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 512
116 #define RTE_TEST_TX_DESC_DEFAULT 512
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 .ignore_offload_bitfield = 1,
188 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
189 DEV_RX_OFFLOAD_CHECKSUM),
194 .rss_hf = ETH_RSS_UDP,
198 .mq_mode = ETH_MQ_TX_NONE,
205 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
208 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
211 #include <rte_hash_crc.h>
212 #define DEFAULT_HASH_FUNC rte_hash_crc
214 #include <rte_jhash.h>
215 #define DEFAULT_HASH_FUNC rte_jhash
224 } __attribute__((__packed__));
227 uint8_t ip_dst[IPV6_ADDR_LEN];
228 uint8_t ip_src[IPV6_ADDR_LEN];
232 } __attribute__((__packed__));
234 struct ipv4_l3fwd_route {
235 struct ipv4_5tuple key;
239 struct ipv6_l3fwd_route {
240 struct ipv6_5tuple key;
244 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
245 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
246 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
247 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
248 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
251 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
254 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
255 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
256 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
257 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
263 typedef struct rte_hash lookup_struct_t;
264 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
265 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
267 #define L3FWD_HASH_ENTRIES 1024
269 #define IPV4_L3FWD_NUM_ROUTES \
270 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
272 #define IPV6_L3FWD_NUM_ROUTES \
273 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
275 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
276 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
279 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
280 struct ipv4_l3fwd_route {
286 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
287 {IPv4(1,1,1,0), 24, 0},
288 {IPv4(2,1,1,0), 24, 1},
289 {IPv4(3,1,1,0), 24, 2},
290 {IPv4(4,1,1,0), 24, 3},
291 {IPv4(5,1,1,0), 24, 4},
292 {IPv4(6,1,1,0), 24, 5},
293 {IPv4(7,1,1,0), 24, 6},
294 {IPv4(8,1,1,0), 24, 7},
297 #define IPV4_L3FWD_NUM_ROUTES \
298 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
300 #define IPV4_L3FWD_LPM_MAX_RULES 1024
302 typedef struct rte_lpm lookup_struct_t;
303 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
308 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
310 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
311 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
312 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
313 lookup_struct_t * ipv4_lookup_struct;
314 lookup_struct_t * ipv6_lookup_struct;
315 } __rte_cache_aligned;
318 /* total sleep time in ms since last frequency scaling down */
320 /* number of long sleep recently */
321 uint32_t nb_long_sleep;
322 /* freq. scaling up trend */
324 /* total packet processed recently */
325 uint64_t nb_rx_processed;
326 /* total iterations looped recently */
327 uint64_t nb_iteration_looped;
329 } __rte_cache_aligned;
331 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
332 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
333 static struct rte_timer power_timers[RTE_MAX_LCORE];
335 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
336 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
337 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
339 /* exit signal handler */
341 signal_exit_now(int sigtype)
344 unsigned int portid, nb_ports;
347 if (sigtype == SIGINT) {
348 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
349 if (rte_lcore_is_enabled(lcore_id) == 0)
352 /* init power management library */
353 ret = rte_power_exit(lcore_id);
355 rte_exit(EXIT_FAILURE, "Power management "
356 "library de-initialization failed on "
357 "core%u\n", lcore_id);
360 nb_ports = rte_eth_dev_count();
361 for (portid = 0; portid < nb_ports; portid++) {
362 if ((enabled_port_mask & (1 << portid)) == 0)
365 rte_eth_dev_stop(portid);
366 rte_eth_dev_close(portid);
370 rte_exit(EXIT_SUCCESS, "User forced exit\n");
373 /* Freqency scale down timer callback */
375 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
376 __attribute__((unused)) void *arg)
379 float sleep_time_ratio;
380 unsigned lcore_id = rte_lcore_id();
382 /* accumulate total execution time in us when callback is invoked */
383 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
384 (float)SCALING_PERIOD;
386 * check whether need to scale down frequency a step if it sleep a lot.
388 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
389 if (rte_power_freq_down)
390 rte_power_freq_down(lcore_id);
392 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
393 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
395 * scale down a step if average packet per iteration less
398 if (rte_power_freq_down)
399 rte_power_freq_down(lcore_id);
403 * initialize another timer according to current frequency to ensure
404 * timer interval is relatively fixed.
