1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 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 #include "perf_core.h"
50 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
52 #define MAX_PKT_BURST 32
54 #define MIN_ZERO_POLL_COUNT 10
57 #define TIMER_NUMBER_PER_SECOND 10
59 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
60 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
62 #define APP_LOOKUP_EXACT_MATCH 0
63 #define APP_LOOKUP_LPM 1
64 #define DO_RFC_1812_CHECKS
66 #ifndef APP_LOOKUP_METHOD
67 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
70 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
72 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
75 #error "APP_LOOKUP_METHOD set to incorrect value"
79 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
80 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
81 #define IPv6_BYTES(addr) \
82 addr[0], addr[1], addr[2], addr[3], \
83 addr[4], addr[5], addr[6], addr[7], \
84 addr[8], addr[9], addr[10], addr[11],\
85 addr[12], addr[13],addr[14], addr[15]
88 #define MAX_JUMBO_PKT_LEN 9600
90 #define IPV6_ADDR_LEN 16
92 #define MEMPOOL_CACHE_SIZE 256
95 * This expression is used to calculate the number of mbufs needed depending on
96 * user input, taking into account memory for rx and tx hardware rings, cache
97 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
98 * NB_MBUF never goes below a minimum value of 8192.
101 #define NB_MBUF RTE_MAX ( \
102 (nb_ports*nb_rx_queue*nb_rxd + \
103 nb_ports*nb_lcores*MAX_PKT_BURST + \
104 nb_ports*n_tx_queue*nb_txd + \
105 nb_lcores*MEMPOOL_CACHE_SIZE), \
108 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
112 /* Configure how many packets ahead to prefetch, when reading packets */
113 #define PREFETCH_OFFSET 3
116 * Configurable number of RX/TX ring descriptors
118 #define RTE_TEST_RX_DESC_DEFAULT 1024
119 #define RTE_TEST_TX_DESC_DEFAULT 1024
120 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
121 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
123 /* ethernet addresses of ports */
124 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
126 /* ethernet addresses of ports */
127 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
129 /* mask of enabled ports */
130 static uint32_t enabled_port_mask = 0;
131 /* Ports set in promiscuous mode off by default. */
132 static int promiscuous_on = 0;
133 /* NUMA is enabled by default. */
134 static int numa_on = 1;
135 static int parse_ptype; /**< Parse packet type using rx callback, and */
136 /**< disabled by default */
138 enum freq_scale_hint_t
146 struct lcore_rx_queue {
149 enum freq_scale_hint_t freq_up_hint;
150 uint32_t zero_rx_packet_count;
152 } __rte_cache_aligned;
154 #define MAX_RX_QUEUE_PER_LCORE 16
155 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
156 #define MAX_RX_QUEUE_PER_PORT 128
158 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
161 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
162 static struct lcore_params lcore_params_array_default[] = {
174 struct lcore_params *lcore_params = lcore_params_array_default;
175 uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
176 sizeof(lcore_params_array_default[0]);
178 static struct rte_eth_conf port_conf = {
180 .mq_mode = ETH_MQ_RX_RSS,
181 .max_rx_pkt_len = ETHER_MAX_LEN,
183 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
184 DEV_RX_OFFLOAD_CHECKSUM),
189 .rss_hf = ETH_RSS_UDP,
193 .mq_mode = ETH_MQ_TX_NONE,
200 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
203 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
206 #include <rte_hash_crc.h>
207 #define DEFAULT_HASH_FUNC rte_hash_crc
209 #include <rte_jhash.h>
210 #define DEFAULT_HASH_FUNC rte_jhash
219 } __attribute__((__packed__));
222 uint8_t ip_dst[IPV6_ADDR_LEN];
223 uint8_t ip_src[IPV6_ADDR_LEN];
227 } __attribute__((__packed__));
229 struct ipv4_l3fwd_route {
230 struct ipv4_5tuple key;
234 struct ipv6_l3fwd_route {
235 struct ipv6_5tuple key;
239 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
240 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
241 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
242 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
243 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
246 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
249 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
250 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
251 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
252 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
258 typedef struct rte_hash lookup_struct_t;
259 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
260 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
262 #define L3FWD_HASH_ENTRIES 1024
