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>
47 #include <rte_common.h>
48 #include <rte_byteorder.h>
50 #include <rte_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_tailq.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
70 #include <rte_mempool.h>
75 #include <rte_string_fns.h>
76 #include <rte_timer.h>
77 #include <rte_power.h>
81 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
83 #define MAX_PKT_BURST 32
85 #define MIN_ZERO_POLL_COUNT 5
87 /* around 100ms at 2 Ghz */
88 #define TIMER_RESOLUTION_CYCLES 200000000ULL
90 #define TIMER_NUMBER_PER_SECOND 10
92 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
93 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
95 #define APP_LOOKUP_EXACT_MATCH 0
96 #define APP_LOOKUP_LPM 1
97 #define DO_RFC_1812_CHECKS
99 #ifndef APP_LOOKUP_METHOD
100 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
103 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
104 #include <rte_hash.h>
105 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
108 #error "APP_LOOKUP_METHOD set to incorrect value"
112 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
113 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
114 #define IPv6_BYTES(addr) \
115 addr[0], addr[1], addr[2], addr[3], \
116 addr[4], addr[5], addr[6], addr[7], \
117 addr[8], addr[9], addr[10], addr[11],\
118 addr[12], addr[13],addr[14], addr[15]
121 #define MAX_JUMBO_PKT_LEN 9600
123 #define IPV6_ADDR_LEN 16
125 #define MEMPOOL_CACHE_SIZE 256
127 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
130 * This expression is used to calculate the number of mbufs needed depending on
131 * user input, taking into account memory for rx and tx hardware rings, cache
132 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
133 * NB_MBUF never goes below a minimum value of 8192.
136 #define NB_MBUF RTE_MAX ( \
137 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
138 nb_ports*nb_lcores*MAX_PKT_BURST + \
139 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
140 nb_lcores*MEMPOOL_CACHE_SIZE), \
144 * RX and TX Prefetch, Host, and Write-back threshold values should be
145 * carefully set for optimal performance. Consult the network
146 * controller's datasheet and supporting DPDK documentation for guidance
147 * on how these parameters should be set.
149 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
150 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
151 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
154 * These default values are optimized for use with the Intel(R) 82599 10 GbE
155 * Controller and the DPDK ixgbe PMD. Consider using other values for other
156 * network controllers and/or network drivers.
158 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
159 #define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
160 #define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */
162 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
166 /* Configure how many packets ahead to prefetch, when reading packets */
167 #define PREFETCH_OFFSET 3
170 * Configurable number of RX/TX ring descriptors
172 #define RTE_TEST_RX_DESC_DEFAULT 128
173 #define RTE_TEST_TX_DESC_DEFAULT 512
174 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
175 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
177 /* ethernet addresses of ports */
178 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
180 /* mask of enabled ports */
181 static uint32_t enabled_port_mask = 0;
182 /* Ports set in promiscuous mode off by default. */
183 static int promiscuous_on = 0;
184 /* NUMA is enabled by default. */
185 static int numa_on = 1;
187 enum freq_scale_hint_t
197 struct rte_mbuf *m_table[MAX_PKT_BURST];
200 struct lcore_rx_queue {
203 enum freq_scale_hint_t freq_up_hint;
204 uint32_t zero_rx_packet_count;
206 } __rte_cache_aligned;
208 #define MAX_RX_QUEUE_PER_LCORE 16
209 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
210 #define MAX_RX_QUEUE_PER_PORT 128
212 #define MAX_LCORE_PARAMS 1024
213 struct lcore_params {
217 } __rte_cache_aligned;
219 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
220 static struct lcore_params lcore_params_array_default[] = {
232 static struct lcore_params * lcore_params = lcore_params_array_default;
233 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
234 sizeof(lcore_params_array_default[0]);
236 static struct rte_eth_conf port_conf = {
238 .max_rx_pkt_len = ETHER_MAX_LEN,
240 .header_split = 0, /**< Header Split disabled */
241 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
242 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
243 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
244 .hw_strip_crc = 0, /**< CRC stripped by hardware */
249 .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
253 .mq_mode = ETH_DCB_NONE,
257 static const struct rte_eth_rxconf rx_conf = {
