4 * Copyright(c) 2010-2014 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_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
54 #include <rte_per_lcore.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>
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;
168 enum freq_scale_hint_t
178 struct rte_mbuf *m_table[MAX_PKT_BURST];
181 struct lcore_rx_queue {
184 enum freq_scale_hint_t freq_up_hint;
185 uint32_t zero_rx_packet_count;
187 } __rte_cache_aligned;
189 #define MAX_RX_QUEUE_PER_LCORE 16
190 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
191 #define MAX_RX_QUEUE_PER_PORT 128
193 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
196 #define MAX_LCORE_PARAMS 1024
197 struct lcore_params {
201 } __rte_cache_aligned;
203 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
204 static struct lcore_params lcore_params_array_default[] = {
216 static struct lcore_params * lcore_params = lcore_params_array_default;
217 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
218 sizeof(lcore_params_array_default[0]);
220 static struct rte_eth_conf port_conf = {
222 .mq_mode = ETH_MQ_RX_RSS,
223 .max_rx_pkt_len = ETHER_MAX_LEN,
225 .header_split = 0, /**< Header Split disabled */
226 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
227 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
228 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
229 .hw_strip_crc = 0, /**< CRC stripped by hardware */
234 .rss_hf = ETH_RSS_UDP,
238 .mq_mode = ETH_MQ_TX_NONE,
246 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
249 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
251 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
252 #include <rte_hash_crc.h>
253 #define DEFAULT_HASH_FUNC rte_hash_crc
255 #include <rte_jhash.h>
256 #define DEFAULT_HASH_FUNC rte_jhash
265 } __attribute__((__packed__));
268 uint8_t ip_dst[IPV6_ADDR_LEN];
269 uint8_t ip_src[IPV6_ADDR_LEN];
273 } __attribute__((__packed__));
275 struct ipv4_l3fwd_route {
276 struct ipv4_5tuple key;
280 struct ipv6_l3fwd_route {
281 struct ipv6_5tuple key;
285 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
286 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
287 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
288 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
289 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
292 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
295 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
296 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
297 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
298 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
304 typedef struct rte_hash lookup_struct_t;
305 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
306 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
308 #define L3FWD_HASH_ENTRIES 1024
310 #define IPV4_L3FWD_NUM_ROUTES \
311 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
313 #define IPV6_L3FWD_NUM_ROUTES \
314 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
316 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
317 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
320 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
321 struct ipv4_l3fwd_route {
327 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
328 {IPv4(1,1,1,0), 24, 0},
329 {IPv4(2,1,1,0), 24, 1},
330 {IPv4(3,1,1,0), 24, 2},
331 {IPv4(4,1,1,0), 24, 3},
332 {IPv4(5,1,1,0), 24, 4},
333 {IPv4(6,1,1,0), 24, 5},
334 {IPv4(7,1,1,0), 24, 6},
335 {IPv4(8,1,1,0), 24, 7},
338 #define IPV4_L3FWD_NUM_ROUTES \
339 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
341 #define IPV4_L3FWD_LPM_MAX_RULES 1024
343 typedef struct rte_lpm lookup_struct_t;
344 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
349 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
350 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
351 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
352 lookup_struct_t * ipv4_lookup_struct;
353 lookup_struct_t * ipv6_lookup_struct;
354 } __rte_cache_aligned;
357 /* total sleep time in ms since last frequency scaling down */
359 /* number of long sleep recently */
360 uint32_t nb_long_sleep;
361 /* freq. scaling up trend */
363 /* total packet processed recently */
364 uint64_t nb_rx_processed;
365 /* total iterations looped recently */
366 uint64_t nb_iteration_looped;
368 } __rte_cache_aligned;
370 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
371 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
372 static struct rte_timer power_timers[RTE_MAX_LCORE];
374 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
375 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
376 unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
378 /* exit signal handler */
380 signal_exit_now(int sigtype)
385 if (sigtype == SIGINT) {
386 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
387 if (rte_lcore_is_enabled(lcore_id) == 0)
390 /* init power management library */
391 ret = rte_power_exit(lcore_id);
393 rte_exit(EXIT_FAILURE, "Power management "
394 "library de-initialization failed on "
395 "core%u\n", lcore_id);
399 rte_exit(EXIT_SUCCESS, "User forced exit\n");
402 /* Freqency scale down timer callback */
404 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
405 __attribute__((unused)) void *arg)
408 float sleep_time_ratio;
409 unsigned lcore_id = rte_lcore_id();
411 /* accumulate total execution time in us when callback is invoked */
412 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
413 (float)SCALING_PERIOD;
415 * check whether need to scale down frequency a step if it sleep a lot.
