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>
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>
75 #include <rte_timer.h>
76 #include <rte_power.h>
78 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
80 #define MAX_PKT_BURST 32
82 #define MIN_ZERO_POLL_COUNT 5
84 /* around 100ms at 2 Ghz */
85 #define TIMER_RESOLUTION_CYCLES 200000000ULL
87 #define TIMER_NUMBER_PER_SECOND 10
89 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
90 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
92 #define APP_LOOKUP_EXACT_MATCH 0
93 #define APP_LOOKUP_LPM 1
94 #define DO_RFC_1812_CHECKS
96 #ifndef APP_LOOKUP_METHOD
97 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
100 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
101 #include <rte_hash.h>
102 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
105 #error "APP_LOOKUP_METHOD set to incorrect value"
109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
110 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
111 #define IPv6_BYTES(addr) \
112 addr[0], addr[1], addr[2], addr[3], \
113 addr[4], addr[5], addr[6], addr[7], \
114 addr[8], addr[9], addr[10], addr[11],\
115 addr[12], addr[13],addr[14], addr[15]
118 #define MAX_JUMBO_PKT_LEN 9600
120 #define IPV6_ADDR_LEN 16
122 #define MEMPOOL_CACHE_SIZE 256
125 * This expression is used to calculate the number of mbufs needed depending on
126 * user input, taking into account memory for rx and tx hardware rings, cache
127 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
128 * NB_MBUF never goes below a minimum value of 8192.
131 #define NB_MBUF RTE_MAX ( \
132 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
133 nb_ports*nb_lcores*MAX_PKT_BURST + \
134 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
135 nb_lcores*MEMPOOL_CACHE_SIZE), \
138 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
142 /* Configure how many packets ahead to prefetch, when reading packets */
143 #define PREFETCH_OFFSET 3
146 * Configurable number of RX/TX ring descriptors
148 #define RTE_TEST_RX_DESC_DEFAULT 128
149 #define RTE_TEST_TX_DESC_DEFAULT 512
150 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
151 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
153 /* ethernet addresses of ports */
154 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
156 /* mask of enabled ports */
157 static uint32_t enabled_port_mask = 0;
158 /* Ports set in promiscuous mode off by default. */
159 static int promiscuous_on = 0;
160 /* NUMA is enabled by default. */
161 static int numa_on = 1;
163 enum freq_scale_hint_t
173 struct rte_mbuf *m_table[MAX_PKT_BURST];
176 struct lcore_rx_queue {
179 enum freq_scale_hint_t freq_up_hint;
180 uint32_t zero_rx_packet_count;
182 } __rte_cache_aligned;
184 #define MAX_RX_QUEUE_PER_LCORE 16
185 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
186 #define MAX_RX_QUEUE_PER_PORT 128
188 #define MAX_LCORE_PARAMS 1024
189 struct lcore_params {
193 } __rte_cache_aligned;
195 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
196 static struct lcore_params lcore_params_array_default[] = {
208 static struct lcore_params * lcore_params = lcore_params_array_default;
209 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
210 sizeof(lcore_params_array_default[0]);
212 static struct rte_eth_conf port_conf = {
214 .mq_mode = ETH_MQ_RX_RSS,
215 .max_rx_pkt_len = ETHER_MAX_LEN,
217 .header_split = 0, /**< Header Split disabled */
218 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
219 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
220 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
221 .hw_strip_crc = 0, /**< CRC stripped by hardware */
226 .rss_hf = ETH_RSS_IP,
230 .mq_mode = ETH_DCB_NONE,
234 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
237 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
239 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
240 #include <rte_hash_crc.h>
241 #define DEFAULT_HASH_FUNC rte_hash_crc
243 #include <rte_jhash.h>
244 #define DEFAULT_HASH_FUNC rte_jhash
253 } __attribute__((__packed__));
256 uint8_t ip_dst[IPV6_ADDR_LEN];
257 uint8_t ip_src[IPV6_ADDR_LEN];
261 } __attribute__((__packed__));
263 struct ipv4_l3fwd_route {
264 struct ipv4_5tuple key;
268 struct ipv6_l3fwd_route {
269 struct ipv6_5tuple key;
273 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
274 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
275 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
276 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
277 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
280 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
283 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
284 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
285 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
