4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 #include <sys/types.h>
40 #include <sys/queue.h>
47 #include <rte_common.h>
48 #include <rte_byteorder.h>
50 #include <rte_malloc.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
55 #include <rte_launch.h>
56 #include <rte_atomic.h>
57 #include <rte_cycles.h>
58 #include <rte_prefetch.h>
59 #include <rte_lcore.h>
60 #include <rte_per_lcore.h>
61 #include <rte_branch_prediction.h>
62 #include <rte_interrupts.h>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
68 #include <rte_mempool.h>
73 #include <rte_string_fns.h>
74 #include <rte_timer.h>
75 #include <rte_power.h>
76 #include <rte_spinlock.h>
78 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
80 #define MAX_PKT_BURST 32
82 #define MIN_ZERO_POLL_COUNT 10
84 /* around 100ms at 2 Ghz */
85 #define TIMER_RESOLUTION_CYCLES 200000000ULL
87 #define TIMER_NUMBER_PER_SECOND 10
89 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
90 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
92 #define APP_LOOKUP_EXACT_MATCH 0
93 #define APP_LOOKUP_LPM 1
94 #define DO_RFC_1812_CHECKS
96 #ifndef APP_LOOKUP_METHOD
97 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
100 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
101 #include <rte_hash.h>
102 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
105 #error "APP_LOOKUP_METHOD set to incorrect value"
109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
110 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
111 #define IPv6_BYTES(addr) \
112 addr[0], addr[1], addr[2], addr[3], \
113 addr[4], addr[5], addr[6], addr[7], \
114 addr[8], addr[9], addr[10], addr[11],\
115 addr[12], addr[13],addr[14], addr[15]
118 #define MAX_JUMBO_PKT_LEN 9600
120 #define IPV6_ADDR_LEN 16
122 #define MEMPOOL_CACHE_SIZE 256
125 * This expression is used to calculate the number of mbufs needed depending on
126 * user input, taking into account memory for rx and tx hardware rings, cache
127 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
128 * NB_MBUF never goes below a minimum value of 8192.
131 #define NB_MBUF RTE_MAX ( \
132 (nb_ports*nb_rx_queue*nb_rxd + \
133 nb_ports*nb_lcores*MAX_PKT_BURST + \
134 nb_ports*n_tx_queue*nb_txd + \
135 nb_lcores*MEMPOOL_CACHE_SIZE), \
138 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
142 /* Configure how many packets ahead to prefetch, when reading packets */
143 #define PREFETCH_OFFSET 3
146 * Configurable number of RX/TX ring descriptors
148 #define RTE_TEST_RX_DESC_DEFAULT 512
149 #define RTE_TEST_TX_DESC_DEFAULT 512
150 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
151 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
153 /* ethernet addresses of ports */
154 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
156 /* ethernet addresses of ports */
157 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
159 /* mask of enabled ports */
160 static uint32_t enabled_port_mask = 0;
161 /* Ports set in promiscuous mode off by default. */
162 static int promiscuous_on = 0;
163 /* NUMA is enabled by default. */
164 static int numa_on = 1;
165 static int parse_ptype; /**< Parse packet type using rx callback, and */
166 /**< disabled by default */
168 enum freq_scale_hint_t
176 struct lcore_rx_queue {
179 enum freq_scale_hint_t freq_up_hint;
180 uint32_t zero_rx_packet_count;
182 } __rte_cache_aligned;
184 #define MAX_RX_QUEUE_PER_LCORE 16
185 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
186 #define MAX_RX_QUEUE_PER_PORT 128
188 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
191 #define MAX_LCORE_PARAMS 1024
192 struct lcore_params {
196 } __rte_cache_aligned;
198 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
199 static struct lcore_params lcore_params_array_default[] = {
211 static struct lcore_params * lcore_params = lcore_params_array_default;
212 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
213 sizeof(lcore_params_array_default[0]);
215 static struct rte_eth_conf port_conf = {
217 .mq_mode = ETH_MQ_RX_RSS,
218 .max_rx_pkt_len = ETHER_MAX_LEN,
220 .header_split = 0, /**< Header Split disabled */
221 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
222 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
223 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
224 .hw_strip_crc = 1, /**< CRC stripped by hardware */
229 .rss_hf = ETH_RSS_UDP,
233 .mq_mode = ETH_MQ_TX_NONE,
241 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
244 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
247 #include <rte_hash_crc.h>
248 #define DEFAULT_HASH_FUNC rte_hash_crc
250 #include <rte_jhash.h>
251 #define DEFAULT_HASH_FUNC rte_jhash
260 } __attribute__((__packed__));
263 uint8_t ip_dst[IPV6_ADDR_LEN];
264 uint8_t ip_src[IPV6_ADDR_LEN];
268 } __attribute__((__packed__));
270 struct ipv4_l3fwd_route {
271 struct ipv4_5tuple key;
275 struct ipv6_l3fwd_route {
276 struct ipv6_5tuple key;
280 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
281 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
282 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
283 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
284 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
287 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
