xref: /dpdk/app/test-pmd/testpmd.c (revision 2a7bb4fdf61e9edfb7adbaecb50e728b82da9e23)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #include <stdarg.h>
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <signal.h>
9 #include <string.h>
10 #include <time.h>
11 #include <fcntl.h>
12 #include <sys/mman.h>
13 #include <sys/types.h>
14 #include <errno.h>
15 #include <stdbool.h>
16 
17 #include <sys/queue.h>
18 #include <sys/stat.h>
19 
20 #include <stdint.h>
21 #include <unistd.h>
22 #include <inttypes.h>
23 
24 #include <rte_common.h>
25 #include <rte_errno.h>
26 #include <rte_byteorder.h>
27 #include <rte_log.h>
28 #include <rte_debug.h>
29 #include <rte_cycles.h>
30 #include <rte_malloc_heap.h>
31 #include <rte_memory.h>
32 #include <rte_memcpy.h>
33 #include <rte_launch.h>
34 #include <rte_eal.h>
35 #include <rte_alarm.h>
36 #include <rte_per_lcore.h>
37 #include <rte_lcore.h>
38 #include <rte_atomic.h>
39 #include <rte_branch_prediction.h>
40 #include <rte_mempool.h>
41 #include <rte_malloc.h>
42 #include <rte_mbuf.h>
43 #include <rte_mbuf_pool_ops.h>
44 #include <rte_interrupts.h>
45 #include <rte_pci.h>
46 #include <rte_ether.h>
47 #include <rte_ethdev.h>
48 #include <rte_dev.h>
49 #include <rte_string_fns.h>
50 #ifdef RTE_LIBRTE_IXGBE_PMD
51 #include <rte_pmd_ixgbe.h>
52 #endif
53 #ifdef RTE_LIBRTE_PDUMP
54 #include <rte_pdump.h>
55 #endif
56 #include <rte_flow.h>
57 #include <rte_metrics.h>
58 #ifdef RTE_LIBRTE_BITRATE
59 #include <rte_bitrate.h>
60 #endif
61 #ifdef RTE_LIBRTE_LATENCY_STATS
62 #include <rte_latencystats.h>
63 #endif
64 
65 #include "testpmd.h"
66 
67 #ifndef MAP_HUGETLB
68 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
69 #define HUGE_FLAG (0x40000)
70 #else
71 #define HUGE_FLAG MAP_HUGETLB
72 #endif
73 
74 #ifndef MAP_HUGE_SHIFT
75 /* older kernels (or FreeBSD) will not have this define */
76 #define HUGE_SHIFT (26)
77 #else
78 #define HUGE_SHIFT MAP_HUGE_SHIFT
79 #endif
80 
81 #define EXTMEM_HEAP_NAME "extmem"
82 
83 uint16_t verbose_level = 0; /**< Silent by default. */
84 int testpmd_logtype; /**< Log type for testpmd logs */
85 
86 /* use master core for command line ? */
87 uint8_t interactive = 0;
88 uint8_t auto_start = 0;
89 uint8_t tx_first;
90 char cmdline_filename[PATH_MAX] = {0};
91 
92 /*
93  * NUMA support configuration.
94  * When set, the NUMA support attempts to dispatch the allocation of the
95  * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
96  * probed ports among the CPU sockets 0 and 1.
97  * Otherwise, all memory is allocated from CPU socket 0.
98  */
99 uint8_t numa_support = 1; /**< numa enabled by default */
100 
101 /*
102  * In UMA mode,all memory is allocated from socket 0 if --socket-num is
103  * not configured.
104  */
105 uint8_t socket_num = UMA_NO_CONFIG;
106 
107 /*
108  * Select mempool allocation type:
109  * - native: use regular DPDK memory
110  * - anon: use regular DPDK memory to create mempool, but populate using
111  *         anonymous memory (may not be IOVA-contiguous)
112  * - xmem: use externally allocated hugepage memory
113  */
114 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
115 
116 /*
117  * Store specified sockets on which memory pool to be used by ports
118  * is allocated.
119  */
120 uint8_t port_numa[RTE_MAX_ETHPORTS];
121 
122 /*
123  * Store specified sockets on which RX ring to be used by ports
124  * is allocated.
125  */
126 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
127 
128 /*
129  * Store specified sockets on which TX ring to be used by ports
130  * is allocated.
131  */
132 uint8_t txring_numa[RTE_MAX_ETHPORTS];
133 
134 /*
135  * Record the Ethernet address of peer target ports to which packets are
136  * forwarded.
137  * Must be instantiated with the ethernet addresses of peer traffic generator
138  * ports.
139  */
140 struct ether_addr peer_eth_addrs[RTE_MAX_ETHPORTS];
141 portid_t nb_peer_eth_addrs = 0;
142 
143 /*
144  * Probed Target Environment.
145  */
146 struct rte_port *ports;	       /**< For all probed ethernet ports. */
147 portid_t nb_ports;             /**< Number of probed ethernet ports. */
148 struct fwd_lcore **fwd_lcores; /**< For all probed logical cores. */
149 lcoreid_t nb_lcores;           /**< Number of probed logical cores. */
150 
151 portid_t ports_ids[RTE_MAX_ETHPORTS]; /**< Store all port ids. */
152 
153 /*
154  * Test Forwarding Configuration.
155  *    nb_fwd_lcores <= nb_cfg_lcores <= nb_lcores
156  *    nb_fwd_ports  <= nb_cfg_ports  <= nb_ports
157  */
158 lcoreid_t nb_cfg_lcores; /**< Number of configured logical cores. */
159 lcoreid_t nb_fwd_lcores; /**< Number of forwarding logical cores. */
160 portid_t  nb_cfg_ports;  /**< Number of configured ports. */
161 portid_t  nb_fwd_ports;  /**< Number of forwarding ports. */
162 
163 unsigned int fwd_lcores_cpuids[RTE_MAX_LCORE]; /**< CPU ids configuration. */
164 portid_t fwd_ports_ids[RTE_MAX_ETHPORTS];      /**< Port ids configuration. */
165 
166 struct fwd_stream **fwd_streams; /**< For each RX queue of each port. */
167 streamid_t nb_fwd_streams;       /**< Is equal to (nb_ports * nb_rxq). */
168 
169 /*
170  * Forwarding engines.
