xref: /dpdk/app/test-pmd/testpmd.c (revision 2befc67ff6798a38c9f52f1f09b3f87bdd4afb4a)
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_memory.h>
31 #include <rte_memcpy.h>
32 #include <rte_launch.h>
33 #include <rte_eal.h>
34 #include <rte_alarm.h>
35 #include <rte_per_lcore.h>
36 #include <rte_lcore.h>
37 #include <rte_atomic.h>
38 #include <rte_branch_prediction.h>
39 #include <rte_mempool.h>
40 #include <rte_malloc.h>
41 #include <rte_mbuf.h>
42 #include <rte_mbuf_pool_ops.h>
43 #include <rte_interrupts.h>
44 #include <rte_pci.h>
45 #include <rte_ether.h>
46 #include <rte_ethdev.h>
47 #include <rte_dev.h>
48 #include <rte_string_fns.h>
49 #ifdef RTE_LIBRTE_IXGBE_PMD
50 #include <rte_pmd_ixgbe.h>
51 #endif
52 #ifdef RTE_LIBRTE_PDUMP
53 #include <rte_pdump.h>
54 #endif
55 #include <rte_flow.h>
56 #include <rte_metrics.h>
57 #ifdef RTE_LIBRTE_BITRATESTATS
58 #include <rte_bitrate.h>
59 #endif
60 #ifdef RTE_LIBRTE_LATENCY_STATS
61 #include <rte_latencystats.h>
62 #endif
63 
64 #include "testpmd.h"
65 
66 #ifndef MAP_HUGETLB
67 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
68 #define HUGE_FLAG (0x40000)
69 #else
70 #define HUGE_FLAG MAP_HUGETLB
71 #endif
72 
73 #ifndef MAP_HUGE_SHIFT
74 /* older kernels (or FreeBSD) will not have this define */
75 #define HUGE_SHIFT (26)
76 #else
77 #define HUGE_SHIFT MAP_HUGE_SHIFT
78 #endif
79 
80 #define EXTMEM_HEAP_NAME "extmem"
81 #define EXTBUF_ZONE_SIZE RTE_PGSIZE_2M
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 rte_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 	&five_tuple_swap_fwd_engine,
183 #ifdef RTE_LIBRTE_IEEE1588
184 	&ieee1588_fwd_engine,
185 #endif
186 	NULL,
187 };
188 
189 struct rte_mempool *mempools[RTE_MAX_NUMA_NODES * MAX_SEGS_BUFFER_SPLIT];
190 uint16_t mempool_flags;
191 
192 struct fwd_config cur_fwd_config;
193 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
194 uint32_t retry_enabled;
195 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
196 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
197 
198 uint32_t mbuf_data_size_n = 1; /* Number of specified mbuf sizes. */
199 uint16_t mbuf_data_size[MAX_SEGS_BUFFER_SPLIT] = {
200 	DEFAULT_MBUF_DATA_SIZE
201 }; /**< Mbuf data space size. */
202 uint32_t param_total_num_mbufs = 0;  /**< number of mbufs in all pools - if
203                                       * specified on command-line. */
204 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
205 
206 /*
207  * In container, it cannot terminate the process which running with 'stats-period'
208  * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
209  */
210 uint8_t f_quit;
211 
212 /*
213  * Configuration of packet segments used to scatter received packets
214  * if some of split features is configured.
215  */
216 uint16_t rx_pkt_seg_lengths[MAX_SEGS_BUFFER_SPLIT];
217 uint8_t  rx_pkt_nb_segs; /**< Number of segments to split */
218 uint16_t rx_pkt_seg_offsets[MAX_SEGS_BUFFER_SPLIT];
219 uint8_t  rx_pkt_nb_offs; /**< Number of specified offsets */
220 
221 /*
222  * Configuration of packet segments used by the "txonly" processing engine.
223  */
224 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
225 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
226 	TXONLY_DEF_PACKET_LEN,
227 };
228 uint8_t  tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
229 
230 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
231 /**< Split policy for packets to TX. */
232 
233 uint8_t txonly_multi_flow;
234 /**< Whether multiple flows are generated in TXONLY mode. */
235 
236 uint32_t tx_pkt_times_inter;
237 /**< Timings for send scheduling in TXONLY mode, time between bursts. */
238 
239 uint32_t tx_pkt_times_intra;
240 /**< Timings for send scheduling in TXONLY mode, time between packets. */
241 
242 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
243 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
244 
245 /* current configuration is in DCB or not,0 means it is not in DCB mode */
246 uint8_t dcb_config = 0;
247 
248 /* Whether the dcb is in testing status */
249 uint8_t dcb_test = 0;
250 
251 /*
252  * Configurable number of RX/TX queues.
253  */
254 queueid_t nb_hairpinq; /**< Number of hairpin queues per port. */
255 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
256 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
257 
258 /*
259  * Configurable number of RX/TX ring descriptors.
260  * Defaults are supplied by drivers via ethdev.
261  */
262 #define RTE_TEST_RX_DESC_DEFAULT 0
263 #define RTE_TEST_TX_DESC_DEFAULT 0
264 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
265 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
266 
267 #define RTE_PMD_PARAM_UNSET -1
268 /*
269  * Configurable values of RX and TX ring threshold registers.
270  */
271 
272 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
273 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
274 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
275 
276 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
277 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
278 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
279 
280 /*
281  * Configurable value of RX free threshold.
282  */
283 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
284 
285 /*
286  * Configurable value of RX drop enable.
287  */
288 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
289 
290 /*
291  * Configurable value of TX free threshold.
292  */
293 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
294 
295 /*
296  * Configurable value of TX RS bit threshold.
297  */
298 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
299 
300 /*
301  * Configurable value of buffered packets before sending.
302  */
303 uint16_t noisy_tx_sw_bufsz;
304 
305 /*
306  * Configurable value of packet buffer timeout.
307  */
308 uint16_t noisy_tx_sw_buf_flush_time;
309 
310 /*
311  * Configurable value for size of VNF internal memory area
312  * used for simulating noisy neighbour behaviour
313  */
314 uint64_t noisy_lkup_mem_sz;
315 
316 /*
317  * Configurable value of number of random writes done in
318  * VNF simulation memory area.
319  */
320 uint64_t noisy_lkup_num_writes;
321 
322 /*
323  * Configurable value of number of random reads done in
324  * VNF simulation memory area.
325  */
326 uint64_t noisy_lkup_num_reads;
327 
328 /*
329  * Configurable value of number of random reads/writes done in
330  * VNF simulation memory area.
331  */
332 uint64_t noisy_lkup_num_reads_writes;
333 
334 /*
335  * Receive Side Scaling (RSS) configuration.
336  */
337 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
338 
339 /*
340  * Port topology configuration
341  */
342 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
343 
344 /*
345  * Avoids to flush all the RX streams before starts forwarding.
346  */
347 uint8_t no_flush_rx = 0; /* flush by default */
348 
349 /*
350  * Flow API isolated mode.
351  */
352 uint8_t flow_isolate_all;
353 
354 /*
355  * Avoids to check link status when starting/stopping a port.
356  */
357 uint8_t no_link_check = 0; /* check by default */
358 
359 /*
360  * Don't automatically start all ports in interactive mode.
361  */
362 uint8_t no_device_start = 0;
363 
364 /*
365  * Enable link status change notification
366  */
367 uint8_t lsc_interrupt = 1; /* enabled by default */
368 
369 /*
370  * Enable device removal notification.
371  */
372 uint8_t rmv_interrupt = 1; /* enabled by default */
373 
374 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
375 
376 /* After attach, port setup is called on event or by iterator */
377 bool setup_on_probe_event = true;
378 
379 /* Clear ptypes on port initialization. */
380 uint8_t clear_ptypes = true;
381 
382 /* Hairpin ports configuration mode. */
383 uint16_t hairpin_mode;
384 
385 /* Pretty printing of ethdev events */
386 static const char * const eth_event_desc[] = {
387 	[RTE_ETH_EVENT_UNKNOWN] = "unknown",
388 	[RTE_ETH_EVENT_INTR_LSC] = "link state change",
389 	[RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
390 	[RTE_ETH_EVENT_INTR_RESET] = "reset",
391 	[RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
392 	[RTE_ETH_EVENT_IPSEC] = "IPsec",
393 	[RTE_ETH_EVENT_MACSEC] = "MACsec",
394 	[RTE_ETH_EVENT_INTR_RMV] = "device removal",
395 	[RTE_ETH_EVENT_NEW] = "device probed",
396 	[RTE_ETH_EVENT_DESTROY] = "device released",
397 	[RTE_ETH_EVENT_FLOW_AGED] = "flow aged",
398 	[RTE_ETH_EVENT_MAX] = NULL,
399 };
400 
401 /*
402  * Display or mask ether events
403  * Default to all events except VF_MBOX
404  */
405 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
406 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
407 			    (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
408 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
409 			    (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
410 			    (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
411 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV) |
412 			    (UINT32_C(1) << RTE_ETH_EVENT_FLOW_AGED);
413 /*
414  * Decide if all memory are locked for performance.
415  */
416 int do_mlockall = 0;
417 
418 /*
419  * NIC bypass mode configuration options.
420  */
421 
422 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
423 /* The NIC bypass watchdog timeout. */
424 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
425 #endif
426 
427 
428 #ifdef RTE_LIBRTE_LATENCY_STATS
429 
430 /*
431  * Set when latency stats is enabled in the commandline
432  */
433 uint8_t latencystats_enabled;
434 
435 /*
436  * Lcore ID to serive latency statistics.
437  */
438 lcoreid_t latencystats_lcore_id = -1;
439 
440 #endif
441 
442 /*
443  * Ethernet device configuration.
444  */
445 struct rte_eth_rxmode rx_mode = {
446 	.max_rx_pkt_len = RTE_ETHER_MAX_LEN,
447 		/**< Default maximum frame length. */
448 };
449 
450 struct rte_eth_txmode tx_mode = {
451 	.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
452 };
453 
454 struct rte_fdir_conf fdir_conf = {
455 	.mode = RTE_FDIR_MODE_NONE,
456 	.pballoc = RTE_FDIR_PBALLOC_64K,
457 	.status = RTE_FDIR_REPORT_STATUS,
458 	.mask = {
459 		.vlan_tci_mask = 0xFFEF,
460 		.ipv4_mask     = {
461 			.src_ip = 0xFFFFFFFF,
462 			.dst_ip = 0xFFFFFFFF,
463 		},
464 		.ipv6_mask     = {
465 			.src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
466 			.dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
467 		},
468 		.src_port_mask = 0xFFFF,
469 		.dst_port_mask = 0xFFFF,
470 		.mac_addr_byte_mask = 0xFF,
471 		.tunnel_type_mask = 1,
472 		.tunnel_id_mask = 0xFFFFFFFF,
473 	},
474 	.drop_queue = 127,
475 };
476 
477 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
478 
479 struct queue_stats_mappings tx_queue_stats_mappings_array[MAX_TX_QUEUE_STATS_MAPPINGS];
480 struct queue_stats_mappings rx_queue_stats_mappings_array[MAX_RX_QUEUE_STATS_MAPPINGS];
481 
482 struct queue_stats_mappings *tx_queue_stats_mappings = tx_queue_stats_mappings_array;
483 struct queue_stats_mappings *rx_queue_stats_mappings = rx_queue_stats_mappings_array;
484 
485 uint16_t nb_tx_queue_stats_mappings = 0;
486 uint16_t nb_rx_queue_stats_mappings = 0;
487 
488 /*
489  * Display zero values by default for xstats
490  */
491 uint8_t xstats_hide_zero;
492 
493 /*
494  * Measure of CPU cycles disabled by default
495  */
496 uint8_t record_core_cycles;
497 
498 /*
499  * Display of RX and TX bursts disabled by default
500  */
501 uint8_t record_burst_stats;
502 
503 unsigned int num_sockets = 0;
504 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
505 
506 #ifdef RTE_LIBRTE_BITRATESTATS
507 /* Bitrate statistics */
508 struct rte_stats_bitrates *bitrate_data;
509 lcoreid_t bitrate_lcore_id;
510 uint8_t bitrate_enabled;
511 #endif
512 
513 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
514 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
515 
516 /*
517  * hexadecimal bitmask of RX mq mode can be enabled.
518  */
519 enum rte_eth_rx_mq_mode rx_mq_mode = ETH_MQ_RX_VMDQ_DCB_RSS;
520 
521 /* Forward function declarations */
522 static void setup_attached_port(portid_t pi);
523 static void map_port_queue_stats_mapping_registers(portid_t pi,
524 						   struct rte_port *port);
525 static void check_all_ports_link_status(uint32_t port_mask);
526 static int eth_event_callback(portid_t port_id,
527 			      enum rte_eth_event_type type,
528 			      void *param, void *ret_param);
529 static void dev_event_callback(const char *device_name,
530 				enum rte_dev_event_type type,
531 				void *param);
532 
533 /*
534  * Check if all the ports are started.
535  * If yes, return positive value. If not, return zero.
536  */
537 static int all_ports_started(void);
538 
539 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
540 uint16_t gso_max_segment_size = RTE_ETHER_MAX_LEN - RTE_ETHER_CRC_LEN;
541 
542 /* Holds the registered mbuf dynamic flags names. */
543 char dynf_names[64][RTE_MBUF_DYN_NAMESIZE];
544 
545 /*
546  * Helper function to check if socket is already discovered.
547  * If yes, return positive value. If not, return zero.
548  */
549 int
550 new_socket_id(unsigned int socket_id)
551 {
552 	unsigned int i;
553 
554 	for (i = 0; i < num_sockets; i++) {
555 		if (socket_ids[i] == socket_id)
556 			return 0;
557 	}
558 	return 1;
559 }
560 
561 /*
562  * Setup default configuration.
563  */
564 static void
565 set_default_fwd_lcores_config(void)
566 {
567 	unsigned int i;
568 	unsigned int nb_lc;
569 	unsigned int sock_num;
570 
571 	nb_lc = 0;
572 	for (i = 0; i < RTE_MAX_LCORE; i++) {
573 		if (!rte_lcore_is_enabled(i))
574 			continue;
575 		sock_num = rte_lcore_to_socket_id(i);
576 		if (new_socket_id(sock_num)) {
577 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
578 				rte_exit(EXIT_FAILURE,
579 					 "Total sockets greater than %u\n",
580 					 RTE_MAX_NUMA_NODES);
581 			}
582 			socket_ids[num_sockets++] = sock_num;
583 		}
584 		if (i == rte_get_master_lcore())
585 			continue;
586 		fwd_lcores_cpuids[nb_lc++] = i;
587 	}
588 	nb_lcores = (lcoreid_t) nb_lc;
589 	nb_cfg_lcores = nb_lcores;
590 	nb_fwd_lcores = 1;
591 }
592 
593 static void
594 set_def_peer_eth_addrs(void)
595 {
596 	portid_t i;
597 
598 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
599 		peer_eth_addrs[i].addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
600 		peer_eth_addrs[i].addr_bytes[5] = i;
601 	}
602 }
603 
604 static void
605 set_default_fwd_ports_config(void)
606 {
607 	portid_t pt_id;
608 	int i = 0;
609 
610 	RTE_ETH_FOREACH_DEV(pt_id) {
611 		fwd_ports_ids[i++] = pt_id;
612 
613 		/* Update sockets info according to the attached device */
614 		int socket_id = rte_eth_dev_socket_id(pt_id);
615 		if (socket_id >= 0 && new_socket_id(socket_id)) {
616 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
617 				rte_exit(EXIT_FAILURE,
618 					 "Total sockets greater than %u\n",
619 					 RTE_MAX_NUMA_NODES);
620 			}
621 			socket_ids[num_sockets++] = socket_id;
622 		}
623 	}
624 
625 	nb_cfg_ports = nb_ports;
626 	nb_fwd_ports = nb_ports;
627 }
628 
629 void
630 set_def_fwd_config(void)
631 {
632 	set_default_fwd_lcores_config();
633 	set_def_peer_eth_addrs();
634 	set_default_fwd_ports_config();
635 }
636 
637 /* extremely pessimistic estimation of memory required to create a mempool */
638 static int
639 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
640 {
641 	unsigned int n_pages, mbuf_per_pg, leftover;
642 	uint64_t total_mem, mbuf_mem, obj_sz;
643 
644 	/* there is no good way to predict how much space the mempool will
645 	 * occupy because it will allocate chunks on the fly, and some of those
646 	 * will come from default DPDK memory while some will come from our
647 	 * external memory, so just assume 128MB will be enough for everyone.