290 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
291 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
292 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
293 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
299 typedef struct rte_hash lookup_struct_t;
300 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
301 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
303 #define L3FWD_HASH_ENTRIES 1024
305 #define IPV4_L3FWD_NUM_ROUTES \
306 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
308 #define IPV6_L3FWD_NUM_ROUTES \
309 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
311 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
312 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
315 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
316 struct ipv4_l3fwd_route {
322 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
323 {IPv4(1,1,1,0), 24, 0},
324 {IPv4(2,1,1,0), 24, 1},
325 {IPv4(3,1,1,0), 24, 2},
326 {IPv4(4,1,1,0), 24, 3},
327 {IPv4(5,1,1,0), 24, 4},
328 {IPv4(6,1,1,0), 24, 5},
329 {IPv4(7,1,1,0), 24, 6},
330 {IPv4(8,1,1,0), 24, 7},
333 #define IPV4_L3FWD_NUM_ROUTES \
334 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
336 #define IPV4_L3FWD_LPM_MAX_RULES 1024
338 typedef struct rte_lpm lookup_struct_t;
339 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
344 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
346 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
347 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
348 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
349 lookup_struct_t * ipv4_lookup_struct;
350 lookup_struct_t * ipv6_lookup_struct;
351 } __rte_cache_aligned;
354 /* total sleep time in ms since last frequency scaling down */
356 /* number of long sleep recently */
357 uint32_t nb_long_sleep;
358 /* freq. scaling up trend */
360 /* total packet processed recently */
361 uint64_t nb_rx_processed;
362 /* total iterations looped recently */
363 uint64_t nb_iteration_looped;
365 } __rte_cache_aligned;
367 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
368 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
369 static struct rte_timer power_timers[RTE_MAX_LCORE];
371 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
372 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
373 unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
375 /* exit signal handler */
377 signal_exit_now(int sigtype)
380 unsigned int portid, nb_ports;
383 if (sigtype == SIGINT) {
384 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
385 if (rte_lcore_is_enabled(lcore_id) == 0)
388 /* init power management library */
389 ret = rte_power_exit(lcore_id);
391 rte_exit(EXIT_FAILURE, "Power management "
392 "library de-initialization failed on "
393 "core%u\n", lcore_id);
396 nb_ports = rte_eth_dev_count();
397 for (portid = 0; portid < nb_ports; portid++) {
398 if ((enabled_port_mask & (1 << portid)) == 0)
401 rte_eth_dev_stop(portid);
402 rte_eth_dev_close(portid);
406 rte_exit(EXIT_SUCCESS, "User forced exit\n");
409 /* Freqency scale down timer callback */
411 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
412 __attribute__((unused)) void *arg)
415 float sleep_time_ratio;
416 unsigned lcore_id = rte_lcore_id();
418 /* accumulate total execution time in us when callback is invoked */
419 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
420 (float)SCALING_PERIOD;
422 * check whether need to scale down frequency a step if it sleep a lot.
424 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
425 if (rte_power_freq_down)
426 rte_power_freq_down(lcore_id);
428 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
429 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
431 * scale down a step if average packet per iteration less
434 if (rte_power_freq_down)
435 rte_power_freq_down(lcore_id);
439 * initialize another timer according to current frequency to ensure
440 * timer interval is relatively fixed.
442 hz = rte_get_timer_hz();
443 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
444 SINGLE, lcore_id, power_timer_cb, NULL);
446 stats[lcore_id].nb_rx_processed = 0;
447 stats[lcore_id].nb_iteration_looped = 0;
449 stats[lcore_id].sleep_time = 0;
452 /* Enqueue a single packet, and send burst if queue is filled */
454 send_single_packet(struct rte_mbuf *m, uint8_t port)
457 struct lcore_conf *qconf;
459 lcore_id = rte_lcore_id();
460 qconf = &lcore_conf[lcore_id];
462 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
463 qconf->tx_buffer[port], m);
468 #ifdef DO_RFC_1812_CHECKS
470 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
472 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
474 * 1. The packet length reported by the Link Layer must be large
475 * enough to hold the minimum length legal IP datagram (20 bytes).
477 if (link_len < sizeof(struct ipv4_hdr))
480 /* 2. The IP checksum must be correct. */
481 /* this is checked in H/W */
484 * 3. The IP version number must be 4. If the version number is not 4
485 * then the packet may be another version of IP, such as IPng or
488 if (((pkt->version_ihl) >> 4) != 4)
491 * 4. The IP header length field must be large enough to hold the
492 * minimum length legal IP datagram (20 bytes = 5 words).
