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 #include <rte_spinlock.h>
80 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
82 #define MAX_PKT_BURST 32
84 #define MIN_ZERO_POLL_COUNT 10
86 /* around 100ms at 2 Ghz */
87 #define TIMER_RESOLUTION_CYCLES 200000000ULL
89 #define TIMER_NUMBER_PER_SECOND 10
91 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
92 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
94 #define APP_LOOKUP_EXACT_MATCH 0
95 #define APP_LOOKUP_LPM 1
96 #define DO_RFC_1812_CHECKS
98 #ifndef APP_LOOKUP_METHOD
99 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
102 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
103 #include <rte_hash.h>
104 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
107 #error "APP_LOOKUP_METHOD set to incorrect value"
111 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
112 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
113 #define IPv6_BYTES(addr) \
114 addr[0], addr[1], addr[2], addr[3], \
115 addr[4], addr[5], addr[6], addr[7], \
116 addr[8], addr[9], addr[10], addr[11],\
117 addr[12], addr[13],addr[14], addr[15]
120 #define MAX_JUMBO_PKT_LEN 9600
122 #define IPV6_ADDR_LEN 16
124 #define MEMPOOL_CACHE_SIZE 256
127 * This expression is used to calculate the number of mbufs needed depending on
128 * user input, taking into account memory for rx and tx hardware rings, cache
129 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
130 * NB_MBUF never goes below a minimum value of 8192.
133 #define NB_MBUF RTE_MAX ( \
134 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
135 nb_ports*nb_lcores*MAX_PKT_BURST + \
136 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
137 nb_lcores*MEMPOOL_CACHE_SIZE), \
140 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
144 /* Configure how many packets ahead to prefetch, when reading packets */
145 #define PREFETCH_OFFSET 3
148 * Configurable number of RX/TX ring descriptors
150 #define RTE_TEST_RX_DESC_DEFAULT 128
151 #define RTE_TEST_TX_DESC_DEFAULT 512
152 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
153 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
155 /* ethernet addresses of ports */
156 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
158 /* ethernet addresses of ports */
159 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
161 /* mask of enabled ports */
162 static uint32_t enabled_port_mask = 0;
163 /* Ports set in promiscuous mode off by default. */
164 static int promiscuous_on = 0;
165 /* NUMA is enabled by default. */
166 static int numa_on = 1;
168 enum freq_scale_hint_t
178 struct rte_mbuf *m_table[MAX_PKT_BURST];
181 struct lcore_rx_queue {
184 enum freq_scale_hint_t freq_up_hint;
185 uint32_t zero_rx_packet_count;
187 } __rte_cache_aligned;
189 #define MAX_RX_QUEUE_PER_LCORE 16
190 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
191 #define MAX_RX_QUEUE_PER_PORT 128
193 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
196 #define MAX_LCORE_PARAMS 1024
197 struct lcore_params {
201 } __rte_cache_aligned;
203 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
204 static struct lcore_params lcore_params_array_default[] = {
216 static struct lcore_params * lcore_params = lcore_params_array_default;
217 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
218 sizeof(lcore_params_array_default[0]);
220 static struct rte_eth_conf port_conf = {
222 .mq_mode = ETH_MQ_RX_RSS,
223 .max_rx_pkt_len = ETHER_MAX_LEN,
225 .header_split = 0, /**< Header Split disabled */
226 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
227 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
228 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
229 .hw_strip_crc = 0, /**< CRC stripped by hardware */
234 .rss_hf = ETH_RSS_UDP,
238 .mq_mode = ETH_MQ_TX_NONE,
246 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
249 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
251 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
252 #include <rte_hash_crc.h>
253 #define DEFAULT_HASH_FUNC rte_hash_crc
255 #include <rte_jhash.h>
256 #define DEFAULT_HASH_FUNC rte_jhash
265 } __attribute__((__packed__));
268 uint8_t ip_dst[IPV6_ADDR_LEN];
269 uint8_t ip_src[IPV6_ADDR_LEN];
273 } __attribute__((__packed__));
275 struct ipv4_l3fwd_route {
276 struct ipv4_5tuple key;
280 struct ipv6_l3fwd_route {
281 struct ipv6_5tuple key;
285 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
286 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
287 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
288 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
289 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
292 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
295 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
296 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
297 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
298 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
304 typedef struct rte_hash lookup_struct_t;
305 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
306 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