648 	 */
649 	uint64_t hdr_mem = 128 << 20;
650 
651 	/* account for possible non-contiguousness */
652 	obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
653 	if (obj_sz > pgsz) {
654 		TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
655 		return -1;
656 	}
657 
658 	mbuf_per_pg = pgsz / obj_sz;
659 	leftover = (nb_mbufs % mbuf_per_pg) > 0;
660 	n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
661 
662 	mbuf_mem = n_pages * pgsz;
663 
664 	total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
665 
666 	if (total_mem > SIZE_MAX) {
667 		TESTPMD_LOG(ERR, "Memory size too big\n");
668 		return -1;
669 	}
670 	*out = (size_t)total_mem;
671 
672 	return 0;
673 }
674 
675 static int
676 pagesz_flags(uint64_t page_sz)
677 {
678 	/* as per mmap() manpage, all page sizes are log2 of page size
679 	 * shifted by MAP_HUGE_SHIFT
680 	 */
681 	int log2 = rte_log2_u64(page_sz);
682 
683 	return (log2 << HUGE_SHIFT);
684 }
685 
686 static void *
687 alloc_mem(size_t memsz, size_t pgsz, bool huge)
688 {
689 	void *addr;
690 	int flags;
691 
692 	/* allocate anonymous hugepages */
693 	flags = MAP_ANONYMOUS | MAP_PRIVATE;
694 	if (huge)
695 		flags |= HUGE_FLAG | pagesz_flags(pgsz);
696 
697 	addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
698 	if (addr == MAP_FAILED)
699 		return NULL;
700 
701 	return addr;
702 }
703 
704 struct extmem_param {
705 	void *addr;
706 	size_t len;
707 	size_t pgsz;
708 	rte_iova_t *iova_table;
709 	unsigned int iova_table_len;
710 };
711 
712 static int
713 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
714 		bool huge)
715 {
716 	uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
717 			RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
718 	unsigned int cur_page, n_pages, pgsz_idx;
719 	size_t mem_sz, cur_pgsz;
720 	rte_iova_t *iovas = NULL;
721 	void *addr;
722 	int ret;
723 
724 	for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
725 		/* skip anything that is too big */
726 		if (pgsizes[pgsz_idx] > SIZE_MAX)
727 			continue;
728 
729 		cur_pgsz = pgsizes[pgsz_idx];
730 
731 		/* if we were told not to allocate hugepages, override */
732 		if (!huge)
733 			cur_pgsz = sysconf(_SC_PAGESIZE);
734 
735 		ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
736 		if (ret < 0) {
737 			TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
738 			return -1;
739 		}
740 
741 		/* allocate our memory */
742 		addr = alloc_mem(mem_sz, cur_pgsz, huge);
743 
744 		/* if we couldn't allocate memory with a specified page size,
745 		 * that doesn't mean we can't do it with other page sizes, so
746 		 * try another one.
747 		 */
748 		if (addr == NULL)
749 			continue;
750 
751 		/* store IOVA addresses for every page in this memory area */
752 		n_pages = mem_sz / cur_pgsz;
753 
754 		iovas = malloc(sizeof(*iovas) * n_pages);
755 
756 		if (iovas == NULL) {
757 			TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
758 			goto fail;
759 		}
760 		/* lock memory if it's not huge pages */
761 		if (!huge)
762 			mlock(addr, mem_sz);
763 
764 		/* populate IOVA addresses */
765 		for (cur_page = 0; cur_page < n_pages; cur_page++) {
766 			rte_iova_t iova;
767 			size_t offset;
768 			void *cur;
769 
770 			offset = cur_pgsz * cur_page;
771 			cur = RTE_PTR_ADD(addr, offset);
772 
773 			/* touch the page before getting its IOVA */
774 			*(volatile char *)cur = 0;
775 
776 			iova = rte_mem_virt2iova(cur);
777 
778 			iovas[cur_page] = iova;
779 		}
780 
781 		break;
782 	}
783 	/* if we couldn't allocate anything */
784 	if (iovas == NULL)
785 		return -1;
786 
787 	param->addr = addr;
788 	param->len = mem_sz;
789 	param->pgsz = cur_pgsz;
790 	param->iova_table = iovas;
791 	param->iova_table_len = n_pages;
792 
793 	return 0;
794 fail:
795 	if (iovas)
796 		free(iovas);
797 	if (addr)
798 		munmap(addr, mem_sz);
799 
800 	return -1;
801 }
802 
803 static int
804 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
805 {
806 	struct extmem_param param;
807 	int socket_id, ret;
808 
809 	memset(&param, 0, sizeof(param));
810 
811 	/* check if our heap exists */
812 	socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
813 	if (socket_id < 0) {
814 		/* create our heap */
815 		ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
816 		if (ret < 0) {
817 			TESTPMD_LOG(ERR, "Cannot create heap\n");
818 			return -1;
819 		}
820 	}
821 
822 	ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
823 	if (ret < 0) {
824 		TESTPMD_LOG(ERR, "Cannot create memory area\n");
825 		return -1;
826 	}
827 
828 	/* we now have a valid memory area, so add it to heap */
829 	ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
830 			param.addr, param.len, param.iova_table,
831 			param.iova_table_len, param.pgsz);
832 
833 	/* when using VFIO, memory is automatically mapped for DMA by EAL */
834 
835 	/* not needed any more */
836 	free(param.iova_table);
837 
838 	if (ret < 0) {
839 		TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
840 		munmap(param.addr, param.len);
841 		return -1;
842 	}
843 
844 	/* success */
845 
846 	TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
847 			param.len >> 20);
848 
849 	return 0;
850 }
851 static void
852 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
853 	     struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
854 {
855 	uint16_t pid = 0;
856 	int ret;
857 
858 	RTE_ETH_FOREACH_DEV(pid) {
859 		struct rte_eth_dev *dev =
860 			&rte_eth_devices[pid];
861 
862 		ret = rte_dev_dma_unmap(dev->device, memhdr->addr, 0,
863 					memhdr->len);
864 		if (ret) {
865 			TESTPMD_LOG(DEBUG,
866 				    "unable to DMA unmap addr 0x%p "
867 				    "for device %s\n",
868 				    memhdr->addr, dev->data->name);
869 		}
870 	}
871 	ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
872 	if (ret) {
873 		TESTPMD_LOG(DEBUG,
874 			    "unable to un-register addr 0x%p\n", memhdr->addr);
875 	}
876 }
877 
878 static void
879 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
880 	   struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
881 {
882 	uint16_t pid = 0;
883 	size_t page_size = sysconf(_SC_PAGESIZE);
884 	int ret;
885 
886 	ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
887 				  page_size);
888 	if (ret) {
889 		TESTPMD_LOG(DEBUG,
890 			    "unable to register addr 0x%p\n", memhdr->addr);
891 		return;
892 	}
893 	RTE_ETH_FOREACH_DEV(pid) {
894 		struct rte_eth_dev *dev =
895 			&rte_eth_devices[pid];
896 
897 		ret = rte_dev_dma_map(dev->device, memhdr->addr, 0,
898 				      memhdr->len);
899 		if (ret) {
900 			TESTPMD_LOG(DEBUG,
901 				    "unable to DMA map addr 0x%p "
902 				    "for device %s\n",
903 				    memhdr->addr, dev->data->name);
904 		}
905 	}
906 }
907 
908 static unsigned int
909 setup_extbuf(uint32_t nb_mbufs, uint16_t mbuf_sz, unsigned int socket_id,
910 	    char *pool_name, struct rte_pktmbuf_extmem **ext_mem)
911 {
912 	struct rte_pktmbuf_extmem *xmem;
913 	unsigned int ext_num, zone_num, elt_num;
914 	uint16_t elt_size;
915 
916 	elt_size = RTE_ALIGN_CEIL(mbuf_sz, RTE_CACHE_LINE_SIZE);
917 	elt_num = EXTBUF_ZONE_SIZE / elt_size;
918 	zone_num = (nb_mbufs + elt_num - 1) / elt_num;
919 
920 	xmem = malloc(sizeof(struct rte_pktmbuf_extmem) * zone_num);
921 	if (xmem == NULL) {
922 		TESTPMD_LOG(ERR, "Cannot allocate memory for "
923 				 "external buffer descriptors\n");
924 		*ext_mem = NULL;
925 		return 0;
926 	}
927 	for (ext_num = 0; ext_num < zone_num; ext_num++) {
928 		struct rte_pktmbuf_extmem *xseg = xmem + ext_num;
929 		const struct rte_memzone *mz;
930 		char mz_name[RTE_MEMZONE_NAMESIZE];
931 		int ret;
932 
933 		ret = snprintf(mz_name, sizeof(mz_name),
934 			RTE_MEMPOOL_MZ_FORMAT "_xb_%u", pool_name, ext_num);
935 		if (ret < 0 || ret >= (int)sizeof(mz_name)) {
936 			errno = ENAMETOOLONG;
937 			ext_num = 0;
938 			break;
939 		}
940 		mz = rte_memzone_reserve_aligned(mz_name, EXTBUF_ZONE_SIZE,
941 						 socket_id,
942 						 RTE_MEMZONE_IOVA_CONTIG |
943 						 RTE_MEMZONE_1GB |
944 						 RTE_MEMZONE_SIZE_HINT_ONLY,
945 						 EXTBUF_ZONE_SIZE);
946 		if (mz == NULL) {
947 			/*
948 			 * The caller exits on external buffer creation
949 			 * error, so there is no need to free memzones.
950 			 */
951 			errno = ENOMEM;
952 			ext_num = 0;
953 			break;
954 		}
955 		xseg->buf_ptr = mz->addr;
956 		xseg->buf_iova = mz->iova;
957 		xseg->buf_len = EXTBUF_ZONE_SIZE;
958 		xseg->elt_size = elt_size;
959 	}
960 	if (ext_num == 0 && xmem != NULL) {
961 		free(xmem);
962 		xmem = NULL;
963 	}
964 	*ext_mem = xmem;
965 	return ext_num;
966 }
967 
968 /*
969  * Configuration initialisation done once at init time.
970  */
971 static struct rte_mempool *
972 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
973 		 unsigned int socket_id, uint16_t size_idx)
974 {
975 	char pool_name[RTE_MEMPOOL_NAMESIZE];
976 	struct rte_mempool *rte_mp = NULL;
977 	uint32_t mb_size;
978 
979 	mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
980 	mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name), size_idx);
981 
982 	TESTPMD_LOG(INFO,
983 		"create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
984 		pool_name, nb_mbuf, mbuf_seg_size, socket_id);
985 
986 	switch (mp_alloc_type) {
987 	case MP_ALLOC_NATIVE:
988 		{
989 			/* wrapper to rte_mempool_create() */
990 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
991 					rte_mbuf_best_mempool_ops());
992 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
993 				mb_mempool_cache, 0, mbuf_seg_size, socket_id);
994 			break;
995 		}
996 	case MP_ALLOC_ANON:
997 		{
998 			rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
999 				mb_size, (unsigned int) mb_mempool_cache,
1000 				sizeof(struct rte_pktmbuf_pool_private),
1001 				socket_id, mempool_flags);
1002 			if (rte_mp == NULL)
1003 				goto err;
1004 
1005 			if (rte_mempool_populate_anon(rte_mp) == 0) {
1006 				rte_mempool_free(rte_mp);
1007 				rte_mp = NULL;
1008 				goto err;
1009 			}
1010 			rte_pktmbuf_pool_init(rte_mp, NULL);
1011 			rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
1012 			rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
1013 			break;
1014 		}
1015 	case MP_ALLOC_XMEM:
1016 	case MP_ALLOC_XMEM_HUGE:
1017 		{
1018 			int heap_socket;
1019 			bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
1020 
1021 			if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
1022 				rte_exit(EXIT_FAILURE, "Could not create external memory\n");
1023 
1024 			heap_socket =
1025 				rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
1026 			if (heap_socket < 0)
1027 				rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
1028 
1029 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1030 					rte_mbuf_best_mempool_ops());
1031 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1032 					mb_mempool_cache, 0, mbuf_seg_size,
1033 					heap_socket);
1034 			break;
1035 		}
1036 	case MP_ALLOC_XBUF:
1037 		{
1038 			struct rte_pktmbuf_extmem *ext_mem;
1039 			unsigned int ext_num;
1040 
1041 			ext_num = setup_extbuf(nb_mbuf,	mbuf_seg_size,
1042 					       socket_id, pool_name, &ext_mem);
1043 			if (ext_num == 0)
1044 				rte_exit(EXIT_FAILURE,
1045 					 "Can't create pinned data buffers\n");
1046 
1047 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1048 					rte_mbuf_best_mempool_ops());
1049 			rte_mp = rte_pktmbuf_pool_create_extbuf
1050 					(pool_name, nb_mbuf, mb_mempool_cache,
1051 					 0, mbuf_seg_size, socket_id,
1052 					 ext_mem, ext_num);
1053 			free(ext_mem);
1054 			break;
1055 		}
1056 	default:
1057 		{
1058 			rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
1059 		}
1060 	}
1061 
1062 err:
1063 	if (rte_mp == NULL) {
1064 		rte_exit(EXIT_FAILURE,
1065 			"Creation of mbuf pool for socket %u failed: %s\n",
1066 			socket_id, rte_strerror(rte_errno));
1067 	} else if (verbose_level > 0) {
1068 		rte_mempool_dump(stdout, rte_mp);
1069 	}
1070 	return rte_mp;
1071 }
1072 
1073 /*
1074  * Check given socket id is valid or not with NUMA mode,
1075  * if valid, return 0, else return -1
1076  */
1077 static int
1078 check_socket_id(const unsigned int socket_id)
1079 {
1080 	static int warning_once = 0;
1081 
1082 	if (new_socket_id(socket_id)) {
1083 		if (!warning_once && numa_support)
1084 			printf("Warning: NUMA should be configured manually by"
1085 			       " using --port-numa-config and"
1086 			       " --ring-numa-config parameters along with"
1087 			       " --numa.\n");
1088 		warning_once = 1;
1089 		return -1;
1090 	}
1091 	return 0;
1092 }
1093 
1094 /*
1095  * Get the allowed maximum number of RX queues.
1096  * *pid return the port id which has minimal value of
1097  * max_rx_queues in all ports.
1098  */
1099 queueid_t
1100 get_allowed_max_nb_rxq(portid_t *pid)
1101 {
1102 	queueid_t allowed_max_rxq = RTE_MAX_QUEUES_PER_PORT;
1103 	bool max_rxq_valid = false;
1104 	portid_t pi;
1105 	struct rte_eth_dev_info dev_info;
1106 
1107 	RTE_ETH_FOREACH_DEV(pi) {
1108 		if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1109 			continue;
1110 
1111 		max_rxq_valid = true;
1112 		if (dev_info.max_rx_queues < allowed_max_rxq) {
1113 			allowed_max_rxq = dev_info.max_rx_queues;
1114 			*pid = pi;
1115 		}
1116 	}
1117 	return max_rxq_valid ? allowed_max_rxq : 0;
1118 }
1119 
1120 /*
1121  * Check input rxq is valid or not.
1122  * If input rxq is not greater than any of maximum number
1123  * of RX queues of all ports, it is valid.
1124  * if valid, return 0, else return -1
1125  */
1126 int
1127 check_nb_rxq(queueid_t rxq)
1128 {
1129 	queueid_t allowed_max_rxq;
1130 	portid_t pid = 0;
1131 
1132 	allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1133 	if (rxq > allowed_max_rxq) {
1134 		printf("Fail: input rxq (%u) can't be greater "
1135 		       "than max_rx_queues (%u) of port %u\n",
1136 		       rxq,
1137 		       allowed_max_rxq,
1138 		       pid);
1139 		return -1;
1140 	}
1141 	return 0;
1142 }
1143 
1144 /*
1145  * Get the allowed maximum number of TX queues.
1146  * *pid return the port id which has minimal value of
1147  * max_tx_queues in all ports.
1148  */
1149 queueid_t
1150 get_allowed_max_nb_txq(portid_t *pid)
1151 {
1152 	queueid_t allowed_max_txq = RTE_MAX_QUEUES_PER_PORT;
1153 	bool max_txq_valid = false;
1154 	portid_t pi;
1155 	struct rte_eth_dev_info dev_info;
1156 
1157 	RTE_ETH_FOREACH_DEV(pi) {
1158 		if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1159 			continue;
1160 
1161 		max_txq_valid = true;
1162 		if (dev_info.max_tx_queues < allowed_max_txq) {
1163 			allowed_max_txq = dev_info.max_tx_queues;
1164 			*pid = pi;
1165 		}
1166 	}
1167 	return max_txq_valid ? allowed_max_txq : 0;
1168 }
1169 
1170 /*
1171  * Check input txq is valid or not.
1172  * If input txq is not greater than any of maximum number
1173  * of TX queues of all ports, it is valid.