417 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
418 if (rte_power_freq_down)
419 rte_power_freq_down(lcore_id);
421 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
422 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
424 * scale down a step if average packet per iteration less
427 if (rte_power_freq_down)
428 rte_power_freq_down(lcore_id);
432 * initialize another timer according to current frequency to ensure
433 * timer interval is relatively fixed.
435 hz = rte_get_timer_hz();
436 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
437 SINGLE, lcore_id, power_timer_cb, NULL);
439 stats[lcore_id].nb_rx_processed = 0;
440 stats[lcore_id].nb_iteration_looped = 0;
442 stats[lcore_id].sleep_time = 0;
445 /* Send burst of packets on an output interface */
447 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
449 struct rte_mbuf **m_table;
453 queueid = qconf->tx_queue_id[port];
454 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
456 ret = rte_eth_tx_burst(port, queueid, m_table, n);
457 if (unlikely(ret < n)) {
459 rte_pktmbuf_free(m_table[ret]);
466 /* Enqueue a single packet, and send burst if queue is filled */
468 send_single_packet(struct rte_mbuf *m, uint8_t port)
472 struct lcore_conf *qconf;
474 lcore_id = rte_lcore_id();
476 qconf = &lcore_conf[lcore_id];
477 len = qconf->tx_mbufs[port].len;
478 qconf->tx_mbufs[port].m_table[len] = m;
481 /* enough pkts to be sent */
482 if (unlikely(len == MAX_PKT_BURST)) {
483 send_burst(qconf, MAX_PKT_BURST, port);
487 qconf->tx_mbufs[port].len = len;
491 #ifdef DO_RFC_1812_CHECKS
493 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
495 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
497 * 1. The packet length reported by the Link Layer must be large
498 * enough to hold the minimum length legal IP datagram (20 bytes).
500 if (link_len < sizeof(struct ipv4_hdr))
503 /* 2. The IP checksum must be correct. */
504 /* this is checked in H/W */
507 * 3. The IP version number must be 4. If the version number is not 4
508 * then the packet may be another version of IP, such as IPng or
511 if (((pkt->version_ihl) >> 4) != 4)
514 * 4. The IP header length field must be large enough to hold the
515 * minimum length legal IP datagram (20 bytes = 5 words).
517 if ((pkt->version_ihl & 0xf) < 5)
521 * 5. The IP total length field must be large enough to hold the IP
522 * datagram header, whose length is specified in the IP header length
525 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
532 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
534 print_ipv4_key(struct ipv4_5tuple key)
536 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
537 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
538 key.port_dst, key.port_src, key.proto);
541 print_ipv6_key(struct ipv6_5tuple key)
543 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
544 "port dst = %d, port src = %d, proto = %d\n",
545 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
546 key.port_dst, key.port_src, key.proto);
549 static inline uint8_t
550 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
551 lookup_struct_t * ipv4_l3fwd_lookup_struct)
553 struct ipv4_5tuple key;
558 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
559 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
560 key.proto = ipv4_hdr->next_proto_id;
562 switch (ipv4_hdr->next_proto_id) {
564 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
565 sizeof(struct ipv4_hdr));
566 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
567 key.port_src = rte_be_to_cpu_16(tcp->src_port);
571 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
572 sizeof(struct ipv4_hdr));
573 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
574 key.port_src = rte_be_to_cpu_16(udp->src_port);
583 /* Find destination port */
584 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
585 return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
588 static inline uint8_t
589 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint8_t portid,
590 lookup_struct_t *ipv6_l3fwd_lookup_struct)
592 struct ipv6_5tuple key;
597 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
598 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
600 key.proto = ipv6_hdr->proto;
602 switch (ipv6_hdr->proto) {
604 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
605 sizeof(struct ipv6_hdr));
606 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
607 key.port_src = rte_be_to_cpu_16(tcp->src_port);
611 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
612 sizeof(struct ipv6_hdr));
613 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
614 key.port_src = rte_be_to_cpu_16(udp->src_port);
623 /* Find destination port */
624 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