259 .pthresh = RX_PTHRESH,
260 .hthresh = RX_HTHRESH,
261 .wthresh = RX_WTHRESH,
263 .rx_free_thresh = 32,
266 static const struct rte_eth_txconf tx_conf = {
268 .pthresh = TX_PTHRESH,
269 .hthresh = TX_HTHRESH,
270 .wthresh = TX_WTHRESH,
272 .tx_free_thresh = 0, /* Use PMD default values */
273 .tx_rs_thresh = 0, /* Use PMD default values */
277 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
280 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
282 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
283 #include <rte_hash_crc.h>
284 #define DEFAULT_HASH_FUNC rte_hash_crc
286 #include <rte_jhash.h>
287 #define DEFAULT_HASH_FUNC rte_jhash
296 } __attribute__((__packed__));
299 uint8_t ip_dst[IPV6_ADDR_LEN];
300 uint8_t ip_src[IPV6_ADDR_LEN];
304 } __attribute__((__packed__));
306 struct ipv4_l3fwd_route {
307 struct ipv4_5tuple key;
311 struct ipv6_l3fwd_route {
312 struct ipv6_5tuple key;
316 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
317 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
318 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
319 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
320 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
323 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
326 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
327 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
328 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
329 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
335 typedef struct rte_hash lookup_struct_t;
336 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
337 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
339 #define L3FWD_HASH_ENTRIES 1024
341 #define IPV4_L3FWD_NUM_ROUTES \
342 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
344 #define IPV6_L3FWD_NUM_ROUTES \
345 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
347 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
348 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
351 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
352 struct ipv4_l3fwd_route {
358 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
359 {IPv4(1,1,1,0), 24, 0},
360 {IPv4(2,1,1,0), 24, 1},
361 {IPv4(3,1,1,0), 24, 2},
362 {IPv4(4,1,1,0), 24, 3},
363 {IPv4(5,1,1,0), 24, 4},
364 {IPv4(6,1,1,0), 24, 5},
365 {IPv4(7,1,1,0), 24, 6},
366 {IPv4(8,1,1,0), 24, 7},
369 #define IPV4_L3FWD_NUM_ROUTES \
370 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
372 #define IPV4_L3FWD_LPM_MAX_RULES 1024
374 typedef struct rte_lpm lookup_struct_t;
375 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
380 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
381 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
382 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
383 lookup_struct_t * ipv4_lookup_struct;
384 lookup_struct_t * ipv6_lookup_struct;
385 } __rte_cache_aligned;
388 /* total sleep time in ms since last frequency scaling down */
390 /* number of long sleep recently */
391 uint32_t nb_long_sleep;
392 /* freq. scaling up trend */
394 /* total packet processed recently */
395 uint64_t nb_rx_processed;
396 /* total iterations looped recently */
397 uint64_t nb_iteration_looped;
399 } __rte_cache_aligned;
401 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
402 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
403 static struct rte_timer power_timers[RTE_MAX_LCORE];
405 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
406 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
407 unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
409 /* exit signal handler */
411 signal_exit_now(int sigtype)
416 if (sigtype == SIGINT) {
417 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
418 if (rte_lcore_is_enabled(lcore_id) == 0)
421 /* init power management library */
422 ret = rte_power_exit(lcore_id);
424 rte_exit(EXIT_FAILURE, "Power management "
425 "library de-initialization failed on "
426 "core%u\n", lcore_id);
430 rte_exit(EXIT_SUCCESS, "User forced exit\n");
433 /* Freqency scale down timer callback */
435 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
436 __attribute__((unused)) void *arg)
439 float sleep_time_ratio;
440 unsigned lcore_id = rte_lcore_id();
442 /* accumulate total execution time in us when callback is invoked */
443 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
444 (float)SCALING_PERIOD;
447 * check whether need to scale down frequency a step if it sleep a lot.
449 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
450 rte_power_freq_down(lcore_id);
451 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
452 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
454 * scale down a step if average packet per iteration less
457 rte_power_freq_down(lcore_id);
460 * initialize another timer according to current frequency to ensure
461 * timer interval is relatively fixed.