1174  * if valid, return 0, else return -1
1175  */
1176 int
1177 check_nb_txq(queueid_t txq)
1178 {
1179 	queueid_t allowed_max_txq;
1180 	portid_t pid = 0;
1181 
1182 	allowed_max_txq = get_allowed_max_nb_txq(&pid);
1183 	if (txq > allowed_max_txq) {
1184 		printf("Fail: input txq (%u) can't be greater "
1185 		       "than max_tx_queues (%u) of port %u\n",
1186 		       txq,
1187 		       allowed_max_txq,
1188 		       pid);
1189 		return -1;
1190 	}
1191 	return 0;
1192 }
1193 
1194 /*
1195  * Get the allowed maximum number of RXDs of every rx queue.
1196  * *pid return the port id which has minimal value of
1197  * max_rxd in all queues of all ports.
1198  */
1199 static uint16_t
1200 get_allowed_max_nb_rxd(portid_t *pid)
1201 {
1202 	uint16_t allowed_max_rxd = UINT16_MAX;
1203 	portid_t pi;
1204 	struct rte_eth_dev_info dev_info;
1205 
1206 	RTE_ETH_FOREACH_DEV(pi) {
1207 		if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1208 			continue;
1209 
1210 		if (dev_info.rx_desc_lim.nb_max < allowed_max_rxd) {
1211 			allowed_max_rxd = dev_info.rx_desc_lim.nb_max;
1212 			*pid = pi;
1213 		}
1214 	}
1215 	return allowed_max_rxd;
1216 }
1217 
1218 /*
1219  * Get the allowed minimal number of RXDs of every rx queue.
1220  * *pid return the port id which has minimal value of
1221  * min_rxd in all queues of all ports.
1222  */
1223 static uint16_t
1224 get_allowed_min_nb_rxd(portid_t *pid)
1225 {
1226 	uint16_t allowed_min_rxd = 0;
1227 	portid_t pi;
1228 	struct rte_eth_dev_info dev_info;
1229 
1230 	RTE_ETH_FOREACH_DEV(pi) {
1231 		if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1232 			continue;
1233 
1234 		if (dev_info.rx_desc_lim.nb_min > allowed_min_rxd) {
1235 			allowed_min_rxd = dev_info.rx_desc_lim.nb_min;
1236 			*pid = pi;
1237 		}
1238 	}
1239 
1240 	return allowed_min_rxd;
1241 }
1242 
1243 /*
1244  * Check input rxd is valid or not.
1245  * If input rxd is not greater than any of maximum number
1246  * of RXDs of every Rx queues and is not less than any of
1247  * minimal number of RXDs of every Rx queues, it is valid.
1248  * if valid, return 0, else return -1
1249  */
1250 int
1251 check_nb_rxd(queueid_t rxd)
1252 {
1253 	uint16_t allowed_max_rxd;
1254 	uint16_t allowed_min_rxd;
1255 	portid_t pid = 0;
1256 
1257 	allowed_max_rxd = get_allowed_max_nb_rxd(&pid);
1258 	if (rxd > allowed_max_rxd) {
1259 		printf("Fail: input rxd (%u) can't be greater "
1260 		       "than max_rxds (%u) of port %u\n",
1261 		       rxd,
1262 		       allowed_max_rxd,
1263 		       pid);
1264 		return -1;
1265 	}
1266 
1267 	allowed_min_rxd = get_allowed_min_nb_rxd(&pid);
1268 	if (rxd < allowed_min_rxd) {
1269 		printf("Fail: input rxd (%u) can't be less "
1270 		       "than min_rxds (%u) of port %u\n",
1271 		       rxd,
1272 		       allowed_min_rxd,
1273 		       pid);
1274 		return -1;
1275 	}
1276 
1277 	return 0;
1278 }
1279 
1280 /*
1281  * Get the allowed maximum number of TXDs of every rx queues.
1282  * *pid return the port id which has minimal value of
1283  * max_txd in every tx queue.
1284  */
1285 static uint16_t
1286 get_allowed_max_nb_txd(portid_t *pid)
1287 {
1288 	uint16_t allowed_max_txd = UINT16_MAX;
1289 	portid_t pi;
1290 	struct rte_eth_dev_info dev_info;
1291 
1292 	RTE_ETH_FOREACH_DEV(pi) {
1293 		if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1294 			continue;
1295 
1296 		if (dev_info.tx_desc_lim.nb_max < allowed_max_txd) {
1297 			allowed_max_txd = dev_info.tx_desc_lim.nb_max;
1298 			*pid = pi;
1299 		}
1300 	}
1301 	return allowed_max_txd;
1302 }
1303 
1304 /*
1305  * Get the allowed maximum number of TXDs of every tx queues.
1306  * *pid return the port id which has minimal value of
1307  * min_txd in every tx queue.
1308  */
1309 static uint16_t
1310 get_allowed_min_nb_txd(portid_t *pid)
1311 {
1312 	uint16_t allowed_min_txd = 0;
1313 	portid_t pi;
1314 	struct rte_eth_dev_info dev_info;
1315 
1316 	RTE_ETH_FOREACH_DEV(pi) {
1317 		if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1318 			continue;
1319 
1320 		if (dev_info.tx_desc_lim.nb_min > allowed_min_txd) {
1321 			allowed_min_txd = dev_info.tx_desc_lim.nb_min;
1322 			*pid = pi;
1323 		}
1324 	}
1325 
1326 	return allowed_min_txd;
1327 }
1328 
1329 /*
1330  * Check input txd is valid or not.
1331  * If input txd is not greater than any of maximum number
1332  * of TXDs of every Rx queues, it is valid.
1333  * if valid, return 0, else return -1
1334  */
1335 int
1336 check_nb_txd(queueid_t txd)
1337 {
1338 	uint16_t allowed_max_txd;
1339 	uint16_t allowed_min_txd;
1340 	portid_t pid = 0;
1341 
1342 	allowed_max_txd = get_allowed_max_nb_txd(&pid);
1343 	if (txd > allowed_max_txd) {
1344 		printf("Fail: input txd (%u) can't be greater "
1345 		       "than max_txds (%u) of port %u\n",
1346 		       txd,
1347 		       allowed_max_txd,
1348 		       pid);
1349 		return -1;
1350 	}
1351 
1352 	allowed_min_txd = get_allowed_min_nb_txd(&pid);
1353 	if (txd < allowed_min_txd) {
1354 		printf("Fail: input txd (%u) can't be less "
1355 		       "than min_txds (%u) of port %u\n",
1356 		       txd,
1357 		       allowed_min_txd,
1358 		       pid);
1359 		return -1;
1360 	}
1361 	return 0;
1362 }
1363 
1364 
1365 /*
1366  * Get the allowed maximum number of hairpin queues.
1367  * *pid return the port id which has minimal value of
1368  * max_hairpin_queues in all ports.
1369  */
1370 queueid_t
1371 get_allowed_max_nb_hairpinq(portid_t *pid)
1372 {
1373 	queueid_t allowed_max_hairpinq = RTE_MAX_QUEUES_PER_PORT;
1374 	portid_t pi;
1375 	struct rte_eth_hairpin_cap cap;
1376 
1377 	RTE_ETH_FOREACH_DEV(pi) {
1378 		if (rte_eth_dev_hairpin_capability_get(pi, &cap) != 0) {
1379 			*pid = pi;
1380 			return 0;
1381 		}
1382 		if (cap.max_nb_queues < allowed_max_hairpinq) {
1383 			allowed_max_hairpinq = cap.max_nb_queues;
1384 			*pid = pi;
1385 		}
1386 	}
1387 	return allowed_max_hairpinq;
1388 }
1389 
1390 /*
1391  * Check input hairpin is valid or not.
1392  * If input hairpin is not greater than any of maximum number
1393  * of hairpin queues of all ports, it is valid.
1394  * if valid, return 0, else return -1
1395  */
1396 int
1397 check_nb_hairpinq(queueid_t hairpinq)
1398 {
1399 	queueid_t allowed_max_hairpinq;
1400 	portid_t pid = 0;
1401 
1402 	allowed_max_hairpinq = get_allowed_max_nb_hairpinq(&pid);
1403 	if (hairpinq > allowed_max_hairpinq) {
1404 		printf("Fail: input hairpin (%u) can't be greater "
1405 		       "than max_hairpin_queues (%u) of port %u\n",
1406 		       hairpinq, allowed_max_hairpinq, pid);
1407 		return -1;
1408 	}
1409 	return 0;
1410 }
1411 
1412 static void
1413 init_config(void)
1414 {
1415 	portid_t pid;
1416 	struct rte_port *port;
1417 	struct rte_mempool *mbp;
1418 	unsigned int nb_mbuf_per_pool;
1419 	lcoreid_t  lc_id;
1420 	uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
1421 	struct rte_gro_param gro_param;
1422 	uint32_t gso_types;
1423 	uint16_t data_size;
1424 	bool warning = 0;
1425 	int k;
1426 	int ret;
1427 
1428 	memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
1429 
1430 	/* Configuration of logical cores. */
1431 	fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1432 				sizeof(struct fwd_lcore *) * nb_lcores,
1433 				RTE_CACHE_LINE_SIZE);
1434 	if (fwd_lcores == NULL) {
1435 		rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1436 							"failed\n", nb_lcores);
1437 	}
1438 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1439 		fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1440 					       sizeof(struct fwd_lcore),
1441 					       RTE_CACHE_LINE_SIZE);
1442 		if (fwd_lcores[lc_id] == NULL) {
1443 			rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1444 								"failed\n");
1445 		}
1446 		fwd_lcores[lc_id]->cpuid_idx = lc_id;
1447 	}
1448 
1449 	RTE_ETH_FOREACH_DEV(pid) {
1450 		port = &ports[pid];
1451 		/* Apply default TxRx configuration for all ports */
1452 		port->dev_conf.txmode = tx_mode;
1453 		port->dev_conf.rxmode = rx_mode;
1454 
1455 		ret = eth_dev_info_get_print_err(pid, &port->dev_info);
1456 		if (ret != 0)
1457 			rte_exit(EXIT_FAILURE,
1458 				 "rte_eth_dev_info_get() failed\n");
1459 
1460 		if (!(port->dev_info.tx_offload_capa &
1461 		      DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1462 			port->dev_conf.txmode.offloads &=
1463 				~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1464 		if (numa_support) {
1465 			if (port_numa[pid] != NUMA_NO_CONFIG)
1466 				port_per_socket[port_numa[pid]]++;
1467 			else {
1468 				uint32_t socket_id = rte_eth_dev_socket_id(pid);
1469 
1470 				/*
1471 				 * if socket_id is invalid,
1472 				 * set to the first available socket.
1473 				 */
1474 				if (check_socket_id(socket_id) < 0)
1475 					socket_id = socket_ids[0];
1476 				port_per_socket[socket_id]++;
1477 			}
1478 		}
1479 
1480 		/* Apply Rx offloads configuration */
1481 		for (k = 0; k < port->dev_info.max_rx_queues; k++)
1482 			port->rx_conf[k].offloads =
1483 				port->dev_conf.rxmode.offloads;
1484 		/* Apply Tx offloads configuration */
1485 		for (k = 0; k < port->dev_info.max_tx_queues; k++)
1486 			port->tx_conf[k].offloads =
1487 				port->dev_conf.txmode.offloads;
1488 
1489 		/* set flag to initialize port/queue */
1490 		port->need_reconfig = 1;
1491 		port->need_reconfig_queues = 1;
1492 		port->tx_metadata = 0;
1493 
1494 		/* Check for maximum number of segments per MTU. Accordingly
1495 		 * update the mbuf data size.
1496 		 */
1497 		if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
1498 				port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
1499 			data_size = rx_mode.max_rx_pkt_len /
1500 				port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1501 
1502 			if ((data_size + RTE_PKTMBUF_HEADROOM) >
1503 							mbuf_data_size[0]) {
1504 				mbuf_data_size[0] = data_size +
1505 						 RTE_PKTMBUF_HEADROOM;
1506 				warning = 1;
1507 			}
1508 		}
1509 	}
1510 
1511 	if (warning)
1512 		TESTPMD_LOG(WARNING,
1513 			    "Configured mbuf size of the first segment %hu\n",
1514 			    mbuf_data_size[0]);
1515 	/*
1516 	 * Create pools of mbuf.
1517 	 * If NUMA support is disabled, create a single pool of mbuf in
1518 	 * socket 0 memory by default.
1519 	 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1520 	 *
1521 	 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1522 	 * nb_txd can be configured at run time.
1523 	 */
1524 	if (param_total_num_mbufs)
1525 		nb_mbuf_per_pool = param_total_num_mbufs;
1526 	else {
1527 		nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1528 			(nb_lcores * mb_mempool_cache) +
1529 			RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1530 		nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1531 	}
1532 
1533 	if (numa_support) {
1534 		uint8_t i, j;
1535 
1536 		for (i = 0; i < num_sockets; i++)
1537 			for (j = 0; j < mbuf_data_size_n; j++)
1538 				mempools[i * MAX_SEGS_BUFFER_SPLIT + j] =
1539 					mbuf_pool_create(mbuf_data_size[j],
1540 							  nb_mbuf_per_pool,
1541 							  socket_ids[i], j);
1542 	} else {
1543 		uint8_t i;
1544 
1545 		for (i = 0; i < mbuf_data_size_n; i++)
1546 			mempools[i] = mbuf_pool_create
1547 					(mbuf_data_size[i],
1548 					 nb_mbuf_per_pool,
1549 					 socket_num == UMA_NO_CONFIG ?
1550 					 0 : socket_num, i);
1551 	}
1552 
1553 	init_port_config();
1554 
1555 	gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1556 		DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1557 	/*
1558 	 * Records which Mbuf pool to use by each logical core, if needed.
1559 	 */
1560 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1561 		mbp = mbuf_pool_find(
1562 			rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]), 0);
1563 
1564 		if (mbp == NULL)
1565 			mbp = mbuf_pool_find(0, 0);
1566 		fwd_lcores[lc_id]->mbp = mbp;
1567 		/* initialize GSO context */
1568 		fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1569 		fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1570 		fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1571 		fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1572 			RTE_ETHER_CRC_LEN;
1573 		fwd_lcores[lc_id]->gso_ctx.flag = 0;
1574 	}
1575 
1576 	/* Configuration of packet forwarding streams. */
1577 	if (init_fwd_streams() < 0)
1578 		rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
1579 
1580 	fwd_config_setup();
1581 
1582 	/* create a gro context for each lcore */
1583 	gro_param.gro_types = RTE_GRO_TCP_IPV4;
1584 	gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1585 	gro_param.max_item_per_flow = MAX_PKT_BURST;
1586 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1587 		gro_param.socket_id = rte_lcore_to_socket_id(
1588 				fwd_lcores_cpuids[lc_id]);
1589 		fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1590 		if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1591 			rte_exit(EXIT_FAILURE,
1592 					"rte_gro_ctx_create() failed\n");
1593 		}
1594 	}
1595 }
1596 
1597 
1598 void
1599 reconfig(portid_t new_port_id, unsigned socket_id)
1600 {
1601 	struct rte_port *port;
1602 	int ret;
1603 
1604 	/* Reconfiguration of Ethernet ports. */
1605 	port = &ports[new_port_id];
1606 
1607 	ret = eth_dev_info_get_print_err(new_port_id, &port->dev_info);
1608 	if (ret != 0)
1609 		return;
1610 
1611 	/* set flag to initialize port/queue */
1612 	port->need_reconfig = 1;
1613 	port->need_reconfig_queues = 1;
1614 	port->socket_id = socket_id;
1615 
1616 	init_port_config();
1617 }
1618 
1619 
1620 int
1621 init_fwd_streams(void)
1622 {
1623 	portid_t pid;
1624 	struct rte_port *port;
1625 	streamid_t sm_id, nb_fwd_streams_new;
1626 	queueid_t q;
1627 
1628 	/* set socket id according to numa or not */
1629 	RTE_ETH_FOREACH_DEV(pid) {
1630 		port = &ports[pid];
1631 		if (nb_rxq > port->dev_info.max_rx_queues) {
1632 			printf("Fail: nb_rxq(%d) is greater than "
1633 				"max_rx_queues(%d)\n", nb_rxq,
1634 				port->dev_info.max_rx_queues);
1635 			return -1;
1636 		}
1637 		if (nb_txq > port->dev_info.max_tx_queues) {
1638 			printf("Fail: nb_txq(%d) is greater than "
1639 				"max_tx_queues(%d)\n", nb_txq,
1640 				port->dev_info.max_tx_queues);
1641 			return -1;
1642 		}
1643 		if (numa_support) {
1644 			if (port_numa[pid] != NUMA_NO_CONFIG)
1645 				port->socket_id = port_numa[pid];
1646 			else {
1647 				port->socket_id = rte_eth_dev_socket_id(pid);
1648 
1649 				/*
1650 				 * if socket_id is invalid,
1651 				 * set to the first available socket.