625 return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
629 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
630 static inline uint8_t
631 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
632 lookup_struct_t *ipv4_l3fwd_lookup_struct)
636 return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
637 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
643 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
644 struct lcore_conf *qconf)
646 struct ether_hdr *eth_hdr;
647 struct ipv4_hdr *ipv4_hdr;
651 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
653 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
654 /* Handle IPv4 headers.*/
656 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
657 sizeof(struct ether_hdr));
659 #ifdef DO_RFC_1812_CHECKS
660 /* Check to make sure the packet is valid (RFC1812) */
661 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
667 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
668 qconf->ipv4_lookup_struct);
669 if (dst_port >= RTE_MAX_ETHPORTS ||
670 (enabled_port_mask & 1 << dst_port) == 0)
673 /* 02:00:00:00:00:xx */
674 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
675 *((uint64_t *)d_addr_bytes) =
676 0x000000000002 + ((uint64_t)dst_port << 40);
678 #ifdef DO_RFC_1812_CHECKS
679 /* Update time to live and header checksum */
680 --(ipv4_hdr->time_to_live);
681 ++(ipv4_hdr->hdr_checksum);
685 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
687 send_single_packet(m, dst_port);
688 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
689 /* Handle IPv6 headers.*/
690 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
691 struct ipv6_hdr *ipv6_hdr;
694 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
695 sizeof(struct ether_hdr));
697 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
698 qconf->ipv6_lookup_struct);
700 if (dst_port >= RTE_MAX_ETHPORTS ||
701 (enabled_port_mask & 1 << dst_port) == 0)
704 /* 02:00:00:00:00:xx */
705 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
706 *((uint64_t *)d_addr_bytes) =
707 0x000000000002 + ((uint64_t)dst_port << 40);
710 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
712 send_single_packet(m, dst_port);
714 /* We don't currently handle IPv6 packets in LPM mode. */
721 #define MINIMUM_SLEEP_TIME 1
722 #define SUSPEND_THRESHOLD 300
724 static inline uint32_t
725 power_idle_heuristic(uint32_t zero_rx_packet_count)
727 /* If zero count is less than 100, sleep 1us */
728 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
729 return MINIMUM_SLEEP_TIME;
730 /* If zero count is less than 1000, sleep 100 us which is the
731 minimum latency switching from C3/C6 to C0
734 return SUSPEND_THRESHOLD;
739 static inline enum freq_scale_hint_t
740 power_freq_scaleup_heuristic(unsigned lcore_id,
745 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
748 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
749 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
750 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
751 #define FREQ_UP_TREND1_ACC 1
752 #define FREQ_UP_TREND2_ACC 100
753 #define FREQ_UP_THRESHOLD 10000
755 if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
756 FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
757 stats[lcore_id].trend = 0;
759 } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
760 FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
761 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
762 else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
763 FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
764 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
766 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
767 stats[lcore_id].trend = 0;
775 * force polling thread sleep until one-shot rx interrupt triggers
784 sleep_until_rx_interrupt(int num)
786 struct rte_epoll_event event[num];
788 uint8_t port_id, queue_id;
791 RTE_LOG(INFO, L3FWD_POWER,
792 "lcore %u sleeps until interrupt triggers\n",
795 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
796 for (i = 0; i < n; i++) {
797 data = event[i].epdata.data;
798 port_id = ((uintptr_t)data) >> CHAR_BIT;
799 queue_id = ((uintptr_t)data) &
800 RTE_LEN2MASK(CHAR_BIT, uint8_t);
801 RTE_LOG(INFO, L3FWD_POWER,
802 "lcore %u is waked up from rx interrupt on"
803 " port %d queue %d\n",
804 rte_lcore_id(), port_id, queue_id);
810 static void turn_on_intr(struct lcore_conf *qconf)
813 struct lcore_rx_queue *rx_queue;
814 uint8_t port_id, queue_id;
816 for (i = 0; i < qconf->n_rx_queue; ++i) {
817 rx_queue = &(qconf->rx_queue_list[i]);
818 port_id = rx_queue->port_id;
819 queue_id = rx_queue->queue_id;
821 rte_spinlock_lock(&(locks[port_id]));
822 rte_eth_dev_rx_intr_enable(port_id, queue_id);
823 rte_spinlock_unlock(&(locks[port_id]));
