xref: /dpdk/app/test-pmd/testpmd.c (revision 23ea57a2a0cecbb904a5feeb0650db116585ee28)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #include <stdarg.h>
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <signal.h>
9 #include <string.h>
10 #include <time.h>
11 #include <fcntl.h>
12 #include <sys/mman.h>
13 #include <sys/types.h>
14 #include <errno.h>
15 #include <stdbool.h>
16 
17 #include <sys/queue.h>
18 #include <sys/stat.h>
19 
20 #include <stdint.h>
21 #include <unistd.h>
22 #include <inttypes.h>
23 
24 #include <rte_common.h>
25 #include <rte_errno.h>
26 #include <rte_byteorder.h>
27 #include <rte_log.h>
28 #include <rte_debug.h>
29 #include <rte_cycles.h>
30 #include <rte_malloc_heap.h>
31 #include <rte_memory.h>
32 #include <rte_memcpy.h>
33 #include <rte_launch.h>
34 #include <rte_eal.h>
35 #include <rte_alarm.h>
36 #include <rte_per_lcore.h>
37 #include <rte_lcore.h>
38 #include <rte_atomic.h>
39 #include <rte_branch_prediction.h>
40 #include <rte_mempool.h>
41 #include <rte_malloc.h>
42 #include <rte_mbuf.h>
43 #include <rte_mbuf_pool_ops.h>
44 #include <rte_interrupts.h>
45 #include <rte_pci.h>
46 #include <rte_ether.h>
47 #include <rte_ethdev.h>
48 #include <rte_dev.h>
49 #include <rte_string_fns.h>
50 #ifdef RTE_LIBRTE_IXGBE_PMD
51 #include <rte_pmd_ixgbe.h>
52 #endif
53 #ifdef RTE_LIBRTE_PDUMP
54 #include <rte_pdump.h>
55 #endif
56 #include <rte_flow.h>
57 #include <rte_metrics.h>
58 #ifdef RTE_LIBRTE_BITRATE
59 #include <rte_bitrate.h>
60 #endif
61 #ifdef RTE_LIBRTE_LATENCY_STATS
62 #include <rte_latencystats.h>
63 #endif
64 
65 #include "testpmd.h"
66 
67 #ifndef MAP_HUGETLB
68 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
69 #define HUGE_FLAG (0x40000)
70 #else
71 #define HUGE_FLAG MAP_HUGETLB
72 #endif
73 
74 #ifndef MAP_HUGE_SHIFT
75 /* older kernels (or FreeBSD) will not have this define */
76 #define HUGE_SHIFT (26)
77 #else
78 #define HUGE_SHIFT MAP_HUGE_SHIFT
79 #endif
80 
81 #define EXTMEM_HEAP_NAME "extmem"
82 
83 uint16_t verbose_level = 0; /**< Silent by default. */
84 int testpmd_logtype; /**< Log type for testpmd logs */
85 
86 /* use master core for command line ? */
87 uint8_t interactive = 0;
88 uint8_t auto_start = 0;
89 uint8_t tx_first;
90 char cmdline_filename[PATH_MAX] = {0};
91 
92 /*
93  * NUMA support configuration.
94  * When set, the NUMA support attempts to dispatch the allocation of the
95  * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
96  * probed ports among the CPU sockets 0 and 1.
97  * Otherwise, all memory is allocated from CPU socket 0.
98  */
99 uint8_t numa_support = 1; /**< numa enabled by default */
100 
101 /*
102  * In UMA mode,all memory is allocated from socket 0 if --socket-num is
103  * not configured.
104  */
105 uint8_t socket_num = UMA_NO_CONFIG;
106 
107 /*
108  * Select mempool allocation type:
109  * - native: use regular DPDK memory
110  * - anon: use regular DPDK memory to create mempool, but populate using
111  *         anonymous memory (may not be IOVA-contiguous)
112  * - xmem: use externally allocated hugepage memory
113  */
114 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
115 
116 /*
117  * Store specified sockets on which memory pool to be used by ports
118  * is allocated.
119  */
120 uint8_t port_numa[RTE_MAX_ETHPORTS];
121 
122 /*
123  * Store specified sockets on which RX ring to be used by ports
124  * is allocated.
125  */
126 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
127 
128 /*
129  * Store specified sockets on which TX ring to be used by ports
130  * is allocated.
131  */
132 uint8_t txring_numa[RTE_MAX_ETHPORTS];
133 
134 /*
135  * Record the Ethernet address of peer target ports to which packets are
136  * forwarded.
137  * Must be instantiated with the ethernet addresses of peer traffic generator
138  * ports.
139  */
140 struct ether_addr peer_eth_addrs[RTE_MAX_ETHPORTS];
141 portid_t nb_peer_eth_addrs = 0;
142 
143 /*
144  * Probed Target Environment.
145  */
146 struct rte_port *ports;	       /**< For all probed ethernet ports. */
147 portid_t nb_ports;             /**< Number of probed ethernet ports. */
148 struct fwd_lcore **fwd_lcores; /**< For all probed logical cores. */
149 lcoreid_t nb_lcores;           /**< Number of probed logical cores. */
150 
151 portid_t ports_ids[RTE_MAX_ETHPORTS]; /**< Store all port ids. */
152 
153 /*
154  * Test Forwarding Configuration.
155  *    nb_fwd_lcores <= nb_cfg_lcores <= nb_lcores
156  *    nb_fwd_ports  <= nb_cfg_ports  <= nb_ports
157  */
158 lcoreid_t nb_cfg_lcores; /**< Number of configured logical cores. */
159 lcoreid_t nb_fwd_lcores; /**< Number of forwarding logical cores. */
160 portid_t  nb_cfg_ports;  /**< Number of configured ports. */
161 portid_t  nb_fwd_ports;  /**< Number of forwarding ports. */
162 
163 unsigned int fwd_lcores_cpuids[RTE_MAX_LCORE]; /**< CPU ids configuration. */
164 portid_t fwd_ports_ids[RTE_MAX_ETHPORTS];      /**< Port ids configuration. */
165 
166 struct fwd_stream **fwd_streams; /**< For each RX queue of each port. */
167 streamid_t nb_fwd_streams;       /**< Is equal to (nb_ports * nb_rxq). */
168 
169 /*
170  * Forwarding engines.
171  */
172 struct fwd_engine * fwd_engines[] = {
173 	&io_fwd_engine,
174 	&mac_fwd_engine,
175 	&mac_swap_engine,
176 	&flow_gen_engine,
177 	&rx_only_engine,
178 	&tx_only_engine,
179 	&csum_fwd_engine,
180 	&icmp_echo_engine,
181 	&noisy_vnf_engine,
182 #if defined RTE_LIBRTE_PMD_SOFTNIC
183 	&softnic_fwd_engine,
184 #endif
185 #ifdef RTE_LIBRTE_IEEE1588
186 	&ieee1588_fwd_engine,
187 #endif
188 	NULL,
189 };
190 
191 struct fwd_config cur_fwd_config;
192 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
193 uint32_t retry_enabled;
194 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
195 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
196 
197 uint16_t mbuf_data_size = DEFAULT_MBUF_DATA_SIZE; /**< Mbuf data space size. */
198 uint32_t param_total_num_mbufs = 0;  /**< number of mbufs in all pools - if
199                                       * specified on command-line. */
200 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
201 
202 /*
203  * In container, it cannot terminate the process which running with 'stats-period'
204  * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
205  */
206 uint8_t f_quit;
207 
208 /*
209  * Configuration of packet segments used by the "txonly" processing engine.
210  */
211 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
212 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
213 	TXONLY_DEF_PACKET_LEN,
214 };
215 uint8_t  tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
216 
217 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
218 /**< Split policy for packets to TX. */
219 
220 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
221 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
222 
223 /* current configuration is in DCB or not,0 means it is not in DCB mode */
224 uint8_t dcb_config = 0;
225 
226 /* Whether the dcb is in testing status */
227 uint8_t dcb_test = 0;
228 
229 /*
230  * Configurable number of RX/TX queues.
231  */
232 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
233 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
234 
235 /*
236  * Configurable number of RX/TX ring descriptors.
237  * Defaults are supplied by drivers via ethdev.
238  */
239 #define RTE_TEST_RX_DESC_DEFAULT 0
240 #define RTE_TEST_TX_DESC_DEFAULT 0
241 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
242 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
243 
244 #define RTE_PMD_PARAM_UNSET -1
245 /*
246  * Configurable values of RX and TX ring threshold registers.
247  */
248 
249 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
250 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
251 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
252 
253 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
254 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
255 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
256 
257 /*
258  * Configurable value of RX free threshold.
259  */
260 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
261 
262 /*
263  * Configurable value of RX drop enable.
264  */
265 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
266 
267 /*
268  * Configurable value of TX free threshold.
269  */
270 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
271 
272 /*
273  * Configurable value of TX RS bit threshold.
274  */
275 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
276 
277 /*
278  * Configurable value of buffered packets before sending.
279  */
280 uint16_t noisy_tx_sw_bufsz;
281 
282 /*
283  * Configurable value of packet buffer timeout.
284  */
285 uint16_t noisy_tx_sw_buf_flush_time;
286 
287 /*
288  * Configurable value for size of VNF internal memory area
289  * used for simulating noisy neighbour behaviour
290  */
291 uint64_t noisy_lkup_mem_sz;
292 
293 /*
294  * Configurable value of number of random writes done in
295  * VNF simulation memory area.
296  */
297 uint64_t noisy_lkup_num_writes;
298 
299 /*
300  * Configurable value of number of random reads done in
301  * VNF simulation memory area.
302  */
303 uint64_t noisy_lkup_num_reads;
304 
305 /*
306  * Configurable value of number of random reads/writes done in
307  * VNF simulation memory area.
308  */
309 uint64_t noisy_lkup_num_reads_writes;
310 
311 /*
312  * Receive Side Scaling (RSS) configuration.
313  */
314 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
315 
316 /*
317  * Port topology configuration
318  */
319 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
320 
321 /*
322  * Avoids to flush all the RX streams before starts forwarding.
323  */
324 uint8_t no_flush_rx = 0; /* flush by default */
325 
326 /*
327  * Flow API isolated mode.
328  */
329 uint8_t flow_isolate_all;
330 
331 /*
332  * Avoids to check link status when starting/stopping a port.
333  */
334 uint8_t no_link_check = 0; /* check by default */
335 
336 /*
337  * Enable link status change notification
338  */
339 uint8_t lsc_interrupt = 1; /* enabled by default */
340 
341 /*
342  * Enable device removal notification.
343  */
344 uint8_t rmv_interrupt = 1; /* enabled by default */
345 
346 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
347 
348 /*
349  * Display or mask ether events
350  * Default to all events except VF_MBOX
351  */
352 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
353 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
354 			    (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
355 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
356 			    (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
357 			    (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
358 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV);
359 /*
360  * Decide if all memory are locked for performance.
361  */
362 int do_mlockall = 0;
363 
364 /*
365  * NIC bypass mode configuration options.
366  */
367 
368 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
369 /* The NIC bypass watchdog timeout. */
370 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
371 #endif
372 
373 
374 #ifdef RTE_LIBRTE_LATENCY_STATS
375 
376 /*
377  * Set when latency stats is enabled in the commandline
378  */
379 uint8_t latencystats_enabled;
380 
381 /*
382  * Lcore ID to serive latency statistics.
383  */
384 lcoreid_t latencystats_lcore_id = -1;
385 
386 #endif
387 
388 /*
389  * Ethernet device configuration.
390  */
391 struct rte_eth_rxmode rx_mode = {
392 	.max_rx_pkt_len = ETHER_MAX_LEN, /**< Default maximum frame length. */
393 };
394 
395 struct rte_eth_txmode tx_mode = {
396 	.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
397 };
398 
399 struct rte_fdir_conf fdir_conf = {
400 	.mode = RTE_FDIR_MODE_NONE,
401 	.pballoc = RTE_FDIR_PBALLOC_64K,
402 	.status = RTE_FDIR_REPORT_STATUS,
403 	.mask = {
404 		.vlan_tci_mask = 0xFFEF,
405 		.ipv4_mask     = {
406 			.src_ip = 0xFFFFFFFF,
407 			.dst_ip = 0xFFFFFFFF,
408 		},
409 		.ipv6_mask     = {
410 			.src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
411 			.dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
412 		},
413 		.src_port_mask = 0xFFFF,
414 		.dst_port_mask = 0xFFFF,
415 		.mac_addr_byte_mask = 0xFF,
416 		.tunnel_type_mask = 1,
417 		.tunnel_id_mask = 0xFFFFFFFF,
418 	},
419 	.drop_queue = 127,
420 };
421 
422 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
423 
424 struct queue_stats_mappings tx_queue_stats_mappings_array[MAX_TX_QUEUE_STATS_MAPPINGS];
425 struct queue_stats_mappings rx_queue_stats_mappings_array[MAX_RX_QUEUE_STATS_MAPPINGS];
426 
427 struct queue_stats_mappings *tx_queue_stats_mappings = tx_queue_stats_mappings_array;
428 struct queue_stats_mappings *rx_queue_stats_mappings = rx_queue_stats_mappings_array;
429 
430 uint16_t nb_tx_queue_stats_mappings = 0;
431 uint16_t nb_rx_queue_stats_mappings = 0;
432 
433 /*
434  * Display zero values by default for xstats
435  */
436 uint8_t xstats_hide_zero;
437 
438 unsigned int num_sockets = 0;
439 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
440 
441 #ifdef RTE_LIBRTE_BITRATE
442 /* Bitrate statistics */
443 struct rte_stats_bitrates *bitrate_data;
444 lcoreid_t bitrate_lcore_id;
445 uint8_t bitrate_enabled;
446 #endif
447 
448 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
449 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
450 
451 struct vxlan_encap_conf vxlan_encap_conf = {
452 	.select_ipv4 = 1,
453 	.select_vlan = 0,
454 	.vni = "\x00\x00\x00",
455 	.udp_src = 0,
456 	.udp_dst = RTE_BE16(4789),
457 	.ipv4_src = IPv4(127, 0, 0, 1),
458 	.ipv4_dst = IPv4(255, 255, 255, 255),
459 	.ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
460 		"\x00\x00\x00\x00\x00\x00\x00\x01",
461 	.ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
462 		"\x00\x00\x00\x00\x00\x00\x11\x11",
463 	.vlan_tci = 0,
464 	.eth_src = "\x00\x00\x00\x00\x00\x00",
465 	.eth_dst = "\xff\xff\xff\xff\xff\xff",
466 };
467 
468 struct nvgre_encap_conf nvgre_encap_conf = {
469 	.select_ipv4 = 1,
470 	.select_vlan = 0,
471 	.tni = "\x00\x00\x00",
472 	.ipv4_src = IPv4(127, 0, 0, 1),
473 	.ipv4_dst = IPv4(255, 255, 255, 255),
474 	.ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
475 		"\x00\x00\x00\x00\x00\x00\x00\x01",
476 	.ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
477 		"\x00\x00\x00\x00\x00\x00\x11\x11",
478 	.vlan_tci = 0,
479 	.eth_src = "\x00\x00\x00\x00\x00\x00",
480 	.eth_dst = "\xff\xff\xff\xff\xff\xff",
481 };
482 
483 /* Forward function declarations */
484 static void map_port_queue_stats_mapping_registers(portid_t pi,
485 						   struct rte_port *port);
486 static void check_all_ports_link_status(uint32_t port_mask);
487 static int eth_event_callback(portid_t port_id,
488 			      enum rte_eth_event_type type,
489 			      void *param, void *ret_param);
490 static void eth_dev_event_callback(const char *device_name,
491 				enum rte_dev_event_type type,
492 				void *param);
493 
494 /*
495  * Check if all the ports are started.
