xref: /dpdk/app/test-pmd/testpmd.c (revision 33f9630fc23dd550c0b2bca85f6c2b1aecc4a4ff)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #include <stdarg.h>
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <signal.h>
9 #include <string.h>
10 #include <time.h>
11 #include <fcntl.h>
12 #include <sys/mman.h>
13 #include <sys/types.h>
14 #include <errno.h>
15 #include <stdbool.h>
16 
17 #include <sys/queue.h>
18 #include <sys/stat.h>
19 
20 #include <stdint.h>
21 #include <unistd.h>
22 #include <inttypes.h>
23 
24 #include <rte_common.h>
25 #include <rte_errno.h>
26 #include <rte_byteorder.h>
27 #include <rte_log.h>
28 #include <rte_debug.h>
29 #include <rte_cycles.h>
30 #include <rte_malloc_heap.h>
31 #include <rte_memory.h>
32 #include <rte_memcpy.h>
33 #include <rte_launch.h>
34 #include <rte_eal.h>
35 #include <rte_alarm.h>
36 #include <rte_per_lcore.h>
37 #include <rte_lcore.h>
38 #include <rte_atomic.h>
39 #include <rte_branch_prediction.h>
40 #include <rte_mempool.h>
41 #include <rte_malloc.h>
42 #include <rte_mbuf.h>
43 #include <rte_mbuf_pool_ops.h>
44 #include <rte_interrupts.h>
45 #include <rte_pci.h>
46 #include <rte_ether.h>
47 #include <rte_ethdev.h>
48 #include <rte_dev.h>
49 #include <rte_string_fns.h>
50 #ifdef RTE_LIBRTE_IXGBE_PMD
51 #include <rte_pmd_ixgbe.h>
52 #endif
53 #ifdef RTE_LIBRTE_PDUMP
54 #include <rte_pdump.h>
55 #endif
56 #include <rte_flow.h>
57 #include <rte_metrics.h>
58 #ifdef RTE_LIBRTE_BITRATE
59 #include <rte_bitrate.h>
60 #endif
61 #ifdef RTE_LIBRTE_LATENCY_STATS
62 #include <rte_latencystats.h>
63 #endif
64 
65 #include "testpmd.h"
66 
67 #ifndef MAP_HUGETLB
68 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
69 #define HUGE_FLAG (0x40000)
70 #else
71 #define HUGE_FLAG MAP_HUGETLB
72 #endif
73 
74 #ifndef MAP_HUGE_SHIFT
75 /* older kernels (or FreeBSD) will not have this define */
76 #define HUGE_SHIFT (26)
77 #else
78 #define HUGE_SHIFT MAP_HUGE_SHIFT
79 #endif
80 
81 #define EXTMEM_HEAP_NAME "extmem"
82 
83 uint16_t verbose_level = 0; /**< Silent by default. */
84 int testpmd_logtype; /**< Log type for testpmd logs */
85 
86 /* use master core for command line ? */
87 uint8_t interactive = 0;
88 uint8_t auto_start = 0;
89 uint8_t tx_first;
90 char cmdline_filename[PATH_MAX] = {0};
91 
92 /*
93  * NUMA support configuration.
94  * When set, the NUMA support attempts to dispatch the allocation of the
95  * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
96  * probed ports among the CPU sockets 0 and 1.
97  * Otherwise, all memory is allocated from CPU socket 0.
98  */
99 uint8_t numa_support = 1; /**< numa enabled by default */
100 
101 /*
102  * In UMA mode,all memory is allocated from socket 0 if --socket-num is
103  * not configured.
104  */
105 uint8_t socket_num = UMA_NO_CONFIG;
106 
107 /*
108  * Select mempool allocation type:
109  * - native: use regular DPDK memory
110  * - anon: use regular DPDK memory to create mempool, but populate using
111  *         anonymous memory (may not be IOVA-contiguous)
112  * - xmem: use externally allocated hugepage memory
113  */
114 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
115 
116 /*
117  * Store specified sockets on which memory pool to be used by ports
118  * is allocated.
119  */
120 uint8_t port_numa[RTE_MAX_ETHPORTS];
121 
122 /*
123  * Store specified sockets on which RX ring to be used by ports
124  * is allocated.
125  */
126 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
127 
128 /*
129  * Store specified sockets on which TX ring to be used by ports
130  * is allocated.
131  */
132 uint8_t txring_numa[RTE_MAX_ETHPORTS];
133 
134 /*
135  * Record the Ethernet address of peer target ports to which packets are
136  * forwarded.
137  * Must be instantiated with the ethernet addresses of peer traffic generator
138  * ports.
139  */
140 struct 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 #if defined RTE_LIBRTE_PMD_SOFTNIC
183 	&softnic_fwd_engine,
184 #endif
185 #ifdef RTE_LIBRTE_IEEE1588
186 	&ieee1588_fwd_engine,
187 #endif
188 	NULL,
189 };
190 
191 struct rte_mempool *mempools[RTE_MAX_NUMA_NODES];
192 uint16_t mempool_flags;
193 
194 struct fwd_config cur_fwd_config;
195 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
196 uint32_t retry_enabled;
197 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
198 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
199 
200 uint16_t mbuf_data_size = DEFAULT_MBUF_DATA_SIZE; /**< Mbuf data space size. */
201 uint32_t param_total_num_mbufs = 0;  /**< number of mbufs in all pools - if
202                                       * specified on command-line. */
203 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
204 
205 /*
206  * In container, it cannot terminate the process which running with 'stats-period'
207  * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
208  */
209 uint8_t f_quit;
210 
211 /*
212  * Configuration of packet segments used by the "txonly" processing engine.
213  */
214 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
215 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
216 	TXONLY_DEF_PACKET_LEN,
217 };
218 uint8_t  tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
219 
220 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
221 /**< Split policy for packets to TX. */
222 
223 uint8_t txonly_multi_flow;
224 /**< Whether multiple flows are generated in TXONLY mode. */
225 
226 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
227 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
228 
229 /* current configuration is in DCB or not,0 means it is not in DCB mode */
230 uint8_t dcb_config = 0;
231 
232 /* Whether the dcb is in testing status */
233 uint8_t dcb_test = 0;
234 
235 /*
236  * Configurable number of RX/TX queues.
237  */
238 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
239 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
240 
241 /*
242  * Configurable number of RX/TX ring descriptors.
243  * Defaults are supplied by drivers via ethdev.
244  */
245 #define RTE_TEST_RX_DESC_DEFAULT 0
246 #define RTE_TEST_TX_DESC_DEFAULT 0
247 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
248 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
249 
250 #define RTE_PMD_PARAM_UNSET -1
251 /*
252  * Configurable values of RX and TX ring threshold registers.
253  */
254 
255 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
256 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
257 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
258 
259 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
260 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
261 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
262 
263 /*
264  * Configurable value of RX free threshold.
265  */
266 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
267 
268 /*
269  * Configurable value of RX drop enable.
270  */
271 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
272 
273 /*
274  * Configurable value of TX free threshold.
275  */
276 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
277 
278 /*
279  * Configurable value of TX RS bit threshold.
280  */
281 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
282 
283 /*
284  * Configurable value of buffered packets before sending.
285  */
286 uint16_t noisy_tx_sw_bufsz;
287 
288 /*
289  * Configurable value of packet buffer timeout.
290  */
291 uint16_t noisy_tx_sw_buf_flush_time;
292 
293 /*
294  * Configurable value for size of VNF internal memory area
295  * used for simulating noisy neighbour behaviour
296  */
297 uint64_t noisy_lkup_mem_sz;
298 
299 /*
300  * Configurable value of number of random writes done in
301  * VNF simulation memory area.
302  */
303 uint64_t noisy_lkup_num_writes;
304 
305 /*
306  * Configurable value of number of random reads done in
307  * VNF simulation memory area.
308  */
309 uint64_t noisy_lkup_num_reads;
310 
311 /*
312  * Configurable value of number of random reads/writes done in
313  * VNF simulation memory area.
314  */
315 uint64_t noisy_lkup_num_reads_writes;
316 
317 /*
318  * Receive Side Scaling (RSS) configuration.
319  */
320 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
321 
322 /*
323  * Port topology configuration
324  */
325 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
326 
327 /*
328  * Avoids to flush all the RX streams before starts forwarding.
329  */
330 uint8_t no_flush_rx = 0; /* flush by default */
331 
332 /*
333  * Flow API isolated mode.
334  */
335 uint8_t flow_isolate_all;
336 
337 /*
338  * Avoids to check link status when starting/stopping a port.
339  */
340 uint8_t no_link_check = 0; /* check by default */
341 
342 /*
343  * Enable link status change notification
344  */
345 uint8_t lsc_interrupt = 1; /* enabled by default */
346 
347 /*
348  * Enable device removal notification.
349  */
350 uint8_t rmv_interrupt = 1; /* enabled by default */
351 
352 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
353 
354 /* After attach, port setup is called on event or by iterator */
355 bool setup_on_probe_event = true;
356 
357 /* Pretty printing of ethdev events */
358 static const char * const eth_event_desc[] = {
359 	[RTE_ETH_EVENT_UNKNOWN] = "unknown",
360 	[RTE_ETH_EVENT_INTR_LSC] = "link state change",
361 	[RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
362 	[RTE_ETH_EVENT_INTR_RESET] = "reset",
363 	[RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
364 	[RTE_ETH_EVENT_IPSEC] = "IPsec",
365 	[RTE_ETH_EVENT_MACSEC] = "MACsec",
366 	[RTE_ETH_EVENT_INTR_RMV] = "device removal",
367 	[RTE_ETH_EVENT_NEW] = "device probed",
368 	[RTE_ETH_EVENT_DESTROY] = "device released",
369 	[RTE_ETH_EVENT_MAX] = NULL,
370 };
371 
372 /*
373  * Display or mask ether events
374  * Default to all events except VF_MBOX
375  */
376 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
377 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
378 			    (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
379 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
380 			    (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
381 			    (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
382 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV);
383 /*
384  * Decide if all memory are locked for performance.
385  */
386 int do_mlockall = 0;
387 
388 /*
389  * NIC bypass mode configuration options.
390  */
391 
392 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
393 /* The NIC bypass watchdog timeout. */
394 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
395 #endif
396 
397 
398 #ifdef RTE_LIBRTE_LATENCY_STATS
399 
400 /*
401  * Set when latency stats is enabled in the commandline
402  */
403 uint8_t latencystats_enabled;
404 
405 /*
406  * Lcore ID to serive latency statistics.
407  */
408 lcoreid_t latencystats_lcore_id = -1;
409 
410 #endif
411 
412 /*
413  * Ethernet device configuration.
414  */
415 struct rte_eth_rxmode rx_mode = {
416 	.max_rx_pkt_len = RTE_ETHER_MAX_LEN,
417 		/**< Default maximum frame length. */
418 };
419 
420 struct rte_eth_txmode tx_mode = {
421 	.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
422 };
423 
424 struct rte_fdir_conf fdir_conf = {
425 	.mode = RTE_FDIR_MODE_NONE,
426 	.pballoc = RTE_FDIR_PBALLOC_64K,
427 	.status = RTE_FDIR_REPORT_STATUS,
428 	.mask = {
429 		.vlan_tci_mask = 0xFFEF,
430 		.ipv4_mask     = {
431 			.src_ip = 0xFFFFFFFF,
432 			.dst_ip = 0xFFFFFFFF,
433 		},
434 		.ipv6_mask     = {
435 			.src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
436 			.dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
437 		},
438 		.src_port_mask = 0xFFFF,
439 		.dst_port_mask = 0xFFFF,
440 		.mac_addr_byte_mask = 0xFF,
441 		.tunnel_type_mask = 1,
442 		.tunnel_id_mask = 0xFFFFFFFF,
443 	},
444 	.drop_queue = 127,
445 };
446 
447 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
448 
449 struct queue_stats_mappings tx_queue_stats_mappings_array[MAX_TX_QUEUE_STATS_MAPPINGS];
450 struct queue_stats_mappings rx_queue_stats_mappings_array[MAX_RX_QUEUE_STATS_MAPPINGS];
451 
452 struct queue_stats_mappings *tx_queue_stats_mappings = tx_queue_stats_mappings_array;
453 struct queue_stats_mappings *rx_queue_stats_mappings = rx_queue_stats_mappings_array;
454 
455 uint16_t nb_tx_queue_stats_mappings = 0;
456 uint16_t nb_rx_queue_stats_mappings = 0;
457 
458 /*
459  * Display zero values by default for xstats
460  */
461 uint8_t xstats_hide_zero;
462 
463 unsigned int num_sockets = 0;
464 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
465 
466 #ifdef RTE_LIBRTE_BITRATE
467 /* Bitrate statistics */
468 struct rte_stats_bitrates *bitrate_data;
469 lcoreid_t bitrate_lcore_id;
470 uint8_t bitrate_enabled;
471 #endif
472 
473 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
474 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
475 
476 struct vxlan_encap_conf vxlan_encap_conf = {
477 	.select_ipv4 = 1,
478 	.select_vlan = 0,
479 	.select_tos_ttl = 0,
480 	.vni = "\x00\x00\x00",
481 	.udp_src = 0,
482 	.udp_dst = RTE_BE16(4789),
483 	.ipv4_src = RTE_IPV4(127, 0, 0, 1),
484 	.ipv4_dst = RTE_IPV4(255, 255, 255, 255),
485 	.ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
486 		"\x00\x00\x00\x00\x00\x00\x00\x01",
487 	.ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
488 		"\x00\x00\x00\x00\x00\x00\x11\x11",
489 	.vlan_tci = 0,
490 	.ip_tos = 0,
491 	.ip_ttl = 255,
492 	.eth_src = "\x00\x00\x00\x00\x00\x00",
493 	.eth_dst = "\xff\xff\xff\xff\xff\xff",
494 };
495 
496 struct nvgre_encap_conf nvgre_encap_conf = {
497 	.select_ipv4 = 1,
498 	.select_vlan = 0,
499 	.tni = "\x00\x00\x00",
500 	.ipv4_src = RTE_IPV4(127, 0, 0, 1),
501 	.ipv4_dst = RTE_IPV4(255, 255, 255, 255),
502 	.ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
503 		"\x00\x00\x00\x00\x00\x00\x00\x01",
504 	.ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
505 		"\x00\x00\x00\x00\x00\x00\x11\x11",
506 	.vlan_tci = 0,
507 	.eth_src = "\x00\x00\x00\x00\x00\x00",
508 	.eth_dst = "\xff\xff\xff\xff\xff\xff",
509 };
510 
511 /* Forward function declarations */
512 static void setup_attached_port(portid_t pi);
513 static void map_port_queue_stats_mapping_registers(portid_t pi,
514 						   struct rte_port *port);
515 static void check_all_ports_link_status(uint32_t port_mask);
516 static int eth_event_callback(portid_t port_id,
517 			      enum rte_eth_event_type type,
518 			      void *param, void *ret_param);
519 static void dev_event_callback(const char *device_name,
520 				enum rte_dev_event_type type,
521 				void *param);
522 
523 /*
524  * Check if all the ports are started.