406 hz = rte_get_timer_hz();
407 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
408 SINGLE, lcore_id, power_timer_cb, NULL);
410 stats[lcore_id].nb_rx_processed = 0;
411 stats[lcore_id].nb_iteration_looped = 0;
413 stats[lcore_id].sleep_time = 0;
416 /* Enqueue a single packet, and send burst if queue is filled */
418 send_single_packet(struct rte_mbuf *m, uint16_t port)
421 struct lcore_conf *qconf;
423 lcore_id = rte_lcore_id();
424 qconf = &lcore_conf[lcore_id];
426 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
427 qconf->tx_buffer[port], m);
432 #ifdef DO_RFC_1812_CHECKS
434 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
436 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
438 * 1. The packet length reported by the Link Layer must be large
439 * enough to hold the minimum length legal IP datagram (20 bytes).
441 if (link_len < sizeof(struct ipv4_hdr))
444 /* 2. The IP checksum must be correct. */
445 /* this is checked in H/W */
448 * 3. The IP version number must be 4. If the version number is not 4
449 * then the packet may be another version of IP, such as IPng or
452 if (((pkt->version_ihl) >> 4) != 4)
455 * 4. The IP header length field must be large enough to hold the
456 * minimum length legal IP datagram (20 bytes = 5 words).
458 if ((pkt->version_ihl & 0xf) < 5)
462 * 5. The IP total length field must be large enough to hold the IP
463 * datagram header, whose length is specified in the IP header length
466 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
473 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
475 print_ipv4_key(struct ipv4_5tuple key)
477 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
478 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
479 key.port_dst, key.port_src, key.proto);
482 print_ipv6_key(struct ipv6_5tuple key)
484 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
485 "port dst = %d, port src = %d, proto = %d\n",
486 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
487 key.port_dst, key.port_src, key.proto);
490 static inline uint16_t
491 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
492 lookup_struct_t * ipv4_l3fwd_lookup_struct)
494 struct ipv4_5tuple key;
499 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
500 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
501 key.proto = ipv4_hdr->next_proto_id;
503 switch (ipv4_hdr->next_proto_id) {
505 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
506 sizeof(struct ipv4_hdr));
507 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
508 key.port_src = rte_be_to_cpu_16(tcp->src_port);
512 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
513 sizeof(struct ipv4_hdr));
514 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
515 key.port_src = rte_be_to_cpu_16(udp->src_port);
524 /* Find destination port */
525 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
526 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
529 static inline uint16_t
530 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
531 lookup_struct_t *ipv6_l3fwd_lookup_struct)
533 struct ipv6_5tuple key;
538 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
539 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
541 key.proto = ipv6_hdr->proto;
543 switch (ipv6_hdr->proto) {
545 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
546 sizeof(struct ipv6_hdr));
547 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
548 key.port_src = rte_be_to_cpu_16(tcp->src_port);
552 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
553 sizeof(struct ipv6_hdr));
554 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
555 key.port_src = rte_be_to_cpu_16(udp->src_port);
564 /* Find destination port */
565 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
566 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
570 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
571 static inline uint16_t
572 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
573 lookup_struct_t *ipv4_l3fwd_lookup_struct)
577 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
578 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
584 parse_ptype_one(struct rte_mbuf *m)
586 struct ether_hdr *eth_hdr;
587 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
590 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
591 ether_type = eth_hdr->ether_type;
592 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
593 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
594 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
595 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
597 m->packet_type = packet_type;
601 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
