examples/l3fwd: skip Tx queue drain on first FIB iteration
[dpdk.git] / examples / l3fwd / l3fwd_fib.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2021 Intel Corporation
3  */
4
5 #include <stdio.h>
6 #include <stddef.h>
7 #include <stdint.h>
8 #include <sys/socket.h>
9 #include <arpa/inet.h>
10
11 #include <rte_fib.h>
12 #include <rte_fib6.h>
13
14 #include "l3fwd.h"
15 #if defined RTE_ARCH_X86
16 #include "l3fwd_sse.h"
17 #elif defined __ARM_NEON
18 #include "l3fwd_neon.h"
19 #elif defined RTE_ARCH_PPC_64
20 #include "l3fwd_altivec.h"
21 #endif
22 #include "l3fwd_event.h"
23 #include "l3fwd_route.h"
24
25 /* Configure how many packets ahead to prefetch for fib. */
26 #define FIB_PREFETCH_OFFSET 4
27
28 /* A non-existent portid is needed to denote a default hop for fib. */
29 #define FIB_DEFAULT_HOP 999
30
31 /*
32  * If the machine has SSE, NEON or PPC 64 then multiple packets
33  * can be sent at once if not only single packets will be sent
34  */
35 #if defined RTE_ARCH_X86 || defined __ARM_NEON \
36                 || defined RTE_ARCH_PPC_64
37 #define FIB_SEND_MULTI
38 #endif
39
40 static struct rte_fib *ipv4_l3fwd_fib_lookup_struct[NB_SOCKETS];
41 static struct rte_fib6 *ipv6_l3fwd_fib_lookup_struct[NB_SOCKETS];
42
43 /* Parse packet type and ip address. */
44 static inline void
45 fib_parse_packet(struct rte_mbuf *mbuf,
46                 uint32_t *ipv4, uint32_t *ipv4_cnt,
47                 uint8_t ipv6[RTE_FIB6_IPV6_ADDR_SIZE],
48                 uint32_t *ipv6_cnt, uint8_t *ip_type)
49 {
50         struct rte_ether_hdr *eth_hdr;
51         struct rte_ipv4_hdr *ipv4_hdr;
52         struct rte_ipv6_hdr *ipv6_hdr;
53
54         eth_hdr = rte_pktmbuf_mtod(mbuf, struct rte_ether_hdr *);
55         /* IPv4 */
56         if (mbuf->packet_type & RTE_PTYPE_L3_IPV4) {
57                 ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
58                 *ipv4 = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
59                 /* Store type of packet in type_arr (IPv4=1, IPv6=0). */
60                 *ip_type = 1;
61                 (*ipv4_cnt)++;
62         }
63         /* IPv6 */
64         else {
65                 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
66                 rte_mov16(ipv6, (const uint8_t *)ipv6_hdr->dst_addr);
67                 *ip_type = 0;
68                 (*ipv6_cnt)++;
69         }
70 }
71
72 /*
73  * If the machine does not have SSE, NEON or PPC 64 then the packets
74  * are sent one at a time using send_single_packet()
75  */
76 #if !defined FIB_SEND_MULTI
77 static inline void
78 fib_send_single(int nb_tx, struct lcore_conf *qconf,
79                 struct rte_mbuf **pkts_burst, uint16_t hops[nb_tx])
80 {
81         int32_t j;
82         struct rte_ether_hdr *eth_hdr;
83
84         for (j = 0; j < nb_tx; j++) {
85                 /* Run rfc1812 if packet is ipv4 and checks enabled. */
86 #if defined DO_RFC_1812_CHECKS
87                 rfc1812_process((struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(
88                                 pkts_burst[j], struct rte_ether_hdr *) + 1),
89                                 &hops[j], pkts_burst[j]->packet_type);
90 #endif
91
92                 /* Set MAC addresses. */
93                 eth_hdr = rte_pktmbuf_mtod(pkts_burst[j],
94                                 struct rte_ether_hdr *);
95                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[hops[j]];
96                 rte_ether_addr_copy(&ports_eth_addr[hops[j]],
97                                 &eth_hdr->s_addr);
98
99                 /* Send single packet. */
100                 send_single_packet(qconf, pkts_burst[j], hops[j]);
101         }
102 }
103 #endif
104
105 /* Bulk parse, fib lookup and send. */
106 static inline void
107 fib_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
108                 uint16_t portid, struct lcore_conf *qconf)
109 {
110         uint32_t ipv4_arr[nb_rx];
111         uint8_t ipv6_arr[nb_rx][RTE_FIB6_IPV6_ADDR_SIZE];
112         uint16_t hops[nb_rx];
113         uint64_t hopsv4[nb_rx], hopsv6[nb_rx];
114         uint8_t type_arr[nb_rx];
115         uint32_t ipv4_cnt = 0, ipv6_cnt = 0;
116         uint32_t ipv4_arr_assem = 0, ipv6_arr_assem = 0;
117         uint16_t nh;
118         int32_t i;
119
120         /* Prefetch first packets. */
