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_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
74 #include <rte_string_fns.h>
75 #include <rte_timer.h>
76 #include <rte_power.h>
78 #include <rte_spinlock.h>
80 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
82 #define MAX_PKT_BURST 32
84 #define MIN_ZERO_POLL_COUNT 10
86 /* around 100ms at 2 Ghz */
87 #define TIMER_RESOLUTION_CYCLES 200000000ULL
89 #define TIMER_NUMBER_PER_SECOND 10
91 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
92 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
94 #define APP_LOOKUP_EXACT_MATCH 0
95 #define APP_LOOKUP_LPM 1
96 #define DO_RFC_1812_CHECKS
98 #ifndef APP_LOOKUP_METHOD
99 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
102 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
103 #include <rte_hash.h>
104 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
107 #error "APP_LOOKUP_METHOD set to incorrect value"
111 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
112 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
113 #define IPv6_BYTES(addr) \
114 addr[0], addr[1], addr[2], addr[3], \
115 addr[4], addr[5], addr[6], addr[7], \
116 addr[8], addr[9], addr[10], addr[11],\
117 addr[12], addr[13],addr[14], addr[15]
120 #define MAX_JUMBO_PKT_LEN 9600
122 #define IPV6_ADDR_LEN 16
124 #define MEMPOOL_CACHE_SIZE 256
127 * This expression is used to calculate the number of mbufs needed depending on
128 * user input, taking into account memory for rx and tx hardware rings, cache
129 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
130 * NB_MBUF never goes below a minimum value of 8192.
133 #define NB_MBUF RTE_MAX ( \
134 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
135 nb_ports*nb_lcores*MAX_PKT_BURST + \
136 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
137 nb_lcores*MEMPOOL_CACHE_SIZE), \
140 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
144 /* Configure how many packets ahead to prefetch, when reading packets */
145 #define PREFETCH_OFFSET 3
148 * Configurable number of RX/TX ring descriptors
150 #define RTE_TEST_RX_DESC_DEFAULT 128
151 #define RTE_TEST_TX_DESC_DEFAULT 512
152 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
153 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
155 /* ethernet addresses of ports */
156 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
158 /* ethernet addresses of ports */
159 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
161 /* mask of enabled ports */
162 static uint32_t enabled_port_mask = 0;
163 /* Ports set in promiscuous mode off by default. */
164 static int promiscuous_on = 0;
165 /* NUMA is enabled by default. */
166 static int numa_on = 1;
167 static int parse_ptype; /**< Parse packet type using rx callback, and */
168 /**< disabled by default */
170 enum freq_scale_hint_t
178 struct lcore_rx_queue {
181 enum freq_scale_hint_t freq_up_hint;
182 uint32_t zero_rx_packet_count;
184 } __rte_cache_aligned;
186 #define MAX_RX_QUEUE_PER_LCORE 16
187 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
188 #define MAX_RX_QUEUE_PER_PORT 128
190 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
193 #define MAX_LCORE_PARAMS 1024
194 struct lcore_params {
198 } __rte_cache_aligned;
200 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
201 static struct lcore_params lcore_params_array_default[] = {
213 static struct lcore_params * lcore_params = lcore_params_array_default;
214 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
215 sizeof(lcore_params_array_default[0]);
217 static struct rte_eth_conf port_conf = {
219 .mq_mode = ETH_MQ_RX_RSS,
220 .max_rx_pkt_len = ETHER_MAX_LEN,
222 .header_split = 0, /**< Header Split disabled */
223 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
224 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
225 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
226 .hw_strip_crc = 0, /**< CRC stripped by hardware */
231 .rss_hf = ETH_RSS_UDP,
235 .mq_mode = ETH_MQ_TX_NONE,
243 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
246 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
248 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
249 #include <rte_hash_crc.h>
250 #define DEFAULT_HASH_FUNC rte_hash_crc
252 #include <rte_jhash.h>
253 #define DEFAULT_HASH_FUNC rte_jhash
262 } __attribute__((__packed__));
265 uint8_t ip_dst[IPV6_ADDR_LEN];
266 uint8_t ip_src[IPV6_ADDR_LEN];
270 } __attribute__((__packed__));
272 struct ipv4_l3fwd_route {
273 struct ipv4_5tuple key;
277 struct ipv6_l3fwd_route {
278 struct ipv6_5tuple key;
282 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
283 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
284 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
285 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
286 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
289 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
292 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