525  * If yes, return positive value. If not, return zero.
526  */
527 static int all_ports_started(void);
528 
529 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
530 uint16_t gso_max_segment_size = RTE_ETHER_MAX_LEN - RTE_ETHER_CRC_LEN;
531 
532 /*
533  * Helper function to check if socket is already discovered.
534  * If yes, return positive value. If not, return zero.
535  */
536 int
537 new_socket_id(unsigned int socket_id)
538 {
539 	unsigned int i;
540 
541 	for (i = 0; i < num_sockets; i++) {
542 		if (socket_ids[i] == socket_id)
543 			return 0;
544 	}
545 	return 1;
546 }
547 
548 /*
549  * Setup default configuration.
550  */
551 static void
552 set_default_fwd_lcores_config(void)
553 {
554 	unsigned int i;
555 	unsigned int nb_lc;
556 	unsigned int sock_num;
557 
558 	nb_lc = 0;
559 	for (i = 0; i < RTE_MAX_LCORE; i++) {
560 		if (!rte_lcore_is_enabled(i))
561 			continue;
562 		sock_num = rte_lcore_to_socket_id(i);
563 		if (new_socket_id(sock_num)) {
564 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
565 				rte_exit(EXIT_FAILURE,
566 					 "Total sockets greater than %u\n",
567 					 RTE_MAX_NUMA_NODES);
568 			}
569 			socket_ids[num_sockets++] = sock_num;
570 		}
571 		if (i == rte_get_master_lcore())
572 			continue;
573 		fwd_lcores_cpuids[nb_lc++] = i;
574 	}
575 	nb_lcores = (lcoreid_t) nb_lc;
576 	nb_cfg_lcores = nb_lcores;
577 	nb_fwd_lcores = 1;
578 }
579 
580 static void
581 set_def_peer_eth_addrs(void)
582 {
583 	portid_t i;
584 
585 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
586 		peer_eth_addrs[i].addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
587 		peer_eth_addrs[i].addr_bytes[5] = i;
588 	}
589 }
590 
591 static void
592 set_default_fwd_ports_config(void)
593 {
594 	portid_t pt_id;
595 	int i = 0;
596 
597 	RTE_ETH_FOREACH_DEV(pt_id) {
598 		fwd_ports_ids[i++] = pt_id;
599 
600 		/* Update sockets info according to the attached device */
601 		int socket_id = rte_eth_dev_socket_id(pt_id);
602 		if (socket_id >= 0 && new_socket_id(socket_id)) {
603 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
604 				rte_exit(EXIT_FAILURE,
605 					 "Total sockets greater than %u\n",
606 					 RTE_MAX_NUMA_NODES);
607 			}
608 			socket_ids[num_sockets++] = socket_id;
609 		}
610 	}
611 
612 	nb_cfg_ports = nb_ports;
613 	nb_fwd_ports = nb_ports;
614 }
615 
616 void
617 set_def_fwd_config(void)
618 {
619 	set_default_fwd_lcores_config();
620 	set_def_peer_eth_addrs();
621 	set_default_fwd_ports_config();
622 }
623 
624 /* extremely pessimistic estimation of memory required to create a mempool */
625 static int
626 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
627 {
628 	unsigned int n_pages, mbuf_per_pg, leftover;
629 	uint64_t total_mem, mbuf_mem, obj_sz;
630 
631 	/* there is no good way to predict how much space the mempool will
632 	 * occupy because it will allocate chunks on the fly, and some of those
633 	 * will come from default DPDK memory while some will come from our
634 	 * external memory, so just assume 128MB will be enough for everyone.
635 	 */
636 	uint64_t hdr_mem = 128 << 20;
637 
638 	/* account for possible non-contiguousness */
639 	obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
640 	if (obj_sz > pgsz) {
641 		TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
642 		return -1;
643 	}
644 
645 	mbuf_per_pg = pgsz / obj_sz;
646 	leftover = (nb_mbufs % mbuf_per_pg) > 0;
647 	n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
648 
649 	mbuf_mem = n_pages * pgsz;
650 
651 	total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
652 
653 	if (total_mem > SIZE_MAX) {
654 		TESTPMD_LOG(ERR, "Memory size too big\n");
655 		return -1;
656 	}
657 	*out = (size_t)total_mem;
658 
659 	return 0;
660 }
661 
662 static int
663 pagesz_flags(uint64_t page_sz)
664 {
665 	/* as per mmap() manpage, all page sizes are log2 of page size
666 	 * shifted by MAP_HUGE_SHIFT
667 	 */
668 	int log2 = rte_log2_u64(page_sz);
669 
670 	return (log2 << HUGE_SHIFT);
671 }
672 
673 static void *
674 alloc_mem(size_t memsz, size_t pgsz, bool huge)
675 {
676 	void *addr;
677 	int flags;
678 
679 	/* allocate anonymous hugepages */
680 	flags = MAP_ANONYMOUS | MAP_PRIVATE;
681 	if (huge)
682 		flags |= HUGE_FLAG | pagesz_flags(pgsz);
683 
684 	addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
685 	if (addr == MAP_FAILED)
686 		return NULL;
687 
688 	return addr;
689 }
690 
691 struct extmem_param {
692 	void *addr;
693 	size_t len;
694 	size_t pgsz;
695 	rte_iova_t *iova_table;
696 	unsigned int iova_table_len;
697 };
698 
699 static int
700 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
701 		bool huge)
702 {
703 	uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
704 			RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
705 	unsigned int cur_page, n_pages, pgsz_idx;
706 	size_t mem_sz, cur_pgsz;
707 	rte_iova_t *iovas = NULL;
708 	void *addr;
709 	int ret;
710 
711 	for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
712 		/* skip anything that is too big */
713 		if (pgsizes[pgsz_idx] > SIZE_MAX)
714 			continue;
715 
716 		cur_pgsz = pgsizes[pgsz_idx];
717 
718 		/* if we were told not to allocate hugepages, override */
719 		if (!huge)
720 			cur_pgsz = sysconf(_SC_PAGESIZE);
721 
722 		ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
723 		if (ret < 0) {
724 			TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
725 			return -1;
726 		}
727 
728 		/* allocate our memory */
729 		addr = alloc_mem(mem_sz, cur_pgsz, huge);
730 
731 		/* if we couldn't allocate memory with a specified page size,
732 		 * that doesn't mean we can't do it with other page sizes, so
733 		 * try another one.
734 		 */
735 		if (addr == NULL)
736 			continue;
737 
738 		/* store IOVA addresses for every page in this memory area */
739 		n_pages = mem_sz / cur_pgsz;
740 
741 		iovas = malloc(sizeof(*iovas) * n_pages);
742 
743 		if (iovas == NULL) {
744 			TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
745 			goto fail;
746 		}
747 		/* lock memory if it's not huge pages */
748 		if (!huge)
749 			mlock(addr, mem_sz);
750 
751 		/* populate IOVA addresses */
752 		for (cur_page = 0; cur_page < n_pages; cur_page++) {
753 			rte_iova_t iova;
754 			size_t offset;
755 			void *cur;
756 
757 			offset = cur_pgsz * cur_page;
758 			cur = RTE_PTR_ADD(addr, offset);
759 
760 			/* touch the page before getting its IOVA */
761 			*(volatile char *)cur = 0;
762 
763 			iova = rte_mem_virt2iova(cur);
764 
765 			iovas[cur_page] = iova;
766 		}
767 
768 		break;
769 	}
770 	/* if we couldn't allocate anything */
771 	if (iovas == NULL)
772 		return -1;
773 
774 	param->addr = addr;
775 	param->len = mem_sz;
776 	param->pgsz = cur_pgsz;
777 	param->iova_table = iovas;
778 	param->iova_table_len = n_pages;
779 
780 	return 0;
781 fail:
782 	if (iovas)
783 		free(iovas);
784 	if (addr)
785 		munmap(addr, mem_sz);
786 
787 	return -1;
788 }
789 
790 static int
791 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
792 {
793 	struct extmem_param param;
794 	int socket_id, ret;
795 
796 	memset(&param, 0, sizeof(param));
797 
798 	/* check if our heap exists */
799 	socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
800 	if (socket_id < 0) {
801 		/* create our heap */
802 		ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
803 		if (ret < 0) {
804 			TESTPMD_LOG(ERR, "Cannot create heap\n");
805 			return -1;
806 		}
807 	}
808 
809 	ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
810 	if (ret < 0) {
811 		TESTPMD_LOG(ERR, "Cannot create memory area\n");
812 		return -1;
813 	}
814 
815 	/* we now have a valid memory area, so add it to heap */
816 	ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
817 			param.addr, param.len, param.iova_table,
818 			param.iova_table_len, param.pgsz);
819 
820 	/* when using VFIO, memory is automatically mapped for DMA by EAL */
821 
822 	/* not needed any more */
823 	free(param.iova_table);
824 
825 	if (ret < 0) {
826 		TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
827 		munmap(param.addr, param.len);
828 		return -1;
829 	}
830 
831 	/* success */
832 
833 	TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
834 			param.len >> 20);
835 
836 	return 0;
837 }
838 static void
839 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
840 	     struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
841 {
842 	uint16_t pid = 0;
843 	int ret;
844 
845 	RTE_ETH_FOREACH_DEV(pid) {
846 		struct rte_eth_dev *dev =
847 			&rte_eth_devices[pid];
848 
849 		ret = rte_dev_dma_unmap(dev->device, memhdr->addr, 0,
850 					memhdr->len);
851 		if (ret) {
852 			TESTPMD_LOG(DEBUG,
853 				    "unable to DMA unmap addr 0x%p "
854 				    "for device %s\n",
855 				    memhdr->addr, dev->data->name);
856 		}
857 	}
858 	ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
859 	if (ret) {
860 		TESTPMD_LOG(DEBUG,
861 			    "unable to un-register addr 0x%p\n", memhdr->addr);
862 	}
863 }
864 
865 static void
866 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
867 	   struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
868 {
869 	uint16_t pid = 0;
870 	size_t page_size = sysconf(_SC_PAGESIZE);
871 	int ret;
872 
873 	ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
874 				  page_size);
875 	if (ret) {
876 		TESTPMD_LOG(DEBUG,
877 			    "unable to register addr 0x%p\n", memhdr->addr);
878 		return;
879 	}
880 	RTE_ETH_FOREACH_DEV(pid) {
881 		struct rte_eth_dev *dev =
882 			&rte_eth_devices[pid];
883 
884 		ret = rte_dev_dma_map(dev->device, memhdr->addr, 0,
885 				      memhdr->len);
886 		if (ret) {
887 			TESTPMD_LOG(DEBUG,
888 				    "unable to DMA map addr 0x%p "
889 				    "for device %s\n",
890 				    memhdr->addr, dev->data->name);
891 		}
892 	}
893 }
894 
895 /*
896  * Configuration initialisation done once at init time.
897  */
898 static struct rte_mempool *
899 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
900 		 unsigned int socket_id)
901 {
902 	char pool_name[RTE_MEMPOOL_NAMESIZE];
903 	struct rte_mempool *rte_mp = NULL;
904 	uint32_t mb_size;
905 
906 	mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
907 	mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name));
908 
909 	TESTPMD_LOG(INFO,
910 		"create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
911 		pool_name, nb_mbuf, mbuf_seg_size, socket_id);
912 
913 	switch (mp_alloc_type) {
914 	case MP_ALLOC_NATIVE:
915 		{
916 			/* wrapper to rte_mempool_create() */
917 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
918 					rte_mbuf_best_mempool_ops());
919 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
920 				mb_mempool_cache, 0, mbuf_seg_size, socket_id);
921 			break;
922 		}
923 	case MP_ALLOC_ANON:
924 		{
925 			rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
926 				mb_size, (unsigned int) mb_mempool_cache,
927 				sizeof(struct rte_pktmbuf_pool_private),
928 				socket_id, mempool_flags);
929 			if (rte_mp == NULL)
930 				goto err;
931 
932 			if (rte_mempool_populate_anon(rte_mp) == 0) {
933 				rte_mempool_free(rte_mp);
934 				rte_mp = NULL;
935 				goto err;
936 			}
937 			rte_pktmbuf_pool_init(rte_mp, NULL);
938 			rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
939 			rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
940 			break;
941 		}
942 	case MP_ALLOC_XMEM:
943 	case MP_ALLOC_XMEM_HUGE:
944 		{
945 			int heap_socket;
946 			bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
947 
948 			if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
949 				rte_exit(EXIT_FAILURE, "Could not create external memory\n");
950 
951 			heap_socket =
952 				rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
953 			if (heap_socket < 0)
954 				rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
955 
956 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
957 					rte_mbuf_best_mempool_ops());
958 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
959 					mb_mempool_cache, 0, mbuf_seg_size,
960 					heap_socket);
961 			break;
962 		}
963 	default:
964 		{
965 			rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
966 		}
967 	}
968 
969 err:
970 	if (rte_mp == NULL) {
971 		rte_exit(EXIT_FAILURE,
972 			"Creation of mbuf pool for socket %u failed: %s\n",
973 			socket_id, rte_strerror(rte_errno));
974 	} else if (verbose_level > 0) {
975 		rte_mempool_dump(stdout, rte_mp);
976 	}
977 	return rte_mp;
978 }
979 
980 /*
981  * Check given socket id is valid or not with NUMA mode,
982  * if valid, return 0, else return -1
983  */
984 static int
985 check_socket_id(const unsigned int socket_id)
986 {
987 	static int warning_once = 0;
988 
989 	if (new_socket_id(socket_id)) {
990 		if (!warning_once && numa_support)
991 			printf("Warning: NUMA should be configured manually by"
992 			       " using --port-numa-config and"
993 			       " --ring-numa-config parameters along with"
994 			       " --numa.\n");
995 		warning_once = 1;
996 		return -1;
997 	}
998 	return 0;
999 }
1000 
1001 /*
1002  * Get the allowed maximum number of RX queues.
