main.c revision 8b25d1ad
1/*-
2 *   BSD LICENSE
3 *
4 *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5 *   All rights reserved.
6 *
7 *   Redistribution and use in source and binary forms, with or without
8 *   modification, are permitted provided that the following conditions
9 *   are met:
10 *
11 *     * Redistributions of source code must retain the above copyright
12 *       notice, this list of conditions and the following disclaimer.
13 *     * Redistributions in binary form must reproduce the above copyright
14 *       notice, this list of conditions and the following disclaimer in
15 *       the documentation and/or other materials provided with the
16 *       distribution.
17 *     * Neither the name of Intel Corporation nor the names of its
18 *       contributors may be used to endorse or promote products derived
19 *       from this software without specific prior written permission.
20 *
21 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34#include <locale.h>
35#include <stdlib.h>
36#include <string.h>
37#include <stdint.h>
38#include <ctype.h>
39#include <getopt.h>
40
41#include <rte_alarm.h>
42#include <rte_common.h>
43#include <rte_log.h>
44#include <rte_malloc.h>
45#include <rte_memory.h>
46#include <rte_memcpy.h>
47#include <rte_memzone.h>
48#include <rte_eal.h>
49#include <rte_per_lcore.h>
50#include <rte_launch.h>
51#include <rte_atomic.h>
52#include <rte_cycles.h>
53#include <rte_prefetch.h>
54#include <rte_lcore.h>
55#include <rte_per_lcore.h>
56#include <rte_branch_prediction.h>
57#include <rte_interrupts.h>
58#include <rte_pci.h>
59#include <rte_debug.h>
60#include <rte_ether.h>
61#include <rte_ethdev.h>
62#include <rte_ring.h>
63#include <rte_mempool.h>
64#include <rte_mbuf.h>
65#include <rte_spinlock.h>
66
67#include <rte_errno.h>
68#include <rte_jobstats.h>
69#include <rte_timer.h>
70#include <rte_alarm.h>
71
72#define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
73
74#define NB_MBUF   8192
75
76#define MAX_PKT_BURST 32
77#define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
78
79/*
80 * Configurable number of RX/TX ring descriptors
81 */
82#define RTE_TEST_RX_DESC_DEFAULT 128
83#define RTE_TEST_TX_DESC_DEFAULT 512
84static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
85static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
86
87/* ethernet addresses of ports */
88static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
89
90/* mask of enabled ports */
91static uint32_t l2fwd_enabled_port_mask;
92
93/* list of enabled ports */
94static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
95
96#define UPDATE_STEP_UP 1
97#define UPDATE_STEP_DOWN 32
98
99static unsigned int l2fwd_rx_queue_per_lcore = 1;
100
101#define MAX_RX_QUEUE_PER_LCORE 16
102#define MAX_TX_QUEUE_PER_PORT 16
103struct lcore_queue_conf {
104	unsigned n_rx_port;
105	unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
106	uint64_t next_flush_time[RTE_MAX_ETHPORTS];
107
108	struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
109	struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
110
111	struct rte_timer flush_timer;
112	struct rte_jobstats flush_job;
113	struct rte_jobstats idle_job;
114	struct rte_jobstats_context jobs_context;
115
116	rte_atomic16_t stats_read_pending;
117	rte_spinlock_t lock;
118} __rte_cache_aligned;
119struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
120
121struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
122
123static const struct rte_eth_conf port_conf = {
124	.rxmode = {
125		.split_hdr_size = 0,
126		.header_split   = 0, /**< Header Split disabled */
127		.hw_ip_checksum = 0, /**< IP checksum offload disabled */
128		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
129		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
130		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
131	},
132	.txmode = {
133		.mq_mode = ETH_MQ_TX_NONE,
134	},
135};
136
137struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
138
139/* Per-port statistics struct */
140struct l2fwd_port_statistics {
141	uint64_t tx;
142	uint64_t rx;
143	uint64_t dropped;
144} __rte_cache_aligned;
145struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
146
147/* 1 day max */
148#define MAX_TIMER_PERIOD 86400
149/* default period is 10 seconds */