496  * If yes, return positive value. If not, return zero.
497  */
498 static int all_ports_started(void);
499 
500 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
501 uint16_t gso_max_segment_size = ETHER_MAX_LEN - ETHER_CRC_LEN;
502 
503 /*
504  * Helper function to check if socket is already discovered.
505  * If yes, return positive value. If not, return zero.
506  */
507 int
508 new_socket_id(unsigned int socket_id)
509 {
510 	unsigned int i;
511 
512 	for (i = 0; i < num_sockets; i++) {
513 		if (socket_ids[i] == socket_id)
514 			return 0;
515 	}
516 	return 1;
517 }
518 
519 /*
520  * Setup default configuration.
521  */
522 static void
523 set_default_fwd_lcores_config(void)
524 {
525 	unsigned int i;
526 	unsigned int nb_lc;
527 	unsigned int sock_num;
528 
529 	nb_lc = 0;
530 	for (i = 0; i < RTE_MAX_LCORE; i++) {
531 		if (!rte_lcore_is_enabled(i))
532 			continue;
533 		sock_num = rte_lcore_to_socket_id(i);
534 		if (new_socket_id(sock_num)) {
535 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
536 				rte_exit(EXIT_FAILURE,
537 					 "Total sockets greater than %u\n",
538 					 RTE_MAX_NUMA_NODES);
539 			}
540 			socket_ids[num_sockets++] = sock_num;
541 		}
542 		if (i == rte_get_master_lcore())
543 			continue;
544 		fwd_lcores_cpuids[nb_lc++] = i;
545 	}
546 	nb_lcores = (lcoreid_t) nb_lc;
547 	nb_cfg_lcores = nb_lcores;
548 	nb_fwd_lcores = 1;
549 }
550 
551 static void
552 set_def_peer_eth_addrs(void)
553 {
554 	portid_t i;
555 
556 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
557 		peer_eth_addrs[i].addr_bytes[0] = ETHER_LOCAL_ADMIN_ADDR;
558 		peer_eth_addrs[i].addr_bytes[5] = i;
559 	}
560 }
561 
562 static void
563 set_default_fwd_ports_config(void)
564 {
565 	portid_t pt_id;
566 	int i = 0;
567 
568 	RTE_ETH_FOREACH_DEV(pt_id) {
569 		fwd_ports_ids[i++] = pt_id;
570 
571 		/* Update sockets info according to the attached device */
572 		int socket_id = rte_eth_dev_socket_id(pt_id);
573 		if (socket_id >= 0 && new_socket_id(socket_id)) {
574 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
575 				rte_exit(EXIT_FAILURE,
576 					 "Total sockets greater than %u\n",
577 					 RTE_MAX_NUMA_NODES);
578 			}
579 			socket_ids[num_sockets++] = socket_id;
580 		}
581 	}
582 
583 	nb_cfg_ports = nb_ports;
584 	nb_fwd_ports = nb_ports;
585 }
586 
587 void
588 set_def_fwd_config(void)
589 {
590 	set_default_fwd_lcores_config();
591 	set_def_peer_eth_addrs();
592 	set_default_fwd_ports_config();
593 }
594 
595 /* extremely pessimistic estimation of memory required to create a mempool */
596 static int
597 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
598 {
599 	unsigned int n_pages, mbuf_per_pg, leftover;
600 	uint64_t total_mem, mbuf_mem, obj_sz;
601 
602 	/* there is no good way to predict how much space the mempool will
603 	 * occupy because it will allocate chunks on the fly, and some of those
604 	 * will come from default DPDK memory while some will come from our
605 	 * external memory, so just assume 128MB will be enough for everyone.
606 	 */
607 	uint64_t hdr_mem = 128 << 20;
608 
609 	/* account for possible non-contiguousness */
610 	obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
611 	if (obj_sz > pgsz) {
612 		TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
613 		return -1;
614 	}
615 
616 	mbuf_per_pg = pgsz / obj_sz;
617 	leftover = (nb_mbufs % mbuf_per_pg) > 0;
618 	n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
619 
620 	mbuf_mem = n_pages * pgsz;
621 
622 	total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
623 
624 	if (total_mem > SIZE_MAX) {
625 		TESTPMD_LOG(ERR, "Memory size too big\n");
626 		return -1;
627 	}
628 	*out = (size_t)total_mem;
629 
630 	return 0;
631 }
632 
633 static inline uint32_t
634 bsf64(uint64_t v)
635 {
636 	return (uint32_t)__builtin_ctzll(v);
637 }
638 
639 static inline uint32_t
640 log2_u64(uint64_t v)
641 {
642 	if (v == 0)
643 		return 0;
644 	v = rte_align64pow2(v);
645 	return bsf64(v);
646 }
647 
648 static int
649 pagesz_flags(uint64_t page_sz)
650 {
651 	/* as per mmap() manpage, all page sizes are log2 of page size
652 	 * shifted by MAP_HUGE_SHIFT
653 	 */
654 	int log2 = log2_u64(page_sz);
655 
656 	return (log2 << HUGE_SHIFT);
657 }
658 
659 static void *
660 alloc_mem(size_t memsz, size_t pgsz, bool huge)
661 {
662 	void *addr;
663 	int flags;
664 
665 	/* allocate anonymous hugepages */
666 	flags = MAP_ANONYMOUS | MAP_PRIVATE;
667 	if (huge)
668 		flags |= HUGE_FLAG | pagesz_flags(pgsz);
669 
670 	addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
671 	if (addr == MAP_FAILED)
672 		return NULL;
673 
674 	return addr;
675 }
676 
677 struct extmem_param {
678 	void *addr;
679 	size_t len;
680 	size_t pgsz;
681 	rte_iova_t *iova_table;
682 	unsigned int iova_table_len;
683 };
684 
685 static int
686 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
687 		bool huge)
688 {
689 	uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
690 			RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
691 	unsigned int cur_page, n_pages, pgsz_idx;
692 	size_t mem_sz, cur_pgsz;
693 	rte_iova_t *iovas = NULL;
694 	void *addr;
695 	int ret;
696 
697 	for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
698 		/* skip anything that is too big */
699 		if (pgsizes[pgsz_idx] > SIZE_MAX)
700 			continue;
701 
702 		cur_pgsz = pgsizes[pgsz_idx];
703 
704 		/* if we were told not to allocate hugepages, override */
705 		if (!huge)
706 			cur_pgsz = sysconf(_SC_PAGESIZE);
707 
708 		ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
709 		if (ret < 0) {
710 			TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
711 			return -1;
712 		}
713 
714 		/* allocate our memory */
715 		addr = alloc_mem(mem_sz, cur_pgsz, huge);
716 
717 		/* if we couldn't allocate memory with a specified page size,
718 		 * that doesn't mean we can't do it with other page sizes, so
719 		 * try another one.
720 		 */
721 		if (addr == NULL)
722 			continue;
723 
724 		/* store IOVA addresses for every page in this memory area */
725 		n_pages = mem_sz / cur_pgsz;
726 
727 		iovas = malloc(sizeof(*iovas) * n_pages);
728 
729 		if (iovas == NULL) {
730 			TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
731 			goto fail;
732 		}
733 		/* lock memory if it's not huge pages */
734 		if (!huge)
735 			mlock(addr, mem_sz);
736 
737 		/* populate IOVA addresses */
738 		for (cur_page = 0; cur_page < n_pages; cur_page++) {
739 			rte_iova_t iova;
740 			size_t offset;
741 			void *cur;
742 
743 			offset = cur_pgsz * cur_page;
744 			cur = RTE_PTR_ADD(addr, offset);
745 
746 			/* touch the page before getting its IOVA */
747 			*(volatile char *)cur = 0;
748 
749 			iova = rte_mem_virt2iova(cur);
750 
751 			iovas[cur_page] = iova;
752 		}
753 
754 		break;
755 	}
756 	/* if we couldn't allocate anything */
757 	if (iovas == NULL)
758 		return -1;
759 
760 	param->addr = addr;
761 	param->len = mem_sz;
762 	param->pgsz = cur_pgsz;
763 	param->iova_table = iovas;
764 	param->iova_table_len = n_pages;
765 
766 	return 0;
767 fail:
768 	if (iovas)
769 		free(iovas);
770 	if (addr)
771 		munmap(addr, mem_sz);
772 
773 	return -1;
774 }
775 
776 static int
777 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
778 {
779 	struct extmem_param param;
780 	int socket_id, ret;
781 
782 	memset(&param, 0, sizeof(param));
783 
784 	/* check if our heap exists */
785 	socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
786 	if (socket_id < 0) {
787 		/* create our heap */
788 		ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
789 		if (ret < 0) {
790 			TESTPMD_LOG(ERR, "Cannot create heap\n");
791 			return -1;
792 		}
793 	}
794 
795 	ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
796 	if (ret < 0) {
797 		TESTPMD_LOG(ERR, "Cannot create memory area\n");
798 		return -1;
799 	}
800 
801 	/* we now have a valid memory area, so add it to heap */
802 	ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
803 			param.addr, param.len, param.iova_table,
804 			param.iova_table_len, param.pgsz);
805 
806 	/* when using VFIO, memory is automatically mapped for DMA by EAL */
807 
808 	/* not needed any more */
809 	free(param.iova_table);
810 
811 	if (ret < 0) {
812 		TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
813 		munmap(param.addr, param.len);
814 		return -1;
815 	}
816 
817 	/* success */
818 
819 	TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
820 			param.len >> 20);
821 
822 	return 0;
823 }
824 
825 /*
826  * Configuration initialisation done once at init time.
827  */
828 static void
829 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
830 		 unsigned int socket_id)
831 {
832 	char pool_name[RTE_MEMPOOL_NAMESIZE];
833 	struct rte_mempool *rte_mp = NULL;
834 	uint32_t mb_size;
835 
836 	mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
837 	mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name));
838 
839 	TESTPMD_LOG(INFO,
840 		"create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
841 		pool_name, nb_mbuf, mbuf_seg_size, socket_id);
842 
843 	switch (mp_alloc_type) {
844 	case MP_ALLOC_NATIVE:
845 		{
846 			/* wrapper to rte_mempool_create() */
847 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
848 					rte_mbuf_best_mempool_ops());
849 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
850 				mb_mempool_cache, 0, mbuf_seg_size, socket_id);
851 			break;
852 		}
853 	case MP_ALLOC_ANON:
854 		{
855 			rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
856 				mb_size, (unsigned int) mb_mempool_cache,
857 				sizeof(struct rte_pktmbuf_pool_private),
858 				socket_id, 0);
859 			if (rte_mp == NULL)
860 				goto err;
861 
862 			if (rte_mempool_populate_anon(rte_mp) == 0) {
863 				rte_mempool_free(rte_mp);
864 				rte_mp = NULL;
865 				goto err;
866 			}
867 			rte_pktmbuf_pool_init(rte_mp, NULL);
868 			rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
869 			break;
870 		}
871 	case MP_ALLOC_XMEM:
872 	case MP_ALLOC_XMEM_HUGE:
873 		{
874 			int heap_socket;
875 			bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
876 
877 			if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
878 				rte_exit(EXIT_FAILURE, "Could not create external memory\n");
879 
880 			heap_socket =
881 				rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
882 			if (heap_socket < 0)
883 				rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
884 
885 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
886 					rte_mbuf_best_mempool_ops());
887 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
888 					mb_mempool_cache, 0, mbuf_seg_size,
889 					heap_socket);
890 			break;
891 		}
892 	default:
893 		{
894 			rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
895 		}
896 	}
897 
898 err:
899 	if (rte_mp == NULL) {
900 		rte_exit(EXIT_FAILURE,
901 			"Creation of mbuf pool for socket %u failed: %s\n",
902 			socket_id, rte_strerror(rte_errno));
903 	} else if (verbose_level > 0) {
904 		rte_mempool_dump(stdout, rte_mp);
905 	}
906 }
907 
908 /*
909  * Check given socket id is valid or not with NUMA mode,
910  * if valid, return 0, else return -1
911  */
912 static int
913 check_socket_id(const unsigned int socket_id)
914 {
915 	static int warning_once = 0;
916 
917 	if (new_socket_id(socket_id)) {
918 		if (!warning_once && numa_support)
919 			printf("Warning: NUMA should be configured manually by"
920 			       " using --port-numa-config and"
921 			       " --ring-numa-config parameters along with"
922 			       " --numa.\n");
923 		warning_once = 1;
924 		return -1;
925 	}
926 	return 0;
927 }
928 
929 /*
930  * Get the allowed maximum number of RX queues.