1652 				 */
1653 				if (check_socket_id(port->socket_id) < 0)
1654 					port->socket_id = socket_ids[0];
1655 			}
1656 		}
1657 		else {
1658 			if (socket_num == UMA_NO_CONFIG)
1659 				port->socket_id = 0;
1660 			else
1661 				port->socket_id = socket_num;
1662 		}
1663 	}
1664 
1665 	q = RTE_MAX(nb_rxq, nb_txq);
1666 	if (q == 0) {
1667 		printf("Fail: Cannot allocate fwd streams as number of queues is 0\n");
1668 		return -1;
1669 	}
1670 	nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1671 	if (nb_fwd_streams_new == nb_fwd_streams)
1672 		return 0;
1673 	/* clear the old */
1674 	if (fwd_streams != NULL) {
1675 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1676 			if (fwd_streams[sm_id] == NULL)
1677 				continue;
1678 			rte_free(fwd_streams[sm_id]);
1679 			fwd_streams[sm_id] = NULL;
1680 		}
1681 		rte_free(fwd_streams);
1682 		fwd_streams = NULL;
1683 	}
1684 
1685 	/* init new */
1686 	nb_fwd_streams = nb_fwd_streams_new;
1687 	if (nb_fwd_streams) {
1688 		fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1689 			sizeof(struct fwd_stream *) * nb_fwd_streams,
1690 			RTE_CACHE_LINE_SIZE);
1691 		if (fwd_streams == NULL)
1692 			rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1693 				 " (struct fwd_stream *)) failed\n",
1694 				 nb_fwd_streams);
1695 
1696 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1697 			fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1698 				" struct fwd_stream", sizeof(struct fwd_stream),
1699 				RTE_CACHE_LINE_SIZE);
1700 			if (fwd_streams[sm_id] == NULL)
1701 				rte_exit(EXIT_FAILURE, "rte_zmalloc"
1702 					 "(struct fwd_stream) failed\n");
1703 		}
1704 	}
1705 
1706 	return 0;
1707 }
1708 
1709 static void
1710 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1711 {
1712 	uint64_t total_burst, sburst;
1713 	uint64_t nb_burst;
1714 	uint64_t burst_stats[4];
1715 	uint16_t pktnb_stats[4];
1716 	uint16_t nb_pkt;
1717 	int burst_percent[4], sburstp;
1718 	int i;
1719 
1720 	/*
1721 	 * First compute the total number of packet bursts and the
1722 	 * two highest numbers of bursts of the same number of packets.
1723 	 */
1724 	memset(&burst_stats, 0x0, sizeof(burst_stats));
1725 	memset(&pktnb_stats, 0x0, sizeof(pktnb_stats));
1726 
1727 	/* Show stats for 0 burst size always */
1728 	total_burst = pbs->pkt_burst_spread[0];
1729 	burst_stats[0] = pbs->pkt_burst_spread[0];
1730 	pktnb_stats[0] = 0;
1731 
1732 	/* Find the next 2 burst sizes with highest occurrences. */
1733 	for (nb_pkt = 1; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1734 		nb_burst = pbs->pkt_burst_spread[nb_pkt];
1735 
1736 		if (nb_burst == 0)
1737 			continue;
1738 
1739 		total_burst += nb_burst;
1740 
1741 		if (nb_burst > burst_stats[1]) {
1742 			burst_stats[2] = burst_stats[1];
1743 			pktnb_stats[2] = pktnb_stats[1];
1744 			burst_stats[1] = nb_burst;
1745 			pktnb_stats[1] = nb_pkt;
1746 		} else if (nb_burst > burst_stats[2]) {
1747 			burst_stats[2] = nb_burst;
1748 			pktnb_stats[2] = nb_pkt;
1749 		}
1750 	}
1751 	if (total_burst == 0)
1752 		return;
1753 
1754 	printf("  %s-bursts : %"PRIu64" [", rx_tx, total_burst);
1755 	for (i = 0, sburst = 0, sburstp = 0; i < 4; i++) {
1756 		if (i == 3) {
1757 			printf("%d%% of other]\n", 100 - sburstp);
1758 			return;
1759 		}
1760 
1761 		sburst += burst_stats[i];
1762 		if (sburst == total_burst) {
1763 			printf("%d%% of %d pkts]\n",
1764 				100 - sburstp, (int) pktnb_stats[i]);
1765 			return;
1766 		}
1767 
1768 		burst_percent[i] =
1769 			(double)burst_stats[i] / total_burst * 100;
1770 		printf("%d%% of %d pkts + ",
1771 			burst_percent[i], (int) pktnb_stats[i]);
1772 		sburstp += burst_percent[i];
1773 	}
1774 }
1775 
1776 static void
1777 fwd_stream_stats_display(streamid_t stream_id)
1778 {
1779 	struct fwd_stream *fs;
1780 	static const char *fwd_top_stats_border = "-------";
1781 
1782 	fs = fwd_streams[stream_id];
1783 	if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1784 	    (fs->fwd_dropped == 0))
1785 		return;
1786 	printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1787 	       "TX Port=%2d/Queue=%2d %s\n",
1788 	       fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1789 	       fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1790 	printf("  RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1791 	       " TX-dropped: %-14"PRIu64,
1792 	       fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1793 
1794 	/* if checksum mode */
1795 	if (cur_fwd_eng == &csum_fwd_engine) {
1796 		printf("  RX- bad IP checksum: %-14"PRIu64
1797 		       "  Rx- bad L4 checksum: %-14"PRIu64
1798 		       " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1799 			fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1800 			fs->rx_bad_outer_l4_csum);
1801 	} else {
1802 		printf("\n");
1803 	}
1804 
1805 	if (record_burst_stats) {
1806 		pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1807 		pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1808 	}
1809 }
1810 
1811 void
1812 fwd_stats_display(void)
1813 {
1814 	static const char *fwd_stats_border = "----------------------";
1815 	static const char *acc_stats_border = "+++++++++++++++";
1816 	struct {
1817 		struct fwd_stream *rx_stream;
1818 		struct fwd_stream *tx_stream;
1819 		uint64_t tx_dropped;
1820 		uint64_t rx_bad_ip_csum;
1821 		uint64_t rx_bad_l4_csum;
1822 		uint64_t rx_bad_outer_l4_csum;
1823 	} ports_stats[RTE_MAX_ETHPORTS];
1824 	uint64_t total_rx_dropped = 0;
1825 	uint64_t total_tx_dropped = 0;
1826 	uint64_t total_rx_nombuf = 0;
1827 	struct rte_eth_stats stats;
1828 	uint64_t fwd_cycles = 0;
1829 	uint64_t total_recv = 0;
1830 	uint64_t total_xmit = 0;
1831 	struct rte_port *port;
1832 	streamid_t sm_id;
1833 	portid_t pt_id;
1834 	int i;
1835 
1836 	memset(ports_stats, 0, sizeof(ports_stats));
1837 
1838 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1839 		struct fwd_stream *fs = fwd_streams[sm_id];
1840 
1841 		if (cur_fwd_config.nb_fwd_streams >
1842 		    cur_fwd_config.nb_fwd_ports) {
1843 			fwd_stream_stats_display(sm_id);
1844 		} else {
1845 			ports_stats[fs->tx_port].tx_stream = fs;
1846 			ports_stats[fs->rx_port].rx_stream = fs;
1847 		}
1848 
1849 		ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1850 
1851 		ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1852 		ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1853 		ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1854 				fs->rx_bad_outer_l4_csum;
1855 
1856 		if (record_core_cycles)
1857 			fwd_cycles += fs->core_cycles;
1858 	}
1859 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1860 		uint8_t j;
1861 
1862 		pt_id = fwd_ports_ids[i];
1863 		port = &ports[pt_id];
1864 
1865 		rte_eth_stats_get(pt_id, &stats);
1866 		stats.ipackets -= port->stats.ipackets;
1867 		stats.opackets -= port->stats.opackets;
1868 		stats.ibytes -= port->stats.ibytes;
1869 		stats.obytes -= port->stats.obytes;
1870 		stats.imissed -= port->stats.imissed;
1871 		stats.oerrors -= port->stats.oerrors;
1872 		stats.rx_nombuf -= port->stats.rx_nombuf;
1873 
1874 		total_recv += stats.ipackets;
1875 		total_xmit += stats.opackets;
1876 		total_rx_dropped += stats.imissed;
1877 		total_tx_dropped += ports_stats[pt_id].tx_dropped;
1878 		total_tx_dropped += stats.oerrors;
1879 		total_rx_nombuf  += stats.rx_nombuf;
1880 
1881 		printf("\n  %s Forward statistics for port %-2d %s\n",
1882 		       fwd_stats_border, pt_id, fwd_stats_border);
1883 
1884 		if (!port->rx_queue_stats_mapping_enabled &&
1885 		    !port->tx_queue_stats_mapping_enabled) {
1886 			printf("  RX-packets: %-14"PRIu64
1887 			       " RX-dropped: %-14"PRIu64
1888 			       "RX-total: %-"PRIu64"\n",
1889 			       stats.ipackets, stats.imissed,
1890 			       stats.ipackets + stats.imissed);
1891 
1892 			if (cur_fwd_eng == &csum_fwd_engine)
1893 				printf("  Bad-ipcsum: %-14"PRIu64
1894 				       " Bad-l4csum: %-14"PRIu64
1895 				       "Bad-outer-l4csum: %-14"PRIu64"\n",
1896 				       ports_stats[pt_id].rx_bad_ip_csum,
1897 				       ports_stats[pt_id].rx_bad_l4_csum,
1898 				       ports_stats[pt_id].rx_bad_outer_l4_csum);
1899 			if (stats.ierrors + stats.rx_nombuf > 0) {
1900 				printf("  RX-error: %-"PRIu64"\n",
1901 				       stats.ierrors);
1902 				printf("  RX-nombufs: %-14"PRIu64"\n",
1903 				       stats.rx_nombuf);
1904 			}
1905 
1906 			printf("  TX-packets: %-14"PRIu64
1907 			       " TX-dropped: %-14"PRIu64
1908 			       "TX-total: %-"PRIu64"\n",
1909 			       stats.opackets, ports_stats[pt_id].tx_dropped,
1910 			       stats.opackets + ports_stats[pt_id].tx_dropped);
1911 		} else {
1912 			printf("  RX-packets:             %14"PRIu64
1913 			       "    RX-dropped:%14"PRIu64
1914 			       "    RX-total:%14"PRIu64"\n",
1915 			       stats.ipackets, stats.imissed,
1916 			       stats.ipackets + stats.imissed);
1917 
1918 			if (cur_fwd_eng == &csum_fwd_engine)
1919 				printf("  Bad-ipcsum:%14"PRIu64
1920 				       "    Bad-l4csum:%14"PRIu64
1921 				       "    Bad-outer-l4csum: %-14"PRIu64"\n",
1922 				       ports_stats[pt_id].rx_bad_ip_csum,
1923 				       ports_stats[pt_id].rx_bad_l4_csum,
1924 				       ports_stats[pt_id].rx_bad_outer_l4_csum);
1925 			if ((stats.ierrors + stats.rx_nombuf) > 0) {
1926 				printf("  RX-error:%"PRIu64"\n", stats.ierrors);
1927 				printf("  RX-nombufs:             %14"PRIu64"\n",
1928 				       stats.rx_nombuf);
1929 			}
1930 
1931 			printf("  TX-packets:             %14"PRIu64
1932 			       "    TX-dropped:%14"PRIu64
1933 			       "    TX-total:%14"PRIu64"\n",
1934 			       stats.opackets, ports_stats[pt_id].tx_dropped,
1935 			       stats.opackets + ports_stats[pt_id].tx_dropped);
1936 		}
1937 
1938 		if (record_burst_stats) {
1939 			if (ports_stats[pt_id].rx_stream)
1940 				pkt_burst_stats_display("RX",
1941 					&ports_stats[pt_id].rx_stream->rx_burst_stats);
1942 			if (ports_stats[pt_id].tx_stream)
1943 				pkt_burst_stats_display("TX",
1944 					&ports_stats[pt_id].tx_stream->tx_burst_stats);
1945 		}
1946 
1947 		if (port->rx_queue_stats_mapping_enabled) {
1948 			printf("\n");
1949 			for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1950 				printf("  Stats reg %2d RX-packets:%14"PRIu64
1951 				       "     RX-errors:%14"PRIu64
1952 				       "    RX-bytes:%14"PRIu64"\n",
1953 				       j, stats.q_ipackets[j],
1954 				       stats.q_errors[j], stats.q_ibytes[j]);
1955 			}
1956 			printf("\n");
1957 		}
1958 		if (port->tx_queue_stats_mapping_enabled) {
1959 			for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1960 				printf("  Stats reg %2d TX-packets:%14"PRIu64
1961 				       "                                 TX-bytes:%14"
1962 				       PRIu64"\n",
1963 				       j, stats.q_opackets[j],
1964 				       stats.q_obytes[j]);
1965 			}
1966 		}
1967 
1968 		printf("  %s--------------------------------%s\n",
1969 		       fwd_stats_border, fwd_stats_border);
1970 	}
1971 
1972 	printf("\n  %s Accumulated forward statistics for all ports"
1973 	       "%s\n",
1974 	       acc_stats_border, acc_stats_border);
1975 	printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1976 	       "%-"PRIu64"\n"
1977 	       "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1978 	       "%-"PRIu64"\n",
1979 	       total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1980 	       total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1981 	if (total_rx_nombuf > 0)
1982 		printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1983 	printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
1984 	       "%s\n",
1985 	       acc_stats_border, acc_stats_border);
1986 	if (record_core_cycles) {
1987 #define CYC_PER_MHZ 1E6
1988 		if (total_recv > 0 || total_xmit > 0) {
1989 			uint64_t total_pkts = 0;
1990 			if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 ||
1991 			    strcmp(cur_fwd_eng->fwd_mode_name, "flowgen") == 0)
1992 				total_pkts = total_xmit;
1993 			else
1994 				total_pkts = total_recv;
1995 
1996 			printf("\n  CPU cycles/packet=%.2F (total cycles="
1997 			       "%"PRIu64" / total %s packets=%"PRIu64") at %"PRIu64
1998 			       " MHz Clock\n",
1999 			       (double) fwd_cycles / total_pkts,
2000 			       fwd_cycles, cur_fwd_eng->fwd_mode_name, total_pkts,
2001 			       (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
2002 		}
2003 	}
2004 }
2005 
2006 void
2007 fwd_stats_reset(void)
2008 {
2009 	streamid_t sm_id;
2010 	portid_t pt_id;
2011 	int i;
2012 
2013 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2014 		pt_id = fwd_ports_ids[i];
2015 		rte_eth_stats_get(pt_id, &ports[pt_id].stats);
2016 	}
2017 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
2018 		struct fwd_stream *fs = fwd_streams[sm_id];
2019 
2020 		fs->rx_packets = 0;
2021 		fs->tx_packets = 0;
2022 		fs->fwd_dropped = 0;
2023 		fs->rx_bad_ip_csum = 0;
2024 		fs->rx_bad_l4_csum = 0;
2025 		fs->rx_bad_outer_l4_csum = 0;
2026 
2027 		memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
2028 		memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
2029 		fs->core_cycles = 0;
2030 	}
2031 }
2032 
2033 static void
2034 flush_fwd_rx_queues(void)
2035 {
2036 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2037 	portid_t  rxp;
2038 	portid_t port_id;
2039 	queueid_t rxq;
2040 	uint16_t  nb_rx;
2041 	uint16_t  i;
2042 	uint8_t   j;
2043 	uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
2044 	uint64_t timer_period;
2045 
2046 	/* convert to number of cycles */
2047 	timer_period = rte_get_timer_hz(); /* 1 second timeout */
2048 
2049 	for (j = 0; j < 2; j++) {
2050 		for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
2051 			for (rxq = 0; rxq < nb_rxq; rxq++) {
2052 				port_id = fwd_ports_ids[rxp];
2053 				/**
2054 				* testpmd can stuck in the below do while loop
2055 				* if rte_eth_rx_burst() always returns nonzero
2056 				* packets. So timer is added to exit this loop
2057 				* after 1sec timer expiry.