293 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
294 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
295 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
301 typedef struct rte_hash lookup_struct_t;
302 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
303 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
305 #define L3FWD_HASH_ENTRIES 1024
307 #define IPV4_L3FWD_NUM_ROUTES \
308 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
310 #define IPV6_L3FWD_NUM_ROUTES \
311 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
313 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
314 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
317 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
318 struct ipv4_l3fwd_route {
324 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
325 {IPv4(1,1,1,0), 24, 0},
326 {IPv4(2,1,1,0), 24, 1},
327 {IPv4(3,1,1,0), 24, 2},
328 {IPv4(4,1,1,0), 24, 3},
329 {IPv4(5,1,1,0), 24, 4},
330 {IPv4(6,1,1,0), 24, 5},
331 {IPv4(7,1,1,0), 24, 6},
332 {IPv4(8,1,1,0), 24, 7},
335 #define IPV4_L3FWD_NUM_ROUTES \
336 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
338 #define IPV4_L3FWD_LPM_MAX_RULES 1024
340 typedef struct rte_lpm lookup_struct_t;
341 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
346 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
348 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
349 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
350 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
351 lookup_struct_t * ipv4_lookup_struct;
352 lookup_struct_t * ipv6_lookup_struct;
353 } __rte_cache_aligned;
356 /* total sleep time in ms since last frequency scaling down */
358 /* number of long sleep recently */
359 uint32_t nb_long_sleep;
360 /* freq. scaling up trend */
362 /* total packet processed recently */
363 uint64_t nb_rx_processed;
364 /* total iterations looped recently */
365 uint64_t nb_iteration_looped;
367 } __rte_cache_aligned;
369 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
370 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
371 static struct rte_timer power_timers[RTE_MAX_LCORE];
373 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
374 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
375 unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
377 /* exit signal handler */
379 signal_exit_now(int sigtype)
384 if (sigtype == SIGINT) {
385 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
386 if (rte_lcore_is_enabled(lcore_id) == 0)
389 /* init power management library */
390 ret = rte_power_exit(lcore_id);
392 rte_exit(EXIT_FAILURE, "Power management "
393 "library de-initialization failed on "
394 "core%u\n", lcore_id);
398 rte_exit(EXIT_SUCCESS, "User forced exit\n");
401 /* Freqency scale down timer callback */
403 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
404 __attribute__((unused)) void *arg)
407 float sleep_time_ratio;
408 unsigned lcore_id = rte_lcore_id();
410 /* accumulate total execution time in us when callback is invoked */
411 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
412 (float)SCALING_PERIOD;
414 * check whether need to scale down frequency a step if it sleep a lot.
416 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
417 if (rte_power_freq_down)
418 rte_power_freq_down(lcore_id);
420 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
421 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
423 * scale down a step if average packet per iteration less
426 if (rte_power_freq_down)
427 rte_power_freq_down(lcore_id);
431 * initialize another timer according to current frequency to ensure
432 * timer interval is relatively fixed.
434 hz = rte_get_timer_hz();
435 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
436 SINGLE, lcore_id, power_timer_cb, NULL);
438 stats[lcore_id].nb_rx_processed = 0;
439 stats[lcore_id].nb_iteration_looped = 0;
441 stats[lcore_id].sleep_time = 0;
444 /* Enqueue a single packet, and send burst if queue is filled */
446 send_single_packet(struct rte_mbuf *m, uint8_t port)
449 struct lcore_conf *qconf;
451 lcore_id = rte_lcore_id();
452 qconf = &lcore_conf[lcore_id];
454 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
455 qconf->tx_buffer[port], m);
460 #ifdef DO_RFC_1812_CHECKS
462 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
464 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
466 * 1. The packet length reported by the Link Layer must be large
467 * enough to hold the minimum length legal IP datagram (20 bytes).
469 if (link_len < sizeof(struct ipv4_hdr))
472 /* 2. The IP checksum must be correct. */
473 /* this is checked in H/W */
476 * 3. The IP version number must be 4. If the version number is not 4
477 * then the packet may be another version of IP, such as IPng or
480 if (((pkt->version_ihl) >> 4) != 4)
483 * 4. The IP header length field must be large enough to hold the
484 * minimum length legal IP datagram (20 bytes = 5 words).
