1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation
11 #include <sys/queue.h>
19 #include <rte_common.h>
20 #include <rte_byteorder.h>
22 #include <rte_malloc.h>
23 #include <rte_memory.h>
24 #include <rte_memcpy.h>
26 #include <rte_launch.h>
27 #include <rte_atomic.h>
28 #include <rte_cycles.h>
29 #include <rte_prefetch.h>
30 #include <rte_lcore.h>
31 #include <rte_per_lcore.h>
32 #include <rte_branch_prediction.h>
33 #include <rte_interrupts.h>
34 #include <rte_random.h>
35 #include <rte_debug.h>
36 #include <rte_ether.h>
37 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
43 #include <rte_string_fns.h>
44 #include <rte_timer.h>
45 #include <rte_power.h>
46 #include <rte_spinlock.h>
47 #include <rte_power_empty_poll.h>
48 #include <rte_metrics.h>
50 #include "perf_core.h"
53 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
55 #define MAX_PKT_BURST 32
57 #define MIN_ZERO_POLL_COUNT 10
60 #define TIMER_NUMBER_PER_SECOND 10
62 #define INTERVALS_PER_SECOND 100
64 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND)
65 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
67 #define APP_LOOKUP_EXACT_MATCH 0
68 #define APP_LOOKUP_LPM 1
69 #define DO_RFC_1812_CHECKS
71 #ifndef APP_LOOKUP_METHOD
72 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM
75 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
77 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
80 #error "APP_LOOKUP_METHOD set to incorrect value"
84 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
85 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
86 #define IPv6_BYTES(addr) \
87 addr[0], addr[1], addr[2], addr[3], \
88 addr[4], addr[5], addr[6], addr[7], \
89 addr[8], addr[9], addr[10], addr[11],\
90 addr[12], addr[13],addr[14], addr[15]
93 #define MAX_JUMBO_PKT_LEN 9600
95 #define IPV6_ADDR_LEN 16
97 #define MEMPOOL_CACHE_SIZE 256
100 * This expression is used to calculate the number of mbufs needed depending on
101 * user input, taking into account memory for rx and tx hardware rings, cache
102 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
103 * NB_MBUF never goes below a minimum value of 8192.
106 #define NB_MBUF RTE_MAX ( \
107 (nb_ports*nb_rx_queue*nb_rxd + \
108 nb_ports*nb_lcores*MAX_PKT_BURST + \
109 nb_ports*n_tx_queue*nb_txd + \
110 nb_lcores*MEMPOOL_CACHE_SIZE), \
113 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
117 /* Configure how many packets ahead to prefetch, when reading packets */
118 #define PREFETCH_OFFSET 3
121 * Configurable number of RX/TX ring descriptors
123 #define RTE_TEST_RX_DESC_DEFAULT 1024
124 #define RTE_TEST_TX_DESC_DEFAULT 1024
127 * These two thresholds were decided on by running the training algorithm on
128 * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
129 * for the med_threshold and high_threshold parameters on the command line.
131 #define EMPTY_POLL_MED_THRESHOLD 350000UL
132 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
136 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
137 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
139 /* ethernet addresses of ports */
140 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
142 /* ethernet addresses of ports */
143 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
145 /* mask of enabled ports */
146 static uint32_t enabled_port_mask = 0;
147 /* Ports set in promiscuous mode off by default. */
148 static int promiscuous_on = 0;
149 /* NUMA is enabled by default. */
150 static int numa_on = 1;
151 static bool empty_poll_stop;
152 static bool empty_poll_train;
153 volatile bool quit_signal;
154 static struct ep_params *ep_params;
155 static struct ep_policy policy;
156 static long ep_med_edpi, ep_hgh_edpi;
157 /* timer to update telemetry every 500ms */
158 static struct rte_timer telemetry_timer;
160 /* stats index returned by metrics lib */
163 struct telstats_name {
164 char name[RTE_ETH_XSTATS_NAME_SIZE];
167 /* telemetry stats to be reported */
168 const struct telstats_name telstats_strings[] = {
174 /* core busyness in percentage */
181 /* reference poll count to measure core busyness */
182 #define DEFAULT_COUNT 10000
184 * reference CYCLES to be used to
185 * measure core busyness based on poll count
187 #define MIN_CYCLES 1500000ULL
188 #define MAX_CYCLES 22000000ULL
191 #define TELEMETRY_INTERVALS_PER_SEC 2
193 static int parse_ptype; /**< Parse packet type using rx callback, and */
194 /**< disabled by default */
202 enum appmode app_mode;
204 enum freq_scale_hint_t
212 struct lcore_rx_queue {
215 enum freq_scale_hint_t freq_up_hint;
216 uint32_t zero_rx_packet_count;
218 } __rte_cache_aligned;
220 #define MAX_RX_QUEUE_PER_LCORE 16
221 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
222 #define MAX_RX_QUEUE_PER_PORT 128
224 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
227 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
228 static struct lcore_params lcore_params_array_default[] = {
240 struct lcore_params *lcore_params = lcore_params_array_default;
241 uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
243 static struct rte_eth_conf port_conf = {
245 .mq_mode = ETH_MQ_RX_RSS,
246 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
248 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
253 .rss_hf = ETH_RSS_UDP,
257 .mq_mode = ETH_MQ_TX_NONE,
264 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
267 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
270 #include <rte_hash_crc.h>
271 #define DEFAULT_HASH_FUNC rte_hash_crc
273 #include <rte_jhash.h>
274 #define DEFAULT_HASH_FUNC rte_jhash
286 uint8_t ip_dst[IPV6_ADDR_LEN];
287 uint8_t ip_src[IPV6_ADDR_LEN];
293 struct ipv4_l3fwd_route {
294 struct ipv4_5tuple key;
298 struct ipv6_l3fwd_route {
299 struct ipv6_5tuple key;
303 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
304 {{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
305 {{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
306 {{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
307 {{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
310 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
313 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
314 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
315 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
316 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
322 typedef struct rte_hash lookup_struct_t;
323 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
324 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
326 #define L3FWD_HASH_ENTRIES 1024
328 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
329 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
332 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
333 struct ipv4_l3fwd_route {
339 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
340 {RTE_IPV4(1,1,1,0), 24, 0},
341 {RTE_IPV4(2,1,1,0), 24, 1},
342 {RTE_IPV4(3,1,1,0), 24, 2},
343 {RTE_IPV4(4,1,1,0), 24, 3},
344 {RTE_IPV4(5,1,1,0), 24, 4},
345 {RTE_IPV4(6,1,1,0), 24, 5},
346 {RTE_IPV4(7,1,1,0), 24, 6},
347 {RTE_IPV4(8,1,1,0), 24, 7},
350 #define IPV4_L3FWD_LPM_MAX_RULES 1024
352 typedef struct rte_lpm lookup_struct_t;
353 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
358 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
360 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
361 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
362 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
363 lookup_struct_t * ipv4_lookup_struct;
364 lookup_struct_t * ipv6_lookup_struct;
365 } __rte_cache_aligned;
368 /* total sleep time in ms since last frequency scaling down */
370 /* number of long sleep recently */
371 uint32_t nb_long_sleep;
372 /* freq. scaling up trend */
374 /* total packet processed recently */
375 uint64_t nb_rx_processed;
376 /* total iterations looped recently */
377 uint64_t nb_iteration_looped;
379 * Represents empty and non empty polls
380 * of rte_eth_rx_burst();
381 * ep_nep[0] holds non empty polls
382 * i.e. 0 < nb_rx <= MAX_BURST
383 * ep_nep[1] holds empty polls.