264 #define IPV4_L3FWD_NUM_ROUTES \
265 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
267 #define IPV6_L3FWD_NUM_ROUTES \
268 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
270 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
271 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
274 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
275 struct ipv4_l3fwd_route {
281 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
282 {IPv4(1,1,1,0), 24, 0},
283 {IPv4(2,1,1,0), 24, 1},
284 {IPv4(3,1,1,0), 24, 2},
285 {IPv4(4,1,1,0), 24, 3},
286 {IPv4(5,1,1,0), 24, 4},
287 {IPv4(6,1,1,0), 24, 5},
288 {IPv4(7,1,1,0), 24, 6},
289 {IPv4(8,1,1,0), 24, 7},
292 #define IPV4_L3FWD_NUM_ROUTES \
293 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
295 #define IPV4_L3FWD_LPM_MAX_RULES 1024
297 typedef struct rte_lpm lookup_struct_t;
298 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
303 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
305 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
306 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
307 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
308 lookup_struct_t * ipv4_lookup_struct;
309 lookup_struct_t * ipv6_lookup_struct;
310 } __rte_cache_aligned;
313 /* total sleep time in ms since last frequency scaling down */
315 /* number of long sleep recently */
316 uint32_t nb_long_sleep;
317 /* freq. scaling up trend */
319 /* total packet processed recently */
320 uint64_t nb_rx_processed;
321 /* total iterations looped recently */
322 uint64_t nb_iteration_looped;
324 } __rte_cache_aligned;
326 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
327 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
328 static struct rte_timer power_timers[RTE_MAX_LCORE];
330 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
331 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
332 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
334 /* exit signal handler */
336 signal_exit_now(int sigtype)
342 if (sigtype == SIGINT) {
343 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
344 if (rte_lcore_is_enabled(lcore_id) == 0)
347 /* init power management library */
348 ret = rte_power_exit(lcore_id);
350 rte_exit(EXIT_FAILURE, "Power management "
351 "library de-initialization failed on "
352 "core%u\n", lcore_id);
355 RTE_ETH_FOREACH_DEV(portid) {
356 if ((enabled_port_mask & (1 << portid)) == 0)
359 rte_eth_dev_stop(portid);
360 rte_eth_dev_close(portid);
364 rte_exit(EXIT_SUCCESS, "User forced exit\n");
367 /* Freqency scale down timer callback */
369 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
370 __attribute__((unused)) void *arg)
373 float sleep_time_ratio;
374 unsigned lcore_id = rte_lcore_id();
376 /* accumulate total execution time in us when callback is invoked */
377 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
378 (float)SCALING_PERIOD;
380 * check whether need to scale down frequency a step if it sleep a lot.
382 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
383 if (rte_power_freq_down)
384 rte_power_freq_down(lcore_id);
386 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
387 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
389 * scale down a step if average packet per iteration less
392 if (rte_power_freq_down)
393 rte_power_freq_down(lcore_id);
397 * initialize another timer according to current frequency to ensure
398 * timer interval is relatively fixed.
400 hz = rte_get_timer_hz();
401 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
402 SINGLE, lcore_id, power_timer_cb, NULL);
404 stats[lcore_id].nb_rx_processed = 0;
405 stats[lcore_id].nb_iteration_looped = 0;
407 stats[lcore_id].sleep_time = 0;
410 /* Enqueue a single packet, and send burst if queue is filled */
412 send_single_packet(struct rte_mbuf *m, uint16_t port)
415 struct lcore_conf *qconf;
417 lcore_id = rte_lcore_id();
418 qconf = &lcore_conf[lcore_id];
420 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
421 qconf->tx_buffer[port], m);
426 #ifdef DO_RFC_1812_CHECKS
428 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
430 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
432 * 1. The packet length reported by the Link Layer must be large
433 * enough to hold the minimum length legal IP datagram (20 bytes).
435 if (link_len < sizeof(struct ipv4_hdr))
438 /* 2. The IP checksum must be correct. */
439 /* this is checked in H/W */
442 * 3. The IP version number must be 4. If the version number is not 4
443 * then the packet may be another version of IP, such as IPng or
446 if (((pkt->version_ihl) >> 4) != 4)
449 * 4. The IP header length field must be large enough to hold the
450 * minimum length legal IP datagram (20 bytes = 5 words).