486 if ((pkt->version_ihl & 0xf) < 5)
490 * 5. The IP total length field must be large enough to hold the IP
491 * datagram header, whose length is specified in the IP header length
494 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
501 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
503 print_ipv4_key(struct ipv4_5tuple key)
505 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
506 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
507 key.port_dst, key.port_src, key.proto);
510 print_ipv6_key(struct ipv6_5tuple key)
512 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
513 "port dst = %d, port src = %d, proto = %d\n",
514 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
515 key.port_dst, key.port_src, key.proto);
518 static inline uint8_t
519 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
520 lookup_struct_t * ipv4_l3fwd_lookup_struct)
522 struct ipv4_5tuple key;
527 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
528 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
529 key.proto = ipv4_hdr->next_proto_id;
531 switch (ipv4_hdr->next_proto_id) {
533 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
534 sizeof(struct ipv4_hdr));
535 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
536 key.port_src = rte_be_to_cpu_16(tcp->src_port);
540 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
541 sizeof(struct ipv4_hdr));
542 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
543 key.port_src = rte_be_to_cpu_16(udp->src_port);
552 /* Find destination port */
553 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
554 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
557 static inline uint8_t
558 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
559 lookup_struct_t *ipv6_l3fwd_lookup_struct)
561 struct ipv6_5tuple key;
566 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
567 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
569 key.proto = ipv6_hdr->proto;
571 switch (ipv6_hdr->proto) {
573 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
574 sizeof(struct ipv6_hdr));
575 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
576 key.port_src = rte_be_to_cpu_16(tcp->src_port);
580 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
581 sizeof(struct ipv6_hdr));
582 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
583 key.port_src = rte_be_to_cpu_16(udp->src_port);
592 /* Find destination port */
593 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
594 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
598 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
599 static inline uint8_t
600 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
601 lookup_struct_t *ipv4_l3fwd_lookup_struct)
605 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
606 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
612 parse_ptype_one(struct rte_mbuf *m)
614 struct ether_hdr *eth_hdr;
615 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
618 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
619 ether_type = eth_hdr->ether_type;
620 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
621 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
622 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
623 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
625 m->packet_type = packet_type;
629 cb_parse_ptype(uint8_t port __rte_unused, uint16_t queue __rte_unused,
630 struct rte_mbuf *pkts[], uint16_t nb_pkts,
631 uint16_t max_pkts __rte_unused,
632 void *user_param __rte_unused)
636 for (i = 0; i < nb_pkts; ++i)
637 parse_ptype_one(pkts[i]);
643 add_cb_parse_ptype(uint8_t portid, uint16_t queueid)
645 printf("Port %d: softly parse packet type info\n", portid);
646 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
649 printf("Failed to add rx callback: port=%d\n", portid);
654 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
655 struct lcore_conf *qconf)
657 struct ether_hdr *eth_hdr;
658 struct ipv4_hdr *ipv4_hdr;
662 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
664 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
665 /* Handle IPv4 headers.*/
667 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
668 sizeof(struct ether_hdr));
670 #ifdef DO_RFC_1812_CHECKS
671 /* Check to make sure the packet is valid (RFC1812) */
672 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
678 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
679 qconf->ipv4_lookup_struct);
680 if (dst_port >= RTE_MAX_ETHPORTS ||
681 (enabled_port_mask & 1 << dst_port) == 0)
684 /* 02:00:00:00:00:xx */
685 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
686 *((uint64_t *)d_addr_bytes) =
687 0x000000000002 + ((uint64_t)dst_port << 40);
689 #ifdef DO_RFC_1812_CHECKS
690 /* Update time to live and header checksum */
691 --(ipv4_hdr->time_to_live);
692 ++(ipv4_hdr->hdr_checksum);
696 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
698 send_single_packet(m, dst_port);
699 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
700 /* Handle IPv6 headers.*/
701 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
702 struct ipv6_hdr *ipv6_hdr;
705 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