463 hz = rte_get_timer_hz();
464 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
465 SINGLE, lcore_id, power_timer_cb, NULL);
467 stats[lcore_id].nb_rx_processed = 0;
468 stats[lcore_id].nb_iteration_looped = 0;
470 stats[lcore_id].sleep_time = 0;
473 /* Send burst of packets on an output interface */
475 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
477 struct rte_mbuf **m_table;
481 queueid = qconf->tx_queue_id[port];
482 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
484 ret = rte_eth_tx_burst(port, queueid, m_table, n);
485 if (unlikely(ret < n)) {
487 rte_pktmbuf_free(m_table[ret]);
494 /* Enqueue a single packet, and send burst if queue is filled */
496 send_single_packet(struct rte_mbuf *m, uint8_t port)
500 struct lcore_conf *qconf;
502 lcore_id = rte_lcore_id();
504 qconf = &lcore_conf[lcore_id];
505 len = qconf->tx_mbufs[port].len;
506 qconf->tx_mbufs[port].m_table[len] = m;
509 /* enough pkts to be sent */
510 if (unlikely(len == MAX_PKT_BURST)) {
511 send_burst(qconf, MAX_PKT_BURST, port);
515 qconf->tx_mbufs[port].len = len;
519 #ifdef DO_RFC_1812_CHECKS
521 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
523 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
525 * 1. The packet length reported by the Link Layer must be large
526 * enough to hold the minimum length legal IP datagram (20 bytes).
528 if (link_len < sizeof(struct ipv4_hdr))
531 /* 2. The IP checksum must be correct. */
532 /* this is checked in H/W */
535 * 3. The IP version number must be 4. If the version number is not 4
536 * then the packet may be another version of IP, such as IPng or
539 if (((pkt->version_ihl) >> 4) != 4)
542 * 4. The IP header length field must be large enough to hold the
543 * minimum length legal IP datagram (20 bytes = 5 words).
545 if ((pkt->version_ihl & 0xf) < 5)
549 * 5. The IP total length field must be large enough to hold the IP
550 * datagram header, whose length is specified in the IP header length
553 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
560 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
562 print_ipv4_key(struct ipv4_5tuple key)
564 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
565 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
566 key.port_dst, key.port_src, key.proto);
569 print_ipv6_key(struct ipv6_5tuple key)
571 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
572 "port dst = %d, port src = %d, proto = %d\n",
573 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
574 key.port_dst, key.port_src, key.proto);
577 static inline uint8_t
578 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
579 lookup_struct_t * ipv4_l3fwd_lookup_struct)
581 struct ipv4_5tuple key;
586 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
587 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
588 key.proto = ipv4_hdr->next_proto_id;
590 switch (ipv4_hdr->next_proto_id) {
592 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
593 sizeof(struct ipv4_hdr));
594 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
595 key.port_src = rte_be_to_cpu_16(tcp->src_port);
599 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
600 sizeof(struct ipv4_hdr));
601 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
602 key.port_src = rte_be_to_cpu_16(udp->src_port);
611 /* Find destination port */
612 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
613 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
616 static inline uint8_t
617 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
618 lookup_struct_t *ipv6_l3fwd_lookup_struct)
620 struct ipv6_5tuple key;
625 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
626 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
628 key.proto = ipv6_hdr->proto;
630 switch (ipv6_hdr->proto) {
632 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
633 sizeof(struct ipv6_hdr));
634 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
635 key.port_src = rte_be_to_cpu_16(tcp->src_port);
639 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
640 sizeof(struct ipv6_hdr));
641 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
642 key.port_src = rte_be_to_cpu_16(udp->src_port);
651 /* Find destination port */
652 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
653 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
657 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
658 static inline uint8_t
659 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
660 lookup_struct_t *ipv4_l3fwd_lookup_struct)
664 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
665 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
671 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