1003  * *pid return the port id which has minimal value of
1004  * max_rx_queues in all ports.
1005  */
1006 queueid_t
1007 get_allowed_max_nb_rxq(portid_t *pid)
1008 {
1009 	queueid_t allowed_max_rxq = MAX_QUEUE_ID;
1010 	portid_t pi;
1011 	struct rte_eth_dev_info dev_info;
1012 
1013 	RTE_ETH_FOREACH_DEV(pi) {
1014 		rte_eth_dev_info_get(pi, &dev_info);
1015 		if (dev_info.max_rx_queues < allowed_max_rxq) {
1016 			allowed_max_rxq = dev_info.max_rx_queues;
1017 			*pid = pi;
1018 		}
1019 	}
1020 	return allowed_max_rxq;
1021 }
1022 
1023 /*
1024  * Check input rxq is valid or not.
1025  * If input rxq is not greater than any of maximum number
1026  * of RX queues of all ports, it is valid.
1027  * if valid, return 0, else return -1
1028  */
1029 int
1030 check_nb_rxq(queueid_t rxq)
1031 {
1032 	queueid_t allowed_max_rxq;
1033 	portid_t pid = 0;
1034 
1035 	allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1036 	if (rxq > allowed_max_rxq) {
1037 		printf("Fail: input rxq (%u) can't be greater "
1038 		       "than max_rx_queues (%u) of port %u\n",
1039 		       rxq,
1040 		       allowed_max_rxq,
1041 		       pid);
1042 		return -1;
1043 	}
1044 	return 0;
1045 }
1046 
1047 /*
1048  * Get the allowed maximum number of TX queues.
1049  * *pid return the port id which has minimal value of
1050  * max_tx_queues in all ports.
1051  */
1052 queueid_t
1053 get_allowed_max_nb_txq(portid_t *pid)
1054 {
1055 	queueid_t allowed_max_txq = MAX_QUEUE_ID;
1056 	portid_t pi;
1057 	struct rte_eth_dev_info dev_info;
1058 
1059 	RTE_ETH_FOREACH_DEV(pi) {
1060 		rte_eth_dev_info_get(pi, &dev_info);
1061 		if (dev_info.max_tx_queues < allowed_max_txq) {
1062 			allowed_max_txq = dev_info.max_tx_queues;
1063 			*pid = pi;
1064 		}
1065 	}
1066 	return allowed_max_txq;
1067 }
1068 
1069 /*
1070  * Check input txq is valid or not.
1071  * If input txq is not greater than any of maximum number
1072  * of TX queues of all ports, it is valid.
1073  * if valid, return 0, else return -1
1074  */
1075 int
1076 check_nb_txq(queueid_t txq)
1077 {
1078 	queueid_t allowed_max_txq;
1079 	portid_t pid = 0;
1080 
1081 	allowed_max_txq = get_allowed_max_nb_txq(&pid);
1082 	if (txq > allowed_max_txq) {
1083 		printf("Fail: input txq (%u) can't be greater "
1084 		       "than max_tx_queues (%u) of port %u\n",
1085 		       txq,
1086 		       allowed_max_txq,
1087 		       pid);
1088 		return -1;
1089 	}
1090 	return 0;
1091 }
1092 
1093 static void
1094 init_config(void)
1095 {
1096 	portid_t pid;
1097 	struct rte_port *port;
1098 	struct rte_mempool *mbp;
1099 	unsigned int nb_mbuf_per_pool;
1100 	lcoreid_t  lc_id;
1101 	uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
1102 	struct rte_gro_param gro_param;
1103 	uint32_t gso_types;
1104 	uint16_t data_size;
1105 	bool warning = 0;
1106 	int k;
1107 
1108 	memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
1109 
1110 	/* Configuration of logical cores. */
1111 	fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1112 				sizeof(struct fwd_lcore *) * nb_lcores,
1113 				RTE_CACHE_LINE_SIZE);
1114 	if (fwd_lcores == NULL) {
1115 		rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1116 							"failed\n", nb_lcores);
1117 	}
1118 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1119 		fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1120 					       sizeof(struct fwd_lcore),
1121 					       RTE_CACHE_LINE_SIZE);
1122 		if (fwd_lcores[lc_id] == NULL) {
1123 			rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1124 								"failed\n");
1125 		}
1126 		fwd_lcores[lc_id]->cpuid_idx = lc_id;
1127 	}
1128 
1129 	RTE_ETH_FOREACH_DEV(pid) {
1130 		port = &ports[pid];
1131 		/* Apply default TxRx configuration for all ports */
1132 		port->dev_conf.txmode = tx_mode;
1133 		port->dev_conf.rxmode = rx_mode;
1134 		rte_eth_dev_info_get(pid, &port->dev_info);
1135 
1136 		if (!(port->dev_info.tx_offload_capa &
1137 		      DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1138 			port->dev_conf.txmode.offloads &=
1139 				~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1140 		if (!(port->dev_info.tx_offload_capa &
1141 			DEV_TX_OFFLOAD_MATCH_METADATA))
1142 			port->dev_conf.txmode.offloads &=
1143 				~DEV_TX_OFFLOAD_MATCH_METADATA;
1144 		if (numa_support) {
1145 			if (port_numa[pid] != NUMA_NO_CONFIG)
1146 				port_per_socket[port_numa[pid]]++;
1147 			else {
1148 				uint32_t socket_id = rte_eth_dev_socket_id(pid);
1149 
1150 				/*
1151 				 * if socket_id is invalid,
1152 				 * set to the first available socket.
1153 				 */
1154 				if (check_socket_id(socket_id) < 0)
1155 					socket_id = socket_ids[0];
1156 				port_per_socket[socket_id]++;
1157 			}
1158 		}
1159 
1160 		/* Apply Rx offloads configuration */
1161 		for (k = 0; k < port->dev_info.max_rx_queues; k++)
1162 			port->rx_conf[k].offloads =
1163 				port->dev_conf.rxmode.offloads;
1164 		/* Apply Tx offloads configuration */
1165 		for (k = 0; k < port->dev_info.max_tx_queues; k++)
1166 			port->tx_conf[k].offloads =
1167 				port->dev_conf.txmode.offloads;
1168 
1169 		/* set flag to initialize port/queue */
1170 		port->need_reconfig = 1;
1171 		port->need_reconfig_queues = 1;
1172 		port->tx_metadata = 0;
1173 
1174 		/* Check for maximum number of segments per MTU. Accordingly
1175 		 * update the mbuf data size.
1176 		 */
1177 		if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX) {
1178 			data_size = rx_mode.max_rx_pkt_len /
1179 				port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1180 
1181 			if ((data_size + RTE_PKTMBUF_HEADROOM) >
1182 							mbuf_data_size) {
1183 				mbuf_data_size = data_size +
1184 						 RTE_PKTMBUF_HEADROOM;
1185 				warning = 1;
1186 			}
1187 		}
1188 	}
1189 
1190 	if (warning)
1191 		TESTPMD_LOG(WARNING, "Configured mbuf size %hu\n",
1192 			    mbuf_data_size);
1193 
1194 	/*
1195 	 * Create pools of mbuf.
1196 	 * If NUMA support is disabled, create a single pool of mbuf in
1197 	 * socket 0 memory by default.
1198 	 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1199 	 *
1200 	 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1201 	 * nb_txd can be configured at run time.
1202 	 */
1203 	if (param_total_num_mbufs)
1204 		nb_mbuf_per_pool = param_total_num_mbufs;
1205 	else {
1206 		nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1207 			(nb_lcores * mb_mempool_cache) +
1208 			RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1209 		nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1210 	}
1211 
1212 	if (numa_support) {
1213 		uint8_t i;
1214 
1215 		for (i = 0; i < num_sockets; i++)
1216 			mempools[i] = mbuf_pool_create(mbuf_data_size,
1217 						       nb_mbuf_per_pool,
1218 						       socket_ids[i]);
1219 	} else {
1220 		if (socket_num == UMA_NO_CONFIG)
1221 			mempools[0] = mbuf_pool_create(mbuf_data_size,
1222 						       nb_mbuf_per_pool, 0);
1223 		else
1224 			mempools[socket_num] = mbuf_pool_create
1225 							(mbuf_data_size,
1226 							 nb_mbuf_per_pool,
1227 							 socket_num);
1228 	}
1229 
1230 	init_port_config();
1231 
1232 	gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1233 		DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1234 	/*
1235 	 * Records which Mbuf pool to use by each logical core, if needed.
1236 	 */
1237 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1238 		mbp = mbuf_pool_find(
1239 			rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]));
1240 
1241 		if (mbp == NULL)
1242 			mbp = mbuf_pool_find(0);
1243 		fwd_lcores[lc_id]->mbp = mbp;
1244 		/* initialize GSO context */
1245 		fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1246 		fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1247 		fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1248 		fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1249 			RTE_ETHER_CRC_LEN;
1250 		fwd_lcores[lc_id]->gso_ctx.flag = 0;
1251 	}
1252 
1253 	/* Configuration of packet forwarding streams. */
1254 	if (init_fwd_streams() < 0)
1255 		rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
1256 
1257 	fwd_config_setup();
1258 
1259 	/* create a gro context for each lcore */
1260 	gro_param.gro_types = RTE_GRO_TCP_IPV4;
1261 	gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1262 	gro_param.max_item_per_flow = MAX_PKT_BURST;
1263 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1264 		gro_param.socket_id = rte_lcore_to_socket_id(
1265 				fwd_lcores_cpuids[lc_id]);
1266 		fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1267 		if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1268 			rte_exit(EXIT_FAILURE,
1269 					"rte_gro_ctx_create() failed\n");
1270 		}
1271 	}
1272 
1273 #if defined RTE_LIBRTE_PMD_SOFTNIC
1274 	if (strcmp(cur_fwd_eng->fwd_mode_name, "softnic") == 0) {
1275 		RTE_ETH_FOREACH_DEV(pid) {
1276 			port = &ports[pid];
1277 			const char *driver = port->dev_info.driver_name;
1278 
1279 			if (strcmp(driver, "net_softnic") == 0)
1280 				port->softport.fwd_lcore_arg = fwd_lcores;
1281 		}
1282 	}
1283 #endif
1284 
1285 }
1286 
1287 
1288 void
1289 reconfig(portid_t new_port_id, unsigned socket_id)
1290 {
1291 	struct rte_port *port;
1292 
1293 	/* Reconfiguration of Ethernet ports. */
1294 	port = &ports[new_port_id];
1295 	rte_eth_dev_info_get(new_port_id, &port->dev_info);
1296 
1297 	/* set flag to initialize port/queue */
1298 	port->need_reconfig = 1;
1299 	port->need_reconfig_queues = 1;
1300 	port->socket_id = socket_id;
1301 
1302 	init_port_config();
1303 }
1304 
1305 
1306 int
1307 init_fwd_streams(void)
1308 {
1309 	portid_t pid;
1310 	struct rte_port *port;
1311 	streamid_t sm_id, nb_fwd_streams_new;
1312 	queueid_t q;
1313 
1314 	/* set socket id according to numa or not */
1315 	RTE_ETH_FOREACH_DEV(pid) {
1316 		port = &ports[pid];
1317 		if (nb_rxq > port->dev_info.max_rx_queues) {
1318 			printf("Fail: nb_rxq(%d) is greater than "
1319 				"max_rx_queues(%d)\n", nb_rxq,
1320 				port->dev_info.max_rx_queues);
1321 			return -1;
1322 		}
1323 		if (nb_txq > port->dev_info.max_tx_queues) {
1324 			printf("Fail: nb_txq(%d) is greater than "
1325 				"max_tx_queues(%d)\n", nb_txq,
1326 				port->dev_info.max_tx_queues);
1327 			return -1;
1328 		}
1329 		if (numa_support) {
1330 			if (port_numa[pid] != NUMA_NO_CONFIG)
1331 				port->socket_id = port_numa[pid];
1332 			else {
1333 				port->socket_id = rte_eth_dev_socket_id(pid);
1334 
1335 				/*
1336 				 * if socket_id is invalid,
1337 				 * set to the first available socket.
1338 				 */
1339 				if (check_socket_id(port->socket_id) < 0)
1340 					port->socket_id = socket_ids[0];
1341 			}
1342 		}
1343 		else {
1344 			if (socket_num == UMA_NO_CONFIG)
1345 				port->socket_id = 0;
1346 			else
1347 				port->socket_id = socket_num;
1348 		}
1349 	}
1350 
1351 	q = RTE_MAX(nb_rxq, nb_txq);
1352 	if (q == 0) {
1353 		printf("Fail: Cannot allocate fwd streams as number of queues is 0\n");
1354 		return -1;
1355 	}
1356 	nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1357 	if (nb_fwd_streams_new == nb_fwd_streams)
1358 		return 0;
1359 	/* clear the old */
1360 	if (fwd_streams != NULL) {
1361 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1362 			if (fwd_streams[sm_id] == NULL)
1363 				continue;
1364 			rte_free(fwd_streams[sm_id]);
1365 			fwd_streams[sm_id] = NULL;
1366 		}
1367 		rte_free(fwd_streams);
1368 		fwd_streams = NULL;
1369 	}
1370 
1371 	/* init new */
1372 	nb_fwd_streams = nb_fwd_streams_new;
1373 	if (nb_fwd_streams) {
1374 		fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1375 			sizeof(struct fwd_stream *) * nb_fwd_streams,
1376 			RTE_CACHE_LINE_SIZE);
1377 		if (fwd_streams == NULL)
1378 			rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1379 				 " (struct fwd_stream *)) failed\n",
1380 				 nb_fwd_streams);
1381 
1382 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1383 			fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1384 				" struct fwd_stream", sizeof(struct fwd_stream),
1385 				RTE_CACHE_LINE_SIZE);
1386 			if (fwd_streams[sm_id] == NULL)
1387 				rte_exit(EXIT_FAILURE, "rte_zmalloc"
1388 					 "(struct fwd_stream) failed\n");
1389 		}
1390 	}
1391 
1392 	return 0;
1393 }
1394 
1395 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1396 static void
1397 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1398 {
1399 	unsigned int total_burst;
1400 	unsigned int nb_burst;
1401 	unsigned int burst_stats[3];
1402 	uint16_t pktnb_stats[3];
1403 	uint16_t nb_pkt;
1404 	int burst_percent[3];
1405 
1406 	/*
1407 	 * First compute the total number of packet bursts and the
1408 	 * two highest numbers of bursts of the same number of packets.
