 static const struct rte_eth_rxconf rx_conf = {
260 .pthresh = RX_PTHRESH,
261 .hthresh = RX_HTHRESH,
262 .wthresh = RX_WTHRESH,
264 .rx_free_thresh = 32,
267 static const struct rte_eth_txconf tx_conf = {
269 .pthresh = TX_PTHRESH,
270 .hthresh = TX_HTHRESH,
271 .wthresh = TX_WTHRESH,
273 .tx_free_thresh = 0, /* Use PMD default values */
274 .tx_rs_thresh = 0, /* Use PMD default values */
278 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
281 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
283 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
284 #include <rte_hash_crc.h>
285 #define DEFAULT_HASH_FUNC rte_hash_crc
287 #include <rte_jhash.h>
288 #define DEFAULT_HASH_FUNC rte_jhash
297 } __attribute__((__packed__));
300 uint8_t ip_dst[IPV6_ADDR_LEN];
301 uint8_t ip_src[IPV6_ADDR_LEN];
305 } __attribute__((__packed__));
307 struct ipv4_l3fwd_route {
308 struct ipv4_5tuple key;
312 struct ipv6_l3fwd_route {
313 struct ipv6_5tuple key;
317 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
318 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
319 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
320 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
321 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
324 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
327 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
328 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
329 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
330 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
336 typedef struct rte_hash lookup_struct_t;
337 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
338 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
340 #define L3FWD_HASH_ENTRIES 1024
342 #define IPV4_L3FWD_NUM_ROUTES \
343 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
345 #define IPV6_L3FWD_NUM_ROUTES \
346 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
348 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
349 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
352 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
353 struct ipv4_l3fwd_route {
359 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
360 {IPv4(1,1,1,0), 24, 0},
361 {IPv4(2,1,1,0), 24, 1},
362 {IPv4(3,1,1,0), 24, 2},
363 {IPv4(4,1,1,0), 24, 3},
364 {IPv4(5,1,1,0), 24, 4},
365 {IPv4(6,1,1,0), 24, 5},
366 {IPv4(7,1,1,0), 24, 6},
367 {IPv4(8,1,1,0), 24, 7},
370 #define IPV4_L3FWD_NUM_ROUTES \
371 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
373 #define IPV4_L3FWD_LPM_MAX_RULES 1024
375 typedef struct rte_lpm lookup_struct_t;
376 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
381 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
382 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
383 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
384 lookup_struct_t * ipv4_lookup_struct;
385 lookup_struct_t * ipv6_lookup_struct;
386 } __rte_cache_aligned;
389 /* total sleep time in ms since last frequency scaling down */
391 /* number of long sleep recently */
392 uint32_t nb_long_sleep;
393 /* freq. scaling up trend */
395 /* total packet processed recently */
396 uint64_t nb_rx_processed;
397 /* total iterations looped recently */
398 uint64_t nb_iteration_looped;
400 } __rte_cache_aligned;
402 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
403 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
404 static struct rte_timer power_timers[RTE_MAX_LCORE];
406 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
407 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
408 unsigned lcore_id, uint32_t rx_ring_length);
410 /* exit signal handler */
412 signal_exit_now(int sigtype)
417 if (sigtype == SIGINT) {
418 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
419 if (rte_lcore_is_enabled(lcore_id) == 0)
422 /* init power management library */
423 ret = rte_power_exit(lcore_id);
425 rte_exit(EXIT_FAILURE, "Power management "
426 "library de-initialization failed on "
427 "core%u\n", lcore_id);
431 rte_exit(EXIT_SUCCESS, "User forced exit\n");
434 /* Freqency scale down timer callback */
436 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
437 __attribute__((unused)) void *arg)
440 float sleep_time_ratio;
441 unsigned lcore_id = rte_lcore_id();
443 /* accumulate total execution time in us when callback is invoked */
444 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
445 (float)SCALING_PERIOD;
448 * check whether need to scale down frequency a step if it sleep a lot.