308 #define L3FWD_HASH_ENTRIES 1024
310 #define IPV4_L3FWD_NUM_ROUTES \
311 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
313 #define IPV6_L3FWD_NUM_ROUTES \
314 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
316 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
317 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
320 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
321 struct ipv4_l3fwd_route {
327 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
328 {IPv4(1,1,1,0), 24, 0},
329 {IPv4(2,1,1,0), 24, 1},
330 {IPv4(3,1,1,0), 24, 2},
331 {IPv4(4,1,1,0), 24, 3},
332 {IPv4(5,1,1,0), 24, 4},
333 {IPv4(6,1,1,0), 24, 5},
334 {IPv4(7,1,1,0), 24, 6},
335 {IPv4(8,1,1,0), 24, 7},
338 #define IPV4_L3FWD_NUM_ROUTES \
339 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
341 #define IPV4_L3FWD_LPM_MAX_RULES 1024
343 typedef struct rte_lpm lookup_struct_t;
344 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
349 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
350 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
351 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
352 lookup_struct_t * ipv4_lookup_struct;
353 lookup_struct_t * ipv6_lookup_struct;
354 } __rte_cache_aligned;
357 /* total sleep time in ms since last frequency scaling down */
359 /* number of long sleep recently */
360 uint32_t nb_long_sleep;
361 /* freq. scaling up trend */
363 /* total packet processed recently */
364 uint64_t nb_rx_processed;
365 /* total iterations looped recently */
366 uint64_t nb_iteration_looped;
368 } __rte_cache_aligned;
370 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
371 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
372 static struct rte_timer power_timers[RTE_MAX_LCORE];
374 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
375 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
376 unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
378 /* exit signal handler */
380 signal_exit_now(int sigtype)
385 if (sigtype == SIGINT) {
386 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
387 if (rte_lcore_is_enabled(lcore_id) == 0)
390 /* init power management library */
391 ret = rte_power_exit(lcore_id);
393 rte_exit(EXIT_FAILURE, "Power management "
394 "library de-initialization failed on "
395 "core%u\n", lcore_id);
399 rte_exit(EXIT_SUCCESS, "User forced exit\n");
402 /* Freqency scale down timer callback */
404 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
405 __attribute__((unused)) void *arg)
408 float sleep_time_ratio;
409 unsigned lcore_id = rte_lcore_id();
411 /* accumulate total execution time in us when callback is invoked */
412 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
413 (float)SCALING_PERIOD;
415 * check whether need to scale down frequency a step if it sleep a lot.
417 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
418 if (rte_power_freq_down)
419 rte_power_freq_down(lcore_id);
421 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
422 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
424 * scale down a step if average packet per iteration less
427 if (rte_power_freq_down)
428 rte_power_freq_down(lcore_id);
432 * initialize another timer according to current frequency to ensure
433 * timer interval is relatively fixed.
435 hz = rte_get_timer_hz();
436 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
437 SINGLE, lcore_id, power_timer_cb, NULL);
439 stats[lcore_id].nb_rx_processed = 0;
440 stats[lcore_id].nb_iteration_looped = 0;
442 stats[lcore_id].sleep_time = 0;
445 /* Send burst of packets on an output interface */
447 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
449 struct rte_mbuf **m_table;
453 queueid = qconf->tx_queue_id[port];
454 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
456 ret = rte_eth_tx_burst(port, queueid, m_table, n);
457 if (unlikely(ret < n)) {
459 rte_pktmbuf_free(m_table[ret]);
466 /* Enqueue a single packet, and send burst if queue is filled */
468 send_single_packet(struct rte_mbuf *m, uint8_t port)
472 struct lcore_conf *qconf;
474 lcore_id = rte_lcore_id();
476 qconf = &lcore_conf[lcore_id];
477 len = qconf->tx_mbufs[port].len;
478 qconf->tx_mbufs[port].m_table[len] = m;
481 /* enough pkts to be sent */
482 if (unlikely(len == MAX_PKT_BURST)) {
483 send_burst(qconf, MAX_PKT_BURST, port);
487 qconf->tx_mbufs[port].len = len;
491 #ifdef DO_RFC_1812_CHECKS
493 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
495 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
497 * 1. The packet length reported by the Link Layer must be large
498 * enough to hold the minimum length legal IP datagram (20 bytes).
500 if (link_len < sizeof(struct ipv4_hdr))
503 /* 2. The IP checksum must be correct. */
504 /* this is checked in H/W */
507 * 3. The IP version number must be 4. If the version number is not 4
508 * then the packet may be another version of IP, such as IPng or
511 if (((pkt->version_ihl) >> 4) != 4)
514 * 4. The IP header length field must be large enough to hold the
515 * minimum length legal IP datagram (20 bytes = 5 words).