1409 	 */
1410 	total_burst = 0;
1411 	burst_stats[0] = burst_stats[1] = burst_stats[2] = 0;
1412 	pktnb_stats[0] = pktnb_stats[1] = pktnb_stats[2] = 0;
1413 	for (nb_pkt = 0; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1414 		nb_burst = pbs->pkt_burst_spread[nb_pkt];
1415 		if (nb_burst == 0)
1416 			continue;
1417 		total_burst += nb_burst;
1418 		if (nb_burst > burst_stats[0]) {
1419 			burst_stats[1] = burst_stats[0];
1420 			pktnb_stats[1] = pktnb_stats[0];
1421 			burst_stats[0] = nb_burst;
1422 			pktnb_stats[0] = nb_pkt;
1423 		} else if (nb_burst > burst_stats[1]) {
1424 			burst_stats[1] = nb_burst;
1425 			pktnb_stats[1] = nb_pkt;
1426 		}
1427 	}
1428 	if (total_burst == 0)
1429 		return;
1430 	burst_percent[0] = (burst_stats[0] * 100) / total_burst;
1431 	printf("  %s-bursts : %u [%d%% of %d pkts", rx_tx, total_burst,
1432 	       burst_percent[0], (int) pktnb_stats[0]);
1433 	if (burst_stats[0] == total_burst) {
1434 		printf("]\n");
1435 		return;
1436 	}
1437 	if (burst_stats[0] + burst_stats[1] == total_burst) {
1438 		printf(" + %d%% of %d pkts]\n",
1439 		       100 - burst_percent[0], pktnb_stats[1]);
1440 		return;
1441 	}
1442 	burst_percent[1] = (burst_stats[1] * 100) / total_burst;
1443 	burst_percent[2] = 100 - (burst_percent[0] + burst_percent[1]);
1444 	if ((burst_percent[1] == 0) || (burst_percent[2] == 0)) {
1445 		printf(" + %d%% of others]\n", 100 - burst_percent[0]);
1446 		return;
1447 	}
1448 	printf(" + %d%% of %d pkts + %d%% of others]\n",
1449 	       burst_percent[1], (int) pktnb_stats[1], burst_percent[2]);
1450 }
1451 #endif /* RTE_TEST_PMD_RECORD_BURST_STATS */
1452 
1453 static void
1454 fwd_stream_stats_display(streamid_t stream_id)
1455 {
1456 	struct fwd_stream *fs;
1457 	static const char *fwd_top_stats_border = "-------";
1458 
1459 	fs = fwd_streams[stream_id];
1460 	if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1461 	    (fs->fwd_dropped == 0))
1462 		return;
1463 	printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1464 	       "TX Port=%2d/Queue=%2d %s\n",
1465 	       fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1466 	       fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1467 	printf("  RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1468 	       " TX-dropped: %-14"PRIu64,
1469 	       fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1470 
1471 	/* if checksum mode */
1472 	if (cur_fwd_eng == &csum_fwd_engine) {
1473 		printf("  RX- bad IP checksum: %-14"PRIu64
1474 		       "  Rx- bad L4 checksum: %-14"PRIu64
1475 		       " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1476 			fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1477 			fs->rx_bad_outer_l4_csum);
1478 	} else {
1479 		printf("\n");
1480 	}
1481 
1482 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1483 	pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1484 	pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1485 #endif
1486 }
1487 
1488 void
1489 fwd_stats_display(void)
1490 {
1491 	static const char *fwd_stats_border = "----------------------";
1492 	static const char *acc_stats_border = "+++++++++++++++";
1493 	struct {
1494 		struct fwd_stream *rx_stream;
1495 		struct fwd_stream *tx_stream;
1496 		uint64_t tx_dropped;
1497 		uint64_t rx_bad_ip_csum;
1498 		uint64_t rx_bad_l4_csum;
1499 		uint64_t rx_bad_outer_l4_csum;
1500 	} ports_stats[RTE_MAX_ETHPORTS];
1501 	uint64_t total_rx_dropped = 0;
1502 	uint64_t total_tx_dropped = 0;
1503 	uint64_t total_rx_nombuf = 0;
1504 	struct rte_eth_stats stats;
1505 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1506 	uint64_t fwd_cycles = 0;
1507 #endif
1508 	uint64_t total_recv = 0;
1509 	uint64_t total_xmit = 0;
1510 	struct rte_port *port;
1511 	streamid_t sm_id;
1512 	portid_t pt_id;
1513 	int i;
1514 
1515 	memset(ports_stats, 0, sizeof(ports_stats));
1516 
1517 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1518 		struct fwd_stream *fs = fwd_streams[sm_id];
1519 
1520 		if (cur_fwd_config.nb_fwd_streams >
1521 		    cur_fwd_config.nb_fwd_ports) {
1522 			fwd_stream_stats_display(sm_id);
1523 		} else {
1524 			ports_stats[fs->tx_port].tx_stream = fs;
1525 			ports_stats[fs->rx_port].rx_stream = fs;
1526 		}
1527 
1528 		ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1529 
1530 		ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1531 		ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1532 		ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1533 				fs->rx_bad_outer_l4_csum;
1534 
1535 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1536 		fwd_cycles += fs->core_cycles;
1537 #endif
1538 	}
1539 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1540 		uint8_t j;
1541 
1542 		pt_id = fwd_ports_ids[i];
1543 		port = &ports[pt_id];
1544 
1545 		rte_eth_stats_get(pt_id, &stats);
1546 		stats.ipackets -= port->stats.ipackets;
1547 		stats.opackets -= port->stats.opackets;
1548 		stats.ibytes -= port->stats.ibytes;
1549 		stats.obytes -= port->stats.obytes;
1550 		stats.imissed -= port->stats.imissed;
1551 		stats.oerrors -= port->stats.oerrors;
1552 		stats.rx_nombuf -= port->stats.rx_nombuf;
1553 
1554 		total_recv += stats.ipackets;
1555 		total_xmit += stats.opackets;
1556 		total_rx_dropped += stats.imissed;
1557 		total_tx_dropped += ports_stats[pt_id].tx_dropped;
1558 		total_tx_dropped += stats.oerrors;
1559 		total_rx_nombuf  += stats.rx_nombuf;
1560 
1561 		printf("\n  %s Forward statistics for port %-2d %s\n",
1562 		       fwd_stats_border, pt_id, fwd_stats_border);
1563 
1564 		if (!port->rx_queue_stats_mapping_enabled &&
1565 		    !port->tx_queue_stats_mapping_enabled) {
1566 			printf("  RX-packets: %-14"PRIu64
1567 			       " RX-dropped: %-14"PRIu64
1568 			       "RX-total: %-"PRIu64"\n",
1569 			       stats.ipackets, stats.imissed,
1570 			       stats.ipackets + stats.imissed);
1571 
1572 			if (cur_fwd_eng == &csum_fwd_engine)
1573 				printf("  Bad-ipcsum: %-14"PRIu64
1574 				       " Bad-l4csum: %-14"PRIu64
1575 				       "Bad-outer-l4csum: %-14"PRIu64"\n",
1576 				       ports_stats[pt_id].rx_bad_ip_csum,
1577 				       ports_stats[pt_id].rx_bad_l4_csum,
1578 				       ports_stats[pt_id].rx_bad_outer_l4_csum);
1579 			if (stats.ierrors + stats.rx_nombuf > 0) {
1580 				printf("  RX-error: %-"PRIu64"\n",
1581 				       stats.ierrors);
1582 				printf("  RX-nombufs: %-14"PRIu64"\n",
1583 				       stats.rx_nombuf);
1584 			}
1585 
1586 			printf("  TX-packets: %-14"PRIu64
1587 			       " TX-dropped: %-14"PRIu64
1588 			       "TX-total: %-"PRIu64"\n",
1589 			       stats.opackets, ports_stats[pt_id].tx_dropped,
1590 			       stats.opackets + ports_stats[pt_id].tx_dropped);
1591 		} else {
1592 			printf("  RX-packets:             %14"PRIu64
1593 			       "    RX-dropped:%14"PRIu64
1594 			       "    RX-total:%14"PRIu64"\n",
1595 			       stats.ipackets, stats.imissed,
1596 			       stats.ipackets + stats.imissed);
1597 
1598 			if (cur_fwd_eng == &csum_fwd_engine)
1599 				printf("  Bad-ipcsum:%14"PRIu64
1600 				       "    Bad-l4csum:%14"PRIu64
1601 				       "    Bad-outer-l4csum: %-14"PRIu64"\n",
1602 				       ports_stats[pt_id].rx_bad_ip_csum,
1603 				       ports_stats[pt_id].rx_bad_l4_csum,
1604 				       ports_stats[pt_id].rx_bad_outer_l4_csum);
1605 			if ((stats.ierrors + stats.rx_nombuf) > 0) {
1606 				printf("  RX-error:%"PRIu64"\n", stats.ierrors);
1607 				printf("  RX-nombufs:             %14"PRIu64"\n",
1608 				       stats.rx_nombuf);
1609 			}
1610 
1611 			printf("  TX-packets:             %14"PRIu64
1612 			       "    TX-dropped:%14"PRIu64
1613 			       "    TX-total:%14"PRIu64"\n",
1614 			       stats.opackets, ports_stats[pt_id].tx_dropped,
1615 			       stats.opackets + ports_stats[pt_id].tx_dropped);
1616 		}
1617 
1618 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1619 		if (ports_stats[pt_id].rx_stream)
1620 			pkt_burst_stats_display("RX",
1621 				&ports_stats[pt_id].rx_stream->rx_burst_stats);
1622 		if (ports_stats[pt_id].tx_stream)
1623 			pkt_burst_stats_display("TX",
1624 				&ports_stats[pt_id].tx_stream->tx_burst_stats);
1625 #endif
1626 
1627 		if (port->rx_queue_stats_mapping_enabled) {
1628 			printf("\n");
1629 			for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1630 				printf("  Stats reg %2d RX-packets:%14"PRIu64
1631 				       "     RX-errors:%14"PRIu64
1632 				       "    RX-bytes:%14"PRIu64"\n",
1633 				       j, stats.q_ipackets[j],
1634 				       stats.q_errors[j], stats.q_ibytes[j]);
1635 			}
1636 			printf("\n");
1637 		}
1638 		if (port->tx_queue_stats_mapping_enabled) {
1639 			for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1640 				printf("  Stats reg %2d TX-packets:%14"PRIu64
1641 				       "                                 TX-bytes:%14"
1642 				       PRIu64"\n",
1643 				       j, stats.q_opackets[j],
1644 				       stats.q_obytes[j]);
1645 			}
1646 		}
1647 
1648 		printf("  %s--------------------------------%s\n",
1649 		       fwd_stats_border, fwd_stats_border);
1650 	}
1651 
1652 	printf("\n  %s Accumulated forward statistics for all ports"
1653 	       "%s\n",
1654 	       acc_stats_border, acc_stats_border);
1655 	printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1656 	       "%-"PRIu64"\n"
1657 	       "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1658 	       "%-"PRIu64"\n",
1659 	       total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1660 	       total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1661 	if (total_rx_nombuf > 0)
1662 		printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1663 	printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
1664 	       "%s\n",
1665 	       acc_stats_border, acc_stats_border);
1666 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1667 	if (total_recv > 0)
1668 		printf("\n  CPU cycles/packet=%u (total cycles="
1669 		       "%"PRIu64" / total RX packets=%"PRIu64")\n",
1670 		       (unsigned int)(fwd_cycles / total_recv),
1671 		       fwd_cycles, total_recv);
1672 #endif
1673 }
1674 
1675 void
1676 fwd_stats_reset(void)
1677 {
1678 	streamid_t sm_id;
1679 	portid_t pt_id;
1680 	int i;
1681 
1682 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1683 		pt_id = fwd_ports_ids[i];
1684 		rte_eth_stats_get(pt_id, &ports[pt_id].stats);
1685 	}
1686 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1687 		struct fwd_stream *fs = fwd_streams[sm_id];
1688 
1689 		fs->rx_packets = 0;
1690 		fs->tx_packets = 0;
1691 		fs->fwd_dropped = 0;
1692 		fs->rx_bad_ip_csum = 0;
1693 		fs->rx_bad_l4_csum = 0;
1694 		fs->rx_bad_outer_l4_csum = 0;
1695 
1696 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1697 		memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
1698 		memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
1699 #endif
1700 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1701 		fs->core_cycles = 0;
1702 #endif
1703 	}
1704 }
1705 
1706 static void
1707 flush_fwd_rx_queues(void)
1708 {
1709 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1710 	portid_t  rxp;
1711 	portid_t port_id;
1712 	queueid_t rxq;
1713 	uint16_t  nb_rx;
1714 	uint16_t  i;
1715 	uint8_t   j;
1716 	uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
1717 	uint64_t timer_period;
1718 
1719 	/* convert to number of cycles */
1720 	timer_period = rte_get_timer_hz(); /* 1 second timeout */
1721 
1722 	for (j = 0; j < 2; j++) {
1723 		for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
1724 			for (rxq = 0; rxq < nb_rxq; rxq++) {
1725 				port_id = fwd_ports_ids[rxp];
1726 				/**
1727 				* testpmd can stuck in the below do while loop
1728 				* if rte_eth_rx_burst() always returns nonzero
1729 				* packets. So timer is added to exit this loop
1730 				* after 1sec timer expiry.