450 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
451 rte_power_freq_down(lcore_id);
452 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
453 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
455 * scale down a step if average packet per iteration less
458 rte_power_freq_down(lcore_id);
461 * initialize another timer according to current frequency to ensure
462 * timer interval is relatively fixed.
464 hz = rte_get_timer_hz();
465 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
466 SINGLE, lcore_id, power_timer_cb, NULL);
468 stats[lcore_id].nb_rx_processed = 0;
469 stats[lcore_id].nb_iteration_looped = 0;
471 stats[lcore_id].sleep_time = 0;
474 /* Send burst of packets on an output interface */
476 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
478 struct rte_mbuf **m_table;
482 queueid = qconf->tx_queue_id[port];
483 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
485 ret = rte_eth_tx_burst(port, queueid, m_table, n);
486 if (unlikely(ret < n)) {
488 rte_pktmbuf_free(m_table[ret]);
495 /* Enqueue a single packet, and send burst if queue is filled */
497 send_single_packet(struct rte_mbuf *m, uint8_t port)
501 struct lcore_conf *qconf;
503 lcore_id = rte_lcore_id();
505 qconf = &lcore_conf[lcore_id];
506 len = qconf->tx_mbufs[port].len;
507 qconf->tx_mbufs[port].m_table[len] = m;
510 /* enough pkts to be sent */
511 if (unlikely(len == MAX_PKT_BURST)) {
512 send_burst(qconf, MAX_PKT_BURST, port);
516 qconf->tx_mbufs[port].len = len;
520 #ifdef DO_RFC_1812_CHECKS
522 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
524 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
526 * 1. The packet length reported by the Link Layer must be large
527 * enough to hold the minimum length legal IP datagram (20 bytes).
529 if (link_len < sizeof(struct ipv4_hdr))
532 /* 2. The IP checksum must be correct. */
533 /* this is checked in H/W */
536 * 3. The IP version number must be 4. If the version number is not 4
537 * then the packet may be another version of IP, such as IPng or
540 if (((pkt->version_ihl) >> 4) != 4)
543 * 4. The IP header length field must be large enough to hold the
544 * minimum length legal IP datagram (20 bytes = 5 words).
546 if ((pkt->version_ihl & 0xf) < 5)
550 * 5. The IP total length field must be large enough to hold the IP
551 * datagram header, whose length is specified in the IP header length
554 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
561 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
563 print_ipv4_key(struct ipv4_5tuple key)
565 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
566 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
567 key.port_dst, key.port_src, key.proto);
570 print_ipv6_key(struct ipv6_5tuple key)
572 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
573 "port dst = %d, port src = %d, proto = %d\n",
574 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
575 key.port_dst, key.port_src, key.proto);
578 static inline uint8_t
579 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
580 lookup_struct_t * ipv4_l3fwd_lookup_struct)
582 struct ipv4_5tuple key;
587 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
588 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
589 key.proto = ipv4_hdr->next_proto_id;
591 switch (ipv4_hdr->next_proto_id) {
593 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
594 sizeof(struct ipv4_hdr));
595 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
596 key.port_src = rte_be_to_cpu_16(tcp->src_port);
600 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
601 sizeof(struct ipv4_hdr));
602 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
603 key.port_src = rte_be_to_cpu_16(udp->src_port);
612 /* Find destination port */
613 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
614 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
617 static inline uint8_t
618 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
619 lookup_struct_t *ipv6_l3fwd_lookup_struct)
621 struct ipv6_5tuple key;
626 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
627 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
629 key.proto = ipv6_hdr->proto;
631 switch (ipv6_hdr->proto) {
633 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
634 sizeof(struct ipv6_hdr));
635 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