706 sizeof(struct ether_hdr));
708 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
709 qconf->ipv6_lookup_struct);
711 if (dst_port >= RTE_MAX_ETHPORTS ||
712 (enabled_port_mask & 1 << dst_port) == 0)
715 /* 02:00:00:00:00:xx */
716 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
717 *((uint64_t *)d_addr_bytes) =
718 0x000000000002 + ((uint64_t)dst_port << 40);
721 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
723 send_single_packet(m, dst_port);
725 /* We don't currently handle IPv6 packets in LPM mode. */
733 #define MINIMUM_SLEEP_TIME 1
734 #define SUSPEND_THRESHOLD 300
736 static inline uint32_t
737 power_idle_heuristic(uint32_t zero_rx_packet_count)
739 /* If zero count is less than 100, sleep 1us */
740 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
741 return MINIMUM_SLEEP_TIME;
742 /* If zero count is less than 1000, sleep 100 us which is the
743 minimum latency switching from C3/C6 to C0
746 return SUSPEND_THRESHOLD;
751 static inline enum freq_scale_hint_t
752 power_freq_scaleup_heuristic(unsigned lcore_id,
757 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
760 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
761 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
762 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
763 #define FREQ_UP_TREND1_ACC 1
764 #define FREQ_UP_TREND2_ACC 100
765 #define FREQ_UP_THRESHOLD 10000
767 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
768 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
769 stats[lcore_id].trend = 0;
771 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
772 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
773 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
774 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
775 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
776 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
778 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
779 stats[lcore_id].trend = 0;
787 * force polling thread sleep until one-shot rx interrupt triggers
796 sleep_until_rx_interrupt(int num)
798 struct rte_epoll_event event[num];
800 uint8_t port_id, queue_id;
803 RTE_LOG(INFO, L3FWD_POWER,
804 "lcore %u sleeps until interrupt triggers\n",
807 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
808 for (i = 0; i < n; i++) {
809 data = event[i].epdata.data;
810 port_id = ((uintptr_t)data) >> CHAR_BIT;
811 queue_id = ((uintptr_t)data) &
812 RTE_LEN2MASK(CHAR_BIT, uint8_t);
813 rte_eth_dev_rx_intr_disable(port_id, queue_id);
814 RTE_LOG(INFO, L3FWD_POWER,
815 "lcore %u is waked up from rx interrupt on"
816 " port %d queue %d\n",
817 rte_lcore_id(), port_id, queue_id);
823 static void turn_on_intr(struct lcore_conf *qconf)
826 struct lcore_rx_queue *rx_queue;
827 uint8_t port_id, queue_id;
829 for (i = 0; i < qconf->n_rx_queue; ++i) {
830 rx_queue = &(qconf->rx_queue_list[i]);
831 port_id = rx_queue->port_id;
832 queue_id = rx_queue->queue_id;
834 rte_spinlock_lock(&(locks[port_id]));
835 rte_eth_dev_rx_intr_enable(port_id, queue_id);
836 rte_spinlock_unlock(&(locks[port_id]));
840 static int event_register(struct lcore_conf *qconf)
842 struct lcore_rx_queue *rx_queue;
843 uint8_t portid, queueid;
848 for (i = 0; i < qconf->n_rx_queue; ++i) {
849 rx_queue = &(qconf->rx_queue_list[i]);
850 portid = rx_queue->port_id;
851 queueid = rx_queue->queue_id;
852 data = portid << CHAR_BIT | queueid;
854 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
855 RTE_EPOLL_PER_THREAD,
857 (void *)((uintptr_t)data));
865 /* main processing loop */
867 main_loop(__attribute__((unused)) void *dummy)
869 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
871 uint64_t prev_tsc, diff_tsc, cur_tsc;
872 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
874 uint8_t portid, queueid;
875 struct lcore_conf *qconf;
876 struct lcore_rx_queue *rx_queue;
877 enum freq_scale_hint_t lcore_scaleup_hint;
878 uint32_t lcore_rx_idle_count = 0;
879 uint32_t lcore_idle_hint = 0;
882 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
886 lcore_id = rte_lcore_id();
887 qconf = &lcore_conf[lcore_id];
889 if (qconf->n_rx_queue == 0) {
890 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
894 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
896 for (i = 0; i < qconf->n_rx_queue; i++) {
897 portid = qconf->rx_queue_list[i].port_id;
898 queueid = qconf->rx_queue_list[i].queue_id;
899 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
900 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
903 /* add into event wait list */
904 if (event_register(qconf) == 0)
907 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
910 stats[lcore_id].nb_iteration_looped++;
912 cur_tsc = rte_rdtsc();
913 cur_tsc_power = cur_tsc;
916 * TX burst queue drain
918 diff_tsc = cur_tsc - prev_tsc;
919 if (unlikely(diff_tsc > drain_tsc)) {
920 for (i = 0; i < qconf->n_tx_port; ++i) {
921 portid = qconf->tx_port_id[i];
922 rte_eth_tx_buffer_flush(portid,
923 qconf->tx_queue_id[portid],
924 qconf->tx_buffer[portid]);
929 diff_tsc_power = cur_tsc_power - prev_tsc_power;
930 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
932 prev_tsc_power = cur_tsc_power;