2058 				*/
2059 				prev_tsc = rte_rdtsc();
2060 				do {
2061 					nb_rx = rte_eth_rx_burst(port_id, rxq,
2062 						pkts_burst, MAX_PKT_BURST);
2063 					for (i = 0; i < nb_rx; i++)
2064 						rte_pktmbuf_free(pkts_burst[i]);
2065 
2066 					cur_tsc = rte_rdtsc();
2067 					diff_tsc = cur_tsc - prev_tsc;
2068 					timer_tsc += diff_tsc;
2069 				} while ((nb_rx > 0) &&
2070 					(timer_tsc < timer_period));
2071 				timer_tsc = 0;
2072 			}
2073 		}
2074 		rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
2075 	}
2076 }
2077 
2078 static void
2079 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
2080 {
2081 	struct fwd_stream **fsm;
2082 	streamid_t nb_fs;
2083 	streamid_t sm_id;
2084 #ifdef RTE_LIBRTE_BITRATESTATS
2085 	uint64_t tics_per_1sec;
2086 	uint64_t tics_datum;
2087 	uint64_t tics_current;
2088 	uint16_t i, cnt_ports;
2089 
2090 	cnt_ports = nb_ports;
2091 	tics_datum = rte_rdtsc();
2092 	tics_per_1sec = rte_get_timer_hz();
2093 #endif
2094 	fsm = &fwd_streams[fc->stream_idx];
2095 	nb_fs = fc->stream_nb;
2096 	do {
2097 		for (sm_id = 0; sm_id < nb_fs; sm_id++)
2098 			(*pkt_fwd)(fsm[sm_id]);
2099 #ifdef RTE_LIBRTE_BITRATESTATS
2100 		if (bitrate_enabled != 0 &&
2101 				bitrate_lcore_id == rte_lcore_id()) {
2102 			tics_current = rte_rdtsc();
2103 			if (tics_current - tics_datum >= tics_per_1sec) {
2104 				/* Periodic bitrate calculation */
2105 				for (i = 0; i < cnt_ports; i++)
2106 					rte_stats_bitrate_calc(bitrate_data,
2107 						ports_ids[i]);
2108 				tics_datum = tics_current;
2109 			}
2110 		}
2111 #endif
2112 #ifdef RTE_LIBRTE_LATENCY_STATS
2113 		if (latencystats_enabled != 0 &&
2114 				latencystats_lcore_id == rte_lcore_id())
2115 			rte_latencystats_update();
2116 #endif
2117 
2118 	} while (! fc->stopped);
2119 }
2120 
2121 static int
2122 start_pkt_forward_on_core(void *fwd_arg)
2123 {
2124 	run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
2125 			     cur_fwd_config.fwd_eng->packet_fwd);
2126 	return 0;
2127 }
2128 
2129 /*
2130  * Run the TXONLY packet forwarding engine to send a single burst of packets.
2131  * Used to start communication flows in network loopback test configurations.
2132  */
2133 static int
2134 run_one_txonly_burst_on_core(void *fwd_arg)
2135 {
2136 	struct fwd_lcore *fwd_lc;
2137 	struct fwd_lcore tmp_lcore;
2138 
2139 	fwd_lc = (struct fwd_lcore *) fwd_arg;
2140 	tmp_lcore = *fwd_lc;
2141 	tmp_lcore.stopped = 1;
2142 	run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
2143 	return 0;
2144 }
2145 
2146 /*
2147  * Launch packet forwarding:
2148  *     - Setup per-port forwarding context.
2149  *     - launch logical cores with their forwarding configuration.
2150  */
2151 static void
2152 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
2153 {
2154 	port_fwd_begin_t port_fwd_begin;
2155 	unsigned int i;
2156 	unsigned int lc_id;
2157 	int diag;
2158 
2159 	port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2160 	if (port_fwd_begin != NULL) {
2161 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2162 			(*port_fwd_begin)(fwd_ports_ids[i]);
2163 	}
2164 	for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2165 		lc_id = fwd_lcores_cpuids[i];
2166 		if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2167 			fwd_lcores[i]->stopped = 0;
2168 			diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2169 						     fwd_lcores[i], lc_id);
2170 			if (diag != 0)
2171 				printf("launch lcore %u failed - diag=%d\n",
2172 				       lc_id, diag);
2173 		}
2174 	}
2175 }
2176 
2177 /*
2178  * Launch packet forwarding configuration.
2179  */
2180 void
2181 start_packet_forwarding(int with_tx_first)
2182 {
2183 	port_fwd_begin_t port_fwd_begin;
2184 	port_fwd_end_t  port_fwd_end;
2185 	struct rte_port *port;
2186 	unsigned int i;
2187 	portid_t   pt_id;
2188 
2189 	if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2190 		rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2191 
2192 	if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2193 		rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2194 
2195 	if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2196 		strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2197 		(!nb_rxq || !nb_txq))
2198 		rte_exit(EXIT_FAILURE,
2199 			"Either rxq or txq are 0, cannot use %s fwd mode\n",
2200 			cur_fwd_eng->fwd_mode_name);
2201 
2202 	if (all_ports_started() == 0) {
2203 		printf("Not all ports were started\n");
2204 		return;
2205 	}
2206 	if (test_done == 0) {
2207 		printf("Packet forwarding already started\n");
2208 		return;
2209 	}
2210 
2211 
2212 	if(dcb_test) {
2213 		for (i = 0; i < nb_fwd_ports; i++) {
2214 			pt_id = fwd_ports_ids[i];
2215 			port = &ports[pt_id];
2216 			if (!port->dcb_flag) {
2217 				printf("In DCB mode, all forwarding ports must "
2218                                        "be configured in this mode.\n");
2219 				return;
2220 			}
2221 		}
2222 		if (nb_fwd_lcores == 1) {
2223 			printf("In DCB mode,the nb forwarding cores "
2224                                "should be larger than 1.\n");
2225 			return;
2226 		}
2227 	}
2228 	test_done = 0;
2229 
2230 	fwd_config_setup();
2231 
2232 	if(!no_flush_rx)
2233 		flush_fwd_rx_queues();
2234 
2235 	pkt_fwd_config_display(&cur_fwd_config);
2236 	rxtx_config_display();
2237 
2238 	fwd_stats_reset();
2239 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2240 		pt_id = fwd_ports_ids[i];
2241 		port = &ports[pt_id];
2242 		map_port_queue_stats_mapping_registers(pt_id, port);
2243 	}
2244 	if (with_tx_first) {
2245 		port_fwd_begin = tx_only_engine.port_fwd_begin;
2246 		if (port_fwd_begin != NULL) {
2247 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2248 				(*port_fwd_begin)(fwd_ports_ids[i]);
2249 		}
2250 		while (with_tx_first--) {
2251 			launch_packet_forwarding(
2252 					run_one_txonly_burst_on_core);
2253 			rte_eal_mp_wait_lcore();
2254 		}
2255 		port_fwd_end = tx_only_engine.port_fwd_end;
2256 		if (port_fwd_end != NULL) {
2257 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2258 				(*port_fwd_end)(fwd_ports_ids[i]);
2259 		}
2260 	}
2261 	launch_packet_forwarding(start_pkt_forward_on_core);
2262 }
2263 
2264 void
2265 stop_packet_forwarding(void)
2266 {
2267 	port_fwd_end_t port_fwd_end;
2268 	lcoreid_t lc_id;
2269 	portid_t pt_id;
2270 	int i;
2271 
2272 	if (test_done) {
2273 		printf("Packet forwarding not started\n");
2274 		return;
2275 	}
2276 	printf("Telling cores to stop...");
2277 	for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2278 		fwd_lcores[lc_id]->stopped = 1;
2279 	printf("\nWaiting for lcores to finish...\n");
2280 	rte_eal_mp_wait_lcore();
2281 	port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2282 	if (port_fwd_end != NULL) {
2283 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2284 			pt_id = fwd_ports_ids[i];
2285 			(*port_fwd_end)(pt_id);
2286 		}
2287 	}
2288 
2289 	fwd_stats_display();
2290 
2291 	printf("\nDone.\n");
2292 	test_done = 1;
2293 }
2294 
2295 void
2296 dev_set_link_up(portid_t pid)
2297 {
2298 	if (rte_eth_dev_set_link_up(pid) < 0)
2299 		printf("\nSet link up fail.\n");
2300 }
2301 
2302 void
2303 dev_set_link_down(portid_t pid)
2304 {
2305 	if (rte_eth_dev_set_link_down(pid) < 0)
2306 		printf("\nSet link down fail.\n");
2307 }
2308 
2309 static int
2310 all_ports_started(void)
2311 {
2312 	portid_t pi;
2313 	struct rte_port *port;
2314 
2315 	RTE_ETH_FOREACH_DEV(pi) {
2316 		port = &ports[pi];
2317 		/* Check if there is a port which is not started */
2318 		if ((port->port_status != RTE_PORT_STARTED) &&
2319 			(port->slave_flag == 0))
2320 			return 0;
2321 	}
2322 
2323 	/* No port is not started */
2324 	return 1;
2325 }
2326 
2327 int
2328 port_is_stopped(portid_t port_id)
2329 {
2330 	struct rte_port *port = &ports[port_id];
2331 
2332 	if ((port->port_status != RTE_PORT_STOPPED) &&
2333 	    (port->slave_flag == 0))
2334 		return 0;
2335 	return 1;
2336 }
2337 
2338 int
2339 all_ports_stopped(void)
2340 {
2341 	portid_t pi;
2342 
2343 	RTE_ETH_FOREACH_DEV(pi) {
2344 		if (!port_is_stopped(pi))
2345 			return 0;
2346 	}
2347 
2348 	return 1;
2349 }
2350 
2351 int
2352 port_is_started(portid_t port_id)
2353 {
2354 	if (port_id_is_invalid(port_id, ENABLED_WARN))
2355 		return 0;
2356 
2357 	if (ports[port_id].port_status != RTE_PORT_STARTED)
2358 		return 0;
2359 
2360 	return 1;
2361 }
2362 
2363 /* Configure the Rx and Tx hairpin queues for the selected port. */
2364 static int
2365 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2366 {
2367 	queueid_t qi;
2368 	struct rte_eth_hairpin_conf hairpin_conf = {
2369 		.peer_count = 1,
2370 	};
2371 	int i;
2372 	int diag;
2373 	struct rte_port *port = &ports[pi];
2374 	uint16_t peer_rx_port = pi;
2375 	uint16_t peer_tx_port = pi;
2376 	uint32_t manual = 1;
2377 	uint32_t tx_exp = hairpin_mode & 0x10;
2378 
2379 	if (!(hairpin_mode & 0xf)) {
2380 		peer_rx_port = pi;
2381 		peer_tx_port = pi;
2382 		manual = 0;
2383 	} else if (hairpin_mode & 0x1) {
2384 		peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2385 						       RTE_ETH_DEV_NO_OWNER);
2386 		if (peer_tx_port >= RTE_MAX_ETHPORTS)
2387 			peer_tx_port = rte_eth_find_next_owned_by(0,
2388 						RTE_ETH_DEV_NO_OWNER);
2389 		if (p_pi != RTE_MAX_ETHPORTS) {
2390 			peer_rx_port = p_pi;
2391 		} else {
2392 			uint16_t next_pi;
2393 
2394 			/* Last port will be the peer RX port of the first. */
2395 			RTE_ETH_FOREACH_DEV(next_pi)
2396 				peer_rx_port = next_pi;
2397 		}
2398 		manual = 1;
2399 	} else if (hairpin_mode & 0x2) {
2400 		if (cnt_pi & 0x1) {
2401 			peer_rx_port = p_pi;
2402 		} else {
2403 			peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2404 						RTE_ETH_DEV_NO_OWNER);
2405 			if (peer_rx_port >= RTE_MAX_ETHPORTS)
2406 				peer_rx_port = pi;
2407 		}
2408 		peer_tx_port = peer_rx_port;
2409 		manual = 1;
2410 	}
2411 
2412 	for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2413 		hairpin_conf.peers[0].port = peer_rx_port;
2414 		hairpin_conf.peers[0].queue = i + nb_rxq;
2415 		hairpin_conf.manual_bind = !!manual;
2416 		hairpin_conf.tx_explicit = !!tx_exp;
2417 		diag = rte_eth_tx_hairpin_queue_setup
2418 			(pi, qi, nb_txd, &hairpin_conf);
2419 		i++;
2420 		if (diag == 0)
2421 			continue;
2422 
2423 		/* Fail to setup rx queue, return */
2424 		if (rte_atomic16_cmpset(&(port->port_status),
2425 					RTE_PORT_HANDLING,
2426 					RTE_PORT_STOPPED) == 0)
2427 			printf("Port %d can not be set back "
2428 					"to stopped\n", pi);
2429 		printf("Fail to configure port %d hairpin "
2430 				"queues\n", pi);
2431 		/* try to reconfigure queues next time */
2432 		port->need_reconfig_queues = 1;
2433 		return -1;
2434 	}
2435 	for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2436 		hairpin_conf.peers[0].port = peer_tx_port;
2437 		hairpin_conf.peers[0].queue = i + nb_txq;
2438 		hairpin_conf.manual_bind = !!manual;
2439 		hairpin_conf.tx_explicit = !!tx_exp;
2440 		diag = rte_eth_rx_hairpin_queue_setup
2441 			(pi, qi, nb_rxd, &hairpin_conf);
2442 		i++;
2443 		if (diag == 0)
2444 			continue;
2445 
2446 		/* Fail to setup rx queue, return */
2447 		if (rte_atomic16_cmpset(&(port->port_status),
2448 					RTE_PORT_HANDLING,
2449 					RTE_PORT_STOPPED) == 0)
2450 			printf("Port %d can not be set back "
2451 					"to stopped\n", pi);
2452 		printf("Fail to configure port %d hairpin "
2453 				"queues\n", pi);
2454 		/* try to reconfigure queues next time */
2455 		port->need_reconfig_queues = 1;
2456 		return -1;
2457 	}
2458 	return 0;
2459 }
2460 
2461 /* Configure the Rx with optional split. */
2462 int
2463 rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
2464 	       uint16_t nb_rx_desc, unsigned int socket_id,
2465 	       struct rte_eth_rxconf *rx_conf, struct rte_mempool *mp)
2466 {
2467 	union rte_eth_rxseg rx_useg[MAX_SEGS_BUFFER_SPLIT] = {};
2468 	unsigned int i, mp_n;
2469 	int ret;
2470 
2471 	if (rx_pkt_nb_segs <= 1 ||
2472 	    (rx_conf->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) == 0) {
2473 		rx_conf->rx_seg = NULL;
2474 		rx_conf->rx_nseg = 0;
2475 		ret = rte_eth_rx_queue_setup(port_id, rx_queue_id,
2476 					     nb_rx_desc, socket_id,
2477 					     rx_conf, mp);
2478 		return ret;
2479 	}
2480 	for (i = 0; i < rx_pkt_nb_segs; i++) {
2481 		struct rte_eth_rxseg_split *rx_seg = &rx_useg[i].split;
2482 		struct rte_mempool *mpx;
2483 		/*
2484 		 * Use last valid pool for the segments with number
2485 		 * exceeding the pool index.
2486 		 */
2487 		mp_n = (i > mbuf_data_size_n) ? mbuf_data_size_n - 1 : i;
2488 		mpx = mbuf_pool_find(socket_id, mp_n);
2489 		/* Handle zero as mbuf data buffer size. */
2490 		rx_seg->length = rx_pkt_seg_lengths[i] ?
2491 				   rx_pkt_seg_lengths[i] :
2492 				   mbuf_data_size[mp_n];
2493 		rx_seg->offset = i < rx_pkt_nb_offs ?