636 key.port_src = rte_be_to_cpu_16(tcp->src_port);
640 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
641 sizeof(struct ipv6_hdr));
642 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
643 key.port_src = rte_be_to_cpu_16(udp->src_port);
652 /* Find destination port */
653 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
654 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
658 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
659 static inline uint8_t
660 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
661 lookup_struct_t *ipv4_l3fwd_lookup_struct)
665 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
666 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
672 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
673 struct lcore_conf *qconf)
675 struct ether_hdr *eth_hdr;
676 struct ipv4_hdr *ipv4_hdr;
680 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
682 if (m->ol_flags & PKT_RX_IPV4_HDR) {
683 /* Handle IPv4 headers.*/
685 (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
686 + sizeof(struct ether_hdr));
688 #ifdef DO_RFC_1812_CHECKS
689 /* Check to make sure the packet is valid (RFC1812) */
690 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
696 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
697 qconf->ipv4_lookup_struct);
698 if (dst_port >= RTE_MAX_ETHPORTS ||
699 (enabled_port_mask & 1 << dst_port) == 0)
702 /* 02:00:00:00:00:xx */
703 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
704 *((uint64_t *)d_addr_bytes) =
705 0x000000000002 + ((uint64_t)dst_port << 40);
707 #ifdef DO_RFC_1812_CHECKS
708 /* Update time to live and header checksum */
709 --(ipv4_hdr->time_to_live);
710 ++(ipv4_hdr->hdr_checksum);
714 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
716 send_single_packet(m, dst_port);
719 /* Handle IPv6 headers.*/
720 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
721 struct ipv6_hdr *ipv6_hdr;
724 (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
725 + sizeof(struct ether_hdr));
727 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
728 qconf->ipv6_lookup_struct);
730 if (dst_port >= RTE_MAX_ETHPORTS ||
731 (enabled_port_mask & 1 << dst_port) == 0)
734 /* 02:00:00:00:00:xx */
735 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
736 *((uint64_t *)d_addr_bytes) =
737 0x000000000002 + ((uint64_t)dst_port << 40);
740 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
742 send_single_packet(m, dst_port);
744 /* We don't currently handle IPv6 packets in LPM mode. */
751 #define SLEEP_GEAR1_THRESHOLD 100
752 #define SLEEP_GEAR2_THRESHOLD 1000
754 static inline uint32_t
755 power_idle_heuristic(uint32_t zero_rx_packet_count)
757 /* If zero count is less than 100, use it as the sleep time in us */
758 if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
759 return zero_rx_packet_count;
760 /* If zero count is less than 1000, sleep time should be 100 us */
761 else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
762 (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
763 return SLEEP_GEAR1_THRESHOLD;
764 /* If zero count is greater than 1000, sleep time should be 1000 us */
765 else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
766 return SLEEP_GEAR2_THRESHOLD;
771 static inline enum freq_scale_hint_t
772 power_freq_scaleup_heuristic(unsigned lcore_id, uint32_t rx_ring_length)
775 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
778 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
779 #define FREQ_GEAR2_RX_PACKET_THRESHOLD 64
780 #define FREQ_GEAR3_RX_PACKET_THRESHOLD 96
781 #define FREQ_UP_TREND1_ACC 1
782 #define FREQ_UP_TREND2_ACC 100
783 #define FREQ_UP_THRESHOLD 10000
786 * there are received packets to process, staying at C0 state while
787 * trying to scale up frequency depending on how many entries on h/w
788 * queue. Determine frequency scaleup trend based on availiable entries
791 if (rx_ring_length > FREQ_GEAR3_RX_PACKET_THRESHOLD) {
792 stats[lcore_id].trend = 0;
794 } else if (rx_ring_length > FREQ_GEAR2_RX_PACKET_THRESHOLD)
795 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
796 else if (rx_ring_length > FREQ_GEAR1_RX_PACKET_THRESHOLD)
797 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
799 if (stats[lcore_id].trend > FREQ_UP_THRESHOLD) {
800 stats[lcore_id].trend = 0;