672 struct lcore_conf *qconf)
674 struct ether_hdr *eth_hdr;
675 struct ipv4_hdr *ipv4_hdr;
679 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
681 if (m->ol_flags & PKT_RX_IPV4_HDR) {
682 /* Handle IPv4 headers.*/
684 (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
685 + sizeof(struct ether_hdr));
687 #ifdef DO_RFC_1812_CHECKS
688 /* Check to make sure the packet is valid (RFC1812) */
689 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
695 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
696 qconf->ipv4_lookup_struct);
697 if (dst_port >= RTE_MAX_ETHPORTS ||
698 (enabled_port_mask & 1 << dst_port) == 0)
701 /* 02:00:00:00:00:xx */
702 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
703 *((uint64_t *)d_addr_bytes) =
704 0x000000000002 + ((uint64_t)dst_port << 40);
706 #ifdef DO_RFC_1812_CHECKS
707 /* Update time to live and header checksum */
708 --(ipv4_hdr->time_to_live);
709 ++(ipv4_hdr->hdr_checksum);
713 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
715 send_single_packet(m, dst_port);
718 /* Handle IPv6 headers.*/
719 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
720 struct ipv6_hdr *ipv6_hdr;
723 (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
724 + sizeof(struct ether_hdr));
726 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
727 qconf->ipv6_lookup_struct);
729 if (dst_port >= RTE_MAX_ETHPORTS ||
730 (enabled_port_mask & 1 << dst_port) == 0)
733 /* 02:00:00:00:00:xx */
734 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
735 *((uint64_t *)d_addr_bytes) =
736 0x000000000002 + ((uint64_t)dst_port << 40);
739 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
741 send_single_packet(m, dst_port);
743 /* We don't currently handle IPv6 packets in LPM mode. */
750 #define SLEEP_GEAR1_THRESHOLD 100
751 #define SLEEP_GEAR2_THRESHOLD 1000
753 static inline uint32_t
754 power_idle_heuristic(uint32_t zero_rx_packet_count)
756 /* If zero count is less than 100, use it as the sleep time in us */
757 if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
758 return zero_rx_packet_count;
759 /* If zero count is less than 1000, sleep time should be 100 us */
760 else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
761 (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
762 return SLEEP_GEAR1_THRESHOLD;
763 /* If zero count is greater than 1000, sleep time should be 1000 us */
764 else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
765 return SLEEP_GEAR2_THRESHOLD;
770 static inline enum freq_scale_hint_t
771 power_freq_scaleup_heuristic(unsigned lcore_id,
776 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
779 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
780 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
781 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
782 #define FREQ_UP_TREND1_ACC 1
783 #define FREQ_UP_TREND2_ACC 100
784 #define FREQ_UP_THRESHOLD 10000
786 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
787 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
788 stats[lcore_id].trend = 0;
790 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
791 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
792 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
793 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
794 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
795 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
797 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
798 stats[lcore_id].trend = 0;
805 /* main processing loop */
807 main_loop(__attribute__((unused)) void *dummy)
809 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
811 uint64_t prev_tsc, diff_tsc, cur_tsc;
812 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
814 uint8_t portid, queueid;
815 struct lcore_conf *qconf;
816 struct lcore_rx_queue *rx_queue;
817 enum freq_scale_hint_t lcore_scaleup_hint;
819 uint32_t lcore_rx_idle_count = 0;
820 uint32_t lcore_idle_hint = 0;
822 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
826 lcore_id = rte_lcore_id();
827 qconf = &lcore_conf[lcore_id];
829 if (qconf->n_rx_queue == 0) {
830 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
834 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
836 for (i = 0; i < qconf->n_rx_queue; i++) {
838 portid = qconf->rx_queue_list[i].port_id;
839 queueid = qconf->rx_queue_list[i].queue_id;
840 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
841 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
845 stats[lcore_id].nb_iteration_looped++;
847 cur_tsc = rte_rdtsc();
848 cur_tsc_power = cur_tsc;
851 * TX burst queue drain
853 diff_tsc = cur_tsc - prev_tsc;
854 if (unlikely(diff_tsc > drain_tsc)) {