494 if ((pkt->version_ihl & 0xf) < 5)
498 * 5. The IP total length field must be large enough to hold the IP
499 * datagram header, whose length is specified in the IP header length
502 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
509 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
511 print_ipv4_key(struct ipv4_5tuple key)
513 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
514 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
515 key.port_dst, key.port_src, key.proto);
518 print_ipv6_key(struct ipv6_5tuple key)
520 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
521 "port dst = %d, port src = %d, proto = %d\n",
522 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
523 key.port_dst, key.port_src, key.proto);
526 static inline uint8_t
527 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
528 lookup_struct_t * ipv4_l3fwd_lookup_struct)
530 struct ipv4_5tuple key;
535 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
536 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
537 key.proto = ipv4_hdr->next_proto_id;
539 switch (ipv4_hdr->next_proto_id) {
541 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
542 sizeof(struct ipv4_hdr));
543 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
544 key.port_src = rte_be_to_cpu_16(tcp->src_port);
548 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
549 sizeof(struct ipv4_hdr));
550 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
551 key.port_src = rte_be_to_cpu_16(udp->src_port);
560 /* Find destination port */
561 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
562 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
565 static inline uint8_t
566 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
567 lookup_struct_t *ipv6_l3fwd_lookup_struct)
569 struct ipv6_5tuple key;
574 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
575 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
577 key.proto = ipv6_hdr->proto;
579 switch (ipv6_hdr->proto) {
581 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
582 sizeof(struct ipv6_hdr));
583 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
584 key.port_src = rte_be_to_cpu_16(tcp->src_port);
588 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
589 sizeof(struct ipv6_hdr));
590 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
591 key.port_src = rte_be_to_cpu_16(udp->src_port);
600 /* Find destination port */
601 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
602 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
606 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
607 static inline uint8_t
608 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
609 lookup_struct_t *ipv4_l3fwd_lookup_struct)
613 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
614 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
620 parse_ptype_one(struct rte_mbuf *m)
622 struct ether_hdr *eth_hdr;
623 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
626 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
627 ether_type = eth_hdr->ether_type;
628 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
629 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
630 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
631 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
633 m->packet_type = packet_type;
637 cb_parse_ptype(uint8_t port __rte_unused, uint16_t queue __rte_unused,
638 struct rte_mbuf *pkts[], uint16_t nb_pkts,
639 uint16_t max_pkts __rte_unused,
640 void *user_param __rte_unused)
644 for (i = 0; i < nb_pkts; ++i)
645 parse_ptype_one(pkts[i]);
651 add_cb_parse_ptype(uint8_t portid, uint16_t queueid)
653 printf("Port %d: softly parse packet type info\n", portid);
654 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
657 printf("Failed to add rx callback: port=%d\n", portid);
662 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
663 struct lcore_conf *qconf)
665 struct ether_hdr *eth_hdr;
666 struct ipv4_hdr *ipv4_hdr;
670 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
672 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
673 /* Handle IPv4 headers.*/
675 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
676 sizeof(struct ether_hdr));
678 #ifdef DO_RFC_1812_CHECKS
679 /* Check to make sure the packet is valid (RFC1812) */
680 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
686 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
687 qconf->ipv4_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);
697 #ifdef DO_RFC_1812_CHECKS
698 /* Update time to live and header checksum */
699 --(ipv4_hdr->time_to_live);
700 ++(ipv4_hdr->hdr_checksum);
704 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
706 send_single_packet(m, dst_port);
707 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
708 /* Handle IPv6 headers.*/
709 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
710 struct ipv6_hdr *ipv6_hdr;
713 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
714 sizeof(struct ether_hdr));
716 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
717 qconf->ipv6_lookup_struct);
719 if (dst_port >= RTE_MAX_ETHPORTS ||
720 (enabled_port_mask & 1 << dst_port) == 0)
723 /* 02:00:00:00:00:xx */