827 static int event_register(struct lcore_conf *qconf)
829 struct lcore_rx_queue *rx_queue;
830 uint8_t portid, queueid;
835 for (i = 0; i < qconf->n_rx_queue; ++i) {
836 rx_queue = &(qconf->rx_queue_list[i]);
837 portid = rx_queue->port_id;
838 queueid = rx_queue->queue_id;
839 data = portid << CHAR_BIT | queueid;
841 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
842 RTE_EPOLL_PER_THREAD,
844 (void *)((uintptr_t)data));
852 /* main processing loop */
854 main_loop(__attribute__((unused)) void *dummy)
856 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
858 uint64_t prev_tsc, diff_tsc, cur_tsc;
859 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
861 uint8_t portid, queueid;
862 struct lcore_conf *qconf;
863 struct lcore_rx_queue *rx_queue;
864 enum freq_scale_hint_t lcore_scaleup_hint;
865 uint32_t lcore_rx_idle_count = 0;
866 uint32_t lcore_idle_hint = 0;
869 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
873 lcore_id = rte_lcore_id();
874 qconf = &lcore_conf[lcore_id];
876 if (qconf->n_rx_queue == 0) {
877 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
881 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
883 for (i = 0; i < qconf->n_rx_queue; i++) {
884 portid = qconf->rx_queue_list[i].port_id;
885 queueid = qconf->rx_queue_list[i].queue_id;
886 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
887 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
890 /* add into event wait list */
891 if (event_register(qconf) == 0)
894 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
897 stats[lcore_id].nb_iteration_looped++;
899 cur_tsc = rte_rdtsc();
900 cur_tsc_power = cur_tsc;
903 * TX burst queue drain
905 diff_tsc = cur_tsc - prev_tsc;
906 if (unlikely(diff_tsc > drain_tsc)) {
909 * This could be optimized (use queueid instead of
910 * portid), but it is not called so often
912 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
913 if (qconf->tx_mbufs[portid].len == 0)
915 send_burst(&lcore_conf[lcore_id],
916 qconf->tx_mbufs[portid].len,
918 qconf->tx_mbufs[portid].len = 0;
924 diff_tsc_power = cur_tsc_power - prev_tsc_power;
925 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
927 prev_tsc_power = cur_tsc_power;
932 * Read packet from RX queues
934 lcore_scaleup_hint = FREQ_CURRENT;
935 lcore_rx_idle_count = 0;
936 for (i = 0; i < qconf->n_rx_queue; ++i) {
937 rx_queue = &(qconf->rx_queue_list[i]);
938 rx_queue->idle_hint = 0;
939 portid = rx_queue->port_id;
940 queueid = rx_queue->queue_id;
942 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
945 stats[lcore_id].nb_rx_processed += nb_rx;
946 if (unlikely(nb_rx == 0)) {
948 * no packet received from rx queue, try to
949 * sleep for a while forcing CPU enter deeper
952 rx_queue->zero_rx_packet_count++;
954 if (rx_queue->zero_rx_packet_count <=
958 rx_queue->idle_hint = power_idle_heuristic(\
959 rx_queue->zero_rx_packet_count);
960 lcore_rx_idle_count++;
962 rx_queue->zero_rx_packet_count = 0;
965 * do not scale up frequency immediately as
966 * user to kernel space communication is costly
967 * which might impact packet I/O for received
970 rx_queue->freq_up_hint =
971 power_freq_scaleup_heuristic(lcore_id,
975 /* Prefetch first packets */
976 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
977 rte_prefetch0(rte_pktmbuf_mtod(
978 pkts_burst[j], void *));
981 /* Prefetch and forward already prefetched packets */
982 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
983 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
984 j + PREFETCH_OFFSET], void *));
985 l3fwd_simple_forward(pkts_burst[j], portid,
989 /* Forward remaining prefetched packets */
990 for (; j < nb_rx; j++) {
991 l3fwd_simple_forward(pkts_burst[j], portid,
996 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
997 for (i = 1, lcore_scaleup_hint =
998 qconf->rx_queue_list[0].freq_up_hint;
999 i < qconf->n_rx_queue; ++i) {
1000 rx_queue = &(qconf->rx_queue_list[i]);
1001 if (rx_queue->freq_up_hint >
1003 lcore_scaleup_hint =
1004 rx_queue->freq_up_hint;
1007 if (lcore_scaleup_hint == FREQ_HIGHEST) {
1008 if (rte_power_freq_max)
1009 rte_power_freq_max(lcore_id);
1010 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1011 if (rte_power_freq_up)
1012 rte_power_freq_up(lcore_id);
1016 * All Rx queues empty in recent consecutive polls,
1017 * sleep in a conservative manner, meaning sleep as
1020 for (i = 1, lcore_idle_hint =
1021 qconf->rx_queue_list[0].idle_hint;
1022 i < qconf->n_rx_queue; ++i) {
1023 rx_queue = &(qconf->rx_queue_list[i]);
1024 if (rx_queue->idle_hint < lcore_idle_hint)
1025 lcore_idle_hint = rx_queue->idle_hint;
1028 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1030 * execute "pause" instruction to avoid context
1031 * switch which generally take hundred of
1032 * microseconds for short sleep.
