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
53 /* around 100ms at 2 Ghz */
54 #define TIMER_RESOLUTION_CYCLES 200000000ULL
56 #define TIMER_NUMBER_PER_SECOND 10
58 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
59 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
61 #define APP_LOOKUP_EXACT_MATCH 0
62 #define APP_LOOKUP_LPM 1
63 #define DO_RFC_1812_CHECKS
65 #ifndef APP_LOOKUP_METHOD
66 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
69 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
71 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
74 #error "APP_LOOKUP_METHOD set to incorrect value"
78 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
79 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
80 #define IPv6_BYTES(addr) \
81 addr[0], addr[1], addr[2], addr[3], \
82 addr[4], addr[5], addr[6], addr[7], \
83 addr[8], addr[9], addr[10], addr[11],\
84 addr[12], addr[13],addr[14], addr[15]
87 #define MAX_JUMBO_PKT_LEN 9600
89 #define IPV6_ADDR_LEN 16
91 #define MEMPOOL_CACHE_SIZE 256
94 * This expression is used to calculate the number of mbufs needed depending on
95 * user input, taking into account memory for rx and tx hardware rings, cache
96 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
97 * NB_MBUF never goes below a minimum value of 8192.
100 #define NB_MBUF RTE_MAX ( \
101 (nb_ports*nb_rx_queue*nb_rxd + \
102 nb_ports*nb_lcores*MAX_PKT_BURST + \
103 nb_ports*n_tx_queue*nb_txd + \
104 nb_lcores*MEMPOOL_CACHE_SIZE), \
107 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
111 /* Configure how many packets ahead to prefetch, when reading packets */
112 #define PREFETCH_OFFSET 3
115 * Configurable number of RX/TX ring descriptors
117 #define RTE_TEST_RX_DESC_DEFAULT 512
118 #define RTE_TEST_TX_DESC_DEFAULT 512
119 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
120 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
122 /* ethernet addresses of ports */
123 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
125 /* ethernet addresses of ports */
126 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
128 /* mask of enabled ports */
129 static uint32_t enabled_port_mask = 0;
130 /* Ports set in promiscuous mode off by default. */
131 static int promiscuous_on = 0;
132 /* NUMA is enabled by default. */
133 static int numa_on = 1;
134 static int parse_ptype; /**< Parse packet type using rx callback, and */
135 /**< disabled by default */
137 enum freq_scale_hint_t
145 struct lcore_rx_queue {
148 enum freq_scale_hint_t freq_up_hint;
149 uint32_t zero_rx_packet_count;
151 } __rte_cache_aligned;
153 #define MAX_RX_QUEUE_PER_LCORE 16
154 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
155 #define MAX_RX_QUEUE_PER_PORT 128
157 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
160 #define MAX_LCORE_PARAMS 1024
161 struct lcore_params {
165 } __rte_cache_aligned;
167 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
168 static struct lcore_params lcore_params_array_default[] = {
180 static struct lcore_params * lcore_params = lcore_params_array_default;
181 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
182 sizeof(lcore_params_array_default[0]);
184 static struct rte_eth_conf port_conf = {
186 .mq_mode = ETH_MQ_RX_RSS,
187 .max_rx_pkt_len = ETHER_MAX_LEN,
189 .header_split = 0, /**< Header Split disabled */
190 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
191 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
192 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
193 .hw_strip_crc = 1, /**< CRC stripped by hardware */
198 .rss_hf = ETH_RSS_UDP,
202 .mq_mode = ETH_MQ_TX_NONE,
210 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
213 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
216 #include <rte_hash_crc.h>
217 #define DEFAULT_HASH_FUNC rte_hash_crc
219 #include <rte_jhash.h>
220 #define DEFAULT_HASH_FUNC rte_jhash
229 } __attribute__((__packed__));
232 uint8_t ip_dst[IPV6_ADDR_LEN];
233 uint8_t ip_src[IPV6_ADDR_LEN];
237 } __attribute__((__packed__));
239 struct ipv4_l3fwd_route {
240 struct ipv4_5tuple key;
244 struct ipv6_l3fwd_route {
245 struct ipv6_5tuple key;
249 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