602 struct rte_mbuf *pkts[], uint16_t nb_pkts,
603 uint16_t max_pkts __rte_unused,
604 void *user_param __rte_unused)
608 for (i = 0; i < nb_pkts; ++i)
609 parse_ptype_one(pkts[i]);
615 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
617 printf("Port %d: softly parse packet type info\n", portid);
618 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
621 printf("Failed to add rx callback: port=%d\n", portid);
626 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
627 struct lcore_conf *qconf)
629 struct ether_hdr *eth_hdr;
630 struct ipv4_hdr *ipv4_hdr;
634 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
636 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
637 /* Handle IPv4 headers.*/
639 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
640 sizeof(struct ether_hdr));
642 #ifdef DO_RFC_1812_CHECKS
643 /* Check to make sure the packet is valid (RFC1812) */
644 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
650 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
651 qconf->ipv4_lookup_struct);
652 if (dst_port >= RTE_MAX_ETHPORTS ||
653 (enabled_port_mask & 1 << dst_port) == 0)
656 /* 02:00:00:00:00:xx */
657 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
658 *((uint64_t *)d_addr_bytes) =
659 0x000000000002 + ((uint64_t)dst_port << 40);
661 #ifdef DO_RFC_1812_CHECKS
662 /* Update time to live and header checksum */
663 --(ipv4_hdr->time_to_live);
664 ++(ipv4_hdr->hdr_checksum);
668 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
670 send_single_packet(m, dst_port);
671 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
672 /* Handle IPv6 headers.*/
673 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
674 struct ipv6_hdr *ipv6_hdr;
677 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
678 sizeof(struct ether_hdr));
680 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
681 qconf->ipv6_lookup_struct);
683 if (dst_port >= RTE_MAX_ETHPORTS ||
684 (enabled_port_mask & 1 << dst_port) == 0)
687 /* 02:00:00:00:00:xx */
688 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
689 *((uint64_t *)d_addr_bytes) =
690 0x000000000002 + ((uint64_t)dst_port << 40);
693 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
695 send_single_packet(m, dst_port);
697 /* We don't currently handle IPv6 packets in LPM mode. */
705 #define MINIMUM_SLEEP_TIME 1
706 #define SUSPEND_THRESHOLD 300
708 static inline uint32_t
709 power_idle_heuristic(uint32_t zero_rx_packet_count)
711 /* If zero count is less than 100, sleep 1us */
712 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
713 return MINIMUM_SLEEP_TIME;
714 /* If zero count is less than 1000, sleep 100 us which is the
715 minimum latency switching from C3/C6 to C0
718 return SUSPEND_THRESHOLD;
721 static inline enum freq_scale_hint_t
722 power_freq_scaleup_heuristic(unsigned lcore_id,
726 uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
728 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
731 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
732 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
733 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
734 #define FREQ_UP_TREND1_ACC 1
735 #define FREQ_UP_TREND2_ACC 100
736 #define FREQ_UP_THRESHOLD 10000
738 if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
739 stats[lcore_id].trend = 0;
741 } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
742 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
743 else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
744 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
746 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
747 stats[lcore_id].trend = 0;
755 * force polling thread sleep until one-shot rx interrupt triggers
764 sleep_until_rx_interrupt(int num)
766 struct rte_epoll_event event[num];
772 RTE_LOG(INFO, L3FWD_POWER,
773 "lcore %u sleeps until interrupt triggers\n",
776 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
777 for (i = 0; i < n; i++) {
778 data = event[i].epdata.data;
779 port_id = ((uintptr_t)data) >> CHAR_BIT;
780 queue_id = ((uintptr_t)data) &
781 RTE_LEN2MASK(CHAR_BIT, uint8_t);
782 rte_eth_dev_rx_intr_disable(port_id, queue_id);
783 RTE_LOG(INFO, L3FWD_POWER,
784 "lcore %u is waked up from rx interrupt on"
785 " port %d queue %d\n",
786 rte_lcore_id(), port_id, queue_id);
792 static void turn_on_intr(struct lcore_conf *qconf)
795 struct lcore_rx_queue *rx_queue;
799 for (i = 0; i < qconf->n_rx_queue; ++i) {
800 rx_queue = &(qconf->rx_queue_list[i]);
801 port_id = rx_queue->port_id;
802 queue_id = rx_queue->queue_id;
804 rte_spinlock_lock(&(locks[port_id]));
805 rte_eth_dev_rx_intr_enable(port_id, queue_id);
806 rte_spinlock_unlock(&(locks[port_id]));