807 /* main processing loop */
809 main_loop(__attribute__((unused)) void *dummy)
811 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
813 uint64_t prev_tsc, diff_tsc, cur_tsc;
814 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
816 uint8_t portid, queueid;
817 struct lcore_conf *qconf;
818 struct lcore_rx_queue *rx_queue;
819 uint32_t rx_ring_length;
820 enum freq_scale_hint_t lcore_scaleup_hint;
822 uint32_t lcore_rx_idle_count = 0;
823 uint32_t lcore_idle_hint = 0;
825 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
829 lcore_id = rte_lcore_id();
830 qconf = &lcore_conf[lcore_id];
832 if (qconf->n_rx_queue == 0) {
833 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
837 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
839 for (i = 0; i < qconf->n_rx_queue; i++) {
841 portid = qconf->rx_queue_list[i].port_id;
842 queueid = qconf->rx_queue_list[i].queue_id;
843 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
844 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
848 stats[lcore_id].nb_iteration_looped++;
850 cur_tsc = rte_rdtsc();
851 cur_tsc_power = cur_tsc;
854 * TX burst queue drain
856 diff_tsc = cur_tsc - prev_tsc;
857 if (unlikely(diff_tsc > drain_tsc)) {
860 * This could be optimized (use queueid instead of
861 * portid), but it is not called so often
863 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
864 if (qconf->tx_mbufs[portid].len == 0)
866 send_burst(&lcore_conf[lcore_id],
867 qconf->tx_mbufs[portid].len,
869 qconf->tx_mbufs[portid].len = 0;
875 diff_tsc_power = cur_tsc_power - prev_tsc_power;
876 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
878 prev_tsc_power = cur_tsc_power;
882 * Read packet from RX queues
884 lcore_scaleup_hint = FREQ_CURRENT;
885 lcore_rx_idle_count = 0;
886 for (i = 0; i < qconf->n_rx_queue; ++i) {
887 rx_queue = &(qconf->rx_queue_list[i]);
888 rx_queue->idle_hint = 0;
889 portid = rx_queue->port_id;
890 queueid = rx_queue->queue_id;
892 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
894 stats[lcore_id].nb_rx_processed += nb_rx;
895 if (unlikely(nb_rx == 0)) {
897 * no packet received from rx queue, try to
898 * sleep for a while forcing CPU enter deeper
901 rx_queue->zero_rx_packet_count++;
903 if (rx_queue->zero_rx_packet_count <=
907 rx_queue->idle_hint = power_idle_heuristic(\
908 rx_queue->zero_rx_packet_count);
909 lcore_rx_idle_count++;
912 * get availiable descriptor number via MMIO read is costly,
913 * so only do it when recent poll returns maximum number.
915 if (nb_rx >= MAX_PKT_BURST)
916 rx_ring_length = rte_eth_rx_queue_count(portid, queueid);
920 rx_queue->zero_rx_packet_count = 0;
923 * do not scale up frequency immediately as
924 * user to kernel space communication is costly
925 * which might impact packet I/O for received
928 rx_queue->freq_up_hint =
929 power_freq_scaleup_heuristic(lcore_id,
933 /* Prefetch first packets */
934 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
935 rte_prefetch0(rte_pktmbuf_mtod(
936 pkts_burst[j], void *));
939 /* Prefetch and forward already prefetched packets */
940 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
941 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
942 j + PREFETCH_OFFSET], void *));
943 l3fwd_simple_forward(pkts_burst[j], portid,
947 /* Forward remaining prefetched packets */
948 for (; j < nb_rx; j++) {
949 l3fwd_simple_forward(pkts_burst[j], portid,
954 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
955 for (i = 1, lcore_scaleup_hint =
956 qconf->rx_queue_list[0].freq_up_hint;
957 i < qconf->n_rx_queue; ++i) {
958 rx_queue = &(qconf->rx_queue_list[i]);
959 if (rx_queue->freq_up_hint >
962 rx_queue->freq_up_hint;
965 if (lcore_scaleup_hint == FREQ_HIGHEST)
966 rte_power_freq_max(lcore_id);
967 else if (lcore_scaleup_hint == FREQ_HIGHER)
968 rte_power_freq_up(lcore_id);
971 * All Rx queues empty in recent consecutive polls,
972 * sleep in a conservative manner, meaning sleep as
975 for (i = 1, lcore_idle_hint =
976 qconf->rx_queue_list[0].idle_hint;
977 i < qconf->n_rx_queue; ++i) {
978 rx_queue = &(qconf->rx_queue_list[i]);
979 if (rx_queue->idle_hint < lcore_idle_hint)
980 lcore_idle_hint = rx_queue->idle_hint;
983 if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
985 * execute "pause" instruction to avoid context
986 * switch for short sleep.