517 if ((pkt->version_ihl & 0xf) < 5)
521 * 5. The IP total length field must be large enough to hold the IP
522 * datagram header, whose length is specified in the IP header length
525 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
532 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
534 print_ipv4_key(struct ipv4_5tuple key)
536 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
537 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
538 key.port_dst, key.port_src, key.proto);
541 print_ipv6_key(struct ipv6_5tuple key)
543 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
544 "port dst = %d, port src = %d, proto = %d\n",
545 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
546 key.port_dst, key.port_src, key.proto);
549 static inline uint8_t
550 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
551 lookup_struct_t * ipv4_l3fwd_lookup_struct)
553 struct ipv4_5tuple key;
558 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
559 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
560 key.proto = ipv4_hdr->next_proto_id;
562 switch (ipv4_hdr->next_proto_id) {
564 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
565 sizeof(struct ipv4_hdr));
566 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
567 key.port_src = rte_be_to_cpu_16(tcp->src_port);
571 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
572 sizeof(struct ipv4_hdr));
573 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
574 key.port_src = rte_be_to_cpu_16(udp->src_port);
583 /* Find destination port */
584 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
585 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
588 static inline uint8_t
589 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
590 lookup_struct_t *ipv6_l3fwd_lookup_struct)
592 struct ipv6_5tuple key;
597 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
598 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
600 key.proto = ipv6_hdr->proto;
602 switch (ipv6_hdr->proto) {
604 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
605 sizeof(struct ipv6_hdr));
606 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
607 key.port_src = rte_be_to_cpu_16(tcp->src_port);
611 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
612 sizeof(struct ipv6_hdr));
613 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
614 key.port_src = rte_be_to_cpu_16(udp->src_port);
623 /* Find destination port */
624 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
625 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
629 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
630 static inline uint8_t
631 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
632 lookup_struct_t *ipv4_l3fwd_lookup_struct)
636 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
637 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
643 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
644 struct lcore_conf *qconf)
646 struct ether_hdr *eth_hdr;
647 struct ipv4_hdr *ipv4_hdr;
651 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
654 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
656 if (m->ol_flags & PKT_RX_IPV4_HDR) {
658 /* Handle IPv4 headers.*/
660 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
661 sizeof(struct ether_hdr));
663 #ifdef DO_RFC_1812_CHECKS
664 /* Check to make sure the packet is valid (RFC1812) */
665 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
671 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
672 qconf->ipv4_lookup_struct);
673 if (dst_port >= RTE_MAX_ETHPORTS ||
674 (enabled_port_mask & 1 << dst_port) == 0)
677 /* 02:00:00:00:00:xx */
678 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
679 *((uint64_t *)d_addr_bytes) =
680 0x000000000002 + ((uint64_t)dst_port << 40);
682 #ifdef DO_RFC_1812_CHECKS
683 /* Update time to live and header checksum */
684 --(ipv4_hdr->time_to_live);
685 ++(ipv4_hdr->hdr_checksum);
689 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
691 send_single_packet(m, dst_port);
693 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
698 /* Handle IPv6 headers.*/
699 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
700 struct ipv6_hdr *ipv6_hdr;
703 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
704 sizeof(struct ether_hdr));
706 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
707 qconf->ipv6_lookup_struct);
709 if (dst_port >= RTE_MAX_ETHPORTS ||
710 (enabled_port_mask & 1 << dst_port) == 0)
713 /* 02:00:00:00:00:xx */
714 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
715 *((uint64_t *)d_addr_bytes) =
716 0x000000000002 + ((uint64_t)dst_port << 40);
719 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