250 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
251 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
252 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
253 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
256 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
259 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
260 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
261 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
262 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
268 typedef struct rte_hash lookup_struct_t;
269 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
270 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
272 #define L3FWD_HASH_ENTRIES 1024
274 #define IPV4_L3FWD_NUM_ROUTES \
275 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
277 #define IPV6_L3FWD_NUM_ROUTES \
278 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
280 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
281 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
284 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
285 struct ipv4_l3fwd_route {
291 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
292 {IPv4(1,1,1,0), 24, 0},
293 {IPv4(2,1,1,0), 24, 1},
294 {IPv4(3,1,1,0), 24, 2},
295 {IPv4(4,1,1,0), 24, 3},
296 {IPv4(5,1,1,0), 24, 4},
297 {IPv4(6,1,1,0), 24, 5},
298 {IPv4(7,1,1,0), 24, 6},
299 {IPv4(8,1,1,0), 24, 7},
302 #define IPV4_L3FWD_NUM_ROUTES \
303 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
305 #define IPV4_L3FWD_LPM_MAX_RULES 1024
307 typedef struct rte_lpm lookup_struct_t;
308 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
313 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
315 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
316 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
317 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
318 lookup_struct_t * ipv4_lookup_struct;
319 lookup_struct_t * ipv6_lookup_struct;
320 } __rte_cache_aligned;
323 /* total sleep time in ms since last frequency scaling down */
325 /* number of long sleep recently */
326 uint32_t nb_long_sleep;
327 /* freq. scaling up trend */
329 /* total packet processed recently */
330 uint64_t nb_rx_processed;
331 /* total iterations looped recently */
332 uint64_t nb_iteration_looped;
334 } __rte_cache_aligned;
336 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
337 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
338 static struct rte_timer power_timers[RTE_MAX_LCORE];
340 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
341 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
342 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
344 /* exit signal handler */
346 signal_exit_now(int sigtype)
349 unsigned int portid, nb_ports;
352 if (sigtype == SIGINT) {
353 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
354 if (rte_lcore_is_enabled(lcore_id) == 0)
357 /* init power management library */
358 ret = rte_power_exit(lcore_id);
360 rte_exit(EXIT_FAILURE, "Power management "
361 "library de-initialization failed on "
362 "core%u\n", lcore_id);
365 nb_ports = rte_eth_dev_count();
366 for (portid = 0; portid < nb_ports; portid++) {
367 if ((enabled_port_mask & (1 << portid)) == 0)
370 rte_eth_dev_stop(portid);
371 rte_eth_dev_close(portid);
375 rte_exit(EXIT_SUCCESS, "User forced exit\n");
378 /* Freqency scale down timer callback */
380 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
381 __attribute__((unused)) void *arg)
384 float sleep_time_ratio;
385 unsigned lcore_id = rte_lcore_id();
387 /* accumulate total execution time in us when callback is invoked */
388 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
389 (float)SCALING_PERIOD;
391 * check whether need to scale down frequency a step if it sleep a lot.
393 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
394 if (rte_power_freq_down)
395 rte_power_freq_down(lcore_id);
397 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
398 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
400 * scale down a step if average packet per iteration less
403 if (rte_power_freq_down)
404 rte_power_freq_down(lcore_id);
408 * initialize another timer according to current frequency to ensure
409 * timer interval is relatively fixed.
