net/mlx5: add flow prepare function
[dpdk.git] / examples / l2fwd-jobstats / main.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4
5 #include <locale.h>
6 #include <stdlib.h>
7 #include <string.h>
8 #include <stdint.h>
9 #include <ctype.h>
10 #include <getopt.h>
11
12 #include <rte_common.h>
13 #include <rte_log.h>
14 #include <rte_malloc.h>
15 #include <rte_memory.h>
16 #include <rte_memcpy.h>
17 #include <rte_eal.h>
18 #include <rte_launch.h>
19 #include <rte_atomic.h>
20 #include <rte_cycles.h>
21 #include <rte_prefetch.h>
22 #include <rte_lcore.h>
23 #include <rte_per_lcore.h>
24 #include <rte_branch_prediction.h>
25 #include <rte_interrupts.h>
26 #include <rte_debug.h>
27 #include <rte_ether.h>
28 #include <rte_ethdev.h>
29 #include <rte_mempool.h>
30 #include <rte_mbuf.h>
31 #include <rte_spinlock.h>
32
33 #include <rte_errno.h>
34 #include <rte_jobstats.h>
35 #include <rte_timer.h>
36 #include <rte_alarm.h>
37 #include <rte_pause.h>
38
39 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
40
41 #define NB_MBUF   8192
42
43 #define MAX_PKT_BURST 32
44 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
45
46 /*
47  * Configurable number of RX/TX ring descriptors
48  */
49 #define RTE_TEST_RX_DESC_DEFAULT 1024
50 #define RTE_TEST_TX_DESC_DEFAULT 1024
51 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
52 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
53
54 /* ethernet addresses of ports */
55 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
56
57 /* mask of enabled ports */
58 static uint32_t l2fwd_enabled_port_mask;
59
60 /* list of enabled ports */
61 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
62
63 #define UPDATE_STEP_UP 1
64 #define UPDATE_STEP_DOWN 32
65
66 static unsigned int l2fwd_rx_queue_per_lcore = 1;
67
68 #define MAX_RX_QUEUE_PER_LCORE 16
69 #define MAX_TX_QUEUE_PER_PORT 16
70 struct lcore_queue_conf {
71         unsigned n_rx_port;
72         unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
73         uint64_t next_flush_time[RTE_MAX_ETHPORTS];
74
75         struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
76         struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
77
78         struct rte_timer flush_timer;
79         struct rte_jobstats flush_job;
80         struct rte_jobstats idle_job;
81         struct rte_jobstats_context jobs_context;
82
83         rte_atomic16_t stats_read_pending;
84         rte_spinlock_t lock;
85 } __rte_cache_aligned;
86 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
87
88 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
89
90 static struct rte_eth_conf port_conf = {
91         .rxmode = {
92                 .split_hdr_size = 0,
93         },
94         .txmode = {
95                 .mq_mode = ETH_MQ_TX_NONE,
96         },
97 };
98
99 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
100
101 /* Per-port statistics struct */
102 struct l2fwd_port_statistics {
103         uint64_t tx;
104         uint64_t rx;
105         uint64_t dropped;
106 } __rte_cache_aligned;
107 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
108
109 /* 1 day max */
110 #define MAX_TIMER_PERIOD 86400
111 /* default period is 10 seconds */
112 static int64_t timer_period = 10;
113 /* default timer frequency */
114 static double hz;
115 /* BURST_TX_DRAIN_US converted to cycles */
116 uint64_t drain_tsc;
117 /* Convert cycles to ns */
118 static inline double
119 cycles_to_ns(uint64_t cycles)
120 {
121         double t = cycles;
122
123         t *= (double)NS_PER_S;
124         t /= hz;
125         return t;
126 }
127
128 static void
129 show_lcore_stats(unsigned lcore_id)
130 {
131         struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
132         struct rte_jobstats_context *ctx = &qconf->jobs_context;
133         struct rte_jobstats *job;
134         uint8_t i;
135
136         /* LCore statistics. */
137         uint64_t stats_period, loop_count;
138         uint64_t exec, exec_min, exec_max;
139         uint64_t management, management_min, management_max;
140         uint64_t busy, busy_min, busy_max;
141
142         /* Jobs statistics. */
143         const uint16_t port_cnt = qconf->n_rx_port;
144         uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
145         uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
146                                 jobs_exec_max[port_cnt];
147
148         uint64_t flush_exec_cnt, flush_period;
149         uint64_t flush_exec, flush_exec_min, flush_exec_max;
150
151         uint64_t idle_exec_cnt;
152         uint64_t idle_exec, idle_exec_min, idle_exec_max;
153         uint64_t collection_time = rte_get_timer_cycles();
154
155         /* Ask forwarding thread to give us stats. */
156         rte_atomic16_set(&qconf->stats_read_pending, 1);
157         rte_spinlock_lock(&qconf->lock);
158         rte_atomic16_set(&qconf->stats_read_pending, 0);
159
160         /* Collect context statistics. */