724 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
725 *((uint64_t *)d_addr_bytes) =
726 0x000000000002 + ((uint64_t)dst_port << 40);
729 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
731 send_single_packet(m, dst_port);
733 /* We don't currently handle IPv6 packets in LPM mode. */
741 #define MINIMUM_SLEEP_TIME 1
742 #define SUSPEND_THRESHOLD 300
744 static inline uint32_t
745 power_idle_heuristic(uint32_t zero_rx_packet_count)
747 /* If zero count is less than 100, sleep 1us */
748 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
749 return MINIMUM_SLEEP_TIME;
750 /* If zero count is less than 1000, sleep 100 us which is the
751 minimum latency switching from C3/C6 to C0
754 return SUSPEND_THRESHOLD;
759 static inline enum freq_scale_hint_t
760 power_freq_scaleup_heuristic(unsigned lcore_id,
765 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
768 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
769 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
770 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
771 #define FREQ_UP_TREND1_ACC 1
772 #define FREQ_UP_TREND2_ACC 100
773 #define FREQ_UP_THRESHOLD 10000
775 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
776 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
777 stats[lcore_id].trend = 0;
779 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
780 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
781 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
782 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
783 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
784 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
786 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
787 stats[lcore_id].trend = 0;
795 * force polling thread sleep until one-shot rx interrupt triggers
804 sleep_until_rx_interrupt(int num)
806 struct rte_epoll_event event[num];
808 uint8_t port_id, queue_id;
811 RTE_LOG(INFO, L3FWD_POWER,
812 "lcore %u sleeps until interrupt triggers\n",
815 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
816 for (i = 0; i < n; i++) {
817 data = event[i].epdata.data;
818 port_id = ((uintptr_t)data) >> CHAR_BIT;
819 queue_id = ((uintptr_t)data) &
820 RTE_LEN2MASK(CHAR_BIT, uint8_t);
821 rte_eth_dev_rx_intr_disable(port_id, queue_id);
822 RTE_LOG(INFO, L3FWD_POWER,
823 "lcore %u is waked up from rx interrupt on"
824 " port %d queue %d\n",
825 rte_lcore_id(), port_id, queue_id);
831 static void turn_on_intr(struct lcore_conf *qconf)
834 struct lcore_rx_queue *rx_queue;
835 uint8_t port_id, queue_id;
837 for (i = 0; i < qconf->n_rx_queue; ++i) {
838 rx_queue = &(qconf->rx_queue_list[i]);
839 port_id = rx_queue->port_id;
840 queue_id = rx_queue->queue_id;
842 rte_spinlock_lock(&(locks[port_id]));
843 rte_eth_dev_rx_intr_enable(port_id, queue_id);
844 rte_spinlock_unlock(&(locks[port_id]));
848 static int event_register(struct lcore_conf *qconf)
850 struct lcore_rx_queue *rx_queue;
851 uint8_t portid, queueid;
856 for (i = 0; i < qconf->n_rx_queue; ++i) {
857 rx_queue = &(qconf->rx_queue_list[i]);
858 portid = rx_queue->port_id;
859 queueid = rx_queue->queue_id;
860 data = portid << CHAR_BIT | queueid;
862 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
863 RTE_EPOLL_PER_THREAD,
865 (void *)((uintptr_t)data));
873 /* main processing loop */
875 main_loop(__attribute__((unused)) void *dummy)
877 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
879 uint64_t prev_tsc, diff_tsc, cur_tsc;
880 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
882 uint8_t portid, queueid;
883 struct lcore_conf *qconf;
884 struct lcore_rx_queue *rx_queue;
885 enum freq_scale_hint_t lcore_scaleup_hint;
886 uint32_t lcore_rx_idle_count = 0;
887 uint32_t lcore_idle_hint = 0;
890 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
894 lcore_id = rte_lcore_id();
895 qconf = &lcore_conf[lcore_id];
897 if (qconf->n_rx_queue == 0) {
898 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
902 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
904 for (i = 0; i < qconf->n_rx_queue; i++) {
905 portid = qconf->rx_queue_list[i].port_id;
906 queueid = qconf->rx_queue_list[i].queue_id;
907 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
908 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
911 /* add into event wait list */
912 if (event_register(qconf) == 0)
915 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
918 stats[lcore_id].nb_iteration_looped++;
920 cur_tsc = rte_rdtsc();
921 cur_tsc_power = cur_tsc;
924 * TX burst queue drain
926 diff_tsc = cur_tsc - prev_tsc;
927 if (unlikely(diff_tsc > drain_tsc)) {
928 for (i = 0; i < qconf->n_tx_port; ++i) {
929 portid = qconf->tx_port_id[i];
930 rte_eth_tx_buffer_flush(portid,
931 qconf->tx_queue_id[portid],
932 qconf->tx_buffer[portid]);
937 diff_tsc_power = cur_tsc_power - prev_tsc_power;
938 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
940 prev_tsc_power = cur_tsc_power;
945 * Read packet from RX queues
947 lcore_scaleup_hint = FREQ_CURRENT;
948 lcore_rx_idle_count = 0;
949 for (i = 0; i < qconf->n_rx_queue; ++i) {
950 rx_queue = &(qconf->rx_queue_list[i]);
951 rx_queue->idle_hint = 0;
952 portid = rx_queue->port_id;
953 queueid = rx_queue->queue_id;