411 hz = rte_get_timer_hz();
412 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
413 SINGLE, lcore_id, power_timer_cb, NULL);
415 stats[lcore_id].nb_rx_processed = 0;
416 stats[lcore_id].nb_iteration_looped = 0;
418 stats[lcore_id].sleep_time = 0;
421 /* Enqueue a single packet, and send burst if queue is filled */
423 send_single_packet(struct rte_mbuf *m, uint16_t port)
426 struct lcore_conf *qconf;
428 lcore_id = rte_lcore_id();
429 qconf = &lcore_conf[lcore_id];
431 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
432 qconf->tx_buffer[port], m);
437 #ifdef DO_RFC_1812_CHECKS
439 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
441 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
443 * 1. The packet length reported by the Link Layer must be large
444 * enough to hold the minimum length legal IP datagram (20 bytes).
446 if (link_len < sizeof(struct ipv4_hdr))
449 /* 2. The IP checksum must be correct. */
450 /* this is checked in H/W */
453 * 3. The IP version number must be 4. If the version number is not 4
454 * then the packet may be another version of IP, such as IPng or
457 if (((pkt->version_ihl) >> 4) != 4)
460 * 4. The IP header length field must be large enough to hold the
461 * minimum length legal IP datagram (20 bytes = 5 words).
463 if ((pkt->version_ihl & 0xf) < 5)
467 * 5. The IP total length field must be large enough to hold the IP
468 * datagram header, whose length is specified in the IP header length
471 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
478 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
480 print_ipv4_key(struct ipv4_5tuple key)
482 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
483 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
484 key.port_dst, key.port_src, key.proto);
487 print_ipv6_key(struct ipv6_5tuple key)
489 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
490 "port dst = %d, port src = %d, proto = %d\n",
491 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
492 key.port_dst, key.port_src, key.proto);
495 static inline uint16_t
496 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
497 lookup_struct_t * ipv4_l3fwd_lookup_struct)
499 struct ipv4_5tuple key;
504 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
505 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
506 key.proto = ipv4_hdr->next_proto_id;
508 switch (ipv4_hdr->next_proto_id) {
510 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
511 sizeof(struct ipv4_hdr));
512 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
513 key.port_src = rte_be_to_cpu_16(tcp->src_port);
517 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
518 sizeof(struct ipv4_hdr));
519 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
520 key.port_src = rte_be_to_cpu_16(udp->src_port);
529 /* Find destination port */
530 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
531 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
534 static inline uint16_t
535 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
536 lookup_struct_t *ipv6_l3fwd_lookup_struct)
538 struct ipv6_5tuple key;
543 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
544 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
546 key.proto = ipv6_hdr->proto;
548 switch (ipv6_hdr->proto) {
550 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
551 sizeof(struct ipv6_hdr));
552 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
553 key.port_src = rte_be_to_cpu_16(tcp->src_port);
557 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
558 sizeof(struct ipv6_hdr));
559 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
560 key.port_src = rte_be_to_cpu_16(udp->src_port);
569 /* Find destination port */
570 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
571 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
575 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
576 static inline uint16_t
577 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
578 lookup_struct_t *ipv4_l3fwd_lookup_struct)
582 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
583 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
589 parse_ptype_one(struct rte_mbuf *m)
591 struct ether_hdr *eth_hdr;
592 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
595 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
596 ether_type = eth_hdr->ether_type;
597 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
598 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
599 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
600 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
602 m->packet_type = packet_type;
606 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