931  * *pid return the port id which has minimal value of
932  * max_rx_queues in all ports.
933  */
934 queueid_t
935 get_allowed_max_nb_rxq(portid_t *pid)
936 {
937 	queueid_t allowed_max_rxq = MAX_QUEUE_ID;
938 	portid_t pi;
939 	struct rte_eth_dev_info dev_info;
940 
941 	RTE_ETH_FOREACH_DEV(pi) {
942 		rte_eth_dev_info_get(pi, &dev_info);
943 		if (dev_info.max_rx_queues < allowed_max_rxq) {
944 			allowed_max_rxq = dev_info.max_rx_queues;
945 			*pid = pi;
946 		}
947 	}
948 	return allowed_max_rxq;
949 }
950 
951 /*
952  * Check input rxq is valid or not.
953  * If input rxq is not greater than any of maximum number
954  * of RX queues of all ports, it is valid.
955  * if valid, return 0, else return -1
956  */
957 int
958 check_nb_rxq(queueid_t rxq)
959 {
960 	queueid_t allowed_max_rxq;
961 	portid_t pid = 0;
962 
963 	allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
964 	if (rxq > allowed_max_rxq) {
965 		printf("Fail: input rxq (%u) can't be greater "
966 		       "than max_rx_queues (%u) of port %u\n",
967 		       rxq,
968 		       allowed_max_rxq,
969 		       pid);
970 		return -1;
971 	}
972 	return 0;
973 }
974 
975 /*
976  * Get the allowed maximum number of TX queues.
977  * *pid return the port id which has minimal value of
978  * max_tx_queues in all ports.
979  */
980 queueid_t
981 get_allowed_max_nb_txq(portid_t *pid)
982 {
983 	queueid_t allowed_max_txq = MAX_QUEUE_ID;
984 	portid_t pi;
985 	struct rte_eth_dev_info dev_info;
986 
987 	RTE_ETH_FOREACH_DEV(pi) {
988 		rte_eth_dev_info_get(pi, &dev_info);
989 		if (dev_info.max_tx_queues < allowed_max_txq) {
990 			allowed_max_txq = dev_info.max_tx_queues;
991 			*pid = pi;
992 		}
993 	}
994 	return allowed_max_txq;
995 }
996 
997 /*
998  * Check input txq is valid or not.
999  * If input txq is not greater than any of maximum number
1000  * of TX queues of all ports, it is valid.
1001  * if valid, return 0, else return -1
1002  */
1003 int
1004 check_nb_txq(queueid_t txq)
1005 {
1006 	queueid_t allowed_max_txq;
1007 	portid_t pid = 0;
1008 
1009 	allowed_max_txq = get_allowed_max_nb_txq(&pid);
1010 	if (txq > allowed_max_txq) {
1011 		printf("Fail: input txq (%u) can't be greater "
1012 		       "than max_tx_queues (%u) of port %u\n",
1013 		       txq,
1014 		       allowed_max_txq,
1015 		       pid);
1016 		return -1;
1017 	}
1018 	return 0;
1019 }
1020 
1021 static void
1022 init_config(void)
1023 {
1024 	portid_t pid;
1025 	struct rte_port *port;
1026 	struct rte_mempool *mbp;
1027 	unsigned int nb_mbuf_per_pool;
1028 	lcoreid_t  lc_id;
1029 	uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
1030 	struct rte_gro_param gro_param;
1031 	uint32_t gso_types;
1032 	int k;
1033 
1034 	memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
1035 
1036 	/* Configuration of logical cores. */
1037 	fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1038 				sizeof(struct fwd_lcore *) * nb_lcores,
1039 				RTE_CACHE_LINE_SIZE);
1040 	if (fwd_lcores == NULL) {
1041 		rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1042 							"failed\n", nb_lcores);
1043 	}
1044 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1045 		fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1046 					       sizeof(struct fwd_lcore),
1047 					       RTE_CACHE_LINE_SIZE);
1048 		if (fwd_lcores[lc_id] == NULL) {
1049 			rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1050 								"failed\n");
1051 		}
1052 		fwd_lcores[lc_id]->cpuid_idx = lc_id;
1053 	}
1054 
1055 	RTE_ETH_FOREACH_DEV(pid) {
1056 		port = &ports[pid];
1057 		/* Apply default TxRx configuration for all ports */
1058 		port->dev_conf.txmode = tx_mode;
1059 		port->dev_conf.rxmode = rx_mode;
1060 		rte_eth_dev_info_get(pid, &port->dev_info);
1061 
1062 		if (!(port->dev_info.tx_offload_capa &
1063 		      DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1064 			port->dev_conf.txmode.offloads &=
1065 				~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1066 		if (numa_support) {
1067 			if (port_numa[pid] != NUMA_NO_CONFIG)
1068 				port_per_socket[port_numa[pid]]++;
1069 			else {
1070 				uint32_t socket_id = rte_eth_dev_socket_id(pid);
1071 
1072 				/*
1073 				 * if socket_id is invalid,
1074 				 * set to the first available socket.
1075 				 */
1076 				if (check_socket_id(socket_id) < 0)
1077 					socket_id = socket_ids[0];
1078 				port_per_socket[socket_id]++;
1079 			}
1080 		}
1081 
1082 		/* Apply Rx offloads configuration */
1083 		for (k = 0; k < port->dev_info.max_rx_queues; k++)
1084 			port->rx_conf[k].offloads =
1085 				port->dev_conf.rxmode.offloads;
1086 		/* Apply Tx offloads configuration */
1087 		for (k = 0; k < port->dev_info.max_tx_queues; k++)
1088 			port->tx_conf[k].offloads =
1089 				port->dev_conf.txmode.offloads;
1090 
1091 		/* set flag to initialize port/queue */
1092 		port->need_reconfig = 1;
1093 		port->need_reconfig_queues = 1;
1094 	}
1095 
1096 	/*
1097 	 * Create pools of mbuf.
1098 	 * If NUMA support is disabled, create a single pool of mbuf in
1099 	 * socket 0 memory by default.
1100 	 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1101 	 *
1102 	 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1103 	 * nb_txd can be configured at run time.
1104 	 */
1105 	if (param_total_num_mbufs)
1106 		nb_mbuf_per_pool = param_total_num_mbufs;
1107 	else {
1108 		nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1109 			(nb_lcores * mb_mempool_cache) +
1110 			RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1111 		nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1112 	}
1113 
1114 	if (numa_support) {
1115 		uint8_t i;
1116 
1117 		for (i = 0; i < num_sockets; i++)
1118 			mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool,
1119 					 socket_ids[i]);
1120 	} else {
1121 		if (socket_num == UMA_NO_CONFIG)
1122 			mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool, 0);
1123 		else
1124 			mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool,
1125 						 socket_num);
1126 	}
1127 
1128 	init_port_config();
1129 
1130 	gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1131 		DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1132 	/*
1133 	 * Records which Mbuf pool to use by each logical core, if needed.
1134 	 */
1135 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1136 		mbp = mbuf_pool_find(
1137 			rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]));
1138 
1139 		if (mbp == NULL)
1140 			mbp = mbuf_pool_find(0);
1141 		fwd_lcores[lc_id]->mbp = mbp;
1142 		/* initialize GSO context */
1143 		fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1144 		fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1145 		fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1146 		fwd_lcores[lc_id]->gso_ctx.gso_size = ETHER_MAX_LEN -
1147 			ETHER_CRC_LEN;
1148 		fwd_lcores[lc_id]->gso_ctx.flag = 0;
1149 	}
1150 
1151 	/* Configuration of packet forwarding streams. */
1152 	if (init_fwd_streams() < 0)
1153 		rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
1154 
1155 	fwd_config_setup();
1156 
1157 	/* create a gro context for each lcore */
1158 	gro_param.gro_types = RTE_GRO_TCP_IPV4;
1159 	gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1160 	gro_param.max_item_per_flow = MAX_PKT_BURST;
1161 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1162 		gro_param.socket_id = rte_lcore_to_socket_id(
1163 				fwd_lcores_cpuids[lc_id]);
1164 		fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1165 		if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1166 			rte_exit(EXIT_FAILURE,
1167 					"rte_gro_ctx_create() failed\n");
1168 		}
1169 	}
1170 
1171 #if defined RTE_LIBRTE_PMD_SOFTNIC
1172 	if (strcmp(cur_fwd_eng->fwd_mode_name, "softnic") == 0) {
1173 		RTE_ETH_FOREACH_DEV(pid) {
1174 			port = &ports[pid];
1175 			const char *driver = port->dev_info.driver_name;
1176 
1177 			if (strcmp(driver, "net_softnic") == 0)
1178 				port->softport.fwd_lcore_arg = fwd_lcores;
1179 		}
1180 	}
1181 #endif
1182 
1183 }
1184 
1185 
1186 void
1187 reconfig(portid_t new_port_id, unsigned socket_id)
1188 {
1189 	struct rte_port *port;
1190 
1191 	/* Reconfiguration of Ethernet ports. */
1192 	port = &ports[new_port_id];
1193 	rte_eth_dev_info_get(new_port_id, &port->dev_info);
1194 
1195 	/* set flag to initialize port/queue */
1196 	port->need_reconfig = 1;
1197 	port->need_reconfig_queues = 1;
1198 	port->socket_id = socket_id;
1199 
1200 	init_port_config();
1201 }
1202 
1203 
1204 int
1205 init_fwd_streams(void)
1206 {
1207 	portid_t pid;
1208 	struct rte_port *port;
1209 	streamid_t sm_id, nb_fwd_streams_new;
1210 	queueid_t q;
1211 
1212 	/* set socket id according to numa or not */
1213 	RTE_ETH_FOREACH_DEV(pid) {
1214 		port = &ports[pid];
1215 		if (nb_rxq > port->dev_info.max_rx_queues) {
1216 			printf("Fail: nb_rxq(%d) is greater than "
1217 				"max_rx_queues(%d)\n", nb_rxq,
1218 				port->dev_info.max_rx_queues);
1219 			return -1;
1220 		}
1221 		if (nb_txq > port->dev_info.max_tx_queues) {
1222 			printf("Fail: nb_txq(%d) is greater than "
1223 				"max_tx_queues(%d)\n", nb_txq,
1224 				port->dev_info.max_tx_queues);
1225 			return -1;
1226 		}
1227 		if (numa_support) {
1228 			if (port_numa[pid] != NUMA_NO_CONFIG)
1229 				port->socket_id = port_numa[pid];
1230 			else {
1231 				port->socket_id = rte_eth_dev_socket_id(pid);
1232 
1233 				/*
1234 				 * if socket_id is invalid,
1235 				 * set to the first available socket.
1236 				 */
1237 				if (check_socket_id(port->socket_id) < 0)
1238 					port->socket_id = socket_ids[0];
1239 			}
1240 		}
1241 		else {
1242 			if (socket_num == UMA_NO_CONFIG)
1243 				port->socket_id = 0;
1244 			else
1245 				port->socket_id = socket_num;
1246 		}
1247 	}
1248 
1249 	q = RTE_MAX(nb_rxq, nb_txq);
1250 	if (q == 0) {
1251 		printf("Fail: Cannot allocate fwd streams as number of queues is 0\n");
1252 		return -1;
1253 	}
1254 	nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1255 	if (nb_fwd_streams_new == nb_fwd_streams)
1256 		return 0;
1257 	/* clear the old */
1258 	if (fwd_streams != NULL) {
1259 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1260 			if (fwd_streams[sm_id] == NULL)
1261 				continue;
1262 			rte_free(fwd_streams[sm_id]);
1263 			fwd_streams[sm_id] = NULL;
1264 		}
1265 		rte_free(fwd_streams);
1266 		fwd_streams = NULL;
1267 	}
1268 
1269 	/* init new */
1270 	nb_fwd_streams = nb_fwd_streams_new;
1271 	if (nb_fwd_streams) {
1272 		fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1273 			sizeof(struct fwd_stream *) * nb_fwd_streams,
1274 			RTE_CACHE_LINE_SIZE);
1275 		if (fwd_streams == NULL)
1276 			rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1277 				 " (struct fwd_stream *)) failed\n",
1278 				 nb_fwd_streams);
1279 
1280 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1281 			fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1282 				" struct fwd_stream", sizeof(struct fwd_stream),
1283 				RTE_CACHE_LINE_SIZE);
1284 			if (fwd_streams[sm_id] == NULL)
1285 				rte_exit(EXIT_FAILURE, "rte_zmalloc"
1286 					 "(struct fwd_stream) failed\n");
1287 		}
1288 	}
1289 
1290 	return 0;
1291 }
1292 
1293 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1294 static void
1295 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1296 {
1297 	unsigned int total_burst;
1298 	unsigned int nb_burst;
1299 	unsigned int burst_stats[3];
1300 	uint16_t pktnb_stats[3];
1301 	uint16_t nb_pkt;
1302 	int burst_percent[3];
1303 
1304 	/*
1305 	 * First compute the total number of packet bursts and the
1306 	 * two highest numbers of bursts of the same number of packets.