286 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
292 typedef struct rte_hash lookup_struct_t;
293 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
294 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
296 #define L3FWD_HASH_ENTRIES 1024
298 #define IPV4_L3FWD_NUM_ROUTES \
299 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
301 #define IPV6_L3FWD_NUM_ROUTES \
302 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
304 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
305 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
308 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
309 struct ipv4_l3fwd_route {
315 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
316 {IPv4(1,1,1,0), 24, 0},
317 {IPv4(2,1,1,0), 24, 1},
318 {IPv4(3,1,1,0), 24, 2},
319 {IPv4(4,1,1,0), 24, 3},
320 {IPv4(5,1,1,0), 24, 4},
321 {IPv4(6,1,1,0), 24, 5},
322 {IPv4(7,1,1,0), 24, 6},
323 {IPv4(8,1,1,0), 24, 7},
326 #define IPV4_L3FWD_NUM_ROUTES \
327 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
329 #define IPV4_L3FWD_LPM_MAX_RULES 1024
331 typedef struct rte_lpm lookup_struct_t;
332 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
337 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
338 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
339 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
340 lookup_struct_t * ipv4_lookup_struct;
341 lookup_struct_t * ipv6_lookup_struct;
342 } __rte_cache_aligned;
345 /* total sleep time in ms since last frequency scaling down */
347 /* number of long sleep recently */
348 uint32_t nb_long_sleep;
349 /* freq. scaling up trend */
351 /* total packet processed recently */
352 uint64_t nb_rx_processed;
353 /* total iterations looped recently */
354 uint64_t nb_iteration_looped;
356 } __rte_cache_aligned;
358 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
359 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
360 static struct rte_timer power_timers[RTE_MAX_LCORE];
362 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
363 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
364 unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
366 /* exit signal handler */
368 signal_exit_now(int sigtype)
373 if (sigtype == SIGINT) {
374 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
375 if (rte_lcore_is_enabled(lcore_id) == 0)
378 /* init power management library */
379 ret = rte_power_exit(lcore_id);
381 rte_exit(EXIT_FAILURE, "Power management "
382 "library de-initialization failed on "
383 "core%u\n", lcore_id);
387 rte_exit(EXIT_SUCCESS, "User forced exit\n");
390 /* Freqency scale down timer callback */
392 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
393 __attribute__((unused)) void *arg)
396 float sleep_time_ratio;
397 unsigned lcore_id = rte_lcore_id();
399 /* accumulate total execution time in us when callback is invoked */
400 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
401 (float)SCALING_PERIOD;
404 * check whether need to scale down frequency a step if it sleep a lot.
406 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
407 rte_power_freq_down(lcore_id);
408 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
409 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
411 * scale down a step if average packet per iteration less
414 rte_power_freq_down(lcore_id);
417 * initialize another timer according to current frequency to ensure
418 * timer interval is relatively fixed.
420 hz = rte_get_timer_hz();
421 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
422 SINGLE, lcore_id, power_timer_cb, NULL);
424 stats[lcore_id].nb_rx_processed = 0;
425 stats[lcore_id].nb_iteration_looped = 0;
427 stats[lcore_id].sleep_time = 0;
430 /* Send burst of packets on an output interface */
432 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
434 struct rte_mbuf **m_table;
438 queueid = qconf->tx_queue_id[port];
439 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
441 ret = rte_eth_tx_burst(port, queueid, m_table, n);
442 if (unlikely(ret < n)) {
444 rte_pktmbuf_free(m_table[ret]);
451 /* Enqueue a single packet, and send burst if queue is filled */
453 send_single_packet(struct rte_mbuf *m, uint8_t port)
457 struct lcore_conf *qconf;
459 lcore_id = rte_lcore_id();
461 qconf = &lcore_conf[lcore_id];
462 len = qconf->tx_mbufs[port].len;
463 qconf->tx_mbufs[port].m_table[len] = m;
466 /* enough pkts to be sent */
467 if (unlikely(len == MAX_PKT_BURST)) {
468 send_burst(qconf, MAX_PKT_BURST, port);
472 qconf->tx_mbufs[port].len = len;
476 #ifdef DO_RFC_1812_CHECKS
478 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
480 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
482 * 1. The packet length reported by the Link Layer must be large
483 * enough to hold the minimum length legal IP datagram (20 bytes).