161         stats_period = ctx->state_time - ctx->start_time;
162         loop_count = ctx->loop_cnt;
163
164         exec = ctx->exec_time;
165         exec_min = ctx->min_exec_time;
166         exec_max = ctx->max_exec_time;
167
168         management = ctx->management_time;
169         management_min = ctx->min_management_time;
170         management_max = ctx->max_management_time;
171
172         rte_jobstats_context_reset(ctx);
173
174         for (i = 0; i < port_cnt; i++) {
175                 job = &qconf->port_fwd_jobs[i];
176
177                 jobs_exec_cnt[i] = job->exec_cnt;
178                 jobs_period[i] = job->period;
179
180                 jobs_exec[i] = job->exec_time;
181                 jobs_exec_min[i] = job->min_exec_time;
182                 jobs_exec_max[i] = job->max_exec_time;
183
184                 rte_jobstats_reset(job);
185         }
186
187         flush_exec_cnt = qconf->flush_job.exec_cnt;
188         flush_period = qconf->flush_job.period;
189         flush_exec = qconf->flush_job.exec_time;
190         flush_exec_min = qconf->flush_job.min_exec_time;
191         flush_exec_max = qconf->flush_job.max_exec_time;
192         rte_jobstats_reset(&qconf->flush_job);
193
194         idle_exec_cnt = qconf->idle_job.exec_cnt;
195         idle_exec = qconf->idle_job.exec_time;
196         idle_exec_min = qconf->idle_job.min_exec_time;
197         idle_exec_max = qconf->idle_job.max_exec_time;
198         rte_jobstats_reset(&qconf->idle_job);
199
200         rte_spinlock_unlock(&qconf->lock);
201
202         exec -= idle_exec;
203         busy = exec + management;
204         busy_min = exec_min + management_min;
205         busy_max = exec_max + management_max;
206
207
208         collection_time = rte_get_timer_cycles() - collection_time;
209
210 #define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
211
212         printf("\n----------------"
213                         "\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
214                         "\n%-18s %14s %7s %10s %10s %10s "
215                         "\n%-18s %'14.0f"
216                         "\n%-18s %'14" PRIu64
217                         STAT_FMT /* Exec */
218                         STAT_FMT /* Management */
219                         STAT_FMT /* Busy */
220                         STAT_FMT, /* Idle  */
221                         lcore_id, cycles_to_ns(collection_time),
222                         "Stat type", "total", "%total", "avg", "min", "max",
223                         "Stats duration:", cycles_to_ns(stats_period),
224                         "Loop count:", loop_count,
225                         "Exec time",
226                         cycles_to_ns(exec), exec * 100.0 / stats_period,
227                         cycles_to_ns(loop_count  ? exec / loop_count : 0),
228                         cycles_to_ns(exec_min),
229                         cycles_to_ns(exec_max),
230                         "Management time",
231                         cycles_to_ns(management), management * 100.0 / stats_period,
232                         cycles_to_ns(loop_count  ? management / loop_count : 0),
233                         cycles_to_ns(management_min),
234                         cycles_to_ns(management_max),
235                         "Exec + management",
236                         cycles_to_ns(busy),  busy * 100.0 / stats_period,
237                         cycles_to_ns(loop_count ? busy / loop_count : 0),
238                         cycles_to_ns(busy_min),
239                         cycles_to_ns(busy_max),
240                         "Idle (job)",
241                         cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
242                         cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
243                         cycles_to_ns(idle_exec_min),
244                         cycles_to_ns(idle_exec_max));
245
246         for (i = 0; i < qconf->n_rx_port; i++) {
247                 job = &qconf->port_fwd_jobs[i];
248                 printf("\n\nJob %" PRIu32 ": %-20s "
249                                 "\n%-18s %'14" PRIu64
250                                 "\n%-18s %'14.0f"
251                                 STAT_FMT,
252                                 i, job->name,
253                                 "Exec count:", jobs_exec_cnt[i],
254                                 "Exec period: ", cycles_to_ns(jobs_period[i]),
255                                 "Exec time",
256                                 cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
257                                 cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
258                                                 : 0),
259                                 cycles_to_ns(jobs_exec_min[i]),
260                                 cycles_to_ns(jobs_exec_max[i]));
261         }
262
263         if (qconf->n_rx_port > 0) {