857 * This could be optimized (use queueid instead of
858 * portid), but it is not called so often
860 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
861 if (qconf->tx_mbufs[portid].len == 0)
863 send_burst(&lcore_conf[lcore_id],
864 qconf->tx_mbufs[portid].len,
866 qconf->tx_mbufs[portid].len = 0;
872 diff_tsc_power = cur_tsc_power - prev_tsc_power;
873 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
875 prev_tsc_power = cur_tsc_power;
879 * Read packet from RX queues
881 lcore_scaleup_hint = FREQ_CURRENT;
882 lcore_rx_idle_count = 0;
883 for (i = 0; i < qconf->n_rx_queue; ++i) {
884 rx_queue = &(qconf->rx_queue_list[i]);
885 rx_queue->idle_hint = 0;
886 portid = rx_queue->port_id;
887 queueid = rx_queue->queue_id;
889 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
891 stats[lcore_id].nb_rx_processed += nb_rx;
892 if (unlikely(nb_rx == 0)) {
894 * no packet received from rx queue, try to
895 * sleep for a while forcing CPU enter deeper
898 rx_queue->zero_rx_packet_count++;
900 if (rx_queue->zero_rx_packet_count <=
904 rx_queue->idle_hint = power_idle_heuristic(\
905 rx_queue->zero_rx_packet_count);
906 lcore_rx_idle_count++;
908 rx_queue->zero_rx_packet_count = 0;
911 * do not scale up frequency immediately as
912 * user to kernel space communication is costly
913 * which might impact packet I/O for received
916 rx_queue->freq_up_hint =
917 power_freq_scaleup_heuristic(lcore_id,
921 /* Prefetch first packets */
922 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
923 rte_prefetch0(rte_pktmbuf_mtod(
924 pkts_burst[j], void *));
927 /* Prefetch and forward already prefetched packets */
928 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
929 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
930 j + PREFETCH_OFFSET], void *));
931 l3fwd_simple_forward(pkts_burst[j], portid,
935 /* Forward remaining prefetched packets */
936 for (; j < nb_rx; j++) {
937 l3fwd_simple_forward(pkts_burst[j], portid,
942 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
943 for (i = 1, lcore_scaleup_hint =
944 qconf->rx_queue_list[0].freq_up_hint;
945 i < qconf->n_rx_queue; ++i) {
946 rx_queue = &(qconf->rx_queue_list[i]);
947 if (rx_queue->freq_up_hint >
950 rx_queue->freq_up_hint;
953 if (lcore_scaleup_hint == FREQ_HIGHEST)
954 rte_power_freq_max(lcore_id);
955 else if (lcore_scaleup_hint == FREQ_HIGHER)
956 rte_power_freq_up(lcore_id);
959 * All Rx queues empty in recent consecutive polls,
960 * sleep in a conservative manner, meaning sleep as
963 for (i = 1, lcore_idle_hint =
964 qconf->rx_queue_list[0].idle_hint;
965 i < qconf->n_rx_queue; ++i) {
966 rx_queue = &(qconf->rx_queue_list[i]);
967 if (rx_queue->idle_hint < lcore_idle_hint)
968 lcore_idle_hint = rx_queue->idle_hint;
971 if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
973 * execute "pause" instruction to avoid context
974 * switch for short sleep.
976 rte_delay_us(lcore_idle_hint);
978 /* long sleep force runing thread to suspend */
979 usleep(lcore_idle_hint);
981 stats[lcore_id].sleep_time += lcore_idle_hint;
987 check_lcore_params(void)
989 uint8_t queue, lcore;
993 for (i = 0; i < nb_lcore_params; ++i) {
994 queue = lcore_params[i].queue_id;
995 if (queue >= MAX_RX_QUEUE_PER_PORT) {
996 printf("invalid queue number: %hhu\n", queue);
999 lcore = lcore_params[i].lcore_id;
1000 if (!rte_lcore_is_enabled(lcore)) {
1001 printf("error: lcore %hhu is not enabled in lcore "
1005 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1007 printf("warning: lcore %hhu is on socket %d with numa "
1008 "off\n", lcore, socketid);
1015 check_port_config(const unsigned nb_ports)
1020 for (i = 0; i < nb_lcore_params; ++i) {
1021 portid = lcore_params[i].port_id;
1022 if ((enabled_port_mask & (1 << portid)) == 0) {
1023 printf("port %u is not enabled in port mask\n",
1027 if (portid >= nb_ports) {
1028 printf("port %u is not present on the board\n",
1037 get_port_n_rx_queues(const uint8_t port)
1042 for (i = 0; i < nb_lcore_params; ++i) {
1043 if (lcore_params[i].port_id == port &&
1044 lcore_params[i].queue_id > queue)
1045 queue = lcore_params[i].queue_id;
1047 return (uint8_t)(++queue);
1051 init_lcore_rx_queues(void)
1053 uint16_t i, nb_rx_queue;
1056 for (i = 0; i < nb_lcore_params; ++i) {
1057 lcore = lcore_params[i].lcore_id;
1058 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1059 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1060 printf("error: too many queues (%u) for lcore: %u\n",
1061 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1064 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1065 lcore_params[i].port_id;
1066 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1067 lcore_params[i].queue_id;
