1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/queue.h>
18 #include <rte_common.h>
19 #include <rte_byteorder.h>
21 #include <rte_malloc.h>
22 #include <rte_memory.h>
23 #include <rte_memcpy.h>
25 #include <rte_launch.h>
26 #include <rte_atomic.h>
27 #include <rte_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
37 #include <rte_mempool.h>
42 #include <rte_string_fns.h>
43 #include <rte_timer.h>
44 #include <rte_power.h>
45 #include <rte_spinlock.h>
47 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
49 #define MAX_PKT_BURST 32
51 #define MIN_ZERO_POLL_COUNT 10
54 #define TIMER_NUMBER_PER_SECOND 10
56 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
57 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
59 #define APP_LOOKUP_EXACT_MATCH 0
60 #define APP_LOOKUP_LPM 1
61 #define DO_RFC_1812_CHECKS
63 #ifndef APP_LOOKUP_METHOD
64 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
67 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
69 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
72 #error "APP_LOOKUP_METHOD set to incorrect value"
76 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
77 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
78 #define IPv6_BYTES(addr) \
79 addr[0], addr[1], addr[2], addr[3], \
80 addr[4], addr[5], addr[6], addr[7], \
81 addr[8], addr[9], addr[10], addr[11],\
82 addr[12], addr[13],addr[14], addr[15]
85 #define MAX_JUMBO_PKT_LEN 9600
87 #define IPV6_ADDR_LEN 16
89 #define MEMPOOL_CACHE_SIZE 256
92 * This expression is used to calculate the number of mbufs needed depending on
93 * user input, taking into account memory for rx and tx hardware rings, cache
94 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
95 * NB_MBUF never goes below a minimum value of 8192.
98 #define NB_MBUF RTE_MAX ( \
99 (nb_ports*nb_rx_queue*nb_rxd + \
100 nb_ports*nb_lcores*MAX_PKT_BURST + \
101 nb_ports*n_tx_queue*nb_txd + \
102 nb_lcores*MEMPOOL_CACHE_SIZE), \
105 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
109 /* Configure how many packets ahead to prefetch, when reading packets */
110 #define PREFETCH_OFFSET 3
113 * Configurable number of RX/TX ring descriptors
115 #define RTE_TEST_RX_DESC_DEFAULT 1024
116 #define RTE_TEST_TX_DESC_DEFAULT 1024
117 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
118 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
120 /* ethernet addresses of ports */
121 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
123 /* ethernet addresses of ports */
124 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
126 /* mask of enabled ports */
127 static uint32_t enabled_port_mask = 0;
128 /* Ports set in promiscuous mode off by default. */
129 static int promiscuous_on = 0;
130 /* NUMA is enabled by default. */
131 static int numa_on = 1;
132 static int parse_ptype; /**< Parse packet type using rx callback, and */
133 /**< disabled by default */
135 enum freq_scale_hint_t
143 struct lcore_rx_queue {
146 enum freq_scale_hint_t freq_up_hint;
147 uint32_t zero_rx_packet_count;
149 } __rte_cache_aligned;
151 #define MAX_RX_QUEUE_PER_LCORE 16
152 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
153 #define MAX_RX_QUEUE_PER_PORT 128
155 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
158 #define MAX_LCORE_PARAMS 1024
159 struct lcore_params {
163 } __rte_cache_aligned;
165 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
166 static struct lcore_params lcore_params_array_default[] = {
178 static struct lcore_params * lcore_params = lcore_params_array_default;
179 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
180 sizeof(lcore_params_array_default[0]);
182 static struct rte_eth_conf port_conf = {
184 .mq_mode = ETH_MQ_RX_RSS,
185 .max_rx_pkt_len = ETHER_MAX_LEN,
187 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
188 DEV_RX_OFFLOAD_CHECKSUM),
193 .rss_hf = ETH_RSS_UDP,
197 .mq_mode = ETH_MQ_TX_NONE,
204 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
207 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
210 #include <rte_hash_crc.h>
211 #define DEFAULT_HASH_FUNC rte_hash_crc
213 #include <rte_jhash.h>
214 #define DEFAULT_HASH_FUNC rte_jhash
223 } __attribute__((__packed__));
226 uint8_t ip_dst[IPV6_ADDR_LEN];
227 uint8_t ip_src[IPV6_ADDR_LEN];
231 } __attribute__((__packed__));
233 struct ipv4_l3fwd_route {
234 struct ipv4_5tuple key;
238 struct ipv6_l3fwd_route {
239 struct ipv6_5tuple key;
243 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
244 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
245 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
246 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
247 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
250 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
253 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
254 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
255 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
256 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