171  */
172 struct fwd_engine * fwd_engines[] = {
173 	&io_fwd_engine,
174 	&mac_fwd_engine,
175 	&mac_swap_engine,
176 	&flow_gen_engine,
177 	&rx_only_engine,
178 	&tx_only_engine,
179 	&csum_fwd_engine,
180 	&icmp_echo_engine,
181 	&noisy_vnf_engine,
182 #if defined RTE_LIBRTE_PMD_SOFTNIC
183 	&softnic_fwd_engine,
184 #endif
185 #ifdef RTE_LIBRTE_IEEE1588
186 	&ieee1588_fwd_engine,
187 #endif
188 	NULL,
189 };
190 
191 struct fwd_config cur_fwd_config;
192 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
193 uint32_t retry_enabled;
194 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
195 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
196 
197 uint16_t mbuf_data_size = DEFAULT_MBUF_DATA_SIZE; /**< Mbuf data space size. */
198 uint32_t param_total_num_mbufs = 0;  /**< number of mbufs in all pools - if
199                                       * specified on command-line. */
200 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
201 
202 /*
203  * In container, it cannot terminate the process which running with 'stats-period'
204  * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
205  */
206 uint8_t f_quit;
207 
208 /*
209  * Configuration of packet segments used by the "txonly" processing engine.
210  */
211 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
212 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
213 	TXONLY_DEF_PACKET_LEN,
214 };
215 uint8_t  tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
216 
217 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
218 /**< Split policy for packets to TX. */
219 
220 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
221 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
222 
223 /* current configuration is in DCB or not,0 means it is not in DCB mode */
224 uint8_t dcb_config = 0;
225 
226 /* Whether the dcb is in testing status */
227 uint8_t dcb_test = 0;
228 
229 /*
230  * Configurable number of RX/TX queues.
231  */
232 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
233 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
234 
235 /*
236  * Configurable number of RX/TX ring descriptors.
237  * Defaults are supplied by drivers via ethdev.
238  */
239 #define RTE_TEST_RX_DESC_DEFAULT 0
240 #define RTE_TEST_TX_DESC_DEFAULT 0
241 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
242 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
243 
244 #define RTE_PMD_PARAM_UNSET -1
245 /*
246  * Configurable values of RX and TX ring threshold registers.
247  */
248 
249 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
250 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
251 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
252 
253 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
254 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
255 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
256 
257 /*
258  * Configurable value of RX free threshold.
259  */
260 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
261 
262 /*
263  * Configurable value of RX drop enable.
264  */
265 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
266 
267 /*
268  * Configurable value of TX free threshold.
269  */
270 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
271 
272 /*
273  * Configurable value of TX RS bit threshold.
274  */
275 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
276 
277 /*
278  * Configurable value of buffered packets before sending.
279  */
280 uint16_t noisy_tx_sw_bufsz;
281 
282 /*
283  * Configurable value of packet buffer timeout.
284  */
285 uint16_t noisy_tx_sw_buf_flush_time;
286 
287 /*
288  * Configurable value for size of VNF internal memory area
289  * used for simulating noisy neighbour behaviour
290  */
291 uint64_t noisy_lkup_mem_sz;
292 
293 /*
294  * Configurable value of number of random writes done in
295  * VNF simulation memory area.
296  */
297 uint64_t noisy_lkup_num_writes;
298 
299 /*
300  * Configurable value of number of random reads done in
301  * VNF simulation memory area.
302  */
303 uint64_t noisy_lkup_num_reads;
304 
305 /*
306  * Configurable value of number of random reads/writes done in
307  * VNF simulation memory area.
308  */
309 uint64_t noisy_lkup_num_reads_writes;
310 
311 /*
312  * Receive Side Scaling (RSS) configuration.
313  */
314 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
315 
316 /*
317  * Port topology configuration
318  */
319 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
320 
321 /*
322  * Avoids to flush all the RX streams before starts forwarding.
323  */
324 uint8_t no_flush_rx = 0; /* flush by default */
325 
326 /*
327  * Flow API isolated mode.
328  */
329 uint8_t flow_isolate_all;
330 
331 /*
332  * Avoids to check link status when starting/stopping a port.
333  */
334 uint8_t no_link_check = 0; /* check by default */
335 
336 /*
337  * Enable link status change notification
338  */
339 uint8_t lsc_interrupt = 1; /* enabled by default */
340 
341 /*
342  * Enable device removal notification.
343  */
344 uint8_t rmv_interrupt = 1; /* enabled by default */
345 
346 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
347 
348 /* After attach, port setup is called on event or by iterator */
349 bool setup_on_probe_event = true;
350 
351 /* Pretty printing of ethdev events */
352 static const char * const eth_event_desc[] = {
353 	[RTE_ETH_EVENT_UNKNOWN] = "unknown",
354 	[RTE_ETH_EVENT_INTR_LSC] = "link state change",
355 	[RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
356 	[RTE_ETH_EVENT_INTR_RESET] = "reset",
357 	[RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
358 	[RTE_ETH_EVENT_IPSEC] = "IPsec",
359 	[RTE_ETH_EVENT_MACSEC] = "MACsec",
360 	[RTE_ETH_EVENT_INTR_RMV] = "device removal",
361 	[RTE_ETH_EVENT_NEW] = "device probed",
362 	[RTE_ETH_EVENT_DESTROY] = "device released",
363 	[RTE_ETH_EVENT_MAX] = NULL,
364 };
365 
366 /*
367  * Display or mask ether events
368  * Default to all events except VF_MBOX
369  */
370 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
371 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
372 			    (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
373 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
374 			    (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
375 			    (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
376 			    (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV);
377 /*
378  * Decide if all memory are locked for performance.
379  */
380 int do_mlockall = 0;
381 
382 /*
383  * NIC bypass mode configuration options.
384  */
385 
386 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
387 /* The NIC bypass watchdog timeout. */
388 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
389 #endif
390 
391 
392 #ifdef RTE_LIBRTE_LATENCY_STATS
393 
394 /*
395  * Set when latency stats is enabled in the commandline
396  */
397 uint8_t latencystats_enabled;
398 
399 /*
400  * Lcore ID to serive latency statistics.
401  */
402 lcoreid_t latencystats_lcore_id = -1;
403 
404 #endif
405 
406 /*
407  * Ethernet device configuration.