485 if (link_len < sizeof(struct ipv4_hdr))
488 /* 2. The IP checksum must be correct. */
489 /* this is checked in H/W */
492 * 3. The IP version number must be 4. If the version number is not 4
493 * then the packet may be another version of IP, such as IPng or
496 if (((pkt->version_ihl) >> 4) != 4)
499 * 4. The IP header length field must be large enough to hold the
500 * minimum length legal IP datagram (20 bytes = 5 words).
502 if ((pkt->version_ihl & 0xf) < 5)
506 * 5. The IP total length field must be large enough to hold the IP
507 * datagram header, whose length is specified in the IP header length
510 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
517 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
519 print_ipv4_key(struct ipv4_5tuple key)
521 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
522 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
523 key.port_dst, key.port_src, key.proto);
526 print_ipv6_key(struct ipv6_5tuple key)
528 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
529 "port dst = %d, port src = %d, proto = %d\n",
530 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
531 key.port_dst, key.port_src, key.proto);
534 static inline uint8_t
535 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
536 lookup_struct_t * ipv4_l3fwd_lookup_struct)
538 struct ipv4_5tuple key;
543 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
544 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
545 key.proto = ipv4_hdr->next_proto_id;
547 switch (ipv4_hdr->next_proto_id) {
549 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
550 sizeof(struct ipv4_hdr));
551 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
552 key.port_src = rte_be_to_cpu_16(tcp->src_port);
556 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
557 sizeof(struct ipv4_hdr));
558 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
559 key.port_src = rte_be_to_cpu_16(udp->src_port);
568 /* Find destination port */
569 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
570 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
573 static inline uint8_t
574 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
575 lookup_struct_t *ipv6_l3fwd_lookup_struct)
577 struct ipv6_5tuple key;
582 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
583 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
585 key.proto = ipv6_hdr->proto;
587 switch (ipv6_hdr->proto) {
589 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
590 sizeof(struct ipv6_hdr));
591 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
592 key.port_src = rte_be_to_cpu_16(tcp->src_port);
596 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
597 sizeof(struct ipv6_hdr));
598 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
599 key.port_src = rte_be_to_cpu_16(udp->src_port);
608 /* Find destination port */
609 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
610 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
614 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
615 static inline uint8_t
616 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
617 lookup_struct_t *ipv4_l3fwd_lookup_struct)
621 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
622 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
628 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
629 struct lcore_conf *qconf)
631 struct ether_hdr *eth_hdr;
632 struct ipv4_hdr *ipv4_hdr;
636 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
638 if (m->ol_flags & PKT_RX_IPV4_HDR) {
639 /* Handle IPv4 headers.*/
641 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
642 sizeof(struct ether_hdr));
644 #ifdef DO_RFC_1812_CHECKS
645 /* Check to make sure the packet is valid (RFC1812) */
646 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
652 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
653 qconf->ipv4_lookup_struct);
654 if (dst_port >= RTE_MAX_ETHPORTS ||
655 (enabled_port_mask & 1 << dst_port) == 0)
658 /* 02:00:00:00:00:xx */
659 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
660 *((uint64_t *)d_addr_bytes) =
661 0x000000000002 + ((uint64_t)dst_port << 40);
663 #ifdef DO_RFC_1812_CHECKS
664 /* Update time to live and header checksum */
665 --(ipv4_hdr->time_to_live);
666 ++(ipv4_hdr->hdr_checksum);
670 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
672 send_single_packet(m, dst_port);
675 /* Handle IPv6 headers.*/