1068 lcore_conf[lcore].n_rx_queue++;
1076 print_usage(const char *prgname)
1078 printf ("%s [EAL options] -- -p PORTMASK -P"
1079 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1080 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1081 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1082 " -P : enable promiscuous mode\n"
1083 " --config (port,queue,lcore): rx queues configuration\n"
1084 " --no-numa: optional, disable numa awareness\n"
1085 " --enable-jumbo: enable jumbo frame"
1086 " which max packet len is PKTLEN in decimal (64-9600)\n",
1090 static int parse_max_pkt_len(const char *pktlen)
1095 /* parse decimal string */
1096 len = strtoul(pktlen, &end, 10);
1097 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1107 parse_portmask(const char *portmask)
1112 /* parse hexadecimal string */
1113 pm = strtoul(portmask, &end, 16);
1114 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1124 parse_config(const char *q_arg)
1127 const char *p, *p0 = q_arg;
1135 unsigned long int_fld[_NUM_FLD];
1136 char *str_fld[_NUM_FLD];
1140 nb_lcore_params = 0;
1142 while ((p = strchr(p0,'(')) != NULL) {
1144 if((p0 = strchr(p,')')) == NULL)
1148 if(size >= sizeof(s))
1151 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1152 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1155 for (i = 0; i < _NUM_FLD; i++){
1157 int_fld[i] = strtoul(str_fld[i], &end, 0);
1158 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1162 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1163 printf("exceeded max number of lcore params: %hu\n",
1167 lcore_params_array[nb_lcore_params].port_id =
1168 (uint8_t)int_fld[FLD_PORT];
1169 lcore_params_array[nb_lcore_params].queue_id =
1170 (uint8_t)int_fld[FLD_QUEUE];
1171 lcore_params_array[nb_lcore_params].lcore_id =
1172 (uint8_t)int_fld[FLD_LCORE];
1175 lcore_params = lcore_params_array;
1180 /* Parse the argument given in the command line of the application */
1182 parse_args(int argc, char **argv)
1187 char *prgname = argv[0];
1188 static struct option lgopts[] = {
1189 {"config", 1, 0, 0},
1190 {"no-numa", 0, 0, 0},
1191 {"enable-jumbo", 0, 0, 0},
1197 while ((opt = getopt_long(argc, argvopt, "p:P",
1198 lgopts, &option_index)) != EOF) {
1203 enabled_port_mask = parse_portmask(optarg);
1204 if (enabled_port_mask == 0) {
1205 printf("invalid portmask\n");
1206 print_usage(prgname);
1211 printf("Promiscuous mode selected\n");
1217 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1218 ret = parse_config(optarg);
1220 printf("invalid config\n");
1221 print_usage(prgname);
1226 if (!strncmp(lgopts[option_index].name,
1228 printf("numa is disabled \n");
1232 if (!strncmp(lgopts[option_index].name,
1233 "enable-jumbo", 12)) {
1234 struct option lenopts =
1235 {"max-pkt-len", required_argument, \
1238 printf("jumbo frame is enabled \n");
1239 port_conf.rxmode.jumbo_frame = 1;
1242 * if no max-pkt-len set, use the default value
1245 if (0 == getopt_long(argc, argvopt, "",
1246 &lenopts, &option_index)) {
1247 ret = parse_max_pkt_len(optarg);
1249 (ret > MAX_JUMBO_PKT_LEN)){
1250 printf("invalid packet "
1252 print_usage(prgname);
1255 port_conf.rxmode.max_rx_pkt_len = ret;
1257 printf("set jumbo frame "
1258 "max packet length to %u\n",
1259 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1265 print_usage(prgname);
1271 argv[optind-1] = prgname;
1274 optind = 0; /* reset getopt lib */
1279 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1281 printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1282 eth_addr->addr_bytes[0],
1283 eth_addr->addr_bytes[1],
1284 eth_addr->addr_bytes[2],
1285 eth_addr->addr_bytes[3],
1286 eth_addr->addr_bytes[4],
1287 eth_addr->addr_bytes[5]);
1290 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1292 setup_hash(int socketid)
1294 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1296 .entries = L3FWD_HASH_ENTRIES,
1297 .bucket_entries = 4,
1298 .key_len = sizeof(struct ipv4_5tuple),
1299 .hash_func = DEFAULT_HASH_FUNC,
1300 .hash_func_init_val = 0,
1303 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1305 .entries = L3FWD_HASH_ENTRIES,
1306 .bucket_entries = 4,
1307 .key_len = sizeof(struct ipv6_5tuple),
1308 .hash_func = DEFAULT_HASH_FUNC,
1309 .hash_func_init_val = 0,
1316 /* create ipv4 hash */
1317 rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1318 ipv4_l3fwd_hash_params.name = s;
1319 ipv4_l3fwd_hash_params.socket_id = socketid;
1320 ipv4_l3fwd_lookup_struct[socketid] =
1321 rte_hash_create(&ipv4_l3fwd_hash_params);
1322 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1323 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1324 "socket %d\n", socketid);
1326 /* create ipv6 hash */
