1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/queue.h>
16 #include <sys/socket.h>
17 #include <arpa/inet.h>
19 #include <rte_debug.h>
20 #include <rte_ether.h>
21 #include <rte_ethdev.h>
22 #include <rte_cycles.h>
31 #include "l3fwd_event.h"
33 struct ipv4_l3fwd_lpm_route {
39 struct ipv6_l3fwd_lpm_route {
45 /* 198.18.0.0/16 are set aside for RFC2544 benchmarking (RFC5735). */
46 static struct ipv4_l3fwd_lpm_route ipv4_l3fwd_lpm_route_array[] = {
47 {RTE_IPV4(198, 18, 0, 0), 24, 0},
48 {RTE_IPV4(198, 18, 1, 0), 24, 1},
49 {RTE_IPV4(198, 18, 2, 0), 24, 2},
50 {RTE_IPV4(198, 18, 3, 0), 24, 3},
51 {RTE_IPV4(198, 18, 4, 0), 24, 4},
52 {RTE_IPV4(198, 18, 5, 0), 24, 5},
53 {RTE_IPV4(198, 18, 6, 0), 24, 6},
54 {RTE_IPV4(198, 18, 7, 0), 24, 7},
57 /* 2001:0200::/48 is IANA reserved range for IPv6 benchmarking (RFC5180) */
58 static struct ipv6_l3fwd_lpm_route ipv6_l3fwd_lpm_route_array[] = {
59 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, 48, 0},
60 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0}, 48, 1},
61 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0}, 48, 2},
62 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0}, 48, 3},
63 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0}, 48, 4},
64 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0}, 48, 5},
65 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0}, 48, 6},
66 {{32, 1, 2, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0}, 48, 7},
69 #define IPV4_L3FWD_LPM_MAX_RULES 1024
70 #define IPV4_L3FWD_LPM_NUMBER_TBL8S (1 << 8)
71 #define IPV6_L3FWD_LPM_MAX_RULES 1024
72 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
74 struct rte_lpm *ipv4_l3fwd_lpm_lookup_struct[NB_SOCKETS];
75 struct rte_lpm6 *ipv6_l3fwd_lpm_lookup_struct[NB_SOCKETS];
77 static inline uint16_t
78 lpm_get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, void *lookup_struct)
81 struct rte_lpm *ipv4_l3fwd_lookup_struct =
82 (struct rte_lpm *)lookup_struct;
84 return (uint16_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
85 rte_be_to_cpu_32(((struct rte_ipv4_hdr *)ipv4_hdr)->dst_addr),
86 &next_hop) == 0) ? next_hop : portid);
89 static inline uint16_t
90 lpm_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)
93 struct rte_lpm6 *ipv6_l3fwd_lookup_struct =
94 (struct rte_lpm6 *)lookup_struct;
96 return (uint16_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
97 ((struct rte_ipv6_hdr *)ipv6_hdr)->dst_addr,
98 &next_hop) == 0) ? next_hop : portid);
101 static __rte_always_inline uint16_t
102 lpm_get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
105 struct rte_ipv6_hdr *ipv6_hdr;
106 struct rte_ipv4_hdr *ipv4_hdr;
107 struct rte_ether_hdr *eth_hdr;
109 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
111 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
112 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
114 return lpm_get_ipv4_dst_port(ipv4_hdr, portid,
115 qconf->ipv4_lookup_struct);
116 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
118 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
119 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
121 return lpm_get_ipv6_dst_port(ipv6_hdr, portid,
122 qconf->ipv6_lookup_struct);
129 * lpm_get_dst_port optimized routine for packets where dst_ipv4 is already
130 * precalculated. If packet is ipv6 dst_addr is taken directly from packet
131 * header and dst_ipv4 value is not used.