452 if ((pkt->version_ihl & 0xf) < 5)
456 * 5. The IP total length field must be large enough to hold the IP
457 * datagram header, whose length is specified in the IP header length
460 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
467 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
469 print_ipv4_key(struct ipv4_5tuple key)
471 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
472 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
473 key.port_dst, key.port_src, key.proto);
476 print_ipv6_key(struct ipv6_5tuple key)
478 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
479 "port dst = %d, port src = %d, proto = %d\n",
480 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
481 key.port_dst, key.port_src, key.proto);
484 static inline uint16_t
485 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
486 lookup_struct_t * ipv4_l3fwd_lookup_struct)
488 struct ipv4_5tuple key;
493 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
494 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
495 key.proto = ipv4_hdr->next_proto_id;
497 switch (ipv4_hdr->next_proto_id) {
499 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
500 sizeof(struct ipv4_hdr));
501 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
502 key.port_src = rte_be_to_cpu_16(tcp->src_port);
506 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
507 sizeof(struct ipv4_hdr));
508 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
509 key.port_src = rte_be_to_cpu_16(udp->src_port);
518 /* Find destination port */
519 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
520 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
523 static inline uint16_t
524 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
525 lookup_struct_t *ipv6_l3fwd_lookup_struct)
527 struct ipv6_5tuple key;
532 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
533 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
535 key.proto = ipv6_hdr->proto;
537 switch (ipv6_hdr->proto) {
539 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
540 sizeof(struct ipv6_hdr));
541 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
542 key.port_src = rte_be_to_cpu_16(tcp->src_port);
546 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
547 sizeof(struct ipv6_hdr));
548 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
549 key.port_src = rte_be_to_cpu_16(udp->src_port);
558 /* Find destination port */
559 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
560 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
564 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
565 static inline uint16_t
566 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
567 lookup_struct_t *ipv4_l3fwd_lookup_struct)
571 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
572 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
578 parse_ptype_one(struct rte_mbuf *m)
580 struct ether_hdr *eth_hdr;
581 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
584 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
585 ether_type = eth_hdr->ether_type;
586 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
587 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
588 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
589 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
591 m->packet_type = packet_type;
595 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
596 struct rte_mbuf *pkts[], uint16_t nb_pkts,
597 uint16_t max_pkts __rte_unused,
598 void *user_param __rte_unused)
602 for (i = 0; i < nb_pkts; ++i)
603 parse_ptype_one(pkts[i]);
609 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
611 printf("Port %d: softly parse packet type info\n", portid);
612 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
615 printf("Failed to add rx callback: port=%d\n", portid);
620 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
621 struct lcore_conf *qconf)
623 struct ether_hdr *eth_hdr;
624 struct ipv4_hdr *ipv4_hdr;
628 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
630 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
631 /* Handle IPv4 headers.*/
633 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
634 sizeof(struct ether_hdr));
636 #ifdef DO_RFC_1812_CHECKS
637 /* Check to make sure the packet is valid (RFC1812) */
638 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
644 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
645 qconf->ipv4_lookup_struct);
646 if (dst_port >= RTE_MAX_ETHPORTS ||
647 (enabled_port_mask & 1 << dst_port) == 0)
650 /* 02:00:00:00:00:xx */
651 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
652 *((uint64_t *)d_addr_bytes) =
653 0x000000000002 + ((uint64_t)dst_port << 40);
655 #ifdef DO_RFC_1812_CHECKS
656 /* Update time to live and header checksum */
657 --(ipv4_hdr->time_to_live);
658 ++(ipv4_hdr->hdr_checksum);
662 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
664 send_single_packet(m, dst_port);
665 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
666 /* Handle IPv6 headers.*/
667 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
668 struct ipv6_hdr *ipv6_hdr;
671 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
672 sizeof(struct ether_hdr));
674 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
675 qconf->ipv6_lookup_struct);
677 if (dst_port >= RTE_MAX_ETHPORTS ||
678 (enabled_port_mask & 1 << dst_port) == 0)
681 /* 02:00:00:00:00:xx */
682 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
683 *((uint64_t *)d_addr_bytes) =
684 0x000000000002 + ((uint64_t)dst_port << 40);
687 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
689 send_single_packet(m, dst_port);
691 /* We don't currently handle IPv6 packets in LPM mode. */
699 #define MINIMUM_SLEEP_TIME 1
700 #define SUSPEND_THRESHOLD 300
702 static inline uint32_t
703 power_idle_heuristic(uint32_t zero_rx_packet_count)
705 /* If zero count is less than 100, sleep 1us */
706 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
707 return MINIMUM_SLEEP_TIME;
708 /* If zero count is less than 1000, sleep 100 us which is the
709 minimum latency switching from C3/C6 to C0
712 return SUSPEND_THRESHOLD;
715 static inline enum freq_scale_hint_t
716 power_freq_scaleup_heuristic(unsigned lcore_id,
720 uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
722 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