988 rte_delay_us(lcore_idle_hint);
990 /* long sleep force runing thread to suspend */
991 usleep(lcore_idle_hint);
993 stats[lcore_id].sleep_time += lcore_idle_hint;
999 check_lcore_params(void)
1001 uint8_t queue, lcore;
1005 for (i = 0; i < nb_lcore_params; ++i) {
1006 queue = lcore_params[i].queue_id;
1007 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1008 printf("invalid queue number: %hhu\n", queue);
1011 lcore = lcore_params[i].lcore_id;
1012 if (!rte_lcore_is_enabled(lcore)) {
1013 printf("error: lcore %hhu is not enabled in lcore "
1017 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1019 printf("warning: lcore %hhu is on socket %d with numa "
1020 "off\n", lcore, socketid);
1027 check_port_config(const unsigned nb_ports)
1032 for (i = 0; i < nb_lcore_params; ++i) {
1033 portid = lcore_params[i].port_id;
1034 if ((enabled_port_mask & (1 << portid)) == 0) {
1035 printf("port %u is not enabled in port mask\n",
1039 if (portid >= nb_ports) {
1040 printf("port %u is not present on the board\n",
1049 get_port_n_rx_queues(const uint8_t port)
1054 for (i = 0; i < nb_lcore_params; ++i) {
1055 if (lcore_params[i].port_id == port &&
1056 lcore_params[i].queue_id > queue)
1057 queue = lcore_params[i].queue_id;
1059 return (uint8_t)(++queue);
1063 init_lcore_rx_queues(void)
1065 uint16_t i, nb_rx_queue;
1068 for (i = 0; i < nb_lcore_params; ++i) {
1069 lcore = lcore_params[i].lcore_id;
1070 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1071 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1072 printf("error: too many queues (%u) for lcore: %u\n",
1073 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1076 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1077 lcore_params[i].port_id;
1078 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1079 lcore_params[i].queue_id;
1080 lcore_conf[lcore].n_rx_queue++;
1088 print_usage(const char *prgname)
1090 printf ("%s [EAL options] -- -p PORTMASK -P"
1091 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1092 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1093 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1094 " -P : enable promiscuous mode\n"
1095 " --config (port,queue,lcore): rx queues configuration\n"
1096 " --no-numa: optional, disable numa awareness\n"
1097 " --enable-jumbo: enable jumbo frame"
1098 " which max packet len is PKTLEN in decimal (64-9600)\n",
1102 static int parse_max_pkt_len(const char *pktlen)
1107 /* parse decimal string */
1108 len = strtoul(pktlen, &end, 10);
1109 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1119 parse_portmask(const char *portmask)
1124 /* parse hexadecimal string */
1125 pm = strtoul(portmask, &end, 16);
1126 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1136 parse_config(const char *q_arg)
1139 const char *p, *p0 = q_arg;
1147 unsigned long int_fld[_NUM_FLD];
1148 char *str_fld[_NUM_FLD];
1152 nb_lcore_params = 0;
1154 while ((p = strchr(p0,'(')) != NULL) {
1156 if((p0 = strchr(p,')')) == NULL)
1160 if(size >= sizeof(s))
1163 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1164 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1167 for (i = 0; i < _NUM_FLD; i++){
1169 int_fld[i] = strtoul(str_fld[i], &end, 0);
1170 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1174 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1175 printf("exceeded max number of lcore params: %hu\n",
1179 lcore_params_array[nb_lcore_params].port_id =
1180 (uint8_t)int_fld[FLD_PORT];
1181 lcore_params_array[nb_lcore_params].queue_id =
1182 (uint8_t)int_fld[FLD_QUEUE];
1183 lcore_params_array[nb_lcore_params].lcore_id =
1184 (uint8_t)int_fld[FLD_LCORE];
1187 lcore_params = lcore_params_array;
1192 /* Parse the argument given in the command line of the application */
1194 parse_args(int argc, char **argv)
1199 char *prgname = argv[0];
1200 static struct option lgopts[] = {
1201 {"config", 1, 0, 0},
1202 {"no-numa", 0, 0, 0},
1203 {"enable-jumbo", 0, 0, 0},