133 static __rte_always_inline uint16_t
134 lpm_get_dst_port_with_ipv4(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
135 uint32_t dst_ipv4, uint16_t portid)
138 struct rte_ipv6_hdr *ipv6_hdr;
139 struct rte_ether_hdr *eth_hdr;
141 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
142 return (uint16_t) ((rte_lpm_lookup(qconf->ipv4_lookup_struct,
143 dst_ipv4, &next_hop) == 0)
144 ? next_hop : portid);
146 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
148 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
149 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
151 return (uint16_t) ((rte_lpm6_lookup(qconf->ipv6_lookup_struct,
152 ipv6_hdr->dst_addr, &next_hop) == 0)
153 ? next_hop : portid);
160 #if defined(RTE_ARCH_X86)
161 #include "l3fwd_lpm_sse.h"
162 #elif defined RTE_MACHINE_CPUFLAG_NEON
163 #include "l3fwd_lpm_neon.h"
164 #elif defined(RTE_ARCH_PPC_64)
165 #include "l3fwd_lpm_altivec.h"
167 #include "l3fwd_lpm.h"
170 /* main processing loop */
172 lpm_main_loop(__attribute__((unused)) void *dummy)
174 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
176 uint64_t prev_tsc, diff_tsc, cur_tsc;
180 struct lcore_conf *qconf;
181 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
182 US_PER_S * BURST_TX_DRAIN_US;
186 lcore_id = rte_lcore_id();
187 qconf = &lcore_conf[lcore_id];
189 if (qconf->n_rx_queue == 0) {
190 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
194 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
196 for (i = 0; i < qconf->n_rx_queue; i++) {
198 portid = qconf->rx_queue_list[i].port_id;
199 queueid = qconf->rx_queue_list[i].queue_id;
201 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
202 lcore_id, portid, queueid);
205 while (!force_quit) {
207 cur_tsc = rte_rdtsc();
210 * TX burst queue drain
212 diff_tsc = cur_tsc - prev_tsc;
213 if (unlikely(diff_tsc > drain_tsc)) {
215 for (i = 0; i < qconf->n_tx_port; ++i) {
216 portid = qconf->tx_port_id[i];
217 if (qconf->tx_mbufs[portid].len == 0)
220 qconf->tx_mbufs[portid].len,
222 qconf->tx_mbufs[portid].len = 0;
229 * Read packet from RX queues
231 for (i = 0; i < qconf->n_rx_queue; ++i) {
232 portid = qconf->rx_queue_list[i].port_id;
233 queueid = qconf->rx_queue_list[i].queue_id;
234 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
239 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON \
240 || defined RTE_ARCH_PPC_64
241 l3fwd_lpm_send_packets(nb_rx, pkts_burst,
244 l3fwd_lpm_no_opt_send_packets(nb_rx, pkts_burst,
253 static __rte_always_inline uint16_t
254 lpm_process_event_pkt(const struct lcore_conf *lconf, struct rte_mbuf *mbuf)
256 mbuf->port = lpm_get_dst_port(lconf, mbuf, mbuf->port);
258 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON \
259 || defined RTE_ARCH_PPC_64
260 process_packet(mbuf, &mbuf->port);
263 struct rte_ether_hdr *eth_hdr = rte_pktmbuf_mtod(mbuf,
264 struct rte_ether_hdr *);
265 #ifdef DO_RFC_1812_CHECKS
266 struct rte_ipv4_hdr *ipv4_hdr;
267 if (RTE_ETH_IS_IPV4_HDR(mbuf->packet_type)) {
268 /* Handle IPv4 headers.*/
269 ipv4_hdr = rte_pktmbuf_mtod_offset(mbuf,
270 struct rte_ipv4_hdr *,
271 sizeof(struct rte_ether_hdr));
273 if (is_valid_ipv4_pkt(ipv4_hdr, mbuf->pkt_len)
275 mbuf->port = BAD_PORT;
278 /* Update time to live and header checksum */
279 --(ipv4_hdr->time_to_live);
280 ++(ipv4_hdr->hdr_checksum);
284 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[mbuf->port];
287 rte_ether_addr_copy(&ports_eth_addr[mbuf->port],
293 static __rte_always_inline void
294 lpm_event_loop_single(struct l3fwd_event_resources *evt_rsrc,
297 const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
298 const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
299 evt_rsrc->evq.nb_queues - 1];
300 const uint8_t event_d_id = evt_rsrc->event_d_id;
301 struct lcore_conf *lconf;
302 unsigned int lcore_id;