121         for (i = 0; i < FIB_PREFETCH_OFFSET && i < nb_rx; i++)
122                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i], void *));
123
124         /* Parse packet info and prefetch. */
125         for (i = 0; i < (nb_rx - FIB_PREFETCH_OFFSET); i++) {
126                 /* Prefetch packet. */
127                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
128                                 i + FIB_PREFETCH_OFFSET], void *));
129                 fib_parse_packet(pkts_burst[i],
130                                 &ipv4_arr[ipv4_cnt], &ipv4_cnt,
131                                 ipv6_arr[ipv6_cnt], &ipv6_cnt,
132                                 &type_arr[i]);
133         }
134
135         /* Parse remaining packet info. */
136         for (; i < nb_rx; i++)
137                 fib_parse_packet(pkts_burst[i],
138                                 &ipv4_arr[ipv4_cnt], &ipv4_cnt,
139                                 ipv6_arr[ipv6_cnt], &ipv6_cnt,
140                                 &type_arr[i]);
141
142         /* Lookup IPv4 hops if IPv4 packets are present. */
143         if (likely(ipv4_cnt > 0))
144                 rte_fib_lookup_bulk(qconf->ipv4_lookup_struct,
145                                 ipv4_arr, hopsv4, ipv4_cnt);
146
147         /* Lookup IPv6 hops if IPv6 packets are present. */
148         if (ipv6_cnt > 0)
149                 rte_fib6_lookup_bulk(qconf->ipv6_lookup_struct,
150                                 ipv6_arr, hopsv6, ipv6_cnt);
151
152         /* Add IPv4 and IPv6 hops to one array depending on type. */
153         for (i = 0; i < nb_rx; i++) {
154                 if (type_arr[i])
155                         nh = (uint16_t)hopsv4[ipv4_arr_assem++];
156                 else
157                         nh = (uint16_t)hopsv6[ipv6_arr_assem++];
158                 hops[i] = nh != FIB_DEFAULT_HOP ? nh : portid;
159         }
160
161 #if defined FIB_SEND_MULTI
162         send_packets_multi(qconf, pkts_burst, hops, nb_rx);
163 #else
164         fib_send_single(nb_rx, qconf, pkts_burst, hops);
165 #endif
166 }
167
168 /* Main fib processing loop. */
169 int
170 fib_main_loop(__rte_unused void *dummy)
171 {
172         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
173         unsigned int lcore_id;
174         uint64_t prev_tsc, diff_tsc, cur_tsc;
175         int i, nb_rx;
176         uint16_t portid;
177         uint8_t queueid;
178         struct lcore_conf *qconf;
179         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
180                         US_PER_S * BURST_TX_DRAIN_US;
181
182         lcore_id = rte_lcore_id();
183         qconf = &lcore_conf[lcore_id];
184
185         if (qconf->n_rx_queue == 0) {
186                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
187                 return 0;
188         }
189
190         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
191
192         for (i = 0; i < qconf->n_rx_queue; i++) {
193
194                 portid = qconf->rx_queue_list[i].port_id;
195                 queueid = qconf->rx_queue_list[i].queue_id;
196                 RTE_LOG(INFO, L3FWD,
197                                 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
198                                 lcore_id, portid, queueid);
199         }
200
201         cur_tsc = rte_rdtsc();
202         prev_tsc = cur_tsc;
203
204         while (!force_quit) {
205
206                 /* TX burst queue drain. */
207                 diff_tsc = cur_tsc - prev_tsc;
208                 if (unlikely(diff_tsc > drain_tsc)) {
209
210                         for (i = 0; i < qconf->n_tx_port; ++i) {
211                                 portid = qconf->tx_port_id[i];
212                                 if (qconf->tx_mbufs[portid].len == 0)
213                                         continue;
214                                 send_burst(qconf,
215                                         qconf->tx_mbufs[portid].len,
216                                         portid);
217                                 qconf->tx_mbufs[portid].len = 0;
218                         }
219
220                         prev_tsc = cur_tsc;
221                 }
222
223                 /* Read packet from RX queues. */
224                 for (i = 0; i < qconf->n_rx_queue; ++i) {
225                         portid = qconf->rx_queue_list[i].port_id;