150static int64_t timer_period = 10;
151/* default timer frequency */
152static double hz;
153/* BURST_TX_DRAIN_US converted to cycles */
154uint64_t drain_tsc;
155/* Convert cycles to ns */
156static inline double
157cycles_to_ns(uint64_t cycles)
158{
159	double t = cycles;
160
161	t *= (double)NS_PER_S;
162	t /= hz;
163	return t;
164}
165
166static void
167show_lcore_stats(unsigned lcore_id)
168{
169	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
170	struct rte_jobstats_context *ctx = &qconf->jobs_context;
171	struct rte_jobstats *job;
172	uint8_t i;
173
174	/* LCore statistics. */
175	uint64_t stats_period, loop_count;
176	uint64_t exec, exec_min, exec_max;
177	uint64_t management, management_min, management_max;
178	uint64_t busy, busy_min, busy_max;
179
180	/* Jobs statistics. */
181	const uint8_t port_cnt = qconf->n_rx_port;
182	uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
183	uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
184				jobs_exec_max[port_cnt];
185
186	uint64_t flush_exec_cnt, flush_period;
187	uint64_t flush_exec, flush_exec_min, flush_exec_max;
188
189	uint64_t idle_exec_cnt;
190	uint64_t idle_exec, idle_exec_min, idle_exec_max;
191	uint64_t collection_time = rte_get_timer_cycles();
192
193	/* Ask forwarding thread to give us stats. */
194	rte_atomic16_set(&qconf->stats_read_pending, 1);
195	rte_spinlock_lock(&qconf->lock);
196	rte_atomic16_set(&qconf->stats_read_pending, 0);
197
198	/* Collect context statistics. */
199	stats_period = ctx->state_time - ctx->start_time;
200	loop_count = ctx->loop_cnt;
201
202	exec = ctx->exec_time;
203	exec_min = ctx->min_exec_time;
204	exec_max = ctx->max_exec_time;
205
206	management = ctx->management_time;
207	management_min = ctx->min_management_time;
208	management_max = ctx->max_management_time;
209
210	rte_jobstats_context_reset(ctx);
211
212	for (i = 0; i < port_cnt; i++) {
213		job = &qconf->port_fwd_jobs[i];
214
215		jobs_exec_cnt[i] = job->exec_cnt;
216		jobs_period[i] = job->period;
217
218		jobs_exec[i] = job->exec_time;
219		jobs_exec_min[i] = job->min_exec_time;
220		jobs_exec_max[i] = job->max_exec_time;
221
222		rte_jobstats_reset(job);
223	}
224
225	flush_exec_cnt = qconf->flush_job.exec_cnt;
226	flush_period = qconf->flush_job.period;
227	flush_exec = qconf->flush_job.exec_time;
228	flush_exec_min = qconf->flush_job.min_exec_time;
229	flush_exec_max = qconf->flush_job.max_exec_time;
230	rte_jobstats_reset(&qconf->flush_job);
231
232	idle_exec_cnt = qconf->idle_job.exec_cnt;
233	idle_exec = qconf->idle_job.exec_time;
234	idle_exec_min = qconf->idle_job.min_exec_time;
235	idle_exec_max = qconf->idle_job.max_exec_time;
236	rte_jobstats_reset(&qconf->idle_job);
237
238	rte_spinlock_unlock(&qconf->lock);
239
240	exec -= idle_exec;
241	busy = exec + management;
242	busy_min = exec_min + management_min;
243	busy_max = exec_max + management_max;
244
245
246	collection_time = rte_get_timer_cycles() - collection_time;
247
248#define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
249
250	printf("\n----------------"
251			"\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
252			"\n%-18s %14s %7s %10s %10s %10s "
253			"\n%-18s %'14.0f"
254			"\n%-18s %'14" PRIu64
255			STAT_FMT /* Exec */
256			STAT_FMT /* Management */
257			STAT_FMT /* Busy */
258			STAT_FMT, /* Idle  */
259			lcore_id, cycles_to_ns(collection_time),
260			"Stat type", "total", "%total", "avg", "min", "max",
261			"Stats duration:", cycles_to_ns(stats_period),
262			"Loop count:", loop_count,
263			"Exec time",
264			cycles_to_ns(exec), exec * 100.0 / stats_period,
265			cycles_to_ns(loop_count  ? exec / loop_count : 0),
266			cycles_to_ns(exec_min),
267			cycles_to_ns(exec_max),
268			"Management time",
269			cycles_to_ns(management), management * 100.0 / stats_period,
270			cycles_to_ns(loop_count  ? management / loop_count : 0),
271			cycles_to_ns(management_min),
272			cycles_to_ns(management_max),
273			"Exec + management",
274			cycles_to_ns(busy),  busy * 100.0 / stats_period,
275			cycles_to_ns(loop_count ? busy / loop_count : 0),
276			cycles_to_ns(busy_min),
277			cycles_to_ns(busy_max),
278			"Idle (job)",
279			cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
280			cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
281			cycles_to_ns(idle_exec_min),