262 typedef struct rte_hash lookup_struct_t;
263 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
264 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
266 #define L3FWD_HASH_ENTRIES 1024
268 #define IPV4_L3FWD_NUM_ROUTES \
269 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
271 #define IPV6_L3FWD_NUM_ROUTES \
272 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
274 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
275 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
278 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
279 struct ipv4_l3fwd_route {
285 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
286 {IPv4(1,1,1,0), 24, 0},
287 {IPv4(2,1,1,0), 24, 1},
288 {IPv4(3,1,1,0), 24, 2},
289 {IPv4(4,1,1,0), 24, 3},
290 {IPv4(5,1,1,0), 24, 4},
291 {IPv4(6,1,1,0), 24, 5},
292 {IPv4(7,1,1,0), 24, 6},
293 {IPv4(8,1,1,0), 24, 7},
296 #define IPV4_L3FWD_NUM_ROUTES \
297 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
299 #define IPV4_L3FWD_LPM_MAX_RULES 1024
301 typedef struct rte_lpm lookup_struct_t;
302 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
307 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
309 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
310 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
311 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
312 lookup_struct_t * ipv4_lookup_struct;
313 lookup_struct_t * ipv6_lookup_struct;
314 } __rte_cache_aligned;
317 /* total sleep time in ms since last frequency scaling down */
319 /* number of long sleep recently */
320 uint32_t nb_long_sleep;
321 /* freq. scaling up trend */
323 /* total packet processed recently */
324 uint64_t nb_rx_processed;
325 /* total iterations looped recently */
326 uint64_t nb_iteration_looped;
328 } __rte_cache_aligned;
330 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
331 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
332 static struct rte_timer power_timers[RTE_MAX_LCORE];
334 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
335 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
336 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
338 /* exit signal handler */
340 signal_exit_now(int sigtype)
346 if (sigtype == SIGINT) {
347 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
348 if (rte_lcore_is_enabled(lcore_id) == 0)
351 /* init power management library */
352 ret = rte_power_exit(lcore_id);
354 rte_exit(EXIT_FAILURE, "Power management "
355 "library de-initialization failed on "
356 "core%u\n", lcore_id);
359 RTE_ETH_FOREACH_DEV(portid) {
360 if ((enabled_port_mask & (1 << portid)) == 0)
363 rte_eth_dev_stop(portid);
364 rte_eth_dev_close(portid);
368 rte_exit(EXIT_SUCCESS, "User forced exit\n");
371 /* Freqency scale down timer callback */
373 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
374 __attribute__((unused)) void *arg)
377 float sleep_time_ratio;
378 unsigned lcore_id = rte_lcore_id();
380 /* accumulate total execution time in us when callback is invoked */
381 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
382 (float)SCALING_PERIOD;
384 * check whether need to scale down frequency a step if it sleep a lot.
386 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
387 if (rte_power_freq_down)
388 rte_power_freq_down(lcore_id);
390 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
391 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
393 * scale down a step if average packet per iteration less
396 if (rte_power_freq_down)
397 rte_power_freq_down(lcore_id);
401 * initialize another timer according to current frequency to ensure
402 * timer interval is relatively fixed.
404 hz = rte_get_timer_hz();
405 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
406 SINGLE, lcore_id, power_timer_cb, NULL);
408 stats[lcore_id].nb_rx_processed = 0;
409 stats[lcore_id].nb_iteration_looped = 0;
411 stats[lcore_id].sleep_time = 0;
414 /* Enqueue a single packet, and send burst if queue is filled */
416 send_single_packet(struct rte_mbuf *m, uint16_t port)
419 struct lcore_conf *qconf;
421 lcore_id = rte_lcore_id();
422 qconf = &lcore_conf[lcore_id];
424 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
425 qconf->tx_buffer[port], m);
430 #ifdef DO_RFC_1812_CHECKS
432 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
434 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
436 * 1. The packet length reported by the Link Layer must be large
437 * enough to hold the minimum length legal IP datagram (20 bytes).
439 if (link_len < sizeof(struct ipv4_hdr))
442 /* 2. The IP checksum must be correct. */
443 /* this is checked in H/W */
446 * 3. The IP version number must be 4. If the version number is not 4
447 * then the packet may be another version of IP, such as IPng or
450 if (((pkt->version_ihl) >> 4) != 4)
453 * 4. The IP header length field must be large enough to hold the
454 * minimum length legal IP datagram (20 bytes = 5 words).