2494 				   rx_pkt_seg_offsets[i] : 0;
2495 		rx_seg->mp = mpx ? mpx : mp;
2496 	}
2497 	rx_conf->rx_nseg = rx_pkt_nb_segs;
2498 	rx_conf->rx_seg = rx_useg;
2499 	ret = rte_eth_rx_queue_setup(port_id, rx_queue_id, nb_rx_desc,
2500 				    socket_id, rx_conf, NULL);
2501 	rx_conf->rx_seg = NULL;
2502 	rx_conf->rx_nseg = 0;
2503 	return ret;
2504 }
2505 
2506 int
2507 start_port(portid_t pid)
2508 {
2509 	int diag, need_check_link_status = -1;
2510 	portid_t pi;
2511 	portid_t p_pi = RTE_MAX_ETHPORTS;
2512 	portid_t pl[RTE_MAX_ETHPORTS];
2513 	portid_t peer_pl[RTE_MAX_ETHPORTS];
2514 	uint16_t cnt_pi = 0;
2515 	uint16_t cfg_pi = 0;
2516 	int peer_pi;
2517 	queueid_t qi;
2518 	struct rte_port *port;
2519 	struct rte_ether_addr mac_addr;
2520 	struct rte_eth_hairpin_cap cap;
2521 
2522 	if (port_id_is_invalid(pid, ENABLED_WARN))
2523 		return 0;
2524 
2525 	if(dcb_config)
2526 		dcb_test = 1;
2527 	RTE_ETH_FOREACH_DEV(pi) {
2528 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2529 			continue;
2530 
2531 		need_check_link_status = 0;
2532 		port = &ports[pi];
2533 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2534 						 RTE_PORT_HANDLING) == 0) {
2535 			printf("Port %d is now not stopped\n", pi);
2536 			continue;
2537 		}
2538 
2539 		if (port->need_reconfig > 0) {
2540 			port->need_reconfig = 0;
2541 
2542 			if (flow_isolate_all) {
2543 				int ret = port_flow_isolate(pi, 1);
2544 				if (ret) {
2545 					printf("Failed to apply isolated"
2546 					       " mode on port %d\n", pi);
2547 					return -1;
2548 				}
2549 			}
2550 			configure_rxtx_dump_callbacks(0);
2551 			printf("Configuring Port %d (socket %u)\n", pi,
2552 					port->socket_id);
2553 			if (nb_hairpinq > 0 &&
2554 			    rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2555 				printf("Port %d doesn't support hairpin "
2556 				       "queues\n", pi);
2557 				return -1;
2558 			}
2559 			/* configure port */
2560 			diag = rte_eth_dev_configure(pi, nb_rxq + nb_hairpinq,
2561 						     nb_txq + nb_hairpinq,
2562 						     &(port->dev_conf));
2563 			if (diag != 0) {
2564 				if (rte_atomic16_cmpset(&(port->port_status),
2565 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2566 					printf("Port %d can not be set back "
2567 							"to stopped\n", pi);
2568 				printf("Fail to configure port %d\n", pi);
2569 				/* try to reconfigure port next time */
2570 				port->need_reconfig = 1;
2571 				return -1;
2572 			}
2573 		}
2574 		if (port->need_reconfig_queues > 0) {
2575 			port->need_reconfig_queues = 0;
2576 			/* setup tx queues */
2577 			for (qi = 0; qi < nb_txq; qi++) {
2578 				if ((numa_support) &&
2579 					(txring_numa[pi] != NUMA_NO_CONFIG))
2580 					diag = rte_eth_tx_queue_setup(pi, qi,
2581 						port->nb_tx_desc[qi],
2582 						txring_numa[pi],
2583 						&(port->tx_conf[qi]));
2584 				else
2585 					diag = rte_eth_tx_queue_setup(pi, qi,
2586 						port->nb_tx_desc[qi],
2587 						port->socket_id,
2588 						&(port->tx_conf[qi]));
2589 
2590 				if (diag == 0)
2591 					continue;
2592 
2593 				/* Fail to setup tx queue, return */
2594 				if (rte_atomic16_cmpset(&(port->port_status),
2595 							RTE_PORT_HANDLING,
2596 							RTE_PORT_STOPPED) == 0)
2597 					printf("Port %d can not be set back "
2598 							"to stopped\n", pi);
2599 				printf("Fail to configure port %d tx queues\n",
2600 				       pi);
2601 				/* try to reconfigure queues next time */
2602 				port->need_reconfig_queues = 1;
2603 				return -1;
2604 			}
2605 			for (qi = 0; qi < nb_rxq; qi++) {
2606 				/* setup rx queues */
2607 				if ((numa_support) &&
2608 					(rxring_numa[pi] != NUMA_NO_CONFIG)) {
2609 					struct rte_mempool * mp =
2610 						mbuf_pool_find
2611 							(rxring_numa[pi], 0);
2612 					if (mp == NULL) {
2613 						printf("Failed to setup RX queue:"
2614 							"No mempool allocation"
2615 							" on the socket %d\n",
2616 							rxring_numa[pi]);
2617 						return -1;
2618 					}
2619 
2620 					diag = rx_queue_setup(pi, qi,
2621 					     port->nb_rx_desc[qi],
2622 					     rxring_numa[pi],
2623 					     &(port->rx_conf[qi]),
2624 					     mp);
2625 				} else {
2626 					struct rte_mempool *mp =
2627 						mbuf_pool_find
2628 							(port->socket_id, 0);
2629 					if (mp == NULL) {
2630 						printf("Failed to setup RX queue:"
2631 							"No mempool allocation"
2632 							" on the socket %d\n",
2633 							port->socket_id);
2634 						return -1;
2635 					}
2636 					diag = rx_queue_setup(pi, qi,
2637 					     port->nb_rx_desc[qi],
2638 					     port->socket_id,
2639 					     &(port->rx_conf[qi]),
2640 					     mp);
2641 				}
2642 				if (diag == 0)
2643 					continue;
2644 
2645 				/* Fail to setup rx queue, return */
2646 				if (rte_atomic16_cmpset(&(port->port_status),
2647 							RTE_PORT_HANDLING,
2648 							RTE_PORT_STOPPED) == 0)
2649 					printf("Port %d can not be set back "
2650 							"to stopped\n", pi);
2651 				printf("Fail to configure port %d rx queues\n",
2652 				       pi);
2653 				/* try to reconfigure queues next time */
2654 				port->need_reconfig_queues = 1;
2655 				return -1;
2656 			}
2657 			/* setup hairpin queues */
2658 			if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2659 				return -1;
2660 		}
2661 		configure_rxtx_dump_callbacks(verbose_level);
2662 		if (clear_ptypes) {
2663 			diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2664 					NULL, 0);
2665 			if (diag < 0)
2666 				printf(
2667 				"Port %d: Failed to disable Ptype parsing\n",
2668 				pi);
2669 		}
2670 
2671 		p_pi = pi;
2672 		cnt_pi++;
2673 
2674 		/* start port */
2675 		if (rte_eth_dev_start(pi) < 0) {
2676 			printf("Fail to start port %d\n", pi);
2677 
2678 			/* Fail to setup rx queue, return */
2679 			if (rte_atomic16_cmpset(&(port->port_status),
2680 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2681 				printf("Port %d can not be set back to "
2682 							"stopped\n", pi);
2683 			continue;
2684 		}
2685 
2686 		if (rte_atomic16_cmpset(&(port->port_status),
2687 			RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2688 			printf("Port %d can not be set into started\n", pi);
2689 
2690 		if (eth_macaddr_get_print_err(pi, &mac_addr) == 0)
2691 			printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2692 				mac_addr.addr_bytes[0], mac_addr.addr_bytes[1],
2693 				mac_addr.addr_bytes[2], mac_addr.addr_bytes[3],
2694 				mac_addr.addr_bytes[4], mac_addr.addr_bytes[5]);
2695 
2696 		/* at least one port started, need checking link status */
2697 		need_check_link_status = 1;
2698 
2699 		pl[cfg_pi++] = pi;
2700 	}
2701 
2702 	if (need_check_link_status == 1 && !no_link_check)
2703 		check_all_ports_link_status(RTE_PORT_ALL);
2704 	else if (need_check_link_status == 0)
2705 		printf("Please stop the ports first\n");
2706 
2707 	if (hairpin_mode & 0xf) {
2708 		uint16_t i;
2709 		int j;
2710 
2711 		/* bind all started hairpin ports */
2712 		for (i = 0; i < cfg_pi; i++) {
2713 			pi = pl[i];
2714 			/* bind current Tx to all peer Rx */
2715 			peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2716 							RTE_MAX_ETHPORTS, 1);
2717 			if (peer_pi < 0)
2718 				return peer_pi;
2719 			for (j = 0; j < peer_pi; j++) {
2720 				if (!port_is_started(peer_pl[j]))
2721 					continue;
2722 				diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2723 				if (diag < 0) {
2724 					printf("Error during binding hairpin"
2725 					       " Tx port %u to %u: %s\n",
2726 					       pi, peer_pl[j],
2727 					       rte_strerror(-diag));
2728 					return -1;
2729 				}
2730 			}
2731 			/* bind all peer Tx to current Rx */
2732 			peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2733 							RTE_MAX_ETHPORTS, 0);
2734 			if (peer_pi < 0)
2735 				return peer_pi;
2736 			for (j = 0; j < peer_pi; j++) {
2737 				if (!port_is_started(peer_pl[j]))
2738 					continue;
2739 				diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2740 				if (diag < 0) {
2741 					printf("Error during binding hairpin"
2742 					       " Tx port %u to %u: %s\n",
2743 					       peer_pl[j], pi,
2744 					       rte_strerror(-diag));
2745 					return -1;
2746 				}
2747 			}
2748 		}
2749 	}
2750 
2751 	printf("Done\n");
2752 	return 0;
2753 }
2754 
2755 void
2756 stop_port(portid_t pid)
2757 {
2758 	portid_t pi;
2759 	struct rte_port *port;
2760 	int need_check_link_status = 0;
2761 	portid_t peer_pl[RTE_MAX_ETHPORTS];
2762 	int peer_pi;
2763 
2764 	if (dcb_test) {
2765 		dcb_test = 0;
2766 		dcb_config = 0;
2767 	}
2768 
2769 	if (port_id_is_invalid(pid, ENABLED_WARN))
2770 		return;
2771 
2772 	printf("Stopping ports...\n");
2773 
2774 	RTE_ETH_FOREACH_DEV(pi) {
2775 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2776 			continue;
2777 
2778 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2779 			printf("Please remove port %d from forwarding configuration.\n", pi);
2780 			continue;
2781 		}
2782 
2783 		if (port_is_bonding_slave(pi)) {
2784 			printf("Please remove port %d from bonded device.\n", pi);
2785 			continue;
2786 		}
2787 
2788 		port = &ports[pi];
2789 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2790 						RTE_PORT_HANDLING) == 0)
2791 			continue;
2792 
2793 		if (hairpin_mode & 0xf) {
2794 			int j;
2795 
2796 			rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
2797 			/* unbind all peer Tx from current Rx */
2798 			peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2799 							RTE_MAX_ETHPORTS, 0);
2800 			if (peer_pi < 0)
2801 				continue;
2802 			for (j = 0; j < peer_pi; j++) {
2803 				if (!port_is_started(peer_pl[j]))
2804 					continue;
2805 				rte_eth_hairpin_unbind(peer_pl[j], pi);
2806 			}
2807 		}
2808 
2809 		rte_eth_dev_stop(pi);
2810 
2811 		if (rte_atomic16_cmpset(&(port->port_status),
2812 			RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2813 			printf("Port %d can not be set into stopped\n", pi);
2814 		need_check_link_status = 1;
2815 	}
2816 	if (need_check_link_status && !no_link_check)
2817 		check_all_ports_link_status(RTE_PORT_ALL);
2818 
2819 	printf("Done\n");
2820 }
2821 
2822 static void
2823 remove_invalid_ports_in(portid_t *array, portid_t *total)
2824 {
2825 	portid_t i;
2826 	portid_t new_total = 0;
2827 
2828 	for (i = 0; i < *total; i++)
2829 		if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2830 			array[new_total] = array[i];
2831 			new_total++;
2832 		}
2833 	*total = new_total;
2834 }
2835 
2836 static void
2837 remove_invalid_ports(void)
2838 {
2839 	remove_invalid_ports_in(ports_ids, &nb_ports);
2840 	remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2841 	nb_cfg_ports = nb_fwd_ports;
2842 }
2843 
2844 void
2845 close_port(portid_t pid)
2846 {
2847 	portid_t pi;
2848 	struct rte_port *port;
2849 
2850 	if (port_id_is_invalid(pid, ENABLED_WARN))
2851 		return;
2852 
2853 	printf("Closing ports...\n");
2854 
2855 	RTE_ETH_FOREACH_DEV(pi) {
2856 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2857 			continue;
2858 
2859 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2860 			printf("Please remove port %d from forwarding configuration.\n", pi);
2861 			continue;
2862 		}
2863 
2864 		if (port_is_bonding_slave(pi)) {
2865 			printf("Please remove port %d from bonded device.\n", pi);
2866 			continue;
2867 		}
2868 
2869 		port = &ports[pi];
2870 		if (rte_atomic16_cmpset(&(port->port_status),
2871 			RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2872 			printf("Port %d is already closed\n", pi);
2873 			continue;
2874 		}
2875 
2876 		port_flow_flush(pi);
2877 		rte_eth_dev_close(pi);
2878 	}
2879 
2880 	remove_invalid_ports();
2881 	printf("Done\n");
2882 }
2883 
2884 void
2885 reset_port(portid_t pid)
2886 {
2887 	int diag;
2888 	portid_t pi;
2889 	struct rte_port *port;
2890 
2891 	if (port_id_is_invalid(pid, ENABLED_WARN))
2892 		return;
2893 
2894 	if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
2895 		(pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
2896 		printf("Can not reset port(s), please stop port(s) first.\n");
2897 		return;
2898 	}
2899 
2900 	printf("Resetting ports...\n");
2901 
2902 	RTE_ETH_FOREACH_DEV(pi) {
2903 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2904 			continue;
2905 
2906 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2907 			printf("Please remove port %d from forwarding "
2908 			       "configuration.\n", pi);
2909 			continue;
2910 		}
2911 
2912 		if (port_is_bonding_slave(pi)) {
2913 			printf("Please remove port %d from bonded device.\n",
2914 			       pi);
2915 			continue;
2916 		}
2917 
2918 		diag = rte_eth_dev_reset(pi);
2919 		if (diag == 0) {
2920 			port = &ports[pi];
2921 			port->need_reconfig = 1;
2922 			port->need_reconfig_queues = 1;
2923 		} else {
2924 			printf("Failed to reset port %d. diag=%d\n", pi, diag);
2925 		}
2926 	}
2927 
2928 	printf("Done\n");
2929 }
2930 
2931 void
2932 attach_port(char *identifier)
2933 {
2934 	portid_t pi;
2935 	struct rte_dev_iterator iterator;
2936 
2937 	printf("Attaching a new port...\n");
2938 
2939 	if (identifier == NULL) {
2940 		printf("Invalid parameters are specified\n");
2941 		return;
2942 	}
2943 
2944 	if (rte_dev_probe(identifier) < 0) {
2945 		TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2946 		return;
2947 	}
2948 
2949 	/* first attach mode: event */
2950 	if (setup_on_probe_event) {
2951 		/* new ports are detected on RTE_ETH_EVENT_NEW event */
2952 		for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2953 			if (ports[pi].port_status == RTE_PORT_HANDLING &&
2954 					ports[pi].need_setup != 0)
2955 				setup_attached_port(pi);
2956 		return;
2957 	}
2958 
2959 	/* second attach mode: iterator */
2960 	RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2961 		/* setup ports matching the devargs used for probing */
2962 		if (port_is_forwarding(pi))
2963 			continue; /* port was already attached before */
2964 		setup_attached_port(pi);
2965 	}
2966 }
2967 
2968 static void
2969 setup_attached_port(portid_t pi)
2970 {
2971 	unsigned int socket_id;
2972 	int ret;
2973 
2974 	socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2975 	/* if socket_id is invalid, set to the first available socket. */
2976 	if (check_socket_id(socket_id) < 0)
2977 		socket_id = socket_ids[0];
2978 	reconfig(pi, socket_id);
2979 	ret = rte_eth_promiscuous_enable(pi);
2980 	if (ret != 0)
2981 		printf("Error during enabling promiscuous mode for port %u: %s - ignore\n",
2982 			pi, rte_strerror(-ret));
2983 
2984 	ports_ids[nb_ports++] = pi;
2985 	fwd_ports_ids[nb_fwd_ports++] = pi;
2986 	nb_cfg_ports = nb_fwd_ports;
2987 	ports[pi].need_setup = 0;
2988 	ports[pi].port_status = RTE_PORT_STOPPED;
2989 
2990 	printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2991 	printf("Done\n");
2992 }
2993 
2994 static void
2995 detach_device(struct rte_device *dev)
2996 {
2997 	portid_t sibling;
2998 
2999 	if (dev == NULL) {
3000 		printf("Device already removed\n");
3001 		return;
3002 	}
3003 
3004 	printf("Removing a device...\n");
3005 
3006 	RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
3007 		if (ports[sibling].port_status != RTE_PORT_CLOSED) {
3008 			if (ports[sibling].port_status != RTE_PORT_STOPPED) {
3009 				printf("Port %u not stopped\n", sibling);
3010 				return;
3011 			}
3012 			port_flow_flush(sibling);
3013 		}
3014 	}
3015 
3016 	if (rte_dev_remove(dev) < 0) {
3017 		TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
3018 		return;
3019 	}
3020 	remove_invalid_ports();
3021 
3022 	printf("Device is detached\n");
3023 	printf("Now total ports is %d\n", nb_ports);
3024 	printf("Done\n");
3025 	return;
3026 }
3027 
3028 void
3029 detach_port_device(portid_t port_id)
3030 {
3031 	if (port_id_is_invalid(port_id, ENABLED_WARN))
3032 		return;
3033 
3034 	if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3035 		if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3036 			printf("Port not stopped\n");
3037 			return;
3038 		}
3039 		printf("Port was not closed\n");
3040 	}
3041 
3042 	detach_device(rte_eth_devices[port_id].device);
3043 }
3044 
3045 void
3046 detach_devargs(char *identifier)
3047 {