955 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
958 stats[lcore_id].nb_rx_processed += nb_rx;
959 if (unlikely(nb_rx == 0)) {
961 * no packet received from rx queue, try to
962 * sleep for a while forcing CPU enter deeper
965 rx_queue->zero_rx_packet_count++;
967 if (rx_queue->zero_rx_packet_count <=
971 rx_queue->idle_hint = power_idle_heuristic(\
972 rx_queue->zero_rx_packet_count);
973 lcore_rx_idle_count++;
975 rx_queue->zero_rx_packet_count = 0;
978 * do not scale up frequency immediately as
979 * user to kernel space communication is costly
980 * which might impact packet I/O for received
983 rx_queue->freq_up_hint =
984 power_freq_scaleup_heuristic(lcore_id,
988 /* Prefetch first packets */
989 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
990 rte_prefetch0(rte_pktmbuf_mtod(
991 pkts_burst[j], void *));
994 /* Prefetch and forward already prefetched packets */
995 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
996 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
997 j + PREFETCH_OFFSET], void *));
998 l3fwd_simple_forward(pkts_burst[j], portid,
1002 /* Forward remaining prefetched packets */
1003 for (; j < nb_rx; j++) {
1004 l3fwd_simple_forward(pkts_burst[j], portid,
1009 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1010 for (i = 1, lcore_scaleup_hint =
1011 qconf->rx_queue_list[0].freq_up_hint;
1012 i < qconf->n_rx_queue; ++i) {
1013 rx_queue = &(qconf->rx_queue_list[i]);
1014 if (rx_queue->freq_up_hint >
1016 lcore_scaleup_hint =
1017 rx_queue->freq_up_hint;
1020 if (lcore_scaleup_hint == FREQ_HIGHEST) {
1021 if (rte_power_freq_max)
1022 rte_power_freq_max(lcore_id);
1023 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1024 if (rte_power_freq_up)
1025 rte_power_freq_up(lcore_id);
1029 * All Rx queues empty in recent consecutive polls,
1030 * sleep in a conservative manner, meaning sleep as
1033 for (i = 1, lcore_idle_hint =
1034 qconf->rx_queue_list[0].idle_hint;
1035 i < qconf->n_rx_queue; ++i) {
1036 rx_queue = &(qconf->rx_queue_list[i]);
1037 if (rx_queue->idle_hint < lcore_idle_hint)
1038 lcore_idle_hint = rx_queue->idle_hint;
1041 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1043 * execute "pause" instruction to avoid context
1044 * switch which generally take hundred of
1045 * microseconds for short sleep.
1047 rte_delay_us(lcore_idle_hint);
1049 /* suspend until rx interrupt trigges */
1051 turn_on_intr(qconf);
1052 sleep_until_rx_interrupt(
1055 /* start receiving packets immediately */
1058 stats[lcore_id].sleep_time += lcore_idle_hint;
1064 check_lcore_params(void)
1066 uint8_t queue, lcore;
1070 for (i = 0; i < nb_lcore_params; ++i) {
1071 queue = lcore_params[i].queue_id;
1072 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1073 printf("invalid queue number: %hhu\n", queue);
1076 lcore = lcore_params[i].lcore_id;
1077 if (!rte_lcore_is_enabled(lcore)) {
1078 printf("error: lcore %hhu is not enabled in lcore "
1082 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1084 printf("warning: lcore %hhu is on socket %d with numa "
1085 "off\n", lcore, socketid);
1092 check_port_config(const unsigned nb_ports)
1097 for (i = 0; i < nb_lcore_params; ++i) {
1098 portid = lcore_params[i].port_id;
1099 if ((enabled_port_mask & (1 << portid)) == 0) {
1100 printf("port %u is not enabled in port mask\n",
1104 if (portid >= nb_ports) {
1105 printf("port %u is not present on the board\n",
1114 get_port_n_rx_queues(const uint8_t port)
1119 for (i = 0; i < nb_lcore_params; ++i) {
1120 if (lcore_params[i].port_id == port &&
1121 lcore_params[i].queue_id > queue)
1122 queue = lcore_params[i].queue_id;
1124 return (uint8_t)(++queue);
1128 init_lcore_rx_queues(void)
1130 uint16_t i, nb_rx_queue;
1133 for (i = 0; i < nb_lcore_params; ++i) {
1134 lcore = lcore_params[i].lcore_id;
1135 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1136 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1137 printf("error: too many queues (%u) for lcore: %u\n",
1138 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1141 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1142 lcore_params[i].port_id;
1143 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1144 lcore_params[i].queue_id;
1145 lcore_conf[lcore].n_rx_queue++;
1153 print_usage(const char *prgname)
1155 printf ("%s [EAL options] -- -p PORTMASK -P"
1156 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1157 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1158 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1159 " -P : enable promiscuous mode\n"
1160 " --config (port,queue,lcore): rx queues configuration\n"
1161 " --no-numa: optional, disable numa awareness\n"
1162 " --enable-jumbo: enable jumbo frame"
1163 " which max packet len is PKTLEN in decimal (64-9600)\n"
1164 " --parse-ptype: parse packet type by software\n",
1168 static int parse_max_pkt_len(const char *pktlen)
1173 /* parse decimal string */
1174 len = strtoul(pktlen, &end, 10);
1175 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1185 parse_portmask(const char *portmask)
1190 /* parse hexadecimal string */
1191 pm = strtoul(portmask, &end, 16);
1192 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1202 parse_config(const char *q_arg)