282			cycles_to_ns(idle_exec_max));
283
284	for (i = 0; i < qconf->n_rx_port; i++) {
285		job = &qconf->port_fwd_jobs[i];
286		printf("\n\nJob %" PRIu32 ": %-20s "
287				"\n%-18s %'14" PRIu64
288				"\n%-18s %'14.0f"
289				STAT_FMT,
290				i, job->name,
291				"Exec count:", jobs_exec_cnt[i],
292				"Exec period: ", cycles_to_ns(jobs_period[i]),
293				"Exec time",
294				cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
295				cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
296						: 0),
297				cycles_to_ns(jobs_exec_min[i]),
298				cycles_to_ns(jobs_exec_max[i]));
299	}
300
301	if (qconf->n_rx_port > 0) {
302		job = &qconf->flush_job;
303		printf("\n\nJob %" PRIu32 ": %-20s "
304				"\n%-18s %'14" PRIu64
305				"\n%-18s %'14.0f"
306				STAT_FMT,
307				i, job->name,
308				"Exec count:", flush_exec_cnt,
309				"Exec period: ", cycles_to_ns(flush_period),
310				"Exec time",
311				cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
312				cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
313				cycles_to_ns(flush_exec_min),
314				cycles_to_ns(flush_exec_max));
315	}
316}
317
318/* Print out statistics on packets dropped */
319static void
320show_stats_cb(__rte_unused void *param)
321{
322	uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
323	unsigned portid, lcore_id;
324
325	total_packets_dropped = 0;
326	total_packets_tx = 0;
327	total_packets_rx = 0;
328
329	const char clr[] = { 27, '[', '2', 'J', '\0' };
330	const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
331
332	/* Clear screen and move to top left */
333	printf("%s%s"
334			"\nPort statistics ===================================",
335			clr, topLeft);
336
337	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
338		/* skip disabled ports */
339		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
340			continue;
341		printf("\nStatistics for port %u ------------------------------"
342				"\nPackets sent: %24"PRIu64
343				"\nPackets received: %20"PRIu64
344				"\nPackets dropped: %21"PRIu64,
345				portid,
346				port_statistics[portid].tx,
347				port_statistics[portid].rx,
348				port_statistics[portid].dropped);
349
350		total_packets_dropped += port_statistics[portid].dropped;
351		total_packets_tx += port_statistics[portid].tx;
352		total_packets_rx += port_statistics[portid].rx;
353	}
354
355	printf("\nAggregate statistics ==============================="
356			"\nTotal packets sent: %18"PRIu64
357			"\nTotal packets received: %14"PRIu64
358			"\nTotal packets dropped: %15"PRIu64
359			"\n====================================================",
360			total_packets_tx,
361			total_packets_rx,
362			total_packets_dropped);
363
364	RTE_LCORE_FOREACH(lcore_id) {
365		if (lcore_queue_conf[lcore_id].n_rx_port > 0)
366			show_lcore_stats(lcore_id);
367	}
368
369	printf("\n====================================================\n");
370	rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
371}
372
373static void
374l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
375{
376	struct ether_hdr *eth;
377	void *tmp;
378	int sent;
379	unsigned dst_port;
380	struct rte_eth_dev_tx_buffer *buffer;
381
382	dst_port = l2fwd_dst_ports[portid];
383	eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
384
385	/* 02:00:00:00:00:xx */
386	tmp = &eth->d_addr.addr_bytes[0];
387	*((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
388
389	/* src addr */
390	ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->s_addr);
391
392	buffer = tx_buffer[dst_port];
393	sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
394	if (sent)
395		port_statistics[dst_port].tx += sent;
396}
397
398static void
399l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
400{
401	int64_t err = job->target - result;
402	int64_t histeresis = job->target / 8;
403
404	if (err < -histeresis) {
405		if (job->min_period + UPDATE_STEP_DOWN < job->period)
406			job->period -= UPDATE_STEP_DOWN;
407	} else if (err > histeresis) {
408		if (job->period + UPDATE_STEP_UP < job->max_period)
409			job->period += UPDATE_STEP_UP;
410	}
411}
412
413static void
414l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
415{
416	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
417	struct rte_mbuf *m;
418
419	const uint8_t port_idx = (uintptr_t) arg;
420	const unsigned lcore_id = rte_lcore_id();