308 lcore_id = rte_lcore_id();
309 lconf = &lcore_conf[lcore_id];
311 RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
312 while (!force_quit) {
313 if (!rte_event_dequeue_burst(event_d_id, event_p_id, &ev, 1, 0))
316 if (lpm_process_event_pkt(lconf, ev.mbuf) == BAD_PORT) {
317 rte_pktmbuf_free(ev.mbuf);
321 if (flags & L3FWD_EVENT_TX_ENQ) {
322 ev.queue_id = tx_q_id;
323 ev.op = RTE_EVENT_OP_FORWARD;
324 while (rte_event_enqueue_burst(event_d_id, event_p_id,
325 &ev, 1) && !force_quit)
329 if (flags & L3FWD_EVENT_TX_DIRECT) {
330 rte_event_eth_tx_adapter_txq_set(ev.mbuf, 0);
331 while (!rte_event_eth_tx_adapter_enqueue(event_d_id,
332 event_p_id, &ev, 1, 0) &&
339 static __rte_always_inline void
340 lpm_event_loop_burst(struct l3fwd_event_resources *evt_rsrc,
343 const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
344 const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
345 evt_rsrc->evq.nb_queues - 1];
346 const uint8_t event_d_id = evt_rsrc->event_d_id;
347 const uint16_t deq_len = evt_rsrc->deq_depth;
348 struct rte_event events[MAX_PKT_BURST];
349 struct lcore_conf *lconf;
350 unsigned int lcore_id;
351 int i, nb_enq, nb_deq;
356 lcore_id = rte_lcore_id();
358 lconf = &lcore_conf[lcore_id];
360 RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
362 while (!force_quit) {
363 /* Read events from RX queues */
364 nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id,
371 for (i = 0; i < nb_deq; i++) {
372 if (flags & L3FWD_EVENT_TX_ENQ) {
373 events[i].queue_id = tx_q_id;
374 events[i].op = RTE_EVENT_OP_FORWARD;
377 if (flags & L3FWD_EVENT_TX_DIRECT)
378 rte_event_eth_tx_adapter_txq_set(events[i].mbuf,
381 lpm_process_event_pkt(lconf, events[i].mbuf);
384 if (flags & L3FWD_EVENT_TX_ENQ) {
385 nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
387 while (nb_enq < nb_deq && !force_quit)
388 nb_enq += rte_event_enqueue_burst(event_d_id,
389 event_p_id, events + nb_enq,
393 if (flags & L3FWD_EVENT_TX_DIRECT) {
394 nb_enq = rte_event_eth_tx_adapter_enqueue(event_d_id,
395 event_p_id, events, nb_deq, 0);
396 while (nb_enq < nb_deq && !force_quit)
397 nb_enq += rte_event_eth_tx_adapter_enqueue(
398 event_d_id, event_p_id,
405 static __rte_always_inline void
406 lpm_event_loop(struct l3fwd_event_resources *evt_rsrc,
409 if (flags & L3FWD_EVENT_SINGLE)
410 lpm_event_loop_single(evt_rsrc, flags);
411 if (flags & L3FWD_EVENT_BURST)
412 lpm_event_loop_burst(evt_rsrc, flags);
416 lpm_event_main_loop_tx_d(__attribute__((unused)) void *dummy)
418 struct l3fwd_event_resources *evt_rsrc =
419 l3fwd_get_eventdev_rsrc();
421 lpm_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT | L3FWD_EVENT_SINGLE);
426 lpm_event_main_loop_tx_d_burst(__attribute__((unused)) void *dummy)
428 struct l3fwd_event_resources *evt_rsrc =
429 l3fwd_get_eventdev_rsrc();
431 lpm_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT | L3FWD_EVENT_BURST);
436 lpm_event_main_loop_tx_q(__attribute__((unused)) void *dummy)
438 struct l3fwd_event_resources *evt_rsrc =
439 l3fwd_get_eventdev_rsrc();
441 lpm_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ | L3FWD_EVENT_SINGLE);
446 lpm_event_main_loop_tx_q_burst(__attribute__((unused)) void *dummy)
448 struct l3fwd_event_resources *evt_rsrc =
449 l3fwd_get_eventdev_rsrc();
451 lpm_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ | L3FWD_EVENT_BURST);
456 setup_lpm(const int socketid)
458 struct rte_lpm6_config config;
459 struct rte_lpm_config config_ipv4;
463 char abuf[INET6_ADDRSTRLEN];
465 /* create the LPM table */
466 config_ipv4.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
467 config_ipv4.number_tbl8s = IPV4_L3FWD_LPM_NUMBER_TBL8S;
468 config_ipv4.flags = 0;
469 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
470 ipv4_l3fwd_lpm_lookup_struct[socketid] =
471 rte_lpm_create(s, socketid, &config_ipv4);