725 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
726 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
727 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
728 #define FREQ_UP_TREND1_ACC 1
729 #define FREQ_UP_TREND2_ACC 100
730 #define FREQ_UP_THRESHOLD 10000
732 if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
733 stats[lcore_id].trend = 0;
735 } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
736 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
737 else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
738 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
740 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
741 stats[lcore_id].trend = 0;
749 * force polling thread sleep until one-shot rx interrupt triggers
758 sleep_until_rx_interrupt(int num)
760 struct rte_epoll_event event[num];
766 RTE_LOG(INFO, L3FWD_POWER,
767 "lcore %u sleeps until interrupt triggers\n",
770 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
771 for (i = 0; i < n; i++) {
772 data = event[i].epdata.data;
773 port_id = ((uintptr_t)data) >> CHAR_BIT;
774 queue_id = ((uintptr_t)data) &
775 RTE_LEN2MASK(CHAR_BIT, uint8_t);
776 rte_eth_dev_rx_intr_disable(port_id, queue_id);
777 RTE_LOG(INFO, L3FWD_POWER,
778 "lcore %u is waked up from rx interrupt on"
779 " port %d queue %d\n",
780 rte_lcore_id(), port_id, queue_id);
786 static void turn_on_intr(struct lcore_conf *qconf)
789 struct lcore_rx_queue *rx_queue;
793 for (i = 0; i < qconf->n_rx_queue; ++i) {
794 rx_queue = &(qconf->rx_queue_list[i]);
795 port_id = rx_queue->port_id;
796 queue_id = rx_queue->queue_id;
798 rte_spinlock_lock(&(locks[port_id]));
799 rte_eth_dev_rx_intr_enable(port_id, queue_id);
800 rte_spinlock_unlock(&(locks[port_id]));
804 static int event_register(struct lcore_conf *qconf)
806 struct lcore_rx_queue *rx_queue;
813 for (i = 0; i < qconf->n_rx_queue; ++i) {
814 rx_queue = &(qconf->rx_queue_list[i]);
815 portid = rx_queue->port_id;
816 queueid = rx_queue->queue_id;
817 data = portid << CHAR_BIT | queueid;
819 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
820 RTE_EPOLL_PER_THREAD,
822 (void *)((uintptr_t)data));
830 /* main processing loop */
832 main_loop(__attribute__((unused)) void *dummy)
834 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
836 uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
837 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
841 struct lcore_conf *qconf;
842 struct lcore_rx_queue *rx_queue;
843 enum freq_scale_hint_t lcore_scaleup_hint;
844 uint32_t lcore_rx_idle_count = 0;
845 uint32_t lcore_idle_hint = 0;
848 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
851 hz = rte_get_timer_hz();
852 tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
854 lcore_id = rte_lcore_id();
855 qconf = &lcore_conf[lcore_id];
857 if (qconf->n_rx_queue == 0) {
858 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
862 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
864 for (i = 0; i < qconf->n_rx_queue; i++) {
865 portid = qconf->rx_queue_list[i].port_id;
866 queueid = qconf->rx_queue_list[i].queue_id;
867 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
868 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
871 /* add into event wait list */
872 if (event_register(qconf) == 0)
875 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
878 stats[lcore_id].nb_iteration_looped++;
880 cur_tsc = rte_rdtsc();
881 cur_tsc_power = cur_tsc;
884 * TX burst queue drain
886 diff_tsc = cur_tsc - prev_tsc;
887 if (unlikely(diff_tsc > drain_tsc)) {
888 for (i = 0; i < qconf->n_tx_port; ++i) {
889 portid = qconf->tx_port_id[i];
890 rte_eth_tx_buffer_flush(portid,
891 qconf->tx_queue_id[portid],
892 qconf->tx_buffer[portid]);
897 diff_tsc_power = cur_tsc_power - prev_tsc_power;
898 if (diff_tsc_power > tim_res_tsc) {
900 prev_tsc_power = cur_tsc_power;
905 * Read packet from RX queues
907 lcore_scaleup_hint = FREQ_CURRENT;
908 lcore_rx_idle_count = 0;
909 for (i = 0; i < qconf->n_rx_queue; ++i) {
910 rx_queue = &(qconf->rx_queue_list[i]);
911 rx_queue->idle_hint = 0;
912 portid = rx_queue->port_id;
913 queueid = rx_queue->queue_id;
915 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
918 stats[lcore_id].nb_rx_processed += nb_rx;
919 if (unlikely(nb_rx == 0)) {
921 * no packet received from rx queue, try to
922 * sleep for a while forcing CPU enter deeper
925 rx_queue->zero_rx_packet_count++;
927 if (rx_queue->zero_rx_packet_count <=
931 rx_queue->idle_hint = power_idle_heuristic(\
932 rx_queue->zero_rx_packet_count);
933 lcore_rx_idle_count++;
935 rx_queue->zero_rx_packet_count = 0;
938 * do not scale up frequency immediately as
939 * user to kernel space communication is costly
940 * which might impact packet I/O for received
943 rx_queue->freq_up_hint =
944 power_freq_scaleup_heuristic(lcore_id,
948 /* Prefetch first packets */
949 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
950 rte_prefetch0(rte_pktmbuf_mtod(
951 pkts_burst[j], void *));
954 /* Prefetch and forward already prefetched packets */
955 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
956 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
957 j + PREFETCH_OFFSET], void *));
958 l3fwd_simple_forward(pkts_burst[j], portid,
962 /* Forward remaining prefetched packets */
963 for (; j < nb_rx; j++) {
964 l3fwd_simple_forward(pkts_burst[j], portid,
969 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
970 for (i = 1, lcore_scaleup_hint =
971 qconf->rx_queue_list[0].freq_up_hint;
972 i < qconf->n_rx_queue; ++i) {
973 rx_queue = &(qconf->rx_queue_list[i]);
974 if (rx_queue->freq_up_hint >
977 rx_queue->freq_up_hint;
980 if (lcore_scaleup_hint == FREQ_HIGHEST) {
981 if (rte_power_freq_max)
982 rte_power_freq_max(lcore_id);
983 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
984 if (rte_power_freq_up)
985 rte_power_freq_up(lcore_id);
989 * All Rx queues empty in recent consecutive polls,
990 * sleep in a conservative manner, meaning sleep as
993 for (i = 1, lcore_idle_hint =
994 qconf->rx_queue_list[0].idle_hint;
995 i < qconf->n_rx_queue; ++i) {
996 rx_queue = &(qconf->rx_queue_list[i]);
997 if (rx_queue->idle_hint < lcore_idle_hint)
998 lcore_idle_hint = rx_queue->idle_hint;
1001 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1003 * execute "pause" instruction to avoid context
1004 * switch which generally take hundred of
1005 * microseconds for short sleep.