1731 				*/
1732 				prev_tsc = rte_rdtsc();
1733 				do {
1734 					nb_rx = rte_eth_rx_burst(port_id, rxq,
1735 						pkts_burst, MAX_PKT_BURST);
1736 					for (i = 0; i < nb_rx; i++)
1737 						rte_pktmbuf_free(pkts_burst[i]);
1738 
1739 					cur_tsc = rte_rdtsc();
1740 					diff_tsc = cur_tsc - prev_tsc;
1741 					timer_tsc += diff_tsc;
1742 				} while ((nb_rx > 0) &&
1743 					(timer_tsc < timer_period));
1744 				timer_tsc = 0;
1745 			}
1746 		}
1747 		rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
1748 	}
1749 }
1750 
1751 static void
1752 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
1753 {
1754 	struct fwd_stream **fsm;
1755 	streamid_t nb_fs;
1756 	streamid_t sm_id;
1757 #ifdef RTE_LIBRTE_BITRATE
1758 	uint64_t tics_per_1sec;
1759 	uint64_t tics_datum;
1760 	uint64_t tics_current;
1761 	uint16_t i, cnt_ports;
1762 
1763 	cnt_ports = nb_ports;
1764 	tics_datum = rte_rdtsc();
1765 	tics_per_1sec = rte_get_timer_hz();
1766 #endif
1767 	fsm = &fwd_streams[fc->stream_idx];
1768 	nb_fs = fc->stream_nb;
1769 	do {
1770 		for (sm_id = 0; sm_id < nb_fs; sm_id++)
1771 			(*pkt_fwd)(fsm[sm_id]);
1772 #ifdef RTE_LIBRTE_BITRATE
1773 		if (bitrate_enabled != 0 &&
1774 				bitrate_lcore_id == rte_lcore_id()) {
1775 			tics_current = rte_rdtsc();
1776 			if (tics_current - tics_datum >= tics_per_1sec) {
1777 				/* Periodic bitrate calculation */
1778 				for (i = 0; i < cnt_ports; i++)
1779 					rte_stats_bitrate_calc(bitrate_data,
1780 						ports_ids[i]);
1781 				tics_datum = tics_current;
1782 			}
1783 		}
1784 #endif
1785 #ifdef RTE_LIBRTE_LATENCY_STATS
1786 		if (latencystats_enabled != 0 &&
1787 				latencystats_lcore_id == rte_lcore_id())
1788 			rte_latencystats_update();
1789 #endif
1790 
1791 	} while (! fc->stopped);
1792 }
1793 
1794 static int
1795 start_pkt_forward_on_core(void *fwd_arg)
1796 {
1797 	run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
1798 			     cur_fwd_config.fwd_eng->packet_fwd);
1799 	return 0;
1800 }
1801 
1802 /*
1803  * Run the TXONLY packet forwarding engine to send a single burst of packets.
1804  * Used to start communication flows in network loopback test configurations.
1805  */
1806 static int
1807 run_one_txonly_burst_on_core(void *fwd_arg)
1808 {
1809 	struct fwd_lcore *fwd_lc;
1810 	struct fwd_lcore tmp_lcore;
1811 
1812 	fwd_lc = (struct fwd_lcore *) fwd_arg;
1813 	tmp_lcore = *fwd_lc;
1814 	tmp_lcore.stopped = 1;
1815 	run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
1816 	return 0;
1817 }
1818 
1819 /*
1820  * Launch packet forwarding:
1821  *     - Setup per-port forwarding context.
1822  *     - launch logical cores with their forwarding configuration.
1823  */
1824 static void
1825 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
1826 {
1827 	port_fwd_begin_t port_fwd_begin;
1828 	unsigned int i;
1829 	unsigned int lc_id;
1830 	int diag;
1831 
1832 	port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
1833 	if (port_fwd_begin != NULL) {
1834 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1835 			(*port_fwd_begin)(fwd_ports_ids[i]);
1836 	}
1837 	for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
1838 		lc_id = fwd_lcores_cpuids[i];
1839 		if ((interactive == 0) || (lc_id != rte_lcore_id())) {
1840 			fwd_lcores[i]->stopped = 0;
1841 			diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
1842 						     fwd_lcores[i], lc_id);
1843 			if (diag != 0)
1844 				printf("launch lcore %u failed - diag=%d\n",
1845 				       lc_id, diag);
1846 		}
1847 	}
1848 }
1849 
1850 /*
1851  * Launch packet forwarding configuration.
1852  */
1853 void
1854 start_packet_forwarding(int with_tx_first)
1855 {
1856 	port_fwd_begin_t port_fwd_begin;
1857 	port_fwd_end_t  port_fwd_end;
1858 	struct rte_port *port;
1859 	unsigned int i;
1860 	portid_t   pt_id;
1861 
1862 	if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
1863 		rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
1864 
1865 	if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
1866 		rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
1867 
1868 	if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
1869 		strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
1870 		(!nb_rxq || !nb_txq))
1871 		rte_exit(EXIT_FAILURE,
1872 			"Either rxq or txq are 0, cannot use %s fwd mode\n",
1873 			cur_fwd_eng->fwd_mode_name);
1874 
1875 	if (all_ports_started() == 0) {
1876 		printf("Not all ports were started\n");
1877 		return;
1878 	}
1879 	if (test_done == 0) {
1880 		printf("Packet forwarding already started\n");
1881 		return;
1882 	}
1883 
1884 
1885 	if(dcb_test) {
1886 		for (i = 0; i < nb_fwd_ports; i++) {
1887 			pt_id = fwd_ports_ids[i];
1888 			port = &ports[pt_id];
1889 			if (!port->dcb_flag) {
1890 				printf("In DCB mode, all forwarding ports must "
1891                                        "be configured in this mode.\n");
1892 				return;
1893 			}
1894 		}
1895 		if (nb_fwd_lcores == 1) {
1896 			printf("In DCB mode,the nb forwarding cores "
1897                                "should be larger than 1.\n");
1898 			return;
1899 		}
1900 	}
1901 	test_done = 0;
1902 
1903 	fwd_config_setup();
1904 
1905 	if(!no_flush_rx)
1906 		flush_fwd_rx_queues();
1907 
1908 	pkt_fwd_config_display(&cur_fwd_config);
1909 	rxtx_config_display();
1910 
1911 	fwd_stats_reset();
1912 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1913 		pt_id = fwd_ports_ids[i];
1914 		port = &ports[pt_id];
1915 		map_port_queue_stats_mapping_registers(pt_id, port);
1916 	}
1917 	if (with_tx_first) {
1918 		port_fwd_begin = tx_only_engine.port_fwd_begin;
1919 		if (port_fwd_begin != NULL) {
1920 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1921 				(*port_fwd_begin)(fwd_ports_ids[i]);
1922 		}
1923 		while (with_tx_first--) {
1924 			launch_packet_forwarding(
1925 					run_one_txonly_burst_on_core);
1926 			rte_eal_mp_wait_lcore();
1927 		}
1928 		port_fwd_end = tx_only_engine.port_fwd_end;
1929 		if (port_fwd_end != NULL) {
1930 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1931 				(*port_fwd_end)(fwd_ports_ids[i]);
1932 		}
1933 	}
1934 	launch_packet_forwarding(start_pkt_forward_on_core);
1935 }
1936 
1937 void
1938 stop_packet_forwarding(void)
1939 {
1940 	port_fwd_end_t port_fwd_end;
1941 	lcoreid_t lc_id;
1942 	portid_t pt_id;
1943 	int i;
1944 
1945 	if (test_done) {
1946 		printf("Packet forwarding not started\n");
1947 		return;
1948 	}
1949 	printf("Telling cores to stop...");
1950 	for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
1951 		fwd_lcores[lc_id]->stopped = 1;
1952 	printf("\nWaiting for lcores to finish...\n");
1953 	rte_eal_mp_wait_lcore();
1954 	port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
1955 	if (port_fwd_end != NULL) {
1956 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1957 			pt_id = fwd_ports_ids[i];
1958 			(*port_fwd_end)(pt_id);
1959 		}
1960 	}
1961 
1962 	fwd_stats_display();
1963 
1964 	printf("\nDone.\n");
1965 	test_done = 1;
1966 }
1967 
1968 void
1969 dev_set_link_up(portid_t pid)
1970 {
1971 	if (rte_eth_dev_set_link_up(pid) < 0)
1972 		printf("\nSet link up fail.\n");
1973 }
1974 
1975 void
1976 dev_set_link_down(portid_t pid)
1977 {
1978 	if (rte_eth_dev_set_link_down(pid) < 0)
1979 		printf("\nSet link down fail.\n");
1980 }
1981 
1982 static int
1983 all_ports_started(void)
1984 {
1985 	portid_t pi;
1986 	struct rte_port *port;
1987 
1988 	RTE_ETH_FOREACH_DEV(pi) {
1989 		port = &ports[pi];
1990 		/* Check if there is a port which is not started */
1991 		if ((port->port_status != RTE_PORT_STARTED) &&
1992 			(port->slave_flag == 0))
1993 			return 0;
1994 	}
1995 
1996 	/* No port is not started */
1997 	return 1;
1998 }
1999 
2000 int
2001 port_is_stopped(portid_t port_id)
2002 {
2003 	struct rte_port *port = &ports[port_id];
2004 
2005 	if ((port->port_status != RTE_PORT_STOPPED) &&
2006 	    (port->slave_flag == 0))
2007 		return 0;
2008 	return 1;
2009 }
2010 
2011 int
2012 all_ports_stopped(void)
2013 {
2014 	portid_t pi;
2015 
2016 	RTE_ETH_FOREACH_DEV(pi) {
2017 		if (!port_is_stopped(pi))
2018 			return 0;
2019 	}
2020 
2021 	return 1;
2022 }
2023 
2024 int
2025 port_is_started(portid_t port_id)
2026 {
2027 	if (port_id_is_invalid(port_id, ENABLED_WARN))
2028 		return 0;
2029 
2030 	if (ports[port_id].port_status != RTE_PORT_STARTED)
2031 		return 0;
2032 
2033 	return 1;
2034 }
2035 
2036 int
2037 start_port(portid_t pid)
2038 {
2039 	int diag, need_check_link_status = -1;
2040 	portid_t pi;
2041 	queueid_t qi;
2042 	struct rte_port *port;
2043 	struct rte_ether_addr mac_addr;
2044 
2045 	if (port_id_is_invalid(pid, ENABLED_WARN))
2046 		return 0;
2047 
2048 	if(dcb_config)
2049 		dcb_test = 1;
2050 	RTE_ETH_FOREACH_DEV(pi) {
2051 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2052 			continue;
2053 
2054 		need_check_link_status = 0;
2055 		port = &ports[pi];
2056 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2057 						 RTE_PORT_HANDLING) == 0) {
2058 			printf("Port %d is now not stopped\n", pi);
2059 			continue;
2060 		}
2061 
2062 		if (port->need_reconfig > 0) {
2063 			port->need_reconfig = 0;
2064 
2065 			if (flow_isolate_all) {
2066 				int ret = port_flow_isolate(pi, 1);
2067 				if (ret) {
2068 					printf("Failed to apply isolated"
2069 					       " mode on port %d\n", pi);
2070 					return -1;
2071 				}
2072 			}
2073 			configure_rxtx_dump_callbacks(0);
2074 			printf("Configuring Port %d (socket %u)\n", pi,
2075 					port->socket_id);
2076 			/* configure port */
2077 			diag = rte_eth_dev_configure(pi, nb_rxq, nb_txq,
2078 						&(port->dev_conf));
2079 			if (diag != 0) {
2080 				if (rte_atomic16_cmpset(&(port->port_status),
2081 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2082 					printf("Port %d can not be set back "
2083 							"to stopped\n", pi);
2084 				printf("Fail to configure port %d\n", pi);
2085 				/* try to reconfigure port next time */
2086 				port->need_reconfig = 1;
2087 				return -1;
2088 			}
2089 		}
2090 		if (port->need_reconfig_queues > 0) {
2091 			port->need_reconfig_queues = 0;
2092 			/* setup tx queues */
2093 			for (qi = 0; qi < nb_txq; qi++) {
2094 				if ((numa_support) &&
2095 					(txring_numa[pi] != NUMA_NO_CONFIG))
2096 					diag = rte_eth_tx_queue_setup(pi, qi,
2097 						port->nb_tx_desc[qi],
2098 						txring_numa[pi],
2099 						&(port->tx_conf[qi]));
2100 				else
2101 					diag = rte_eth_tx_queue_setup(pi, qi,
2102 						port->nb_tx_desc[qi],
2103 						port->socket_id,
2104 						&(port->tx_conf[qi]));
2105 
2106 				if (diag == 0)
2107 					continue;
2108 
2109 				/* Fail to setup tx queue, return */
2110 				if (rte_atomic16_cmpset(&(port->port_status),
2111 							RTE_PORT_HANDLING,
2112 							RTE_PORT_STOPPED) == 0)
2113 					printf("Port %d can not be set back "
2114 							"to stopped\n", pi);
2115 				printf("Fail to configure port %d tx queues\n",
2116 				       pi);
2117 				/* try to reconfigure queues next time */
2118 				port->need_reconfig_queues = 1;
2119 				return -1;
2120 			}
2121 			for (qi = 0; qi < nb_rxq; qi++) {
2122 				/* setup rx queues */
2123 				if ((numa_support) &&
2124 					(rxring_numa[pi] != NUMA_NO_CONFIG)) {
2125 					struct rte_mempool * mp =
2126 						mbuf_pool_find(rxring_numa[pi]);
2127 					if (mp == NULL) {
2128 						printf("Failed to setup RX queue:"
2129 							"No mempool allocation"
2130 							" on the socket %d\n",
2131 							rxring_numa[pi]);
2132 						return -1;
2133 					}
2134 
2135 					diag = rte_eth_rx_queue_setup(pi, qi,
2136 					     port->nb_rx_desc[qi],
2137 					     rxring_numa[pi],
2138 					     &(port->rx_conf[qi]),
2139 					     mp);
2140 				} else {
2141 					struct rte_mempool *mp =
2142 						mbuf_pool_find(port->socket_id);
2143 					if (mp == NULL) {
2144 						printf("Failed to setup RX queue:"
2145 							"No mempool allocation"
2146 							" on the socket %d\n",
2147 							port->socket_id);
2148 						return -1;
2149 					}
2150 					diag = rte_eth_rx_queue_setup(pi, qi,
2151 					     port->nb_rx_desc[qi],
2152 					     port->socket_id,
2153 					     &(port->rx_conf[qi]),
2154 					     mp);
2155 				}
2156 				if (diag == 0)
2157 					continue;
2158 
2159 				/* Fail to setup rx queue, return */
2160 				if (rte_atomic16_cmpset(&(port->port_status),
2161 							RTE_PORT_HANDLING,
2162 							RTE_PORT_STOPPED) == 0)
2163 					printf("Port %d can not be set back "
2164 							"to stopped\n", pi);
2165 				printf("Fail to configure port %d rx queues\n",
2166 				       pi);
2167 				/* try to reconfigure queues next time */
2168 				port->need_reconfig_queues = 1;
2169 				return -1;
2170 			}
2171 		}
2172 		configure_rxtx_dump_callbacks(verbose_level);