264                 job = &qconf->flush_job;
265                 printf("\n\nJob %" PRIu32 ": %-20s "
266                                 "\n%-18s %'14" PRIu64
267                                 "\n%-18s %'14.0f"
268                                 STAT_FMT,
269                                 i, job->name,
270                                 "Exec count:", flush_exec_cnt,
271                                 "Exec period: ", cycles_to_ns(flush_period),
272                                 "Exec time",
273                                 cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
274                                 cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
275                                 cycles_to_ns(flush_exec_min),
276                                 cycles_to_ns(flush_exec_max));
277         }
278 }
279
280 /* Print out statistics on packets dropped */
281 static void
282 show_stats_cb(__rte_unused void *param)
283 {
284         uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
285         unsigned portid, lcore_id;
286
287         total_packets_dropped = 0;
288         total_packets_tx = 0;
289         total_packets_rx = 0;
290
291         const char clr[] = { 27, '[', '2', 'J', '\0' };
292         const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
293
294         /* Clear screen and move to top left */
295         printf("%s%s"
296                         "\nPort statistics ===================================",
297                         clr, topLeft);
298
299         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
300                 /* skip disabled ports */
301                 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
302                         continue;
303                 printf("\nStatistics for port %u ------------------------------"
304                                 "\nPackets sent: %24"PRIu64
305                                 "\nPackets received: %20"PRIu64
306                                 "\nPackets dropped: %21"PRIu64,
307                                 portid,
308                                 port_statistics[portid].tx,
309                                 port_statistics[portid].rx,
310                                 port_statistics[portid].dropped);
311
312                 total_packets_dropped += port_statistics[portid].dropped;
313                 total_packets_tx += port_statistics[portid].tx;
314                 total_packets_rx += port_statistics[portid].rx;
315         }
316
317         printf("\nAggregate statistics ==============================="
318                         "\nTotal packets sent: %18"PRIu64
319                         "\nTotal packets received: %14"PRIu64
320                         "\nTotal packets dropped: %15"PRIu64
321                         "\n====================================================",
322                         total_packets_tx,
323                         total_packets_rx,
324                         total_packets_dropped);
325
326         RTE_LCORE_FOREACH(lcore_id) {
327                 if (lcore_queue_conf[lcore_id].n_rx_port > 0)
328                         show_lcore_stats(lcore_id);
329         }
330
331         printf("\n====================================================\n");
332         rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
333 }
334
335 static void
336 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
337 {
338         struct ether_hdr *eth;
339         void *tmp;
340         int sent;
341         unsigned dst_port;
342         struct rte_eth_dev_tx_buffer *buffer;
343
344         dst_port = l2fwd_dst_ports[portid];
345         eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
346
347         /* 02:00:00:00:00:xx */
348         tmp = &eth->d_addr.addr_bytes[0];
349         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
350
351         /* src addr */
352         ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->s_addr);
353
354         buffer = tx_buffer[dst_port];
355         sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
356         if (sent)
357                 port_statistics[dst_port].tx += sent;
358 }
359
360 static void
361 l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
362 {
363         int64_t err = job->target - result;
364         int64_t histeresis = job->target / 8;
365
366         if (err < -histeresis) {
367                 if (job->min_period + UPDATE_STEP_DOWN < job->period)
368                         job->period -= UPDATE_STEP_DOWN;
369         } else if (err > histeresis) {
370                 if (job->period + UPDATE_STEP_UP < job->max_period)
371                         job->period += UPDATE_STEP_UP;
372         }
373 }
374
375 static void
376 l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
377 {
378         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
379         struct rte_mbuf *m;
380
381         const uint16_t port_idx = (uintptr_t) arg;