1007 rte_delay_us(lcore_idle_hint);
1009 /* suspend until rx interrupt trigges */
1011 turn_on_intr(qconf);
1012 sleep_until_rx_interrupt(
1015 * start receiving packets immediately
1020 stats[lcore_id].sleep_time += lcore_idle_hint;
1026 check_lcore_params(void)
1028 uint8_t queue, lcore;
1032 for (i = 0; i < nb_lcore_params; ++i) {
1033 queue = lcore_params[i].queue_id;
1034 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1035 printf("invalid queue number: %hhu\n", queue);
1038 lcore = lcore_params[i].lcore_id;
1039 if (!rte_lcore_is_enabled(lcore)) {
1040 printf("error: lcore %hhu is not enabled in lcore "
1044 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1046 printf("warning: lcore %hhu is on socket %d with numa "
1047 "off\n", lcore, socketid);
1054 check_port_config(void)
1059 for (i = 0; i < nb_lcore_params; ++i) {
1060 portid = lcore_params[i].port_id;
1061 if ((enabled_port_mask & (1 << portid)) == 0) {
1062 printf("port %u is not enabled in port mask\n",
1066 if (!rte_eth_dev_is_valid_port(portid)) {
1067 printf("port %u is not present on the board\n",
1076 get_port_n_rx_queues(const uint16_t port)
1081 for (i = 0; i < nb_lcore_params; ++i) {
1082 if (lcore_params[i].port_id == port &&
1083 lcore_params[i].queue_id > queue)
1084 queue = lcore_params[i].queue_id;
1086 return (uint8_t)(++queue);
1090 init_lcore_rx_queues(void)
1092 uint16_t i, nb_rx_queue;
1095 for (i = 0; i < nb_lcore_params; ++i) {
1096 lcore = lcore_params[i].lcore_id;
1097 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1098 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1099 printf("error: too many queues (%u) for lcore: %u\n",
1100 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1103 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1104 lcore_params[i].port_id;
1105 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1106 lcore_params[i].queue_id;
1107 lcore_conf[lcore].n_rx_queue++;
1115 print_usage(const char *prgname)
1117 printf ("%s [EAL options] -- -p PORTMASK -P"
1118 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1119 " [--high-perf-cores CORELIST"
1120 " [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1121 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1122 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1123 " -P : enable promiscuous mode\n"
1124 " --config (port,queue,lcore): rx queues configuration\n"
1125 " --high-perf-cores CORELIST: list of high performance cores\n"
1126 " --perf-config: similar as config, cores specified as indices"
1127 " for bins containing high or regular performance cores\n"
1128 " --no-numa: optional, disable numa awareness\n"
1129 " --enable-jumbo: enable jumbo frame"
1130 " which max packet len is PKTLEN in decimal (64-9600)\n"
1131 " --parse-ptype: parse packet type by software\n",
1135 static int parse_max_pkt_len(const char *pktlen)
1140 /* parse decimal string */
1141 len = strtoul(pktlen, &end, 10);
1142 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1152 parse_portmask(const char *portmask)
1157 /* parse hexadecimal string */
1158 pm = strtoul(portmask, &end, 16);
1159 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1169 parse_config(const char *q_arg)
1172 const char *p, *p0 = q_arg;
1180 unsigned long int_fld[_NUM_FLD];
1181 char *str_fld[_NUM_FLD];
1185 nb_lcore_params = 0;
1187 while ((p = strchr(p0,'(')) != NULL) {
1189 if((p0 = strchr(p,')')) == NULL)
1193 if(size >= sizeof(s))
1196 snprintf(s, sizeof(s), "%.*s", size, p);
1197 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1200 for (i = 0; i < _NUM_FLD; i++){
1202 int_fld[i] = strtoul(str_fld[i], &end, 0);
1203 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1207 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1208 printf("exceeded max number of lcore params: %hu\n",
1212 lcore_params_array[nb_lcore_params].port_id =
1213 (uint8_t)int_fld[FLD_PORT];
1214 lcore_params_array[nb_lcore_params].queue_id =
1215 (uint8_t)int_fld[FLD_QUEUE];
1216 lcore_params_array[nb_lcore_params].lcore_id =
1217 (uint8_t)int_fld[FLD_LCORE];
1220 lcore_params = lcore_params_array;
1225 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1227 /* Parse the argument given in the command line of the application */
1229 parse_args(int argc, char **argv)
1234 char *prgname = argv[0];
1235 static struct option lgopts[] = {
1236 {"config", 1, 0, 0},
1237 {"perf-config", 1, 0, 0},
1238 {"high-perf-cores", 1, 0, 0},
1239 {"no-numa", 0, 0, 0},
1240 {"enable-jumbo", 0, 0, 0},
1241 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1247 while ((opt = getopt_long(argc, argvopt, "p:P",
1248 lgopts, &option_index)) != EOF) {
1253 enabled_port_mask = parse_portmask(optarg);
1254 if (enabled_port_mask == 0) {
1255 printf("invalid portmask\n");
1256 print_usage(prgname);
1261 printf("Promiscuous mode selected\n");
1267 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1268 ret = parse_config(optarg);
1270 printf("invalid config\n");
1271 print_usage(prgname);
1276 if (!strncmp(lgopts[option_index].name,