721 send_single_packet(m, dst_port);
723 /* We don't currently handle IPv6 packets in LPM mode. */
730 #define MINIMUM_SLEEP_TIME 1
731 #define SUSPEND_THRESHOLD 300
733 static inline uint32_t
734 power_idle_heuristic(uint32_t zero_rx_packet_count)
736 /* If zero count is less than 100, sleep 1us */
737 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
738 return MINIMUM_SLEEP_TIME;
739 /* If zero count is less than 1000, sleep 100 us which is the
740 minimum latency switching from C3/C6 to C0
743 return SUSPEND_THRESHOLD;
748 static inline enum freq_scale_hint_t
749 power_freq_scaleup_heuristic(unsigned lcore_id,
754 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
757 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
758 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
759 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
760 #define FREQ_UP_TREND1_ACC 1
761 #define FREQ_UP_TREND2_ACC 100
762 #define FREQ_UP_THRESHOLD 10000
764 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
765 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
766 stats[lcore_id].trend = 0;
768 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
769 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
770 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
771 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
772 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
773 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
775 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
776 stats[lcore_id].trend = 0;
784 * force polling thread sleep until one-shot rx interrupt triggers
793 sleep_until_rx_interrupt(int num)
795 struct rte_epoll_event event[num];
797 uint8_t port_id, queue_id;
800 RTE_LOG(INFO, L3FWD_POWER,
801 "lcore %u sleeps until interrupt triggers\n",
804 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
805 for (i = 0; i < n; i++) {
806 data = event[i].epdata.data;
807 port_id = ((uintptr_t)data) >> CHAR_BIT;
808 queue_id = ((uintptr_t)data) &
809 RTE_LEN2MASK(CHAR_BIT, uint8_t);
810 RTE_LOG(INFO, L3FWD_POWER,
811 "lcore %u is waked up from rx interrupt on"
812 " port %d queue %d\n",
813 rte_lcore_id(), port_id, queue_id);
819 static void turn_on_intr(struct lcore_conf *qconf)
822 struct lcore_rx_queue *rx_queue;
823 uint8_t port_id, queue_id;
825 for (i = 0; i < qconf->n_rx_queue; ++i) {
826 rx_queue = &(qconf->rx_queue_list[i]);
827 port_id = rx_queue->port_id;
828 queue_id = rx_queue->queue_id;
830 rte_spinlock_lock(&(locks[port_id]));
831 rte_eth_dev_rx_intr_enable(port_id, queue_id);
832 rte_spinlock_unlock(&(locks[port_id]));
836 static int event_register(struct lcore_conf *qconf)
838 struct lcore_rx_queue *rx_queue;
839 uint8_t portid, queueid;
844 for (i = 0; i < qconf->n_rx_queue; ++i) {
845 rx_queue = &(qconf->rx_queue_list[i]);
846 portid = rx_queue->port_id;
847 queueid = rx_queue->queue_id;
848 data = portid << CHAR_BIT | queueid;
850 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
851 RTE_EPOLL_PER_THREAD,
853 (void *)((uintptr_t)data));
861 /* main processing loop */
863 main_loop(__attribute__((unused)) void *dummy)
865 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
867 uint64_t prev_tsc, diff_tsc, cur_tsc;
868 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
870 uint8_t portid, queueid;
871 struct lcore_conf *qconf;
872 struct lcore_rx_queue *rx_queue;
873 enum freq_scale_hint_t lcore_scaleup_hint;
874 uint32_t lcore_rx_idle_count = 0;
875 uint32_t lcore_idle_hint = 0;
878 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
882 lcore_id = rte_lcore_id();
883 qconf = &lcore_conf[lcore_id];
885 if (qconf->n_rx_queue == 0) {
886 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
890 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
892 for (i = 0; i < qconf->n_rx_queue; i++) {
893 portid = qconf->rx_queue_list[i].port_id;
894 queueid = qconf->rx_queue_list[i].queue_id;
895 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
896 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
899 /* add into event wait list */
900 if (event_register(qconf) == 0)
903 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
906 stats[lcore_id].nb_iteration_looped++;
908 cur_tsc = rte_rdtsc();
909 cur_tsc_power = cur_tsc;
912 * TX burst queue drain
914 diff_tsc = cur_tsc - prev_tsc;
915 if (unlikely(diff_tsc > drain_tsc)) {
918 * This could be optimized (use queueid instead of
919 * portid), but it is not called so often
921 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
922 if (qconf->tx_mbufs[portid].len == 0)
924 send_burst(&lcore_conf[lcore_id],
925 qconf->tx_mbufs[portid].len,
927 qconf->tx_mbufs[portid].len = 0;