1327 rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1328 ipv6_l3fwd_hash_params.name = s;
1329 ipv6_l3fwd_hash_params.socket_id = socketid;
1330 ipv6_l3fwd_lookup_struct[socketid] =
1331 rte_hash_create(&ipv6_l3fwd_hash_params);
1332 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1333 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1334 "socket %d\n", socketid);
1337 /* populate the ipv4 hash */
1338 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1339 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1340 (void *) &ipv4_l3fwd_route_array[i].key);
1342 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1343 "l3fwd hash on socket %d\n", i, socketid);
1345 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1346 printf("Hash: Adding key\n");
1347 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1350 /* populate the ipv6 hash */
1351 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1352 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1353 (void *) &ipv6_l3fwd_route_array[i].key);
1355 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1356 "l3fwd hash on socket %d\n", i, socketid);
1358 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1359 printf("Hash: Adding key\n");
1360 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1365 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1367 setup_lpm(int socketid)
1373 /* create the LPM table */
1374 rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1375 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1376 IPV4_L3FWD_LPM_MAX_RULES, 0);
1377 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1378 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1379 " on socket %d\n", socketid);
1381 /* populate the LPM table */
1382 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1383 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1384 ipv4_l3fwd_route_array[i].ip,
1385 ipv4_l3fwd_route_array[i].depth,
1386 ipv4_l3fwd_route_array[i].if_out);
1389 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1390 "l3fwd LPM table on socket %d\n",
1394 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1395 (unsigned)ipv4_l3fwd_route_array[i].ip,
1396 ipv4_l3fwd_route_array[i].depth,
1397 ipv4_l3fwd_route_array[i].if_out);
1403 init_mem(unsigned nb_mbuf)
1405 struct lcore_conf *qconf;
1410 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1411 if (rte_lcore_is_enabled(lcore_id) == 0)
1415 socketid = rte_lcore_to_socket_id(lcore_id);
1419 if (socketid >= NB_SOCKETS) {
1420 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1421 "out of range %d\n", socketid,
1422 lcore_id, NB_SOCKETS);
1424 if (pktmbuf_pool[socketid] == NULL) {
1425 rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1426 pktmbuf_pool[socketid] =
1427 rte_mempool_create(s, nb_mbuf,
1428 MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1429 sizeof(struct rte_pktmbuf_pool_private),
1430 rte_pktmbuf_pool_init, NULL,
1431 rte_pktmbuf_init, NULL,
1433 if (pktmbuf_pool[socketid] == NULL)
1434 rte_exit(EXIT_FAILURE,
1435 "Cannot init mbuf pool on socket %d\n",
1438 printf("Allocated mbuf pool on socket %d\n",
1441 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1442 setup_lpm(socketid);
1444 setup_hash(socketid);
1447 qconf = &lcore_conf[lcore_id];
1448 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1449 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1450 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1456 /* Check the link status of all ports in up to 9s, and print them finally */
1458 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1460 #define CHECK_INTERVAL 100 /* 100ms */
1461 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1462 uint8_t portid, count, all_ports_up, print_flag = 0;
1463 struct rte_eth_link link;
1465 printf("\nChecking link status");
1467 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1469 for (portid = 0; portid < port_num; portid++) {
1470 if ((port_mask & (1 << portid)) == 0)
1472 memset(&link, 0, sizeof(link));
1473 rte_eth_link_get_nowait(portid, &link);
1474 /* print link status if flag set */
1475 if (print_flag == 1) {
1476 if (link.link_status)
1477 printf("Port %d Link Up - speed %u "
1478 "Mbps - %s\n", (uint8_t)portid,
1479 (unsigned)link.link_speed,
1480 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1481 ("full-duplex") : ("half-duplex\n"));
1483 printf("Port %d Link Down\n",
1487 /* clear all_ports_up flag if any link down */
1488 if (link.link_status == 0) {
1493 /* after finally printing all link status, get out */
1494 if (print_flag == 1)
1497 if (all_ports_up == 0) {
1500 rte_delay_ms(CHECK_INTERVAL);
1503 /* set the print_flag if all ports up or timeout */