388 * Represents full and empty+partial
389 * polls of rte_eth_rx_burst();
390 * ep_nep[0] holds empty+partial polls.
391 * i.e. 0 <= nb_rx < MAX_BURST
392 * ep_nep[1] holds full polls
393 * i.e. nb_rx == MAX_BURST
397 rte_spinlock_t telemetry_lock;
398 } __rte_cache_aligned;
400 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
401 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
402 static struct rte_timer power_timers[RTE_MAX_LCORE];
404 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
405 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
406 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
410 * These defaults are using the max frequency index (1), a medium index (9)
411 * and a typical low frequency index (14). These can be adjusted to use
412 * different indexes using the relevant command line parameters.
414 static uint8_t freq_tlb[] = {14, 9, 1};
416 static int is_done(void)
421 /* exit signal handler */
423 signal_exit_now(int sigtype)
429 if (sigtype == SIGINT) {
430 if (app_mode == APP_MODE_EMPTY_POLL ||
431 app_mode == APP_MODE_TELEMETRY)
435 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
436 if (rte_lcore_is_enabled(lcore_id) == 0)
439 /* init power management library */
440 ret = rte_power_exit(lcore_id);
442 rte_exit(EXIT_FAILURE, "Power management "
443 "library de-initialization failed on "
444 "core%u\n", lcore_id);
447 if (app_mode != APP_MODE_EMPTY_POLL) {
448 RTE_ETH_FOREACH_DEV(portid) {
449 if ((enabled_port_mask & (1 << portid)) == 0)
452 rte_eth_dev_stop(portid);
453 rte_eth_dev_close(portid);
458 if (app_mode != APP_MODE_EMPTY_POLL)
459 rte_exit(EXIT_SUCCESS, "User forced exit\n");
462 /* Freqency scale down timer callback */
464 power_timer_cb(__rte_unused struct rte_timer *tim,
465 __rte_unused void *arg)
468 float sleep_time_ratio;
469 unsigned lcore_id = rte_lcore_id();
471 /* accumulate total execution time in us when callback is invoked */
472 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
473 (float)SCALING_PERIOD;
475 * check whether need to scale down frequency a step if it sleep a lot.
477 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
478 if (rte_power_freq_down)
479 rte_power_freq_down(lcore_id);
481 else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
482 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
484 * scale down a step if average packet per iteration less
487 if (rte_power_freq_down)
488 rte_power_freq_down(lcore_id);
492 * initialize another timer according to current frequency to ensure
493 * timer interval is relatively fixed.
495 hz = rte_get_timer_hz();
496 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
497 SINGLE, lcore_id, power_timer_cb, NULL);
499 stats[lcore_id].nb_rx_processed = 0;
500 stats[lcore_id].nb_iteration_looped = 0;
502 stats[lcore_id].sleep_time = 0;
505 /* Enqueue a single packet, and send burst if queue is filled */
507 send_single_packet(struct rte_mbuf *m, uint16_t port)
510 struct lcore_conf *qconf;
512 lcore_id = rte_lcore_id();
513 qconf = &lcore_conf[lcore_id];
515 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
516 qconf->tx_buffer[port], m);
521 #ifdef DO_RFC_1812_CHECKS
523 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
525 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
527 * 1. The packet length reported by the Link Layer must be large
528 * enough to hold the minimum length legal IP datagram (20 bytes).
530 if (link_len < sizeof(struct rte_ipv4_hdr))
533 /* 2. The IP checksum must be correct. */
534 /* this is checked in H/W */
537 * 3. The IP version number must be 4. If the version number is not 4
538 * then the packet may be another version of IP, such as IPng or
541 if (((pkt->version_ihl) >> 4) != 4)
544 * 4. The IP header length field must be large enough to hold the
545 * minimum length legal IP datagram (20 bytes = 5 words).
547 if ((pkt->version_ihl & 0xf) < 5)
551 * 5. The IP total length field must be large enough to hold the IP
552 * datagram header, whose length is specified in the IP header length
555 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
562 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
564 print_ipv4_key(struct ipv4_5tuple key)
566 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
567 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
568 key.port_dst, key.port_src, key.proto);
571 print_ipv6_key(struct ipv6_5tuple key)
573 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
574 "port dst = %d, port src = %d, proto = %d\n",
575 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
576 key.port_dst, key.port_src, key.proto);
579 static inline uint16_t
580 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
581 lookup_struct_t * ipv4_l3fwd_lookup_struct)
583 struct ipv4_5tuple key;
584 struct rte_tcp_hdr *tcp;
585 struct rte_udp_hdr *udp;
588 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
589 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
590 key.proto = ipv4_hdr->next_proto_id;
592 switch (ipv4_hdr->next_proto_id) {
594 tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
595 sizeof(struct rte_ipv4_hdr));
596 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
597 key.port_src = rte_be_to_cpu_16(tcp->src_port);
601 udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
602 sizeof(struct rte_ipv4_hdr));
603 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
604 key.port_src = rte_be_to_cpu_16(udp->src_port);
613 /* Find destination port */
614 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
615 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
618 static inline uint16_t
619 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
620 lookup_struct_t *ipv6_l3fwd_lookup_struct)
622 struct ipv6_5tuple key;
623 struct rte_tcp_hdr *tcp;
624 struct rte_udp_hdr *udp;
627 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
628 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
630 key.proto = ipv6_hdr->proto;
632 switch (ipv6_hdr->proto) {
634 tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
635 sizeof(struct rte_ipv6_hdr));
636 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
637 key.port_src = rte_be_to_cpu_16(tcp->src_port);
641 udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
642 sizeof(struct rte_ipv6_hdr));
643 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
644 key.port_src = rte_be_to_cpu_16(udp->src_port);
653 /* Find destination port */
654 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
655 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
659 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
660 static inline uint16_t
661 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
662 lookup_struct_t *ipv4_l3fwd_lookup_struct)
666 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
667 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
673 parse_ptype_one(struct rte_mbuf *m)
675 struct rte_ether_hdr *eth_hdr;
676 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
679 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
680 ether_type = eth_hdr->ether_type;
681 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
682 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
683 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