408  */
409 struct rte_eth_rxmode rx_mode = {
410 	.max_rx_pkt_len = ETHER_MAX_LEN, /**< Default maximum frame length. */
411 };
412 
413 struct rte_eth_txmode tx_mode = {
414 	.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
415 };
416 
417 struct rte_fdir_conf fdir_conf = {
418 	.mode = RTE_FDIR_MODE_NONE,
419 	.pballoc = RTE_FDIR_PBALLOC_64K,
420 	.status = RTE_FDIR_REPORT_STATUS,
421 	.mask = {
422 		.vlan_tci_mask = 0xFFEF,
423 		.ipv4_mask     = {
424 			.src_ip = 0xFFFFFFFF,
425 			.dst_ip = 0xFFFFFFFF,
426 		},
427 		.ipv6_mask     = {
428 			.src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
429 			.dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
430 		},
431 		.src_port_mask = 0xFFFF,
432 		.dst_port_mask = 0xFFFF,
433 		.mac_addr_byte_mask = 0xFF,
434 		.tunnel_type_mask = 1,
435 		.tunnel_id_mask = 0xFFFFFFFF,
436 	},
437 	.drop_queue = 127,
438 };
439 
440 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
441 
442 struct queue_stats_mappings tx_queue_stats_mappings_array[MAX_TX_QUEUE_STATS_MAPPINGS];
443 struct queue_stats_mappings rx_queue_stats_mappings_array[MAX_RX_QUEUE_STATS_MAPPINGS];
444 
445 struct queue_stats_mappings *tx_queue_stats_mappings = tx_queue_stats_mappings_array;
446 struct queue_stats_mappings *rx_queue_stats_mappings = rx_queue_stats_mappings_array;
447 
448 uint16_t nb_tx_queue_stats_mappings = 0;
449 uint16_t nb_rx_queue_stats_mappings = 0;
450 
451 /*
452  * Display zero values by default for xstats
453  */
454 uint8_t xstats_hide_zero;
455 
456 unsigned int num_sockets = 0;
457 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
458 
459 #ifdef RTE_LIBRTE_BITRATE
460 /* Bitrate statistics */
461 struct rte_stats_bitrates *bitrate_data;
462 lcoreid_t bitrate_lcore_id;
463 uint8_t bitrate_enabled;
464 #endif
465 
466 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
467 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
468 
469 struct vxlan_encap_conf vxlan_encap_conf = {
470 	.select_ipv4 = 1,
471 	.select_vlan = 0,
472 	.vni = "\x00\x00\x00",
473 	.udp_src = 0,
474 	.udp_dst = RTE_BE16(4789),
475 	.ipv4_src = IPv4(127, 0, 0, 1),
476 	.ipv4_dst = IPv4(255, 255, 255, 255),
477 	.ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
478 		"\x00\x00\x00\x00\x00\x00\x00\x01",
479 	.ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
480 		"\x00\x00\x00\x00\x00\x00\x11\x11",
481 	.vlan_tci = 0,
482 	.eth_src = "\x00\x00\x00\x00\x00\x00",
483 	.eth_dst = "\xff\xff\xff\xff\xff\xff",
484 };
485 
486 struct nvgre_encap_conf nvgre_encap_conf = {
487 	.select_ipv4 = 1,
488 	.select_vlan = 0,
489 	.tni = "\x00\x00\x00",
490 	.ipv4_src = IPv4(127, 0, 0, 1),
491 	.ipv4_dst = IPv4(255, 255, 255, 255),
492 	.ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
493 		"\x00\x00\x00\x00\x00\x00\x00\x01",
494 	.ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
495 		"\x00\x00\x00\x00\x00\x00\x11\x11",
496 	.vlan_tci = 0,
497 	.eth_src = "\x00\x00\x00\x00\x00\x00",
498 	.eth_dst = "\xff\xff\xff\xff\xff\xff",
499 };
500 
501 /* Forward function declarations */
502 static void setup_attached_port(portid_t pi);
503 static void map_port_queue_stats_mapping_registers(portid_t pi,
504 						   struct rte_port *port);
505 static void check_all_ports_link_status(uint32_t port_mask);
506 static int eth_event_callback(portid_t port_id,
507 			      enum rte_eth_event_type type,
508 			      void *param, void *ret_param);
509 static void dev_event_callback(const char *device_name,
510 				enum rte_dev_event_type type,
511 				void *param);
512 
513 /*
514  * Check if all the ports are started.
515  * If yes, return positive value. If not, return zero.
516  */
517 static int all_ports_started(void);
518 
519 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
520 uint16_t gso_max_segment_size = ETHER_MAX_LEN - ETHER_CRC_LEN;
521 
522 /*
523  * Helper function to check if socket is already discovered.
524  * If yes, return positive value. If not, return zero.
525  */
526 int
527 new_socket_id(unsigned int socket_id)
528 {
529 	unsigned int i;
530 
531 	for (i = 0; i < num_sockets; i++) {
532 		if (socket_ids[i] == socket_id)
533 			return 0;
534 	}
535 	return 1;
536 }
537 
538 /*
539  * Setup default configuration.
540  */
541 static void
542 set_default_fwd_lcores_config(void)
543 {
544 	unsigned int i;
545 	unsigned int nb_lc;
546 	unsigned int sock_num;
547 
548 	nb_lc = 0;
549 	for (i = 0; i < RTE_MAX_LCORE; i++) {
550 		if (!rte_lcore_is_enabled(i))
551 			continue;
552 		sock_num = rte_lcore_to_socket_id(i);
553 		if (new_socket_id(sock_num)) {
554 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
555 				rte_exit(EXIT_FAILURE,
556 					 "Total sockets greater than %u\n",
557 					 RTE_MAX_NUMA_NODES);
558 			}
559 			socket_ids[num_sockets++] = sock_num;
560 		}
561 		if (i == rte_get_master_lcore())
562 			continue;
563 		fwd_lcores_cpuids[nb_lc++] = i;
564 	}
565 	nb_lcores = (lcoreid_t) nb_lc;
566 	nb_cfg_lcores = nb_lcores;
567 	nb_fwd_lcores = 1;
568 }
569 
570 static void
571 set_def_peer_eth_addrs(void)
572 {
573 	portid_t i;
574 
575 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
576 		peer_eth_addrs[i].addr_bytes[0] = ETHER_LOCAL_ADMIN_ADDR;
577 		peer_eth_addrs[i].addr_bytes[5] = i;
578 	}
579 }
580 
581 static void
582 set_default_fwd_ports_config(void)
583 {
584 	portid_t pt_id;
585 	int i = 0;
586 
587 	RTE_ETH_FOREACH_DEV(pt_id) {
588 		fwd_ports_ids[i++] = pt_id;
589 
590 		/* Update sockets info according to the attached device */
591 		int socket_id = rte_eth_dev_socket_id(pt_id);
592 		if (socket_id >= 0 && new_socket_id(socket_id)) {
593 			if (num_sockets >= RTE_MAX_NUMA_NODES) {
594 				rte_exit(EXIT_FAILURE,
595 					 "Total sockets greater than %u\n",
596 					 RTE_MAX_NUMA_NODES);
597 			}
598 			socket_ids[num_sockets++] = socket_id;
599 		}
600 	}
601 
602 	nb_cfg_ports = nb_ports;
603 	nb_fwd_ports = nb_ports;
604 }
605 
606 void
607 set_def_fwd_config(void)
608 {
609 	set_default_fwd_lcores_config();
610 	set_def_peer_eth_addrs();
611 	set_default_fwd_ports_config();
612 }
613 
614 /* extremely pessimistic estimation of memory required to create a mempool */
615 static int
616 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
617 {
618 	unsigned int n_pages, mbuf_per_pg, leftover;
619 	uint64_t total_mem, mbuf_mem, obj_sz;
620 
621 	/* there is no good way to predict how much space the mempool will
622 	 * occupy because it will allocate chunks on the fly, and some of those
623 	 * will come from default DPDK memory while some will come from our
624 	 * external memory, so just assume 128MB will be enough for everyone.