456 if ((pkt->version_ihl & 0xf) < 5)
460 * 5. The IP total length field must be large enough to hold the IP
461 * datagram header, whose length is specified in the IP header length
464 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
471 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
473 print_ipv4_key(struct ipv4_5tuple key)
475 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
476 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
477 key.port_dst, key.port_src, key.proto);
480 print_ipv6_key(struct ipv6_5tuple key)
482 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
483 "port dst = %d, port src = %d, proto = %d\n",
484 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
485 key.port_dst, key.port_src, key.proto);
488 static inline uint16_t
489 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
490 lookup_struct_t * ipv4_l3fwd_lookup_struct)
492 struct ipv4_5tuple key;
497 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
498 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
499 key.proto = ipv4_hdr->next_proto_id;
501 switch (ipv4_hdr->next_proto_id) {
503 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
504 sizeof(struct ipv4_hdr));
505 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
506 key.port_src = rte_be_to_cpu_16(tcp->src_port);
510 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
511 sizeof(struct ipv4_hdr));
512 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
513 key.port_src = rte_be_to_cpu_16(udp->src_port);
522 /* Find destination port */
523 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
524 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
527 static inline uint16_t
528 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
529 lookup_struct_t *ipv6_l3fwd_lookup_struct)
531 struct ipv6_5tuple key;
536 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
537 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
539 key.proto = ipv6_hdr->proto;
541 switch (ipv6_hdr->proto) {
543 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
544 sizeof(struct ipv6_hdr));
545 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
546 key.port_src = rte_be_to_cpu_16(tcp->src_port);
550 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
551 sizeof(struct ipv6_hdr));
552 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
553 key.port_src = rte_be_to_cpu_16(udp->src_port);
562 /* Find destination port */
563 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
564 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
568 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
569 static inline uint16_t
570 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
571 lookup_struct_t *ipv4_l3fwd_lookup_struct)
575 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
576 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
582 parse_ptype_one(struct rte_mbuf *m)
584 struct ether_hdr *eth_hdr;
585 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
588 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
589 ether_type = eth_hdr->ether_type;
590 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
591 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
592 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
593 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
595 m->packet_type = packet_type;
599 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
600 struct rte_mbuf *pkts[], uint16_t nb_pkts,
601 uint16_t max_pkts __rte_unused,
602 void *user_param __rte_unused)
606 for (i = 0; i < nb_pkts; ++i)
607 parse_ptype_one(pkts[i]);
613 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
615 printf("Port %d: softly parse packet type info\n", portid);
616 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
619 printf("Failed to add rx callback: port=%d\n", portid);
624 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
625 struct lcore_conf *qconf)
627 struct ether_hdr *eth_hdr;
628 struct ipv4_hdr *ipv4_hdr;
632 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
634 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
635 /* Handle IPv4 headers.*/
637 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
638 sizeof(struct ether_hdr));
640 #ifdef DO_RFC_1812_CHECKS
641 /* Check to make sure the packet is valid (RFC1812) */
642 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
648 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
649 qconf->ipv4_lookup_struct);
650 if (dst_port >= RTE_MAX_ETHPORTS ||
651 (enabled_port_mask & 1 << dst_port) == 0)
654 /* 02:00:00:00:00:xx */
655 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
656 *((uint64_t *)d_addr_bytes) =
657 0x000000000002 + ((uint64_t)dst_port << 40);
659 #ifdef DO_RFC_1812_CHECKS
660 /* Update time to live and header checksum */
661 --(ipv4_hdr->time_to_live);
662 ++(ipv4_hdr->hdr_checksum);
666 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
668 send_single_packet(m, dst_port);
669 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
670 /* Handle IPv6 headers.*/
671 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
672 struct ipv6_hdr *ipv6_hdr;
675 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
676 sizeof(struct ether_hdr));
678 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
679 qconf->ipv6_lookup_struct);
681 if (dst_port >= RTE_MAX_ETHPORTS ||
682 (enabled_port_mask & 1 << dst_port) == 0)
685 /* 02:00:00:00:00:xx */
686 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
687 *((uint64_t *)d_addr_bytes) =
688 0x000000000002 + ((uint64_t)dst_port << 40);
691 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
693 send_single_packet(m, dst_port);
695 /* We don't currently handle IPv6 packets in LPM mode. */
703 #define MINIMUM_SLEEP_TIME 1
704 #define SUSPEND_THRESHOLD 300
706 static inline uint32_t
707 power_idle_heuristic(uint32_t zero_rx_packet_count)
709 /* If zero count is less than 100, sleep 1us */
710 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
711 return MINIMUM_SLEEP_TIME;
712 /* If zero count is less than 1000, sleep 100 us which is the
713 minimum latency switching from C3/C6 to C0
716 return SUSPEND_THRESHOLD;
719 static inline enum freq_scale_hint_t
720 power_freq_scaleup_heuristic(unsigned lcore_id,
724 uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
726 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
729 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