1307 	 */
1308 	total_burst = 0;
1309 	burst_stats[0] = burst_stats[1] = burst_stats[2] = 0;
1310 	pktnb_stats[0] = pktnb_stats[1] = pktnb_stats[2] = 0;
1311 	for (nb_pkt = 0; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1312 		nb_burst = pbs->pkt_burst_spread[nb_pkt];
1313 		if (nb_burst == 0)
1314 			continue;
1315 		total_burst += nb_burst;
1316 		if (nb_burst > burst_stats[0]) {
1317 			burst_stats[1] = burst_stats[0];
1318 			pktnb_stats[1] = pktnb_stats[0];
1319 			burst_stats[0] = nb_burst;
1320 			pktnb_stats[0] = nb_pkt;
1321 		} else if (nb_burst > burst_stats[1]) {
1322 			burst_stats[1] = nb_burst;
1323 			pktnb_stats[1] = nb_pkt;
1324 		}
1325 	}
1326 	if (total_burst == 0)
1327 		return;
1328 	burst_percent[0] = (burst_stats[0] * 100) / total_burst;
1329 	printf("  %s-bursts : %u [%d%% of %d pkts", rx_tx, total_burst,
1330 	       burst_percent[0], (int) pktnb_stats[0]);
1331 	if (burst_stats[0] == total_burst) {
1332 		printf("]\n");
1333 		return;
1334 	}
1335 	if (burst_stats[0] + burst_stats[1] == total_burst) {
1336 		printf(" + %d%% of %d pkts]\n",
1337 		       100 - burst_percent[0], pktnb_stats[1]);
1338 		return;
1339 	}
1340 	burst_percent[1] = (burst_stats[1] * 100) / total_burst;
1341 	burst_percent[2] = 100 - (burst_percent[0] + burst_percent[1]);
1342 	if ((burst_percent[1] == 0) || (burst_percent[2] == 0)) {
1343 		printf(" + %d%% of others]\n", 100 - burst_percent[0]);
1344 		return;
1345 	}
1346 	printf(" + %d%% of %d pkts + %d%% of others]\n",
1347 	       burst_percent[1], (int) pktnb_stats[1], burst_percent[2]);
1348 }
1349 #endif /* RTE_TEST_PMD_RECORD_BURST_STATS */
1350 
1351 static void
1352 fwd_port_stats_display(portid_t port_id, struct rte_eth_stats *stats)
1353 {
1354 	struct rte_port *port;
1355 	uint8_t i;
1356 
1357 	static const char *fwd_stats_border = "----------------------";
1358 
1359 	port = &ports[port_id];
1360 	printf("\n  %s Forward statistics for port %-2d %s\n",
1361 	       fwd_stats_border, port_id, fwd_stats_border);
1362 
1363 	if ((!port->rx_queue_stats_mapping_enabled) && (!port->tx_queue_stats_mapping_enabled)) {
1364 		printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1365 		       "%-"PRIu64"\n",
1366 		       stats->ipackets, stats->imissed,
1367 		       (uint64_t) (stats->ipackets + stats->imissed));
1368 
1369 		if (cur_fwd_eng == &csum_fwd_engine)
1370 			printf("  Bad-ipcsum: %-14"PRIu64" Bad-l4csum: %-14"PRIu64"Bad-outer-l4csum: %-14"PRIu64"\n",
1371 			       port->rx_bad_ip_csum, port->rx_bad_l4_csum,
1372 			       port->rx_bad_outer_l4_csum);
1373 		if ((stats->ierrors + stats->rx_nombuf) > 0) {
1374 			printf("  RX-error: %-"PRIu64"\n",  stats->ierrors);
1375 			printf("  RX-nombufs: %-14"PRIu64"\n", stats->rx_nombuf);
1376 		}
1377 
1378 		printf("  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1379 		       "%-"PRIu64"\n",
1380 		       stats->opackets, port->tx_dropped,
1381 		       (uint64_t) (stats->opackets + port->tx_dropped));
1382 	}
1383 	else {
1384 		printf("  RX-packets:             %14"PRIu64"    RX-dropped:%14"PRIu64"    RX-total:"
1385 		       "%14"PRIu64"\n",
1386 		       stats->ipackets, stats->imissed,
1387 		       (uint64_t) (stats->ipackets + stats->imissed));
1388 
1389 		if (cur_fwd_eng == &csum_fwd_engine)
1390 			printf("  Bad-ipcsum:%14"PRIu64"    Bad-l4csum:%14"PRIu64"    Bad-outer-l4csum: %-14"PRIu64"\n",
1391 			       port->rx_bad_ip_csum, port->rx_bad_l4_csum,
1392 			       port->rx_bad_outer_l4_csum);
1393 		if ((stats->ierrors + stats->rx_nombuf) > 0) {
1394 			printf("  RX-error:%"PRIu64"\n", stats->ierrors);
1395 			printf("  RX-nombufs:             %14"PRIu64"\n",
1396 			       stats->rx_nombuf);
1397 		}
1398 
1399 		printf("  TX-packets:             %14"PRIu64"    TX-dropped:%14"PRIu64"    TX-total:"
1400 		       "%14"PRIu64"\n",
1401 		       stats->opackets, port->tx_dropped,
1402 		       (uint64_t) (stats->opackets + port->tx_dropped));
1403 	}
1404 
1405 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1406 	if (port->rx_stream)
1407 		pkt_burst_stats_display("RX",
1408 			&port->rx_stream->rx_burst_stats);
1409 	if (port->tx_stream)
1410 		pkt_burst_stats_display("TX",
1411 			&port->tx_stream->tx_burst_stats);
1412 #endif
1413 
1414 	if (port->rx_queue_stats_mapping_enabled) {
1415 		printf("\n");
1416 		for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
1417 			printf("  Stats reg %2d RX-packets:%14"PRIu64
1418 			       "     RX-errors:%14"PRIu64
1419 			       "    RX-bytes:%14"PRIu64"\n",
1420 			       i, stats->q_ipackets[i], stats->q_errors[i], stats->q_ibytes[i]);
1421 		}
1422 		printf("\n");
1423 	}
1424 	if (port->tx_queue_stats_mapping_enabled) {
1425 		for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
1426 			printf("  Stats reg %2d TX-packets:%14"PRIu64
1427 			       "                                 TX-bytes:%14"PRIu64"\n",
1428 			       i, stats->q_opackets[i], stats->q_obytes[i]);
1429 		}
1430 	}
1431 
1432 	printf("  %s--------------------------------%s\n",
1433 	       fwd_stats_border, fwd_stats_border);
1434 }
1435 
1436 static void
1437 fwd_stream_stats_display(streamid_t stream_id)
1438 {
1439 	struct fwd_stream *fs;
1440 	static const char *fwd_top_stats_border = "-------";
1441 
1442 	fs = fwd_streams[stream_id];
1443 	if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1444 	    (fs->fwd_dropped == 0))
1445 		return;
1446 	printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1447 	       "TX Port=%2d/Queue=%2d %s\n",
1448 	       fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1449 	       fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1450 	printf("  RX-packets: %-14u TX-packets: %-14u TX-dropped: %-14u",
1451 	       fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1452 
1453 	/* if checksum mode */
1454 	if (cur_fwd_eng == &csum_fwd_engine) {
1455 	       printf("  RX- bad IP checksum: %-14u  Rx- bad L4 checksum: "
1456 			"%-14u Rx- bad outer L4 checksum: %-14u\n",
1457 			fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1458 			fs->rx_bad_outer_l4_csum);
1459 	}
1460 
1461 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1462 	pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1463 	pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1464 #endif
1465 }
1466 
1467 static void
1468 flush_fwd_rx_queues(void)
1469 {
1470 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1471 	portid_t  rxp;
1472 	portid_t port_id;
1473 	queueid_t rxq;
1474 	uint16_t  nb_rx;
1475 	uint16_t  i;
1476 	uint8_t   j;
1477 	uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
1478 	uint64_t timer_period;
1479 
1480 	/* convert to number of cycles */
1481 	timer_period = rte_get_timer_hz(); /* 1 second timeout */
1482 
1483 	for (j = 0; j < 2; j++) {
1484 		for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
1485 			for (rxq = 0; rxq < nb_rxq; rxq++) {
1486 				port_id = fwd_ports_ids[rxp];
1487 				/**
1488 				* testpmd can stuck in the below do while loop
1489 				* if rte_eth_rx_burst() always returns nonzero
1490 				* packets. So timer is added to exit this loop
1491 				* after 1sec timer expiry.
1492 				*/
1493 				prev_tsc = rte_rdtsc();
1494 				do {
1495 					nb_rx = rte_eth_rx_burst(port_id, rxq,
1496 						pkts_burst, MAX_PKT_BURST);
1497 					for (i = 0; i < nb_rx; i++)
1498 						rte_pktmbuf_free(pkts_burst[i]);
1499 
1500 					cur_tsc = rte_rdtsc();
1501 					diff_tsc = cur_tsc - prev_tsc;
1502 					timer_tsc += diff_tsc;
1503 				} while ((nb_rx > 0) &&
1504 					(timer_tsc < timer_period));
1505 				timer_tsc = 0;
1506 			}
1507 		}
1508 		rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
1509 	}
1510 }
1511 
1512 static void
1513 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
1514 {
1515 	struct fwd_stream **fsm;
1516 	streamid_t nb_fs;
1517 	streamid_t sm_id;
1518 #ifdef RTE_LIBRTE_BITRATE
1519 	uint64_t tics_per_1sec;
1520 	uint64_t tics_datum;
1521 	uint64_t tics_current;
1522 	uint16_t i, cnt_ports;
1523 
1524 	cnt_ports = nb_ports;
1525 	tics_datum = rte_rdtsc();
1526 	tics_per_1sec = rte_get_timer_hz();
1527 #endif
1528 	fsm = &fwd_streams[fc->stream_idx];
1529 	nb_fs = fc->stream_nb;
1530 	do {
1531 		for (sm_id = 0; sm_id < nb_fs; sm_id++)
1532 			(*pkt_fwd)(fsm[sm_id]);
1533 #ifdef RTE_LIBRTE_BITRATE
1534 		if (bitrate_enabled != 0 &&
1535 				bitrate_lcore_id == rte_lcore_id()) {
1536 			tics_current = rte_rdtsc();
1537 			if (tics_current - tics_datum >= tics_per_1sec) {
1538 				/* Periodic bitrate calculation */
1539 				for (i = 0; i < cnt_ports; i++)
1540 					rte_stats_bitrate_calc(bitrate_data,
1541 						ports_ids[i]);
1542 				tics_datum = tics_current;
1543 			}
1544 		}
1545 #endif
1546 #ifdef RTE_LIBRTE_LATENCY_STATS
1547 		if (latencystats_enabled != 0 &&
1548 				latencystats_lcore_id == rte_lcore_id())
1549 			rte_latencystats_update();
1550 #endif
1551 
1552 	} while (! fc->stopped);
1553 }
1554 
1555 static int
1556 start_pkt_forward_on_core(void *fwd_arg)
1557 {
1558 	run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
1559 			     cur_fwd_config.fwd_eng->packet_fwd);
1560 	return 0;
1561 }
1562 
1563 /*
1564  * Run the TXONLY packet forwarding engine to send a single burst of packets.
1565  * Used to start communication flows in network loopback test configurations.
1566  */
1567 static int
1568 run_one_txonly_burst_on_core(void *fwd_arg)
1569 {
1570 	struct fwd_lcore *fwd_lc;
1571 	struct fwd_lcore tmp_lcore;
1572 
1573 	fwd_lc = (struct fwd_lcore *) fwd_arg;
1574 	tmp_lcore = *fwd_lc;
1575 	tmp_lcore.stopped = 1;
1576 	run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
1577 	return 0;
1578 }
1579 
1580 /*
1581  * Launch packet forwarding:
1582  *     - Setup per-port forwarding context.
1583  *     - launch logical cores with their forwarding configuration.
1584  */
1585 static void
1586 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
1587 {
1588 	port_fwd_begin_t port_fwd_begin;
1589 	unsigned int i;
1590 	unsigned int lc_id;
1591 	int diag;
1592 
1593 	port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
1594 	if (port_fwd_begin != NULL) {
1595 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1596 			(*port_fwd_begin)(fwd_ports_ids[i]);
1597 	}
1598 	for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
1599 		lc_id = fwd_lcores_cpuids[i];
1600 		if ((interactive == 0) || (lc_id != rte_lcore_id())) {
1601 			fwd_lcores[i]->stopped = 0;
1602 			diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
1603 						     fwd_lcores[i], lc_id);
1604 			if (diag != 0)
1605 				printf("launch lcore %u failed - diag=%d\n",
1606 				       lc_id, diag);
1607 		}
1608 	}
1609 }
1610 
1611 /*
1612  * Update the forward ports list.
1613  */
1614 void
1615 update_fwd_ports(portid_t new_pid)
1616 {
1617 	unsigned int i;
1618 	unsigned int new_nb_fwd_ports = 0;
1619 	int move = 0;
1620 
1621 	for (i = 0; i < nb_fwd_ports; ++i) {
1622 		if (port_id_is_invalid(fwd_ports_ids[i], DISABLED_WARN))
1623 			move = 1;
1624 		else if (move)
1625 			fwd_ports_ids[new_nb_fwd_ports++] = fwd_ports_ids[i];
1626 		else
1627 			new_nb_fwd_ports++;
1628 	}
1629 	if (new_pid < RTE_MAX_ETHPORTS)
1630 		fwd_ports_ids[new_nb_fwd_ports++] = new_pid;
1631 
1632 	nb_fwd_ports = new_nb_fwd_ports;
1633 	nb_cfg_ports = new_nb_fwd_ports;
1634 }
1635 
1636 /*
1637  * Launch packet forwarding configuration.