810 static int event_register(struct lcore_conf *qconf)
812 struct lcore_rx_queue *rx_queue;
819 for (i = 0; i < qconf->n_rx_queue; ++i) {
820 rx_queue = &(qconf->rx_queue_list[i]);
821 portid = rx_queue->port_id;
822 queueid = rx_queue->queue_id;
823 data = portid << CHAR_BIT | queueid;
825 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
826 RTE_EPOLL_PER_THREAD,
828 (void *)((uintptr_t)data));
836 /* main processing loop */
838 main_loop(__attribute__((unused)) void *dummy)
840 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
842 uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
843 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
847 struct lcore_conf *qconf;
848 struct lcore_rx_queue *rx_queue;
849 enum freq_scale_hint_t lcore_scaleup_hint;
850 uint32_t lcore_rx_idle_count = 0;
851 uint32_t lcore_idle_hint = 0;
854 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
857 hz = rte_get_timer_hz();
858 tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
860 lcore_id = rte_lcore_id();
861 qconf = &lcore_conf[lcore_id];
863 if (qconf->n_rx_queue == 0) {
864 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
868 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
870 for (i = 0; i < qconf->n_rx_queue; i++) {
871 portid = qconf->rx_queue_list[i].port_id;
872 queueid = qconf->rx_queue_list[i].queue_id;
873 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
874 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
877 /* add into event wait list */
878 if (event_register(qconf) == 0)
881 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
884 stats[lcore_id].nb_iteration_looped++;
886 cur_tsc = rte_rdtsc();
887 cur_tsc_power = cur_tsc;
890 * TX burst queue drain
892 diff_tsc = cur_tsc - prev_tsc;
893 if (unlikely(diff_tsc > drain_tsc)) {
894 for (i = 0; i < qconf->n_tx_port; ++i) {
895 portid = qconf->tx_port_id[i];
896 rte_eth_tx_buffer_flush(portid,
897 qconf->tx_queue_id[portid],
898 qconf->tx_buffer[portid]);
903 diff_tsc_power = cur_tsc_power - prev_tsc_power;
904 if (diff_tsc_power > tim_res_tsc) {
906 prev_tsc_power = cur_tsc_power;
911 * Read packet from RX queues
913 lcore_scaleup_hint = FREQ_CURRENT;
914 lcore_rx_idle_count = 0;
915 for (i = 0; i < qconf->n_rx_queue; ++i) {
916 rx_queue = &(qconf->rx_queue_list[i]);
917 rx_queue->idle_hint = 0;
918 portid = rx_queue->port_id;
919 queueid = rx_queue->queue_id;
921 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
924 stats[lcore_id].nb_rx_processed += nb_rx;
925 if (unlikely(nb_rx == 0)) {
927 * no packet received from rx queue, try to
928 * sleep for a while forcing CPU enter deeper
931 rx_queue->zero_rx_packet_count++;
933 if (rx_queue->zero_rx_packet_count <=
937 rx_queue->idle_hint = power_idle_heuristic(\
938 rx_queue->zero_rx_packet_count);
939 lcore_rx_idle_count++;
941 rx_queue->zero_rx_packet_count = 0;
944 * do not scale up frequency immediately as
945 * user to kernel space communication is costly
946 * which might impact packet I/O for received
949 rx_queue->freq_up_hint =
950 power_freq_scaleup_heuristic(lcore_id,
954 /* Prefetch first packets */
955 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
956 rte_prefetch0(rte_pktmbuf_mtod(
957 pkts_burst[j], void *));
960 /* Prefetch and forward already prefetched packets */
961 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
962 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
963 j + PREFETCH_OFFSET], void *));
964 l3fwd_simple_forward(pkts_burst[j], portid,
968 /* Forward remaining prefetched packets */
969 for (; j < nb_rx; j++) {
970 l3fwd_simple_forward(pkts_burst[j], portid,
975 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
976 for (i = 1, lcore_scaleup_hint =
977 qconf->rx_queue_list[0].freq_up_hint;
978 i < qconf->n_rx_queue; ++i) {
979 rx_queue = &(qconf->rx_queue_list[i]);
980 if (rx_queue->freq_up_hint >
983 rx_queue->freq_up_hint;
986 if (lcore_scaleup_hint == FREQ_HIGHEST) {
987 if (rte_power_freq_max)
988 rte_power_freq_max(lcore_id);
989 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
990 if (rte_power_freq_up)
991 rte_power_freq_up(lcore_id);
995 * All Rx queues empty in recent consecutive polls,
996 * sleep in a conservative manner, meaning sleep as
999 for (i = 1, lcore_idle_hint =
1000 qconf->rx_queue_list[0].idle_hint;
1001 i < qconf->n_rx_queue; ++i) {
1002 rx_queue = &(qconf->rx_queue_list[i]);
1003 if (rx_queue->idle_hint < lcore_idle_hint)
1004 lcore_idle_hint = rx_queue->idle_hint;
1007 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1009 * execute "pause" instruction to avoid context
1010 * switch which generally take hundred of
1011 * microseconds for short sleep.