933 diff_tsc_power = cur_tsc_power - prev_tsc_power;
934 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
936 prev_tsc_power = cur_tsc_power;
941 * Read packet from RX queues
943 lcore_scaleup_hint = FREQ_CURRENT;
944 lcore_rx_idle_count = 0;
945 for (i = 0; i < qconf->n_rx_queue; ++i) {
946 rx_queue = &(qconf->rx_queue_list[i]);
947 rx_queue->idle_hint = 0;
948 portid = rx_queue->port_id;
949 queueid = rx_queue->queue_id;
951 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
954 stats[lcore_id].nb_rx_processed += nb_rx;
955 if (unlikely(nb_rx == 0)) {
957 * no packet received from rx queue, try to
958 * sleep for a while forcing CPU enter deeper
961 rx_queue->zero_rx_packet_count++;
963 if (rx_queue->zero_rx_packet_count <=
967 rx_queue->idle_hint = power_idle_heuristic(\
968 rx_queue->zero_rx_packet_count);
969 lcore_rx_idle_count++;
971 rx_queue->zero_rx_packet_count = 0;
974 * do not scale up frequency immediately as
975 * user to kernel space communication is costly
976 * which might impact packet I/O for received
979 rx_queue->freq_up_hint =
980 power_freq_scaleup_heuristic(lcore_id,
984 /* Prefetch first packets */
985 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
986 rte_prefetch0(rte_pktmbuf_mtod(
987 pkts_burst[j], void *));
990 /* Prefetch and forward already prefetched packets */
991 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
992 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
993 j + PREFETCH_OFFSET], void *));
994 l3fwd_simple_forward(pkts_burst[j], portid,
998 /* Forward remaining prefetched packets */
999 for (; j < nb_rx; j++) {
1000 l3fwd_simple_forward(pkts_burst[j], portid,
1005 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1006 for (i = 1, lcore_scaleup_hint =
1007 qconf->rx_queue_list[0].freq_up_hint;
1008 i < qconf->n_rx_queue; ++i) {
1009 rx_queue = &(qconf->rx_queue_list[i]);
1010 if (rx_queue->freq_up_hint >
1012 lcore_scaleup_hint =
1013 rx_queue->freq_up_hint;
1016 if (lcore_scaleup_hint == FREQ_HIGHEST) {
1017 if (rte_power_freq_max)
1018 rte_power_freq_max(lcore_id);
1019 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1020 if (rte_power_freq_up)
1021 rte_power_freq_up(lcore_id);
1025 * All Rx queues empty in recent consecutive polls,
1026 * sleep in a conservative manner, meaning sleep as
1029 for (i = 1, lcore_idle_hint =
1030 qconf->rx_queue_list[0].idle_hint;
1031 i < qconf->n_rx_queue; ++i) {
1032 rx_queue = &(qconf->rx_queue_list[i]);
1033 if (rx_queue->idle_hint < lcore_idle_hint)
1034 lcore_idle_hint = rx_queue->idle_hint;
1037 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1039 * execute "pause" instruction to avoid context
1040 * switch which generally take hundred of
1041 * microseconds for short sleep.
1043 rte_delay_us(lcore_idle_hint);
1045 /* suspend until rx interrupt trigges */
1047 turn_on_intr(qconf);
1048 sleep_until_rx_interrupt(
1051 /* start receiving packets immediately */
1054 stats[lcore_id].sleep_time += lcore_idle_hint;
1060 check_lcore_params(void)
1062 uint8_t queue, lcore;
1066 for (i = 0; i < nb_lcore_params; ++i) {
1067 queue = lcore_params[i].queue_id;
1068 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1069 printf("invalid queue number: %hhu\n", queue);
1072 lcore = lcore_params[i].lcore_id;
1073 if (!rte_lcore_is_enabled(lcore)) {
1074 printf("error: lcore %hhu is not enabled in lcore "
1078 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1080 printf("warning: lcore %hhu is on socket %d with numa "
1081 "off\n", lcore, socketid);
1088 check_port_config(const unsigned nb_ports)
1093 for (i = 0; i < nb_lcore_params; ++i) {
1094 portid = lcore_params[i].port_id;
1095 if ((enabled_port_mask & (1 << portid)) == 0) {
1096 printf("port %u is not enabled in port mask\n",
1100 if (portid >= nb_ports) {
1101 printf("port %u is not present on the board\n",
1110 get_port_n_rx_queues(const uint8_t port)
1115 for (i = 0; i < nb_lcore_params; ++i) {
1116 if (lcore_params[i].port_id == port &&
1117 lcore_params[i].queue_id > queue)
1118 queue = lcore_params[i].queue_id;
1120 return (uint8_t)(++queue);
1124 init_lcore_rx_queues(void)
1126 uint16_t i, nb_rx_queue;
1129 for (i = 0; i < nb_lcore_params; ++i) {
1130 lcore = lcore_params[i].lcore_id;
1131 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1132 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1133 printf("error: too many queues (%u) for lcore: %u\n",
1134 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1137 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1138 lcore_params[i].port_id;
1139 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1140 lcore_params[i].queue_id;
1141 lcore_conf[lcore].n_rx_queue++;