1638  */
1639 void
1640 start_packet_forwarding(int with_tx_first)
1641 {
1642 	port_fwd_begin_t port_fwd_begin;
1643 	port_fwd_end_t  port_fwd_end;
1644 	struct rte_port *port;
1645 	unsigned int i;
1646 	portid_t   pt_id;
1647 	streamid_t sm_id;
1648 
1649 	if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
1650 		rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
1651 
1652 	if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
1653 		rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
1654 
1655 	if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
1656 		strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
1657 		(!nb_rxq || !nb_txq))
1658 		rte_exit(EXIT_FAILURE,
1659 			"Either rxq or txq are 0, cannot use %s fwd mode\n",
1660 			cur_fwd_eng->fwd_mode_name);
1661 
1662 	if (all_ports_started() == 0) {
1663 		printf("Not all ports were started\n");
1664 		return;
1665 	}
1666 	if (test_done == 0) {
1667 		printf("Packet forwarding already started\n");
1668 		return;
1669 	}
1670 
1671 
1672 	if(dcb_test) {
1673 		for (i = 0; i < nb_fwd_ports; i++) {
1674 			pt_id = fwd_ports_ids[i];
1675 			port = &ports[pt_id];
1676 			if (!port->dcb_flag) {
1677 				printf("In DCB mode, all forwarding ports must "
1678                                        "be configured in this mode.\n");
1679 				return;
1680 			}
1681 		}
1682 		if (nb_fwd_lcores == 1) {
1683 			printf("In DCB mode,the nb forwarding cores "
1684                                "should be larger than 1.\n");
1685 			return;
1686 		}
1687 	}
1688 	test_done = 0;
1689 
1690 	fwd_config_setup();
1691 
1692 	if(!no_flush_rx)
1693 		flush_fwd_rx_queues();
1694 
1695 	pkt_fwd_config_display(&cur_fwd_config);
1696 	rxtx_config_display();
1697 
1698 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1699 		pt_id = fwd_ports_ids[i];
1700 		port = &ports[pt_id];
1701 		rte_eth_stats_get(pt_id, &port->stats);
1702 		port->tx_dropped = 0;
1703 
1704 		map_port_queue_stats_mapping_registers(pt_id, port);
1705 	}
1706 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1707 		fwd_streams[sm_id]->rx_packets = 0;
1708 		fwd_streams[sm_id]->tx_packets = 0;
1709 		fwd_streams[sm_id]->fwd_dropped = 0;
1710 		fwd_streams[sm_id]->rx_bad_ip_csum = 0;
1711 		fwd_streams[sm_id]->rx_bad_l4_csum = 0;
1712 		fwd_streams[sm_id]->rx_bad_outer_l4_csum = 0;
1713 
1714 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1715 		memset(&fwd_streams[sm_id]->rx_burst_stats, 0,
1716 		       sizeof(fwd_streams[sm_id]->rx_burst_stats));
1717 		memset(&fwd_streams[sm_id]->tx_burst_stats, 0,
1718 		       sizeof(fwd_streams[sm_id]->tx_burst_stats));
1719 #endif
1720 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1721 		fwd_streams[sm_id]->core_cycles = 0;
1722 #endif
1723 	}
1724 	if (with_tx_first) {
1725 		port_fwd_begin = tx_only_engine.port_fwd_begin;
1726 		if (port_fwd_begin != NULL) {
1727 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1728 				(*port_fwd_begin)(fwd_ports_ids[i]);
1729 		}
1730 		while (with_tx_first--) {
1731 			launch_packet_forwarding(
1732 					run_one_txonly_burst_on_core);
1733 			rte_eal_mp_wait_lcore();
1734 		}
1735 		port_fwd_end = tx_only_engine.port_fwd_end;
1736 		if (port_fwd_end != NULL) {
1737 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1738 				(*port_fwd_end)(fwd_ports_ids[i]);
1739 		}
1740 	}
1741 	launch_packet_forwarding(start_pkt_forward_on_core);
1742 }
1743 
1744 void
1745 stop_packet_forwarding(void)
1746 {
1747 	struct rte_eth_stats stats;
1748 	struct rte_port *port;
1749 	port_fwd_end_t  port_fwd_end;
1750 	int i;
1751 	portid_t   pt_id;
1752 	streamid_t sm_id;
1753 	lcoreid_t  lc_id;
1754 	uint64_t total_recv;
1755 	uint64_t total_xmit;
1756 	uint64_t total_rx_dropped;
1757 	uint64_t total_tx_dropped;
1758 	uint64_t total_rx_nombuf;
1759 	uint64_t tx_dropped;
1760 	uint64_t rx_bad_ip_csum;
1761 	uint64_t rx_bad_l4_csum;
1762 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1763 	uint64_t fwd_cycles;
1764 #endif
1765 
1766 	static const char *acc_stats_border = "+++++++++++++++";
1767 
1768 	if (test_done) {
1769 		printf("Packet forwarding not started\n");
1770 		return;
1771 	}
1772 	printf("Telling cores to stop...");
1773 	for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
1774 		fwd_lcores[lc_id]->stopped = 1;
1775 	printf("\nWaiting for lcores to finish...\n");
1776 	rte_eal_mp_wait_lcore();
1777 	port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
1778 	if (port_fwd_end != NULL) {
1779 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1780 			pt_id = fwd_ports_ids[i];
1781 			(*port_fwd_end)(pt_id);
1782 		}
1783 	}
1784 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1785 	fwd_cycles = 0;
1786 #endif
1787 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1788 		if (cur_fwd_config.nb_fwd_streams >
1789 		    cur_fwd_config.nb_fwd_ports) {
1790 			fwd_stream_stats_display(sm_id);
1791 			ports[fwd_streams[sm_id]->tx_port].tx_stream = NULL;
1792 			ports[fwd_streams[sm_id]->rx_port].rx_stream = NULL;
1793 		} else {
1794 			ports[fwd_streams[sm_id]->tx_port].tx_stream =
1795 				fwd_streams[sm_id];
1796 			ports[fwd_streams[sm_id]->rx_port].rx_stream =
1797 				fwd_streams[sm_id];
1798 		}
1799 		tx_dropped = ports[fwd_streams[sm_id]->tx_port].tx_dropped;
1800 		tx_dropped = (uint64_t) (tx_dropped +
1801 					 fwd_streams[sm_id]->fwd_dropped);
1802 		ports[fwd_streams[sm_id]->tx_port].tx_dropped = tx_dropped;
1803 
1804 		rx_bad_ip_csum =
1805 			ports[fwd_streams[sm_id]->rx_port].rx_bad_ip_csum;
1806 		rx_bad_ip_csum = (uint64_t) (rx_bad_ip_csum +
1807 					 fwd_streams[sm_id]->rx_bad_ip_csum);
1808 		ports[fwd_streams[sm_id]->rx_port].rx_bad_ip_csum =
1809 							rx_bad_ip_csum;
1810 
1811 		rx_bad_l4_csum =
1812 			ports[fwd_streams[sm_id]->rx_port].rx_bad_l4_csum;
1813 		rx_bad_l4_csum = (uint64_t) (rx_bad_l4_csum +
1814 					 fwd_streams[sm_id]->rx_bad_l4_csum);
1815 		ports[fwd_streams[sm_id]->rx_port].rx_bad_l4_csum =
1816 							rx_bad_l4_csum;
1817 
1818 		ports[fwd_streams[sm_id]->rx_port].rx_bad_outer_l4_csum +=
1819 				fwd_streams[sm_id]->rx_bad_outer_l4_csum;
1820 
1821 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1822 		fwd_cycles = (uint64_t) (fwd_cycles +
1823 					 fwd_streams[sm_id]->core_cycles);
1824 #endif
1825 	}
1826 	total_recv = 0;
1827 	total_xmit = 0;
1828 	total_rx_dropped = 0;
1829 	total_tx_dropped = 0;
1830 	total_rx_nombuf  = 0;
1831 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1832 		pt_id = fwd_ports_ids[i];
1833 
1834 		port = &ports[pt_id];
1835 		rte_eth_stats_get(pt_id, &stats);
1836 		stats.ipackets -= port->stats.ipackets;
1837 		port->stats.ipackets = 0;
1838 		stats.opackets -= port->stats.opackets;
1839 		port->stats.opackets = 0;
1840 		stats.ibytes   -= port->stats.ibytes;
1841 		port->stats.ibytes = 0;
1842 		stats.obytes   -= port->stats.obytes;
1843 		port->stats.obytes = 0;
1844 		stats.imissed  -= port->stats.imissed;
1845 		port->stats.imissed = 0;
1846 		stats.oerrors  -= port->stats.oerrors;
1847 		port->stats.oerrors = 0;
1848 		stats.rx_nombuf -= port->stats.rx_nombuf;
1849 		port->stats.rx_nombuf = 0;
1850 
1851 		total_recv += stats.ipackets;
1852 		total_xmit += stats.opackets;
1853 		total_rx_dropped += stats.imissed;
1854 		total_tx_dropped += port->tx_dropped;
1855 		total_rx_nombuf  += stats.rx_nombuf;
1856 
1857 		fwd_port_stats_display(pt_id, &stats);
1858 	}
1859 
1860 	printf("\n  %s Accumulated forward statistics for all ports"
1861 	       "%s\n",
1862 	       acc_stats_border, acc_stats_border);
1863 	printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1864 	       "%-"PRIu64"\n"
1865 	       "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1866 	       "%-"PRIu64"\n",
1867 	       total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1868 	       total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1869 	if (total_rx_nombuf > 0)
1870 		printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1871 	printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
1872 	       "%s\n",
1873 	       acc_stats_border, acc_stats_border);
1874 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1875 	if (total_recv > 0)
1876 		printf("\n  CPU cycles/packet=%u (total cycles="
1877 		       "%"PRIu64" / total RX packets=%"PRIu64")\n",
1878 		       (unsigned int)(fwd_cycles / total_recv),
1879 		       fwd_cycles, total_recv);
1880 #endif
1881 	printf("\nDone.\n");
1882 	test_done = 1;
1883 }
1884 
1885 void
1886 dev_set_link_up(portid_t pid)
1887 {
1888 	if (rte_eth_dev_set_link_up(pid) < 0)
1889 		printf("\nSet link up fail.\n");
1890 }
1891 
1892 void
1893 dev_set_link_down(portid_t pid)
1894 {
1895 	if (rte_eth_dev_set_link_down(pid) < 0)
1896 		printf("\nSet link down fail.\n");
1897 }
1898 
1899 static int
1900 all_ports_started(void)
1901 {
1902 	portid_t pi;
1903 	struct rte_port *port;
1904 
1905 	RTE_ETH_FOREACH_DEV(pi) {
1906 		port = &ports[pi];
1907 		/* Check if there is a port which is not started */
1908 		if ((port->port_status != RTE_PORT_STARTED) &&
1909 			(port->slave_flag == 0))
1910 			return 0;
1911 	}
1912 
1913 	/* No port is not started */
1914 	return 1;
1915 }
1916 
1917 int
1918 port_is_stopped(portid_t port_id)
1919 {
1920 	struct rte_port *port = &ports[port_id];
1921 
1922 	if ((port->port_status != RTE_PORT_STOPPED) &&
1923 	    (port->slave_flag == 0))
1924 		return 0;
1925 	return 1;
1926 }
1927 
1928 int
1929 all_ports_stopped(void)
1930 {
1931 	portid_t pi;
1932 
1933 	RTE_ETH_FOREACH_DEV(pi) {
1934 		if (!port_is_stopped(pi))
1935 			return 0;
1936 	}
1937 
1938 	return 1;
1939 }
1940 
1941 int
1942 port_is_started(portid_t port_id)
1943 {
1944 	if (port_id_is_invalid(port_id, ENABLED_WARN))
1945 		return 0;
1946 
1947 	if (ports[port_id].port_status != RTE_PORT_STARTED)
1948 		return 0;
1949 
1950 	return 1;
1951 }
1952 
1953 int
1954 start_port(portid_t pid)
1955 {
1956 	int diag, need_check_link_status = -1;
1957 	portid_t pi;
1958 	queueid_t qi;
1959 	struct rte_port *port;
1960 	struct ether_addr mac_addr;
1961 	enum rte_eth_event_type event_type;
1962 
1963 	if (port_id_is_invalid(pid, ENABLED_WARN))
1964 		return 0;
1965 
1966 	if(dcb_config)
1967 		dcb_test = 1;
1968 	RTE_ETH_FOREACH_DEV(pi) {
1969 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
1970 			continue;
1971 
1972 		need_check_link_status = 0;
1973 		port = &ports[pi];
1974 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
1975 						 RTE_PORT_HANDLING) == 0) {
1976 			printf("Port %d is now not stopped\n", pi);
1977 			continue;
1978 		}
1979 
1980 		if (port->need_reconfig > 0) {
1981 			port->need_reconfig = 0;
1982 
1983 			if (flow_isolate_all) {
1984 				int ret = port_flow_isolate(pi, 1);
1985 				if (ret) {
1986 					printf("Failed to apply isolated"
1987 					       " mode on port %d\n", pi);
1988 					return -1;
1989 				}
1990 			}
1991 			configure_rxtx_dump_callbacks(0);
1992 			printf("Configuring Port %d (socket %u)\n", pi,
1993 					port->socket_id);
1994 			/* configure port */
1995 			diag = rte_eth_dev_configure(pi, nb_rxq, nb_txq,
1996 						&(port->dev_conf));
1997 			if (diag != 0) {
1998 				if (rte_atomic16_cmpset(&(port->port_status),
1999 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2000 					printf("Port %d can not be set back "