625 	 */
626 	uint64_t hdr_mem = 128 << 20;
627 
628 	/* account for possible non-contiguousness */
629 	obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
630 	if (obj_sz > pgsz) {
631 		TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
632 		return -1;
633 	}
634 
635 	mbuf_per_pg = pgsz / obj_sz;
636 	leftover = (nb_mbufs % mbuf_per_pg) > 0;
637 	n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
638 
639 	mbuf_mem = n_pages * pgsz;
640 
641 	total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
642 
643 	if (total_mem > SIZE_MAX) {
644 		TESTPMD_LOG(ERR, "Memory size too big\n");
645 		return -1;
646 	}
647 	*out = (size_t)total_mem;
648 
649 	return 0;
650 }
651 
652 static int
653 pagesz_flags(uint64_t page_sz)
654 {
655 	/* as per mmap() manpage, all page sizes are log2 of page size
656 	 * shifted by MAP_HUGE_SHIFT
657 	 */
658 	int log2 = rte_log2_u64(page_sz);
659 
660 	return (log2 << HUGE_SHIFT);
661 }
662 
663 static void *
664 alloc_mem(size_t memsz, size_t pgsz, bool huge)
665 {
666 	void *addr;
667 	int flags;
668 
669 	/* allocate anonymous hugepages */
670 	flags = MAP_ANONYMOUS | MAP_PRIVATE;
671 	if (huge)
672 		flags |= HUGE_FLAG | pagesz_flags(pgsz);
673 
674 	addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
675 	if (addr == MAP_FAILED)
676 		return NULL;
677 
678 	return addr;
679 }
680 
681 struct extmem_param {
682 	void *addr;
683 	size_t len;
684 	size_t pgsz;
685 	rte_iova_t *iova_table;
686 	unsigned int iova_table_len;
687 };
688 
689 static int
690 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
691 		bool huge)
692 {
693 	uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
694 			RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
695 	unsigned int cur_page, n_pages, pgsz_idx;
696 	size_t mem_sz, cur_pgsz;
697 	rte_iova_t *iovas = NULL;
698 	void *addr;
699 	int ret;
700 
701 	for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
702 		/* skip anything that is too big */
703 		if (pgsizes[pgsz_idx] > SIZE_MAX)
704 			continue;
705 
706 		cur_pgsz = pgsizes[pgsz_idx];
707 
708 		/* if we were told not to allocate hugepages, override */
709 		if (!huge)
710 			cur_pgsz = sysconf(_SC_PAGESIZE);
711 
712 		ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
713 		if (ret < 0) {
714 			TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
715 			return -1;
716 		}
717 
718 		/* allocate our memory */
719 		addr = alloc_mem(mem_sz, cur_pgsz, huge);
720 
721 		/* if we couldn't allocate memory with a specified page size,
722 		 * that doesn't mean we can't do it with other page sizes, so
723 		 * try another one.
724 		 */
725 		if (addr == NULL)
726 			continue;
727 
728 		/* store IOVA addresses for every page in this memory area */
729 		n_pages = mem_sz / cur_pgsz;
730 
731 		iovas = malloc(sizeof(*iovas) * n_pages);
732 
733 		if (iovas == NULL) {
734 			TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
735 			goto fail;
736 		}
737 		/* lock memory if it's not huge pages */
738 		if (!huge)
739 			mlock(addr, mem_sz);
740 
741 		/* populate IOVA addresses */
742 		for (cur_page = 0; cur_page < n_pages; cur_page++) {
743 			rte_iova_t iova;
744 			size_t offset;
745 			void *cur;
746 
747 			offset = cur_pgsz * cur_page;
748 			cur = RTE_PTR_ADD(addr, offset);
749 
750 			/* touch the page before getting its IOVA */
751 			*(volatile char *)cur = 0;
752 
753 			iova = rte_mem_virt2iova(cur);
754 
755 			iovas[cur_page] = iova;
756 		}
757 
758 		break;
759 	}
760 	/* if we couldn't allocate anything */
761 	if (iovas == NULL)
762 		return -1;
763 
764 	param->addr = addr;
765 	param->len = mem_sz;
766 	param->pgsz = cur_pgsz;
767 	param->iova_table = iovas;
768 	param->iova_table_len = n_pages;
769 
770 	return 0;
771 fail:
772 	if (iovas)
773 		free(iovas);
774 	if (addr)
775 		munmap(addr, mem_sz);
776 
777 	return -1;
778 }
779 
780 static int
781 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
782 {
783 	struct extmem_param param;
784 	int socket_id, ret;
785 
786 	memset(&param, 0, sizeof(param));
787 
788 	/* check if our heap exists */
789 	socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
790 	if (socket_id < 0) {
791 		/* create our heap */
792 		ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
793 		if (ret < 0) {
794 			TESTPMD_LOG(ERR, "Cannot create heap\n");
795 			return -1;
796 		}
797 	}
798 
799 	ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
800 	if (ret < 0) {
801 		TESTPMD_LOG(ERR, "Cannot create memory area\n");
802 		return -1;
803 	}
804 
805 	/* we now have a valid memory area, so add it to heap */
806 	ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
807 			param.addr, param.len, param.iova_table,
808 			param.iova_table_len, param.pgsz);
809 
810 	/* when using VFIO, memory is automatically mapped for DMA by EAL */
811 
812 	/* not needed any more */
813 	free(param.iova_table);
814 
815 	if (ret < 0) {
816 		TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
817 		munmap(param.addr, param.len);
818 		return -1;
819 	}
820 
821 	/* success */
822 
823 	TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
824 			param.len >> 20);
825 
826 	return 0;
827 }
828 
829 /*
830  * Configuration initialisation done once at init time.