676 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
677 struct ipv6_hdr *ipv6_hdr;
680 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
681 sizeof(struct ether_hdr));
683 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
684 qconf->ipv6_lookup_struct);
686 if (dst_port >= RTE_MAX_ETHPORTS ||
687 (enabled_port_mask & 1 << dst_port) == 0)
690 /* 02:00:00:00:00:xx */
691 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
692 *((uint64_t *)d_addr_bytes) =
693 0x000000000002 + ((uint64_t)dst_port << 40);
696 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
698 send_single_packet(m, dst_port);
700 /* We don't currently handle IPv6 packets in LPM mode. */
707 #define SLEEP_GEAR1_THRESHOLD 100
708 #define SLEEP_GEAR2_THRESHOLD 1000
710 static inline uint32_t
711 power_idle_heuristic(uint32_t zero_rx_packet_count)
713 /* If zero count is less than 100, use it as the sleep time in us */
714 if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
715 return zero_rx_packet_count;
716 /* If zero count is less than 1000, sleep time should be 100 us */
717 else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
718 (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
719 return SLEEP_GEAR1_THRESHOLD;
720 /* If zero count is greater than 1000, sleep time should be 1000 us */
721 else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
722 return SLEEP_GEAR2_THRESHOLD;
727 static inline enum freq_scale_hint_t
728 power_freq_scaleup_heuristic(unsigned lcore_id,
733 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
736 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
737 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
738 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
739 #define FREQ_UP_TREND1_ACC 1
740 #define FREQ_UP_TREND2_ACC 100
741 #define FREQ_UP_THRESHOLD 10000
743 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
744 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
745 stats[lcore_id].trend = 0;
747 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
748 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
749 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
750 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
751 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
752 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
754 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
755 stats[lcore_id].trend = 0;
762 /* main processing loop */
764 main_loop(__attribute__((unused)) void *dummy)
766 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
768 uint64_t prev_tsc, diff_tsc, cur_tsc;
769 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
771 uint8_t portid, queueid;
772 struct lcore_conf *qconf;
773 struct lcore_rx_queue *rx_queue;
774 enum freq_scale_hint_t lcore_scaleup_hint;
776 uint32_t lcore_rx_idle_count = 0;
777 uint32_t lcore_idle_hint = 0;
779 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
783 lcore_id = rte_lcore_id();
784 qconf = &lcore_conf[lcore_id];
786 if (qconf->n_rx_queue == 0) {
787 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
791 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
793 for (i = 0; i < qconf->n_rx_queue; i++) {
795 portid = qconf->rx_queue_list[i].port_id;
796 queueid = qconf->rx_queue_list[i].queue_id;
797 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
798 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
802 stats[lcore_id].nb_iteration_looped++;
804 cur_tsc = rte_rdtsc();
805 cur_tsc_power = cur_tsc;
808 * TX burst queue drain
810 diff_tsc = cur_tsc - prev_tsc;
811 if (unlikely(diff_tsc > drain_tsc)) {
814 * This could be optimized (use queueid instead of
815 * portid), but it is not called so often
817 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
818 if (qconf->tx_mbufs[portid].len == 0)
820 send_burst(&lcore_conf[lcore_id],
821 qconf->tx_mbufs[portid].len,
823 qconf->tx_mbufs[portid].len = 0;
829 diff_tsc_power = cur_tsc_power - prev_tsc_power;
830 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
832 prev_tsc_power = cur_tsc_power;
836 * Read packet from RX queues
838 lcore_scaleup_hint = FREQ_CURRENT;
839 lcore_rx_idle_count = 0;
840 for (i = 0; i < qconf->n_rx_queue; ++i) {
841 rx_queue = &(qconf->rx_queue_list[i]);
842 rx_queue->idle_hint = 0;
843 portid = rx_queue->port_id;
844 queueid = rx_queue->queue_id;
846 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