607 struct rte_mbuf *pkts[], uint16_t nb_pkts,
608 uint16_t max_pkts __rte_unused,
609 void *user_param __rte_unused)
613 for (i = 0; i < nb_pkts; ++i)
614 parse_ptype_one(pkts[i]);
620 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
622 printf("Port %d: softly parse packet type info\n", portid);
623 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
626 printf("Failed to add rx callback: port=%d\n", portid);
631 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
632 struct lcore_conf *qconf)
634 struct ether_hdr *eth_hdr;
635 struct ipv4_hdr *ipv4_hdr;
639 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
641 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
642 /* Handle IPv4 headers.*/
644 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
645 sizeof(struct ether_hdr));
647 #ifdef DO_RFC_1812_CHECKS
648 /* Check to make sure the packet is valid (RFC1812) */
649 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
655 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
656 qconf->ipv4_lookup_struct);
657 if (dst_port >= RTE_MAX_ETHPORTS ||
658 (enabled_port_mask & 1 << dst_port) == 0)
661 /* 02:00:00:00:00:xx */
662 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
663 *((uint64_t *)d_addr_bytes) =
664 0x000000000002 + ((uint64_t)dst_port << 40);
666 #ifdef DO_RFC_1812_CHECKS
667 /* Update time to live and header checksum */
668 --(ipv4_hdr->time_to_live);
669 ++(ipv4_hdr->hdr_checksum);
673 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
675 send_single_packet(m, dst_port);
676 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
677 /* Handle IPv6 headers.*/
678 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
679 struct ipv6_hdr *ipv6_hdr;
682 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
683 sizeof(struct ether_hdr));
685 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
686 qconf->ipv6_lookup_struct);
688 if (dst_port >= RTE_MAX_ETHPORTS ||
689 (enabled_port_mask & 1 << dst_port) == 0)
692 /* 02:00:00:00:00:xx */
693 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
694 *((uint64_t *)d_addr_bytes) =
695 0x000000000002 + ((uint64_t)dst_port << 40);
698 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
700 send_single_packet(m, dst_port);
702 /* We don't currently handle IPv6 packets in LPM mode. */
710 #define MINIMUM_SLEEP_TIME 1
711 #define SUSPEND_THRESHOLD 300
713 static inline uint32_t
714 power_idle_heuristic(uint32_t zero_rx_packet_count)
716 /* If zero count is less than 100, sleep 1us */
717 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
718 return MINIMUM_SLEEP_TIME;
719 /* If zero count is less than 1000, sleep 100 us which is the
720 minimum latency switching from C3/C6 to C0
723 return SUSPEND_THRESHOLD;
726 static inline enum freq_scale_hint_t
727 power_freq_scaleup_heuristic(unsigned lcore_id,
732 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
735 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
736 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
737 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
738 #define FREQ_UP_TREND1_ACC 1
739 #define FREQ_UP_TREND2_ACC 100
740 #define FREQ_UP_THRESHOLD 10000
742 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
743 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
744 stats[lcore_id].trend = 0;
746 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
747 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
748 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
749 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
750 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
751 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
753 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
754 stats[lcore_id].trend = 0;
762 * force polling thread sleep until one-shot rx interrupt triggers
771 sleep_until_rx_interrupt(int num)
773 struct rte_epoll_event event[num];
779 RTE_LOG(INFO, L3FWD_POWER,
780 "lcore %u sleeps until interrupt triggers\n",
783 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
784 for (i = 0; i < n; i++) {
785 data = event[i].epdata.data;
786 port_id = ((uintptr_t)data) >> CHAR_BIT;
787 queue_id = ((uintptr_t)data) &
788 RTE_LEN2MASK(CHAR_BIT, uint8_t);
789 rte_eth_dev_rx_intr_disable(port_id, queue_id);
790 RTE_LOG(INFO, L3FWD_POWER,
791 "lcore %u is waked up from rx interrupt on"
792 " port %d queue %d\n",
793 rte_lcore_id(), port_id, queue_id);
799 static void turn_on_intr(struct lcore_conf *qconf)
802 struct lcore_rx_queue *rx_queue;
806 for (i = 0; i < qconf->n_rx_queue; ++i) {
807 rx_queue = &(qconf->rx_queue_list[i]);
808 port_id = rx_queue->port_id;
809 queue_id = rx_queue->queue_id;
811 rte_spinlock_lock(&(locks[port_id]));