1034 rte_delay_us(lcore_idle_hint);
1036 /* suspend until rx interrupt trigges */
1038 turn_on_intr(qconf);
1039 sleep_until_rx_interrupt(
1042 /* start receiving packets immediately */
1045 stats[lcore_id].sleep_time += lcore_idle_hint;
1051 check_lcore_params(void)
1053 uint8_t queue, lcore;
1057 for (i = 0; i < nb_lcore_params; ++i) {
1058 queue = lcore_params[i].queue_id;
1059 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1060 printf("invalid queue number: %hhu\n", queue);
1063 lcore = lcore_params[i].lcore_id;
1064 if (!rte_lcore_is_enabled(lcore)) {
1065 printf("error: lcore %hhu is not enabled in lcore "
1069 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1071 printf("warning: lcore %hhu is on socket %d with numa "
1072 "off\n", lcore, socketid);
1079 check_port_config(const unsigned nb_ports)
1084 for (i = 0; i < nb_lcore_params; ++i) {
1085 portid = lcore_params[i].port_id;
1086 if ((enabled_port_mask & (1 << portid)) == 0) {
1087 printf("port %u is not enabled in port mask\n",
1091 if (portid >= nb_ports) {
1092 printf("port %u is not present on the board\n",
1101 get_port_n_rx_queues(const uint8_t port)
1106 for (i = 0; i < nb_lcore_params; ++i) {
1107 if (lcore_params[i].port_id == port &&
1108 lcore_params[i].queue_id > queue)
1109 queue = lcore_params[i].queue_id;
1111 return (uint8_t)(++queue);
1115 init_lcore_rx_queues(void)
1117 uint16_t i, nb_rx_queue;
1120 for (i = 0; i < nb_lcore_params; ++i) {
1121 lcore = lcore_params[i].lcore_id;
1122 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1123 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1124 printf("error: too many queues (%u) for lcore: %u\n",
1125 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1128 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1129 lcore_params[i].port_id;
1130 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1131 lcore_params[i].queue_id;
1132 lcore_conf[lcore].n_rx_queue++;
1140 print_usage(const char *prgname)
1142 printf ("%s [EAL options] -- -p PORTMASK -P"
1143 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1144 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1145 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1146 " -P : enable promiscuous mode\n"
1147 " --config (port,queue,lcore): rx queues configuration\n"
1148 " --no-numa: optional, disable numa awareness\n"
1149 " --enable-jumbo: enable jumbo frame"
1150 " which max packet len is PKTLEN in decimal (64-9600)\n",
1154 static int parse_max_pkt_len(const char *pktlen)
1159 /* parse decimal string */
1160 len = strtoul(pktlen, &end, 10);
1161 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1171 parse_portmask(const char *portmask)
1176 /* parse hexadecimal string */
1177 pm = strtoul(portmask, &end, 16);
1178 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1188 parse_config(const char *q_arg)
1191 const char *p, *p0 = q_arg;
1199 unsigned long int_fld[_NUM_FLD];
1200 char *str_fld[_NUM_FLD];
1204 nb_lcore_params = 0;
1206 while ((p = strchr(p0,'(')) != NULL) {
1208 if((p0 = strchr(p,')')) == NULL)
1212 if(size >= sizeof(s))
1215 snprintf(s, sizeof(s), "%.*s", size, p);
1216 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1219 for (i = 0; i < _NUM_FLD; i++){
1221 int_fld[i] = strtoul(str_fld[i], &end, 0);
1222 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1226 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1227 printf("exceeded max number of lcore params: %hu\n",
1231 lcore_params_array[nb_lcore_params].port_id =
1232 (uint8_t)int_fld[FLD_PORT];
1233 lcore_params_array[nb_lcore_params].queue_id =
1234 (uint8_t)int_fld[FLD_QUEUE];
1235 lcore_params_array[nb_lcore_params].lcore_id =
1236 (uint8_t)int_fld[FLD_LCORE];
1239 lcore_params = lcore_params_array;
1244 /* Parse the argument given in the command line of the application */
1246 parse_args(int argc, char **argv)
1251 char *prgname = argv[0];
1252 static struct option lgopts[] = {
1253 {"config", 1, 0, 0},
1254 {"no-numa", 0, 0, 0},