1205 const char *p, *p0 = q_arg;
1213 unsigned long int_fld[_NUM_FLD];
1214 char *str_fld[_NUM_FLD];
1218 nb_lcore_params = 0;
1220 while ((p = strchr(p0,'(')) != NULL) {
1222 if((p0 = strchr(p,')')) == NULL)
1226 if(size >= sizeof(s))
1229 snprintf(s, sizeof(s), "%.*s", size, p);
1230 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1233 for (i = 0; i < _NUM_FLD; i++){
1235 int_fld[i] = strtoul(str_fld[i], &end, 0);
1236 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1240 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1241 printf("exceeded max number of lcore params: %hu\n",
1245 lcore_params_array[nb_lcore_params].port_id =
1246 (uint8_t)int_fld[FLD_PORT];
1247 lcore_params_array[nb_lcore_params].queue_id =
1248 (uint8_t)int_fld[FLD_QUEUE];
1249 lcore_params_array[nb_lcore_params].lcore_id =
1250 (uint8_t)int_fld[FLD_LCORE];
1253 lcore_params = lcore_params_array;
1258 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1260 /* Parse the argument given in the command line of the application */
1262 parse_args(int argc, char **argv)
1267 char *prgname = argv[0];
1268 static struct option lgopts[] = {
1269 {"config", 1, 0, 0},
1270 {"no-numa", 0, 0, 0},
1271 {"enable-jumbo", 0, 0, 0},
1272 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1278 while ((opt = getopt_long(argc, argvopt, "p:P",
1279 lgopts, &option_index)) != EOF) {
1284 enabled_port_mask = parse_portmask(optarg);
1285 if (enabled_port_mask == 0) {
1286 printf("invalid portmask\n");
1287 print_usage(prgname);
1292 printf("Promiscuous mode selected\n");
1298 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1299 ret = parse_config(optarg);
1301 printf("invalid config\n");
1302 print_usage(prgname);
1307 if (!strncmp(lgopts[option_index].name,
1309 printf("numa is disabled \n");
1313 if (!strncmp(lgopts[option_index].name,
1314 "enable-jumbo", 12)) {
1315 struct option lenopts =
1316 {"max-pkt-len", required_argument, \
1319 printf("jumbo frame is enabled \n");
1320 port_conf.rxmode.jumbo_frame = 1;
1323 * if no max-pkt-len set, use the default value
1326 if (0 == getopt_long(argc, argvopt, "",
1327 &lenopts, &option_index)) {
1328 ret = parse_max_pkt_len(optarg);
1330 (ret > MAX_JUMBO_PKT_LEN)){
1331 printf("invalid packet "
1333 print_usage(prgname);
1336 port_conf.rxmode.max_rx_pkt_len = ret;
1338 printf("set jumbo frame "
1339 "max packet length to %u\n",
1340 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1343 if (!strncmp(lgopts[option_index].name,
1344 CMD_LINE_OPT_PARSE_PTYPE,
1345 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1346 printf("soft parse-ptype is enabled\n");
1353 print_usage(prgname);
1359 argv[optind-1] = prgname;
1362 optind = 1; /* reset getopt lib */
1367 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1369 char buf[ETHER_ADDR_FMT_SIZE];
1370 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1371 printf("%s%s", name, buf);
1374 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1376 setup_hash(int socketid)
1378 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1380 .entries = L3FWD_HASH_ENTRIES,
1381 .key_len = sizeof(struct ipv4_5tuple),
1382 .hash_func = DEFAULT_HASH_FUNC,
1383 .hash_func_init_val = 0,
1386 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1388 .entries = L3FWD_HASH_ENTRIES,
1389 .key_len = sizeof(struct ipv6_5tuple),
1390 .hash_func = DEFAULT_HASH_FUNC,
1391 .hash_func_init_val = 0,
1398 /* create ipv4 hash */
1399 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1400 ipv4_l3fwd_hash_params.name = s;
1401 ipv4_l3fwd_hash_params.socket_id = socketid;
1402 ipv4_l3fwd_lookup_struct[socketid] =
1403 rte_hash_create(&ipv4_l3fwd_hash_params);
1404 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1405 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1406 "socket %d\n", socketid);
1408 /* create ipv6 hash */
1409 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1410 ipv6_l3fwd_hash_params.name = s;
1411 ipv6_l3fwd_hash_params.socket_id = socketid;
1412 ipv6_l3fwd_lookup_struct[socketid] =
1413 rte_hash_create(&ipv6_l3fwd_hash_params);
1414 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1415 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1416 "socket %d\n", socketid);
1419 /* populate the ipv4 hash */
1420 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1421 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1422 (void *) &ipv4_l3fwd_route_array[i].key);
1424 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1425 "l3fwd hash on socket %d\n", i, socketid);
1427 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1428 printf("Hash: Adding key\n");
1429 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1432 /* populate the ipv6 hash */
1433 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1434 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1435 (void *) &ipv6_l3fwd_route_array[i].key);
1437 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1438 "l3fwd hash on socket %d\n", i, socketid);
1440 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1441 printf("Hash: Adding key\n");
1442 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1447 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1449 setup_lpm(int socketid)