3048 	struct rte_dev_iterator iterator;
3049 	struct rte_devargs da;
3050 	portid_t port_id;
3051 
3052 	printf("Removing a device...\n");
3053 
3054 	memset(&da, 0, sizeof(da));
3055 	if (rte_devargs_parsef(&da, "%s", identifier)) {
3056 		printf("cannot parse identifier\n");
3057 		if (da.args)
3058 			free(da.args);
3059 		return;
3060 	}
3061 
3062 	RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3063 		if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3064 			if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3065 				printf("Port %u not stopped\n", port_id);
3066 				rte_eth_iterator_cleanup(&iterator);
3067 				return;
3068 			}
3069 			port_flow_flush(port_id);
3070 		}
3071 	}
3072 
3073 	if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3074 		TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3075 			    da.name, da.bus->name);
3076 		return;
3077 	}
3078 
3079 	remove_invalid_ports();
3080 
3081 	printf("Device %s is detached\n", identifier);
3082 	printf("Now total ports is %d\n", nb_ports);
3083 	printf("Done\n");
3084 }
3085 
3086 void
3087 pmd_test_exit(void)
3088 {
3089 	portid_t pt_id;
3090 	unsigned int i;
3091 	int ret;
3092 
3093 	if (test_done == 0)
3094 		stop_packet_forwarding();
3095 
3096 	for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3097 		if (mempools[i]) {
3098 			if (mp_alloc_type == MP_ALLOC_ANON)
3099 				rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3100 						     NULL);
3101 		}
3102 	}
3103 	if (ports != NULL) {
3104 		no_link_check = 1;
3105 		RTE_ETH_FOREACH_DEV(pt_id) {
3106 			printf("\nStopping port %d...\n", pt_id);
3107 			fflush(stdout);
3108 			stop_port(pt_id);
3109 		}
3110 		RTE_ETH_FOREACH_DEV(pt_id) {
3111 			printf("\nShutting down port %d...\n", pt_id);
3112 			fflush(stdout);
3113 			close_port(pt_id);
3114 		}
3115 	}
3116 
3117 	if (hot_plug) {
3118 		ret = rte_dev_event_monitor_stop();
3119 		if (ret) {
3120 			RTE_LOG(ERR, EAL,
3121 				"fail to stop device event monitor.");
3122 			return;
3123 		}
3124 
3125 		ret = rte_dev_event_callback_unregister(NULL,
3126 			dev_event_callback, NULL);
3127 		if (ret < 0) {
3128 			RTE_LOG(ERR, EAL,
3129 				"fail to unregister device event callback.\n");
3130 			return;
3131 		}
3132 
3133 		ret = rte_dev_hotplug_handle_disable();
3134 		if (ret) {
3135 			RTE_LOG(ERR, EAL,
3136 				"fail to disable hotplug handling.\n");
3137 			return;
3138 		}
3139 	}
3140 	for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3141 		if (mempools[i])
3142 			rte_mempool_free(mempools[i]);
3143 	}
3144 
3145 	printf("\nBye...\n");
3146 }
3147 
3148 typedef void (*cmd_func_t)(void);
3149 struct pmd_test_command {
3150 	const char *cmd_name;
3151 	cmd_func_t cmd_func;
3152 };
3153 
3154 /* Check the link status of all ports in up to 9s, and print them finally */
3155 static void
3156 check_all_ports_link_status(uint32_t port_mask)
3157 {
3158 #define CHECK_INTERVAL 100 /* 100ms */
3159 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3160 	portid_t portid;
3161 	uint8_t count, all_ports_up, print_flag = 0;
3162 	struct rte_eth_link link;
3163 	int ret;
3164 	char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3165 
3166 	printf("Checking link statuses...\n");
3167 	fflush(stdout);
3168 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
3169 		all_ports_up = 1;
3170 		RTE_ETH_FOREACH_DEV(portid) {
3171 			if ((port_mask & (1 << portid)) == 0)
3172 				continue;
3173 			memset(&link, 0, sizeof(link));
3174 			ret = rte_eth_link_get_nowait(portid, &link);
3175 			if (ret < 0) {
3176 				all_ports_up = 0;
3177 				if (print_flag == 1)
3178 					printf("Port %u link get failed: %s\n",
3179 						portid, rte_strerror(-ret));
3180 				continue;
3181 			}
3182 			/* print link status if flag set */
3183 			if (print_flag == 1) {
3184 				rte_eth_link_to_str(link_status,
3185 					sizeof(link_status), &link);
3186 				printf("Port %d %s\n", portid, link_status);
3187 				continue;
3188 			}
3189 			/* clear all_ports_up flag if any link down */
3190 			if (link.link_status == ETH_LINK_DOWN) {
3191 				all_ports_up = 0;
3192 				break;
3193 			}
3194 		}
3195 		/* after finally printing all link status, get out */
3196 		if (print_flag == 1)
3197 			break;
3198 
3199 		if (all_ports_up == 0) {
3200 			fflush(stdout);
3201 			rte_delay_ms(CHECK_INTERVAL);
3202 		}
3203 
3204 		/* set the print_flag if all ports up or timeout */
3205 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3206 			print_flag = 1;
3207 		}
3208 
3209 		if (lsc_interrupt)
3210 			break;
3211 	}
3212 }
3213 
3214 static void
3215 rmv_port_callback(void *arg)
3216 {
3217 	int need_to_start = 0;
3218 	int org_no_link_check = no_link_check;
3219 	portid_t port_id = (intptr_t)arg;
3220 	struct rte_device *dev;
3221 
3222 	RTE_ETH_VALID_PORTID_OR_RET(port_id);
3223 
3224 	if (!test_done && port_is_forwarding(port_id)) {
3225 		need_to_start = 1;
3226 		stop_packet_forwarding();
3227 	}
3228 	no_link_check = 1;
3229 	stop_port(port_id);
3230 	no_link_check = org_no_link_check;
3231 
3232 	/* Save rte_device pointer before closing ethdev port */
3233 	dev = rte_eth_devices[port_id].device;
3234 	close_port(port_id);
3235 	detach_device(dev); /* might be already removed or have more ports */
3236 
3237 	if (need_to_start)
3238 		start_packet_forwarding(0);
3239 }
3240 
3241 /* This function is used by the interrupt thread */
3242 static int
3243 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3244 		  void *ret_param)
3245 {
3246 	RTE_SET_USED(param);
3247 	RTE_SET_USED(ret_param);
3248 
3249 	if (type >= RTE_ETH_EVENT_MAX) {
3250 		fprintf(stderr, "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3251 			port_id, __func__, type);
3252 		fflush(stderr);
3253 	} else if (event_print_mask & (UINT32_C(1) << type)) {
3254 		printf("\nPort %" PRIu16 ": %s event\n", port_id,
3255 			eth_event_desc[type]);
3256 		fflush(stdout);
3257 	}
3258 
3259 	switch (type) {
3260 	case RTE_ETH_EVENT_NEW:
3261 		ports[port_id].need_setup = 1;
3262 		ports[port_id].port_status = RTE_PORT_HANDLING;
3263 		break;
3264 	case RTE_ETH_EVENT_INTR_RMV:
3265 		if (port_id_is_invalid(port_id, DISABLED_WARN))
3266 			break;
3267 		if (rte_eal_alarm_set(100000,
3268 				rmv_port_callback, (void *)(intptr_t)port_id))
3269 			fprintf(stderr, "Could not set up deferred device removal\n");
3270 		break;
3271 	case RTE_ETH_EVENT_DESTROY:
3272 		ports[port_id].port_status = RTE_PORT_CLOSED;
3273 		printf("Port %u is closed\n", port_id);
3274 		break;
3275 	default:
3276 		break;
3277 	}
3278 	return 0;
3279 }
3280 
3281 static int
3282 register_eth_event_callback(void)
3283 {
3284 	int ret;
3285 	enum rte_eth_event_type event;
3286 
3287 	for (event = RTE_ETH_EVENT_UNKNOWN;
3288 			event < RTE_ETH_EVENT_MAX; event++) {
3289 		ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3290 				event,
3291 				eth_event_callback,
3292 				NULL);
3293 		if (ret != 0) {
3294 			TESTPMD_LOG(ERR, "Failed to register callback for "
3295 					"%s event\n", eth_event_desc[event]);
3296 			return -1;
3297 		}
3298 	}
3299 
3300 	return 0;
3301 }
3302 
3303 /* This function is used by the interrupt thread */
3304 static void
3305 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3306 			     __rte_unused void *arg)
3307 {
3308 	uint16_t port_id;
3309 	int ret;
3310 
3311 	if (type >= RTE_DEV_EVENT_MAX) {
3312 		fprintf(stderr, "%s called upon invalid event %d\n",
3313 			__func__, type);
3314 		fflush(stderr);
3315 	}
3316 
3317 	switch (type) {
3318 	case RTE_DEV_EVENT_REMOVE:
3319 		RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3320 			device_name);
3321 		ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3322 		if (ret) {
3323 			RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3324 				device_name);
3325 			return;
3326 		}
3327 		/*
3328 		 * Because the user's callback is invoked in eal interrupt
3329 		 * callback, the interrupt callback need to be finished before
3330 		 * it can be unregistered when detaching device. So finish
3331 		 * callback soon and use a deferred removal to detach device
3332 		 * is need. It is a workaround, once the device detaching be
3333 		 * moved into the eal in the future, the deferred removal could
3334 		 * be deleted.
3335 		 */
3336 		if (rte_eal_alarm_set(100000,
3337 				rmv_port_callback, (void *)(intptr_t)port_id))
3338 			RTE_LOG(ERR, EAL,
3339 				"Could not set up deferred device removal\n");
3340 		break;
3341 	case RTE_DEV_EVENT_ADD:
3342 		RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3343 			device_name);
3344 		/* TODO: After finish kernel driver binding,
3345 		 * begin to attach port.
3346 		 */
3347 		break;
3348 	default:
3349 		break;
3350 	}
3351 }
3352 
3353 static int
3354 set_tx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
3355 {
3356 	uint16_t i;
3357 	int diag;
3358 	uint8_t mapping_found = 0;
3359 
3360 	for (i = 0; i < nb_tx_queue_stats_mappings; i++) {
3361 		if ((tx_queue_stats_mappings[i].port_id == port_id) &&
3362 				(tx_queue_stats_mappings[i].queue_id < nb_txq )) {
3363 			diag = rte_eth_dev_set_tx_queue_stats_mapping(port_id,
3364 					tx_queue_stats_mappings[i].queue_id,
3365 					tx_queue_stats_mappings[i].stats_counter_id);
3366 			if (diag != 0)
3367 				return diag;
3368 			mapping_found = 1;
3369 		}
3370 	}
3371 	if (mapping_found)
3372 		port->tx_queue_stats_mapping_enabled = 1;
3373 	return 0;
3374 }
3375 
3376 static int
3377 set_rx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
3378 {
3379 	uint16_t i;
3380 	int diag;
3381 	uint8_t mapping_found = 0;
3382 
3383 	for (i = 0; i < nb_rx_queue_stats_mappings; i++) {
3384 		if ((rx_queue_stats_mappings[i].port_id == port_id) &&
3385 				(rx_queue_stats_mappings[i].queue_id < nb_rxq )) {
3386 			diag = rte_eth_dev_set_rx_queue_stats_mapping(port_id,
3387 					rx_queue_stats_mappings[i].queue_id,
3388 					rx_queue_stats_mappings[i].stats_counter_id);
3389 			if (diag != 0)
3390 				return diag;
3391 			mapping_found = 1;
3392 		}
3393 	}
3394 	if (mapping_found)
3395 		port->rx_queue_stats_mapping_enabled = 1;
3396 	return 0;
3397 }
3398 
3399 static void
3400 map_port_queue_stats_mapping_registers(portid_t pi, struct rte_port *port)
3401 {
3402 	int diag = 0;
3403 
3404 	diag = set_tx_queue_stats_mapping_registers(pi, port);
3405 	if (diag != 0) {
3406 		if (diag == -ENOTSUP) {
3407 			port->tx_queue_stats_mapping_enabled = 0;
3408 			printf("TX queue stats mapping not supported port id=%d\n", pi);
3409 		}
3410 		else
3411 			rte_exit(EXIT_FAILURE,
3412 					"set_tx_queue_stats_mapping_registers "
3413 					"failed for port id=%d diag=%d\n",
3414 					pi, diag);
3415 	}
3416 
3417 	diag = set_rx_queue_stats_mapping_registers(pi, port);
3418 	if (diag != 0) {
3419 		if (diag == -ENOTSUP) {
3420 			port->rx_queue_stats_mapping_enabled = 0;
3421 			printf("RX queue stats mapping not supported port id=%d\n", pi);
3422 		}
3423 		else
3424 			rte_exit(EXIT_FAILURE,
3425 					"set_rx_queue_stats_mapping_registers "
3426 					"failed for port id=%d diag=%d\n",
3427 					pi, diag);
3428 	}
3429 }
3430 
3431 static void
3432 rxtx_port_config(struct rte_port *port)
3433 {
3434 	uint16_t qid;
3435 	uint64_t offloads;
3436 
3437 	for (qid = 0; qid < nb_rxq; qid++) {
3438 		offloads = port->rx_conf[qid].offloads;
3439 		port->rx_conf[qid] = port->dev_info.default_rxconf;
3440 		if (offloads != 0)
3441 			port->rx_conf[qid].offloads = offloads;
3442 
3443 		/* Check if any Rx parameters have been passed */
3444 		if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3445 			port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3446 
3447 		if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3448 			port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3449 
3450 		if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3451 			port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3452 
3453 		if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3454 			port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3455 
3456 		if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3457 			port->rx_conf[qid].rx_drop_en = rx_drop_en;
3458 
3459 		port->nb_rx_desc[qid] = nb_rxd;
3460 	}
3461 
3462 	for (qid = 0; qid < nb_txq; qid++) {
3463 		offloads = port->tx_conf[qid].offloads;
3464 		port->tx_conf[qid] = port->dev_info.default_txconf;
3465 		if (offloads != 0)
3466 			port->tx_conf[qid].offloads = offloads;
3467 
3468 		/* Check if any Tx parameters have been passed */
3469 		if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3470 			port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3471 
3472 		if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3473 			port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3474 
3475 		if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3476 			port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3477 
3478 		if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3479 			port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3480 
3481 		if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3482 			port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3483 
3484 		port->nb_tx_desc[qid] = nb_txd;
3485 	}
3486 }
3487 
3488 void
3489 init_port_config(void)
3490 {
3491 	portid_t pid;
3492 	struct rte_port *port;
3493 	int ret;
3494 
3495 	RTE_ETH_FOREACH_DEV(pid) {
3496 		port = &ports[pid];
3497 		port->dev_conf.fdir_conf = fdir_conf;
3498 
3499 		ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3500 		if (ret != 0)
3501 			return;
3502 
3503 		if (nb_rxq > 1) {
3504 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3505 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3506 				rss_hf & port->dev_info.flow_type_rss_offloads;
3507 		} else {
3508 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3509 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3510 		}
3511 
3512 		if (port->dcb_flag == 0) {
3513 			if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
3514 				port->dev_conf.rxmode.mq_mode =
3515 					(enum rte_eth_rx_mq_mode)
3516 						(rx_mq_mode & ETH_MQ_RX_RSS);
3517 			else
3518 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
3519 		}
3520 
3521 		rxtx_port_config(port);
3522 
3523 		ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3524 		if (ret != 0)
3525 			return;
3526 
3527 		map_port_queue_stats_mapping_registers(pid, port);
3528 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
3529 		rte_pmd_ixgbe_bypass_init(pid);
3530 #endif
3531 
3532 		if (lsc_interrupt &&
3533 		    (rte_eth_devices[pid].data->dev_flags &
3534 		     RTE_ETH_DEV_INTR_LSC))
3535 			port->dev_conf.intr_conf.lsc = 1;
3536 		if (rmv_interrupt &&
3537 		    (rte_eth_devices[pid].data->dev_flags &
3538 		     RTE_ETH_DEV_INTR_RMV))
3539 			port->dev_conf.intr_conf.rmv = 1;
3540 	}
3541 }
3542 
3543 void set_port_slave_flag(portid_t slave_pid)
3544 {
3545 	struct rte_port *port;
3546 
3547 	port = &ports[slave_pid];
3548 	port->slave_flag = 1;
3549 }
3550 
3551 void clear_port_slave_flag(portid_t slave_pid)
3552 {
3553 	struct rte_port *port;
3554 
3555 	port = &ports[slave_pid];
3556 	port->slave_flag = 0;
3557 }
3558 
3559 uint8_t port_is_bonding_slave(portid_t slave_pid)
3560 {
3561 	struct rte_port *port;
3562 
3563 	port = &ports[slave_pid];
3564 	if ((rte_eth_devices[slave_pid].data->dev_flags &
3565 	    RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3566 		return 1;
3567 	return 0;
3568 }
3569 
3570 const uint16_t vlan_tags[] = {
3571 		0,  1,  2,  3,  4,  5,  6,  7,
3572 		8,  9, 10, 11,  12, 13, 14, 15,
3573 		16, 17, 18, 19, 20, 21, 22, 23,
3574 		24, 25, 26, 27, 28, 29, 30, 31
3575 };
3576 
3577 static  int
3578 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3579 		 enum dcb_mode_enable dcb_mode,
3580 		 enum rte_eth_nb_tcs num_tcs,
3581 		 uint8_t pfc_en)
3582 {
3583 	uint8_t i;
3584 	int32_t rc;
3585 	struct rte_eth_rss_conf rss_conf;
3586 
3587 	/*
3588 	 * Builds up the correct configuration for dcb+vt based on the vlan tags array
3589 	 * given above, and the number of traffic classes available for use.