684 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
686 m->packet_type = packet_type;
690 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
691 struct rte_mbuf *pkts[], uint16_t nb_pkts,
692 uint16_t max_pkts __rte_unused,
693 void *user_param __rte_unused)
697 for (i = 0; i < nb_pkts; ++i)
698 parse_ptype_one(pkts[i]);
704 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
706 printf("Port %d: softly parse packet type info\n", portid);
707 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
710 printf("Failed to add rx callback: port=%d\n", portid);
715 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
716 struct lcore_conf *qconf)
718 struct rte_ether_hdr *eth_hdr;
719 struct rte_ipv4_hdr *ipv4_hdr;
723 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
725 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
726 /* Handle IPv4 headers.*/
728 rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
729 sizeof(struct rte_ether_hdr));
731 #ifdef DO_RFC_1812_CHECKS
732 /* Check to make sure the packet is valid (RFC1812) */
733 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
739 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
740 qconf->ipv4_lookup_struct);
741 if (dst_port >= RTE_MAX_ETHPORTS ||
742 (enabled_port_mask & 1 << dst_port) == 0)
745 /* 02:00:00:00:00:xx */
746 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
747 *((uint64_t *)d_addr_bytes) =
748 0x000000000002 + ((uint64_t)dst_port << 40);
750 #ifdef DO_RFC_1812_CHECKS
751 /* Update time to live and header checksum */
752 --(ipv4_hdr->time_to_live);
753 ++(ipv4_hdr->hdr_checksum);
757 rte_ether_addr_copy(&ports_eth_addr[dst_port],
760 send_single_packet(m, dst_port);
761 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
762 /* Handle IPv6 headers.*/
763 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
764 struct rte_ipv6_hdr *ipv6_hdr;
767 rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
768 sizeof(struct rte_ether_hdr));
770 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
771 qconf->ipv6_lookup_struct);
773 if (dst_port >= RTE_MAX_ETHPORTS ||
774 (enabled_port_mask & 1 << dst_port) == 0)
777 /* 02:00:00:00:00:xx */
778 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
779 *((uint64_t *)d_addr_bytes) =
780 0x000000000002 + ((uint64_t)dst_port << 40);
783 rte_ether_addr_copy(&ports_eth_addr[dst_port],
786 send_single_packet(m, dst_port);
788 /* We don't currently handle IPv6 packets in LPM mode. */
796 #define MINIMUM_SLEEP_TIME 1
797 #define SUSPEND_THRESHOLD 300
799 static inline uint32_t
800 power_idle_heuristic(uint32_t zero_rx_packet_count)
802 /* If zero count is less than 100, sleep 1us */
803 if (zero_rx_packet_count < SUSPEND_THRESHOLD)
804 return MINIMUM_SLEEP_TIME;
805 /* If zero count is less than 1000, sleep 100 us which is the
806 minimum latency switching from C3/C6 to C0
809 return SUSPEND_THRESHOLD;
812 static inline enum freq_scale_hint_t
813 power_freq_scaleup_heuristic(unsigned lcore_id,
817 uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
819 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
822 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST
823 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2)
824 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3)
825 #define FREQ_UP_TREND1_ACC 1
826 #define FREQ_UP_TREND2_ACC 100
827 #define FREQ_UP_THRESHOLD 10000
829 if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
830 stats[lcore_id].trend = 0;
832 } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
833 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
834 else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
835 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
837 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
838 stats[lcore_id].trend = 0;
846 * force polling thread sleep until one-shot rx interrupt triggers
855 sleep_until_rx_interrupt(int num)
857 struct rte_epoll_event event[num];
863 RTE_LOG(INFO, L3FWD_POWER,
864 "lcore %u sleeps until interrupt triggers\n",
867 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
868 for (i = 0; i < n; i++) {
869 data = event[i].epdata.data;
870 port_id = ((uintptr_t)data) >> CHAR_BIT;
871 queue_id = ((uintptr_t)data) &
872 RTE_LEN2MASK(CHAR_BIT, uint8_t);
873 RTE_LOG(INFO, L3FWD_POWER,
874 "lcore %u is waked up from rx interrupt on"
875 " port %d queue %d\n",
876 rte_lcore_id(), port_id, queue_id);
882 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
885 struct lcore_rx_queue *rx_queue;
889 for (i = 0; i < qconf->n_rx_queue; ++i) {
890 rx_queue = &(qconf->rx_queue_list[i]);
891 port_id = rx_queue->port_id;
892 queue_id = rx_queue->queue_id;
894 rte_spinlock_lock(&(locks[port_id]));
896 rte_eth_dev_rx_intr_enable(port_id, queue_id);
898 rte_eth_dev_rx_intr_disable(port_id, queue_id);
899 rte_spinlock_unlock(&(locks[port_id]));
903 static int event_register(struct lcore_conf *qconf)
905 struct lcore_rx_queue *rx_queue;
912 for (i = 0; i < qconf->n_rx_queue; ++i) {
913 rx_queue = &(qconf->rx_queue_list[i]);
914 portid = rx_queue->port_id;
915 queueid = rx_queue->queue_id;
916 data = portid << CHAR_BIT | queueid;
918 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
919 RTE_EPOLL_PER_THREAD,
921 (void *)((uintptr_t)data));
928 /* main processing loop */
930 main_telemetry_loop(__rte_unused void *dummy)
932 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
933 unsigned int lcore_id;
934 uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
938 struct lcore_conf *qconf;
939 struct lcore_rx_queue *rx_queue;
940 uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
944 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
945 US_PER_S * BURST_TX_DRAIN_US;
951 lcore_id = rte_lcore_id();
952 qconf = &lcore_conf[lcore_id];
954 if (qconf->n_rx_queue == 0) {
955 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
960 RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
963 for (i = 0; i < qconf->n_rx_queue; i++) {
964 portid = qconf->rx_queue_list[i].port_id;
965 queueid = qconf->rx_queue_list[i].queue_id;
966 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
967 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
972 cur_tsc = rte_rdtsc();
974 * TX burst queue drain
976 diff_tsc = cur_tsc - prev_tsc;
977 if (unlikely(diff_tsc > drain_tsc)) {
978 for (i = 0; i < qconf->n_tx_port; ++i) {
979 portid = qconf->tx_port_id[i];
980 rte_eth_tx_buffer_flush(portid,
981 qconf->tx_queue_id[portid],
982 qconf->tx_buffer[portid]);
988 * Read packet from RX queues
990 for (i = 0; i < qconf->n_rx_queue; ++i) {
991 rx_queue = &(qconf->rx_queue_list[i]);
992 portid = rx_queue->port_id;
993 queueid = rx_queue->queue_id;
995 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
997 ep_nep[nb_rx == 0]++;
998 fp_nfp[nb_rx == MAX_PKT_BURST]++;
1000 if (unlikely(nb_rx == 0))
1003 /* Prefetch first packets */
1004 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1005 rte_prefetch0(rte_pktmbuf_mtod(
1006 pkts_burst[j], void *));
1009 /* Prefetch and forward already prefetched packets */
1010 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1011 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1012 j + PREFETCH_OFFSET], void *));
1013 l3fwd_simple_forward(pkts_burst[j], portid,
1017 /* Forward remaining prefetched packets */