730 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
731 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
732 #define FREQ_UP_TREND1_ACC 1
733 #define FREQ_UP_TREND2_ACC 100
734 #define FREQ_UP_THRESHOLD 10000
736 if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
737 stats[lcore_id].trend = 0;
739 } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
740 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
741 else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
742 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
744 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
745 stats[lcore_id].trend = 0;
753 * force polling thread sleep until one-shot rx interrupt triggers
762 sleep_until_rx_interrupt(int num)
764 struct rte_epoll_event event[num];
770 RTE_LOG(INFO, L3FWD_POWER,
771 "lcore %u sleeps until interrupt triggers\n",
774 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
775 for (i = 0; i < n; i++) {
776 data = event[i].epdata.data;
777 port_id = ((uintptr_t)data) >> CHAR_BIT;
778 queue_id = ((uintptr_t)data) &
779 RTE_LEN2MASK(CHAR_BIT, uint8_t);
780 rte_eth_dev_rx_intr_disable(port_id, queue_id);
781 RTE_LOG(INFO, L3FWD_POWER,
782 "lcore %u is waked up from rx interrupt on"
783 " port %d queue %d\n",
784 rte_lcore_id(), port_id, queue_id);
790 static void turn_on_intr(struct lcore_conf *qconf)
793 struct lcore_rx_queue *rx_queue;
797 for (i = 0; i < qconf->n_rx_queue; ++i) {
798 rx_queue = &(qconf->rx_queue_list[i]);
799 port_id = rx_queue->port_id;
800 queue_id = rx_queue->queue_id;
802 rte_spinlock_lock(&(locks[port_id]));
803 rte_eth_dev_rx_intr_enable(port_id, queue_id);
804 rte_spinlock_unlock(&(locks[port_id]));
808 static int event_register(struct lcore_conf *qconf)
810 struct lcore_rx_queue *rx_queue;
817 for (i = 0; i < qconf->n_rx_queue; ++i) {
818 rx_queue = &(qconf->rx_queue_list[i]);
819 portid = rx_queue->port_id;
820 queueid = rx_queue->queue_id;
821 data = portid << CHAR_BIT | queueid;
823 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
824 RTE_EPOLL_PER_THREAD,
826 (void *)((uintptr_t)data));
834 /* main processing loop */
836 main_loop(__attribute__((unused)) void *dummy)
838 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
840 uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
841 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
845 struct lcore_conf *qconf;
846 struct lcore_rx_queue *rx_queue;
847 enum freq_scale_hint_t lcore_scaleup_hint;
848 uint32_t lcore_rx_idle_count = 0;
849 uint32_t lcore_idle_hint = 0;
852 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
855 hz = rte_get_timer_hz();
856 tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
858 lcore_id = rte_lcore_id();
859 qconf = &lcore_conf[lcore_id];
861 if (qconf->n_rx_queue == 0) {
862 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
866 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
868 for (i = 0; i < qconf->n_rx_queue; i++) {
869 portid = qconf->rx_queue_list[i].port_id;
870 queueid = qconf->rx_queue_list[i].queue_id;
871 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
872 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
875 /* add into event wait list */
876 if (event_register(qconf) == 0)
879 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
882 stats[lcore_id].nb_iteration_looped++;
884 cur_tsc = rte_rdtsc();
885 cur_tsc_power = cur_tsc;
888 * TX burst queue drain
890 diff_tsc = cur_tsc - prev_tsc;
891 if (unlikely(diff_tsc > drain_tsc)) {
892 for (i = 0; i < qconf->n_tx_port; ++i) {
893 portid = qconf->tx_port_id[i];
894 rte_eth_tx_buffer_flush(portid,
895 qconf->tx_queue_id[portid],
896 qconf->tx_buffer[portid]);
901 diff_tsc_power = cur_tsc_power - prev_tsc_power;
902 if (diff_tsc_power > tim_res_tsc) {
904 prev_tsc_power = cur_tsc_power;
909 * Read packet from RX queues
911 lcore_scaleup_hint = FREQ_CURRENT;
912 lcore_rx_idle_count = 0;
913 for (i = 0; i < qconf->n_rx_queue; ++i) {
914 rx_queue = &(qconf->rx_queue_list[i]);
915 rx_queue->idle_hint = 0;
916 portid = rx_queue->port_id;
917 queueid = rx_queue->queue_id;
919 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
922 stats[lcore_id].nb_rx_processed += nb_rx;
923 if (unlikely(nb_rx == 0)) {
925 * no packet received from rx queue, try to
926 * sleep for a while forcing CPU enter deeper
929 rx_queue->zero_rx_packet_count++;
931 if (rx_queue->zero_rx_packet_count <=
935 rx_queue->idle_hint = power_idle_heuristic(\
936 rx_queue->zero_rx_packet_count);
937 lcore_rx_idle_count++;
939 rx_queue->zero_rx_packet_count = 0;
942 * do not scale up frequency immediately as
943 * user to kernel space communication is costly
944 * which might impact packet I/O for received
947 rx_queue->freq_up_hint =
948 power_freq_scaleup_heuristic(lcore_id,
952 /* Prefetch first packets */
953 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
954 rte_prefetch0(rte_pktmbuf_mtod(
955 pkts_burst[j], void *));
958 /* Prefetch and forward already prefetched packets */
959 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
960 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
961 j + PREFETCH_OFFSET], void *));
962 l3fwd_simple_forward(pkts_burst[j], portid,
966 /* Forward remaining prefetched packets */
967 for (; j < nb_rx; j++) {
968 l3fwd_simple_forward(pkts_burst[j], portid,
973 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
974 for (i = 1, lcore_scaleup_hint =
975 qconf->rx_queue_list[0].freq_up_hint;
976 i < qconf->n_rx_queue; ++i) {
977 rx_queue = &(qconf->rx_queue_list[i]);
978 if (rx_queue->freq_up_hint >
981 rx_queue->freq_up_hint;
984 if (lcore_scaleup_hint == FREQ_HIGHEST) {
985 if (rte_power_freq_max)
986 rte_power_freq_max(lcore_id);
987 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
988 if (rte_power_freq_up)
989 rte_power_freq_up(lcore_id);
993 * All Rx queues empty in recent consecutive polls,
994 * sleep in a conservative manner, meaning sleep as
997 for (i = 1, lcore_idle_hint =
998 qconf->rx_queue_list[0].idle_hint;
999 i < qconf->n_rx_queue; ++i) {
1000 rx_queue = &(qconf->rx_queue_list[i]);
1001 if (rx_queue->idle_hint < lcore_idle_hint)
1002 lcore_idle_hint = rx_queue->idle_hint;
1005 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1007 * execute "pause" instruction to avoid context
1008 * switch which generally take hundred of
1009 * microseconds for short sleep.