226                         queueid = qconf->rx_queue_list[i].queue_id;
227                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
228                                         MAX_PKT_BURST);
229                         if (nb_rx == 0)
230                                 continue;
231
232                         /* Use fib to lookup port IDs and transmit them. */
233                         fib_send_packets(nb_rx, pkts_burst,     portid, qconf);
234                 }
235
236                 cur_tsc = rte_rdtsc();
237         }
238
239         return 0;
240 }
241
242 /* One eventdev loop for single and burst using fib. */
243 static __rte_always_inline void
244 fib_event_loop(struct l3fwd_event_resources *evt_rsrc,
245                 const uint8_t flags)
246 {
247         const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
248         const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
249                         evt_rsrc->evq.nb_queues - 1];
250         const uint8_t event_d_id = evt_rsrc->event_d_id;
251         const uint16_t deq_len = evt_rsrc->deq_depth;
252         struct rte_event events[MAX_PKT_BURST];
253         struct lcore_conf *lconf;
254         unsigned int lcore_id;
255         int nb_enq, nb_deq, i;
256
257         uint32_t ipv4_arr[MAX_PKT_BURST];
258         uint8_t ipv6_arr[MAX_PKT_BURST][RTE_FIB6_IPV6_ADDR_SIZE];
259         uint64_t hopsv4[MAX_PKT_BURST], hopsv6[MAX_PKT_BURST];
260         uint16_t nh;
261         uint8_t type_arr[MAX_PKT_BURST];
262         uint32_t ipv4_cnt, ipv6_cnt;
263         uint32_t ipv4_arr_assem, ipv6_arr_assem;
264
265         if (event_p_id < 0)
266                 return;
267
268         lcore_id = rte_lcore_id();
269
270         lconf = &lcore_conf[lcore_id];
271
272         RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
273
274         while (!force_quit) {
275                 /* Read events from RX queues. */
276                 nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id,
277                                 events, deq_len, 0);
278                 if (nb_deq == 0) {
279                         rte_pause();
280                         continue;
281                 }
282
283                 /* Reset counters. */
284                 ipv4_cnt = 0;
285                 ipv6_cnt = 0;
286                 ipv4_arr_assem = 0;
287                 ipv6_arr_assem = 0;
288
289                 /* Prefetch first packets. */
290                 for (i = 0; i < FIB_PREFETCH_OFFSET && i < nb_deq; i++)
291                         rte_prefetch0(rte_pktmbuf_mtod(events[i].mbuf, void *));
292
293                 /* Parse packet info and prefetch. */
294                 for (i = 0; i < (nb_deq - FIB_PREFETCH_OFFSET); i++) {
295                         if (flags & L3FWD_EVENT_TX_ENQ) {
296                                 events[i].queue_id = tx_q_id;
297                                 events[i].op = RTE_EVENT_OP_FORWARD;
298                         }
299
300                         if (flags & L3FWD_EVENT_TX_DIRECT)
301                                 rte_event_eth_tx_adapter_txq_set(events[i].mbuf,
302                                                 0);
303
304                         /* Prefetch packet. */
305                         rte_prefetch0(rte_pktmbuf_mtod(events[
306                                         i + FIB_PREFETCH_OFFSET].mbuf,
307                                         void *));
308
309                         fib_parse_packet(events[i].mbuf,
310                                         &ipv4_arr[ipv4_cnt], &ipv4_cnt,
311                                         ipv6_arr[ipv6_cnt], &ipv6_cnt,
312                                         &type_arr[i]);
313                 }
314
315                 /* Parse remaining packet info. */
316                 for (; i < nb_deq; i++) {
317                         if (flags & L3FWD_EVENT_TX_ENQ) {
318                                 events[i].queue_id = tx_q_id;
319                                 events[i].op = RTE_EVENT_OP_FORWARD;
320                         }
321
322                         if (flags & L3FWD_EVENT_TX_DIRECT)
323                                 rte_event_eth_tx_adapter_txq_set(events[i].mbuf,
324                                                 0);
325
326                         fib_parse_packet(events[i].mbuf,