1455 /* create the LPM table */
1456 struct rte_lpm_config lpm_ipv4_config;
1458 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1459 lpm_ipv4_config.number_tbl8s = 256;
1460 lpm_ipv4_config.flags = 0;
1462 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1463 ipv4_l3fwd_lookup_struct[socketid] =
1464 rte_lpm_create(s, socketid, &lpm_ipv4_config);
1465 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1466 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1467 " on socket %d\n", socketid);
1469 /* populate the LPM table */
1470 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1471 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1472 ipv4_l3fwd_route_array[i].ip,
1473 ipv4_l3fwd_route_array[i].depth,
1474 ipv4_l3fwd_route_array[i].if_out);
1477 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1478 "l3fwd LPM table on socket %d\n",
1482 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1483 (unsigned)ipv4_l3fwd_route_array[i].ip,
1484 ipv4_l3fwd_route_array[i].depth,
1485 ipv4_l3fwd_route_array[i].if_out);
1491 init_mem(unsigned nb_mbuf)
1493 struct lcore_conf *qconf;
1498 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1499 if (rte_lcore_is_enabled(lcore_id) == 0)
1503 socketid = rte_lcore_to_socket_id(lcore_id);
1507 if (socketid >= NB_SOCKETS) {
1508 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1509 "out of range %d\n", socketid,
1510 lcore_id, NB_SOCKETS);
1512 if (pktmbuf_pool[socketid] == NULL) {
1513 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1514 pktmbuf_pool[socketid] =
1515 rte_pktmbuf_pool_create(s, nb_mbuf,
1516 MEMPOOL_CACHE_SIZE, 0,
1517 RTE_MBUF_DEFAULT_BUF_SIZE,
1519 if (pktmbuf_pool[socketid] == NULL)
1520 rte_exit(EXIT_FAILURE,
1521 "Cannot init mbuf pool on socket %d\n",
1524 printf("Allocated mbuf pool on socket %d\n",
1527 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1528 setup_lpm(socketid);
1530 setup_hash(socketid);
1533 qconf = &lcore_conf[lcore_id];
1534 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1535 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1536 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1542 /* Check the link status of all ports in up to 9s, and print them finally */
1544 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1546 #define CHECK_INTERVAL 100 /* 100ms */
1547 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1548 uint8_t portid, count, all_ports_up, print_flag = 0;
1549 struct rte_eth_link link;
1551 printf("\nChecking link status");
1553 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1555 for (portid = 0; portid < port_num; portid++) {
1556 if ((port_mask & (1 << portid)) == 0)
1558 memset(&link, 0, sizeof(link));
1559 rte_eth_link_get_nowait(portid, &link);
1560 /* print link status if flag set */
1561 if (print_flag == 1) {
1562 if (link.link_status)
1563 printf("Port %d Link Up - speed %u "
1564 "Mbps - %s\n", (uint8_t)portid,
1565 (unsigned)link.link_speed,
1566 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1567 ("full-duplex") : ("half-duplex\n"));
1569 printf("Port %d Link Down\n",
1573 /* clear all_ports_up flag if any link down */
1574 if (link.link_status == ETH_LINK_DOWN) {
1579 /* after finally printing all link status, get out */
1580 if (print_flag == 1)
1583 if (all_ports_up == 0) {
1586 rte_delay_ms(CHECK_INTERVAL);
1589 /* set the print_flag if all ports up or timeout */
1590 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1597 static int check_ptype(uint8_t portid)
1600 int ptype_l3_ipv4 = 0;
1601 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1602 int ptype_l3_ipv6 = 0;
1604 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1606 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1610 uint32_t ptypes[ret];
1612 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1613 for (i = 0; i < ret; ++i) {
1614 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1616 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1617 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1622 if (ptype_l3_ipv4 == 0)
1623 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1625 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1626 if (ptype_l3_ipv6 == 0)
1627 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1630 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1632 #else /* APP_LOOKUP_EXACT_MATCH */
1633 if (ptype_l3_ipv4 && ptype_l3_ipv6)
1642 main(int argc, char **argv)
1644 struct lcore_conf *qconf;
1645 struct rte_eth_dev_info dev_info;
1646 struct rte_eth_txconf *txconf;
1652 uint32_t n_tx_queue, nb_lcores;
1653 uint32_t dev_rxq_num, dev_txq_num;
1654 uint8_t portid, nb_rx_queue, queue, socketid;
1655 uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
1657 /* catch SIGINT and restore cpufreq governor to ondemand */
1658 signal(SIGINT, signal_exit_now);
1661 ret = rte_eal_init(argc, argv);
1663 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1667 /* init RTE timer library to be used late */
1668 rte_timer_subsystem_init();
1670 /* parse application arguments (after the EAL ones) */
1671 ret = parse_args(argc, argv);
1673 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1675 if (check_lcore_params() < 0)