1504 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1512 MAIN(int argc, char **argv)
1514 struct lcore_conf *qconf;
1520 uint32_t n_tx_queue, nb_lcores;
1521 uint8_t portid, nb_rx_queue, queue, socketid;
1523 /* catch SIGINT and restore cpufreq governor to ondemand */
1524 signal(SIGINT, signal_exit_now);
1527 ret = rte_eal_init(argc, argv);
1529 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1533 /* init RTE timer library to be used late */
1534 rte_timer_subsystem_init();
1536 /* parse application arguments (after the EAL ones) */
1537 ret = parse_args(argc, argv);
1539 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1541 if (check_lcore_params() < 0)
1542 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1544 ret = init_lcore_rx_queues();
1546 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1549 /* init driver(s) */
1550 if (rte_pmd_init_all() < 0)
1551 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1553 if (rte_eal_pci_probe() < 0)
1554 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1556 nb_ports = rte_eth_dev_count();
1557 if (nb_ports > RTE_MAX_ETHPORTS)
1558 nb_ports = RTE_MAX_ETHPORTS;
1560 if (check_port_config(nb_ports) < 0)
1561 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1563 nb_lcores = rte_lcore_count();
1565 /* initialize all ports */
1566 for (portid = 0; portid < nb_ports; portid++) {
1567 /* skip ports that are not enabled */
1568 if ((enabled_port_mask & (1 << portid)) == 0) {
1569 printf("\nSkipping disabled port %d\n", portid);
1574 printf("Initializing port %d ... ", portid );
1577 nb_rx_queue = get_port_n_rx_queues(portid);
1578 n_tx_queue = nb_lcores;
1579 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1580 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1581 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1582 nb_rx_queue, (unsigned)n_tx_queue );
1583 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1584 (uint16_t)n_tx_queue, &port_conf);
1586 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1587 "err=%d, port=%d\n", ret, portid);
1589 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1590 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1594 ret = init_mem(NB_MBUF);
1596 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1598 /* init one TX queue per couple (lcore,port) */
1600 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1601 if (rte_lcore_is_enabled(lcore_id) == 0)
1606 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1610 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1612 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1613 socketid, &tx_conf);
1615 rte_exit(EXIT_FAILURE,
1616 "rte_eth_tx_queue_setup: err=%d, "
1617 "port=%d\n", ret, portid);
1619 qconf = &lcore_conf[lcore_id];
1620 qconf->tx_queue_id[portid] = queueid;
1626 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1627 if (rte_lcore_is_enabled(lcore_id) == 0)
1630 /* init power management library */
1631 ret = rte_power_init(lcore_id);
1633 rte_exit(EXIT_FAILURE, "Power management library "
1634 "initialization failed on core%u\n", lcore_id);
1636 /* init timer structures for each enabled lcore */
1637 rte_timer_init(&power_timers[lcore_id]);
1638 hz = rte_get_timer_hz();
1639 rte_timer_reset(&power_timers[lcore_id],
1640 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1641 power_timer_cb, NULL);
1643 qconf = &lcore_conf[lcore_id];
1644 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1646 /* init RX queues */
1647 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1648 portid = qconf->rx_queue_list[queue].port_id;
1649 queueid = qconf->rx_queue_list[queue].queue_id;
1653 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1657 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1660 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1661 socketid, &rx_conf, pktmbuf_pool[socketid]);
1663 rte_exit(EXIT_FAILURE,
1664 "rte_eth_rx_queue_setup: err=%d, "
1665 "port=%d\n", ret, portid);
1672 for (portid = 0; portid < nb_ports; portid++) {
1673 if ((enabled_port_mask & (1 << portid)) == 0) {
1677 ret = rte_eth_dev_start(portid);
1679 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1680 "port=%d\n", ret, portid);
1683 * If enabled, put device in promiscuous mode.
1684 * This allows IO forwarding mode to forward packets
1685 * to itself through 2 cross-connected ports of the
1689 rte_eth_promiscuous_enable(portid);
1692 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1694 /* launch per-lcore init on every lcore */
1695 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1696 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1697 if (rte_eal_wait_lcore(lcore_id) < 0)