1018 for (; j < nb_rx; j++) {
1019 l3fwd_simple_forward(pkts_burst[j], portid,
1023 if (unlikely(poll_count >= DEFAULT_COUNT)) {
1024 diff_tsc = cur_tsc - prev_tel_tsc;
1025 if (diff_tsc >= MAX_CYCLES) {
1027 } else if (diff_tsc > MIN_CYCLES &&
1028 diff_tsc < MAX_CYCLES) {
1029 br = (diff_tsc * 100) / MAX_CYCLES;
1034 prev_tel_tsc = cur_tsc;
1035 /* update stats for telemetry */
1036 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1037 stats[lcore_id].ep_nep[0] = ep_nep[0];
1038 stats[lcore_id].ep_nep[1] = ep_nep[1];
1039 stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1040 stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1041 stats[lcore_id].br = br;
1042 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1048 /* main processing loop */
1050 main_empty_poll_loop(__rte_unused void *dummy)
1052 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1053 unsigned int lcore_id;
1054 uint64_t prev_tsc, diff_tsc, cur_tsc;
1058 struct lcore_conf *qconf;
1059 struct lcore_rx_queue *rx_queue;
1061 const uint64_t drain_tsc =
1062 (rte_get_tsc_hz() + US_PER_S - 1) /
1063 US_PER_S * BURST_TX_DRAIN_US;
1067 lcore_id = rte_lcore_id();
1068 qconf = &lcore_conf[lcore_id];
1070 if (qconf->n_rx_queue == 0) {
1071 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1076 for (i = 0; i < qconf->n_rx_queue; i++) {
1077 portid = qconf->rx_queue_list[i].port_id;
1078 queueid = qconf->rx_queue_list[i].queue_id;
1079 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1080 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1083 while (!is_done()) {
1084 stats[lcore_id].nb_iteration_looped++;
1086 cur_tsc = rte_rdtsc();
1088 * TX burst queue drain
1090 diff_tsc = cur_tsc - prev_tsc;
1091 if (unlikely(diff_tsc > drain_tsc)) {
1092 for (i = 0; i < qconf->n_tx_port; ++i) {
1093 portid = qconf->tx_port_id[i];
1094 rte_eth_tx_buffer_flush(portid,
1095 qconf->tx_queue_id[portid],
1096 qconf->tx_buffer[portid]);
1102 * Read packet from RX queues
1104 for (i = 0; i < qconf->n_rx_queue; ++i) {
1105 rx_queue = &(qconf->rx_queue_list[i]);
1106 rx_queue->idle_hint = 0;
1107 portid = rx_queue->port_id;
1108 queueid = rx_queue->queue_id;
1110 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1113 stats[lcore_id].nb_rx_processed += nb_rx;
1117 rte_power_empty_poll_stat_update(lcore_id);
1121 rte_power_poll_stat_update(lcore_id, nb_rx);
1125 /* Prefetch first packets */
1126 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1127 rte_prefetch0(rte_pktmbuf_mtod(
1128 pkts_burst[j], void *));
1131 /* Prefetch and forward already prefetched packets */
1132 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1133 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1134 j + PREFETCH_OFFSET],
1136 l3fwd_simple_forward(pkts_burst[j], portid,
1140 /* Forward remaining prefetched packets */
1141 for (; j < nb_rx; j++) {
1142 l3fwd_simple_forward(pkts_burst[j], portid,
1152 /* main processing loop */
1154 main_loop(__rte_unused void *dummy)
1156 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1158 uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1159 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1163 struct lcore_conf *qconf;
1164 struct lcore_rx_queue *rx_queue;
1165 enum freq_scale_hint_t lcore_scaleup_hint;
1166 uint32_t lcore_rx_idle_count = 0;
1167 uint32_t lcore_idle_hint = 0;
1170 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1173 hz = rte_get_timer_hz();
1174 tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1176 lcore_id = rte_lcore_id();
1177 qconf = &lcore_conf[lcore_id];
1179 if (qconf->n_rx_queue == 0) {
1180 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1184 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1186 for (i = 0; i < qconf->n_rx_queue; i++) {
1187 portid = qconf->rx_queue_list[i].port_id;
1188 queueid = qconf->rx_queue_list[i].queue_id;
1189 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1190 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1193 /* add into event wait list */
1194 if (event_register(qconf) == 0)
1197 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1200 stats[lcore_id].nb_iteration_looped++;
1202 cur_tsc = rte_rdtsc();
1203 cur_tsc_power = cur_tsc;
1206 * TX burst queue drain
1208 diff_tsc = cur_tsc - prev_tsc;
1209 if (unlikely(diff_tsc > drain_tsc)) {
1210 for (i = 0; i < qconf->n_tx_port; ++i) {
1211 portid = qconf->tx_port_id[i];
1212 rte_eth_tx_buffer_flush(portid,
1213 qconf->tx_queue_id[portid],
1214 qconf->tx_buffer[portid]);
1219 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1220 if (diff_tsc_power > tim_res_tsc) {
1222 prev_tsc_power = cur_tsc_power;
1227 * Read packet from RX queues
1229 lcore_scaleup_hint = FREQ_CURRENT;
1230 lcore_rx_idle_count = 0;
1231 for (i = 0; i < qconf->n_rx_queue; ++i) {
1232 rx_queue = &(qconf->rx_queue_list[i]);
1233 rx_queue->idle_hint = 0;
1234 portid = rx_queue->port_id;
1235 queueid = rx_queue->queue_id;
1237 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1240 stats[lcore_id].nb_rx_processed += nb_rx;
1241 if (unlikely(nb_rx == 0)) {
1243 * no packet received from rx queue, try to
1244 * sleep for a while forcing CPU enter deeper
1247 rx_queue->zero_rx_packet_count++;
1249 if (rx_queue->zero_rx_packet_count <=
1250 MIN_ZERO_POLL_COUNT)
1253 rx_queue->idle_hint = power_idle_heuristic(\
1254 rx_queue->zero_rx_packet_count);
1255 lcore_rx_idle_count++;
1257 rx_queue->zero_rx_packet_count = 0;
1260 * do not scale up frequency immediately as
1261 * user to kernel space communication is costly
1262 * which might impact packet I/O for received
1265 rx_queue->freq_up_hint =
1266 power_freq_scaleup_heuristic(lcore_id,
1270 /* Prefetch first packets */
1271 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1272 rte_prefetch0(rte_pktmbuf_mtod(
1273 pkts_burst[j], void *));
1276 /* Prefetch and forward already prefetched packets */
1277 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1278 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1279 j + PREFETCH_OFFSET], void *));
1280 l3fwd_simple_forward(pkts_burst[j], portid,
1284 /* Forward remaining prefetched packets */
1285 for (; j < nb_rx; j++) {
1286 l3fwd_simple_forward(pkts_burst[j], portid,
1291 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1292 for (i = 1, lcore_scaleup_hint =
1293 qconf->rx_queue_list[0].freq_up_hint;
1294 i < qconf->n_rx_queue; ++i) {
1295 rx_queue = &(qconf->rx_queue_list[i]);
1296 if (rx_queue->freq_up_hint >
1298 lcore_scaleup_hint =
1299 rx_queue->freq_up_hint;
1302 if (lcore_scaleup_hint == FREQ_HIGHEST) {
1303 if (rte_power_freq_max)
1304 rte_power_freq_max(lcore_id);
1305 } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1306 if (rte_power_freq_up)
1307 rte_power_freq_up(lcore_id);
1311 * All Rx queues empty in recent consecutive polls,
1312 * sleep in a conservative manner, meaning sleep as
1315 for (i = 1, lcore_idle_hint =
1316 qconf->rx_queue_list[0].idle_hint;
1317 i < qconf->n_rx_queue; ++i) {
1318 rx_queue = &(qconf->rx_queue_list[i]);
1319 if (rx_queue->idle_hint < lcore_idle_hint)
1320 lcore_idle_hint = rx_queue->idle_hint;
1323 if (lcore_idle_hint < SUSPEND_THRESHOLD)
1325 * execute "pause" instruction to avoid context
1326 * switch which generally take hundred of
1327 * microseconds for short sleep.