1209 while ((opt = getopt_long(argc, argvopt, "p:P",
1210 lgopts, &option_index)) != EOF) {
1215 enabled_port_mask = parse_portmask(optarg);
1216 if (enabled_port_mask == 0) {
1217 printf("invalid portmask\n");
1218 print_usage(prgname);
1223 printf("Promiscuous mode selected\n");
1229 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1230 ret = parse_config(optarg);
1232 printf("invalid config\n");
1233 print_usage(prgname);
1238 if (!strncmp(lgopts[option_index].name,
1240 printf("numa is disabled \n");
1244 if (!strncmp(lgopts[option_index].name,
1245 "enable-jumbo", 12)) {
1246 struct option lenopts =
1247 {"max-pkt-len", required_argument, \
1250 printf("jumbo frame is enabled \n");
1251 port_conf.rxmode.jumbo_frame = 1;
1254 * if no max-pkt-len set, use the default value
1257 if (0 == getopt_long(argc, argvopt, "",
1258 &lenopts, &option_index)) {
1259 ret = parse_max_pkt_len(optarg);
1261 (ret > MAX_JUMBO_PKT_LEN)){
1262 printf("invalid packet "
1264 print_usage(prgname);
1267 port_conf.rxmode.max_rx_pkt_len = ret;
1269 printf("set jumbo frame "
1270 "max packet length to %u\n",
1271 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1277 print_usage(prgname);
1283 argv[optind-1] = prgname;
1286 optind = 0; /* reset getopt lib */
1291 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1293 printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1294 eth_addr->addr_bytes[0],
1295 eth_addr->addr_bytes[1],
1296 eth_addr->addr_bytes[2],
1297 eth_addr->addr_bytes[3],
1298 eth_addr->addr_bytes[4],
1299 eth_addr->addr_bytes[5]);
1302 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1304 setup_hash(int socketid)
1306 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1308 .entries = L3FWD_HASH_ENTRIES,
1309 .bucket_entries = 4,
1310 .key_len = sizeof(struct ipv4_5tuple),
1311 .hash_func = DEFAULT_HASH_FUNC,
1312 .hash_func_init_val = 0,
1315 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1317 .entries = L3FWD_HASH_ENTRIES,
1318 .bucket_entries = 4,
1319 .key_len = sizeof(struct ipv6_5tuple),
1320 .hash_func = DEFAULT_HASH_FUNC,
1321 .hash_func_init_val = 0,
1328 /* create ipv4 hash */
1329 rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1330 ipv4_l3fwd_hash_params.name = s;
1331 ipv4_l3fwd_hash_params.socket_id = socketid;
1332 ipv4_l3fwd_lookup_struct[socketid] =
1333 rte_hash_create(&ipv4_l3fwd_hash_params);
1334 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1335 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1336 "socket %d\n", socketid);
1338 /* create ipv6 hash */
1339 rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1340 ipv6_l3fwd_hash_params.name = s;
1341 ipv6_l3fwd_hash_params.socket_id = socketid;
1342 ipv6_l3fwd_lookup_struct[socketid] =
1343 rte_hash_create(&ipv6_l3fwd_hash_params);
1344 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1345 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1346 "socket %d\n", socketid);
1349 /* populate the ipv4 hash */
1350 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1351 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1352 (void *) &ipv4_l3fwd_route_array[i].key);
1354 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1355 "l3fwd hash on socket %d\n", i, socketid);
1357 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1358 printf("Hash: Adding key\n");
1359 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1362 /* populate the ipv6 hash */
1363 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1364 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1365 (void *) &ipv6_l3fwd_route_array[i].key);
1367 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1368 "l3fwd hash on socket %d\n", i, socketid);
1370 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1371 printf("Hash: Adding key\n");
1372 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1377 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1379 setup_lpm(int socketid)
1385 /* create the LPM table */