382         const unsigned lcore_id = rte_lcore_id();
383         struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
384         struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
385         const uint16_t portid = qconf->rx_port_list[port_idx];
386
387         uint8_t j;
388         uint16_t total_nb_rx;
389
390         rte_jobstats_start(&qconf->jobs_context, job);
391
392         /* Call rx burst 2 times. This allow rte_jobstats logic to see if this
393          * function must be called more frequently. */
394
395         total_nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
396                         MAX_PKT_BURST);
397
398         for (j = 0; j < total_nb_rx; j++) {
399                 m = pkts_burst[j];
400                 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
401                 l2fwd_simple_forward(m, portid);
402         }
403
404         if (total_nb_rx == MAX_PKT_BURST) {
405                 const uint16_t nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
406                                 MAX_PKT_BURST);
407
408                 total_nb_rx += nb_rx;
409                 for (j = 0; j < nb_rx; j++) {
410                         m = pkts_burst[j];
411                         rte_prefetch0(rte_pktmbuf_mtod(m, void *));
412                         l2fwd_simple_forward(m, portid);
413                 }
414         }
415
416         port_statistics[portid].rx += total_nb_rx;
417
418         /* Adjust period time in which we are running here. */
419         if (rte_jobstats_finish(job, total_nb_rx) != 0) {
420                 rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
421                                 lcore_id, l2fwd_fwd_job, arg);
422         }
423 }
424
425 static void
426 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
427 {
428         uint64_t now;
429         unsigned lcore_id;
430         struct lcore_queue_conf *qconf;
431         uint16_t portid;
432         unsigned i;
433         uint32_t sent;
434         struct rte_eth_dev_tx_buffer *buffer;
435
436         lcore_id = rte_lcore_id();
437         qconf = &lcore_queue_conf[lcore_id];
438
439         rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
440
441         now = rte_get_timer_cycles();
442         lcore_id = rte_lcore_id();
443         qconf = &lcore_queue_conf[lcore_id];
444
445         for (i = 0; i < qconf->n_rx_port; i++) {
446                 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
447
448                 if (qconf->next_flush_time[portid] <= now)
449                         continue;
450
451                 buffer = tx_buffer[portid];
452                 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
453                 if (sent)
454                         port_statistics[portid].tx += sent;
455
456                 qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc;
457         }
458
459         /* Pass target to indicate that this job is happy of time interwal
460          * in which it was called. */
461         rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
462 }
463
464 /* main processing loop */
465 static void
466 l2fwd_main_loop(void)
467 {
468         unsigned lcore_id;
469         unsigned i, portid;
470         struct lcore_queue_conf *qconf;
471         uint8_t stats_read_pending = 0;
472         uint8_t need_manage;
473
474         lcore_id = rte_lcore_id();
475         qconf = &lcore_queue_conf[lcore_id];
476
477         if (qconf->n_rx_port == 0) {
478                 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
479                 return;
480         }
481
482         RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
483
484         for (i = 0; i < qconf->n_rx_port; i++) {
485
486                 portid = qconf->rx_port_list[i];
487                 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
488                         portid);
489         }
490
491         rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
492
493         for (;;) {
494                 rte_spinlock_lock(&qconf->lock);
495
496                 do {
497                         rte_jobstats_context_start(&qconf->jobs_context);
498
499                         /* Do the Idle job:
500                          * - Read stats_read_pending flag
501                          * - check if some real job need to be executed
502                          */
503                         rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
504
505                         uint64_t repeats = 0;
506
507                         do {
508                                 uint8_t i;
509                                 uint64_t now = rte_get_timer_cycles();
510
511                                 repeats++;
512                                 need_manage = qconf->flush_timer.expire < now;
513                                 /* Check if we was esked to give a stats. */