1255 {"enable-jumbo", 0, 0, 0},
1261 while ((opt = getopt_long(argc, argvopt, "p:P",
1262 lgopts, &option_index)) != EOF) {
1267 enabled_port_mask = parse_portmask(optarg);
1268 if (enabled_port_mask == 0) {
1269 printf("invalid portmask\n");
1270 print_usage(prgname);
1275 printf("Promiscuous mode selected\n");
1281 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1282 ret = parse_config(optarg);
1284 printf("invalid config\n");
1285 print_usage(prgname);
1290 if (!strncmp(lgopts[option_index].name,
1292 printf("numa is disabled \n");
1296 if (!strncmp(lgopts[option_index].name,
1297 "enable-jumbo", 12)) {
1298 struct option lenopts =
1299 {"max-pkt-len", required_argument, \
1302 printf("jumbo frame is enabled \n");
1303 port_conf.rxmode.jumbo_frame = 1;
1306 * if no max-pkt-len set, use the default value
1309 if (0 == getopt_long(argc, argvopt, "",
1310 &lenopts, &option_index)) {
1311 ret = parse_max_pkt_len(optarg);
1313 (ret > MAX_JUMBO_PKT_LEN)){
1314 printf("invalid packet "
1316 print_usage(prgname);
1319 port_conf.rxmode.max_rx_pkt_len = ret;
1321 printf("set jumbo frame "
1322 "max packet length to %u\n",
1323 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1329 print_usage(prgname);
1335 argv[optind-1] = prgname;
1338 optind = 0; /* reset getopt lib */
1343 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1345 char buf[ETHER_ADDR_FMT_SIZE];
1346 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1347 printf("%s%s", name, buf);
1350 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1352 setup_hash(int socketid)
1354 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1356 .entries = L3FWD_HASH_ENTRIES,
1357 .key_len = sizeof(struct ipv4_5tuple),
1358 .hash_func = DEFAULT_HASH_FUNC,
1359 .hash_func_init_val = 0,
1362 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1364 .entries = L3FWD_HASH_ENTRIES,
1365 .key_len = sizeof(struct ipv6_5tuple),
1366 .hash_func = DEFAULT_HASH_FUNC,
1367 .hash_func_init_val = 0,
1374 /* create ipv4 hash */
1375 rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1376 ipv4_l3fwd_hash_params.name = s;
1377 ipv4_l3fwd_hash_params.socket_id = socketid;
1378 ipv4_l3fwd_lookup_struct[socketid] =
1379 rte_hash_create(&ipv4_l3fwd_hash_params);
1380 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1381 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1382 "socket %d\n", socketid);
1384 /* create ipv6 hash */
1385 rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1386 ipv6_l3fwd_hash_params.name = s;
1387 ipv6_l3fwd_hash_params.socket_id = socketid;
1388 ipv6_l3fwd_lookup_struct[socketid] =
1389 rte_hash_create(&ipv6_l3fwd_hash_params);
1390 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1391 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1392 "socket %d\n", socketid);
1395 /* populate the ipv4 hash */
1396 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1397 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1398 (void *) &ipv4_l3fwd_route_array[i].key);
1400 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1401 "l3fwd hash on socket %d\n", i, socketid);
1403 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1404 printf("Hash: Adding key\n");
1405 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1408 /* populate the ipv6 hash */
1409 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1410 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1411 (void *) &ipv6_l3fwd_route_array[i].key);
1413 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1414 "l3fwd hash on socket %d\n", i, socketid);
1416 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1417 printf("Hash: Adding key\n");
1418 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1423 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1425 setup_lpm(int socketid)
1431 /* create the LPM table */
1432 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1433 ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1434 IPV4_L3FWD_LPM_MAX_RULES, 0);