1277 "perf-config", 11)) {
1278 ret = parse_perf_config(optarg);
1280 printf("invalid perf-config\n");
1281 print_usage(prgname);
1286 if (!strncmp(lgopts[option_index].name,
1287 "high-perf-cores", 15)) {
1288 ret = parse_perf_core_list(optarg);
1290 printf("invalid high-perf-cores\n");
1291 print_usage(prgname);
1296 if (!strncmp(lgopts[option_index].name,
1298 printf("numa is disabled \n");
1302 if (!strncmp(lgopts[option_index].name,
1303 "enable-jumbo", 12)) {
1304 struct option lenopts =
1305 {"max-pkt-len", required_argument, \
1308 printf("jumbo frame is enabled \n");
1309 port_conf.rxmode.offloads |=
1310 DEV_RX_OFFLOAD_JUMBO_FRAME;
1311 port_conf.txmode.offloads |=
1312 DEV_TX_OFFLOAD_MULTI_SEGS;
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);
1335 if (!strncmp(lgopts[option_index].name,
1336 CMD_LINE_OPT_PARSE_PTYPE,
1337 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1338 printf("soft parse-ptype is enabled\n");
1345 print_usage(prgname);
1351 argv[optind-1] = prgname;
1354 optind = 1; /* reset getopt lib */
1359 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1361 char buf[ETHER_ADDR_FMT_SIZE];
1362 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1363 printf("%s%s", name, buf);
1366 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1368 setup_hash(int socketid)
1370 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1372 .entries = L3FWD_HASH_ENTRIES,
1373 .key_len = sizeof(struct ipv4_5tuple),
1374 .hash_func = DEFAULT_HASH_FUNC,
1375 .hash_func_init_val = 0,
1378 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1380 .entries = L3FWD_HASH_ENTRIES,
1381 .key_len = sizeof(struct ipv6_5tuple),
1382 .hash_func = DEFAULT_HASH_FUNC,
1383 .hash_func_init_val = 0,
1390 /* create ipv4 hash */
1391 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1392 ipv4_l3fwd_hash_params.name = s;
1393 ipv4_l3fwd_hash_params.socket_id = socketid;
1394 ipv4_l3fwd_lookup_struct[socketid] =
1395 rte_hash_create(&ipv4_l3fwd_hash_params);
1396 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1397 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1398 "socket %d\n", socketid);
1400 /* create ipv6 hash */
1401 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1402 ipv6_l3fwd_hash_params.name = s;
1403 ipv6_l3fwd_hash_params.socket_id = socketid;
1404 ipv6_l3fwd_lookup_struct[socketid] =
1405 rte_hash_create(&ipv6_l3fwd_hash_params);
1406 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1407 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1408 "socket %d\n", socketid);
1411 /* populate the ipv4 hash */
1412 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1413 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1414 (void *) &ipv4_l3fwd_route_array[i].key);
1416 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1417 "l3fwd hash on socket %d\n", i, socketid);
1419 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1420 printf("Hash: Adding key\n");
1421 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1424 /* populate the ipv6 hash */
1425 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1426 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1427 (void *) &ipv6_l3fwd_route_array[i].key);
1429 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1430 "l3fwd hash on socket %d\n", i, socketid);
1432 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1433 printf("Hash: Adding key\n");
1434 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1439 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1441 setup_lpm(int socketid)
1447 /* create the LPM table */
1448 struct rte_lpm_config lpm_ipv4_config;
1450 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1451 lpm_ipv4_config.number_tbl8s = 256;
1452 lpm_ipv4_config.flags = 0;
1454 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1455 ipv4_l3fwd_lookup_struct[socketid] =
1456 rte_lpm_create(s, socketid, &lpm_ipv4_config);
1457 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1458 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1459 " on socket %d\n", socketid);
1461 /* populate the LPM table */
1462 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1463 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1464 ipv4_l3fwd_route_array[i].ip,
1465 ipv4_l3fwd_route_array[i].depth,
1466 ipv4_l3fwd_route_array[i].if_out);
1469 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1470 "l3fwd LPM table on socket %d\n",
1474 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1475 (unsigned)ipv4_l3fwd_route_array[i].ip,
1476 ipv4_l3fwd_route_array[i].depth,
1477 ipv4_l3fwd_route_array[i].if_out);
1483 init_mem(unsigned nb_mbuf)
1485 struct lcore_conf *qconf;
1490 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1491 if (rte_lcore_is_enabled(lcore_id) == 0)
1495 socketid = rte_lcore_to_socket_id(lcore_id);
1499 if (socketid >= NB_SOCKETS) {
1500 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1501 "out of range %d\n", socketid,
1502 lcore_id, NB_SOCKETS);
1504 if (pktmbuf_pool[socketid] == NULL) {