514                                 stats_read_pending =
515                                                 rte_atomic16_read(&qconf->stats_read_pending);
516                                 need_manage |= stats_read_pending;
517
518                                 for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
519                                         need_manage = qconf->rx_timers[i].expire < now;
520
521                         } while (!need_manage);
522
523                         if (likely(repeats != 1))
524                                 rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
525                         else
526                                 rte_jobstats_abort(&qconf->idle_job);
527
528                         rte_timer_manage();
529                         rte_jobstats_context_finish(&qconf->jobs_context);
530                 } while (likely(stats_read_pending == 0));
531
532                 rte_spinlock_unlock(&qconf->lock);
533                 rte_pause();
534         }
535 }
536
537 static int
538 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
539 {
540         l2fwd_main_loop();
541         return 0;
542 }
543
544 /* display usage */
545 static void
546 l2fwd_usage(const char *prgname)
547 {
548         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
549                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
550                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
551                    "  -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
552                    "  -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
553                prgname);
554 }
555
556 static int
557 l2fwd_parse_portmask(const char *portmask)
558 {
559         char *end = NULL;
560         unsigned long pm;
561
562         /* parse hexadecimal string */
563         pm = strtoul(portmask, &end, 16);
564         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
565                 return -1;
566
567         if (pm == 0)
568                 return -1;
569
570         return pm;
571 }
572
573 static unsigned int
574 l2fwd_parse_nqueue(const char *q_arg)
575 {
576         char *end = NULL;
577         unsigned long n;
578
579         /* parse hexadecimal string */
580         n = strtoul(q_arg, &end, 10);
581         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
582                 return 0;
583         if (n == 0)
584                 return 0;
585         if (n >= MAX_RX_QUEUE_PER_LCORE)
586                 return 0;
587
588         return n;
589 }
590
591 static int
592 l2fwd_parse_timer_period(const char *q_arg)
593 {
594         char *end = NULL;
595         int n;
596
597         /* parse number string */
598         n = strtol(q_arg, &end, 10);
599         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
600                 return -1;
601         if (n >= MAX_TIMER_PERIOD)
602                 return -1;
603
604         return n;
605 }
606
607 /* Parse the argument given in the command line of the application */
608 static int
609 l2fwd_parse_args(int argc, char **argv)
610 {
611         int opt, ret;
612         char **argvopt;
613         int option_index;
614         char *prgname = argv[0];
615         static struct option lgopts[] = {
616                 {NULL, 0, 0, 0}
617         };
618
619         argvopt = argv;
620
621         while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
622                                   lgopts, &option_index)) != EOF) {
623
624                 switch (opt) {
625                 /* portmask */
626                 case 'p':
627                         l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
628                         if (l2fwd_enabled_port_mask == 0) {
629                                 printf("invalid portmask\n");
630                                 l2fwd_usage(prgname);
631                                 return -1;
632                         }
633                         break;
634
635                 /* nqueue */
636                 case 'q':
637                         l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
638                         if (l2fwd_rx_queue_per_lcore == 0) {
639                                 printf("invalid queue number\n");
640                                 l2fwd_usage(prgname);
641                                 return -1;
642                         }
643                         break;
644
645                 /* timer period */
646                 case 'T':
647                         timer_period = l2fwd_parse_timer_period(optarg);
648                         if (timer_period < 0) {
649                                 printf("invalid timer period\n");
650                                 l2fwd_usage(prgname);
651                                 return -1;
652                         }
653                         break;
654
655                 /* For thousands separator in printf. */