2001 							"to stopped\n", pi);
2002 				printf("Fail to configure port %d\n", pi);
2003 				/* try to reconfigure port next time */
2004 				port->need_reconfig = 1;
2005 				return -1;
2006 			}
2007 		}
2008 		if (port->need_reconfig_queues > 0) {
2009 			port->need_reconfig_queues = 0;
2010 			/* setup tx queues */
2011 			for (qi = 0; qi < nb_txq; qi++) {
2012 				if ((numa_support) &&
2013 					(txring_numa[pi] != NUMA_NO_CONFIG))
2014 					diag = rte_eth_tx_queue_setup(pi, qi,
2015 						port->nb_tx_desc[qi],
2016 						txring_numa[pi],
2017 						&(port->tx_conf[qi]));
2018 				else
2019 					diag = rte_eth_tx_queue_setup(pi, qi,
2020 						port->nb_tx_desc[qi],
2021 						port->socket_id,
2022 						&(port->tx_conf[qi]));
2023 
2024 				if (diag == 0)
2025 					continue;
2026 
2027 				/* Fail to setup tx queue, return */
2028 				if (rte_atomic16_cmpset(&(port->port_status),
2029 							RTE_PORT_HANDLING,
2030 							RTE_PORT_STOPPED) == 0)
2031 					printf("Port %d can not be set back "
2032 							"to stopped\n", pi);
2033 				printf("Fail to configure port %d tx queues\n",
2034 				       pi);
2035 				/* try to reconfigure queues next time */
2036 				port->need_reconfig_queues = 1;
2037 				return -1;
2038 			}
2039 			for (qi = 0; qi < nb_rxq; qi++) {
2040 				/* setup rx queues */
2041 				if ((numa_support) &&
2042 					(rxring_numa[pi] != NUMA_NO_CONFIG)) {
2043 					struct rte_mempool * mp =
2044 						mbuf_pool_find(rxring_numa[pi]);
2045 					if (mp == NULL) {
2046 						printf("Failed to setup RX queue:"
2047 							"No mempool allocation"
2048 							" on the socket %d\n",
2049 							rxring_numa[pi]);
2050 						return -1;
2051 					}
2052 
2053 					diag = rte_eth_rx_queue_setup(pi, qi,
2054 					     port->nb_rx_desc[qi],
2055 					     rxring_numa[pi],
2056 					     &(port->rx_conf[qi]),
2057 					     mp);
2058 				} else {
2059 					struct rte_mempool *mp =
2060 						mbuf_pool_find(port->socket_id);
2061 					if (mp == NULL) {
2062 						printf("Failed to setup RX queue:"
2063 							"No mempool allocation"
2064 							" on the socket %d\n",
2065 							port->socket_id);
2066 						return -1;
2067 					}
2068 					diag = rte_eth_rx_queue_setup(pi, qi,
2069 					     port->nb_rx_desc[qi],
2070 					     port->socket_id,
2071 					     &(port->rx_conf[qi]),
2072 					     mp);
2073 				}
2074 				if (diag == 0)
2075 					continue;
2076 
2077 				/* Fail to setup rx queue, return */
2078 				if (rte_atomic16_cmpset(&(port->port_status),
2079 							RTE_PORT_HANDLING,
2080 							RTE_PORT_STOPPED) == 0)
2081 					printf("Port %d can not be set back "
2082 							"to stopped\n", pi);
2083 				printf("Fail to configure port %d rx queues\n",
2084 				       pi);
2085 				/* try to reconfigure queues next time */
2086 				port->need_reconfig_queues = 1;
2087 				return -1;
2088 			}
2089 		}
2090 		configure_rxtx_dump_callbacks(verbose_level);
2091 		/* start port */
2092 		if (rte_eth_dev_start(pi) < 0) {
2093 			printf("Fail to start port %d\n", pi);
2094 
2095 			/* Fail to setup rx queue, return */
2096 			if (rte_atomic16_cmpset(&(port->port_status),
2097 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2098 				printf("Port %d can not be set back to "
2099 							"stopped\n", pi);
2100 			continue;
2101 		}
2102 
2103 		if (rte_atomic16_cmpset(&(port->port_status),
2104 			RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2105 			printf("Port %d can not be set into started\n", pi);
2106 
2107 		rte_eth_macaddr_get(pi, &mac_addr);
2108 		printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2109 				mac_addr.addr_bytes[0], mac_addr.addr_bytes[1],
2110 				mac_addr.addr_bytes[2], mac_addr.addr_bytes[3],
2111 				mac_addr.addr_bytes[4], mac_addr.addr_bytes[5]);
2112 
2113 		/* at least one port started, need checking link status */
2114 		need_check_link_status = 1;
2115 	}
2116 
2117 	for (event_type = RTE_ETH_EVENT_UNKNOWN;
2118 	     event_type < RTE_ETH_EVENT_MAX;
2119 	     event_type++) {
2120 		diag = rte_eth_dev_callback_register(RTE_ETH_ALL,
2121 						event_type,
2122 						eth_event_callback,
2123 						NULL);
2124 		if (diag) {
2125 			printf("Failed to setup even callback for event %d\n",
2126 				event_type);
2127 			return -1;
2128 		}
2129 	}
2130 
2131 	if (need_check_link_status == 1 && !no_link_check)
2132 		check_all_ports_link_status(RTE_PORT_ALL);
2133 	else if (need_check_link_status == 0)
2134 		printf("Please stop the ports first\n");
2135 
2136 	printf("Done\n");
2137 	return 0;
2138 }
2139 
2140 void
2141 stop_port(portid_t pid)
2142 {
2143 	portid_t pi;
2144 	struct rte_port *port;
2145 	int need_check_link_status = 0;
2146 
2147 	if (dcb_test) {
2148 		dcb_test = 0;
2149 		dcb_config = 0;
2150 	}
2151 
2152 	if (port_id_is_invalid(pid, ENABLED_WARN))
2153 		return;
2154 
2155 	printf("Stopping ports...\n");
2156 
2157 	RTE_ETH_FOREACH_DEV(pi) {
2158 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2159 			continue;
2160 
2161 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2162 			printf("Please remove port %d from forwarding configuration.\n", pi);
2163 			continue;
2164 		}
2165 
2166 		if (port_is_bonding_slave(pi)) {
2167 			printf("Please remove port %d from bonded device.\n", pi);
2168 			continue;
2169 		}
2170 
2171 		port = &ports[pi];
2172 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2173 						RTE_PORT_HANDLING) == 0)
2174 			continue;
2175 
2176 		rte_eth_dev_stop(pi);
2177 
2178 		if (rte_atomic16_cmpset(&(port->port_status),
2179 			RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2180 			printf("Port %d can not be set into stopped\n", pi);
2181 		need_check_link_status = 1;
2182 	}
2183 	if (need_check_link_status && !no_link_check)
2184 		check_all_ports_link_status(RTE_PORT_ALL);
2185 
2186 	printf("Done\n");
2187 }
2188 
2189 static void
2190 remove_unused_fwd_ports(void)
2191 {
2192 	int i;
2193 	int last_port_idx = nb_ports - 1;
2194 
2195 	for (i = 0; i <= last_port_idx; i++) { /* iterate in ports_ids */
2196 		if (rte_eth_devices[ports_ids[i]].state != RTE_ETH_DEV_UNUSED)
2197 			continue;
2198 		/* skip unused ports at the end */
2199 		while (i <= last_port_idx &&
2200 				rte_eth_devices[ports_ids[last_port_idx]].state
2201 				== RTE_ETH_DEV_UNUSED)
2202 			last_port_idx--;
2203 		if (last_port_idx < i)
2204 			break;
2205 		/* overwrite unused port with last valid port */
2206 		ports_ids[i] = ports_ids[last_port_idx];
2207 		/* decrease ports count */
2208 		last_port_idx--;
2209 	}
2210 	nb_ports = rte_eth_dev_count_avail();
2211 	update_fwd_ports(RTE_MAX_ETHPORTS);
2212 }
2213 
2214 void
2215 close_port(portid_t pid)
2216 {
2217 	portid_t pi;
2218 	struct rte_port *port;
2219 
2220 	if (port_id_is_invalid(pid, ENABLED_WARN))
2221 		return;
2222 
2223 	printf("Closing ports...\n");
2224 
2225 	RTE_ETH_FOREACH_DEV(pi) {
2226 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2227 			continue;
2228 
2229 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2230 			printf("Please remove port %d from forwarding configuration.\n", pi);
2231 			continue;
2232 		}
2233 
2234 		if (port_is_bonding_slave(pi)) {
2235 			printf("Please remove port %d from bonded device.\n", pi);
2236 			continue;
2237 		}
2238 
2239 		port = &ports[pi];
2240 		if (rte_atomic16_cmpset(&(port->port_status),
2241 			RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2242 			printf("Port %d is already closed\n", pi);
2243 			continue;
2244 		}
2245 
2246 		if (rte_atomic16_cmpset(&(port->port_status),
2247 			RTE_PORT_STOPPED, RTE_PORT_HANDLING) == 0) {
2248 			printf("Port %d is now not stopped\n", pi);
2249 			continue;
2250 		}
2251 
2252 		if (port->flow_list)
2253 			port_flow_flush(pi);
2254 		rte_eth_dev_close(pi);
2255 
2256 		remove_unused_fwd_ports();
2257 
2258 		if (rte_atomic16_cmpset(&(port->port_status),
2259 			RTE_PORT_HANDLING, RTE_PORT_CLOSED) == 0)
2260 			printf("Port %d cannot be set to closed\n", pi);
2261 	}
2262 
2263 	printf("Done\n");
2264 }
2265 
2266 void
2267 reset_port(portid_t pid)
2268 {
2269 	int diag;
2270 	portid_t pi;
2271 	struct rte_port *port;
2272 
2273 	if (port_id_is_invalid(pid, ENABLED_WARN))
2274 		return;
2275 
2276 	printf("Resetting ports...\n");
2277 
2278 	RTE_ETH_FOREACH_DEV(pi) {
2279 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2280 			continue;
2281 
2282 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2283 			printf("Please remove port %d from forwarding "
2284 			       "configuration.\n", pi);
2285 			continue;
2286 		}
2287 
2288 		if (port_is_bonding_slave(pi)) {
2289 			printf("Please remove port %d from bonded device.\n",
2290 			       pi);
2291 			continue;
2292 		}
2293 
2294 		diag = rte_eth_dev_reset(pi);
2295 		if (diag == 0) {
2296 			port = &ports[pi];
2297 			port->need_reconfig = 1;
2298 			port->need_reconfig_queues = 1;
2299 		} else {
2300 			printf("Failed to reset port %d. diag=%d\n", pi, diag);
2301 		}
2302 	}
2303 
2304 	printf("Done\n");
2305 }
2306 
2307 void
2308 attach_port(char *identifier)
2309 {
2310 	portid_t pi = 0;
2311 	unsigned int socket_id;
2312 
2313 	printf("Attaching a new port...\n");
2314 
2315 	if (identifier == NULL) {
2316 		printf("Invalid parameters are specified\n");
2317 		return;
2318 	}
2319 
2320 	if (rte_eth_dev_attach(identifier, &pi))
2321 		return;
2322 
2323 	socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2324 	/* if socket_id is invalid, set to the first available socket. */
2325 	if (check_socket_id(socket_id) < 0)
2326 		socket_id = socket_ids[0];
2327 	reconfig(pi, socket_id);
2328 	rte_eth_promiscuous_enable(pi);
2329 
2330 	ports_ids[nb_ports] = pi;
2331 	nb_ports = rte_eth_dev_count_avail();
2332 
2333 	ports[pi].port_status = RTE_PORT_STOPPED;
2334 
2335 	update_fwd_ports(pi);
2336 
2337 	printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2338 	printf("Done\n");
2339 }
2340 
2341 void
2342 detach_port(portid_t port_id)
2343 {
2344 	char name[RTE_ETH_NAME_MAX_LEN];
2345 
2346 	printf("Detaching a port...\n");
2347 
2348 	if (ports[port_id].port_status != RTE_PORT_CLOSED) {
2349 		if (ports[port_id].port_status != RTE_PORT_STOPPED) {
2350 			printf("Port not stopped\n");
2351 			return;
2352 		}
2353 		printf("Port was not closed\n");
2354 		if (ports[port_id].flow_list)
2355 			port_flow_flush(port_id);
2356 	}
2357 
2358 	if (rte_eth_dev_detach(port_id, name)) {
2359 		TESTPMD_LOG(ERR, "Failed to detach port %u\n", port_id);
2360 		return;
2361 	}
2362 
2363 	remove_unused_fwd_ports();
2364 
2365 	printf("Port %u is detached. Now total ports is %d\n",
2366 			port_id, nb_ports);
2367 	printf("Done\n");
2368 	return;
2369 }
2370 
2371 void
2372 pmd_test_exit(void)
2373 {
2374 	struct rte_device *device;
2375 	portid_t pt_id;
2376 	int ret;
2377 
2378 	if (test_done == 0)
2379 		stop_packet_forwarding();
2380 
2381 	if (ports != NULL) {
2382 		no_link_check = 1;
2383 		RTE_ETH_FOREACH_DEV(pt_id) {
2384 			printf("\nShutting down port %d...\n", pt_id);
2385 			fflush(stdout);
2386 			stop_port(pt_id);
2387 			close_port(pt_id);
2388 
2389 			/*
2390 			 * This is a workaround to fix a virtio-user issue that
2391 			 * requires to call clean-up routine to remove existing
2392 			 * socket.
2393 			 * This workaround valid only for testpmd, needs a fix
2394 			 * valid for all applications.