2173 		/* start port */
2174 		if (rte_eth_dev_start(pi) < 0) {
2175 			printf("Fail to start port %d\n", pi);
2176 
2177 			/* Fail to setup rx queue, return */
2178 			if (rte_atomic16_cmpset(&(port->port_status),
2179 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2180 				printf("Port %d can not be set back to "
2181 							"stopped\n", pi);
2182 			continue;
2183 		}
2184 
2185 		if (rte_atomic16_cmpset(&(port->port_status),
2186 			RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2187 			printf("Port %d can not be set into started\n", pi);
2188 
2189 		rte_eth_macaddr_get(pi, &mac_addr);
2190 		printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2191 				mac_addr.addr_bytes[0], mac_addr.addr_bytes[1],
2192 				mac_addr.addr_bytes[2], mac_addr.addr_bytes[3],
2193 				mac_addr.addr_bytes[4], mac_addr.addr_bytes[5]);
2194 
2195 		/* at least one port started, need checking link status */
2196 		need_check_link_status = 1;
2197 	}
2198 
2199 	if (need_check_link_status == 1 && !no_link_check)
2200 		check_all_ports_link_status(RTE_PORT_ALL);
2201 	else if (need_check_link_status == 0)
2202 		printf("Please stop the ports first\n");
2203 
2204 	printf("Done\n");
2205 	return 0;
2206 }
2207 
2208 void
2209 stop_port(portid_t pid)
2210 {
2211 	portid_t pi;
2212 	struct rte_port *port;
2213 	int need_check_link_status = 0;
2214 
2215 	if (dcb_test) {
2216 		dcb_test = 0;
2217 		dcb_config = 0;
2218 	}
2219 
2220 	if (port_id_is_invalid(pid, ENABLED_WARN))
2221 		return;
2222 
2223 	printf("Stopping ports...\n");
2224 
2225 	RTE_ETH_FOREACH_DEV(pi) {
2226 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2227 			continue;
2228 
2229 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2230 			printf("Please remove port %d from forwarding configuration.\n", pi);
2231 			continue;
2232 		}
2233 
2234 		if (port_is_bonding_slave(pi)) {
2235 			printf("Please remove port %d from bonded device.\n", pi);
2236 			continue;
2237 		}
2238 
2239 		port = &ports[pi];
2240 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2241 						RTE_PORT_HANDLING) == 0)
2242 			continue;
2243 
2244 		rte_eth_dev_stop(pi);
2245 
2246 		if (rte_atomic16_cmpset(&(port->port_status),
2247 			RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2248 			printf("Port %d can not be set into stopped\n", pi);
2249 		need_check_link_status = 1;
2250 	}
2251 	if (need_check_link_status && !no_link_check)
2252 		check_all_ports_link_status(RTE_PORT_ALL);
2253 
2254 	printf("Done\n");
2255 }
2256 
2257 static void
2258 remove_invalid_ports_in(portid_t *array, portid_t *total)
2259 {
2260 	portid_t i;
2261 	portid_t new_total = 0;
2262 
2263 	for (i = 0; i < *total; i++)
2264 		if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2265 			array[new_total] = array[i];
2266 			new_total++;
2267 		}
2268 	*total = new_total;
2269 }
2270 
2271 static void
2272 remove_invalid_ports(void)
2273 {
2274 	remove_invalid_ports_in(ports_ids, &nb_ports);
2275 	remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2276 	nb_cfg_ports = nb_fwd_ports;
2277 }
2278 
2279 void
2280 close_port(portid_t pid)
2281 {
2282 	portid_t pi;
2283 	struct rte_port *port;
2284 
2285 	if (port_id_is_invalid(pid, ENABLED_WARN))
2286 		return;
2287 
2288 	printf("Closing ports...\n");
2289 
2290 	RTE_ETH_FOREACH_DEV(pi) {
2291 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2292 			continue;
2293 
2294 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2295 			printf("Please remove port %d from forwarding configuration.\n", pi);
2296 			continue;
2297 		}
2298 
2299 		if (port_is_bonding_slave(pi)) {
2300 			printf("Please remove port %d from bonded device.\n", pi);
2301 			continue;
2302 		}
2303 
2304 		port = &ports[pi];
2305 		if (rte_atomic16_cmpset(&(port->port_status),
2306 			RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2307 			printf("Port %d is already closed\n", pi);
2308 			continue;
2309 		}
2310 
2311 		if (rte_atomic16_cmpset(&(port->port_status),
2312 			RTE_PORT_STOPPED, RTE_PORT_HANDLING) == 0) {
2313 			printf("Port %d is now not stopped\n", pi);
2314 			continue;
2315 		}
2316 
2317 		if (port->flow_list)
2318 			port_flow_flush(pi);
2319 		rte_eth_dev_close(pi);
2320 
2321 		remove_invalid_ports();
2322 
2323 		if (rte_atomic16_cmpset(&(port->port_status),
2324 			RTE_PORT_HANDLING, RTE_PORT_CLOSED) == 0)
2325 			printf("Port %d cannot be set to closed\n", pi);
2326 	}
2327 
2328 	printf("Done\n");
2329 }
2330 
2331 void
2332 reset_port(portid_t pid)
2333 {
2334 	int diag;
2335 	portid_t pi;
2336 	struct rte_port *port;
2337 
2338 	if (port_id_is_invalid(pid, ENABLED_WARN))
2339 		return;
2340 
2341 	printf("Resetting ports...\n");
2342 
2343 	RTE_ETH_FOREACH_DEV(pi) {
2344 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2345 			continue;
2346 
2347 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2348 			printf("Please remove port %d from forwarding "
2349 			       "configuration.\n", pi);
2350 			continue;
2351 		}
2352 
2353 		if (port_is_bonding_slave(pi)) {
2354 			printf("Please remove port %d from bonded device.\n",
2355 			       pi);
2356 			continue;
2357 		}
2358 
2359 		diag = rte_eth_dev_reset(pi);
2360 		if (diag == 0) {
2361 			port = &ports[pi];
2362 			port->need_reconfig = 1;
2363 			port->need_reconfig_queues = 1;
2364 		} else {
2365 			printf("Failed to reset port %d. diag=%d\n", pi, diag);
2366 		}
2367 	}
2368 
2369 	printf("Done\n");
2370 }
2371 
2372 void
2373 attach_port(char *identifier)
2374 {
2375 	portid_t pi;
2376 	struct rte_dev_iterator iterator;
2377 
2378 	printf("Attaching a new port...\n");
2379 
2380 	if (identifier == NULL) {
2381 		printf("Invalid parameters are specified\n");
2382 		return;
2383 	}
2384 
2385 	if (rte_dev_probe(identifier) != 0) {
2386 		TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2387 		return;
2388 	}
2389 
2390 	/* first attach mode: event */
2391 	if (setup_on_probe_event) {
2392 		/* new ports are detected on RTE_ETH_EVENT_NEW event */
2393 		for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2394 			if (ports[pi].port_status == RTE_PORT_HANDLING &&
2395 					ports[pi].need_setup != 0)
2396 				setup_attached_port(pi);
2397 		return;
2398 	}
2399 
2400 	/* second attach mode: iterator */
2401 	RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2402 		/* setup ports matching the devargs used for probing */
2403 		if (port_is_forwarding(pi))
2404 			continue; /* port was already attached before */
2405 		setup_attached_port(pi);
2406 	}
2407 }
2408 
2409 static void
2410 setup_attached_port(portid_t pi)
2411 {
2412 	unsigned int socket_id;
2413 
2414 	socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2415 	/* if socket_id is invalid, set to the first available socket. */
2416 	if (check_socket_id(socket_id) < 0)
2417 		socket_id = socket_ids[0];
2418 	reconfig(pi, socket_id);
2419 	rte_eth_promiscuous_enable(pi);
2420 
2421 	ports_ids[nb_ports++] = pi;
2422 	fwd_ports_ids[nb_fwd_ports++] = pi;
2423 	nb_cfg_ports = nb_fwd_ports;
2424 	ports[pi].need_setup = 0;
2425 	ports[pi].port_status = RTE_PORT_STOPPED;
2426 
2427 	printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2428 	printf("Done\n");
2429 }
2430 
2431 void
2432 detach_port_device(portid_t port_id)
2433 {
2434 	struct rte_device *dev;
2435 	portid_t sibling;
2436 
2437 	printf("Removing a device...\n");
2438 
2439 	dev = rte_eth_devices[port_id].device;
2440 	if (dev == NULL) {
2441 		printf("Device already removed\n");
2442 		return;
2443 	}
2444 
2445 	if (ports[port_id].port_status != RTE_PORT_CLOSED) {
2446 		if (ports[port_id].port_status != RTE_PORT_STOPPED) {
2447 			printf("Port not stopped\n");
2448 			return;
2449 		}
2450 		printf("Port was not closed\n");
2451 		if (ports[port_id].flow_list)
2452 			port_flow_flush(port_id);
2453 	}
2454 
2455 	if (rte_dev_remove(dev) != 0) {
2456 		TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
2457 		return;
2458 	}
2459 	RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
2460 		/* reset mapping between old ports and removed device */
2461 		rte_eth_devices[sibling].device = NULL;
2462 		if (ports[sibling].port_status != RTE_PORT_CLOSED) {
2463 			/* sibling ports are forced to be closed */
2464 			ports[sibling].port_status = RTE_PORT_CLOSED;
2465 			printf("Port %u is closed\n", sibling);
2466 		}
2467 	}
2468 
2469 	remove_invalid_ports();
2470 
2471 	printf("Device of port %u is detached\n", port_id);
2472 	printf("Now total ports is %d\n", nb_ports);
2473 	printf("Done\n");
2474 	return;
2475 }
2476 
2477 void
2478 pmd_test_exit(void)
2479 {
2480 	struct rte_device *device;
2481 	portid_t pt_id;
2482 	int ret;
2483 	int i;
2484 
2485 	if (test_done == 0)
2486 		stop_packet_forwarding();
2487 
2488 	for (i = 0 ; i < RTE_MAX_NUMA_NODES ; i++) {
2489 		if (mempools[i]) {
2490 			if (mp_alloc_type == MP_ALLOC_ANON)
2491 				rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
2492 						     NULL);
2493 		}
2494 	}
2495 	if (ports != NULL) {
2496 		no_link_check = 1;
2497 		RTE_ETH_FOREACH_DEV(pt_id) {
2498 			printf("\nStopping port %d...\n", pt_id);
2499 			fflush(stdout);
2500 			stop_port(pt_id);
2501 		}
2502 		RTE_ETH_FOREACH_DEV(pt_id) {
2503 			printf("\nShutting down port %d...\n", pt_id);
2504 			fflush(stdout);
2505 			close_port(pt_id);
2506 
2507 			/*
2508 			 * This is a workaround to fix a virtio-user issue that
2509 			 * requires to call clean-up routine to remove existing
2510 			 * socket.
2511 			 * This workaround valid only for testpmd, needs a fix
2512 			 * valid for all applications.
2513 			 * TODO: Implement proper resource cleanup
2514 			 */
2515 			device = rte_eth_devices[pt_id].device;
2516 			if (device && !strcmp(device->driver->name, "net_virtio_user"))
2517 				detach_port_device(pt_id);
2518 		}
2519 	}
2520 
2521 	if (hot_plug) {
2522 		ret = rte_dev_event_monitor_stop();
2523 		if (ret) {
2524 			RTE_LOG(ERR, EAL,
2525 				"fail to stop device event monitor.");
2526 			return;
2527 		}
2528 
2529 		ret = rte_dev_event_callback_unregister(NULL,
2530 			dev_event_callback, NULL);
2531 		if (ret < 0) {
2532 			RTE_LOG(ERR, EAL,
2533 				"fail to unregister device event callback.\n");
2534 			return;
2535 		}
2536 
2537 		ret = rte_dev_hotplug_handle_disable();
2538 		if (ret) {
2539 			RTE_LOG(ERR, EAL,
2540 				"fail to disable hotplug handling.\n");
2541 			return;
2542 		}
2543 	}
2544 	for (i = 0 ; i < RTE_MAX_NUMA_NODES ; i++) {
2545 		if (mempools[i])
2546 			rte_mempool_free(mempools[i]);
2547 	}
2548 
2549 	printf("\nBye...\n");
2550 }
2551 
2552 typedef void (*cmd_func_t)(void);
2553 struct pmd_test_command {
2554 	const char *cmd_name;
2555 	cmd_func_t cmd_func;
2556 };
2557 
2558 #define PMD_TEST_CMD_NB (sizeof(pmd_test_menu) / sizeof(pmd_test_menu[0]))
2559 
2560 /* Check the link status of all ports in up to 9s, and print them finally */
2561 static void
2562 check_all_ports_link_status(uint32_t port_mask)
2563 {
2564 #define CHECK_INTERVAL 100 /* 100ms */
2565 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2566 	portid_t portid;
2567 	uint8_t count, all_ports_up, print_flag = 0;
2568 	struct rte_eth_link link;
2569 
2570 	printf("Checking link statuses...\n");
2571 	fflush(stdout);
2572 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
2573 		all_ports_up = 1;
2574 		RTE_ETH_FOREACH_DEV(portid) {
2575 			if ((port_mask & (1 << portid)) == 0)
2576 				continue;
2577 			memset(&link, 0, sizeof(link));
2578 			rte_eth_link_get_nowait(portid, &link);
2579 			/* print link status if flag set */
2580 			if (print_flag == 1) {
2581 				if (link.link_status)
2582 					printf(
2583 					"Port%d Link Up. speed %u Mbps- %s\n",
2584 					portid, link.link_speed,
2585 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2586 					("full-duplex") : ("half-duplex\n"));
2587 				else
2588 					printf("Port %d Link Down\n", portid);
2589 				continue;
2590 			}
2591 			/* clear all_ports_up flag if any link down */
2592 			if (link.link_status == ETH_LINK_DOWN) {
2593 				all_ports_up = 0;
2594 				break;
2595 			}
2596 		}
2597 		/* after finally printing all link status, get out */
2598 		if (print_flag == 1)
2599 			break;
2600 
2601 		if (all_ports_up == 0) {
2602 			fflush(stdout);
2603 			rte_delay_ms(CHECK_INTERVAL);
2604 		}
2605 
2606 		/* set the print_flag if all ports up or timeout */
2607 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2608 			print_flag = 1;
2609 		}
2610 
2611 		if (lsc_interrupt)
2612 			break;
2613 	}
2614 }
2615 
2616 /*
2617  * This callback is for remove a port for a device. It has limitation because
2618  * it is not for multiple port removal for a device.