656                 case 'l':
657                         setlocale(LC_ALL, "");
658                         break;
659
660                 /* long options */
661                 case 0:
662                         l2fwd_usage(prgname);
663                         return -1;
664
665                 default:
666                         l2fwd_usage(prgname);
667                         return -1;
668                 }
669         }
670
671         if (optind >= 0)
672                 argv[optind-1] = prgname;
673
674         ret = optind-1;
675         optind = 1; /* reset getopt lib */
676         return ret;
677 }
678
679 /* Check the link status of all ports in up to 9s, and print them finally */
680 static void
681 check_all_ports_link_status(uint32_t port_mask)
682 {
683 #define CHECK_INTERVAL 100 /* 100ms */
684 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
685         uint16_t portid;
686         uint8_t count, all_ports_up, print_flag = 0;
687         struct rte_eth_link link;
688
689         printf("\nChecking link status");
690         fflush(stdout);
691         for (count = 0; count <= MAX_CHECK_TIME; count++) {
692                 all_ports_up = 1;
693                 RTE_ETH_FOREACH_DEV(portid) {
694                         if ((port_mask & (1 << portid)) == 0)
695                                 continue;
696                         memset(&link, 0, sizeof(link));
697                         rte_eth_link_get_nowait(portid, &link);
698                         /* print link status if flag set */
699                         if (print_flag == 1) {
700                                 if (link.link_status)
701                                         printf(
702                                         "Port%d Link Up. Speed %u Mbps - %s\n",
703                                                 portid, link.link_speed,
704                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
705                                         ("full-duplex") : ("half-duplex\n"));
706                                 else
707                                         printf("Port %d Link Down\n", portid);
708                                 continue;
709                         }
710                         /* clear all_ports_up flag if any link down */
711                         if (link.link_status == ETH_LINK_DOWN) {
712                                 all_ports_up = 0;
713                                 break;
714                         }
715                 }
716                 /* after finally printing all link status, get out */
717                 if (print_flag == 1)
718                         break;
719
720                 if (all_ports_up == 0) {
721                         printf(".");
722                         fflush(stdout);
723                         rte_delay_ms(CHECK_INTERVAL);
724                 }
725
726                 /* set the print_flag if all ports up or timeout */
727                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
728                         print_flag = 1;
729                         printf("done\n");
730                 }
731         }
732 }
733
734 int
735 main(int argc, char **argv)
736 {
737         struct lcore_queue_conf *qconf;
738         unsigned lcore_id, rx_lcore_id;
739         unsigned nb_ports_in_mask = 0;
740         int ret;
741         char name[RTE_JOBSTATS_NAMESIZE];
742         uint16_t nb_ports;
743         uint16_t nb_ports_available = 0;
744         uint16_t portid, last_port;
745         uint8_t i;
746
747         /* init EAL */
748         ret = rte_eal_init(argc, argv);
749         if (ret < 0)
750                 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
751         argc -= ret;
752         argv += ret;
753
754         /* parse application arguments (after the EAL ones) */
755         ret = l2fwd_parse_args(argc, argv);
756         if (ret < 0)
757                 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
758
759         rte_timer_subsystem_init();
760
761         /* fetch default timer frequency. */
762         hz = rte_get_timer_hz();
763
764         /* create the mbuf pool */
765         l2fwd_pktmbuf_pool =
766                 rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
767                         0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
768         if (l2fwd_pktmbuf_pool == NULL)
769                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
770
771         nb_ports = rte_eth_dev_count_avail();
772         if (nb_ports == 0)
773                 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
774
775         /* reset l2fwd_dst_ports */
776         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
777                 l2fwd_dst_ports[portid] = 0;
778         last_port = 0;
779
780         /*
781          * Each logical core is assigned a dedicated TX queue on each port.