1329 rte_delay_us(lcore_idle_hint);
1331 /* suspend until rx interrupt triggers */
1333 turn_on_off_intr(qconf, 1);
1334 sleep_until_rx_interrupt(
1336 turn_on_off_intr(qconf, 0);
1338 * start receiving packets immediately
1343 stats[lcore_id].sleep_time += lcore_idle_hint;
1349 check_lcore_params(void)
1351 uint8_t queue, lcore;
1355 for (i = 0; i < nb_lcore_params; ++i) {
1356 queue = lcore_params[i].queue_id;
1357 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1358 printf("invalid queue number: %hhu\n", queue);
1361 lcore = lcore_params[i].lcore_id;
1362 if (!rte_lcore_is_enabled(lcore)) {
1363 printf("error: lcore %hhu is not enabled in lcore "
1367 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1369 printf("warning: lcore %hhu is on socket %d with numa "
1370 "off\n", lcore, socketid);
1372 if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1373 printf("cannot enable master core %d in config for telemetry mode\n",
1382 check_port_config(void)
1387 for (i = 0; i < nb_lcore_params; ++i) {
1388 portid = lcore_params[i].port_id;
1389 if ((enabled_port_mask & (1 << portid)) == 0) {
1390 printf("port %u is not enabled in port mask\n",
1394 if (!rte_eth_dev_is_valid_port(portid)) {
1395 printf("port %u is not present on the board\n",
1404 get_port_n_rx_queues(const uint16_t port)
1409 for (i = 0; i < nb_lcore_params; ++i) {
1410 if (lcore_params[i].port_id == port &&
1411 lcore_params[i].queue_id > queue)
1412 queue = lcore_params[i].queue_id;
1414 return (uint8_t)(++queue);
1418 init_lcore_rx_queues(void)
1420 uint16_t i, nb_rx_queue;
1423 for (i = 0; i < nb_lcore_params; ++i) {
1424 lcore = lcore_params[i].lcore_id;
1425 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1426 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1427 printf("error: too many queues (%u) for lcore: %u\n",
1428 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1431 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1432 lcore_params[i].port_id;
1433 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1434 lcore_params[i].queue_id;
1435 lcore_conf[lcore].n_rx_queue++;
1443 print_usage(const char *prgname)
1445 printf ("%s [EAL options] -- -p PORTMASK -P"
1446 " [--config (port,queue,lcore)[,(port,queue,lcore]]"
1447 " [--high-perf-cores CORELIST"
1448 " [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1449 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1450 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1451 " -P : enable promiscuous mode\n"
1452 " --config (port,queue,lcore): rx queues configuration\n"
1453 " --high-perf-cores CORELIST: list of high performance cores\n"
1454 " --perf-config: similar as config, cores specified as indices"
1455 " for bins containing high or regular performance cores\n"
1456 " --no-numa: optional, disable numa awareness\n"
1457 " --enable-jumbo: enable jumbo frame"
1458 " which max packet len is PKTLEN in decimal (64-9600)\n"
1459 " --parse-ptype: parse packet type by software\n"
1460 " --empty-poll: enable empty poll detection"
1461 " follow (training_flag, high_threshold, med_threshold)\n"
1462 " --telemetry: enable telemetry mode, to update"
1463 " empty polls, full polls, and core busyness to telemetry\n",
1467 static int parse_max_pkt_len(const char *pktlen)
1472 /* parse decimal string */
1473 len = strtoul(pktlen, &end, 10);
1474 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1484 parse_portmask(const char *portmask)
1489 /* parse hexadecimal string */
1490 pm = strtoul(portmask, &end, 16);
1491 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1501 parse_config(const char *q_arg)
1504 const char *p, *p0 = q_arg;
1512 unsigned long int_fld[_NUM_FLD];
1513 char *str_fld[_NUM_FLD];
1517 nb_lcore_params = 0;
1519 while ((p = strchr(p0,'(')) != NULL) {
1521 if((p0 = strchr(p,')')) == NULL)
1525 if(size >= sizeof(s))
1528 snprintf(s, sizeof(s), "%.*s", size, p);
1529 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1532 for (i = 0; i < _NUM_FLD; i++){
1534 int_fld[i] = strtoul(str_fld[i], &end, 0);
1535 if (errno != 0 || end == str_fld[i] || int_fld[i] >
1539 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1540 printf("exceeded max number of lcore params: %hu\n",
1544 lcore_params_array[nb_lcore_params].port_id =
1545 (uint8_t)int_fld[FLD_PORT];
1546 lcore_params_array[nb_lcore_params].queue_id =
1547 (uint8_t)int_fld[FLD_QUEUE];
1548 lcore_params_array[nb_lcore_params].lcore_id =
1549 (uint8_t)int_fld[FLD_LCORE];
1552 lcore_params = lcore_params_array;
1557 parse_ep_config(const char *q_arg)
1560 const char *p = q_arg;
1570 ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1571 ep_hgh_edpi = EMPTY_POLL_MED_THRESHOLD;
1573 strlcpy(s, p, sizeof(s));
1575 num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1577 empty_poll_train = false;
1584 training_flag = strtoul(str_fld[0], &end, 0);
1585 med_edpi = strtoul(str_fld[1], &end, 0);
1586 hgh_edpi = strtoul(str_fld[2], &end, 0);
1588 if (training_flag == 1)
1589 empty_poll_train = true;
1592 ep_med_edpi = med_edpi;
1595 ep_hgh_edpi = hgh_edpi;
1605 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1606 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1608 /* Parse the argument given in the command line of the application */
1610 parse_args(int argc, char **argv)
1616 char *prgname = argv[0];
1617 static struct option lgopts[] = {
1618 {"config", 1, 0, 0},
1619 {"perf-config", 1, 0, 0},
1620 {"high-perf-cores", 1, 0, 0},
1621 {"no-numa", 0, 0, 0},
1622 {"enable-jumbo", 0, 0, 0},
1623 {"empty-poll", 1, 0, 0},
1624 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1625 {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1631 while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1632 lgopts, &option_index)) != EOF) {
1637 enabled_port_mask = parse_portmask(optarg);
1638 if (enabled_port_mask == 0) {
1639 printf("invalid portmask\n");
1640 print_usage(prgname);
1645 printf("Promiscuous mode selected\n");
1649 limit = parse_max_pkt_len(optarg);
1650 freq_tlb[LOW] = limit;
1653 limit = parse_max_pkt_len(optarg);
1654 freq_tlb[MED] = limit;
1657 limit = parse_max_pkt_len(optarg);
1658 freq_tlb[HGH] = limit;
1662 if (!strncmp(lgopts[option_index].name, "config", 6)) {
1663 ret = parse_config(optarg);
1665 printf("invalid config\n");
1666 print_usage(prgname);
1671 if (!strncmp(lgopts[option_index].name,
1672 "perf-config", 11)) {
1673 ret = parse_perf_config(optarg);
1675 printf("invalid perf-config\n");
1676 print_usage(prgname);
1681 if (!strncmp(lgopts[option_index].name,
1682 "high-perf-cores", 15)) {
1683 ret = parse_perf_core_list(optarg);
1685 printf("invalid high-perf-cores\n");