1013 rte_delay_us(lcore_idle_hint);
1015 /* suspend until rx interrupt trigges */
1017 turn_on_intr(qconf);
1018 sleep_until_rx_interrupt(
1021 * start receiving packets immediately
1026 stats[lcore_id].sleep_time += lcore_idle_hint;
1032 check_lcore_params(void)
1034 uint8_t queue, lcore;
1038 for (i = 0; i < nb_lcore_params; ++i) {
1039 queue = lcore_params[i].queue_id;
1040 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1041 printf("invalid queue number: %hhu\n", queue);
1044 lcore = lcore_params[i].lcore_id;
1045 if (!rte_lcore_is_enabled(lcore)) {
1046 printf("error: lcore %hhu is not enabled in lcore "
1050 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1052 printf("warning: lcore %hhu is on socket %d with numa "
1053 "off\n", lcore, socketid);
1060 check_port_config(const unsigned nb_ports)
1065 for (i = 0; i < nb_lcore_params; ++i) {
1066 portid = lcore_params[i].port_id;
1067 if ((enabled_port_mask & (1 << portid)) == 0) {
1068 printf("port %u is not enabled in port mask\n",
1072 if (portid >= nb_ports) {
1073 printf("port %u is not present on the board\n",
1082 get_port_n_rx_queues(const uint16_t port)
1087 for (i = 0; i < nb_lcore_params; ++i) {
1088 if (lcore_params[i].port_id == port &&
1089 lcore_params[i].queue_id > queue)
1090 queue = lcore_params[i].queue_id;
1092 return (uint8_t)(++queue);
1096 init_lcore_rx_queues(void)
1098 uint16_t i, nb_rx_queue;
1101 for (i = 0; i < nb_lcore_params; ++i) {
1102 lcore = lcore_params[i].lcore_id;
1103 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1104 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1105 printf("error: too many queues (%u) for lcore: %u\n",
1106 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1109 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1110 lcore_params[i].port_id;
1111 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1112 lcore_params[i].queue_id;
1113 lcore_conf[lcore].n_rx_queue++;
1121 print_usage(const char *prgname)
1123 printf ("%s [EAL options] -- -p PORTMASK -P"
1124 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1125 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1126 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1127 " -P : enable promiscuous mode\n"
1128 " --config (port,queue,lcore): rx queues configuration\n"
1129 " --no-numa: optional, disable numa awareness\n"
1130 " --enable-jumbo: enable jumbo frame"
1131 " which max packet len is PKTLEN in decimal (64-9600)\n"
1132 " --parse-ptype: parse packet type by software\n",
1136 static int parse_max_pkt_len(const char *pktlen)
1141 /* parse decimal string */
1142 len = strtoul(pktlen, &end, 10);
1143 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1153 parse_portmask(const char *portmask)
1158 /* parse hexadecimal string */
1159 pm = strtoul(portmask, &end, 16);
1160 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1170 parse_config(const char *q_arg)
1173 const char *p, *p0 = q_arg;
1181 unsigned long int_fld[_NUM_FLD];
1182 char *str_fld[_NUM_FLD];
1186 nb_lcore_params = 0;
1188 while ((p = strchr(p0,'(')) != NULL) {
1190 if((p0 = strchr(p,')')) == NULL)
1194 if(size >= sizeof(s))
1197 snprintf(s, sizeof(s), "%.*s", size, p);
1198 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1201 for (i = 0; i < _NUM_FLD; i++){
1203 int_fld[i] = strtoul(str_fld[i], &end, 0);
1204 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1208 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1209 printf("exceeded max number of lcore params: %hu\n",
1213 lcore_params_array[nb_lcore_params].port_id =
1214 (uint8_t)int_fld[FLD_PORT];
1215 lcore_params_array[nb_lcore_params].queue_id =
1216 (uint8_t)int_fld[FLD_QUEUE];
1217 lcore_params_array[nb_lcore_params].lcore_id =
1218 (uint8_t)int_fld[FLD_LCORE];
1221 lcore_params = lcore_params_array;
1226 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1228 /* Parse the argument given in the command line of the application */
1230 parse_args(int argc, char **argv)
1235 char *prgname = argv[0];
1236 static struct option lgopts[] = {
1237 {"config", 1, 0, 0},
1238 {"no-numa", 0, 0, 0},
1239 {"enable-jumbo", 0, 0, 0},
1240 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1246 while ((opt = getopt_long(argc, argvopt, "p:P",
1247 lgopts, &option_index)) != EOF) {
1252 enabled_port_mask = parse_portmask(optarg);
1253 if (enabled_port_mask == 0) {
1254 printf("invalid portmask\n");
1255 print_usage(prgname);
1260 printf("Promiscuous mode selected\n");
1266 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1267 ret = parse_config(optarg);
1269 printf("invalid config\n");
1270 print_usage(prgname);
1275 if (!strncmp(lgopts[option_index].name,
1277 printf("numa is disabled \n");
1281 if (!strncmp(lgopts[option_index].name,
1282 "enable-jumbo", 12)) {