421	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
422	struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
423	const uint8_t portid = qconf->rx_port_list[port_idx];
424
425	uint8_t j;
426	uint16_t total_nb_rx;
427
428	rte_jobstats_start(&qconf->jobs_context, job);
429
430	/* Call rx burst 2 times. This allow rte_jobstats logic to see if this
431	 * function must be called more frequently. */
432
433	total_nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
434			MAX_PKT_BURST);
435
436	for (j = 0; j < total_nb_rx; j++) {
437		m = pkts_burst[j];
438		rte_prefetch0(rte_pktmbuf_mtod(m, void *));
439		l2fwd_simple_forward(m, portid);
440	}
441
442	if (total_nb_rx == MAX_PKT_BURST) {
443		const uint16_t nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
444				MAX_PKT_BURST);
445
446		total_nb_rx += nb_rx;
447		for (j = 0; j < nb_rx; j++) {
448			m = pkts_burst[j];
449			rte_prefetch0(rte_pktmbuf_mtod(m, void *));
450			l2fwd_simple_forward(m, portid);
451		}
452	}
453
454	port_statistics[portid].rx += total_nb_rx;
455
456	/* Adjust period time in which we are running here. */
457	if (rte_jobstats_finish(job, total_nb_rx) != 0) {
458		rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
459				lcore_id, l2fwd_fwd_job, arg);
460	}
461}
462
463static void
464l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
465{
466	uint64_t now;
467	unsigned lcore_id;
468	struct lcore_queue_conf *qconf;
469	uint8_t portid;
470	unsigned i;
471	uint32_t sent;
472	struct rte_eth_dev_tx_buffer *buffer;
473
474	lcore_id = rte_lcore_id();
475	qconf = &lcore_queue_conf[lcore_id];
476
477	rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
478
479	now = rte_get_timer_cycles();
480	lcore_id = rte_lcore_id();
481	qconf = &lcore_queue_conf[lcore_id];
482
483	for (i = 0; i < qconf->n_rx_port; i++) {
484		portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
485
486		if (qconf->next_flush_time[portid] <= now)
487			continue;
488
489		buffer = tx_buffer[portid];
490		sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
491		if (sent)
492			port_statistics[portid].tx += sent;
493
494		qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc;
495	}
496
497	/* Pass target to indicate that this job is happy of time interwal
498	 * in which it was called. */
499	rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
500}
501
502/* main processing loop */
503static void
504l2fwd_main_loop(void)
505{
506	unsigned lcore_id;
507	unsigned i, portid;
508	struct lcore_queue_conf *qconf;
509	uint8_t stats_read_pending = 0;
510	uint8_t need_manage;
511
512	lcore_id = rte_lcore_id();
513	qconf = &lcore_queue_conf[lcore_id];
514
515	if (qconf->n_rx_port == 0) {
516		RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
517		return;
518	}
519
520	RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
521
522	for (i = 0; i < qconf->n_rx_port; i++) {
523
524		portid = qconf->rx_port_list[i];
525		RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
526			portid);
527	}
528
529	rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
530
531	for (;;) {
532		rte_spinlock_lock(&qconf->lock);
533
534		do {
535			rte_jobstats_context_start(&qconf->jobs_context);
536
537			/* Do the Idle job:
538			 * - Read stats_read_pending flag
539			 * - check if some real job need to be executed
540			 */
541			rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
542
543			uint64_t repeats = 0;
544
545			do {
546				uint8_t i;
547				uint64_t now = rte_get_timer_cycles();
548
549				repeats++;
550				need_manage = qconf->flush_timer.expire < now;
551				/* Check if we was esked to give a stats. */
552				stats_read_pending =
553						rte_atomic16_read(&qconf->stats_read_pending);
554				need_manage |= stats_read_pending;
555
556				for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
557					need_manage = qconf->rx_timers[i].expire < now;
558
559			} while (!need_manage);
560
561			if (likely(repeats != 1))
562				rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
563			else
564				rte_jobstats_abort(&qconf->idle_job);
565
566			rte_timer_manage();
567			rte_jobstats_context_finish(&qconf->jobs_context);
568		} while (likely(stats_read_pending == 0));
569
570		rte_spinlock_unlock(&qconf->lock);
571		rte_pause();
572	}
573}
574
575static int