1149 print_usage(const char *prgname)
1151 printf ("%s [EAL options] -- -p PORTMASK -P"
1152 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1153 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1154 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1155 " -P : enable promiscuous mode\n"
1156 " --config (port,queue,lcore): rx queues configuration\n"
1157 " --no-numa: optional, disable numa awareness\n"
1158 " --enable-jumbo: enable jumbo frame"
1159 " which max packet len is PKTLEN in decimal (64-9600)\n",
1163 static int parse_max_pkt_len(const char *pktlen)
1168 /* parse decimal string */
1169 len = strtoul(pktlen, &end, 10);
1170 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1180 parse_portmask(const char *portmask)
1185 /* parse hexadecimal string */
1186 pm = strtoul(portmask, &end, 16);
1187 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1197 parse_config(const char *q_arg)
1200 const char *p, *p0 = q_arg;
1208 unsigned long int_fld[_NUM_FLD];
1209 char *str_fld[_NUM_FLD];
1213 nb_lcore_params = 0;
1215 while ((p = strchr(p0,'(')) != NULL) {
1217 if((p0 = strchr(p,')')) == NULL)
1221 if(size >= sizeof(s))
1224 snprintf(s, sizeof(s), "%.*s", size, p);
1225 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1228 for (i = 0; i < _NUM_FLD; i++){
1230 int_fld[i] = strtoul(str_fld[i], &end, 0);
1231 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1235 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1236 printf("exceeded max number of lcore params: %hu\n",
1240 lcore_params_array[nb_lcore_params].port_id =
1241 (uint8_t)int_fld[FLD_PORT];
1242 lcore_params_array[nb_lcore_params].queue_id =
1243 (uint8_t)int_fld[FLD_QUEUE];
1244 lcore_params_array[nb_lcore_params].lcore_id =
1245 (uint8_t)int_fld[FLD_LCORE];
1248 lcore_params = lcore_params_array;
1253 /* Parse the argument given in the command line of the application */
1255 parse_args(int argc, char **argv)
1260 char *prgname = argv[0];
1261 static struct option lgopts[] = {
1262 {"config", 1, 0, 0},
1263 {"no-numa", 0, 0, 0},
1264 {"enable-jumbo", 0, 0, 0},
1270 while ((opt = getopt_long(argc, argvopt, "p:P",
1271 lgopts, &option_index)) != EOF) {
1276 enabled_port_mask = parse_portmask(optarg);
1277 if (enabled_port_mask == 0) {
1278 printf("invalid portmask\n");
1279 print_usage(prgname);
1284 printf("Promiscuous mode selected\n");
1290 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1291 ret = parse_config(optarg);
1293 printf("invalid config\n");
1294 print_usage(prgname);
1299 if (!strncmp(lgopts[option_index].name,
1301 printf("numa is disabled \n");
1305 if (!strncmp(lgopts[option_index].name,
1306 "enable-jumbo", 12)) {
1307 struct option lenopts =
1308 {"max-pkt-len", required_argument, \
1311 printf("jumbo frame is enabled \n");
1312 port_conf.rxmode.jumbo_frame = 1;
1315 * if no max-pkt-len set, use the default value
1318 if (0 == getopt_long(argc, argvopt, "",
1319 &lenopts, &option_index)) {
1320 ret = parse_max_pkt_len(optarg);
1322 (ret > MAX_JUMBO_PKT_LEN)){
1323 printf("invalid packet "
1325 print_usage(prgname);
1328 port_conf.rxmode.max_rx_pkt_len = ret;
1330 printf("set jumbo frame "
1331 "max packet length to %u\n",
1332 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1338 print_usage(prgname);
1344 argv[optind-1] = prgname;
1347 optind = 0; /* reset getopt lib */
1352 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1354 char buf[ETHER_ADDR_FMT_SIZE];
1355 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1356 printf("%s%s", name, buf);
1359 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1361 setup_hash(int socketid)
1363 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1365 .entries = L3FWD_HASH_ENTRIES,
1366 .key_len = sizeof(struct ipv4_5tuple),
1367 .hash_func = DEFAULT_HASH_FUNC,
1368 .hash_func_init_val = 0,
1371 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1373 .entries = L3FWD_HASH_ENTRIES,
1374 .key_len = sizeof(struct ipv6_5tuple),
1375 .hash_func = DEFAULT_HASH_FUNC,
1376 .hash_func_init_val = 0,
1383 /* create ipv4 hash */
1384 rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1385 ipv4_l3fwd_hash_params.name = s;
1386 ipv4_l3fwd_hash_params.socket_id = socketid;
1387 ipv4_l3fwd_lookup_struct[socketid] =
1388 rte_hash_create(&ipv4_l3fwd_hash_params);
1389 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1390 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1391 "socket %d\n", socketid);
1393 /* create ipv6 hash */
1394 rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1395 ipv6_l3fwd_hash_params.name = s;
1396 ipv6_l3fwd_hash_params.socket_id = socketid;
1397 ipv6_l3fwd_lookup_struct[socketid] =