937 * Read packet from RX queues
939 lcore_scaleup_hint = FREQ_CURRENT;
940 lcore_rx_idle_count = 0;
941 for (i = 0; i < qconf->n_rx_queue; ++i) {
942 rx_queue = &(qconf->rx_queue_list[i]);
943 rx_queue->idle_hint = 0;
944 portid = rx_queue->port_id;
945 queueid = rx_queue->queue_id;
947 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
950 stats[lcore_id].nb_rx_processed += nb_rx;
951 if (unlikely(nb_rx == 0)) {
953 * no packet received from rx queue, try to
954 * sleep for a while forcing CPU enter deeper
957 rx_queue->zero_rx_packet_count++;
959 if (rx_queue->zero_rx_packet_count <=
963 rx_queue->idle_hint = power_idle_heuristic(\
964 rx_queue->zero_rx_packet_count);
965 lcore_rx_idle_count++;
967 rx_queue->zero_rx_packet_count = 0;
970 * do not scale up frequency immediately as
971 * user to kernel space communication is costly
972 * which might impact packet I/O for received
975 rx_queue->freq_up_hint =
976 power_freq_scaleup_heuristic(lcore_id,
980 /* Prefetch first packets */
981 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
982 rte_prefetch0(rte_pktmbuf_mtod(
983 pkts_burst[j], void *));
986 /* Prefetch and forward already prefetched packets */
987 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
988 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
989 j + PREFETCH_OFFSET], void *));
990 l3fwd_simple_forward(pkts_burst[j], portid,
994 /* Forward remaining prefetched packets */
995 for (; j < nb_rx; j++) {
996 l3fwd_simple_forward(pkts_burst[j], portid,
1001 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1002 for (i = 1, lcore_scaleup_hint =
1003 qconf->rx_queue_list[0].freq_up_hint;
1004 i < qconf->n_rx_queue; ++i) {
1005 rx_queue = &(qconf->rx_queue_list[i]);
1006 if (rx_queue->freq_up_hint >
1008 lcore_scaleup_hint =
1009 rx_queue->freq_up_hint;
1012 if (lcore_scaleup_hint == FREQ_HIGHEST) {
1013 if (rte_power_freq_max)
1014 rte_power_freq_max(lcore_id);
1015 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1016 if (rte_power_freq_up)
1017 rte_power_freq_up(lcore_id);
1021 * All Rx queues empty in recent consecutive polls,
1022 * sleep in a conservative manner, meaning sleep as
1025 for (i = 1, lcore_idle_hint =
1026 qconf->rx_queue_list[0].idle_hint;
1027 i < qconf->n_rx_queue; ++i) {
1028 rx_queue = &(qconf->rx_queue_list[i]);
1029 if (rx_queue->idle_hint < lcore_idle_hint)
1030 lcore_idle_hint = rx_queue->idle_hint;
1033 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1035 * execute "pause" instruction to avoid context
1036 * switch which generally take hundred of
1037 * microseconds for short sleep.
1039 rte_delay_us(lcore_idle_hint);
1041 /* suspend until rx interrupt trigges */
1043 turn_on_intr(qconf);
1044 sleep_until_rx_interrupt(
1047 /* start receiving packets immediately */
1050 stats[lcore_id].sleep_time += lcore_idle_hint;
1056 check_lcore_params(void)
1058 uint8_t queue, lcore;
1062 for (i = 0; i < nb_lcore_params; ++i) {
1063 queue = lcore_params[i].queue_id;
1064 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1065 printf("invalid queue number: %hhu\n", queue);
1068 lcore = lcore_params[i].lcore_id;
1069 if (!rte_lcore_is_enabled(lcore)) {
1070 printf("error: lcore %hhu is not enabled in lcore "
1074 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1076 printf("warning: lcore %hhu is on socket %d with numa "
1077 "off\n", lcore, socketid);
1084 check_port_config(const unsigned nb_ports)
1089 for (i = 0; i < nb_lcore_params; ++i) {
1090 portid = lcore_params[i].port_id;
1091 if ((enabled_port_mask & (1 << portid)) == 0) {
1092 printf("port %u is not enabled in port mask\n",
1096 if (portid >= nb_ports) {
1097 printf("port %u is not present on the board\n",
1106 get_port_n_rx_queues(const uint8_t port)
1111 for (i = 0; i < nb_lcore_params; ++i) {
1112 if (lcore_params[i].port_id == port &&
1113 lcore_params[i].queue_id > queue)
1114 queue = lcore_params[i].queue_id;
1116 return (uint8_t)(++queue);
1120 init_lcore_rx_queues(void)
1122 uint16_t i, nb_rx_queue;
1125 for (i = 0; i < nb_lcore_params; ++i) {
1126 lcore = lcore_params[i].lcore_id;
1127 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1128 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1129 printf("error: too many queues (%u) for lcore: %u\n",
1130 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1133 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1134 lcore_params[i].port_id;
1135 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1136 lcore_params[i].queue_id;
1137 lcore_conf[lcore].n_rx_queue++;
1145 print_usage(const char *prgname)
1147 printf ("%s [EAL options] -- -p PORTMASK -P"
1148 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1149 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1150 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1151 " -P : enable promiscuous mode\n"
1152 " --config (port,queue,lcore): rx queues configuration\n"
1153 " --no-numa: optional, disable numa awareness\n"
1154 " --enable-jumbo: enable jumbo frame"
1155 " which max packet len is PKTLEN in decimal (64-9600)\n"
1156 " --parse-ptype: parse packet type by software\n",
1160 static int parse_max_pkt_len(const char *pktlen)