1011 rte_delay_us(lcore_idle_hint);
1013 /* suspend until rx interrupt trigges */
1015 turn_on_intr(qconf);
1016 sleep_until_rx_interrupt(
1019 * start receiving packets immediately
1024 stats[lcore_id].sleep_time += lcore_idle_hint;
1030 check_lcore_params(void)
1032 uint8_t queue, lcore;
1036 for (i = 0; i < nb_lcore_params; ++i) {
1037 queue = lcore_params[i].queue_id;
1038 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1039 printf("invalid queue number: %hhu\n", queue);
1042 lcore = lcore_params[i].lcore_id;
1043 if (!rte_lcore_is_enabled(lcore)) {
1044 printf("error: lcore %hhu is not enabled in lcore "
1048 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1050 printf("warning: lcore %hhu is on socket %d with numa "
1051 "off\n", lcore, socketid);
1058 check_port_config(void)
1063 for (i = 0; i < nb_lcore_params; ++i) {
1064 portid = lcore_params[i].port_id;
1065 if ((enabled_port_mask & (1 << portid)) == 0) {
1066 printf("port %u is not enabled in port mask\n",
1070 if (!rte_eth_dev_is_valid_port(portid)) {
1071 printf("port %u is not present on the board\n",
1080 get_port_n_rx_queues(const uint16_t port)
1085 for (i = 0; i < nb_lcore_params; ++i) {
1086 if (lcore_params[i].port_id == port &&
1087 lcore_params[i].queue_id > queue)
1088 queue = lcore_params[i].queue_id;
1090 return (uint8_t)(++queue);
1094 init_lcore_rx_queues(void)
1096 uint16_t i, nb_rx_queue;
1099 for (i = 0; i < nb_lcore_params; ++i) {
1100 lcore = lcore_params[i].lcore_id;
1101 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1102 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1103 printf("error: too many queues (%u) for lcore: %u\n",
1104 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1107 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1108 lcore_params[i].port_id;
1109 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1110 lcore_params[i].queue_id;
1111 lcore_conf[lcore].n_rx_queue++;
1119 print_usage(const char *prgname)
1121 printf ("%s [EAL options] -- -p PORTMASK -P"
1122 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1123 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1124 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1125 " -P : enable promiscuous mode\n"
1126 " --config (port,queue,lcore): rx queues configuration\n"
1127 " --no-numa: optional, disable numa awareness\n"
1128 " --enable-jumbo: enable jumbo frame"
1129 " which max packet len is PKTLEN in decimal (64-9600)\n"
1130 " --parse-ptype: parse packet type by software\n",
1134 static int parse_max_pkt_len(const char *pktlen)
1139 /* parse decimal string */
1140 len = strtoul(pktlen, &end, 10);
1141 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1151 parse_portmask(const char *portmask)
1156 /* parse hexadecimal string */
1157 pm = strtoul(portmask, &end, 16);
1158 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1168 parse_config(const char *q_arg)
1171 const char *p, *p0 = q_arg;
1179 unsigned long int_fld[_NUM_FLD];
1180 char *str_fld[_NUM_FLD];
1184 nb_lcore_params = 0;
1186 while ((p = strchr(p0,'(')) != NULL) {
1188 if((p0 = strchr(p,')')) == NULL)
1192 if(size >= sizeof(s))
1195 snprintf(s, sizeof(s), "%.*s", size, p);
1196 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1199 for (i = 0; i < _NUM_FLD; i++){
1201 int_fld[i] = strtoul(str_fld[i], &end, 0);
1202 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1206 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1207 printf("exceeded max number of lcore params: %hu\n",
1211 lcore_params_array[nb_lcore_params].port_id =
1212 (uint8_t)int_fld[FLD_PORT];
1213 lcore_params_array[nb_lcore_params].queue_id =
1214 (uint8_t)int_fld[FLD_QUEUE];
1215 lcore_params_array[nb_lcore_params].lcore_id =
1216 (uint8_t)int_fld[FLD_LCORE];
1219 lcore_params = lcore_params_array;
1224 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1226 /* Parse the argument given in the command line of the application */
1228 parse_args(int argc, char **argv)
1233 char *prgname = argv[0];
1234 static struct option lgopts[] = {
1235 {"config", 1, 0, 0},
1236 {"no-numa", 0, 0, 0},
1237 {"enable-jumbo", 0, 0, 0},
1238 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1244 while ((opt = getopt_long(argc, argvopt, "p:P",
1245 lgopts, &option_index)) != EOF) {
1250 enabled_port_mask = parse_portmask(optarg);
1251 if (enabled_port_mask == 0) {
1252 printf("invalid portmask\n");
1253 print_usage(prgname);
1258 printf("Promiscuous mode selected\n");
1264 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1265 ret = parse_config(optarg);
1267 printf("invalid config\n");
1268 print_usage(prgname);
1273 if (!strncmp(lgopts[option_index].name,
1275 printf("numa is disabled \n");
1279 if (!strncmp(lgopts[option_index].name,