782          */
783         RTE_ETH_FOREACH_DEV(portid) {
784                 /* skip ports that are not enabled */
785                 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
786                         continue;
787
788                 if (nb_ports_in_mask % 2) {
789                         l2fwd_dst_ports[portid] = last_port;
790                         l2fwd_dst_ports[last_port] = portid;
791                 } else
792                         last_port = portid;
793
794                 nb_ports_in_mask++;
795         }
796         if (nb_ports_in_mask % 2) {
797                 printf("Notice: odd number of ports in portmask.\n");
798                 l2fwd_dst_ports[last_port] = last_port;
799         }
800
801         rx_lcore_id = 0;
802         qconf = NULL;
803
804         /* Initialize the port/queue configuration of each logical core */
805         RTE_ETH_FOREACH_DEV(portid) {
806                 /* skip ports that are not enabled */
807                 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
808                         continue;
809
810                 /* get the lcore_id for this port */
811                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
812                        lcore_queue_conf[rx_lcore_id].n_rx_port ==
813                        l2fwd_rx_queue_per_lcore) {
814                         rx_lcore_id++;
815                         if (rx_lcore_id >= RTE_MAX_LCORE)
816                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
817                 }
818
819                 if (qconf != &lcore_queue_conf[rx_lcore_id])
820                         /* Assigned a new logical core in the loop above. */
821                         qconf = &lcore_queue_conf[rx_lcore_id];
822
823                 qconf->rx_port_list[qconf->n_rx_port] = portid;
824                 qconf->n_rx_port++;
825                 printf("Lcore %u: RX port %u\n", rx_lcore_id, portid);
826         }
827
828         /* Initialise each port */
829         RTE_ETH_FOREACH_DEV(portid) {
830                 struct rte_eth_dev_info dev_info;
831                 struct rte_eth_rxconf rxq_conf;
832                 struct rte_eth_txconf txq_conf;
833                 struct rte_eth_conf local_port_conf = port_conf;
834
835                 /* skip ports that are not enabled */
836                 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
837                         printf("Skipping disabled port %u\n", portid);
838                         continue;
839                 }
840                 nb_ports_available++;
841
842                 /* init port */
843                 printf("Initializing port %u... ", portid);
844                 fflush(stdout);
845                 rte_eth_dev_info_get(portid, &dev_info);
846                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
847                         local_port_conf.txmode.offloads |=
848                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
849                 ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
850                 if (ret < 0)
851                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
852                                   ret, portid);
853
854                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
855                                                        &nb_txd);
856                 if (ret < 0)
857                         rte_exit(EXIT_FAILURE,
858                                  "Cannot adjust number of descriptors: err=%d, port=%u\n",
859                                  ret, portid);
860
861                 rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
862
863                 /* init one RX queue */
864                 fflush(stdout);
865                 rxq_conf = dev_info.default_rxconf;
866                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
867                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
868                                              rte_eth_dev_socket_id(portid),
869                                              &rxq_conf,
870                                              l2fwd_pktmbuf_pool);
871                 if (ret < 0)
872                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
873                                   ret, portid);
874
875                 /* init one TX queue on each port */
876                 txq_conf = dev_info.default_txconf;
877                 txq_conf.offloads = local_port_conf.txmode.offloads;
878                 fflush(stdout);
879                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
880                                 rte_eth_dev_socket_id(portid),
881                                 &txq_conf);
882                 if (ret < 0)
883                         rte_exit(EXIT_FAILURE,
884                         "rte_eth_tx_queue_setup:err=%d, port=%u\n",
885                                 ret, portid);
886
887                 /* Initialize TX buffers */
888                 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
889                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
890                                 rte_eth_dev_socket_id(portid));
891                 if (tx_buffer[portid] == NULL)