812 rte_eth_dev_rx_intr_enable(port_id, queue_id);
813 rte_spinlock_unlock(&(locks[port_id]));
817 static int event_register(struct lcore_conf *qconf)
819 struct lcore_rx_queue *rx_queue;
826 for (i = 0; i < qconf->n_rx_queue; ++i) {
827 rx_queue = &(qconf->rx_queue_list[i]);
828 portid = rx_queue->port_id;
829 queueid = rx_queue->queue_id;
830 data = portid << CHAR_BIT | queueid;
832 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
833 RTE_EPOLL_PER_THREAD,
835 (void *)((uintptr_t)data));
843 /* main processing loop */
845 main_loop(__attribute__((unused)) void *dummy)
847 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
849 uint64_t prev_tsc, diff_tsc, cur_tsc;
850 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
854 struct lcore_conf *qconf;
855 struct lcore_rx_queue *rx_queue;
856 enum freq_scale_hint_t lcore_scaleup_hint;
857 uint32_t lcore_rx_idle_count = 0;
858 uint32_t lcore_idle_hint = 0;
861 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
865 lcore_id = rte_lcore_id();
866 qconf = &lcore_conf[lcore_id];
868 if (qconf->n_rx_queue == 0) {
869 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
873 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
875 for (i = 0; i < qconf->n_rx_queue; i++) {
876 portid = qconf->rx_queue_list[i].port_id;
877 queueid = qconf->rx_queue_list[i].queue_id;
878 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
879 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
882 /* add into event wait list */
883 if (event_register(qconf) == 0)
886 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
889 stats[lcore_id].nb_iteration_looped++;
891 cur_tsc = rte_rdtsc();
892 cur_tsc_power = cur_tsc;
895 * TX burst queue drain
897 diff_tsc = cur_tsc - prev_tsc;
898 if (unlikely(diff_tsc > drain_tsc)) {
899 for (i = 0; i < qconf->n_tx_port; ++i) {
900 portid = qconf->tx_port_id[i];
901 rte_eth_tx_buffer_flush(portid,
902 qconf->tx_queue_id[portid],
903 qconf->tx_buffer[portid]);
908 diff_tsc_power = cur_tsc_power - prev_tsc_power;
909 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
911 prev_tsc_power = cur_tsc_power;
916 * Read packet from RX queues
918 lcore_scaleup_hint = FREQ_CURRENT;
919 lcore_rx_idle_count = 0;
920 for (i = 0; i < qconf->n_rx_queue; ++i) {
921 rx_queue = &(qconf->rx_queue_list[i]);
922 rx_queue->idle_hint = 0;
923 portid = rx_queue->port_id;
924 queueid = rx_queue->queue_id;
926 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
929 stats[lcore_id].nb_rx_processed += nb_rx;
930 if (unlikely(nb_rx == 0)) {
932 * no packet received from rx queue, try to
933 * sleep for a while forcing CPU enter deeper
936 rx_queue->zero_rx_packet_count++;
938 if (rx_queue->zero_rx_packet_count <=
942 rx_queue->idle_hint = power_idle_heuristic(\
943 rx_queue->zero_rx_packet_count);
944 lcore_rx_idle_count++;
946 rx_queue->zero_rx_packet_count = 0;
949 * do not scale up frequency immediately as
950 * user to kernel space communication is costly
951 * which might impact packet I/O for received
954 rx_queue->freq_up_hint =
955 power_freq_scaleup_heuristic(lcore_id,
959 /* Prefetch first packets */
960 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
961 rte_prefetch0(rte_pktmbuf_mtod(
962 pkts_burst[j], void *));
965 /* Prefetch and forward already prefetched packets */
966 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
967 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
968 j + PREFETCH_OFFSET], void *));
969 l3fwd_simple_forward(pkts_burst[j], portid,
973 /* Forward remaining prefetched packets */
974 for (; j < nb_rx; j++) {
975 l3fwd_simple_forward(pkts_burst[j], portid,
980 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
981 for (i = 1, lcore_scaleup_hint =
982 qconf->rx_queue_list[0].freq_up_hint;
983 i < qconf->n_rx_queue; ++i) {
984 rx_queue = &(qconf->rx_queue_list[i]);
985 if (rx_queue->freq_up_hint >
988 rx_queue->freq_up_hint;
991 if (lcore_scaleup_hint == FREQ_HIGHEST) {
992 if (rte_power_freq_max)
993 rte_power_freq_max(lcore_id);
994 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
995 if (rte_power_freq_up)
996 rte_power_freq_up(lcore_id);
1000 * All Rx queues empty in recent consecutive polls,
1001 * sleep in a conservative manner, meaning sleep as
1004 for (i = 1, lcore_idle_hint =
1005 qconf->rx_queue_list[0].idle_hint;
1006 i < qconf->n_rx_queue; ++i) {
1007 rx_queue = &(qconf->rx_queue_list[i]);
1008 if (rx_queue->idle_hint < lcore_idle_hint)
1009 lcore_idle_hint = rx_queue->idle_hint;
1012 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1014 * execute "pause" instruction to avoid context
1015 * switch which generally take hundred of
1016 * microseconds for short sleep.