1686 print_usage(prgname);
1691 if (!strncmp(lgopts[option_index].name,
1693 printf("numa is disabled \n");
1697 if (!strncmp(lgopts[option_index].name,
1698 "empty-poll", 10)) {
1699 if (app_mode == APP_MODE_TELEMETRY) {
1700 printf(" empty-poll cannot be enabled as telemetry mode is enabled\n");
1703 app_mode = APP_MODE_EMPTY_POLL;
1704 ret = parse_ep_config(optarg);
1707 printf("invalid empty poll config\n");
1708 print_usage(prgname);
1711 printf("empty-poll is enabled\n");
1714 if (!strncmp(lgopts[option_index].name,
1715 CMD_LINE_OPT_TELEMETRY,
1716 sizeof(CMD_LINE_OPT_TELEMETRY))) {
1717 if (app_mode == APP_MODE_EMPTY_POLL) {
1718 printf("telemetry mode cannot be enabled as empty poll mode is enabled\n");
1721 app_mode = APP_MODE_TELEMETRY;
1722 printf("telemetry mode is enabled\n");
1725 if (!strncmp(lgopts[option_index].name,
1726 "enable-jumbo", 12)) {
1727 struct option lenopts =
1728 {"max-pkt-len", required_argument, \
1731 printf("jumbo frame is enabled \n");
1732 port_conf.rxmode.offloads |=
1733 DEV_RX_OFFLOAD_JUMBO_FRAME;
1734 port_conf.txmode.offloads |=
1735 DEV_TX_OFFLOAD_MULTI_SEGS;
1738 * if no max-pkt-len set, use the default value
1741 if (0 == getopt_long(argc, argvopt, "",
1742 &lenopts, &option_index)) {
1743 ret = parse_max_pkt_len(optarg);
1745 (ret > MAX_JUMBO_PKT_LEN)){
1746 printf("invalid packet "
1748 print_usage(prgname);
1751 port_conf.rxmode.max_rx_pkt_len = ret;
1753 printf("set jumbo frame "
1754 "max packet length to %u\n",
1755 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1758 if (!strncmp(lgopts[option_index].name,
1759 CMD_LINE_OPT_PARSE_PTYPE,
1760 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1761 printf("soft parse-ptype is enabled\n");
1768 print_usage(prgname);
1774 argv[optind-1] = prgname;
1777 optind = 1; /* reset getopt lib */
1782 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1784 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1785 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1786 printf("%s%s", name, buf);
1789 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1791 setup_hash(int socketid)
1793 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1795 .entries = L3FWD_HASH_ENTRIES,
1796 .key_len = sizeof(struct ipv4_5tuple),
1797 .hash_func = DEFAULT_HASH_FUNC,
1798 .hash_func_init_val = 0,
1801 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1803 .entries = L3FWD_HASH_ENTRIES,
1804 .key_len = sizeof(struct ipv6_5tuple),
1805 .hash_func = DEFAULT_HASH_FUNC,
1806 .hash_func_init_val = 0,
1813 /* create ipv4 hash */
1814 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1815 ipv4_l3fwd_hash_params.name = s;
1816 ipv4_l3fwd_hash_params.socket_id = socketid;
1817 ipv4_l3fwd_lookup_struct[socketid] =
1818 rte_hash_create(&ipv4_l3fwd_hash_params);
1819 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1820 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1821 "socket %d\n", socketid);
1823 /* create ipv6 hash */
1824 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1825 ipv6_l3fwd_hash_params.name = s;
1826 ipv6_l3fwd_hash_params.socket_id = socketid;
1827 ipv6_l3fwd_lookup_struct[socketid] =
1828 rte_hash_create(&ipv6_l3fwd_hash_params);
1829 if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1830 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1831 "socket %d\n", socketid);
1834 /* populate the ipv4 hash */
1835 for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
1836 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1837 (void *) &ipv4_l3fwd_route_array[i].key);
1839 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1840 "l3fwd hash on socket %d\n", i, socketid);
1842 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1843 printf("Hash: Adding key\n");
1844 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1847 /* populate the ipv6 hash */
1848 for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
1849 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1850 (void *) &ipv6_l3fwd_route_array[i].key);
1852 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1853 "l3fwd hash on socket %d\n", i, socketid);
1855 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1856 printf("Hash: Adding key\n");
1857 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1862 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1864 setup_lpm(int socketid)
1870 /* create the LPM table */
1871 struct rte_lpm_config lpm_ipv4_config;
1873 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1874 lpm_ipv4_config.number_tbl8s = 256;
1875 lpm_ipv4_config.flags = 0;
1877 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1878 ipv4_l3fwd_lookup_struct[socketid] =
1879 rte_lpm_create(s, socketid, &lpm_ipv4_config);
1880 if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1881 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1882 " on socket %d\n", socketid);
1884 /* populate the LPM table */
1885 for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
1886 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1887 ipv4_l3fwd_route_array[i].ip,
1888 ipv4_l3fwd_route_array[i].depth,
1889 ipv4_l3fwd_route_array[i].if_out);
1892 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1893 "l3fwd LPM table on socket %d\n",
1897 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1898 (unsigned)ipv4_l3fwd_route_array[i].ip,
1899 ipv4_l3fwd_route_array[i].depth,
1900 ipv4_l3fwd_route_array[i].if_out);
1906 init_mem(unsigned nb_mbuf)
1908 struct lcore_conf *qconf;
1913 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1914 if (rte_lcore_is_enabled(lcore_id) == 0)
1918 socketid = rte_lcore_to_socket_id(lcore_id);
1922 if (socketid >= NB_SOCKETS) {
1923 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1924 "out of range %d\n", socketid,
1925 lcore_id, NB_SOCKETS);
1927 if (pktmbuf_pool[socketid] == NULL) {
1928 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1929 pktmbuf_pool[socketid] =
1930 rte_pktmbuf_pool_create(s, nb_mbuf,
1931 MEMPOOL_CACHE_SIZE, 0,
1932 RTE_MBUF_DEFAULT_BUF_SIZE,
1934 if (pktmbuf_pool[socketid] == NULL)
1935 rte_exit(EXIT_FAILURE,
1936 "Cannot init mbuf pool on socket %d\n",
1939 printf("Allocated mbuf pool on socket %d\n",
1942 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1943 setup_lpm(socketid);
1945 setup_hash(socketid);
1948 qconf = &lcore_conf[lcore_id];
1949 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1950 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1951 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1957 /* Check the link status of all ports in up to 9s, and print them finally */
1959 check_all_ports_link_status(uint32_t port_mask)