831  */
832 static void
833 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
834 		 unsigned int socket_id)
835 {
836 	char pool_name[RTE_MEMPOOL_NAMESIZE];
837 	struct rte_mempool *rte_mp = NULL;
838 	uint32_t mb_size;
839 
840 	mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
841 	mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name));
842 
843 	TESTPMD_LOG(INFO,
844 		"create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
845 		pool_name, nb_mbuf, mbuf_seg_size, socket_id);
846 
847 	switch (mp_alloc_type) {
848 	case MP_ALLOC_NATIVE:
849 		{
850 			/* wrapper to rte_mempool_create() */
851 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
852 					rte_mbuf_best_mempool_ops());
853 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
854 				mb_mempool_cache, 0, mbuf_seg_size, socket_id);
855 			break;
856 		}
857 	case MP_ALLOC_ANON:
858 		{
859 			rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
860 				mb_size, (unsigned int) mb_mempool_cache,
861 				sizeof(struct rte_pktmbuf_pool_private),
862 				socket_id, 0);
863 			if (rte_mp == NULL)
864 				goto err;
865 
866 			if (rte_mempool_populate_anon(rte_mp) == 0) {
867 				rte_mempool_free(rte_mp);
868 				rte_mp = NULL;
869 				goto err;
870 			}
871 			rte_pktmbuf_pool_init(rte_mp, NULL);
872 			rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
873 			break;
874 		}
875 	case MP_ALLOC_XMEM:
876 	case MP_ALLOC_XMEM_HUGE:
877 		{
878 			int heap_socket;
879 			bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
880 
881 			if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
882 				rte_exit(EXIT_FAILURE, "Could not create external memory\n");
883 
884 			heap_socket =
885 				rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
886 			if (heap_socket < 0)
887 				rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
888 
889 			TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
890 					rte_mbuf_best_mempool_ops());
891 			rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
892 					mb_mempool_cache, 0, mbuf_seg_size,
893 					heap_socket);
894 			break;
895 		}
896 	default:
897 		{
898 			rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
899 		}
900 	}
901 
902 err:
903 	if (rte_mp == NULL) {
904 		rte_exit(EXIT_FAILURE,
905 			"Creation of mbuf pool for socket %u failed: %s\n",
906 			socket_id, rte_strerror(rte_errno));
907 	} else if (verbose_level > 0) {
908 		rte_mempool_dump(stdout, rte_mp);
909 	}
910 }
911 
912 /*
913  * Check given socket id is valid or not with NUMA mode,
914  * if valid, return 0, else return -1
915  */
916 static int
917 check_socket_id(const unsigned int socket_id)
918 {
919 	static int warning_once = 0;
920 
921 	if (new_socket_id(socket_id)) {
922 		if (!warning_once && numa_support)
923 			printf("Warning: NUMA should be configured manually by"
924 			       " using --port-numa-config and"
925 			       " --ring-numa-config parameters along with"
926 			       " --numa.\n");
927 		warning_once = 1;
928 		return -1;
929 	}
930 	return 0;
931 }
932 
933 /*
934  * Get the allowed maximum number of RX queues.
935  * *pid return the port id which has minimal value of
936  * max_rx_queues in all ports.
937  */
938 queueid_t
939 get_allowed_max_nb_rxq(portid_t *pid)
940 {
941 	queueid_t allowed_max_rxq = MAX_QUEUE_ID;
942 	portid_t pi;
943 	struct rte_eth_dev_info dev_info;
944 
945 	RTE_ETH_FOREACH_DEV(pi) {
946 		rte_eth_dev_info_get(pi, &dev_info);
947 		if (dev_info.max_rx_queues < allowed_max_rxq) {
948 			allowed_max_rxq = dev_info.max_rx_queues;
949 			*pid = pi;
950 		}
951 	}
952 	return allowed_max_rxq;
953 }
954 
955 /*
956  * Check input rxq is valid or not.
957  * If input rxq is not greater than any of maximum number
958  * of RX queues of all ports, it is valid.
959  * if valid, return 0, else return -1
960  */
961 int
962 check_nb_rxq(queueid_t rxq)
963 {
964 	queueid_t allowed_max_rxq;
965 	portid_t pid = 0;
966 
967 	allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
968 	if (rxq > allowed_max_rxq) {
969 		printf("Fail: input rxq (%u) can't be greater "
970 		       "than max_rx_queues (%u) of port %u\n",
971 		       rxq,
972 		       allowed_max_rxq,
973 		       pid);
974 		return -1;
975 	}
976 	return 0;
977 }
978 
979 /*
980  * Get the allowed maximum number of TX queues.
981  * *pid return the port id which has minimal value of
982  * max_tx_queues in all ports.
983  */
984 queueid_t
985 get_allowed_max_nb_txq(portid_t *pid)
986 {
987 	queueid_t allowed_max_txq = MAX_QUEUE_ID;
988 	portid_t pi;
989 	struct rte_eth_dev_info dev_info;
990 
991 	RTE_ETH_FOREACH_DEV(pi) {
992 		rte_eth_dev_info_get(pi, &dev_info);
993 		if (dev_info.max_tx_queues < allowed_max_txq) {
994 			allowed_max_txq = dev_info.max_tx_queues;
995 			*pid = pi;
996 		}
997 	}
998 	return allowed_max_txq;
999 }
1000 
1001 /*
1002  * Check input txq is valid or not.
1003  * If input txq is not greater than any of maximum number
1004  * of TX queues of all ports, it is valid.