1386 rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1387 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1388 IPV4_L3FWD_LPM_MAX_RULES, 0);
1389 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1390 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1391 " on socket %d\n", socketid);
1393 /* populate the LPM table */
1394 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1395 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1396 ipv4_l3fwd_route_array[i].ip,
1397 ipv4_l3fwd_route_array[i].depth,
1398 ipv4_l3fwd_route_array[i].if_out);
1401 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1402 "l3fwd LPM table on socket %d\n",
1406 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1407 (unsigned)ipv4_l3fwd_route_array[i].ip,
1408 ipv4_l3fwd_route_array[i].depth,
1409 ipv4_l3fwd_route_array[i].if_out);
1415 init_mem(unsigned nb_mbuf)
1417 struct lcore_conf *qconf;
1422 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1423 if (rte_lcore_is_enabled(lcore_id) == 0)
1427 socketid = rte_lcore_to_socket_id(lcore_id);
1431 if (socketid >= NB_SOCKETS) {
1432 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1433 "out of range %d\n", socketid,
1434 lcore_id, NB_SOCKETS);
1436 if (pktmbuf_pool[socketid] == NULL) {
1437 rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1438 pktmbuf_pool[socketid] =
1439 rte_mempool_create(s, nb_mbuf,
1440 MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1441 sizeof(struct rte_pktmbuf_pool_private),
1442 rte_pktmbuf_pool_init, NULL,
1443 rte_pktmbuf_init, NULL,
1445 if (pktmbuf_pool[socketid] == NULL)
1446 rte_exit(EXIT_FAILURE,
1447 "Cannot init mbuf pool on socket %d\n",
1450 printf("Allocated mbuf pool on socket %d\n",
1453 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1454 setup_lpm(socketid);
1456 setup_hash(socketid);
1459 qconf = &lcore_conf[lcore_id];
1460 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1461 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1462 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1468 /* Check the link status of all ports in up to 9s, and print them finally */
1470 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1472 #define CHECK_INTERVAL 100 /* 100ms */
1473 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1474 uint8_t portid, count, all_ports_up, print_flag = 0;
1475 struct rte_eth_link link;
1477 printf("\nChecking link status");
1479 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1481 for (portid = 0; portid < port_num; portid++) {
1482 if ((port_mask & (1 << portid)) == 0)
1484 memset(&link, 0, sizeof(link));
1485 rte_eth_link_get_nowait(portid, &link);
1486 /* print link status if flag set */
1487 if (print_flag == 1) {
1488 if (link.link_status)
1489 printf("Port %d Link Up - speed %u "
1490 "Mbps - %s\n", (uint8_t)portid,
1491 (unsigned)link.link_speed,
1492 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1493 ("full-duplex") : ("half-duplex\n"));
1495 printf("Port %d Link Down\n",
1499 /* clear all_ports_up flag if any link down */
1500 if (link.link_status == 0) {
1505 /* after finally printing all link status, get out */
1506 if (print_flag == 1)
1509 if (all_ports_up == 0) {
1512 rte_delay_ms(CHECK_INTERVAL);
1515 /* set the print_flag if all ports up or timeout */
1516 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1524 MAIN(int argc, char **argv)
1526 struct lcore_conf *qconf;
1532 uint32_t n_tx_queue, nb_lcores;
1533 uint8_t portid, nb_rx_queue, queue, socketid;
1535 /* catch SIGINT and restore cpufreq governor to ondemand */
1536 signal(SIGINT, signal_exit_now);
1539 ret = rte_eal_init(argc, argv);
1541 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1545 /* init RTE timer library to be used late */
1546 rte_timer_subsystem_init();
1548 /* parse application arguments (after the EAL ones) */
1549 ret = parse_args(argc, argv);
1551 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1553 if (check_lcore_params() < 0)