848 stats[lcore_id].nb_rx_processed += nb_rx;
849 if (unlikely(nb_rx == 0)) {
851 * no packet received from rx queue, try to
852 * sleep for a while forcing CPU enter deeper
855 rx_queue->zero_rx_packet_count++;
857 if (rx_queue->zero_rx_packet_count <=
861 rx_queue->idle_hint = power_idle_heuristic(\
862 rx_queue->zero_rx_packet_count);
863 lcore_rx_idle_count++;
865 rx_queue->zero_rx_packet_count = 0;
868 * do not scale up frequency immediately as
869 * user to kernel space communication is costly
870 * which might impact packet I/O for received
873 rx_queue->freq_up_hint =
874 power_freq_scaleup_heuristic(lcore_id,
878 /* Prefetch first packets */
879 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
880 rte_prefetch0(rte_pktmbuf_mtod(
881 pkts_burst[j], void *));
884 /* Prefetch and forward already prefetched packets */
885 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
886 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
887 j + PREFETCH_OFFSET], void *));
888 l3fwd_simple_forward(pkts_burst[j], portid,
892 /* Forward remaining prefetched packets */
893 for (; j < nb_rx; j++) {
894 l3fwd_simple_forward(pkts_burst[j], portid,
899 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
900 for (i = 1, lcore_scaleup_hint =
901 qconf->rx_queue_list[0].freq_up_hint;
902 i < qconf->n_rx_queue; ++i) {
903 rx_queue = &(qconf->rx_queue_list[i]);
904 if (rx_queue->freq_up_hint >
907 rx_queue->freq_up_hint;
910 if (lcore_scaleup_hint == FREQ_HIGHEST)
911 rte_power_freq_max(lcore_id);
912 else if (lcore_scaleup_hint == FREQ_HIGHER)
913 rte_power_freq_up(lcore_id);
916 * All Rx queues empty in recent consecutive polls,
917 * sleep in a conservative manner, meaning sleep as
920 for (i = 1, lcore_idle_hint =
921 qconf->rx_queue_list[0].idle_hint;
922 i < qconf->n_rx_queue; ++i) {
923 rx_queue = &(qconf->rx_queue_list[i]);
924 if (rx_queue->idle_hint < lcore_idle_hint)
925 lcore_idle_hint = rx_queue->idle_hint;
928 if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
930 * execute "pause" instruction to avoid context
931 * switch for short sleep.
933 rte_delay_us(lcore_idle_hint);
935 /* long sleep force runing thread to suspend */
936 usleep(lcore_idle_hint);
938 stats[lcore_id].sleep_time += lcore_idle_hint;
944 check_lcore_params(void)
946 uint8_t queue, lcore;
950 for (i = 0; i < nb_lcore_params; ++i) {
951 queue = lcore_params[i].queue_id;
952 if (queue >= MAX_RX_QUEUE_PER_PORT) {
953 printf("invalid queue number: %hhu\n", queue);
956 lcore = lcore_params[i].lcore_id;
957 if (!rte_lcore_is_enabled(lcore)) {
958 printf("error: lcore %hhu is not enabled in lcore "
962 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
964 printf("warning: lcore %hhu is on socket %d with numa "
965 "off\n", lcore, socketid);
972 check_port_config(const unsigned nb_ports)
977 for (i = 0; i < nb_lcore_params; ++i) {
978 portid = lcore_params[i].port_id;
979 if ((enabled_port_mask & (1 << portid)) == 0) {
980 printf("port %u is not enabled in port mask\n",
984 if (portid >= nb_ports) {
985 printf("port %u is not present on the board\n",
994 get_port_n_rx_queues(const uint8_t port)
999 for (i = 0; i < nb_lcore_params; ++i) {
1000 if (lcore_params[i].port_id == port &&
1001 lcore_params[i].queue_id > queue)
1002 queue = lcore_params[i].queue_id;
1004 return (uint8_t)(++queue);
1008 init_lcore_rx_queues(void)
1010 uint16_t i, nb_rx_queue;
1013 for (i = 0; i < nb_lcore_params; ++i) {
1014 lcore = lcore_params[i].lcore_id;
1015 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1016 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1017 printf("error: too many queues (%u) for lcore: %u\n",
1018 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1021 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1022 lcore_params[i].port_id;
1023 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1024 lcore_params[i].queue_id;
1025 lcore_conf[lcore].n_rx_queue++;
1033 print_usage(const char *prgname)
1035 printf ("%s [EAL options] -- -p PORTMASK -P"
1036 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1037 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1038 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1039 " -P : enable promiscuous mode\n"
1040 " --config (port,queue,lcore): rx queues configuration\n"
1041 " --no-numa: optional, disable numa awareness\n"
1042 " --enable-jumbo: enable jumbo frame"
1043 " which max packet len is PKTLEN in decimal (64-9600)\n",