1505 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1506 pktmbuf_pool[socketid] =
1507 rte_pktmbuf_pool_create(s, nb_mbuf,
1508 MEMPOOL_CACHE_SIZE, 0,
1509 RTE_MBUF_DEFAULT_BUF_SIZE,
1511 if (pktmbuf_pool[socketid] == NULL)
1512 rte_exit(EXIT_FAILURE,
1513 "Cannot init mbuf pool on socket %d\n",
1516 printf("Allocated mbuf pool on socket %d\n",
1519 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1520 setup_lpm(socketid);
1522 setup_hash(socketid);
1525 qconf = &lcore_conf[lcore_id];
1526 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1527 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1528 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1534 /* Check the link status of all ports in up to 9s, and print them finally */
1536 check_all_ports_link_status(uint32_t port_mask)
1538 #define CHECK_INTERVAL 100 /* 100ms */
1539 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1540 uint8_t count, all_ports_up, print_flag = 0;
1542 struct rte_eth_link link;
1544 printf("\nChecking link status");
1546 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1548 RTE_ETH_FOREACH_DEV(portid) {
1549 if ((port_mask & (1 << portid)) == 0)
1551 memset(&link, 0, sizeof(link));
1552 rte_eth_link_get_nowait(portid, &link);
1553 /* print link status if flag set */
1554 if (print_flag == 1) {
1555 if (link.link_status)
1556 printf("Port %d Link Up - speed %u "
1557 "Mbps - %s\n", (uint8_t)portid,
1558 (unsigned)link.link_speed,
1559 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1560 ("full-duplex") : ("half-duplex\n"));
1562 printf("Port %d Link Down\n",
1566 /* clear all_ports_up flag if any link down */
1567 if (link.link_status == ETH_LINK_DOWN) {
1572 /* after finally printing all link status, get out */
1573 if (print_flag == 1)
1576 if (all_ports_up == 0) {
1579 rte_delay_ms(CHECK_INTERVAL);
1582 /* set the print_flag if all ports up or timeout */
1583 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1590 static int check_ptype(uint16_t portid)
1593 int ptype_l3_ipv4 = 0;
1594 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1595 int ptype_l3_ipv6 = 0;
1597 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1599 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1603 uint32_t ptypes[ret];
1605 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1606 for (i = 0; i < ret; ++i) {
1607 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1609 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1610 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1615 if (ptype_l3_ipv4 == 0)
1616 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1618 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1619 if (ptype_l3_ipv6 == 0)
1620 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1623 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1625 #else /* APP_LOOKUP_EXACT_MATCH */
1626 if (ptype_l3_ipv4 && ptype_l3_ipv6)
1635 init_power_library(void)
1637 int ret = 0, lcore_id;
1638 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1639 if (rte_lcore_is_enabled(lcore_id)) {
1640 /* init power management library */
1641 ret = rte_power_init(lcore_id);
1644 "Library initialization failed on core %u\n",
1652 main(int argc, char **argv)
1654 struct lcore_conf *qconf;
1655 struct rte_eth_dev_info dev_info;
1656 struct rte_eth_txconf *txconf;
1662 uint32_t n_tx_queue, nb_lcores;
1663 uint32_t dev_rxq_num, dev_txq_num;
1664 uint8_t nb_rx_queue, queue, socketid;
1667 /* catch SIGINT and restore cpufreq governor to ondemand */
1668 signal(SIGINT, signal_exit_now);
1671 ret = rte_eal_init(argc, argv);
1673 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1677 /* init RTE timer library to be used late */
1678 rte_timer_subsystem_init();
1680 /* parse application arguments (after the EAL ones) */
1681 ret = parse_args(argc, argv);
1683 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1685 if (init_power_library())
1686 rte_exit(EXIT_FAILURE, "init_power_library failed\n");
1688 if (update_lcore_params() < 0)
1689 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
1691 if (check_lcore_params() < 0)
1692 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1694 ret = init_lcore_rx_queues();
1696 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1698 nb_ports = rte_eth_dev_count_avail();
1700 if (check_port_config() < 0)
1701 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1703 nb_lcores = rte_lcore_count();
1705 /* initialize all ports */
1706 RTE_ETH_FOREACH_DEV(portid) {
1707 struct rte_eth_conf local_port_conf = port_conf;
1709 /* skip ports that are not enabled */
1710 if ((enabled_port_mask & (1 << portid)) == 0) {
1711 printf("\nSkipping disabled port %d\n", portid);
1716 printf("Initializing port %d ... ", portid );
1719 rte_eth_dev_info_get(portid, &dev_info);
1720 dev_rxq_num = dev_info.max_rx_queues;
1721 dev_txq_num = dev_info.max_tx_queues;
1723 nb_rx_queue = get_port_n_rx_queues(portid);
1724 if (nb_rx_queue > dev_rxq_num)
1725 rte_exit(EXIT_FAILURE,
1726 "Cannot configure not existed rxq: "