1435 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1436 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1437 " on socket %d\n", socketid);
1439 /* populate the LPM table */
1440 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1441 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1442 ipv4_l3fwd_route_array[i].ip,
1443 ipv4_l3fwd_route_array[i].depth,
1444 ipv4_l3fwd_route_array[i].if_out);
1447 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1448 "l3fwd LPM table on socket %d\n",
1452 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1453 (unsigned)ipv4_l3fwd_route_array[i].ip,
1454 ipv4_l3fwd_route_array[i].depth,
1455 ipv4_l3fwd_route_array[i].if_out);
1461 init_mem(unsigned nb_mbuf)
1463 struct lcore_conf *qconf;
1468 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1469 if (rte_lcore_is_enabled(lcore_id) == 0)
1473 socketid = rte_lcore_to_socket_id(lcore_id);
1477 if (socketid >= NB_SOCKETS) {
1478 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1479 "out of range %d\n", socketid,
1480 lcore_id, NB_SOCKETS);
1482 if (pktmbuf_pool[socketid] == NULL) {
1483 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1484 pktmbuf_pool[socketid] =
1485 rte_pktmbuf_pool_create(s, nb_mbuf,
1486 MEMPOOL_CACHE_SIZE, 0,
1487 RTE_MBUF_DEFAULT_BUF_SIZE,
1489 if (pktmbuf_pool[socketid] == NULL)
1490 rte_exit(EXIT_FAILURE,
1491 "Cannot init mbuf pool on socket %d\n",
1494 printf("Allocated mbuf pool on socket %d\n",
1497 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1498 setup_lpm(socketid);
1500 setup_hash(socketid);
1503 qconf = &lcore_conf[lcore_id];
1504 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1505 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1506 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1512 /* Check the link status of all ports in up to 9s, and print them finally */
1514 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1516 #define CHECK_INTERVAL 100 /* 100ms */
1517 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1518 uint8_t portid, count, all_ports_up, print_flag = 0;
1519 struct rte_eth_link link;
1521 printf("\nChecking link status");
1523 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1525 for (portid = 0; portid < port_num; portid++) {
1526 if ((port_mask & (1 << portid)) == 0)
1528 memset(&link, 0, sizeof(link));
1529 rte_eth_link_get_nowait(portid, &link);
1530 /* print link status if flag set */
1531 if (print_flag == 1) {
1532 if (link.link_status)
1533 printf("Port %d Link Up - speed %u "
1534 "Mbps - %s\n", (uint8_t)portid,
1535 (unsigned)link.link_speed,
1536 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1537 ("full-duplex") : ("half-duplex\n"));
1539 printf("Port %d Link Down\n",
1543 /* clear all_ports_up flag if any link down */
1544 if (link.link_status == 0) {
1549 /* after finally printing all link status, get out */
1550 if (print_flag == 1)
1553 if (all_ports_up == 0) {
1556 rte_delay_ms(CHECK_INTERVAL);
1559 /* set the print_flag if all ports up or timeout */
1560 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1568 main(int argc, char **argv)
1570 struct lcore_conf *qconf;
1571 struct rte_eth_dev_info dev_info;
1572 struct rte_eth_txconf *txconf;
1578 uint32_t n_tx_queue, nb_lcores;
1579 uint32_t dev_rxq_num, dev_txq_num;
1580 uint8_t portid, nb_rx_queue, queue, socketid;
1582 /* catch SIGINT and restore cpufreq governor to ondemand */
1583 signal(SIGINT, signal_exit_now);
1586 ret = rte_eal_init(argc, argv);
1588 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1592 /* init RTE timer library to be used late */
1593 rte_timer_subsystem_init();
1595 /* parse application arguments (after the EAL ones) */
1596 ret = parse_args(argc, argv);
1598 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1600 if (check_lcore_params() < 0)
1601 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1603 ret = init_lcore_rx_queues();
1605 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1608 nb_ports = rte_eth_dev_count();