1047 static int parse_max_pkt_len(const char *pktlen)
1052 /* parse decimal string */
1053 len = strtoul(pktlen, &end, 10);
1054 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1064 parse_portmask(const char *portmask)
1069 /* parse hexadecimal string */
1070 pm = strtoul(portmask, &end, 16);
1071 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1081 parse_config(const char *q_arg)
1084 const char *p, *p0 = q_arg;
1092 unsigned long int_fld[_NUM_FLD];
1093 char *str_fld[_NUM_FLD];
1097 nb_lcore_params = 0;
1099 while ((p = strchr(p0,'(')) != NULL) {
1101 if((p0 = strchr(p,')')) == NULL)
1105 if(size >= sizeof(s))
1108 snprintf(s, sizeof(s), "%.*s", size, p);
1109 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1112 for (i = 0; i < _NUM_FLD; i++){
1114 int_fld[i] = strtoul(str_fld[i], &end, 0);
1115 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1119 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1120 printf("exceeded max number of lcore params: %hu\n",
1124 lcore_params_array[nb_lcore_params].port_id =
1125 (uint8_t)int_fld[FLD_PORT];
1126 lcore_params_array[nb_lcore_params].queue_id =
1127 (uint8_t)int_fld[FLD_QUEUE];
1128 lcore_params_array[nb_lcore_params].lcore_id =
1129 (uint8_t)int_fld[FLD_LCORE];
1132 lcore_params = lcore_params_array;
1137 /* Parse the argument given in the command line of the application */
1139 parse_args(int argc, char **argv)
1144 char *prgname = argv[0];
1145 static struct option lgopts[] = {
1146 {"config", 1, 0, 0},
1147 {"no-numa", 0, 0, 0},
1148 {"enable-jumbo", 0, 0, 0},
1154 while ((opt = getopt_long(argc, argvopt, "p:P",
1155 lgopts, &option_index)) != EOF) {
1160 enabled_port_mask = parse_portmask(optarg);
1161 if (enabled_port_mask == 0) {
1162 printf("invalid portmask\n");
1163 print_usage(prgname);
1168 printf("Promiscuous mode selected\n");
1174 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1175 ret = parse_config(optarg);
1177 printf("invalid config\n");
1178 print_usage(prgname);
1183 if (!strncmp(lgopts[option_index].name,
1185 printf("numa is disabled \n");
1189 if (!strncmp(lgopts[option_index].name,
1190 "enable-jumbo", 12)) {
1191 struct option lenopts =
1192 {"max-pkt-len", required_argument, \
1195 printf("jumbo frame is enabled \n");
1196 port_conf.rxmode.jumbo_frame = 1;
1199 * if no max-pkt-len set, use the default value
1202 if (0 == getopt_long(argc, argvopt, "",
1203 &lenopts, &option_index)) {
1204 ret = parse_max_pkt_len(optarg);
1206 (ret > MAX_JUMBO_PKT_LEN)){
1207 printf("invalid packet "
1209 print_usage(prgname);
1212 port_conf.rxmode.max_rx_pkt_len = ret;
1214 printf("set jumbo frame "
1215 "max packet length to %u\n",
1216 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1222 print_usage(prgname);
1228 argv[optind-1] = prgname;
1231 optind = 0; /* reset getopt lib */
1236 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1238 char buf[ETHER_ADDR_FMT_SIZE];
1239 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1240 printf("%s%s", name, buf);
1243 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1245 setup_hash(int socketid)
1247 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1249 .entries = L3FWD_HASH_ENTRIES,
1250 .bucket_entries = 4,
1251 .key_len = sizeof(struct ipv4_5tuple),
1252 .hash_func = DEFAULT_HASH_FUNC,
1253 .hash_func_init_val = 0,
1256 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1258 .entries = L3FWD_HASH_ENTRIES,
1259 .bucket_entries = 4,
1260 .key_len = sizeof(struct ipv6_5tuple),
1261 .hash_func = DEFAULT_HASH_FUNC,
1262 .hash_func_init_val = 0,
1269 /* create ipv4 hash */
1270 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1271 ipv4_l3fwd_hash_params.name = s;
1272 ipv4_l3fwd_hash_params.socket_id = socketid;
1273 ipv4_l3fwd_lookup_struct[socketid] =
1274 rte_hash_create(&ipv4_l3fwd_hash_params);
1275 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1276 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1277 "socket %d\n", socketid);
1279 /* create ipv6 hash */
1280 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1281 ipv6_l3fwd_hash_params.name = s;
1282 ipv6_l3fwd_hash_params.socket_id = socketid;
1283 ipv6_l3fwd_lookup_struct[socketid] =
1284 rte_hash_create(&ipv6_l3fwd_hash_params);
1285 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1286 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1287 "socket %d\n", socketid);
1290 /* populate the ipv4 hash */