1727 "port=%d\n", portid);
1729 n_tx_queue = nb_lcores;
1730 if (n_tx_queue > dev_txq_num)
1731 n_tx_queue = dev_txq_num;
1732 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1733 nb_rx_queue, (unsigned)n_tx_queue );
1734 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1735 if (nb_rx_queue == 0)
1736 local_port_conf.intr_conf.rxq = 0;
1737 rte_eth_dev_info_get(portid, &dev_info);
1738 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1739 local_port_conf.txmode.offloads |=
1740 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1742 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1743 dev_info.flow_type_rss_offloads;
1744 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1745 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1746 printf("Port %u modified RSS hash function based on hardware support,"
1747 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1749 port_conf.rx_adv_conf.rss_conf.rss_hf,
1750 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1753 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1754 (uint16_t)n_tx_queue, &local_port_conf);
1756 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1757 "err=%d, port=%d\n", ret, portid);
1759 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1762 rte_exit(EXIT_FAILURE,
1763 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1766 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1767 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1771 ret = init_mem(NB_MBUF);
1773 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1775 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1776 if (rte_lcore_is_enabled(lcore_id) == 0)
1779 /* Initialize TX buffers */
1780 qconf = &lcore_conf[lcore_id];
1781 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1782 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1783 rte_eth_dev_socket_id(portid));
1784 if (qconf->tx_buffer[portid] == NULL)
1785 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1788 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1791 /* init one TX queue per couple (lcore,port) */
1793 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1794 if (rte_lcore_is_enabled(lcore_id) == 0)
1797 if (queueid >= dev_txq_num)
1802 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1806 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1809 txconf = &dev_info.default_txconf;
1810 txconf->offloads = local_port_conf.txmode.offloads;
1811 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1814 rte_exit(EXIT_FAILURE,
1815 "rte_eth_tx_queue_setup: err=%d, "
1816 "port=%d\n", ret, portid);
1818 qconf = &lcore_conf[lcore_id];
1819 qconf->tx_queue_id[portid] = queueid;
1822 qconf->tx_port_id[qconf->n_tx_port] = portid;
1828 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1829 if (rte_lcore_is_enabled(lcore_id) == 0)
1832 /* init timer structures for each enabled lcore */
1833 rte_timer_init(&power_timers[lcore_id]);
1834 hz = rte_get_timer_hz();
1835 rte_timer_reset(&power_timers[lcore_id],
1836 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1837 power_timer_cb, NULL);
1839 qconf = &lcore_conf[lcore_id];
1840 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1842 /* init RX queues */
1843 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1844 struct rte_eth_rxconf rxq_conf;
1845 struct rte_eth_dev *dev;
1846 struct rte_eth_conf *conf;
1848 portid = qconf->rx_queue_list[queue].port_id;
1849 queueid = qconf->rx_queue_list[queue].queue_id;
1850 dev = &rte_eth_devices[portid];
1851 conf = &dev->data->dev_conf;
1855 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1859 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1862 rte_eth_dev_info_get(portid, &dev_info);
1863 rxq_conf = dev_info.default_rxconf;
1864 rxq_conf.offloads = conf->rxmode.offloads;
1865 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1866 socketid, &rxq_conf,
1867 pktmbuf_pool[socketid]);
1869 rte_exit(EXIT_FAILURE,
1870 "rte_eth_rx_queue_setup: err=%d, "
1871 "port=%d\n", ret, portid);
1874 if (add_cb_parse_ptype(portid, queueid) < 0)
1875 rte_exit(EXIT_FAILURE,
1876 "Fail to add ptype cb\n");
1877 } else if (!check_ptype(portid))
1878 rte_exit(EXIT_FAILURE,
1879 "PMD can not provide needed ptypes\n");
1886 RTE_ETH_FOREACH_DEV(portid) {
1887 if ((enabled_port_mask & (1 << portid)) == 0) {
1891 ret = rte_eth_dev_start(portid);
1893 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1894 "port=%d\n", ret, portid);
1896 * If enabled, put device in promiscuous mode.
1897 * This allows IO forwarding mode to forward packets
1898 * to itself through 2 cross-connected ports of the
1902 rte_eth_promiscuous_enable(portid);
1903 /* initialize spinlock for each port */
1904 rte_spinlock_init(&(locks[portid]));
1907 check_all_ports_link_status(enabled_port_mask);
1909 /* launch per-lcore init on every lcore */
1910 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1911 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1912 if (rte_eal_wait_lcore(lcore_id) < 0)