1018 rte_delay_us(lcore_idle_hint);
1020 /* suspend until rx interrupt trigges */
1022 turn_on_intr(qconf);
1023 sleep_until_rx_interrupt(
1026 /* start receiving packets immediately */
1029 stats[lcore_id].sleep_time += lcore_idle_hint;
1035 check_lcore_params(void)
1037 uint8_t queue, lcore;
1041 for (i = 0; i < nb_lcore_params; ++i) {
1042 queue = lcore_params[i].queue_id;
1043 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1044 printf("invalid queue number: %hhu\n", queue);
1047 lcore = lcore_params[i].lcore_id;
1048 if (!rte_lcore_is_enabled(lcore)) {
1049 printf("error: lcore %hhu is not enabled in lcore "
1053 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1055 printf("warning: lcore %hhu is on socket %d with numa "
1056 "off\n", lcore, socketid);
1063 check_port_config(const unsigned nb_ports)
1068 for (i = 0; i < nb_lcore_params; ++i) {
1069 portid = lcore_params[i].port_id;
1070 if ((enabled_port_mask & (1 << portid)) == 0) {
1071 printf("port %u is not enabled in port mask\n",
1075 if (portid >= nb_ports) {
1076 printf("port %u is not present on the board\n",
1085 get_port_n_rx_queues(const uint16_t port)
1090 for (i = 0; i < nb_lcore_params; ++i) {
1091 if (lcore_params[i].port_id == port &&
1092 lcore_params[i].queue_id > queue)
1093 queue = lcore_params[i].queue_id;
1095 return (uint8_t)(++queue);
1099 init_lcore_rx_queues(void)
1101 uint16_t i, nb_rx_queue;
1104 for (i = 0; i < nb_lcore_params; ++i) {
1105 lcore = lcore_params[i].lcore_id;
1106 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1107 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1108 printf("error: too many queues (%u) for lcore: %u\n",
1109 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1112 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1113 lcore_params[i].port_id;
1114 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1115 lcore_params[i].queue_id;
1116 lcore_conf[lcore].n_rx_queue++;
1124 print_usage(const char *prgname)
1126 printf ("%s [EAL options] -- -p PORTMASK -P"
1127 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1128 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1129 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1130 " -P : enable promiscuous mode\n"
1131 " --config (port,queue,lcore): rx queues configuration\n"
1132 " --no-numa: optional, disable numa awareness\n"
1133 " --enable-jumbo: enable jumbo frame"
1134 " which max packet len is PKTLEN in decimal (64-9600)\n"
1135 " --parse-ptype: parse packet type by software\n",
1139 static int parse_max_pkt_len(const char *pktlen)
1144 /* parse decimal string */
1145 len = strtoul(pktlen, &end, 10);
1146 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1156 parse_portmask(const char *portmask)
1161 /* parse hexadecimal string */
1162 pm = strtoul(portmask, &end, 16);
1163 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1173 parse_config(const char *q_arg)
1176 const char *p, *p0 = q_arg;
1184 unsigned long int_fld[_NUM_FLD];
1185 char *str_fld[_NUM_FLD];
1189 nb_lcore_params = 0;
1191 while ((p = strchr(p0,'(')) != NULL) {
1193 if((p0 = strchr(p,')')) == NULL)
1197 if(size >= sizeof(s))
1200 snprintf(s, sizeof(s), "%.*s", size, p);
1201 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1204 for (i = 0; i < _NUM_FLD; i++){
1206 int_fld[i] = strtoul(str_fld[i], &end, 0);
1207 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1211 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1212 printf("exceeded max number of lcore params: %hu\n",
1216 lcore_params_array[nb_lcore_params].port_id =