576l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
577{
578	l2fwd_main_loop();
579	return 0;
580}
581
582/* display usage */
583static void
584l2fwd_usage(const char *prgname)
585{
586	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
587	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
588	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
589		   "  -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
590		   "  -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
591	       prgname);
592}
593
594static int
595l2fwd_parse_portmask(const char *portmask)
596{
597	char *end = NULL;
598	unsigned long pm;
599
600	/* parse hexadecimal string */
601	pm = strtoul(portmask, &end, 16);
602	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
603		return -1;
604
605	if (pm == 0)
606		return -1;
607
608	return pm;
609}
610
611static unsigned int
612l2fwd_parse_nqueue(const char *q_arg)
613{
614	char *end = NULL;
615	unsigned long n;
616
617	/* parse hexadecimal string */
618	n = strtoul(q_arg, &end, 10);
619	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
620		return 0;
621	if (n == 0)
622		return 0;
623	if (n >= MAX_RX_QUEUE_PER_LCORE)
624		return 0;
625
626	return n;
627}
628
629static int
630l2fwd_parse_timer_period(const char *q_arg)
631{
632	char *end = NULL;
633	int n;
634
635	/* parse number string */
636	n = strtol(q_arg, &end, 10);
637	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
638		return -1;
639	if (n >= MAX_TIMER_PERIOD)
640		return -1;
641
642	return n;
643}
644
645/* Parse the argument given in the command line of the application */
646static int
647l2fwd_parse_args(int argc, char **argv)
648{
649	int opt, ret;
650	char **argvopt;
651	int option_index;
652	char *prgname = argv[0];
653	static struct option lgopts[] = {
654		{NULL, 0, 0, 0}
655	};
656
657	argvopt = argv;
658
659	while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
660				  lgopts, &option_index)) != EOF) {
661
662		switch (opt) {
663		/* portmask */
664		case 'p':
665			l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
666			if (l2fwd_enabled_port_mask == 0) {
667				printf("invalid portmask\n");
668				l2fwd_usage(prgname);
669				return -1;
670			}
671			break;
672
673		/* nqueue */
674		case 'q':
675			l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
676			if (l2fwd_rx_queue_per_lcore == 0) {
677				printf("invalid queue number\n");
678				l2fwd_usage(prgname);
679				return -1;
680			}
681			break;
682
683		/* timer period */
684		case 'T':
685			timer_period = l2fwd_parse_timer_period(optarg);
686			if (timer_period < 0) {
687				printf("invalid timer period\n");
688				l2fwd_usage(prgname);
689				return -1;
690			}
691			break;
692
693		/* For thousands separator in printf. */
694		case 'l':
695			setlocale(LC_ALL, "");
696			break;
697
698		/* long options */
699		case 0:
700			l2fwd_usage(prgname);
701			return -1;
702
703		default:
704			l2fwd_usage(prgname);
705			return -1;
706		}
707	}
708
709	if (optind >= 0)
710		argv[optind-1] = prgname;
711
712	ret = optind-1;
713	optind = 0; /* reset getopt lib */
714	return ret;
715}
716
717/* Check the link status of all ports in up to 9s, and print them finally */
718static void
719check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
720{
721#define CHECK_INTERVAL 100 /* 100ms */
722#define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
723	uint8_t portid, count, all_ports_up, print_flag = 0;
724	struct rte_eth_link link;
725
726	printf("\nChecking link status");
727	fflush(stdout);
728	for (count = 0; count <= MAX_CHECK_TIME; count++) {
729		all_ports_up = 1;
730		for (portid = 0; portid < port_num; portid++) {
731			if ((port_mask & (1 << portid)) == 0)
732				continue;
733			memset(&link, 0, sizeof(link));
734			rte_eth_link_get_nowait(portid, &link);
735			/* print link status if flag set */
736			if (print_flag == 1) {
737				if (link.link_status)
738					printf("Port %d Link Up - speed %u "
739						"Mbps - %s\n", (uint8_t)portid,
740						(unsigned)link.link_speed,
741				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
742					("full-duplex") : ("half-duplex\n"));
743				else
744					printf("Port %d Link Down\n",
745						(uint8_t)portid);
746				continue;
747			}
748			/* clear all_ports_up flag if any link down */
749			if (link.link_status == ETH_LINK_DOWN) {