327                                         &ipv4_arr[ipv4_cnt], &ipv4_cnt,
328                                         ipv6_arr[ipv6_cnt], &ipv6_cnt,
329                                         &type_arr[i]);
330                 }
331
332                 /* Lookup IPv4 hops if IPv4 packets are present. */
333                 if (likely(ipv4_cnt > 0))
334                         rte_fib_lookup_bulk(lconf->ipv4_lookup_struct,
335                                         ipv4_arr, hopsv4, ipv4_cnt);
336
337                 /* Lookup IPv6 hops if IPv6 packets are present. */
338                 if (ipv6_cnt > 0)
339                         rte_fib6_lookup_bulk(lconf->ipv6_lookup_struct,
340                                         ipv6_arr, hopsv6, ipv6_cnt);
341
342                 /* Assign ports looked up in fib depending on IPv4 or IPv6 */
343                 for (i = 0; i < nb_deq; i++) {
344                         if (type_arr[i])
345                                 nh = (uint16_t)hopsv4[ipv4_arr_assem++];
346                         else
347                                 nh = (uint16_t)hopsv6[ipv6_arr_assem++];
348                         if (nh != FIB_DEFAULT_HOP)
349                                 events[i].mbuf->port = nh;
350                 }
351
352                 if (flags & L3FWD_EVENT_TX_ENQ) {
353                         nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
354                                         events, nb_deq);
355                         while (nb_enq < nb_deq && !force_quit)
356                                 nb_enq += rte_event_enqueue_burst(event_d_id,
357                                                 event_p_id, events + nb_enq,
358                                                 nb_deq - nb_enq);
359                 }
360
361                 if (flags & L3FWD_EVENT_TX_DIRECT) {
362                         nb_enq = rte_event_eth_tx_adapter_enqueue(event_d_id,
363                                         event_p_id, events, nb_deq, 0);
364                         while (nb_enq < nb_deq && !force_quit)
365                                 nb_enq += rte_event_eth_tx_adapter_enqueue(
366                                                 event_d_id, event_p_id,
367                                                 events + nb_enq,
368                                                 nb_deq - nb_enq, 0);
369                 }
370         }
371 }
372
373 int __rte_noinline
374 fib_event_main_loop_tx_d(__rte_unused void *dummy)
375 {
376         struct l3fwd_event_resources *evt_rsrc =
377                         l3fwd_get_eventdev_rsrc();
378
379         fib_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
380         return 0;
381 }
382
383 int __rte_noinline
384 fib_event_main_loop_tx_d_burst(__rte_unused void *dummy)
385 {
386         struct l3fwd_event_resources *evt_rsrc =
387                         l3fwd_get_eventdev_rsrc();
388
389         fib_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
390         return 0;
391 }
392
393 int __rte_noinline
394 fib_event_main_loop_tx_q(__rte_unused void *dummy)
395 {
396         struct l3fwd_event_resources *evt_rsrc =
397                         l3fwd_get_eventdev_rsrc();
398
399         fib_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ);
400         return 0;
401 }
402
403 int __rte_noinline
404 fib_event_main_loop_tx_q_burst(__rte_unused void *dummy)
405 {
406         struct l3fwd_event_resources *evt_rsrc =
407                         l3fwd_get_eventdev_rsrc();
408
409         fib_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ);
410         return 0;
411 }
412
413 /* Function to setup fib. */
414 void
415 setup_fib(const int socketid)
416 {
417         struct rte_fib6_conf config;
418         struct rte_fib_conf config_ipv4;
419         unsigned int i;
420         int ret;
421         char s[64];
422         char abuf[INET6_ADDRSTRLEN];
423
424         /* Create the fib IPv4 table. */
425         config_ipv4.type = RTE_FIB_DIR24_8;
426         config_ipv4.max_routes = (1 << 16);
427         config_ipv4.default_nh = FIB_DEFAULT_HOP;
428         config_ipv4.dir24_8.nh_sz = RTE_FIB_DIR24_8_4B;
429         config_ipv4.dir24_8.num_tbl8 = (1 << 15);
430         snprintf(s, sizeof(s), "IPV4_L3FWD_FIB_%d", socketid);
431         ipv4_l3fwd_fib_lookup_struct[socketid] =
432                         rte_fib_create(s, socketid, &config_ipv4);