1609 if (nb_ports > RTE_MAX_ETHPORTS)
1610 nb_ports = RTE_MAX_ETHPORTS;
1612 if (check_port_config(nb_ports) < 0)
1613 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1615 nb_lcores = rte_lcore_count();
1617 /* initialize all ports */
1618 for (portid = 0; portid < nb_ports; portid++) {
1619 /* skip ports that are not enabled */
1620 if ((enabled_port_mask & (1 << portid)) == 0) {
1621 printf("\nSkipping disabled port %d\n", portid);
1626 printf("Initializing port %d ... ", portid );
1629 rte_eth_dev_info_get(portid, &dev_info);
1630 dev_rxq_num = dev_info.max_rx_queues;
1631 dev_txq_num = dev_info.max_tx_queues;
1633 nb_rx_queue = get_port_n_rx_queues(portid);
1634 if (nb_rx_queue > dev_rxq_num)
1635 rte_exit(EXIT_FAILURE,
1636 "Cannot configure not existed rxq: "
1637 "port=%d\n", portid);
1639 n_tx_queue = nb_lcores;
1640 if (n_tx_queue > dev_txq_num)
1641 n_tx_queue = dev_txq_num;
1642 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1643 nb_rx_queue, (unsigned)n_tx_queue );
1644 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1645 (uint16_t)n_tx_queue, &port_conf);
1647 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1648 "err=%d, port=%d\n", ret, portid);
1650 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1651 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1655 ret = init_mem(NB_MBUF);
1657 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1659 /* init one TX queue per couple (lcore,port) */
1661 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1662 if (rte_lcore_is_enabled(lcore_id) == 0)
1665 if (queueid >= dev_txq_num)
1670 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1674 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1677 rte_eth_dev_info_get(portid, &dev_info);
1678 txconf = &dev_info.default_txconf;
1679 if (port_conf.rxmode.jumbo_frame)
1680 txconf->txq_flags = 0;
1681 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1684 rte_exit(EXIT_FAILURE,
1685 "rte_eth_tx_queue_setup: err=%d, "
1686 "port=%d\n", ret, portid);
1688 qconf = &lcore_conf[lcore_id];
1689 qconf->tx_queue_id[portid] = queueid;
1695 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1696 if (rte_lcore_is_enabled(lcore_id) == 0)
1699 /* init power management library */
1700 ret = rte_power_init(lcore_id);
1703 "Library initialization failed on core %u\n", lcore_id);
1705 /* init timer structures for each enabled lcore */
1706 rte_timer_init(&power_timers[lcore_id]);
1707 hz = rte_get_timer_hz();
1708 rte_timer_reset(&power_timers[lcore_id],
1709 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1710 power_timer_cb, NULL);
1712 qconf = &lcore_conf[lcore_id];
1713 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1715 /* init RX queues */
1716 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1717 portid = qconf->rx_queue_list[queue].port_id;
1718 queueid = qconf->rx_queue_list[queue].queue_id;
1722 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1726 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1729 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1731 pktmbuf_pool[socketid]);
1733 rte_exit(EXIT_FAILURE,
1734 "rte_eth_rx_queue_setup: err=%d, "
1735 "port=%d\n", ret, portid);
1742 for (portid = 0; portid < nb_ports; portid++) {
1743 if ((enabled_port_mask & (1 << portid)) == 0) {
1747 ret = rte_eth_dev_start(portid);
1749 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1750 "port=%d\n", ret, portid);
1752 * If enabled, put device in promiscuous mode.
1753 * This allows IO forwarding mode to forward packets
1754 * to itself through 2 cross-connected ports of the
1758 rte_eth_promiscuous_enable(portid);
1759 /* initialize spinlock for each port */
1760 rte_spinlock_init(&(locks[portid]));
1763 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1765 /* launch per-lcore init on every lcore */
1766 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1767 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1768 if (rte_eal_wait_lcore(lcore_id) < 0)