3590 	 */
3591 	if (dcb_mode == DCB_VT_ENABLED) {
3592 		struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3593 				&eth_conf->rx_adv_conf.vmdq_dcb_conf;
3594 		struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3595 				&eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3596 
3597 		/* VMDQ+DCB RX and TX configurations */
3598 		vmdq_rx_conf->enable_default_pool = 0;
3599 		vmdq_rx_conf->default_pool = 0;
3600 		vmdq_rx_conf->nb_queue_pools =
3601 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3602 		vmdq_tx_conf->nb_queue_pools =
3603 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3604 
3605 		vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3606 		for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3607 			vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3608 			vmdq_rx_conf->pool_map[i].pools =
3609 				1 << (i % vmdq_rx_conf->nb_queue_pools);
3610 		}
3611 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3612 			vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3613 			vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3614 		}
3615 
3616 		/* set DCB mode of RX and TX of multiple queues */
3617 		eth_conf->rxmode.mq_mode =
3618 				(enum rte_eth_rx_mq_mode)
3619 					(rx_mq_mode & ETH_MQ_RX_VMDQ_DCB);
3620 		eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3621 	} else {
3622 		struct rte_eth_dcb_rx_conf *rx_conf =
3623 				&eth_conf->rx_adv_conf.dcb_rx_conf;
3624 		struct rte_eth_dcb_tx_conf *tx_conf =
3625 				&eth_conf->tx_adv_conf.dcb_tx_conf;
3626 
3627 		memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3628 
3629 		rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3630 		if (rc != 0)
3631 			return rc;
3632 
3633 		rx_conf->nb_tcs = num_tcs;
3634 		tx_conf->nb_tcs = num_tcs;
3635 
3636 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3637 			rx_conf->dcb_tc[i] = i % num_tcs;
3638 			tx_conf->dcb_tc[i] = i % num_tcs;
3639 		}
3640 
3641 		eth_conf->rxmode.mq_mode =
3642 				(enum rte_eth_rx_mq_mode)
3643 					(rx_mq_mode & ETH_MQ_RX_DCB_RSS);
3644 		eth_conf->rx_adv_conf.rss_conf = rss_conf;
3645 		eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3646 	}
3647 
3648 	if (pfc_en)
3649 		eth_conf->dcb_capability_en =
3650 				ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3651 	else
3652 		eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3653 
3654 	return 0;
3655 }
3656 
3657 int
3658 init_port_dcb_config(portid_t pid,
3659 		     enum dcb_mode_enable dcb_mode,
3660 		     enum rte_eth_nb_tcs num_tcs,
3661 		     uint8_t pfc_en)
3662 {
3663 	struct rte_eth_conf port_conf;
3664 	struct rte_port *rte_port;
3665 	int retval;
3666 	uint16_t i;
3667 
3668 	rte_port = &ports[pid];
3669 
3670 	memset(&port_conf, 0, sizeof(struct rte_eth_conf));
3671 	/* Enter DCB configuration status */
3672 	dcb_config = 1;
3673 
3674 	port_conf.rxmode = rte_port->dev_conf.rxmode;
3675 	port_conf.txmode = rte_port->dev_conf.txmode;
3676 
3677 	/*set configuration of DCB in vt mode and DCB in non-vt mode*/
3678 	retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3679 	if (retval < 0)
3680 		return retval;
3681 	port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3682 
3683 	/* re-configure the device . */
3684 	retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3685 	if (retval < 0)
3686 		return retval;
3687 
3688 	retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3689 	if (retval != 0)
3690 		return retval;
3691 
3692 	/* If dev_info.vmdq_pool_base is greater than 0,
3693 	 * the queue id of vmdq pools is started after pf queues.
3694 	 */
3695 	if (dcb_mode == DCB_VT_ENABLED &&
3696 	    rte_port->dev_info.vmdq_pool_base > 0) {
3697 		printf("VMDQ_DCB multi-queue mode is nonsensical"
3698 			" for port %d.", pid);
3699 		return -1;
3700 	}
3701 
3702 	/* Assume the ports in testpmd have the same dcb capability
3703 	 * and has the same number of rxq and txq in dcb mode
3704 	 */
3705 	if (dcb_mode == DCB_VT_ENABLED) {
3706 		if (rte_port->dev_info.max_vfs > 0) {
3707 			nb_rxq = rte_port->dev_info.nb_rx_queues;
3708 			nb_txq = rte_port->dev_info.nb_tx_queues;
3709 		} else {
3710 			nb_rxq = rte_port->dev_info.max_rx_queues;
3711 			nb_txq = rte_port->dev_info.max_tx_queues;
3712 		}
3713 	} else {
3714 		/*if vt is disabled, use all pf queues */
3715 		if (rte_port->dev_info.vmdq_pool_base == 0) {
3716 			nb_rxq = rte_port->dev_info.max_rx_queues;
3717 			nb_txq = rte_port->dev_info.max_tx_queues;
3718 		} else {
3719 			nb_rxq = (queueid_t)num_tcs;
3720 			nb_txq = (queueid_t)num_tcs;
3721 
3722 		}
3723 	}
3724 	rx_free_thresh = 64;
3725 
3726 	memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3727 
3728 	rxtx_port_config(rte_port);
3729 	/* VLAN filter */
3730 	rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3731 	for (i = 0; i < RTE_DIM(vlan_tags); i++)
3732 		rx_vft_set(pid, vlan_tags[i], 1);
3733 
3734 	retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
3735 	if (retval != 0)
3736 		return retval;
3737 
3738 	map_port_queue_stats_mapping_registers(pid, rte_port);
3739 
3740 	rte_port->dcb_flag = 1;
3741 
3742 	return 0;
3743 }
3744 
3745 static void
3746 init_port(void)
3747 {
3748 	int i;
3749 
3750 	/* Configuration of Ethernet ports. */
3751 	ports = rte_zmalloc("testpmd: ports",
3752 			    sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3753 			    RTE_CACHE_LINE_SIZE);
3754 	if (ports == NULL) {
3755 		rte_exit(EXIT_FAILURE,
3756 				"rte_zmalloc(%d struct rte_port) failed\n",
3757 				RTE_MAX_ETHPORTS);
3758 	}
3759 	for (i = 0; i < RTE_MAX_ETHPORTS; i++)
3760 		LIST_INIT(&ports[i].flow_tunnel_list);
3761 	/* Initialize ports NUMA structures */
3762 	memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3763 	memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3764 	memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3765 }
3766 
3767 static void
3768 force_quit(void)
3769 {
3770 	pmd_test_exit();
3771 	prompt_exit();
3772 }
3773 
3774 static void
3775 print_stats(void)
3776 {
3777 	uint8_t i;
3778 	const char clr[] = { 27, '[', '2', 'J', '\0' };
3779 	const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3780 
3781 	/* Clear screen and move to top left */
3782 	printf("%s%s", clr, top_left);
3783 
3784 	printf("\nPort statistics ====================================");
3785 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3786 		nic_stats_display(fwd_ports_ids[i]);
3787 
3788 	fflush(stdout);
3789 }
3790 
3791 static void
3792 signal_handler(int signum)
3793 {
3794 	if (signum == SIGINT || signum == SIGTERM) {
3795 		printf("\nSignal %d received, preparing to exit...\n",
3796 				signum);
3797 #ifdef RTE_LIBRTE_PDUMP
3798 		/* uninitialize packet capture framework */
3799 		rte_pdump_uninit();
3800 #endif
3801 #ifdef RTE_LIBRTE_LATENCY_STATS
3802 		if (latencystats_enabled != 0)
3803 			rte_latencystats_uninit();
3804 #endif
3805 		force_quit();
3806 		/* Set flag to indicate the force termination. */
3807 		f_quit = 1;
3808 		/* exit with the expected status */
3809 		signal(signum, SIG_DFL);
3810 		kill(getpid(), signum);
3811 	}
3812 }
3813 
3814 int
3815 main(int argc, char** argv)
3816 {
3817 	int diag;
3818 	portid_t port_id;
3819 	uint16_t count;
3820 	int ret;
3821 
3822 	signal(SIGINT, signal_handler);
3823 	signal(SIGTERM, signal_handler);
3824 
3825 	testpmd_logtype = rte_log_register("testpmd");
3826 	if (testpmd_logtype < 0)
3827 		rte_exit(EXIT_FAILURE, "Cannot register log type");
3828 	rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3829 
3830 	diag = rte_eal_init(argc, argv);
3831 	if (diag < 0)
3832 		rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
3833 			 rte_strerror(rte_errno));
3834 
3835 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
3836 		rte_exit(EXIT_FAILURE,
3837 			 "Secondary process type not supported.\n");
3838 
3839 	ret = register_eth_event_callback();
3840 	if (ret != 0)
3841 		rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
3842 
3843 #ifdef RTE_LIBRTE_PDUMP
3844 	/* initialize packet capture framework */
3845 	rte_pdump_init();
3846 #endif
3847 
3848 	count = 0;
3849 	RTE_ETH_FOREACH_DEV(port_id) {
3850 		ports_ids[count] = port_id;
3851 		count++;
3852 	}
3853 	nb_ports = (portid_t) count;
3854 	if (nb_ports == 0)
3855 		TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3856 
3857 	/* allocate port structures, and init them */
3858 	init_port();
3859 
3860 	set_def_fwd_config();
3861 	if (nb_lcores == 0)
3862 		rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
3863 			 "Check the core mask argument\n");
3864 
3865 	/* Bitrate/latency stats disabled by default */
3866 #ifdef RTE_LIBRTE_BITRATESTATS
3867 	bitrate_enabled = 0;
3868 #endif
3869 #ifdef RTE_LIBRTE_LATENCY_STATS
3870 	latencystats_enabled = 0;
3871 #endif
3872 
3873 	/* on FreeBSD, mlockall() is disabled by default */
3874 #ifdef RTE_EXEC_ENV_FREEBSD
3875 	do_mlockall = 0;
3876 #else
3877 	do_mlockall = 1;
3878 #endif
3879 
3880 	argc -= diag;
3881 	argv += diag;
3882 	if (argc > 1)
3883 		launch_args_parse(argc, argv);
3884 
3885 	if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3886 		TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3887 			strerror(errno));
3888 	}
3889 
3890 	if (tx_first && interactive)
3891 		rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3892 				"interactive mode.\n");
3893 
3894 	if (tx_first && lsc_interrupt) {
3895 		printf("Warning: lsc_interrupt needs to be off when "
3896 				" using tx_first. Disabling.\n");
3897 		lsc_interrupt = 0;
3898 	}
3899 
3900 	if (!nb_rxq && !nb_txq)
3901 		printf("Warning: Either rx or tx queues should be non-zero\n");
3902 
3903 	if (nb_rxq > 1 && nb_rxq > nb_txq)
3904 		printf("Warning: nb_rxq=%d enables RSS configuration, "
3905 		       "but nb_txq=%d will prevent to fully test it.\n",
3906 		       nb_rxq, nb_txq);
3907 
3908 	init_config();
3909 
3910 	if (hot_plug) {
3911 		ret = rte_dev_hotplug_handle_enable();
3912 		if (ret) {
3913 			RTE_LOG(ERR, EAL,
3914 				"fail to enable hotplug handling.");
3915 			return -1;
3916 		}
3917 
3918 		ret = rte_dev_event_monitor_start();
3919 		if (ret) {
3920 			RTE_LOG(ERR, EAL,
3921 				"fail to start device event monitoring.");
3922 			return -1;
3923 		}
3924 
3925 		ret = rte_dev_event_callback_register(NULL,
3926 			dev_event_callback, NULL);
3927 		if (ret) {
3928 			RTE_LOG(ERR, EAL,
3929 				"fail  to register device event callback\n");
3930 			return -1;
3931 		}
3932 	}
3933 
3934 	if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
3935 		rte_exit(EXIT_FAILURE, "Start ports failed\n");
3936 
3937 	/* set all ports to promiscuous mode by default */
3938 	RTE_ETH_FOREACH_DEV(port_id) {
3939 		ret = rte_eth_promiscuous_enable(port_id);
3940 		if (ret != 0)
3941 			printf("Error during enabling promiscuous mode for port %u: %s - ignore\n",
3942 				port_id, rte_strerror(-ret));
3943 	}
3944 
3945 	/* Init metrics library */
3946 	rte_metrics_init(rte_socket_id());
3947 
3948 #ifdef RTE_LIBRTE_LATENCY_STATS
3949 	if (latencystats_enabled != 0) {
3950 		int ret = rte_latencystats_init(1, NULL);
3951 		if (ret)
3952 			printf("Warning: latencystats init()"
3953 				" returned error %d\n",	ret);
3954 		printf("Latencystats running on lcore %d\n",
3955 			latencystats_lcore_id);
3956 	}
3957 #endif
3958 
3959 	/* Setup bitrate stats */
3960 #ifdef RTE_LIBRTE_BITRATESTATS
3961 	if (bitrate_enabled != 0) {
3962 		bitrate_data = rte_stats_bitrate_create();
3963 		if (bitrate_data == NULL)
3964 			rte_exit(EXIT_FAILURE,
3965 				"Could not allocate bitrate data.\n");
3966 		rte_stats_bitrate_reg(bitrate_data);
3967 	}
3968 #endif
3969 
3970 #ifdef RTE_LIBRTE_CMDLINE
3971 	if (strlen(cmdline_filename) != 0)
3972 		cmdline_read_from_file(cmdline_filename);
3973 
3974 	if (interactive == 1) {
3975 		if (auto_start) {
3976 			printf("Start automatic packet forwarding\n");
3977 			start_packet_forwarding(0);
3978 		}
3979 		prompt();
3980 		pmd_test_exit();
3981 	} else
3982 #endif
3983 	{
3984 		char c;
3985 		int rc;
3986 
3987 		f_quit = 0;
3988 
3989 		printf("No commandline core given, start packet forwarding\n");
3990 		start_packet_forwarding(tx_first);
3991 		if (stats_period != 0) {
3992 			uint64_t prev_time = 0, cur_time, diff_time = 0;
3993 			uint64_t timer_period;
3994 
3995 			/* Convert to number of cycles */
3996 			timer_period = stats_period * rte_get_timer_hz();
3997 
3998 			while (f_quit == 0) {
3999 				cur_time = rte_get_timer_cycles();
4000 				diff_time += cur_time - prev_time;
4001 
4002 				if (diff_time >= timer_period) {
4003 					print_stats();
4004 					/* Reset the timer */
4005 					diff_time = 0;
4006 				}
4007 				/* Sleep to avoid unnecessary checks */
4008 				prev_time = cur_time;
4009 				sleep(1);
4010 			}
4011 		}
4012 
4013 		printf("Press enter to exit\n");
4014 		rc = read(0, &c, 1);
4015 		pmd_test_exit();
4016 		if (rc < 0)
4017 			return 1;
4018 	}
4019 
4020 	ret = rte_eal_cleanup();
4021 	if (ret != 0)
4022 		rte_exit(EXIT_FAILURE,
4023 			 "EAL cleanup failed: %s\n", strerror(-ret));
4024 
4025 	return EXIT_SUCCESS;
4026 }
4027