1165 /* parse decimal string */
1166 len = strtoul(pktlen, &end, 10);
1167 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1177 parse_portmask(const char *portmask)
1182 /* parse hexadecimal string */
1183 pm = strtoul(portmask, &end, 16);
1184 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1194 parse_config(const char *q_arg)
1197 const char *p, *p0 = q_arg;
1205 unsigned long int_fld[_NUM_FLD];
1206 char *str_fld[_NUM_FLD];
1210 nb_lcore_params = 0;
1212 while ((p = strchr(p0,'(')) != NULL) {
1214 if((p0 = strchr(p,')')) == NULL)
1218 if(size >= sizeof(s))
1221 snprintf(s, sizeof(s), "%.*s", size, p);
1222 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1225 for (i = 0; i < _NUM_FLD; i++){
1227 int_fld[i] = strtoul(str_fld[i], &end, 0);
1228 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1232 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1233 printf("exceeded max number of lcore params: %hu\n",
1237 lcore_params_array[nb_lcore_params].port_id =
1238 (uint8_t)int_fld[FLD_PORT];
1239 lcore_params_array[nb_lcore_params].queue_id =
1240 (uint8_t)int_fld[FLD_QUEUE];
1241 lcore_params_array[nb_lcore_params].lcore_id =
1242 (uint8_t)int_fld[FLD_LCORE];
1245 lcore_params = lcore_params_array;
1250 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1252 /* Parse the argument given in the command line of the application */
1254 parse_args(int argc, char **argv)
1259 char *prgname = argv[0];
1260 static struct option lgopts[] = {
1261 {"config", 1, 0, 0},
1262 {"no-numa", 0, 0, 0},
1263 {"enable-jumbo", 0, 0, 0},
1264 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1270 while ((opt = getopt_long(argc, argvopt, "p:P",
1271 lgopts, &option_index)) != EOF) {
1276 enabled_port_mask = parse_portmask(optarg);
1277 if (enabled_port_mask == 0) {
1278 printf("invalid portmask\n");
1279 print_usage(prgname);
1284 printf("Promiscuous mode selected\n");
1290 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1291 ret = parse_config(optarg);
1293 printf("invalid config\n");
1294 print_usage(prgname);
1299 if (!strncmp(lgopts[option_index].name,
1301 printf("numa is disabled \n");
1305 if (!strncmp(lgopts[option_index].name,
1306 "enable-jumbo", 12)) {
1307 struct option lenopts =
1308 {"max-pkt-len", required_argument, \
1311 printf("jumbo frame is enabled \n");
1312 port_conf.rxmode.jumbo_frame = 1;
1315 * if no max-pkt-len set, use the default value
1318 if (0 == getopt_long(argc, argvopt, "",
1319 &lenopts, &option_index)) {
1320 ret = parse_max_pkt_len(optarg);
1322 (ret > MAX_JUMBO_PKT_LEN)){
1323 printf("invalid packet "
1325 print_usage(prgname);
1328 port_conf.rxmode.max_rx_pkt_len = ret;
1330 printf("set jumbo frame "
1331 "max packet length to %u\n",
1332 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1335 if (!strncmp(lgopts[option_index].name,
1336 CMD_LINE_OPT_PARSE_PTYPE,
1337 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1338 printf("soft parse-ptype is enabled\n");
1345 print_usage(prgname);
1351 argv[optind-1] = prgname;
1354 optind = 0; /* reset getopt lib */
1359 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1361 char buf[ETHER_ADDR_FMT_SIZE];
1362 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1363 printf("%s%s", name, buf);
1366 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1368 setup_hash(int socketid)
1370 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1372 .entries = L3FWD_HASH_ENTRIES,
1373 .key_len = sizeof(struct ipv4_5tuple),
1374 .hash_func = DEFAULT_HASH_FUNC,
1375 .hash_func_init_val = 0,
1378 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1380 .entries = L3FWD_HASH_ENTRIES,
1381 .key_len = sizeof(struct ipv6_5tuple),
1382 .hash_func = DEFAULT_HASH_FUNC,
1383 .hash_func_init_val = 0,
1390 /* create ipv4 hash */
1391 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1392 ipv4_l3fwd_hash_params.name = s;
1393 ipv4_l3fwd_hash_params.socket_id = socketid;
1394 ipv4_l3fwd_lookup_struct[socketid] =
1395 rte_hash_create(&ipv4_l3fwd_hash_params);
1396 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1397 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1398 "socket %d\n", socketid);
1400 /* create ipv6 hash */
1401 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1402 ipv6_l3fwd_hash_params.name = s;
1403 ipv6_l3fwd_hash_params.socket_id = socketid;
1404 ipv6_l3fwd_lookup_struct[socketid] =
1405 rte_hash_create(&ipv6_l3fwd_hash_params);
1406 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1407 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1408 "socket %d\n", socketid);
1411 /* populate the ipv4 hash */
1412 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1413 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1414 (void *) &ipv4_l3fwd_route_array[i].key);
1416 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1417 "l3fwd hash on socket %d\n", i, socketid);
1419 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1420 printf("Hash: Adding key\n");
1421 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1424 /* populate the ipv6 hash */
1425 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1426 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1427 (void *) &ipv6_l3fwd_route_array[i].key);