1291 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1292 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1293 (void *) &ipv4_l3fwd_route_array[i].key);
1295 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1296 "l3fwd hash on socket %d\n", i, socketid);
1298 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1299 printf("Hash: Adding key\n");
1300 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1303 /* populate the ipv6 hash */
1304 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1305 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1306 (void *) &ipv6_l3fwd_route_array[i].key);
1308 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1309 "l3fwd hash on socket %d\n", i, socketid);
1311 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1312 printf("Hash: Adding key\n");
1313 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1318 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1320 setup_lpm(int socketid)
1326 /* create the LPM table */
1327 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1328 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1329 IPV4_L3FWD_LPM_MAX_RULES, 0);
1330 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1331 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1332 " on socket %d\n", socketid);
1334 /* populate the LPM table */
1335 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1336 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1337 ipv4_l3fwd_route_array[i].ip,
1338 ipv4_l3fwd_route_array[i].depth,
1339 ipv4_l3fwd_route_array[i].if_out);
1342 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1343 "l3fwd LPM table on socket %d\n",
1347 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1348 (unsigned)ipv4_l3fwd_route_array[i].ip,
1349 ipv4_l3fwd_route_array[i].depth,
1350 ipv4_l3fwd_route_array[i].if_out);
1356 init_mem(unsigned nb_mbuf)
1358 struct lcore_conf *qconf;
1363 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1364 if (rte_lcore_is_enabled(lcore_id) == 0)
1368 socketid = rte_lcore_to_socket_id(lcore_id);
1372 if (socketid >= NB_SOCKETS) {
1373 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1374 "out of range %d\n", socketid,
1375 lcore_id, NB_SOCKETS);
1377 if (pktmbuf_pool[socketid] == NULL) {
1378 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1379 pktmbuf_pool[socketid] =
1380 rte_pktmbuf_pool_create(s, nb_mbuf,
1381 MEMPOOL_CACHE_SIZE, 0,
1382 RTE_MBUF_DEFAULT_BUF_SIZE,
1384 if (pktmbuf_pool[socketid] == NULL)
1385 rte_exit(EXIT_FAILURE,
1386 "Cannot init mbuf pool on socket %d\n",
1389 printf("Allocated mbuf pool on socket %d\n",
1392 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1393 setup_lpm(socketid);
1395 setup_hash(socketid);
1398 qconf = &lcore_conf[lcore_id];
1399 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1400 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1401 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1407 /* Check the link status of all ports in up to 9s, and print them finally */
1409 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1411 #define CHECK_INTERVAL 100 /* 100ms */
1412 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1413 uint8_t portid, count, all_ports_up, print_flag = 0;
1414 struct rte_eth_link link;
1416 printf("\nChecking link status");
1418 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1420 for (portid = 0; portid < port_num; portid++) {
1421 if ((port_mask & (1 << portid)) == 0)
1423 memset(&link, 0, sizeof(link));
1424 rte_eth_link_get_nowait(portid, &link);
1425 /* print link status if flag set */
1426 if (print_flag == 1) {
1427 if (link.link_status)
1428 printf("Port %d Link Up - speed %u "
1429 "Mbps - %s\n", (uint8_t)portid,
1430 (unsigned)link.link_speed,
1431 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1432 ("full-duplex") : ("half-duplex\n"));
1434 printf("Port %d Link Down\n",
1438 /* clear all_ports_up flag if any link down */
1439 if (link.link_status == 0) {
1444 /* after finally printing all link status, get out */
1445 if (print_flag == 1)
1448 if (all_ports_up == 0) {
1451 rte_delay_ms(CHECK_INTERVAL);
1454 /* set the print_flag if all ports up or timeout */
1455 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1463 main(int argc, char **argv)
1465 struct lcore_conf *qconf;
1466 struct rte_eth_dev_info dev_info;
1467 struct rte_eth_txconf *txconf;
1473 uint32_t n_tx_queue, nb_lcores;