1283 struct option lenopts =
1284 {"max-pkt-len", required_argument, \
1287 printf("jumbo frame is enabled \n");
1288 port_conf.rxmode.offloads |=
1289 DEV_RX_OFFLOAD_JUMBO_FRAME;
1290 port_conf.txmode.offloads |=
1291 DEV_TX_OFFLOAD_MULTI_SEGS;
1294 * if no max-pkt-len set, use the default value
1297 if (0 == getopt_long(argc, argvopt, "",
1298 &lenopts, &option_index)) {
1299 ret = parse_max_pkt_len(optarg);
1301 (ret > MAX_JUMBO_PKT_LEN)){
1302 printf("invalid packet "
1304 print_usage(prgname);
1307 port_conf.rxmode.max_rx_pkt_len = ret;
1309 printf("set jumbo frame "
1310 "max packet length to %u\n",
1311 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1314 if (!strncmp(lgopts[option_index].name,
1315 CMD_LINE_OPT_PARSE_PTYPE,
1316 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1317 printf("soft parse-ptype is enabled\n");
1324 print_usage(prgname);
1330 argv[optind-1] = prgname;
1333 optind = 1; /* reset getopt lib */
1338 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1340 char buf[ETHER_ADDR_FMT_SIZE];
1341 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1342 printf("%s%s", name, buf);
1345 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1347 setup_hash(int socketid)
1349 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1351 .entries = L3FWD_HASH_ENTRIES,
1352 .key_len = sizeof(struct ipv4_5tuple),
1353 .hash_func = DEFAULT_HASH_FUNC,
1354 .hash_func_init_val = 0,
1357 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1359 .entries = L3FWD_HASH_ENTRIES,
1360 .key_len = sizeof(struct ipv6_5tuple),
1361 .hash_func = DEFAULT_HASH_FUNC,
1362 .hash_func_init_val = 0,
1369 /* create ipv4 hash */
1370 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1371 ipv4_l3fwd_hash_params.name = s;
1372 ipv4_l3fwd_hash_params.socket_id = socketid;
1373 ipv4_l3fwd_lookup_struct[socketid] =
1374 rte_hash_create(&ipv4_l3fwd_hash_params);
1375 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1376 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1377 "socket %d\n", socketid);
1379 /* create ipv6 hash */
1380 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1381 ipv6_l3fwd_hash_params.name = s;
1382 ipv6_l3fwd_hash_params.socket_id = socketid;
1383 ipv6_l3fwd_lookup_struct[socketid] =
1384 rte_hash_create(&ipv6_l3fwd_hash_params);
1385 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1386 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1387 "socket %d\n", socketid);
1390 /* populate the ipv4 hash */
1391 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1392 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1393 (void *) &ipv4_l3fwd_route_array[i].key);
1395 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1396 "l3fwd hash on socket %d\n", i, socketid);
1398 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1399 printf("Hash: Adding key\n");
1400 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1403 /* populate the ipv6 hash */
1404 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1405 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1406 (void *) &ipv6_l3fwd_route_array[i].key);
1408 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1409 "l3fwd hash on socket %d\n", i, socketid);
1411 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1412 printf("Hash: Adding key\n");
1413 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1418 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1420 setup_lpm(int socketid)
1426 /* create the LPM table */
1427 struct rte_lpm_config lpm_ipv4_config;
1429 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1430 lpm_ipv4_config.number_tbl8s = 256;
1431 lpm_ipv4_config.flags = 0;
1433 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1434 ipv4_l3fwd_lookup_struct[socketid] =
1435 rte_lpm_create(s, socketid, &lpm_ipv4_config);
1436 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1437 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1438 " on socket %d\n", socketid);
1440 /* populate the LPM table */
1441 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1442 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1443 ipv4_l3fwd_route_array[i].ip,
1444 ipv4_l3fwd_route_array[i].depth,
1445 ipv4_l3fwd_route_array[i].if_out);
1448 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1449 "l3fwd LPM table on socket %d\n",
1453 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1454 (unsigned)ipv4_l3fwd_route_array[i].ip,
1455 ipv4_l3fwd_route_array[i].depth,
1456 ipv4_l3fwd_route_array[i].if_out);
1462 init_mem(unsigned nb_mbuf)
1464 struct lcore_conf *qconf;
1469 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1470 if (rte_lcore_is_enabled(lcore_id) == 0)
1474 socketid = rte_lcore_to_socket_id(lcore_id);
1478 if (socketid >= NB_SOCKETS) {