1429 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1430 "l3fwd hash on socket %d\n", i, socketid);
1432 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1433 printf("Hash: Adding key\n");
1434 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1439 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1441 setup_lpm(int socketid)
1447 /* create the LPM table */
1448 struct rte_lpm_config lpm_ipv4_config;
1450 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1451 lpm_ipv4_config.number_tbl8s = 256;
1452 lpm_ipv4_config.flags = 0;
1454 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1455 ipv4_l3fwd_lookup_struct[socketid] =
1456 rte_lpm_create(s, socketid, &lpm_ipv4_config);
1457 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1458 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1459 " on socket %d\n", socketid);
1461 /* populate the LPM table */
1462 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1463 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1464 ipv4_l3fwd_route_array[i].ip,
1465 ipv4_l3fwd_route_array[i].depth,
1466 ipv4_l3fwd_route_array[i].if_out);
1469 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1470 "l3fwd LPM table on socket %d\n",
1474 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1475 (unsigned)ipv4_l3fwd_route_array[i].ip,
1476 ipv4_l3fwd_route_array[i].depth,
1477 ipv4_l3fwd_route_array[i].if_out);
1483 init_mem(unsigned nb_mbuf)
1485 struct lcore_conf *qconf;
1490 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1491 if (rte_lcore_is_enabled(lcore_id) == 0)
1495 socketid = rte_lcore_to_socket_id(lcore_id);
1499 if (socketid >= NB_SOCKETS) {
1500 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1501 "out of range %d\n", socketid,
1502 lcore_id, NB_SOCKETS);
1504 if (pktmbuf_pool[socketid] == NULL) {
1505 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1506 pktmbuf_pool[socketid] =
1507 rte_pktmbuf_pool_create(s, nb_mbuf,
1508 MEMPOOL_CACHE_SIZE, 0,
1509 RTE_MBUF_DEFAULT_BUF_SIZE,
1511 if (pktmbuf_pool[socketid] == NULL)
1512 rte_exit(EXIT_FAILURE,
1513 "Cannot init mbuf pool on socket %d\n",
1516 printf("Allocated mbuf pool on socket %d\n",
1519 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1520 setup_lpm(socketid);
1522 setup_hash(socketid);
1525 qconf = &lcore_conf[lcore_id];
1526 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1527 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1528 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1534 /* Check the link status of all ports in up to 9s, and print them finally */
1536 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1538 #define CHECK_INTERVAL 100 /* 100ms */
1539 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1540 uint8_t portid, count, all_ports_up, print_flag = 0;
1541 struct rte_eth_link link;
1543 printf("\nChecking link status");
1545 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1547 for (portid = 0; portid < port_num; portid++) {
1548 if ((port_mask & (1 << portid)) == 0)
1550 memset(&link, 0, sizeof(link));
1551 rte_eth_link_get_nowait(portid, &link);
1552 /* print link status if flag set */
1553 if (print_flag == 1) {
1554 if (link.link_status)
1555 printf("Port %d Link Up - speed %u "
1556 "Mbps - %s\n", (uint8_t)portid,
1557 (unsigned)link.link_speed,
1558 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1559 ("full-duplex") : ("half-duplex\n"));
1561 printf("Port %d Link Down\n",
1565 /* clear all_ports_up flag if any link down */
1566 if (link.link_status == ETH_LINK_DOWN) {
1571 /* after finally printing all link status, get out */
1572 if (print_flag == 1)
1575 if (all_ports_up == 0) {
1578 rte_delay_ms(CHECK_INTERVAL);
1581 /* set the print_flag if all ports up or timeout */
1582 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1589 static int check_ptype(uint8_t portid)
1592 int ptype_l3_ipv4 = 0;
1593 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1594 int ptype_l3_ipv6 = 0;
1596 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1598 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1602 uint32_t ptypes[ret];
1604 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1605 for (i = 0; i < ret; ++i) {
1606 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1608 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1609 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1614 if (ptype_l3_ipv4 == 0)
1615 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1617 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1618 if (ptype_l3_ipv6 == 0)
1619 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1622 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1624 #else /* APP_LOOKUP_EXACT_MATCH */
1625 if (ptype_l3_ipv4 && ptype_l3_ipv6)
1634 main(int argc, char **argv)
1636 struct lcore_conf *qconf;
1637 struct rte_eth_dev_info dev_info;
1638 struct rte_eth_txconf *txconf;
1644 uint32_t n_tx_queue, nb_lcores;
1645 uint32_t dev_rxq_num, dev_txq_num;
1646 uint8_t portid, nb_rx_queue, queue, socketid;
1648 /* catch SIGINT and restore cpufreq governor to ondemand */
1649 signal(SIGINT, signal_exit_now);
1652 ret = rte_eal_init(argc, argv);