1280 "enable-jumbo", 12)) {
1281 struct option lenopts =
1282 {"max-pkt-len", required_argument, \
1285 printf("jumbo frame is enabled \n");
1286 port_conf.rxmode.offloads |=
1287 DEV_RX_OFFLOAD_JUMBO_FRAME;
1288 port_conf.txmode.offloads |=
1289 DEV_TX_OFFLOAD_MULTI_SEGS;
1292 * if no max-pkt-len set, use the default value
1295 if (0 == getopt_long(argc, argvopt, "",
1296 &lenopts, &option_index)) {
1297 ret = parse_max_pkt_len(optarg);
1299 (ret > MAX_JUMBO_PKT_LEN)){
1300 printf("invalid packet "
1302 print_usage(prgname);
1305 port_conf.rxmode.max_rx_pkt_len = ret;
1307 printf("set jumbo frame "
1308 "max packet length to %u\n",
1309 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1312 if (!strncmp(lgopts[option_index].name,
1313 CMD_LINE_OPT_PARSE_PTYPE,
1314 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1315 printf("soft parse-ptype is enabled\n");
1322 print_usage(prgname);
1328 argv[optind-1] = prgname;
1331 optind = 1; /* reset getopt lib */
1336 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1338 char buf[ETHER_ADDR_FMT_SIZE];
1339 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1340 printf("%s%s", name, buf);
1343 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1345 setup_hash(int socketid)
1347 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1349 .entries = L3FWD_HASH_ENTRIES,
1350 .key_len = sizeof(struct ipv4_5tuple),
1351 .hash_func = DEFAULT_HASH_FUNC,
1352 .hash_func_init_val = 0,
1355 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1357 .entries = L3FWD_HASH_ENTRIES,
1358 .key_len = sizeof(struct ipv6_5tuple),
1359 .hash_func = DEFAULT_HASH_FUNC,
1360 .hash_func_init_val = 0,
1367 /* create ipv4 hash */
1368 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1369 ipv4_l3fwd_hash_params.name = s;
1370 ipv4_l3fwd_hash_params.socket_id = socketid;
1371 ipv4_l3fwd_lookup_struct[socketid] =
1372 rte_hash_create(&ipv4_l3fwd_hash_params);
1373 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1374 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1375 "socket %d\n", socketid);
1377 /* create ipv6 hash */
1378 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1379 ipv6_l3fwd_hash_params.name = s;
1380 ipv6_l3fwd_hash_params.socket_id = socketid;
1381 ipv6_l3fwd_lookup_struct[socketid] =
1382 rte_hash_create(&ipv6_l3fwd_hash_params);
1383 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1384 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1385 "socket %d\n", socketid);
1388 /* populate the ipv4 hash */
1389 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1390 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1391 (void *) &ipv4_l3fwd_route_array[i].key);
1393 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1394 "l3fwd hash on socket %d\n", i, socketid);
1396 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1397 printf("Hash: Adding key\n");
1398 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1401 /* populate the ipv6 hash */
1402 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1403 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1404 (void *) &ipv6_l3fwd_route_array[i].key);
1406 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1407 "l3fwd hash on socket %d\n", i, socketid);
1409 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1410 printf("Hash: Adding key\n");
1411 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1416 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1418 setup_lpm(int socketid)
1424 /* create the LPM table */
1425 struct rte_lpm_config lpm_ipv4_config;
1427 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1428 lpm_ipv4_config.number_tbl8s = 256;
1429 lpm_ipv4_config.flags = 0;
1431 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1432 ipv4_l3fwd_lookup_struct[socketid] =
1433 rte_lpm_create(s, socketid, &lpm_ipv4_config);
1434 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1435 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1436 " on socket %d\n", socketid);
1438 /* populate the LPM table */
1439 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1440 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1441 ipv4_l3fwd_route_array[i].ip,
1442 ipv4_l3fwd_route_array[i].depth,
1443 ipv4_l3fwd_route_array[i].if_out);
1446 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1447 "l3fwd LPM table on socket %d\n",
1451 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1452 (unsigned)ipv4_l3fwd_route_array[i].ip,
1453 ipv4_l3fwd_route_array[i].depth,
1454 ipv4_l3fwd_route_array[i].if_out);
1460 init_mem(unsigned nb_mbuf)
1462 struct lcore_conf *qconf;
1467 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1468 if (rte_lcore_is_enabled(lcore_id) == 0)
1472 socketid = rte_lcore_to_socket_id(lcore_id);