750				all_ports_up = 0;
751				break;
752			}
753		}
754		/* after finally printing all link status, get out */
755		if (print_flag == 1)
756			break;
757
758		if (all_ports_up == 0) {
759			printf(".");
760			fflush(stdout);
761			rte_delay_ms(CHECK_INTERVAL);
762		}
763
764		/* set the print_flag if all ports up or timeout */
765		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
766			print_flag = 1;
767			printf("done\n");
768		}
769	}
770}
771
772int
773main(int argc, char **argv)
774{
775	struct lcore_queue_conf *qconf;
776	struct rte_eth_dev_info dev_info;
777	unsigned lcore_id, rx_lcore_id;
778	unsigned nb_ports_in_mask = 0;
779	int ret;
780	char name[RTE_JOBSTATS_NAMESIZE];
781	uint8_t nb_ports;
782	uint8_t nb_ports_available;
783	uint8_t portid, last_port;
784	uint8_t i;
785
786	/* init EAL */
787	ret = rte_eal_init(argc, argv);
788	if (ret < 0)
789		rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
790	argc -= ret;
791	argv += ret;
792
793	/* parse application arguments (after the EAL ones) */
794	ret = l2fwd_parse_args(argc, argv);
795	if (ret < 0)
796		rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
797
798	rte_timer_subsystem_init();
799
800	/* fetch default timer frequency. */
801	hz = rte_get_timer_hz();
802
803	/* create the mbuf pool */
804	l2fwd_pktmbuf_pool =
805		rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
806			0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
807	if (l2fwd_pktmbuf_pool == NULL)
808		rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
809
810	nb_ports = rte_eth_dev_count();
811	if (nb_ports == 0)
812		rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
813
814	/* reset l2fwd_dst_ports */
815	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
816		l2fwd_dst_ports[portid] = 0;
817	last_port = 0;
818
819	/*
820	 * Each logical core is assigned a dedicated TX queue on each port.
821	 */
822	for (portid = 0; portid < nb_ports; portid++) {
823		/* skip ports that are not enabled */
824		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
825			continue;
826
827		if (nb_ports_in_mask % 2) {
828			l2fwd_dst_ports[portid] = last_port;
829			l2fwd_dst_ports[last_port] = portid;
830		} else
831			last_port = portid;
832
833		nb_ports_in_mask++;
834
835		rte_eth_dev_info_get(portid, &dev_info);
836	}
837	if (nb_ports_in_mask % 2) {
838		printf("Notice: odd number of ports in portmask.\n");
839		l2fwd_dst_ports[last_port] = last_port;
840	}
841
842	rx_lcore_id = 0;
843	qconf = NULL;
844
845	/* Initialize the port/queue configuration of each logical core */
846	for (portid = 0; portid < nb_ports; portid++) {
847		/* skip ports that are not enabled */
848		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
849			continue;
850
851		/* get the lcore_id for this port */
852		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
853		       lcore_queue_conf[rx_lcore_id].n_rx_port ==
854		       l2fwd_rx_queue_per_lcore) {
855			rx_lcore_id++;
856			if (rx_lcore_id >= RTE_MAX_LCORE)
857				rte_exit(EXIT_FAILURE, "Not enough cores\n");
858		}
859
860		if (qconf != &lcore_queue_conf[rx_lcore_id])
861			/* Assigned a new logical core in the loop above. */
862			qconf = &lcore_queue_conf[rx_lcore_id];
863
864		qconf->rx_port_list[qconf->n_rx_port] = portid;
865		qconf->n_rx_port++;
866		printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
867	}
868
869	nb_ports_available = nb_ports;
870
871	/* Initialise each port */
872	for (portid = 0; portid < nb_ports; portid++) {
873		/* skip ports that are not enabled */
874		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
875			printf("Skipping disabled port %u\n", (unsigned) portid);
876			nb_ports_available--;
877			continue;
878		}
879		/* init port */
880		printf("Initializing port %u... ", (unsigned) portid);
881		fflush(stdout);
882		ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
883		if (ret < 0)
884			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
885				  ret, (unsigned) portid);
886
887		rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
888
889		/* init one RX queue */
890		fflush(stdout);
891		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
892					     rte_eth_dev_socket_id(portid),
893					     NULL,
894					     l2fwd_pktmbuf_pool);
895		if (ret < 0)