2395 			 * TODO: Implement proper resource cleanup
2396 			 */
2397 			device = rte_eth_devices[pt_id].device;
2398 			if (device && !strcmp(device->driver->name, "net_virtio_user"))
2399 				detach_port(pt_id);
2400 		}
2401 	}
2402 
2403 	if (hot_plug) {
2404 		ret = rte_dev_event_monitor_stop();
2405 		if (ret) {
2406 			RTE_LOG(ERR, EAL,
2407 				"fail to stop device event monitor.");
2408 			return;
2409 		}
2410 
2411 		ret = rte_dev_event_callback_unregister(NULL,
2412 			eth_dev_event_callback, NULL);
2413 		if (ret < 0) {
2414 			RTE_LOG(ERR, EAL,
2415 				"fail to unregister device event callback.\n");
2416 			return;
2417 		}
2418 
2419 		ret = rte_dev_hotplug_handle_disable();
2420 		if (ret) {
2421 			RTE_LOG(ERR, EAL,
2422 				"fail to disable hotplug handling.\n");
2423 			return;
2424 		}
2425 	}
2426 
2427 	printf("\nBye...\n");
2428 }
2429 
2430 typedef void (*cmd_func_t)(void);
2431 struct pmd_test_command {
2432 	const char *cmd_name;
2433 	cmd_func_t cmd_func;
2434 };
2435 
2436 #define PMD_TEST_CMD_NB (sizeof(pmd_test_menu) / sizeof(pmd_test_menu[0]))
2437 
2438 /* Check the link status of all ports in up to 9s, and print them finally */
2439 static void
2440 check_all_ports_link_status(uint32_t port_mask)
2441 {
2442 #define CHECK_INTERVAL 100 /* 100ms */
2443 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2444 	portid_t portid;
2445 	uint8_t count, all_ports_up, print_flag = 0;
2446 	struct rte_eth_link link;
2447 
2448 	printf("Checking link statuses...\n");
2449 	fflush(stdout);
2450 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
2451 		all_ports_up = 1;
2452 		RTE_ETH_FOREACH_DEV(portid) {
2453 			if ((port_mask & (1 << portid)) == 0)
2454 				continue;
2455 			memset(&link, 0, sizeof(link));
2456 			rte_eth_link_get_nowait(portid, &link);
2457 			/* print link status if flag set */
2458 			if (print_flag == 1) {
2459 				if (link.link_status)
2460 					printf(
2461 					"Port%d Link Up. speed %u Mbps- %s\n",
2462 					portid, link.link_speed,
2463 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2464 					("full-duplex") : ("half-duplex\n"));
2465 				else
2466 					printf("Port %d Link Down\n", portid);
2467 				continue;
2468 			}
2469 			/* clear all_ports_up flag if any link down */
2470 			if (link.link_status == ETH_LINK_DOWN) {
2471 				all_ports_up = 0;
2472 				break;
2473 			}
2474 		}
2475 		/* after finally printing all link status, get out */
2476 		if (print_flag == 1)
2477 			break;
2478 
2479 		if (all_ports_up == 0) {
2480 			fflush(stdout);
2481 			rte_delay_ms(CHECK_INTERVAL);
2482 		}
2483 
2484 		/* set the print_flag if all ports up or timeout */
2485 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2486 			print_flag = 1;
2487 		}
2488 
2489 		if (lsc_interrupt)
2490 			break;
2491 	}
2492 }
2493 
2494 static void
2495 rmv_event_callback(void *arg)
2496 {
2497 	int need_to_start = 0;
2498 	int org_no_link_check = no_link_check;
2499 	portid_t port_id = (intptr_t)arg;
2500 
2501 	RTE_ETH_VALID_PORTID_OR_RET(port_id);
2502 
2503 	if (!test_done && port_is_forwarding(port_id)) {
2504 		need_to_start = 1;
2505 		stop_packet_forwarding();
2506 	}
2507 	no_link_check = 1;
2508 	stop_port(port_id);
2509 	no_link_check = org_no_link_check;
2510 	close_port(port_id);
2511 	detach_port(port_id);
2512 	if (need_to_start)
2513 		start_packet_forwarding(0);
2514 }
2515 
2516 /* This function is used by the interrupt thread */
2517 static int
2518 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
2519 		  void *ret_param)
2520 {
2521 	static const char * const event_desc[] = {
2522 		[RTE_ETH_EVENT_UNKNOWN] = "Unknown",
2523 		[RTE_ETH_EVENT_INTR_LSC] = "LSC",
2524 		[RTE_ETH_EVENT_QUEUE_STATE] = "Queue state",
2525 		[RTE_ETH_EVENT_INTR_RESET] = "Interrupt reset",
2526 		[RTE_ETH_EVENT_VF_MBOX] = "VF Mbox",
2527 		[RTE_ETH_EVENT_IPSEC] = "IPsec",
2528 		[RTE_ETH_EVENT_MACSEC] = "MACsec",
2529 		[RTE_ETH_EVENT_INTR_RMV] = "device removal",
2530 		[RTE_ETH_EVENT_NEW] = "device probed",
2531 		[RTE_ETH_EVENT_DESTROY] = "device released",
2532 		[RTE_ETH_EVENT_MAX] = NULL,
2533 	};
2534 
2535 	RTE_SET_USED(param);
2536 	RTE_SET_USED(ret_param);
2537 
2538 	if (type >= RTE_ETH_EVENT_MAX) {
2539 		fprintf(stderr, "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
2540 			port_id, __func__, type);
2541 		fflush(stderr);
2542 	} else if (event_print_mask & (UINT32_C(1) << type)) {
2543 		printf("\nPort %" PRIu16 ": %s event\n", port_id,
2544 			event_desc[type]);
2545 		fflush(stdout);
2546 	}
2547 
2548 	if (port_id_is_invalid(port_id, DISABLED_WARN))
2549 		return 0;
2550 
2551 	switch (type) {
2552 	case RTE_ETH_EVENT_INTR_RMV:
2553 		if (rte_eal_alarm_set(100000,
2554 				rmv_event_callback, (void *)(intptr_t)port_id))
2555 			fprintf(stderr, "Could not set up deferred device removal\n");
2556 		break;
2557 	default:
2558 		break;
2559 	}
2560 	return 0;
2561 }
2562 
2563 /* This function is used by the interrupt thread */
2564 static void
2565 eth_dev_event_callback(const char *device_name, enum rte_dev_event_type type,
2566 			     __rte_unused void *arg)
2567 {
2568 	uint16_t port_id;
2569 	int ret;
2570 
2571 	if (type >= RTE_DEV_EVENT_MAX) {
2572 		fprintf(stderr, "%s called upon invalid event %d\n",
2573 			__func__, type);
2574 		fflush(stderr);
2575 	}
2576 
2577 	switch (type) {
2578 	case RTE_DEV_EVENT_REMOVE:
2579 		RTE_LOG(ERR, EAL, "The device: %s has been removed!\n",
2580 			device_name);
2581 		ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
2582 		if (ret) {
2583 			RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
2584 				device_name);
2585 			return;
2586 		}
2587 		rmv_event_callback((void *)(intptr_t)port_id);
2588 		break;
2589 	case RTE_DEV_EVENT_ADD:
2590 		RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
2591 			device_name);
2592 		/* TODO: After finish kernel driver binding,
2593 		 * begin to attach port.
2594 		 */
2595 		break;
2596 	default:
2597 		break;
2598 	}
2599 }
2600 
2601 static int
2602 set_tx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2603 {
2604 	uint16_t i;
2605 	int diag;
2606 	uint8_t mapping_found = 0;
2607 
2608 	for (i = 0; i < nb_tx_queue_stats_mappings; i++) {
2609 		if ((tx_queue_stats_mappings[i].port_id == port_id) &&
2610 				(tx_queue_stats_mappings[i].queue_id < nb_txq )) {
2611 			diag = rte_eth_dev_set_tx_queue_stats_mapping(port_id,
2612 					tx_queue_stats_mappings[i].queue_id,
2613 					tx_queue_stats_mappings[i].stats_counter_id);
2614 			if (diag != 0)
2615 				return diag;
2616 			mapping_found = 1;
2617 		}
2618 	}
2619 	if (mapping_found)
2620 		port->tx_queue_stats_mapping_enabled = 1;
2621 	return 0;
2622 }
2623 
2624 static int
2625 set_rx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2626 {
2627 	uint16_t i;
2628 	int diag;
2629 	uint8_t mapping_found = 0;
2630 
2631 	for (i = 0; i < nb_rx_queue_stats_mappings; i++) {
2632 		if ((rx_queue_stats_mappings[i].port_id == port_id) &&
2633 				(rx_queue_stats_mappings[i].queue_id < nb_rxq )) {
2634 			diag = rte_eth_dev_set_rx_queue_stats_mapping(port_id,
2635 					rx_queue_stats_mappings[i].queue_id,
2636 					rx_queue_stats_mappings[i].stats_counter_id);
2637 			if (diag != 0)
2638 				return diag;
2639 			mapping_found = 1;
2640 		}
2641 	}
2642 	if (mapping_found)
2643 		port->rx_queue_stats_mapping_enabled = 1;
2644 	return 0;
2645 }
2646 
2647 static void
2648 map_port_queue_stats_mapping_registers(portid_t pi, struct rte_port *port)
2649 {
2650 	int diag = 0;
2651 
2652 	diag = set_tx_queue_stats_mapping_registers(pi, port);
2653 	if (diag != 0) {
2654 		if (diag == -ENOTSUP) {
2655 			port->tx_queue_stats_mapping_enabled = 0;
2656 			printf("TX queue stats mapping not supported port id=%d\n", pi);
2657 		}
2658 		else
2659 			rte_exit(EXIT_FAILURE,
2660 					"set_tx_queue_stats_mapping_registers "
2661 					"failed for port id=%d diag=%d\n",
2662 					pi, diag);
2663 	}
2664 
2665 	diag = set_rx_queue_stats_mapping_registers(pi, port);
2666 	if (diag != 0) {
2667 		if (diag == -ENOTSUP) {
2668 			port->rx_queue_stats_mapping_enabled = 0;
2669 			printf("RX queue stats mapping not supported port id=%d\n", pi);
2670 		}
2671 		else
2672 			rte_exit(EXIT_FAILURE,
2673 					"set_rx_queue_stats_mapping_registers "
2674 					"failed for port id=%d diag=%d\n",
2675 					pi, diag);
2676 	}
2677 }
2678 
2679 static void
2680 rxtx_port_config(struct rte_port *port)
2681 {
2682 	uint16_t qid;
2683 
2684 	for (qid = 0; qid < nb_rxq; qid++) {
2685 		port->rx_conf[qid] = port->dev_info.default_rxconf;
2686 
2687 		/* Check if any Rx parameters have been passed */
2688 		if (rx_pthresh != RTE_PMD_PARAM_UNSET)
2689 			port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
2690 
2691 		if (rx_hthresh != RTE_PMD_PARAM_UNSET)
2692 			port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
2693 
2694 		if (rx_wthresh != RTE_PMD_PARAM_UNSET)
2695 			port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
2696 
2697 		if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
2698 			port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
2699 
2700 		if (rx_drop_en != RTE_PMD_PARAM_UNSET)
2701 			port->rx_conf[qid].rx_drop_en = rx_drop_en;
2702 
2703 		port->nb_rx_desc[qid] = nb_rxd;
2704 	}
2705 
2706 	for (qid = 0; qid < nb_txq; qid++) {
2707 		port->tx_conf[qid] = port->dev_info.default_txconf;
2708 
2709 		/* Check if any Tx parameters have been passed */
2710 		if (tx_pthresh != RTE_PMD_PARAM_UNSET)
2711 			port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
2712 
2713 		if (tx_hthresh != RTE_PMD_PARAM_UNSET)
2714 			port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
2715 
2716 		if (tx_wthresh != RTE_PMD_PARAM_UNSET)
2717 			port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
2718 
2719 		if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
2720 			port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
2721 
2722 		if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
2723 			port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
2724 
2725 		port->nb_tx_desc[qid] = nb_txd;
2726 	}
2727 }
2728 
2729 void
2730 init_port_config(void)
2731 {
2732 	portid_t pid;
2733 	struct rte_port *port;
2734 
2735 	RTE_ETH_FOREACH_DEV(pid) {
2736 		port = &ports[pid];
2737 		port->dev_conf.fdir_conf = fdir_conf;
2738 		rte_eth_dev_info_get(pid, &port->dev_info);
2739 		if (nb_rxq > 1) {
2740 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2741 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
2742 				rss_hf & port->dev_info.flow_type_rss_offloads;
2743 		} else {
2744 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2745 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
2746 		}
2747 
2748 		if (port->dcb_flag == 0) {
2749 			if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
2750 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
2751 			else
2752 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
2753 		}
2754 
2755 		rxtx_port_config(port);
2756 
2757 		rte_eth_macaddr_get(pid, &port->eth_addr);
2758 
2759 		map_port_queue_stats_mapping_registers(pid, port);
2760 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
2761 		rte_pmd_ixgbe_bypass_init(pid);
2762 #endif
2763 
2764 		if (lsc_interrupt &&
2765 		    (rte_eth_devices[pid].data->dev_flags &
2766 		     RTE_ETH_DEV_INTR_LSC))
2767 			port->dev_conf.intr_conf.lsc = 1;
2768 		if (rmv_interrupt &&
2769 		    (rte_eth_devices[pid].data->dev_flags &
2770 		     RTE_ETH_DEV_INTR_RMV))
2771 			port->dev_conf.intr_conf.rmv = 1;
2772 	}
2773 }
2774 
2775 void set_port_slave_flag(portid_t slave_pid)
2776 {
2777 	struct rte_port *port;
2778 
2779 	port = &ports[slave_pid];
2780 	port->slave_flag = 1;
2781 }
2782 
2783 void clear_port_slave_flag(portid_t slave_pid)
2784 {
2785 	struct rte_port *port;
2786 
2787 	port = &ports[slave_pid];
2788 	port->slave_flag = 0;
2789 }
2790 
2791 uint8_t port_is_bonding_slave(portid_t slave_pid)
2792 {
2793 	struct rte_port *port;
2794 
2795 	port = &ports[slave_pid];
2796 	if ((rte_eth_devices[slave_pid].data->dev_flags &
2797 	    RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
2798 		return 1;
2799 	return 0;
2800 }
2801 
2802 const uint16_t vlan_tags[] = {
2803 		0,  1,  2,  3,  4,  5,  6,  7,
2804 		8,  9, 10, 11,  12, 13, 14, 15,
2805 		16, 17, 18, 19, 20, 21, 22, 23,
2806 		24, 25, 26, 27, 28, 29, 30, 31
2807 };
2808 
2809 static  int
2810 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
2811 		 enum dcb_mode_enable dcb_mode,
2812 		 enum rte_eth_nb_tcs num_tcs,
2813 		 uint8_t pfc_en)
2814 {
2815 	uint8_t i;
2816 	int32_t rc;
2817 	struct rte_eth_rss_conf rss_conf;
2818 
2819 	/*
2820 	 * Builds up the correct configuration for dcb+vt based on the vlan tags array
2821 	 * given above, and the number of traffic classes available for use.