1005  * if valid, return 0, else return -1
1006  */
1007 int
1008 check_nb_txq(queueid_t txq)
1009 {
1010 	queueid_t allowed_max_txq;
1011 	portid_t pid = 0;
1012 
1013 	allowed_max_txq = get_allowed_max_nb_txq(&pid);
1014 	if (txq > allowed_max_txq) {
1015 		printf("Fail: input txq (%u) can't be greater "
1016 		       "than max_tx_queues (%u) of port %u\n",
1017 		       txq,
1018 		       allowed_max_txq,
1019 		       pid);
1020 		return -1;
1021 	}
1022 	return 0;
1023 }
1024 
1025 static void
1026 init_config(void)
1027 {
1028 	portid_t pid;
1029 	struct rte_port *port;
1030 	struct rte_mempool *mbp;
1031 	unsigned int nb_mbuf_per_pool;
1032 	lcoreid_t  lc_id;
1033 	uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
1034 	struct rte_gro_param gro_param;
1035 	uint32_t gso_types;
1036 	int k;
1037 
1038 	memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
1039 
1040 	/* Configuration of logical cores. */
1041 	fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1042 				sizeof(struct fwd_lcore *) * nb_lcores,
1043 				RTE_CACHE_LINE_SIZE);
1044 	if (fwd_lcores == NULL) {
1045 		rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1046 							"failed\n", nb_lcores);
1047 	}
1048 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1049 		fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1050 					       sizeof(struct fwd_lcore),
1051 					       RTE_CACHE_LINE_SIZE);
1052 		if (fwd_lcores[lc_id] == NULL) {
1053 			rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1054 								"failed\n");
1055 		}
1056 		fwd_lcores[lc_id]->cpuid_idx = lc_id;
1057 	}
1058 
1059 	RTE_ETH_FOREACH_DEV(pid) {
1060 		port = &ports[pid];
1061 		/* Apply default TxRx configuration for all ports */
1062 		port->dev_conf.txmode = tx_mode;
1063 		port->dev_conf.rxmode = rx_mode;
1064 		rte_eth_dev_info_get(pid, &port->dev_info);
1065 
1066 		if (!(port->dev_info.tx_offload_capa &
1067 		      DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1068 			port->dev_conf.txmode.offloads &=
1069 				~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1070 		if (!(port->dev_info.tx_offload_capa &
1071 			DEV_TX_OFFLOAD_MATCH_METADATA))
1072 			port->dev_conf.txmode.offloads &=
1073 				~DEV_TX_OFFLOAD_MATCH_METADATA;
1074 		if (numa_support) {
1075 			if (port_numa[pid] != NUMA_NO_CONFIG)
1076 				port_per_socket[port_numa[pid]]++;
1077 			else {
1078 				uint32_t socket_id = rte_eth_dev_socket_id(pid);
1079 
1080 				/*
1081 				 * if socket_id is invalid,
1082 				 * set to the first available socket.
1083 				 */
1084 				if (check_socket_id(socket_id) < 0)
1085 					socket_id = socket_ids[0];
1086 				port_per_socket[socket_id]++;
1087 			}
1088 		}
1089 
1090 		/* Apply Rx offloads configuration */
1091 		for (k = 0; k < port->dev_info.max_rx_queues; k++)
1092 			port->rx_conf[k].offloads =
1093 				port->dev_conf.rxmode.offloads;
1094 		/* Apply Tx offloads configuration */
1095 		for (k = 0; k < port->dev_info.max_tx_queues; k++)
1096 			port->tx_conf[k].offloads =
1097 				port->dev_conf.txmode.offloads;
1098 
1099 		/* set flag to initialize port/queue */
1100 		port->need_reconfig = 1;
1101 		port->need_reconfig_queues = 1;
1102 		port->tx_metadata = 0;
1103 	}
1104 
1105 	/*
1106 	 * Create pools of mbuf.
1107 	 * If NUMA support is disabled, create a single pool of mbuf in
1108 	 * socket 0 memory by default.
1109 	 * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1110 	 *
1111 	 * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1112 	 * nb_txd can be configured at run time.
1113 	 */
1114 	if (param_total_num_mbufs)
1115 		nb_mbuf_per_pool = param_total_num_mbufs;
1116 	else {
1117 		nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1118 			(nb_lcores * mb_mempool_cache) +
1119 			RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1120 		nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1121 	}
1122 
1123 	if (numa_support) {
1124 		uint8_t i;
1125 
1126 		for (i = 0; i < num_sockets; i++)
1127 			mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool,
1128 					 socket_ids[i]);
1129 	} else {
1130 		if (socket_num == UMA_NO_CONFIG)
1131 			mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool, 0);
1132 		else
1133 			mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool,
1134 						 socket_num);
1135 	}
1136 
1137 	init_port_config();
1138 
1139 	gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1140 		DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1141 	/*
1142 	 * Records which Mbuf pool to use by each logical core, if needed.
1143 	 */
1144 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1145 		mbp = mbuf_pool_find(
1146 			rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]));
1147 
1148 		if (mbp == NULL)
1149 			mbp = mbuf_pool_find(0);
1150 		fwd_lcores[lc_id]->mbp = mbp;
1151 		/* initialize GSO context */
1152 		fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1153 		fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1154 		fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1155 		fwd_lcores[lc_id]->gso_ctx.gso_size = ETHER_MAX_LEN -
1156 			ETHER_CRC_LEN;
1157 		fwd_lcores[lc_id]->gso_ctx.flag = 0;
1158 	}
1159 
1160 	/* Configuration of packet forwarding streams. */
1161 	if (init_fwd_streams() < 0)
1162 		rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
1163 
1164 	fwd_config_setup();
1165 
1166 	/* create a gro context for each lcore */
1167 	gro_param.gro_types = RTE_GRO_TCP_IPV4;
1168 	gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1169 	gro_param.max_item_per_flow = MAX_PKT_BURST;
1170 	for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1171 		gro_param.socket_id = rte_lcore_to_socket_id(
1172 				fwd_lcores_cpuids[lc_id]);
1173 		fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1174 		if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1175 			rte_exit(EXIT_FAILURE,
1176 					"rte_gro_ctx_create() failed\n");
1177 		}
1178 	}
1179 
1180 #if defined RTE_LIBRTE_PMD_SOFTNIC
1181 	if (strcmp(cur_fwd_eng->fwd_mode_name, "softnic") == 0) {
1182 		RTE_ETH_FOREACH_DEV(pid) {
1183 			port = &ports[pid];
1184 			const char *driver = port->dev_info.driver_name;
1185 
1186 			if (strcmp(driver, "net_softnic") == 0)
1187 				port->softport.fwd_lcore_arg = fwd_lcores;
1188 		}
1189 	}
1190 #endif
1191 
1192 }
1193 
1194 
1195 void
1196 reconfig(portid_t new_port_id, unsigned socket_id)
1197 {
1198 	struct rte_port *port;
1199 
1200 	/* Reconfiguration of Ethernet ports. */
1201 	port = &ports[new_port_id];
1202 	rte_eth_dev_info_get(new_port_id, &port->dev_info);
1203 
1204 	/* set flag to initialize port/queue */
1205 	port->need_reconfig = 1;
1206 	port->need_reconfig_queues = 1;
1207 	port->socket_id = socket_id;
1208 
1209 	init_port_config();
1210 }
1211 
1212 
1213 int
1214 init_fwd_streams(void)
1215 {
1216 	portid_t pid;
1217 	struct rte_port *port;
1218 	streamid_t sm_id, nb_fwd_streams_new;
1219 	queueid_t q;
1220 
1221 	/* set socket id according to numa or not */
1222 	RTE_ETH_FOREACH_DEV(pid) {
1223 		port = &ports[pid];
1224 		if (nb_rxq > port->dev_info.max_rx_queues) {
1225 			printf("Fail: nb_rxq(%d) is greater than "
1226 				"max_rx_queues(%d)\n", nb_rxq,
1227 				port->dev_info.max_rx_queues);
1228 			return -1;
1229 		}
1230 		if (nb_txq > port->dev_info.max_tx_queues) {
1231 			printf("Fail: nb_txq(%d) is greater than "
1232 				"max_tx_queues(%d)\n", nb_txq,
1233 				port->dev_info.max_tx_queues);
1234 			return -1;
1235 		}
1236 		if (numa_support) {
1237 			if (port_numa[pid] != NUMA_NO_CONFIG)
1238 				port->socket_id = port_numa[pid];
1239 			else {
1240 				port->socket_id = rte_eth_dev_socket_id(pid);
1241 
1242 				/*
1243 				 * if socket_id is invalid,
1244 				 * set to the first available socket.