2619  * TODO: the device detach invoke will plan to be removed from user side to
2620  * eal. And convert all PMDs to free port resources on ether device closing.
2621  */
2622 static void
2623 rmv_port_callback(void *arg)
2624 {
2625 	int need_to_start = 0;
2626 	int org_no_link_check = no_link_check;
2627 	portid_t port_id = (intptr_t)arg;
2628 
2629 	RTE_ETH_VALID_PORTID_OR_RET(port_id);
2630 
2631 	if (!test_done && port_is_forwarding(port_id)) {
2632 		need_to_start = 1;
2633 		stop_packet_forwarding();
2634 	}
2635 	no_link_check = 1;
2636 	stop_port(port_id);
2637 	no_link_check = org_no_link_check;
2638 	close_port(port_id);
2639 	detach_port_device(port_id);
2640 	if (need_to_start)
2641 		start_packet_forwarding(0);
2642 }
2643 
2644 /* This function is used by the interrupt thread */
2645 static int
2646 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
2647 		  void *ret_param)
2648 {
2649 	RTE_SET_USED(param);
2650 	RTE_SET_USED(ret_param);
2651 
2652 	if (type >= RTE_ETH_EVENT_MAX) {
2653 		fprintf(stderr, "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
2654 			port_id, __func__, type);
2655 		fflush(stderr);
2656 	} else if (event_print_mask & (UINT32_C(1) << type)) {
2657 		printf("\nPort %" PRIu16 ": %s event\n", port_id,
2658 			eth_event_desc[type]);
2659 		fflush(stdout);
2660 	}
2661 
2662 	switch (type) {
2663 	case RTE_ETH_EVENT_NEW:
2664 		ports[port_id].need_setup = 1;
2665 		ports[port_id].port_status = RTE_PORT_HANDLING;
2666 		break;
2667 	case RTE_ETH_EVENT_INTR_RMV:
2668 		if (port_id_is_invalid(port_id, DISABLED_WARN))
2669 			break;
2670 		if (rte_eal_alarm_set(100000,
2671 				rmv_port_callback, (void *)(intptr_t)port_id))
2672 			fprintf(stderr, "Could not set up deferred device removal\n");
2673 		break;
2674 	default:
2675 		break;
2676 	}
2677 	return 0;
2678 }
2679 
2680 static int
2681 register_eth_event_callback(void)
2682 {
2683 	int ret;
2684 	enum rte_eth_event_type event;
2685 
2686 	for (event = RTE_ETH_EVENT_UNKNOWN;
2687 			event < RTE_ETH_EVENT_MAX; event++) {
2688 		ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
2689 				event,
2690 				eth_event_callback,
2691 				NULL);
2692 		if (ret != 0) {
2693 			TESTPMD_LOG(ERR, "Failed to register callback for "
2694 					"%s event\n", eth_event_desc[event]);
2695 			return -1;
2696 		}
2697 	}
2698 
2699 	return 0;
2700 }
2701 
2702 /* This function is used by the interrupt thread */
2703 static void
2704 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
2705 			     __rte_unused void *arg)
2706 {
2707 	uint16_t port_id;
2708 	int ret;
2709 
2710 	if (type >= RTE_DEV_EVENT_MAX) {
2711 		fprintf(stderr, "%s called upon invalid event %d\n",
2712 			__func__, type);
2713 		fflush(stderr);
2714 	}
2715 
2716 	switch (type) {
2717 	case RTE_DEV_EVENT_REMOVE:
2718 		RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
2719 			device_name);
2720 		ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
2721 		if (ret) {
2722 			RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
2723 				device_name);
2724 			return;
2725 		}
2726 		/*
2727 		 * Because the user's callback is invoked in eal interrupt
2728 		 * callback, the interrupt callback need to be finished before
2729 		 * it can be unregistered when detaching device. So finish
2730 		 * callback soon and use a deferred removal to detach device
2731 		 * is need. It is a workaround, once the device detaching be
2732 		 * moved into the eal in the future, the deferred removal could
2733 		 * be deleted.
2734 		 */
2735 		if (rte_eal_alarm_set(100000,
2736 				rmv_port_callback, (void *)(intptr_t)port_id))
2737 			RTE_LOG(ERR, EAL,
2738 				"Could not set up deferred device removal\n");
2739 		break;
2740 	case RTE_DEV_EVENT_ADD:
2741 		RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
2742 			device_name);
2743 		/* TODO: After finish kernel driver binding,
2744 		 * begin to attach port.
2745 		 */
2746 		break;
2747 	default:
2748 		break;
2749 	}
2750 }
2751 
2752 static int
2753 set_tx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2754 {
2755 	uint16_t i;
2756 	int diag;
2757 	uint8_t mapping_found = 0;
2758 
2759 	for (i = 0; i < nb_tx_queue_stats_mappings; i++) {
2760 		if ((tx_queue_stats_mappings[i].port_id == port_id) &&
2761 				(tx_queue_stats_mappings[i].queue_id < nb_txq )) {
2762 			diag = rte_eth_dev_set_tx_queue_stats_mapping(port_id,
2763 					tx_queue_stats_mappings[i].queue_id,
2764 					tx_queue_stats_mappings[i].stats_counter_id);
2765 			if (diag != 0)
2766 				return diag;
2767 			mapping_found = 1;
2768 		}
2769 	}
2770 	if (mapping_found)
2771 		port->tx_queue_stats_mapping_enabled = 1;
2772 	return 0;
2773 }
2774 
2775 static int
2776 set_rx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2777 {
2778 	uint16_t i;
2779 	int diag;
2780 	uint8_t mapping_found = 0;
2781 
2782 	for (i = 0; i < nb_rx_queue_stats_mappings; i++) {
2783 		if ((rx_queue_stats_mappings[i].port_id == port_id) &&
2784 				(rx_queue_stats_mappings[i].queue_id < nb_rxq )) {
2785 			diag = rte_eth_dev_set_rx_queue_stats_mapping(port_id,
2786 					rx_queue_stats_mappings[i].queue_id,
2787 					rx_queue_stats_mappings[i].stats_counter_id);
2788 			if (diag != 0)
2789 				return diag;
2790 			mapping_found = 1;
2791 		}
2792 	}
2793 	if (mapping_found)
2794 		port->rx_queue_stats_mapping_enabled = 1;
2795 	return 0;
2796 }
2797 
2798 static void
2799 map_port_queue_stats_mapping_registers(portid_t pi, struct rte_port *port)
2800 {
2801 	int diag = 0;
2802 
2803 	diag = set_tx_queue_stats_mapping_registers(pi, port);
2804 	if (diag != 0) {
2805 		if (diag == -ENOTSUP) {
2806 			port->tx_queue_stats_mapping_enabled = 0;
2807 			printf("TX queue stats mapping not supported port id=%d\n", pi);
2808 		}
2809 		else
2810 			rte_exit(EXIT_FAILURE,
2811 					"set_tx_queue_stats_mapping_registers "
2812 					"failed for port id=%d diag=%d\n",
2813 					pi, diag);
2814 	}
2815 
2816 	diag = set_rx_queue_stats_mapping_registers(pi, port);
2817 	if (diag != 0) {
2818 		if (diag == -ENOTSUP) {
2819 			port->rx_queue_stats_mapping_enabled = 0;
2820 			printf("RX queue stats mapping not supported port id=%d\n", pi);
2821 		}
2822 		else
2823 			rte_exit(EXIT_FAILURE,
2824 					"set_rx_queue_stats_mapping_registers "
2825 					"failed for port id=%d diag=%d\n",
2826 					pi, diag);
2827 	}
2828 }
2829 
2830 static void
2831 rxtx_port_config(struct rte_port *port)
2832 {
2833 	uint16_t qid;
2834 	uint64_t offloads;
2835 
2836 	for (qid = 0; qid < nb_rxq; qid++) {
2837 		offloads = port->rx_conf[qid].offloads;
2838 		port->rx_conf[qid] = port->dev_info.default_rxconf;
2839 		port->rx_conf[qid].offloads |= offloads;
2840 
2841 		/* Check if any Rx parameters have been passed */
2842 		if (rx_pthresh != RTE_PMD_PARAM_UNSET)
2843 			port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
2844 
2845 		if (rx_hthresh != RTE_PMD_PARAM_UNSET)
2846 			port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
2847 
2848 		if (rx_wthresh != RTE_PMD_PARAM_UNSET)
2849 			port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
2850 
2851 		if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
2852 			port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
2853 
2854 		if (rx_drop_en != RTE_PMD_PARAM_UNSET)
2855 			port->rx_conf[qid].rx_drop_en = rx_drop_en;
2856 
2857 		port->nb_rx_desc[qid] = nb_rxd;
2858 	}
2859 
2860 	for (qid = 0; qid < nb_txq; qid++) {
2861 		offloads = port->tx_conf[qid].offloads;
2862 		port->tx_conf[qid] = port->dev_info.default_txconf;
2863 		port->tx_conf[qid].offloads |= offloads;
2864 
2865 		/* Check if any Tx parameters have been passed */
2866 		if (tx_pthresh != RTE_PMD_PARAM_UNSET)
2867 			port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
2868 
2869 		if (tx_hthresh != RTE_PMD_PARAM_UNSET)
2870 			port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
2871 
2872 		if (tx_wthresh != RTE_PMD_PARAM_UNSET)
2873 			port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
2874 
2875 		if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
2876 			port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
2877 
2878 		if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
2879 			port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
2880 
2881 		port->nb_tx_desc[qid] = nb_txd;
2882 	}
2883 }
2884 
2885 void
2886 init_port_config(void)
2887 {
2888 	portid_t pid;
2889 	struct rte_port *port;
2890 
2891 	RTE_ETH_FOREACH_DEV(pid) {
2892 		port = &ports[pid];
2893 		port->dev_conf.fdir_conf = fdir_conf;
2894 		rte_eth_dev_info_get(pid, &port->dev_info);
2895 		if (nb_rxq > 1) {
2896 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2897 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
2898 				rss_hf & port->dev_info.flow_type_rss_offloads;
2899 		} else {
2900 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2901 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
2902 		}
2903 
2904 		if (port->dcb_flag == 0) {
2905 			if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
2906 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
2907 			else
2908 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
2909 		}
2910 
2911 		rxtx_port_config(port);
2912 
2913 		rte_eth_macaddr_get(pid, &port->eth_addr);
2914 
2915 		map_port_queue_stats_mapping_registers(pid, port);
2916 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
2917 		rte_pmd_ixgbe_bypass_init(pid);
2918 #endif
2919 
2920 		if (lsc_interrupt &&
2921 		    (rte_eth_devices[pid].data->dev_flags &
2922 		     RTE_ETH_DEV_INTR_LSC))
2923 			port->dev_conf.intr_conf.lsc = 1;
2924 		if (rmv_interrupt &&
2925 		    (rte_eth_devices[pid].data->dev_flags &
2926 		     RTE_ETH_DEV_INTR_RMV))
2927 			port->dev_conf.intr_conf.rmv = 1;
2928 	}
2929 }
2930 
2931 void set_port_slave_flag(portid_t slave_pid)
2932 {
2933 	struct rte_port *port;
2934 
2935 	port = &ports[slave_pid];
2936 	port->slave_flag = 1;
2937 }
2938 
2939 void clear_port_slave_flag(portid_t slave_pid)
2940 {
2941 	struct rte_port *port;
2942 
2943 	port = &ports[slave_pid];
2944 	port->slave_flag = 0;
2945 }
2946 
2947 uint8_t port_is_bonding_slave(portid_t slave_pid)
2948 {
2949 	struct rte_port *port;
2950 
2951 	port = &ports[slave_pid];
2952 	if ((rte_eth_devices[slave_pid].data->dev_flags &
2953 	    RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
2954 		return 1;
2955 	return 0;
2956 }
2957 
2958 const uint16_t vlan_tags[] = {
2959 		0,  1,  2,  3,  4,  5,  6,  7,
2960 		8,  9, 10, 11,  12, 13, 14, 15,
2961 		16, 17, 18, 19, 20, 21, 22, 23,
2962 		24, 25, 26, 27, 28, 29, 30, 31
2963 };
2964 
2965 static  int
2966 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
2967 		 enum dcb_mode_enable dcb_mode,
2968 		 enum rte_eth_nb_tcs num_tcs,
2969 		 uint8_t pfc_en)
2970 {
2971 	uint8_t i;
2972 	int32_t rc;
2973 	struct rte_eth_rss_conf rss_conf;
2974 
2975 	/*
2976 	 * Builds up the correct configuration for dcb+vt based on the vlan tags array
2977 	 * given above, and the number of traffic classes available for use.