1217 (uint8_t)int_fld[FLD_PORT];
1218 lcore_params_array[nb_lcore_params].queue_id =
1219 (uint8_t)int_fld[FLD_QUEUE];
1220 lcore_params_array[nb_lcore_params].lcore_id =
1221 (uint8_t)int_fld[FLD_LCORE];
1224 lcore_params = lcore_params_array;
1229 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1231 /* Parse the argument given in the command line of the application */
1233 parse_args(int argc, char **argv)
1238 char *prgname = argv[0];
1239 static struct option lgopts[] = {
1240 {"config", 1, 0, 0},
1241 {"no-numa", 0, 0, 0},
1242 {"enable-jumbo", 0, 0, 0},
1243 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1249 while ((opt = getopt_long(argc, argvopt, "p:P",
1250 lgopts, &option_index)) != EOF) {
1255 enabled_port_mask = parse_portmask(optarg);
1256 if (enabled_port_mask == 0) {
1257 printf("invalid portmask\n");
1258 print_usage(prgname);
1263 printf("Promiscuous mode selected\n");
1269 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1270 ret = parse_config(optarg);
1272 printf("invalid config\n");
1273 print_usage(prgname);
1278 if (!strncmp(lgopts[option_index].name,
1280 printf("numa is disabled \n");
1284 if (!strncmp(lgopts[option_index].name,
1285 "enable-jumbo", 12)) {
1286 struct option lenopts =
1287 {"max-pkt-len", required_argument, \
1290 printf("jumbo frame is enabled \n");
1291 port_conf.rxmode.jumbo_frame = 1;
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;
1628 uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
1630 /* catch SIGINT and restore cpufreq governor to ondemand */
1631 signal(SIGINT, signal_exit_now);
1634 ret = rte_eal_init(argc, argv);
1636 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1640 /* init RTE timer library to be used late */
1641 rte_timer_subsystem_init();
1643 /* parse application arguments (after the EAL ones) */
1644 ret = parse_args(argc, argv);
1646 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1648 if (check_lcore_params() < 0)
1649 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1651 ret = init_lcore_rx_queues();
1653 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1655 nb_ports = rte_eth_dev_count();
1657 if (check_port_config(nb_ports) < 0)
1658 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1660 nb_lcores = rte_lcore_count();
1662 /* initialize all ports */
1663 for (portid = 0; portid < nb_ports; portid++) {
1664 /* skip ports that are not enabled */
1665 if ((enabled_port_mask & (1 << portid)) == 0) {
1666 printf("\nSkipping disabled port %d\n", portid);
1671 printf("Initializing port %d ... ", portid );
1674 rte_eth_dev_info_get(portid, &dev_info);
1675 dev_rxq_num = dev_info.max_rx_queues;
1676 dev_txq_num = dev_info.max_tx_queues;
1678 nb_rx_queue = get_port_n_rx_queues(portid);
1679 if (nb_rx_queue > dev_rxq_num)
1680 rte_exit(EXIT_FAILURE,
1681 "Cannot configure not existed rxq: "
1682 "port=%d\n", portid);
1684 n_tx_queue = nb_lcores;
1685 if (n_tx_queue > dev_txq_num)
1686 n_tx_queue = dev_txq_num;
1687 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1688 nb_rx_queue, (unsigned)n_tx_queue );
1689 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1690 if (nb_rx_queue == 0)
1691 port_conf.intr_conf.rxq = 0;
1692 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1693 (uint16_t)n_tx_queue, &port_conf);
1694 /* Revert to original value */
1695 port_conf.intr_conf.rxq = org_rxq_intr;
1697 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1698 "err=%d, port=%d\n", ret, portid);
1700 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1703 rte_exit(EXIT_FAILURE,