472 if (ipv4_l3fwd_lpm_lookup_struct[socketid] == NULL)
473 rte_exit(EXIT_FAILURE,
474 "Unable to create the l3fwd LPM table on socket %d\n",
477 /* populate the LPM table */
478 for (i = 0; i < RTE_DIM(ipv4_l3fwd_lpm_route_array); i++) {
481 /* skip unused ports */
482 if ((1 << ipv4_l3fwd_lpm_route_array[i].if_out &
483 enabled_port_mask) == 0)
486 ret = rte_lpm_add(ipv4_l3fwd_lpm_lookup_struct[socketid],
487 ipv4_l3fwd_lpm_route_array[i].ip,
488 ipv4_l3fwd_lpm_route_array[i].depth,
489 ipv4_l3fwd_lpm_route_array[i].if_out);
492 rte_exit(EXIT_FAILURE,
493 "Unable to add entry %u to the l3fwd LPM table on socket %d\n",
497 in.s_addr = htonl(ipv4_l3fwd_lpm_route_array[i].ip);
498 printf("LPM: Adding route %s / %d (%d)\n",
499 inet_ntop(AF_INET, &in, abuf, sizeof(abuf)),
500 ipv4_l3fwd_lpm_route_array[i].depth,
501 ipv4_l3fwd_lpm_route_array[i].if_out);
504 /* create the LPM6 table */
505 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
507 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
508 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
510 ipv6_l3fwd_lpm_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
512 if (ipv6_l3fwd_lpm_lookup_struct[socketid] == NULL)
513 rte_exit(EXIT_FAILURE,
514 "Unable to create the l3fwd LPM table on socket %d\n",
517 /* populate the LPM table */
518 for (i = 0; i < RTE_DIM(ipv6_l3fwd_lpm_route_array); i++) {
520 /* skip unused ports */
521 if ((1 << ipv6_l3fwd_lpm_route_array[i].if_out &
522 enabled_port_mask) == 0)
525 ret = rte_lpm6_add(ipv6_l3fwd_lpm_lookup_struct[socketid],
526 ipv6_l3fwd_lpm_route_array[i].ip,
527 ipv6_l3fwd_lpm_route_array[i].depth,
528 ipv6_l3fwd_lpm_route_array[i].if_out);
531 rte_exit(EXIT_FAILURE,
532 "Unable to add entry %u to the l3fwd LPM table on socket %d\n",
536 printf("LPM: Adding route %s / %d (%d)\n",
537 inet_ntop(AF_INET6, ipv6_l3fwd_lpm_route_array[i].ip,
539 ipv6_l3fwd_lpm_route_array[i].depth,
540 ipv6_l3fwd_lpm_route_array[i].if_out);
545 lpm_check_ptype(int portid)
548 int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
549 uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
551 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
555 uint32_t ptypes[ret];
557 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
558 for (i = 0; i < ret; ++i) {
559 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
561 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
565 if (ptype_l3_ipv4 == 0)
566 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
568 if (ptype_l3_ipv6 == 0)
569 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
571 if (ptype_l3_ipv4 && ptype_l3_ipv6)
579 lpm_parse_ptype(struct rte_mbuf *m)
581 struct rte_ether_hdr *eth_hdr;
582 uint32_t packet_type = RTE_PTYPE_UNKNOWN;
585 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
586 ether_type = eth_hdr->ether_type;
587 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
588 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
589 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
590 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
592 m->packet_type = packet_type;
596 lpm_cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
597 struct rte_mbuf *pkts[], uint16_t nb_pkts,
598 uint16_t max_pkts __rte_unused,
599 void *user_param __rte_unused)
603 if (unlikely(nb_pkts == 0))
605 rte_prefetch0(rte_pktmbuf_mtod(pkts[0], struct ether_hdr *));
606 for (i = 0; i < (unsigned int) (nb_pkts - 1); ++i) {
607 rte_prefetch0(rte_pktmbuf_mtod(pkts[i+1],
608 struct ether_hdr *));
609 lpm_parse_ptype(pkts[i]);
611 lpm_parse_ptype(pkts[i]);
616 /* Return ipv4/ipv6 lpm fwd lookup struct. */
618 lpm_get_ipv4_l3fwd_lookup_struct(const int socketid)
620 return ipv4_l3fwd_lpm_lookup_struct[socketid];
624 lpm_get_ipv6_l3fwd_lookup_struct(const int socketid)
626 return ipv6_l3fwd_lpm_lookup_struct[socketid];