1245 				 */
1246 				if (check_socket_id(port->socket_id) < 0)
1247 					port->socket_id = socket_ids[0];
1248 			}
1249 		}
1250 		else {
1251 			if (socket_num == UMA_NO_CONFIG)
1252 				port->socket_id = 0;
1253 			else
1254 				port->socket_id = socket_num;
1255 		}
1256 	}
1257 
1258 	q = RTE_MAX(nb_rxq, nb_txq);
1259 	if (q == 0) {
1260 		printf("Fail: Cannot allocate fwd streams as number of queues is 0\n");
1261 		return -1;
1262 	}
1263 	nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1264 	if (nb_fwd_streams_new == nb_fwd_streams)
1265 		return 0;
1266 	/* clear the old */
1267 	if (fwd_streams != NULL) {
1268 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1269 			if (fwd_streams[sm_id] == NULL)
1270 				continue;
1271 			rte_free(fwd_streams[sm_id]);
1272 			fwd_streams[sm_id] = NULL;
1273 		}
1274 		rte_free(fwd_streams);
1275 		fwd_streams = NULL;
1276 	}
1277 
1278 	/* init new */
1279 	nb_fwd_streams = nb_fwd_streams_new;
1280 	if (nb_fwd_streams) {
1281 		fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1282 			sizeof(struct fwd_stream *) * nb_fwd_streams,
1283 			RTE_CACHE_LINE_SIZE);
1284 		if (fwd_streams == NULL)
1285 			rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1286 				 " (struct fwd_stream *)) failed\n",
1287 				 nb_fwd_streams);
1288 
1289 		for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1290 			fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1291 				" struct fwd_stream", sizeof(struct fwd_stream),
1292 				RTE_CACHE_LINE_SIZE);
1293 			if (fwd_streams[sm_id] == NULL)
1294 				rte_exit(EXIT_FAILURE, "rte_zmalloc"
1295 					 "(struct fwd_stream) failed\n");
1296 		}
1297 	}
1298 
1299 	return 0;
1300 }
1301 
1302 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1303 static void
1304 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1305 {
1306 	unsigned int total_burst;
1307 	unsigned int nb_burst;
1308 	unsigned int burst_stats[3];
1309 	uint16_t pktnb_stats[3];
1310 	uint16_t nb_pkt;
1311 	int burst_percent[3];
1312 
1313 	/*
1314 	 * First compute the total number of packet bursts and the
1315 	 * two highest numbers of bursts of the same number of packets.
1316 	 */
1317 	total_burst = 0;
1318 	burst_stats[0] = burst_stats[1] = burst_stats[2] = 0;
1319 	pktnb_stats[0] = pktnb_stats[1] = pktnb_stats[2] = 0;
1320 	for (nb_pkt = 0; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1321 		nb_burst = pbs->pkt_burst_spread[nb_pkt];
1322 		if (nb_burst == 0)
1323 			continue;
1324 		total_burst += nb_burst;
1325 		if (nb_burst > burst_stats[0]) {
1326 			burst_stats[1] = burst_stats[0];
1327 			pktnb_stats[1] = pktnb_stats[0];
1328 			burst_stats[0] = nb_burst;
1329 			pktnb_stats[0] = nb_pkt;
1330 		} else if (nb_burst > burst_stats[1]) {
1331 			burst_stats[1] = nb_burst;
1332 			pktnb_stats[1] = nb_pkt;
1333 		}
1334 	}
1335 	if (total_burst == 0)
1336 		return;
1337 	burst_percent[0] = (burst_stats[0] * 100) / total_burst;
1338 	printf("  %s-bursts : %u [%d%% of %d pkts", rx_tx, total_burst,
1339 	       burst_percent[0], (int) pktnb_stats[0]);
1340 	if (burst_stats[0] == total_burst) {
1341 		printf("]\n");
1342 		return;
1343 	}
1344 	if (burst_stats[0] + burst_stats[1] == total_burst) {
1345 		printf(" + %d%% of %d pkts]\n",
1346 		       100 - burst_percent[0], pktnb_stats[1]);
1347 		return;
1348 	}
1349 	burst_percent[1] = (burst_stats[1] * 100) / total_burst;
1350 	burst_percent[2] = 100 - (burst_percent[0] + burst_percent[1]);
1351 	if ((burst_percent[1] == 0) || (burst_percent[2] == 0)) {
1352 		printf(" + %d%% of others]\n", 100 - burst_percent[0]);
1353 		return;
1354 	}
1355 	printf(" + %d%% of %d pkts + %d%% of others]\n",
1356 	       burst_percent[1], (int) pktnb_stats[1], burst_percent[2]);
1357 }
1358 #endif /* RTE_TEST_PMD_RECORD_BURST_STATS */
1359 
1360 static void
1361 fwd_port_stats_display(portid_t port_id, struct rte_eth_stats *stats)
1362 {
1363 	struct rte_port *port;
1364 	uint8_t i;
1365 
1366 	static const char *fwd_stats_border = "----------------------";
1367 
1368 	port = &ports[port_id];
1369 	printf("\n  %s Forward statistics for port %-2d %s\n",
1370 	       fwd_stats_border, port_id, fwd_stats_border);
1371 
1372 	if ((!port->rx_queue_stats_mapping_enabled) && (!port->tx_queue_stats_mapping_enabled)) {
1373 		printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1374 		       "%-"PRIu64"\n",
1375 		       stats->ipackets, stats->imissed,
1376 		       (uint64_t) (stats->ipackets + stats->imissed));
1377 
1378 		if (cur_fwd_eng == &csum_fwd_engine)
1379 			printf("  Bad-ipcsum: %-14"PRIu64" Bad-l4csum: %-14"PRIu64"Bad-outer-l4csum: %-14"PRIu64"\n",
1380 			       port->rx_bad_ip_csum, port->rx_bad_l4_csum,
1381 			       port->rx_bad_outer_l4_csum);