2978 	 */
2979 	if (dcb_mode == DCB_VT_ENABLED) {
2980 		struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
2981 				&eth_conf->rx_adv_conf.vmdq_dcb_conf;
2982 		struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
2983 				&eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
2984 
2985 		/* VMDQ+DCB RX and TX configurations */
2986 		vmdq_rx_conf->enable_default_pool = 0;
2987 		vmdq_rx_conf->default_pool = 0;
2988 		vmdq_rx_conf->nb_queue_pools =
2989 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2990 		vmdq_tx_conf->nb_queue_pools =
2991 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2992 
2993 		vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
2994 		for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
2995 			vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
2996 			vmdq_rx_conf->pool_map[i].pools =
2997 				1 << (i % vmdq_rx_conf->nb_queue_pools);
2998 		}
2999 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3000 			vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3001 			vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3002 		}
3003 
3004 		/* set DCB mode of RX and TX of multiple queues */
3005 		eth_conf->rxmode.mq_mode = ETH_MQ_RX_VMDQ_DCB;
3006 		eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3007 	} else {
3008 		struct rte_eth_dcb_rx_conf *rx_conf =
3009 				&eth_conf->rx_adv_conf.dcb_rx_conf;
3010 		struct rte_eth_dcb_tx_conf *tx_conf =
3011 				&eth_conf->tx_adv_conf.dcb_tx_conf;
3012 
3013 		rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3014 		if (rc != 0)
3015 			return rc;
3016 
3017 		rx_conf->nb_tcs = num_tcs;
3018 		tx_conf->nb_tcs = num_tcs;
3019 
3020 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3021 			rx_conf->dcb_tc[i] = i % num_tcs;
3022 			tx_conf->dcb_tc[i] = i % num_tcs;
3023 		}
3024 
3025 		eth_conf->rxmode.mq_mode = ETH_MQ_RX_DCB_RSS;
3026 		eth_conf->rx_adv_conf.rss_conf = rss_conf;
3027 		eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3028 	}
3029 
3030 	if (pfc_en)
3031 		eth_conf->dcb_capability_en =
3032 				ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3033 	else
3034 		eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3035 
3036 	return 0;
3037 }
3038 
3039 int
3040 init_port_dcb_config(portid_t pid,
3041 		     enum dcb_mode_enable dcb_mode,
3042 		     enum rte_eth_nb_tcs num_tcs,
3043 		     uint8_t pfc_en)
3044 {
3045 	struct rte_eth_conf port_conf;
3046 	struct rte_port *rte_port;
3047 	int retval;
3048 	uint16_t i;
3049 
3050 	rte_port = &ports[pid];
3051 
3052 	memset(&port_conf, 0, sizeof(struct rte_eth_conf));
3053 	/* Enter DCB configuration status */
3054 	dcb_config = 1;
3055 
3056 	port_conf.rxmode = rte_port->dev_conf.rxmode;
3057 	port_conf.txmode = rte_port->dev_conf.txmode;
3058 
3059 	/*set configuration of DCB in vt mode and DCB in non-vt mode*/
3060 	retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3061 	if (retval < 0)
3062 		return retval;
3063 	port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3064 
3065 	/* re-configure the device . */
3066 	retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3067 	if (retval < 0)
3068 		return retval;
3069 	rte_eth_dev_info_get(pid, &rte_port->dev_info);
3070 
3071 	/* If dev_info.vmdq_pool_base is greater than 0,
3072 	 * the queue id of vmdq pools is started after pf queues.
3073 	 */
3074 	if (dcb_mode == DCB_VT_ENABLED &&
3075 	    rte_port->dev_info.vmdq_pool_base > 0) {
3076 		printf("VMDQ_DCB multi-queue mode is nonsensical"
3077 			" for port %d.", pid);
3078 		return -1;
3079 	}
3080 
3081 	/* Assume the ports in testpmd have the same dcb capability
3082 	 * and has the same number of rxq and txq in dcb mode
3083 	 */
3084 	if (dcb_mode == DCB_VT_ENABLED) {
3085 		if (rte_port->dev_info.max_vfs > 0) {
3086 			nb_rxq = rte_port->dev_info.nb_rx_queues;
3087 			nb_txq = rte_port->dev_info.nb_tx_queues;
3088 		} else {
3089 			nb_rxq = rte_port->dev_info.max_rx_queues;
3090 			nb_txq = rte_port->dev_info.max_tx_queues;
3091 		}
3092 	} else {
3093 		/*if vt is disabled, use all pf queues */
3094 		if (rte_port->dev_info.vmdq_pool_base == 0) {
3095 			nb_rxq = rte_port->dev_info.max_rx_queues;
3096 			nb_txq = rte_port->dev_info.max_tx_queues;
3097 		} else {
3098 			nb_rxq = (queueid_t)num_tcs;
3099 			nb_txq = (queueid_t)num_tcs;
3100 
3101 		}
3102 	}
3103 	rx_free_thresh = 64;
3104 
3105 	memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3106 
3107 	rxtx_port_config(rte_port);
3108 	/* VLAN filter */
3109 	rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3110 	for (i = 0; i < RTE_DIM(vlan_tags); i++)
3111 		rx_vft_set(pid, vlan_tags[i], 1);
3112 
3113 	rte_eth_macaddr_get(pid, &rte_port->eth_addr);
3114 	map_port_queue_stats_mapping_registers(pid, rte_port);
3115 
3116 	rte_port->dcb_flag = 1;
3117 
3118 	return 0;
3119 }
3120 
3121 static void
3122 init_port(void)
3123 {
3124 	/* Configuration of Ethernet ports. */
3125 	ports = rte_zmalloc("testpmd: ports",
3126 			    sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3127 			    RTE_CACHE_LINE_SIZE);
3128 	if (ports == NULL) {
3129 		rte_exit(EXIT_FAILURE,
3130 				"rte_zmalloc(%d struct rte_port) failed\n",
3131 				RTE_MAX_ETHPORTS);
3132 	}
3133 
3134 	/* Initialize ports NUMA structures */
3135 	memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3136 	memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3137 	memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3138 }
3139 
3140 static void
3141 force_quit(void)
3142 {
3143 	pmd_test_exit();
3144 	prompt_exit();
3145 }
3146 
3147 static void
3148 print_stats(void)
3149 {
3150 	uint8_t i;
3151 	const char clr[] = { 27, '[', '2', 'J', '\0' };
3152 	const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3153 
3154 	/* Clear screen and move to top left */
3155 	printf("%s%s", clr, top_left);
3156 
3157 	printf("\nPort statistics ====================================");
3158 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3159 		nic_stats_display(fwd_ports_ids[i]);
3160 
3161 	fflush(stdout);
3162 }
3163 
3164 static void
3165 signal_handler(int signum)
3166 {
3167 	if (signum == SIGINT || signum == SIGTERM) {
3168 		printf("\nSignal %d received, preparing to exit...\n",
3169 				signum);
3170 #ifdef RTE_LIBRTE_PDUMP
3171 		/* uninitialize packet capture framework */
3172 		rte_pdump_uninit();
3173 #endif
3174 #ifdef RTE_LIBRTE_LATENCY_STATS
3175 		rte_latencystats_uninit();
3176 #endif
3177 		force_quit();
3178 		/* Set flag to indicate the force termination. */
3179 		f_quit = 1;
3180 		/* exit with the expected status */
3181 		signal(signum, SIG_DFL);
3182 		kill(getpid(), signum);
3183 	}
3184 }
3185 
3186 int
3187 main(int argc, char** argv)
3188 {
3189 	int diag;
3190 	portid_t port_id;
3191 	uint16_t count;
3192 	int ret;
3193 
3194 	signal(SIGINT, signal_handler);
3195 	signal(SIGTERM, signal_handler);
3196 
3197 	diag = rte_eal_init(argc, argv);
3198 	if (diag < 0)
3199 		rte_panic("Cannot init EAL\n");
3200 
3201 	testpmd_logtype = rte_log_register("testpmd");
3202 	if (testpmd_logtype < 0)
3203 		rte_panic("Cannot register log type");
3204 	rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3205 
3206 	ret = register_eth_event_callback();
3207 	if (ret != 0)
3208 		rte_panic("Cannot register for ethdev events");
3209 
3210 #ifdef RTE_LIBRTE_PDUMP
3211 	/* initialize packet capture framework */
3212 	rte_pdump_init();
3213 #endif
3214 
3215 	count = 0;
3216 	RTE_ETH_FOREACH_DEV(port_id) {
3217 		ports_ids[count] = port_id;
3218 		count++;
3219 	}
3220 	nb_ports = (portid_t) count;
3221 	if (nb_ports == 0)
3222 		TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3223 
3224 	/* allocate port structures, and init them */
3225 	init_port();
3226 
3227 	set_def_fwd_config();
3228 	if (nb_lcores == 0)
3229 		rte_panic("Empty set of forwarding logical cores - check the "
3230 			  "core mask supplied in the command parameters\n");
3231 
3232 	/* Bitrate/latency stats disabled by default */
3233 #ifdef RTE_LIBRTE_BITRATE
3234 	bitrate_enabled = 0;
3235 #endif
3236 #ifdef RTE_LIBRTE_LATENCY_STATS
3237 	latencystats_enabled = 0;
3238 #endif
3239 
3240 	/* on FreeBSD, mlockall() is disabled by default */
3241 #ifdef RTE_EXEC_ENV_FREEBSD
3242 	do_mlockall = 0;
3243 #else
3244 	do_mlockall = 1;
3245 #endif
3246 
3247 	argc -= diag;
3248 	argv += diag;
3249 	if (argc > 1)
3250 		launch_args_parse(argc, argv);
3251 
3252 	if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3253 		TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3254 			strerror(errno));
3255 	}
3256 
3257 	if (tx_first && interactive)
3258 		rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3259 				"interactive mode.\n");
3260 
3261 	if (tx_first && lsc_interrupt) {
3262 		printf("Warning: lsc_interrupt needs to be off when "
3263 				" using tx_first. Disabling.\n");
3264 		lsc_interrupt = 0;
3265 	}
3266 
3267 	if (!nb_rxq && !nb_txq)
3268 		printf("Warning: Either rx or tx queues should be non-zero\n");
3269 
3270 	if (nb_rxq > 1 && nb_rxq > nb_txq)
3271 		printf("Warning: nb_rxq=%d enables RSS configuration, "
3272 		       "but nb_txq=%d will prevent to fully test it.\n",
3273 		       nb_rxq, nb_txq);
3274 
3275 	init_config();
3276 
3277 	if (hot_plug) {
3278 		ret = rte_dev_hotplug_handle_enable();
3279 		if (ret) {
3280 			RTE_LOG(ERR, EAL,
3281 				"fail to enable hotplug handling.");
3282 			return -1;
3283 		}
3284 
3285 		ret = rte_dev_event_monitor_start();
3286 		if (ret) {
3287 			RTE_LOG(ERR, EAL,
3288 				"fail to start device event monitoring.");
3289 			return -1;
3290 		}
3291 
3292 		ret = rte_dev_event_callback_register(NULL,
3293 			dev_event_callback, NULL);
3294 		if (ret) {
3295 			RTE_LOG(ERR, EAL,
3296 				"fail  to register device event callback\n");
3297 			return -1;
3298 		}
3299 	}
3300 
3301 	if (start_port(RTE_PORT_ALL) != 0)
3302 		rte_exit(EXIT_FAILURE, "Start ports failed\n");
3303 
3304 	/* set all ports to promiscuous mode by default */
3305 	RTE_ETH_FOREACH_DEV(port_id)
3306 		rte_eth_promiscuous_enable(port_id);
3307 
3308 	/* Init metrics library */
3309 	rte_metrics_init(rte_socket_id());
3310 
3311 #ifdef RTE_LIBRTE_LATENCY_STATS
3312 	if (latencystats_enabled != 0) {
3313 		int ret = rte_latencystats_init(1, NULL);
3314 		if (ret)
3315 			printf("Warning: latencystats init()"
3316 				" returned error %d\n",	ret);
3317 		printf("Latencystats running on lcore %d\n",
3318 			latencystats_lcore_id);
3319 	}
3320 #endif
3321 
3322 	/* Setup bitrate stats */
3323 #ifdef RTE_LIBRTE_BITRATE
3324 	if (bitrate_enabled != 0) {
3325 		bitrate_data = rte_stats_bitrate_create();
3326 		if (bitrate_data == NULL)
3327 			rte_exit(EXIT_FAILURE,
3328 				"Could not allocate bitrate data.\n");
3329 		rte_stats_bitrate_reg(bitrate_data);
3330 	}
3331 #endif
3332 
3333 #ifdef RTE_LIBRTE_CMDLINE
3334 	if (strlen(cmdline_filename) != 0)
3335 		cmdline_read_from_file(cmdline_filename);
3336 
3337 	if (interactive == 1) {
3338 		if (auto_start) {
3339 			printf("Start automatic packet forwarding\n");
3340 			start_packet_forwarding(0);
3341 		}
3342 		prompt();
3343 		pmd_test_exit();
3344 	} else
3345 #endif
3346 	{
3347 		char c;
3348 		int rc;
3349 
3350 		f_quit = 0;
3351 
3352 		printf("No commandline core given, start packet forwarding\n");
3353 		start_packet_forwarding(tx_first);
3354 		if (stats_period != 0) {
3355 			uint64_t prev_time = 0, cur_time, diff_time = 0;
3356 			uint64_t timer_period;
3357 
3358 			/* Convert to number of cycles */
3359 			timer_period = stats_period * rte_get_timer_hz();
3360 
3361 			while (f_quit == 0) {
3362 				cur_time = rte_get_timer_cycles();
3363 				diff_time += cur_time - prev_time;
3364 
3365 				if (diff_time >= timer_period) {
3366 					print_stats();
3367 					/* Reset the timer */
3368 					diff_time = 0;
3369 				}
3370 				/* Sleep to avoid unnecessary checks */
3371 				prev_time = cur_time;
3372 				sleep(1);
3373 			}
3374 		}
3375 
3376 		printf("Press enter to exit\n");
3377 		rc = read(0, &c, 1);
3378 		pmd_test_exit();
3379 		if (rc < 0)
3380 			return 1;
3381 	}
3382 
3383 	return 0;
3384 }
3385