1961 #define CHECK_INTERVAL 100 /* 100ms */
1962 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1963 uint8_t count, all_ports_up, print_flag = 0;
1965 struct rte_eth_link link;
1968 printf("\nChecking link status");
1970 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1972 RTE_ETH_FOREACH_DEV(portid) {
1973 if ((port_mask & (1 << portid)) == 0)
1975 memset(&link, 0, sizeof(link));
1976 ret = rte_eth_link_get_nowait(portid, &link);
1979 if (print_flag == 1)
1980 printf("Port %u link get failed: %s\n",
1981 portid, rte_strerror(-ret));
1984 /* print link status if flag set */
1985 if (print_flag == 1) {
1986 if (link.link_status)
1987 printf("Port %d Link Up - speed %u "
1988 "Mbps - %s\n", (uint8_t)portid,
1989 (unsigned)link.link_speed,
1990 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1991 ("full-duplex") : ("half-duplex\n"));
1993 printf("Port %d Link Down\n",
1997 /* clear all_ports_up flag if any link down */
1998 if (link.link_status == ETH_LINK_DOWN) {
2003 /* after finally printing all link status, get out */
2004 if (print_flag == 1)
2007 if (all_ports_up == 0) {
2010 rte_delay_ms(CHECK_INTERVAL);
2013 /* set the print_flag if all ports up or timeout */
2014 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2021 static int check_ptype(uint16_t portid)
2024 int ptype_l3_ipv4 = 0;
2025 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2026 int ptype_l3_ipv6 = 0;
2028 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2030 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2034 uint32_t ptypes[ret];
2036 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2037 for (i = 0; i < ret; ++i) {
2038 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2040 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2041 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2046 if (ptype_l3_ipv4 == 0)
2047 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2049 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2050 if (ptype_l3_ipv6 == 0)
2051 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2054 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2056 #else /* APP_LOOKUP_EXACT_MATCH */
2057 if (ptype_l3_ipv4 && ptype_l3_ipv6)
2066 init_power_library(void)
2068 int ret = 0, lcore_id;
2069 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2070 if (rte_lcore_is_enabled(lcore_id)) {
2071 /* init power management library */
2072 ret = rte_power_init(lcore_id);
2075 "Library initialization failed on core %u\n",
2082 update_telemetry(__rte_unused struct rte_timer *tim,
2083 __rte_unused void *arg)
2085 unsigned int lcore_id = rte_lcore_id();
2086 struct lcore_conf *qconf;
2087 uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2088 uint64_t values[3] = {0};
2092 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2093 qconf = &lcore_conf[lcore_id];
2094 if (qconf->n_rx_queue == 0)
2097 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2098 app_eps += stats[lcore_id].ep_nep[1];
2099 app_fps += stats[lcore_id].fp_nfp[1];
2100 app_br += stats[lcore_id].br;
2101 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2105 values[0] = app_eps/count;
2106 values[1] = app_fps/count;
2107 values[2] = app_br/count;
2114 ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2115 values, RTE_DIM(values));
2117 RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2120 telemetry_setup_timer(void)
2122 int lcore_id = rte_lcore_id();
2123 uint64_t hz = rte_get_timer_hz();
2126 ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2127 rte_timer_reset_sync(&telemetry_timer,
2135 empty_poll_setup_timer(void)
2137 int lcore_id = rte_lcore_id();
2138 uint64_t hz = rte_get_timer_hz();
2140 struct ep_params *ep_ptr = ep_params;
2142 ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2144 rte_timer_reset_sync(&ep_ptr->timer0,
2145 ep_ptr->interval_ticks,
2148 rte_empty_poll_detection,
2153 launch_timer(unsigned int lcore_id)
2155 int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2157 RTE_SET_USED(lcore_id);
2160 if (rte_get_master_lcore() != lcore_id) {
2161 rte_panic("timer on lcore:%d which is not master core:%d\n",
2163 rte_get_master_lcore());
2166 RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2168 if (app_mode == APP_MODE_EMPTY_POLL)
2169 empty_poll_setup_timer();
2171 telemetry_setup_timer();
2173 cycles_10ms = rte_get_timer_hz() / 100;
2175 while (!is_done()) {
2176 cur_tsc = rte_rdtsc();
2177 diff_tsc = cur_tsc - prev_tsc;
2178 if (diff_tsc > cycles_10ms) {
2181 cycles_10ms = rte_get_timer_hz() / 100;
2185 RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2192 main(int argc, char **argv)
2194 struct lcore_conf *qconf;
2195 struct rte_eth_dev_info dev_info;
2196 struct rte_eth_txconf *txconf;
2202 uint32_t n_tx_queue, nb_lcores;
2203 uint32_t dev_rxq_num, dev_txq_num;
2204 uint8_t nb_rx_queue, queue, socketid;
2206 uint8_t num_telstats = RTE_DIM(telstats_strings);
2207 const char *ptr_strings[num_telstats];
2209 /* catch SIGINT and restore cpufreq governor to ondemand */
2210 signal(SIGINT, signal_exit_now);
2213 ret = rte_eal_init(argc, argv);
2215 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2219 /* init RTE timer library to be used late */
2220 rte_timer_subsystem_init();
2222 /* parse application arguments (after the EAL ones) */
2223 ret = parse_args(argc, argv);
2225 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2227 if (init_power_library())
2228 RTE_LOG(ERR, L3FWD_POWER, "init_power_library failed\n");
2230 if (update_lcore_params() < 0)
2231 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2233 if (check_lcore_params() < 0)
2234 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2236 ret = init_lcore_rx_queues();
2238 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2240 nb_ports = rte_eth_dev_count_avail();
2242 if (check_port_config() < 0)
2243 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2245 nb_lcores = rte_lcore_count();
2247 /* initialize all ports */
2248 RTE_ETH_FOREACH_DEV(portid) {
2249 struct rte_eth_conf local_port_conf = port_conf;
2251 /* skip ports that are not enabled */
2252 if ((enabled_port_mask & (1 << portid)) == 0) {
2253 printf("\nSkipping disabled port %d\n", portid);
2258 printf("Initializing port %d ... ", portid );
2261 ret = rte_eth_dev_info_get(portid, &dev_info);
2263 rte_exit(EXIT_FAILURE,
2264 "Error during getting device (port %u) info: %s\n",
2265 portid, strerror(-ret));
2267 dev_rxq_num = dev_info.max_rx_queues;
2268 dev_txq_num = dev_info.max_tx_queues;
2270 nb_rx_queue = get_port_n_rx_queues(portid);
2271 if (nb_rx_queue > dev_rxq_num)
2272 rte_exit(EXIT_FAILURE,