1476 if (socketid >= NB_SOCKETS) {
1477 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1478 "out of range %d\n", socketid,
1479 lcore_id, NB_SOCKETS);
1481 if (pktmbuf_pool[socketid] == NULL) {
1482 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1483 pktmbuf_pool[socketid] =
1484 rte_pktmbuf_pool_create(s, nb_mbuf,
1485 MEMPOOL_CACHE_SIZE, 0,
1486 RTE_MBUF_DEFAULT_BUF_SIZE,
1488 if (pktmbuf_pool[socketid] == NULL)
1489 rte_exit(EXIT_FAILURE,
1490 "Cannot init mbuf pool on socket %d\n",
1493 printf("Allocated mbuf pool on socket %d\n",
1496 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1497 setup_lpm(socketid);
1499 setup_hash(socketid);
1502 qconf = &lcore_conf[lcore_id];
1503 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1504 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1505 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1511 /* Check the link status of all ports in up to 9s, and print them finally */
1513 check_all_ports_link_status(uint32_t port_mask)
1515 #define CHECK_INTERVAL 100 /* 100ms */
1516 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1517 uint8_t count, all_ports_up, print_flag = 0;
1519 struct rte_eth_link link;
1521 printf("\nChecking link status");
1523 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1525 RTE_ETH_FOREACH_DEV(portid) {
1526 if ((port_mask & (1 << portid)) == 0)
1528 memset(&link, 0, sizeof(link));
1529 rte_eth_link_get_nowait(portid, &link);
1530 /* print link status if flag set */
1531 if (print_flag == 1) {
1532 if (link.link_status)
1533 printf("Port %d Link Up - speed %u "
1534 "Mbps - %s\n", (uint8_t)portid,
1535 (unsigned)link.link_speed,
1536 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1537 ("full-duplex") : ("half-duplex\n"));
1539 printf("Port %d Link Down\n",
1543 /* clear all_ports_up flag if any link down */
1544 if (link.link_status == ETH_LINK_DOWN) {
1549 /* after finally printing all link status, get out */
1550 if (print_flag == 1)
1553 if (all_ports_up == 0) {
1556 rte_delay_ms(CHECK_INTERVAL);
1559 /* set the print_flag if all ports up or timeout */
1560 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1567 static int check_ptype(uint16_t portid)
1570 int ptype_l3_ipv4 = 0;
1571 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1572 int ptype_l3_ipv6 = 0;
1574 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1576 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1580 uint32_t ptypes[ret];
1582 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1583 for (i = 0; i < ret; ++i) {
1584 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1586 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1587 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1592 if (ptype_l3_ipv4 == 0)
1593 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1595 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1596 if (ptype_l3_ipv6 == 0)
1597 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1600 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1602 #else /* APP_LOOKUP_EXACT_MATCH */
1603 if (ptype_l3_ipv4 && ptype_l3_ipv6)
1612 main(int argc, char **argv)
1614 struct lcore_conf *qconf;
1615 struct rte_eth_dev_info dev_info;
1616 struct rte_eth_txconf *txconf;
1622 uint32_t n_tx_queue, nb_lcores;
1623 uint32_t dev_rxq_num, dev_txq_num;
1624 uint8_t nb_rx_queue, queue, socketid;
1627 /* catch SIGINT and restore cpufreq governor to ondemand */
1628 signal(SIGINT, signal_exit_now);
1631 ret = rte_eal_init(argc, argv);
1633 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1637 /* init RTE timer library to be used late */
1638 rte_timer_subsystem_init();
1640 /* parse application arguments (after the EAL ones) */
1641 ret = parse_args(argc, argv);
1643 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1645 if (check_lcore_params() < 0)
1646 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1648 ret = init_lcore_rx_queues();
1650 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1652 nb_ports = rte_eth_dev_count_avail();
1654 if (check_port_config() < 0)
1655 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1657 nb_lcores = rte_lcore_count();
1659 /* initialize all ports */
1660 RTE_ETH_FOREACH_DEV(portid) {
1661 struct rte_eth_conf local_port_conf = port_conf;
1663 /* skip ports that are not enabled */
1664 if ((enabled_port_mask & (1 << portid)) == 0) {
1665 printf("\nSkipping disabled port %d\n", portid);
1670 printf("Initializing port %d ... ", portid );
1673 rte_eth_dev_info_get(portid, &dev_info);
1674 dev_rxq_num = dev_info.max_rx_queues;
1675 dev_txq_num = dev_info.max_tx_queues;
1677 nb_rx_queue = get_port_n_rx_queues(portid);
1678 if (nb_rx_queue > dev_rxq_num)
1679 rte_exit(EXIT_FAILURE,