1474 uint8_t portid, nb_rx_queue, queue, socketid;
1476 /* catch SIGINT and restore cpufreq governor to ondemand */
1477 signal(SIGINT, signal_exit_now);
1480 ret = rte_eal_init(argc, argv);
1482 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1486 /* init RTE timer library to be used late */
1487 rte_timer_subsystem_init();
1489 /* parse application arguments (after the EAL ones) */
1490 ret = parse_args(argc, argv);
1492 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1494 if (check_lcore_params() < 0)
1495 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1497 ret = init_lcore_rx_queues();
1499 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1502 nb_ports = rte_eth_dev_count();
1503 if (nb_ports > RTE_MAX_ETHPORTS)
1504 nb_ports = RTE_MAX_ETHPORTS;
1506 if (check_port_config(nb_ports) < 0)
1507 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1509 nb_lcores = rte_lcore_count();
1511 /* initialize all ports */
1512 for (portid = 0; portid < nb_ports; portid++) {
1513 /* skip ports that are not enabled */
1514 if ((enabled_port_mask & (1 << portid)) == 0) {
1515 printf("\nSkipping disabled port %d\n", portid);
1520 printf("Initializing port %d ... ", portid );
1523 nb_rx_queue = get_port_n_rx_queues(portid);
1524 n_tx_queue = nb_lcores;
1525 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1526 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1527 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1528 nb_rx_queue, (unsigned)n_tx_queue );
1529 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1530 (uint16_t)n_tx_queue, &port_conf);
1532 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1533 "err=%d, port=%d\n", ret, portid);
1535 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1536 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1540 ret = init_mem(NB_MBUF);
1542 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1544 /* init one TX queue per couple (lcore,port) */
1546 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1547 if (rte_lcore_is_enabled(lcore_id) == 0)
1552 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1556 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1559 rte_eth_dev_info_get(portid, &dev_info);
1560 txconf = &dev_info.default_txconf;
1561 if (port_conf.rxmode.jumbo_frame)
1562 txconf->txq_flags = 0;
1563 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1566 rte_exit(EXIT_FAILURE,
1567 "rte_eth_tx_queue_setup: err=%d, "
1568 "port=%d\n", ret, portid);
1570 qconf = &lcore_conf[lcore_id];
1571 qconf->tx_queue_id[portid] = queueid;
1577 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1578 if (rte_lcore_is_enabled(lcore_id) == 0)
1581 /* init power management library */
1582 ret = rte_power_init(lcore_id);
1584 rte_exit(EXIT_FAILURE, "Power management library "
1585 "initialization failed on core%u\n", lcore_id);
1587 /* init timer structures for each enabled lcore */
1588 rte_timer_init(&power_timers[lcore_id]);
1589 hz = rte_get_timer_hz();
1590 rte_timer_reset(&power_timers[lcore_id],
1591 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1592 power_timer_cb, NULL);
1594 qconf = &lcore_conf[lcore_id];
1595 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1597 /* init RX queues */
1598 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1599 portid = qconf->rx_queue_list[queue].port_id;
1600 queueid = qconf->rx_queue_list[queue].queue_id;
1604 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1608 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1611 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1613 pktmbuf_pool[socketid]);
1615 rte_exit(EXIT_FAILURE,
1616 "rte_eth_rx_queue_setup: err=%d, "
1617 "port=%d\n", ret, portid);
1624 for (portid = 0; portid < nb_ports; portid++) {
1625 if ((enabled_port_mask & (1 << portid)) == 0) {
1629 ret = rte_eth_dev_start(portid);
1631 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1632 "port=%d\n", ret, portid);
1635 * If enabled, put device in promiscuous mode.
1636 * This allows IO forwarding mode to forward packets
1637 * to itself through 2 cross-connected ports of the
1641 rte_eth_promiscuous_enable(portid);
1644 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1646 /* launch per-lcore init on every lcore */
1647 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1648 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1649 if (rte_eal_wait_lcore(lcore_id) < 0)