2822 	 */
2823 	if (dcb_mode == DCB_VT_ENABLED) {
2824 		struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
2825 				&eth_conf->rx_adv_conf.vmdq_dcb_conf;
2826 		struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
2827 				&eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
2828 
2829 		/* VMDQ+DCB RX and TX configurations */
2830 		vmdq_rx_conf->enable_default_pool = 0;
2831 		vmdq_rx_conf->default_pool = 0;
2832 		vmdq_rx_conf->nb_queue_pools =
2833 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2834 		vmdq_tx_conf->nb_queue_pools =
2835 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2836 
2837 		vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
2838 		for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
2839 			vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
2840 			vmdq_rx_conf->pool_map[i].pools =
2841 				1 << (i % vmdq_rx_conf->nb_queue_pools);
2842 		}
2843 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2844 			vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
2845 			vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
2846 		}
2847 
2848 		/* set DCB mode of RX and TX of multiple queues */
2849 		eth_conf->rxmode.mq_mode = ETH_MQ_RX_VMDQ_DCB;
2850 		eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
2851 	} else {
2852 		struct rte_eth_dcb_rx_conf *rx_conf =
2853 				&eth_conf->rx_adv_conf.dcb_rx_conf;
2854 		struct rte_eth_dcb_tx_conf *tx_conf =
2855 				&eth_conf->tx_adv_conf.dcb_tx_conf;
2856 
2857 		rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
2858 		if (rc != 0)
2859 			return rc;
2860 
2861 		rx_conf->nb_tcs = num_tcs;
2862 		tx_conf->nb_tcs = num_tcs;
2863 
2864 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2865 			rx_conf->dcb_tc[i] = i % num_tcs;
2866 			tx_conf->dcb_tc[i] = i % num_tcs;
2867 		}
2868 
2869 		eth_conf->rxmode.mq_mode = ETH_MQ_RX_DCB_RSS;
2870 		eth_conf->rx_adv_conf.rss_conf = rss_conf;
2871 		eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
2872 	}
2873 
2874 	if (pfc_en)
2875 		eth_conf->dcb_capability_en =
2876 				ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
2877 	else
2878 		eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
2879 
2880 	return 0;
2881 }
2882 
2883 int
2884 init_port_dcb_config(portid_t pid,
2885 		     enum dcb_mode_enable dcb_mode,
2886 		     enum rte_eth_nb_tcs num_tcs,
2887 		     uint8_t pfc_en)
2888 {
2889 	struct rte_eth_conf port_conf;
2890 	struct rte_port *rte_port;
2891 	int retval;
2892 	uint16_t i;
2893 
2894 	rte_port = &ports[pid];
2895 
2896 	memset(&port_conf, 0, sizeof(struct rte_eth_conf));
2897 	/* Enter DCB configuration status */
2898 	dcb_config = 1;
2899 
2900 	port_conf.rxmode = rte_port->dev_conf.rxmode;
2901 	port_conf.txmode = rte_port->dev_conf.txmode;
2902 
2903 	/*set configuration of DCB in vt mode and DCB in non-vt mode*/
2904 	retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
2905 	if (retval < 0)
2906 		return retval;
2907 	port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
2908 
2909 	/* re-configure the device . */
2910 	rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
2911 
2912 	rte_eth_dev_info_get(pid, &rte_port->dev_info);
2913 
2914 	/* If dev_info.vmdq_pool_base is greater than 0,
2915 	 * the queue id of vmdq pools is started after pf queues.
2916 	 */
2917 	if (dcb_mode == DCB_VT_ENABLED &&
2918 	    rte_port->dev_info.vmdq_pool_base > 0) {
2919 		printf("VMDQ_DCB multi-queue mode is nonsensical"
2920 			" for port %d.", pid);
2921 		return -1;
2922 	}
2923 
2924 	/* Assume the ports in testpmd have the same dcb capability
2925 	 * and has the same number of rxq and txq in dcb mode
2926 	 */
2927 	if (dcb_mode == DCB_VT_ENABLED) {
2928 		if (rte_port->dev_info.max_vfs > 0) {
2929 			nb_rxq = rte_port->dev_info.nb_rx_queues;
2930 			nb_txq = rte_port->dev_info.nb_tx_queues;
2931 		} else {
2932 			nb_rxq = rte_port->dev_info.max_rx_queues;
2933 			nb_txq = rte_port->dev_info.max_tx_queues;
2934 		}
2935 	} else {
2936 		/*if vt is disabled, use all pf queues */
2937 		if (rte_port->dev_info.vmdq_pool_base == 0) {
2938 			nb_rxq = rte_port->dev_info.max_rx_queues;
2939 			nb_txq = rte_port->dev_info.max_tx_queues;
2940 		} else {
2941 			nb_rxq = (queueid_t)num_tcs;
2942 			nb_txq = (queueid_t)num_tcs;
2943 
2944 		}
2945 	}
2946 	rx_free_thresh = 64;
2947 
2948 	memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
2949 
2950 	rxtx_port_config(rte_port);
2951 	/* VLAN filter */
2952 	rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
2953 	for (i = 0; i < RTE_DIM(vlan_tags); i++)
2954 		rx_vft_set(pid, vlan_tags[i], 1);
2955 
2956 	rte_eth_macaddr_get(pid, &rte_port->eth_addr);
2957 	map_port_queue_stats_mapping_registers(pid, rte_port);
2958 
2959 	rte_port->dcb_flag = 1;
2960 
2961 	return 0;
2962 }
2963 
2964 static void
2965 init_port(void)
2966 {
2967 	/* Configuration of Ethernet ports. */
2968 	ports = rte_zmalloc("testpmd: ports",
2969 			    sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
2970 			    RTE_CACHE_LINE_SIZE);
2971 	if (ports == NULL) {
2972 		rte_exit(EXIT_FAILURE,
2973 				"rte_zmalloc(%d struct rte_port) failed\n",
2974 				RTE_MAX_ETHPORTS);
2975 	}
2976 
2977 	/* Initialize ports NUMA structures */
2978 	memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
2979 	memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
2980 	memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
2981 }
2982 
2983 static void
2984 force_quit(void)
2985 {
2986 	pmd_test_exit();
2987 	prompt_exit();
2988 }
2989 
2990 static void
2991 print_stats(void)
2992 {
2993 	uint8_t i;
2994 	const char clr[] = { 27, '[', '2', 'J', '\0' };
2995 	const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
2996 
2997 	/* Clear screen and move to top left */
2998 	printf("%s%s", clr, top_left);
2999 
3000 	printf("\nPort statistics ====================================");
3001 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3002 		nic_stats_display(fwd_ports_ids[i]);
3003 }
3004 
3005 static void
3006 signal_handler(int signum)
3007 {
3008 	if (signum == SIGINT || signum == SIGTERM) {
3009 		printf("\nSignal %d received, preparing to exit...\n",
3010 				signum);
3011 #ifdef RTE_LIBRTE_PDUMP
3012 		/* uninitialize packet capture framework */
3013 		rte_pdump_uninit();
3014 #endif
3015 #ifdef RTE_LIBRTE_LATENCY_STATS
3016 		rte_latencystats_uninit();
3017 #endif
3018 		force_quit();
3019 		/* Set flag to indicate the force termination. */
3020 		f_quit = 1;
3021 		/* exit with the expected status */
3022 		signal(signum, SIG_DFL);
3023 		kill(getpid(), signum);
3024 	}
3025 }
3026 
3027 int
3028 main(int argc, char** argv)
3029 {
3030 	int diag;
3031 	portid_t port_id;
3032 	uint16_t count;
3033 	int ret;
3034 
3035 	signal(SIGINT, signal_handler);
3036 	signal(SIGTERM, signal_handler);
3037 
3038 	diag = rte_eal_init(argc, argv);
3039 	if (diag < 0)
3040 		rte_panic("Cannot init EAL\n");
3041 
3042 	testpmd_logtype = rte_log_register("testpmd");
3043 	if (testpmd_logtype < 0)
3044 		rte_panic("Cannot register log type");
3045 	rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3046 
3047 #ifdef RTE_LIBRTE_PDUMP
3048 	/* initialize packet capture framework */
3049 	rte_pdump_init(NULL);
3050 #endif
3051 
3052 	count = 0;
3053 	RTE_ETH_FOREACH_DEV(port_id) {
3054 		ports_ids[count] = port_id;
3055 		count++;
3056 	}
3057 	nb_ports = (portid_t) count;
3058 	if (nb_ports == 0)
3059 		TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3060 
3061 	/* allocate port structures, and init them */
3062 	init_port();
3063 
3064 	set_def_fwd_config();
3065 	if (nb_lcores == 0)
3066 		rte_panic("Empty set of forwarding logical cores - check the "
3067 			  "core mask supplied in the command parameters\n");
3068 
3069 	/* Bitrate/latency stats disabled by default */
3070 #ifdef RTE_LIBRTE_BITRATE
3071 	bitrate_enabled = 0;
3072 #endif
3073 #ifdef RTE_LIBRTE_LATENCY_STATS
3074 	latencystats_enabled = 0;
3075 #endif
3076 
3077 	/* on FreeBSD, mlockall() is disabled by default */
3078 #ifdef RTE_EXEC_ENV_BSDAPP
3079 	do_mlockall = 0;
3080 #else
3081 	do_mlockall = 1;
3082 #endif
3083 
3084 	argc -= diag;
3085 	argv += diag;
3086 	if (argc > 1)
3087 		launch_args_parse(argc, argv);
3088 
3089 	if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3090 		TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3091 			strerror(errno));
3092 	}
3093 
3094 	if (tx_first && interactive)
3095 		rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3096 				"interactive mode.\n");
3097 
3098 	if (tx_first && lsc_interrupt) {
3099 		printf("Warning: lsc_interrupt needs to be off when "
3100 				" using tx_first. Disabling.\n");
3101 		lsc_interrupt = 0;
3102 	}
3103 
3104 	if (!nb_rxq && !nb_txq)
3105 		printf("Warning: Either rx or tx queues should be non-zero\n");
3106 
3107 	if (nb_rxq > 1 && nb_rxq > nb_txq)
3108 		printf("Warning: nb_rxq=%d enables RSS configuration, "
3109 		       "but nb_txq=%d will prevent to fully test it.\n",
3110 		       nb_rxq, nb_txq);
3111 
3112 	init_config();
3113 
3114 	if (hot_plug) {
3115 		ret = rte_dev_hotplug_handle_enable();
3116 		if (ret) {
3117 			RTE_LOG(ERR, EAL,
3118 				"fail to enable hotplug handling.");
3119 			return -1;
3120 		}
3121 
3122 		ret = rte_dev_event_monitor_start();
3123 		if (ret) {
3124 			RTE_LOG(ERR, EAL,
3125 				"fail to start device event monitoring.");
3126 			return -1;
3127 		}
3128 
3129 		ret = rte_dev_event_callback_register(NULL,
3130 			eth_dev_event_callback, NULL);
3131 		if (ret) {
3132 			RTE_LOG(ERR, EAL,
3133 				"fail  to register device event callback\n");
3134 			return -1;
3135 		}
3136 	}
3137 
3138 	if (start_port(RTE_PORT_ALL) != 0)
3139 		rte_exit(EXIT_FAILURE, "Start ports failed\n");
3140 
3141 	/* set all ports to promiscuous mode by default */
3142 	RTE_ETH_FOREACH_DEV(port_id)
3143 		rte_eth_promiscuous_enable(port_id);
3144 
3145 	/* Init metrics library */
3146 	rte_metrics_init(rte_socket_id());
3147 
3148 #ifdef RTE_LIBRTE_LATENCY_STATS
3149 	if (latencystats_enabled != 0) {
3150 		int ret = rte_latencystats_init(1, NULL);
3151 		if (ret)
3152 			printf("Warning: latencystats init()"
3153 				" returned error %d\n",	ret);
3154 		printf("Latencystats running on lcore %d\n",
3155 			latencystats_lcore_id);
3156 	}
3157 #endif
3158 
3159 	/* Setup bitrate stats */
3160 #ifdef RTE_LIBRTE_BITRATE
3161 	if (bitrate_enabled != 0) {
3162 		bitrate_data = rte_stats_bitrate_create();
3163 		if (bitrate_data == NULL)
3164 			rte_exit(EXIT_FAILURE,
3165 				"Could not allocate bitrate data.\n");
3166 		rte_stats_bitrate_reg(bitrate_data);
3167 	}
3168 #endif
3169 
3170 #ifdef RTE_LIBRTE_CMDLINE
3171 	if (strlen(cmdline_filename) != 0)
3172 		cmdline_read_from_file(cmdline_filename);
3173 
3174 	if (interactive == 1) {
3175 		if (auto_start) {
3176 			printf("Start automatic packet forwarding\n");
3177 			start_packet_forwarding(0);
3178 		}
3179 		prompt();
3180 		pmd_test_exit();
3181 	} else
3182 #endif
3183 	{
3184 		char c;
3185 		int rc;
3186 
3187 		f_quit = 0;
3188 
3189 		printf("No commandline core given, start packet forwarding\n");
3190 		start_packet_forwarding(tx_first);
3191 		if (stats_period != 0) {
3192 			uint64_t prev_time = 0, cur_time, diff_time = 0;
3193 			uint64_t timer_period;
3194 
3195 			/* Convert to number of cycles */
3196 			timer_period = stats_period * rte_get_timer_hz();
3197 
3198 			while (f_quit == 0) {
3199 				cur_time = rte_get_timer_cycles();
3200 				diff_time += cur_time - prev_time;
3201 
3202 				if (diff_time >= timer_period) {
3203 					print_stats();
3204 					/* Reset the timer */
3205 					diff_time = 0;
3206 				}
3207 				/* Sleep to avoid unnecessary checks */
3208 				prev_time = cur_time;
3209 				sleep(1);
3210 			}
3211 		}
3212 
3213 		printf("Press enter to exit\n");
3214 		rc = read(0, &c, 1);
3215 		pmd_test_exit();
3216 		if (rc < 0)
3217 			return 1;
3218 	}
3219 
3220 	return 0;
3221 }
3222