433         if (ipv4_l3fwd_fib_lookup_struct[socketid] == NULL)
434                 rte_exit(EXIT_FAILURE,
435                         "Unable to create the l3fwd FIB table on socket %d\n",
436                         socketid);
437
438         /* Populate the fib ipv4 table. */
439         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
440                 struct in_addr in;
441
442                 /* Skip unused ports. */
443                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
444                                 enabled_port_mask) == 0)
445                         continue;
446
447                 ret = rte_fib_add(ipv4_l3fwd_fib_lookup_struct[socketid],
448                         ipv4_l3fwd_route_array[i].ip,
449                         ipv4_l3fwd_route_array[i].depth,
450                         ipv4_l3fwd_route_array[i].if_out);
451
452                 if (ret < 0) {
453                         rte_exit(EXIT_FAILURE,
454                                         "Unable to add entry %u to the l3fwd FIB table on socket %d\n",
455                                         i, socketid);
456                 }
457
458                 in.s_addr = htonl(ipv4_l3fwd_route_array[i].ip);
459                 if (inet_ntop(AF_INET, &in, abuf, sizeof(abuf)) != NULL) {
460                         printf("FIB: Adding route %s / %d (%d)\n",
461                                 abuf,
462                                 ipv4_l3fwd_route_array[i].depth,
463                                 ipv4_l3fwd_route_array[i].if_out);
464                 } else {
465                         printf("FIB: IPv4 route added to port %d\n",
466                                 ipv4_l3fwd_route_array[i].if_out);
467                 }
468         }
469
470         /* Create the fib IPv6 table. */
471         snprintf(s, sizeof(s), "IPV6_L3FWD_FIB_%d", socketid);
472
473         config.type = RTE_FIB6_TRIE;
474         config.max_routes = (1 << 16) - 1;
475         config.default_nh = FIB_DEFAULT_HOP;
476         config.trie.nh_sz = RTE_FIB6_TRIE_4B;
477         config.trie.num_tbl8 = (1 << 15);
478         ipv6_l3fwd_fib_lookup_struct[socketid] = rte_fib6_create(s, socketid,
479                         &config);
480         if (ipv6_l3fwd_fib_lookup_struct[socketid] == NULL)
481                 rte_exit(EXIT_FAILURE,
482                                 "Unable to create the l3fwd FIB table on socket %d\n",
483                                 socketid);
484
485         /* Populate the fib IPv6 table. */
486         for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
487
488                 /* Skip unused ports. */
489                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
490                                 enabled_port_mask) == 0)
491                         continue;
492
493                 ret = rte_fib6_add(ipv6_l3fwd_fib_lookup_struct[socketid],
494                         ipv6_l3fwd_route_array[i].ip,
495                         ipv6_l3fwd_route_array[i].depth,
496                         ipv6_l3fwd_route_array[i].if_out);
497
498                 if (ret < 0) {
499                         rte_exit(EXIT_FAILURE,
500                                         "Unable to add entry %u to the l3fwd FIB table on socket %d\n",
501                                         i, socketid);
502                 }
503
504                 if (inet_ntop(AF_INET6, ipv6_l3fwd_route_array[i].ip,
505                                 abuf, sizeof(abuf)) != NULL) {
506                         printf("FIB: Adding route %s / %d (%d)\n",
507                                 abuf,
508                                 ipv6_l3fwd_route_array[i].depth,
509                                 ipv6_l3fwd_route_array[i].if_out);
510                 } else {
511                         printf("FIB: IPv6 route added to port %d\n",
512                                 ipv6_l3fwd_route_array[i].if_out);
513                 }
514         }
515 }
516
517 /* Return ipv4 fib lookup struct. */
518 void *
519 fib_get_ipv4_l3fwd_lookup_struct(const int socketid)
520 {
521         return ipv4_l3fwd_fib_lookup_struct[socketid];
522 }
523
524 /* Return ipv6 fib lookup struct. */
525 void *
526 fib_get_ipv6_l3fwd_lookup_struct(const int socketid)
527 {
528         return ipv6_l3fwd_fib_lookup_struct[socketid];
529 }