2273 "Cannot configure not existed rxq: "
2274 "port=%d\n", portid);
2276 n_tx_queue = nb_lcores;
2277 if (n_tx_queue > dev_txq_num)
2278 n_tx_queue = dev_txq_num;
2279 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2280 nb_rx_queue, (unsigned)n_tx_queue );
2281 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2282 if (nb_rx_queue == 0)
2283 local_port_conf.intr_conf.rxq = 0;
2285 ret = rte_eth_dev_info_get(portid, &dev_info);
2287 rte_exit(EXIT_FAILURE,
2288 "Error during getting device (port %u) info: %s\n",
2289 portid, strerror(-ret));
2291 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2292 local_port_conf.txmode.offloads |=
2293 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2295 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2296 dev_info.flow_type_rss_offloads;
2297 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2298 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2299 printf("Port %u modified RSS hash function based on hardware support,"
2300 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2302 port_conf.rx_adv_conf.rss_conf.rss_hf,
2303 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2306 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2307 (uint16_t)n_tx_queue, &local_port_conf);
2309 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2310 "err=%d, port=%d\n", ret, portid);
2312 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2315 rte_exit(EXIT_FAILURE,
2316 "Cannot adjust number of descriptors: err=%d, port=%d\n",
2319 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2321 rte_exit(EXIT_FAILURE,
2322 "Cannot get MAC address: err=%d, port=%d\n",
2325 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2329 ret = init_mem(NB_MBUF);
2331 rte_exit(EXIT_FAILURE, "init_mem failed\n");
2333 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2334 if (rte_lcore_is_enabled(lcore_id) == 0)
2337 /* Initialize TX buffers */
2338 qconf = &lcore_conf[lcore_id];
2339 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2340 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2341 rte_eth_dev_socket_id(portid));
2342 if (qconf->tx_buffer[portid] == NULL)
2343 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2346 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2349 /* init one TX queue per couple (lcore,port) */
2351 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2352 if (rte_lcore_is_enabled(lcore_id) == 0)
2355 if (queueid >= dev_txq_num)
2360 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2364 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2367 txconf = &dev_info.default_txconf;
2368 txconf->offloads = local_port_conf.txmode.offloads;
2369 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2372 rte_exit(EXIT_FAILURE,
2373 "rte_eth_tx_queue_setup: err=%d, "
2374 "port=%d\n", ret, portid);
2376 qconf = &lcore_conf[lcore_id];
2377 qconf->tx_queue_id[portid] = queueid;
2380 qconf->tx_port_id[qconf->n_tx_port] = portid;
2386 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2387 if (rte_lcore_is_enabled(lcore_id) == 0)
2390 if (app_mode == APP_MODE_LEGACY) {
2391 /* init timer structures for each enabled lcore */
2392 rte_timer_init(&power_timers[lcore_id]);
2393 hz = rte_get_timer_hz();
2394 rte_timer_reset(&power_timers[lcore_id],
2395 hz/TIMER_NUMBER_PER_SECOND,
2397 power_timer_cb, NULL);
2399 qconf = &lcore_conf[lcore_id];
2400 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2402 /* init RX queues */
2403 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2404 struct rte_eth_rxconf rxq_conf;
2406 portid = qconf->rx_queue_list[queue].port_id;
2407 queueid = qconf->rx_queue_list[queue].queue_id;
2411 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2415 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2418 ret = rte_eth_dev_info_get(portid, &dev_info);
2420 rte_exit(EXIT_FAILURE,
2421 "Error during getting device (port %u) info: %s\n",
2422 portid, strerror(-ret));
2424 rxq_conf = dev_info.default_rxconf;
2425 rxq_conf.offloads = port_conf.rxmode.offloads;
2426 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2427 socketid, &rxq_conf,
2428 pktmbuf_pool[socketid]);
2430 rte_exit(EXIT_FAILURE,
2431 "rte_eth_rx_queue_setup: err=%d, "
2432 "port=%d\n", ret, portid);
2435 if (add_cb_parse_ptype(portid, queueid) < 0)
2436 rte_exit(EXIT_FAILURE,
2437 "Fail to add ptype cb\n");
2438 } else if (!check_ptype(portid))
2439 rte_exit(EXIT_FAILURE,
2440 "PMD can not provide needed ptypes\n");
2447 RTE_ETH_FOREACH_DEV(portid) {
2448 if ((enabled_port_mask & (1 << portid)) == 0) {
2452 ret = rte_eth_dev_start(portid);
2454 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2455 "port=%d\n", ret, portid);
2457 * If enabled, put device in promiscuous mode.
2458 * This allows IO forwarding mode to forward packets
2459 * to itself through 2 cross-connected ports of the
2462 if (promiscuous_on) {
2463 ret = rte_eth_promiscuous_enable(portid);
2465 rte_exit(EXIT_FAILURE,
2466 "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2467 rte_strerror(-ret), portid);
2469 /* initialize spinlock for each port */
2470 rte_spinlock_init(&(locks[portid]));
2473 check_all_ports_link_status(enabled_port_mask);
2475 if (app_mode == APP_MODE_EMPTY_POLL) {
2477 if (empty_poll_train) {
2478 policy.state = TRAINING;
2480 policy.state = MED_NORMAL;
2481 policy.med_base_edpi = ep_med_edpi;
2482 policy.hgh_base_edpi = ep_hgh_edpi;
2485 ret = rte_power_empty_poll_stat_init(&ep_params,
2489 rte_exit(EXIT_FAILURE, "empty poll init failed");
2493 /* launch per-lcore init on every lcore */
2494 if (app_mode == APP_MODE_LEGACY) {
2495 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2496 } else if (app_mode == APP_MODE_EMPTY_POLL) {
2497 empty_poll_stop = false;
2498 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2503 /* Init metrics library */
2504 rte_metrics_init(rte_socket_id());
2505 /** Register stats with metrics library */
2506 for (i = 0; i < num_telstats; i++)
2507 ptr_strings[i] = telstats_strings[i].name;
2509 ret = rte_metrics_reg_names(ptr_strings, num_telstats);
2511 telstats_index = ret;
2513 rte_exit(EXIT_FAILURE, "failed to register metrics names");
2515 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2516 rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2518 rte_timer_init(&telemetry_timer);
2519 rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2523 if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2524 launch_timer(rte_lcore_id());
2526 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2527 if (rte_eal_wait_lcore(lcore_id) < 0)
2531 if (app_mode == APP_MODE_EMPTY_POLL)
2532 rte_power_empty_poll_stat_free();