1398 rte_hash_create(&ipv6_l3fwd_hash_params);
1399 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1400 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1401 "socket %d\n", socketid);
1404 /* populate the ipv4 hash */
1405 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1406 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1407 (void *) &ipv4_l3fwd_route_array[i].key);
1409 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1410 "l3fwd hash on socket %d\n", i, socketid);
1412 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1413 printf("Hash: Adding key\n");
1414 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1417 /* populate the ipv6 hash */
1418 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1419 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1420 (void *) &ipv6_l3fwd_route_array[i].key);
1422 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1423 "l3fwd hash on socket %d\n", i, socketid);
1425 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1426 printf("Hash: Adding key\n");
1427 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1432 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1434 setup_lpm(int socketid)
1440 /* create the LPM table */
1441 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1442 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1443 IPV4_L3FWD_LPM_MAX_RULES, 0);
1444 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1445 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1446 " on socket %d\n", socketid);
1448 /* populate the LPM table */
1449 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1450 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1451 ipv4_l3fwd_route_array[i].ip,
1452 ipv4_l3fwd_route_array[i].depth,
1453 ipv4_l3fwd_route_array[i].if_out);
1456 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1457 "l3fwd LPM table on socket %d\n",
1461 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1462 (unsigned)ipv4_l3fwd_route_array[i].ip,
1463 ipv4_l3fwd_route_array[i].depth,
1464 ipv4_l3fwd_route_array[i].if_out);
1470 init_mem(unsigned nb_mbuf)
1472 struct lcore_conf *qconf;
1477 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1478 if (rte_lcore_is_enabled(lcore_id) == 0)
1482 socketid = rte_lcore_to_socket_id(lcore_id);
1486 if (socketid >= NB_SOCKETS) {
1487 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1488 "out of range %d\n", socketid,
1489 lcore_id, NB_SOCKETS);
1491 if (pktmbuf_pool[socketid] == NULL) {
1492 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1493 pktmbuf_pool[socketid] =
1494 rte_pktmbuf_pool_create(s, nb_mbuf,
1495 MEMPOOL_CACHE_SIZE, 0,
1496 RTE_MBUF_DEFAULT_BUF_SIZE,
1498 if (pktmbuf_pool[socketid] == NULL)
1499 rte_exit(EXIT_FAILURE,
1500 "Cannot init mbuf pool on socket %d\n",
1503 printf("Allocated mbuf pool on socket %d\n",
1506 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1507 setup_lpm(socketid);
1509 setup_hash(socketid);
1512 qconf = &lcore_conf[lcore_id];
1513 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1514 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1515 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1521 /* Check the link status of all ports in up to 9s, and print them finally */
1523 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1525 #define CHECK_INTERVAL 100 /* 100ms */
1526 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1527 uint8_t portid, count, all_ports_up, print_flag = 0;
1528 struct rte_eth_link link;
1530 printf("\nChecking link status");
1532 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1534 for (portid = 0; portid < port_num; portid++) {
1535 if ((port_mask & (1 << portid)) == 0)
1537 memset(&link, 0, sizeof(link));
1538 rte_eth_link_get_nowait(portid, &link);
1539 /* print link status if flag set */
1540 if (print_flag == 1) {
1541 if (link.link_status)
1542 printf("Port %d Link Up - speed %u "
1543 "Mbps - %s\n", (uint8_t)portid,
1544 (unsigned)link.link_speed,
1545 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1546 ("full-duplex") : ("half-duplex\n"));
1548 printf("Port %d Link Down\n",
1552 /* clear all_ports_up flag if any link down */
1553 if (link.link_status == 0) {
1558 /* after finally printing all link status, get out */
1559 if (print_flag == 1)
1562 if (all_ports_up == 0) {
1565 rte_delay_ms(CHECK_INTERVAL);
1568 /* set the print_flag if all ports up or timeout */
1569 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1577 main(int argc, char **argv)
1579 struct lcore_conf *qconf;
1580 struct rte_eth_dev_info dev_info;
1581 struct rte_eth_txconf *txconf;
1587 uint32_t n_tx_queue, nb_lcores;
1588 uint32_t dev_rxq_num, dev_txq_num;