1479 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1480 "out of range %d\n", socketid,
1481 lcore_id, NB_SOCKETS);
1483 if (pktmbuf_pool[socketid] == NULL) {
1484 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1485 pktmbuf_pool[socketid] =
1486 rte_pktmbuf_pool_create(s, nb_mbuf,
1487 MEMPOOL_CACHE_SIZE, 0,
1488 RTE_MBUF_DEFAULT_BUF_SIZE,
1490 if (pktmbuf_pool[socketid] == NULL)
1491 rte_exit(EXIT_FAILURE,
1492 "Cannot init mbuf pool on socket %d\n",
1495 printf("Allocated mbuf pool on socket %d\n",
1498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1499 setup_lpm(socketid);
1501 setup_hash(socketid);
1504 qconf = &lcore_conf[lcore_id];
1505 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1506 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1507 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1513 /* Check the link status of all ports in up to 9s, and print them finally */
1515 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
1517 #define CHECK_INTERVAL 100 /* 100ms */
1518 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1519 uint8_t count, all_ports_up, print_flag = 0;
1521 struct rte_eth_link link;
1523 printf("\nChecking link status");
1525 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1527 for (portid = 0; portid < port_num; portid++) {
1528 if ((port_mask & (1 << portid)) == 0)
1530 memset(&link, 0, sizeof(link));
1531 rte_eth_link_get_nowait(portid, &link);
1532 /* print link status if flag set */
1533 if (print_flag == 1) {
1534 if (link.link_status)
1535 printf("Port %d Link Up - speed %u "
1536 "Mbps - %s\n", (uint8_t)portid,
1537 (unsigned)link.link_speed,
1538 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1539 ("full-duplex") : ("half-duplex\n"));
1541 printf("Port %d Link Down\n",
1545 /* clear all_ports_up flag if any link down */
1546 if (link.link_status == ETH_LINK_DOWN) {
1551 /* after finally printing all link status, get out */
1552 if (print_flag == 1)
1555 if (all_ports_up == 0) {
1558 rte_delay_ms(CHECK_INTERVAL);
1561 /* set the print_flag if all ports up or timeout */
1562 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1569 static int check_ptype(uint16_t portid)
1572 int ptype_l3_ipv4 = 0;
1573 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1574 int ptype_l3_ipv6 = 0;
1576 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1578 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1582 uint32_t ptypes[ret];
1584 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1585 for (i = 0; i < ret; ++i) {
1586 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1588 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1589 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1594 if (ptype_l3_ipv4 == 0)
1595 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1597 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1598 if (ptype_l3_ipv6 == 0)
1599 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1602 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1604 #else /* APP_LOOKUP_EXACT_MATCH */
1605 if (ptype_l3_ipv4 && ptype_l3_ipv6)
1614 main(int argc, char **argv)
1616 struct lcore_conf *qconf;
1617 struct rte_eth_dev_info dev_info;
1618 struct rte_eth_txconf *txconf;
1624 uint32_t n_tx_queue, nb_lcores;
1625 uint32_t dev_rxq_num, dev_txq_num;
1626 uint8_t nb_rx_queue, queue, socketid;
1629 /* catch SIGINT and restore cpufreq governor to ondemand */
1630 signal(SIGINT, signal_exit_now);
1633 ret = rte_eal_init(argc, argv);
1635 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1639 /* init RTE timer library to be used late */
1640 rte_timer_subsystem_init();
1642 /* parse application arguments (after the EAL ones) */
1643 ret = parse_args(argc, argv);
1645 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1647 if (check_lcore_params() < 0)
1648 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1650 ret = init_lcore_rx_queues();
1652 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1654 nb_ports = rte_eth_dev_count();
1656 if (check_port_config(nb_ports) < 0)
1657 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1659 nb_lcores = rte_lcore_count();
1661 /* initialize all ports */
1662 for (portid = 0; portid < nb_ports; portid++) {
1663 struct rte_eth_conf local_port_conf = port_conf;
1665 /* skip ports that are not enabled */
1666 if ((enabled_port_mask & (1 << portid)) == 0) {
1667 printf("\nSkipping disabled port %d\n", portid);
1672 printf("Initializing port %d ... ", portid );
1675 rte_eth_dev_info_get(portid, &dev_info);
1676 dev_rxq_num = dev_info.max_rx_queues;
1677 dev_txq_num = dev_info.max_tx_queues;
1679 nb_rx_queue = get_port_n_rx_queues(portid);
1680 if (nb_rx_queue > dev_rxq_num)
1681 rte_exit(EXIT_FAILURE,
1682 "Cannot configure not existed rxq: "