1554 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1556 ret = init_lcore_rx_queues();
1558 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1561 /* init driver(s) */
1562 if (rte_pmd_init_all() < 0)
1563 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1565 if (rte_eal_pci_probe() < 0)
1566 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1568 nb_ports = rte_eth_dev_count();
1569 if (nb_ports > RTE_MAX_ETHPORTS)
1570 nb_ports = RTE_MAX_ETHPORTS;
1572 if (check_port_config(nb_ports) < 0)
1573 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1575 nb_lcores = rte_lcore_count();
1577 /* initialize all ports */
1578 for (portid = 0; portid < nb_ports; portid++) {
1579 /* skip ports that are not enabled */
1580 if ((enabled_port_mask & (1 << portid)) == 0) {
1581 printf("\nSkipping disabled port %d\n", portid);
1586 printf("Initializing port %d ... ", portid );
1589 nb_rx_queue = get_port_n_rx_queues(portid);
1590 n_tx_queue = nb_lcores;
1591 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1592 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1593 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1594 nb_rx_queue, (unsigned)n_tx_queue );
1595 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1596 (uint16_t)n_tx_queue, &port_conf);
1598 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1599 "err=%d, port=%d\n", ret, portid);
1601 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1602 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1606 ret = init_mem(NB_MBUF);
1608 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1610 /* init one TX queue per couple (lcore,port) */
1612 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1613 if (rte_lcore_is_enabled(lcore_id) == 0)
1618 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1622 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1624 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1625 socketid, &tx_conf);
1627 rte_exit(EXIT_FAILURE,
1628 "rte_eth_tx_queue_setup: err=%d, "
1629 "port=%d\n", ret, portid);
1631 qconf = &lcore_conf[lcore_id];
1632 qconf->tx_queue_id[portid] = queueid;
1638 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1639 if (rte_lcore_is_enabled(lcore_id) == 0)
1642 /* init power management library */
1643 ret = rte_power_init(lcore_id);
1645 rte_exit(EXIT_FAILURE, "Power management library "
1646 "initialization failed on core%u\n", lcore_id);
1648 /* init timer structures for each enabled lcore */
1649 rte_timer_init(&power_timers[lcore_id]);
1650 hz = rte_get_timer_hz();
1651 rte_timer_reset(&power_timers[lcore_id],
1652 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1653 power_timer_cb, NULL);
1655 qconf = &lcore_conf[lcore_id];
1656 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1658 /* init RX queues */
1659 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1660 portid = qconf->rx_queue_list[queue].port_id;
1661 queueid = qconf->rx_queue_list[queue].queue_id;
1665 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1669 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1672 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1673 socketid, &rx_conf, pktmbuf_pool[socketid]);
1675 rte_exit(EXIT_FAILURE,
1676 "rte_eth_rx_queue_setup: err=%d, "
1677 "port=%d\n", ret, portid);
1684 for (portid = 0; portid < nb_ports; portid++) {
1685 if ((enabled_port_mask & (1 << portid)) == 0) {
1689 ret = rte_eth_dev_start(portid);
1691 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1692 "port=%d\n", ret, portid);
1695 * If enabled, put device in promiscuous mode.
1696 * This allows IO forwarding mode to forward packets
1697 * to itself through 2 cross-connected ports of the
1701 rte_eth_promiscuous_enable(portid);
1704 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1706 /* launch per-lcore init on every lcore */
1707 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1708 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1709 if (rte_eal_wait_lcore(lcore_id) < 0)