1680 "Cannot configure not existed rxq: "
1681 "port=%d\n", portid);
1683 n_tx_queue = nb_lcores;
1684 if (n_tx_queue > dev_txq_num)
1685 n_tx_queue = dev_txq_num;
1686 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1687 nb_rx_queue, (unsigned)n_tx_queue );
1688 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1689 if (nb_rx_queue == 0)
1690 local_port_conf.intr_conf.rxq = 0;
1691 rte_eth_dev_info_get(portid, &dev_info);
1692 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1693 local_port_conf.txmode.offloads |=
1694 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1695 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1696 (uint16_t)n_tx_queue, &local_port_conf);
1698 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1699 "err=%d, port=%d\n", ret, portid);
1701 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1704 rte_exit(EXIT_FAILURE,
1705 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1708 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1709 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1713 ret = init_mem(NB_MBUF);
1715 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1717 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1718 if (rte_lcore_is_enabled(lcore_id) == 0)
1721 /* Initialize TX buffers */
1722 qconf = &lcore_conf[lcore_id];
1723 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1724 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1725 rte_eth_dev_socket_id(portid));
1726 if (qconf->tx_buffer[portid] == NULL)
1727 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1730 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1733 /* init one TX queue per couple (lcore,port) */
1735 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1736 if (rte_lcore_is_enabled(lcore_id) == 0)
1739 if (queueid >= dev_txq_num)
1744 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1748 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1751 txconf = &dev_info.default_txconf;
1752 txconf->offloads = local_port_conf.txmode.offloads;
1753 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1756 rte_exit(EXIT_FAILURE,
1757 "rte_eth_tx_queue_setup: err=%d, "
1758 "port=%d\n", ret, portid);
1760 qconf = &lcore_conf[lcore_id];
1761 qconf->tx_queue_id[portid] = queueid;
1764 qconf->tx_port_id[qconf->n_tx_port] = portid;
1770 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1771 if (rte_lcore_is_enabled(lcore_id) == 0)
1774 /* init power management library */
1775 ret = rte_power_init(lcore_id);
1778 "Library initialization failed on core %u\n", lcore_id);
1780 /* init timer structures for each enabled lcore */
1781 rte_timer_init(&power_timers[lcore_id]);
1782 hz = rte_get_timer_hz();
1783 rte_timer_reset(&power_timers[lcore_id],
1784 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1785 power_timer_cb, NULL);
1787 qconf = &lcore_conf[lcore_id];
1788 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1790 /* init RX queues */
1791 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1792 struct rte_eth_rxconf rxq_conf;
1793 struct rte_eth_dev *dev;
1794 struct rte_eth_conf *conf;
1796 portid = qconf->rx_queue_list[queue].port_id;
1797 queueid = qconf->rx_queue_list[queue].queue_id;
1798 dev = &rte_eth_devices[portid];
1799 conf = &dev->data->dev_conf;
1803 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1807 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1810 rte_eth_dev_info_get(portid, &dev_info);
1811 rxq_conf = dev_info.default_rxconf;
1812 rxq_conf.offloads = conf->rxmode.offloads;
1813 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1814 socketid, &rxq_conf,
1815 pktmbuf_pool[socketid]);
1817 rte_exit(EXIT_FAILURE,
1818 "rte_eth_rx_queue_setup: err=%d, "
1819 "port=%d\n", ret, portid);
1822 if (add_cb_parse_ptype(portid, queueid) < 0)
1823 rte_exit(EXIT_FAILURE,
1824 "Fail to add ptype cb\n");
1825 } else if (!check_ptype(portid))
1826 rte_exit(EXIT_FAILURE,
1827 "PMD can not provide needed ptypes\n");
1834 RTE_ETH_FOREACH_DEV(portid) {
1835 if ((enabled_port_mask & (1 << portid)) == 0) {
1839 ret = rte_eth_dev_start(portid);
1841 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1842 "port=%d\n", ret, portid);
1844 * If enabled, put device in promiscuous mode.
1845 * This allows IO forwarding mode to forward packets
1846 * to itself through 2 cross-connected ports of the
1850 rte_eth_promiscuous_enable(portid);
1851 /* initialize spinlock for each port */
1852 rte_spinlock_init(&(locks[portid]));
1855 check_all_ports_link_status(enabled_port_mask);
1857 /* launch per-lcore init on every lcore */
1858 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1859 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1860 if (rte_eal_wait_lcore(lcore_id) < 0)