1654 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1658 /* init RTE timer library to be used late */
1659 rte_timer_subsystem_init();
1661 /* parse application arguments (after the EAL ones) */
1662 ret = parse_args(argc, argv);
1664 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1666 if (check_lcore_params() < 0)
1667 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1669 ret = init_lcore_rx_queues();
1671 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1673 nb_ports = rte_eth_dev_count();
1675 if (check_port_config(nb_ports) < 0)
1676 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1678 nb_lcores = rte_lcore_count();
1680 /* initialize all ports */
1681 for (portid = 0; portid < nb_ports; portid++) {
1682 /* skip ports that are not enabled */
1683 if ((enabled_port_mask & (1 << portid)) == 0) {
1684 printf("\nSkipping disabled port %d\n", portid);
1689 printf("Initializing port %d ... ", portid );
1692 rte_eth_dev_info_get(portid, &dev_info);
1693 dev_rxq_num = dev_info.max_rx_queues;
1694 dev_txq_num = dev_info.max_tx_queues;
1696 nb_rx_queue = get_port_n_rx_queues(portid);
1697 if (nb_rx_queue > dev_rxq_num)
1698 rte_exit(EXIT_FAILURE,
1699 "Cannot configure not existed rxq: "
1700 "port=%d\n", portid);
1702 n_tx_queue = nb_lcores;
1703 if (n_tx_queue > dev_txq_num)
1704 n_tx_queue = dev_txq_num;
1705 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1706 nb_rx_queue, (unsigned)n_tx_queue );
1707 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1708 (uint16_t)n_tx_queue, &port_conf);
1710 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1711 "err=%d, port=%d\n", ret, portid);
1713 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1714 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1718 ret = init_mem(NB_MBUF);
1720 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1722 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1723 if (rte_lcore_is_enabled(lcore_id) == 0)
1726 /* Initialize TX buffers */
1727 qconf = &lcore_conf[lcore_id];
1728 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1729 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1730 rte_eth_dev_socket_id(portid));
1731 if (qconf->tx_buffer[portid] == NULL)
1732 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1735 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1738 /* init one TX queue per couple (lcore,port) */
1740 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1741 if (rte_lcore_is_enabled(lcore_id) == 0)
1744 if (queueid >= dev_txq_num)
1749 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1753 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1756 rte_eth_dev_info_get(portid, &dev_info);
1757 txconf = &dev_info.default_txconf;
1758 if (port_conf.rxmode.jumbo_frame)
1759 txconf->txq_flags = 0;
1760 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1763 rte_exit(EXIT_FAILURE,
1764 "rte_eth_tx_queue_setup: err=%d, "
1765 "port=%d\n", ret, portid);
1767 qconf = &lcore_conf[lcore_id];
1768 qconf->tx_queue_id[portid] = queueid;
1771 qconf->tx_port_id[qconf->n_tx_port] = portid;
1777 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1778 if (rte_lcore_is_enabled(lcore_id) == 0)
1781 /* init power management library */
1782 ret = rte_power_init(lcore_id);
1785 "Library initialization failed on core %u\n", lcore_id);
1787 /* init timer structures for each enabled lcore */
1788 rte_timer_init(&power_timers[lcore_id]);
1789 hz = rte_get_timer_hz();
1790 rte_timer_reset(&power_timers[lcore_id],
1791 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1792 power_timer_cb, NULL);
1794 qconf = &lcore_conf[lcore_id];
1795 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1797 /* init RX queues */
1798 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1799 portid = qconf->rx_queue_list[queue].port_id;
1800 queueid = qconf->rx_queue_list[queue].queue_id;
1804 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1808 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1811 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1813 pktmbuf_pool[socketid]);
1815 rte_exit(EXIT_FAILURE,
1816 "rte_eth_rx_queue_setup: err=%d, "
1817 "port=%d\n", ret, portid);
1820 if (add_cb_parse_ptype(portid, queueid) < 0)
1821 rte_exit(EXIT_FAILURE,
1822 "Fail to add ptype cb\n");
1823 } else if (!check_ptype(portid))
1824 rte_exit(EXIT_FAILURE,
1825 "PMD can not provide needed ptypes\n");
1832 for (portid = 0; portid < nb_ports; portid++) {
1833 if ((enabled_port_mask & (1 << portid)) == 0) {
1837 ret = rte_eth_dev_start(portid);
1839 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1840 "port=%d\n", ret, portid);
1842 * If enabled, put device in promiscuous mode.
1843 * This allows IO forwarding mode to forward packets
1844 * to itself through 2 cross-connected ports of the
1848 rte_eth_promiscuous_enable(portid);
1849 /* initialize spinlock for each port */
1850 rte_spinlock_init(&(locks[portid]));
1853 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1855 /* launch per-lcore init on every lcore */
1856 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1857 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1858 if (rte_eal_wait_lcore(lcore_id) < 0)