1676 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1678 ret = init_lcore_rx_queues();
1680 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1682 nb_ports = rte_eth_dev_count();
1684 if (check_port_config(nb_ports) < 0)
1685 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1687 nb_lcores = rte_lcore_count();
1689 /* initialize all ports */
1690 for (portid = 0; portid < nb_ports; portid++) {
1691 /* skip ports that are not enabled */
1692 if ((enabled_port_mask & (1 << portid)) == 0) {
1693 printf("\nSkipping disabled port %d\n", portid);
1698 printf("Initializing port %d ... ", portid );
1701 rte_eth_dev_info_get(portid, &dev_info);
1702 dev_rxq_num = dev_info.max_rx_queues;
1703 dev_txq_num = dev_info.max_tx_queues;
1705 nb_rx_queue = get_port_n_rx_queues(portid);
1706 if (nb_rx_queue > dev_rxq_num)
1707 rte_exit(EXIT_FAILURE,
1708 "Cannot configure not existed rxq: "
1709 "port=%d\n", portid);
1711 n_tx_queue = nb_lcores;
1712 if (n_tx_queue > dev_txq_num)
1713 n_tx_queue = dev_txq_num;
1714 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1715 nb_rx_queue, (unsigned)n_tx_queue );
1716 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1717 if (nb_rx_queue == 0)
1718 port_conf.intr_conf.rxq = 0;
1719 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1720 (uint16_t)n_tx_queue, &port_conf);
1721 /* Revert to original value */
1722 port_conf.intr_conf.rxq = org_rxq_intr;
1724 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1725 "err=%d, port=%d\n", ret, portid);
1727 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1730 rte_exit(EXIT_FAILURE,
1731 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1734 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1735 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1739 ret = init_mem(NB_MBUF);
1741 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1743 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1744 if (rte_lcore_is_enabled(lcore_id) == 0)
1747 /* Initialize TX buffers */
1748 qconf = &lcore_conf[lcore_id];
1749 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1750 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1751 rte_eth_dev_socket_id(portid));
1752 if (qconf->tx_buffer[portid] == NULL)
1753 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1756 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1759 /* init one TX queue per couple (lcore,port) */
1761 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1762 if (rte_lcore_is_enabled(lcore_id) == 0)
1765 if (queueid >= dev_txq_num)
1770 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1774 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1777 rte_eth_dev_info_get(portid, &dev_info);
1778 txconf = &dev_info.default_txconf;
1779 if (port_conf.rxmode.jumbo_frame)
1780 txconf->txq_flags = 0;
1781 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1784 rte_exit(EXIT_FAILURE,
1785 "rte_eth_tx_queue_setup: err=%d, "
1786 "port=%d\n", ret, portid);
1788 qconf = &lcore_conf[lcore_id];
1789 qconf->tx_queue_id[portid] = queueid;
1792 qconf->tx_port_id[qconf->n_tx_port] = portid;
1798 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1799 if (rte_lcore_is_enabled(lcore_id) == 0)
1802 /* init power management library */
1803 ret = rte_power_init(lcore_id);
1806 "Library initialization failed on core %u\n", lcore_id);
1808 /* init timer structures for each enabled lcore */
1809 rte_timer_init(&power_timers[lcore_id]);
1810 hz = rte_get_timer_hz();
1811 rte_timer_reset(&power_timers[lcore_id],
1812 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1813 power_timer_cb, NULL);
1815 qconf = &lcore_conf[lcore_id];
1816 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1818 /* init RX queues */
1819 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1820 portid = qconf->rx_queue_list[queue].port_id;
1821 queueid = qconf->rx_queue_list[queue].queue_id;
1825 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1829 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1832 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1834 pktmbuf_pool[socketid]);
1836 rte_exit(EXIT_FAILURE,
1837 "rte_eth_rx_queue_setup: err=%d, "
1838 "port=%d\n", ret, portid);
1841 if (add_cb_parse_ptype(portid, queueid) < 0)
1842 rte_exit(EXIT_FAILURE,
1843 "Fail to add ptype cb\n");
1844 } else if (!check_ptype(portid))
1845 rte_exit(EXIT_FAILURE,
1846 "PMD can not provide needed ptypes\n");
1853 for (portid = 0; portid < nb_ports; portid++) {
1854 if ((enabled_port_mask & (1 << portid)) == 0) {
1858 ret = rte_eth_dev_start(portid);
1860 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1861 "port=%d\n", ret, portid);
1863 * If enabled, put device in promiscuous mode.
1864 * This allows IO forwarding mode to forward packets
1865 * to itself through 2 cross-connected ports of the
1869 rte_eth_promiscuous_enable(portid);
1870 /* initialize spinlock for each port */
1871 rte_spinlock_init(&(locks[portid]));
1874 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1876 /* launch per-lcore init on every lcore */
1877 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1878 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1879 if (rte_eal_wait_lcore(lcore_id) < 0)