896			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
897				  ret, (unsigned) portid);
898
899		/* init one TX queue on each port */
900		fflush(stdout);
901		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
902				rte_eth_dev_socket_id(portid),
903				NULL);
904		if (ret < 0)
905			rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
906				ret, (unsigned) portid);
907
908		/* Initialize TX buffers */
909		tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
910				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
911				rte_eth_dev_socket_id(portid));
912		if (tx_buffer[portid] == NULL)
913			rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
914					(unsigned) portid);
915
916		rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
917
918		ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
919				rte_eth_tx_buffer_count_callback,
920				&port_statistics[portid].dropped);
921		if (ret < 0)
922				rte_exit(EXIT_FAILURE, "Cannot set error callback for "
923						"tx buffer on port %u\n", (unsigned) portid);
924
925		/* Start device */
926		ret = rte_eth_dev_start(portid);
927		if (ret < 0)
928			rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
929				  ret, (unsigned) portid);
930
931		printf("done:\n");
932
933		rte_eth_promiscuous_enable(portid);
934
935		printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
936				(unsigned) portid,
937				l2fwd_ports_eth_addr[portid].addr_bytes[0],
938				l2fwd_ports_eth_addr[portid].addr_bytes[1],
939				l2fwd_ports_eth_addr[portid].addr_bytes[2],
940				l2fwd_ports_eth_addr[portid].addr_bytes[3],
941				l2fwd_ports_eth_addr[portid].addr_bytes[4],
942				l2fwd_ports_eth_addr[portid].addr_bytes[5]);
943
944		/* initialize port stats */
945		memset(&port_statistics, 0, sizeof(port_statistics));
946	}
947
948	if (!nb_ports_available) {
949		rte_exit(EXIT_FAILURE,
950			"All available ports are disabled. Please set portmask.\n");
951	}
952
953	check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
954
955	drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
956
957	RTE_LCORE_FOREACH(lcore_id) {
958		qconf = &lcore_queue_conf[lcore_id];
959
960		rte_spinlock_init(&qconf->lock);
961
962		if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
963			rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
964
965		if (qconf->n_rx_port == 0) {
966			RTE_LOG(INFO, L2FWD,
967				"lcore %u: no ports so no jobs stats context initialization\n",
968				lcore_id);
969			continue;
970		}
971		/* Add flush job.
972		 * Set fixed period by setting min = max = initial period. Set target to
973		 * zero as it is irrelevant for this job. */
974		rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
975				drain_tsc, 0);
976
977		rte_timer_init(&qconf->flush_timer);
978		ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
979				lcore_id, &l2fwd_flush_job, NULL);
980
981		if (ret < 0) {
982			rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
983					lcore_id, rte_strerror(-ret));
984		}
985
986		for (i = 0; i < qconf->n_rx_port; i++) {
987			struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
988
989			portid = qconf->rx_port_list[i];
990			printf("Setting forward job for port %u\n", portid);
991
992			snprintf(name, RTE_DIM(name), "port %u fwd", portid);
993			/* Setup forward job.
994			 * Set min, max and initial period. Set target to MAX_PKT_BURST as
995			 * this is desired optimal RX/TX burst size. */
996			rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
997			rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
998
999			rte_timer_init(&qconf->rx_timers[i]);
1000			ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
1001					&l2fwd_fwd_job, (void *)(uintptr_t)i);
1002
1003			if (ret < 0) {
1004				rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
1005						lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
1006			}
1007		}
1008	}
1009
1010	if (timer_period)
1011		rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
1012	else
1013		RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
1014
1015	/* launch per-lcore init on every lcore */
1016	rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
1017	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1018		if (rte_eal_wait_lcore(lcore_id) < 0)
1019			return -1;
1020	}
1021
1022	return 0;
1023}
1024