1589 uint8_t portid, nb_rx_queue, queue, socketid;
1591 /* catch SIGINT and restore cpufreq governor to ondemand */
1592 signal(SIGINT, signal_exit_now);
1595 ret = rte_eal_init(argc, argv);
1597 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1601 /* init RTE timer library to be used late */
1602 rte_timer_subsystem_init();
1604 /* parse application arguments (after the EAL ones) */
1605 ret = parse_args(argc, argv);
1607 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1609 if (check_lcore_params() < 0)
1610 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1612 ret = init_lcore_rx_queues();
1614 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1617 nb_ports = rte_eth_dev_count();
1618 if (nb_ports > RTE_MAX_ETHPORTS)
1619 nb_ports = RTE_MAX_ETHPORTS;
1621 if (check_port_config(nb_ports) < 0)
1622 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1624 nb_lcores = rte_lcore_count();
1626 /* initialize all ports */
1627 for (portid = 0; portid < nb_ports; portid++) {
1628 /* skip ports that are not enabled */
1629 if ((enabled_port_mask & (1 << portid)) == 0) {
1630 printf("\nSkipping disabled port %d\n", portid);
1635 printf("Initializing port %d ... ", portid );
1638 rte_eth_dev_info_get(portid, &dev_info);
1639 dev_rxq_num = dev_info.max_rx_queues;
1640 dev_txq_num = dev_info.max_tx_queues;
1642 nb_rx_queue = get_port_n_rx_queues(portid);
1643 if (nb_rx_queue > dev_rxq_num)
1644 rte_exit(EXIT_FAILURE,
1645 "Cannot configure not existed rxq: "
1646 "port=%d\n", portid);
1648 n_tx_queue = nb_lcores;
1649 if (n_tx_queue > dev_txq_num)
1650 n_tx_queue = dev_txq_num;
1651 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1652 nb_rx_queue, (unsigned)n_tx_queue );
1653 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1654 (uint16_t)n_tx_queue, &port_conf);
1656 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1657 "err=%d, port=%d\n", ret, portid);
1659 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1660 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1664 ret = init_mem(NB_MBUF);
1666 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1668 /* init one TX queue per couple (lcore,port) */
1670 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1671 if (rte_lcore_is_enabled(lcore_id) == 0)
1674 if (queueid >= dev_txq_num)
1679 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1683 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1686 rte_eth_dev_info_get(portid, &dev_info);
1687 txconf = &dev_info.default_txconf;
1688 if (port_conf.rxmode.jumbo_frame)
1689 txconf->txq_flags = 0;
1690 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1693 rte_exit(EXIT_FAILURE,
1694 "rte_eth_tx_queue_setup: err=%d, "
1695 "port=%d\n", ret, portid);
1697 qconf = &lcore_conf[lcore_id];
1698 qconf->tx_queue_id[portid] = queueid;
1704 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1705 if (rte_lcore_is_enabled(lcore_id) == 0)
1708 /* init power management library */
1709 ret = rte_power_init(lcore_id);
1712 "Library initialization failed on core %u\n", lcore_id);
1714 /* init timer structures for each enabled lcore */
1715 rte_timer_init(&power_timers[lcore_id]);
1716 hz = rte_get_timer_hz();
1717 rte_timer_reset(&power_timers[lcore_id],
1718 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1719 power_timer_cb, NULL);
1721 qconf = &lcore_conf[lcore_id];
1722 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1724 /* init RX queues */
1725 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1726 portid = qconf->rx_queue_list[queue].port_id;
1727 queueid = qconf->rx_queue_list[queue].queue_id;
1731 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1735 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1738 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1740 pktmbuf_pool[socketid]);
1742 rte_exit(EXIT_FAILURE,
1743 "rte_eth_rx_queue_setup: err=%d, "
1744 "port=%d\n", ret, portid);
1751 for (portid = 0; portid < nb_ports; portid++) {
1752 if ((enabled_port_mask & (1 << portid)) == 0) {
1756 ret = rte_eth_dev_start(portid);
1758 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1759 "port=%d\n", ret, portid);
1761 * If enabled, put device in promiscuous mode.
1762 * This allows IO forwarding mode to forward packets
1763 * to itself through 2 cross-connected ports of the
1767 rte_eth_promiscuous_enable(portid);
1768 /* initialize spinlock for each port */
1769 rte_spinlock_init(&(locks[portid]));
1772 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1774 /* launch per-lcore init on every lcore */
1775 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1776 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1777 if (rte_eal_wait_lcore(lcore_id) < 0)