1683 "port=%d\n", portid);
1685 n_tx_queue = nb_lcores;
1686 if (n_tx_queue > dev_txq_num)
1687 n_tx_queue = dev_txq_num;
1688 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1689 nb_rx_queue, (unsigned)n_tx_queue );
1690 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1691 if (nb_rx_queue == 0)
1692 local_port_conf.intr_conf.rxq = 0;
1693 rte_eth_dev_info_get(portid, &dev_info);
1694 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1695 local_port_conf.txmode.offloads |=
1696 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1697 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1698 (uint16_t)n_tx_queue, &local_port_conf);
1700 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1701 "err=%d, port=%d\n", ret, portid);
1703 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1706 rte_exit(EXIT_FAILURE,
1707 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1710 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1711 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1715 ret = init_mem(NB_MBUF);
1717 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1719 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1720 if (rte_lcore_is_enabled(lcore_id) == 0)
1723 /* Initialize TX buffers */
1724 qconf = &lcore_conf[lcore_id];
1725 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1726 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1727 rte_eth_dev_socket_id(portid));
1728 if (qconf->tx_buffer[portid] == NULL)
1729 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1732 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1735 /* init one TX queue per couple (lcore,port) */
1737 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1738 if (rte_lcore_is_enabled(lcore_id) == 0)
1741 if (queueid >= dev_txq_num)
1746 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1750 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1753 txconf = &dev_info.default_txconf;
1754 txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
1755 txconf->offloads = local_port_conf.txmode.offloads;
1756 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1759 rte_exit(EXIT_FAILURE,
1760 "rte_eth_tx_queue_setup: err=%d, "
1761 "port=%d\n", ret, portid);
1763 qconf = &lcore_conf[lcore_id];
1764 qconf->tx_queue_id[portid] = queueid;
1767 qconf->tx_port_id[qconf->n_tx_port] = portid;
1773 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1774 if (rte_lcore_is_enabled(lcore_id) == 0)
1777 /* init power management library */
1778 ret = rte_power_init(lcore_id);
1781 "Library initialization failed on core %u\n", lcore_id);
1783 /* init timer structures for each enabled lcore */
1784 rte_timer_init(&power_timers[lcore_id]);
1785 hz = rte_get_timer_hz();
1786 rte_timer_reset(&power_timers[lcore_id],
1787 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1788 power_timer_cb, NULL);
1790 qconf = &lcore_conf[lcore_id];
1791 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1793 /* init RX queues */
1794 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1795 struct rte_eth_rxconf rxq_conf;
1796 struct rte_eth_dev *dev;
1797 struct rte_eth_conf *conf;
1799 portid = qconf->rx_queue_list[queue].port_id;
1800 queueid = qconf->rx_queue_list[queue].queue_id;
1801 dev = &rte_eth_devices[portid];
1802 conf = &dev->data->dev_conf;
1806 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1810 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1813 rte_eth_dev_info_get(portid, &dev_info);
1814 rxq_conf = dev_info.default_rxconf;
1815 rxq_conf.offloads = conf->rxmode.offloads;
1816 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1817 socketid, &rxq_conf,
1818 pktmbuf_pool[socketid]);
1820 rte_exit(EXIT_FAILURE,
1821 "rte_eth_rx_queue_setup: err=%d, "
1822 "port=%d\n", ret, portid);
1825 if (add_cb_parse_ptype(portid, queueid) < 0)
1826 rte_exit(EXIT_FAILURE,
1827 "Fail to add ptype cb\n");
1828 } else if (!check_ptype(portid))
1829 rte_exit(EXIT_FAILURE,
1830 "PMD can not provide needed ptypes\n");
1837 for (portid = 0; portid < nb_ports; portid++) {
1838 if ((enabled_port_mask & (1 << portid)) == 0) {
1842 ret = rte_eth_dev_start(portid);
1844 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1845 "port=%d\n", ret, portid);
1847 * If enabled, put device in promiscuous mode.
1848 * This allows IO forwarding mode to forward packets
1849 * to itself through 2 cross-connected ports of the
1853 rte_eth_promiscuous_enable(portid);
1854 /* initialize spinlock for each port */
1855 rte_spinlock_init(&(locks[portid]));
1858 check_all_ports_link_status(nb_ports, enabled_port_mask);
1860 /* launch per-lcore init on every lcore */
1861 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1862 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1863 if (rte_eal_wait_lcore(lcore_id) < 0)