1704 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1707 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1708 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1712 ret = init_mem(NB_MBUF);
1714 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1716 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1717 if (rte_lcore_is_enabled(lcore_id) == 0)
1720 /* Initialize TX buffers */
1721 qconf = &lcore_conf[lcore_id];
1722 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1723 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1724 rte_eth_dev_socket_id(portid));
1725 if (qconf->tx_buffer[portid] == NULL)
1726 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1729 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1732 /* init one TX queue per couple (lcore,port) */
1734 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1735 if (rte_lcore_is_enabled(lcore_id) == 0)
1738 if (queueid >= dev_txq_num)
1743 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1747 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1750 rte_eth_dev_info_get(portid, &dev_info);
1751 txconf = &dev_info.default_txconf;
1752 if (port_conf.rxmode.jumbo_frame)
1753 txconf->txq_flags = 0;
1754 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1757 rte_exit(EXIT_FAILURE,
1758 "rte_eth_tx_queue_setup: err=%d, "
1759 "port=%d\n", ret, portid);
1761 qconf = &lcore_conf[lcore_id];
1762 qconf->tx_queue_id[portid] = queueid;
1765 qconf->tx_port_id[qconf->n_tx_port] = portid;
1771 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1772 if (rte_lcore_is_enabled(lcore_id) == 0)
1775 /* init power management library */
1776 ret = rte_power_init(lcore_id);
1779 "Library initialization failed on core %u\n", lcore_id);
1781 /* init timer structures for each enabled lcore */
1782 rte_timer_init(&power_timers[lcore_id]);
1783 hz = rte_get_timer_hz();
1784 rte_timer_reset(&power_timers[lcore_id],
1785 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1786 power_timer_cb, NULL);
1788 qconf = &lcore_conf[lcore_id];
1789 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1791 /* init RX queues */
1792 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1793 portid = qconf->rx_queue_list[queue].port_id;
1794 queueid = qconf->rx_queue_list[queue].queue_id;
1798 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1802 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1805 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1807 pktmbuf_pool[socketid]);
1809 rte_exit(EXIT_FAILURE,
1810 "rte_eth_rx_queue_setup: err=%d, "
1811 "port=%d\n", ret, portid);
1814 if (add_cb_parse_ptype(portid, queueid) < 0)
1815 rte_exit(EXIT_FAILURE,
1816 "Fail to add ptype cb\n");
1817 } else if (!check_ptype(portid))
1818 rte_exit(EXIT_FAILURE,
1819 "PMD can not provide needed ptypes\n");
1826 for (portid = 0; portid < nb_ports; portid++) {
1827 if ((enabled_port_mask & (1 << portid)) == 0) {
1831 ret = rte_eth_dev_start(portid);
1833 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1834 "port=%d\n", ret, portid);
1836 * If enabled, put device in promiscuous mode.
1837 * This allows IO forwarding mode to forward packets
1838 * to itself through 2 cross-connected ports of the
1842 rte_eth_promiscuous_enable(portid);
1843 /* initialize spinlock for each port */
1844 rte_spinlock_init(&(locks[portid]));
1847 check_all_ports_link_status(nb_ports, enabled_port_mask);
1849 /* launch per-lcore init on every lcore */
1850 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1851 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1852 if (rte_eal_wait_lcore(lcore_id) < 0)