892                         rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
893                                         portid);
894
895                 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
896
897                 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
898                                 rte_eth_tx_buffer_count_callback,
899                                 &port_statistics[portid].dropped);
900                 if (ret < 0)
901                         rte_exit(EXIT_FAILURE,
902                         "Cannot set error callback for tx buffer on port %u\n",
903                                  portid);
904
905                 /* Start device */
906                 ret = rte_eth_dev_start(portid);
907                 if (ret < 0)
908                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
909                                   ret, portid);
910
911                 printf("done:\n");
912
913                 rte_eth_promiscuous_enable(portid);
914
915                 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
916                                 portid,
917                                 l2fwd_ports_eth_addr[portid].addr_bytes[0],
918                                 l2fwd_ports_eth_addr[portid].addr_bytes[1],
919                                 l2fwd_ports_eth_addr[portid].addr_bytes[2],
920                                 l2fwd_ports_eth_addr[portid].addr_bytes[3],
921                                 l2fwd_ports_eth_addr[portid].addr_bytes[4],
922                                 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
923
924                 /* initialize port stats */
925                 memset(&port_statistics, 0, sizeof(port_statistics));
926         }
927
928         if (!nb_ports_available) {
929                 rte_exit(EXIT_FAILURE,
930                         "All available ports are disabled. Please set portmask.\n");
931         }
932
933         check_all_ports_link_status(l2fwd_enabled_port_mask);
934
935         drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
936
937         RTE_LCORE_FOREACH(lcore_id) {
938                 qconf = &lcore_queue_conf[lcore_id];
939
940                 rte_spinlock_init(&qconf->lock);
941
942                 if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
943                         rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
944
945                 if (qconf->n_rx_port == 0) {
946                         RTE_LOG(INFO, L2FWD,
947                                 "lcore %u: no ports so no jobs stats context initialization\n",
948                                 lcore_id);
949                         continue;
950                 }
951                 /* Add flush job.
952                  * Set fixed period by setting min = max = initial period. Set target to
953                  * zero as it is irrelevant for this job. */
954                 rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
955                                 drain_tsc, 0);
956
957                 rte_timer_init(&qconf->flush_timer);
958                 ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
959                                 lcore_id, &l2fwd_flush_job, NULL);
960
961                 if (ret < 0) {
962                         rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
963                                         lcore_id, rte_strerror(-ret));
964                 }
965
966                 for (i = 0; i < qconf->n_rx_port; i++) {
967                         struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
968
969                         portid = qconf->rx_port_list[i];
970                         printf("Setting forward job for port %u\n", portid);
971
972                         snprintf(name, RTE_DIM(name), "port %u fwd", portid);
973                         /* Setup forward job.
974                          * Set min, max and initial period. Set target to MAX_PKT_BURST as
975                          * this is desired optimal RX/TX burst size. */
976                         rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
977                         rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
978
979                         rte_timer_init(&qconf->rx_timers[i]);
980                         ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
981                                         &l2fwd_fwd_job, (void *)(uintptr_t)i);
982
983                         if (ret < 0) {
984                                 rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
985                                                 lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
986                         }
987                 }
988         }
989
990         if (timer_period)
991                 rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
992         else
993                 RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
994
995         /* launch per-lcore init on every lcore */
996         rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
997         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
998                 if (rte_eal_wait_lcore(lcore_id) < 0)
999                         return -1;
1000         }
1001
1002         return 0;
1003 }