1382 		if ((stats->ierrors + stats->rx_nombuf) > 0) {
1383 			printf("  RX-error: %-"PRIu64"\n",  stats->ierrors);
1384 			printf("  RX-nombufs: %-14"PRIu64"\n", stats->rx_nombuf);
1385 		}
1386 
1387 		printf("  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1388 		       "%-"PRIu64"\n",
1389 		       stats->opackets, port->tx_dropped,
1390 		       (uint64_t) (stats->opackets + port->tx_dropped));
1391 	}
1392 	else {
1393 		printf("  RX-packets:             %14"PRIu64"    RX-dropped:%14"PRIu64"    RX-total:"
1394 		       "%14"PRIu64"\n",
1395 		       stats->ipackets, stats->imissed,
1396 		       (uint64_t) (stats->ipackets + stats->imissed));
1397 
1398 		if (cur_fwd_eng == &csum_fwd_engine)
1399 			printf("  Bad-ipcsum:%14"PRIu64"    Bad-l4csum:%14"PRIu64"    Bad-outer-l4csum: %-14"PRIu64"\n",
1400 			       port->rx_bad_ip_csum, port->rx_bad_l4_csum,
1401 			       port->rx_bad_outer_l4_csum);
1402 		if ((stats->ierrors + stats->rx_nombuf) > 0) {
1403 			printf("  RX-error:%"PRIu64"\n", stats->ierrors);
1404 			printf("  RX-nombufs:             %14"PRIu64"\n",
1405 			       stats->rx_nombuf);
1406 		}
1407 
1408 		printf("  TX-packets:             %14"PRIu64"    TX-dropped:%14"PRIu64"    TX-total:"
1409 		       "%14"PRIu64"\n",
1410 		       stats->opackets, port->tx_dropped,
1411 		       (uint64_t) (stats->opackets + port->tx_dropped));
1412 	}
1413 
1414 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1415 	if (port->rx_stream)
1416 		pkt_burst_stats_display("RX",
1417 			&port->rx_stream->rx_burst_stats);
1418 	if (port->tx_stream)
1419 		pkt_burst_stats_display("TX",
1420 			&port->tx_stream->tx_burst_stats);
1421 #endif
1422 
1423 	if (port->rx_queue_stats_mapping_enabled) {
1424 		printf("\n");
1425 		for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
1426 			printf("  Stats reg %2d RX-packets:%14"PRIu64
1427 			       "     RX-errors:%14"PRIu64
1428 			       "    RX-bytes:%14"PRIu64"\n",
1429 			       i, stats->q_ipackets[i], stats->q_errors[i], stats->q_ibytes[i]);
1430 		}
1431 		printf("\n");
1432 	}
1433 	if (port->tx_queue_stats_mapping_enabled) {
1434 		for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS; i++) {
1435 			printf("  Stats reg %2d TX-packets:%14"PRIu64
1436 			       "                                 TX-bytes:%14"PRIu64"\n",
1437 			       i, stats->q_opackets[i], stats->q_obytes[i]);
1438 		}
1439 	}
1440 
1441 	printf("  %s--------------------------------%s\n",
1442 	       fwd_stats_border, fwd_stats_border);
1443 }
1444 
1445 static void
1446 fwd_stream_stats_display(streamid_t stream_id)
1447 {
1448 	struct fwd_stream *fs;
1449 	static const char *fwd_top_stats_border = "-------";
1450 
1451 	fs = fwd_streams[stream_id];
1452 	if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1453 	    (fs->fwd_dropped == 0))
1454 		return;
1455 	printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1456 	       "TX Port=%2d/Queue=%2d %s\n",
1457 	       fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1458 	       fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1459 	printf("  RX-packets: %-14u TX-packets: %-14u TX-dropped: %-14u",
1460 	       fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1461 
1462 	/* if checksum mode */
1463 	if (cur_fwd_eng == &csum_fwd_engine) {
1464 	       printf("  RX- bad IP checksum: %-14u  Rx- bad L4 checksum: "
1465 			"%-14u Rx- bad outer L4 checksum: %-14u\n",
1466 			fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1467 			fs->rx_bad_outer_l4_csum);
1468 	}
1469 
1470 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1471 	pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1472 	pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1473 #endif
1474 }
1475 
1476 static void
1477 flush_fwd_rx_queues(void)
1478 {
1479 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1480 	portid_t  rxp;
1481 	portid_t port_id;
1482 	queueid_t rxq;
1483 	uint16_t  nb_rx;
1484 	uint16_t  i;
1485 	uint8_t   j;
1486 	uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
1487 	uint64_t timer_period;
1488 
1489 	/* convert to number of cycles */
1490 	timer_period = rte_get_timer_hz(); /* 1 second timeout */
1491 
1492 	for (j = 0; j < 2; j++) {
1493 		for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
1494 			for (rxq = 0; rxq < nb_rxq; rxq++) {
1495 				port_id = fwd_ports_ids[rxp];
1496 				/**
1497 				* testpmd can stuck in the below do while loop
1498 				* if rte_eth_rx_burst() always returns nonzero
1499 				* packets. So timer is added to exit this loop
1500 				* after 1sec timer expiry.
1501 				*/
1502 				prev_tsc = rte_rdtsc();
1503 				do {
1504 					nb_rx = rte_eth_rx_burst(port_id, rxq,
1505 						pkts_burst, MAX_PKT_BURST);
1506 					for (i = 0; i < nb_rx; i++)
1507 						rte_pktmbuf_free(pkts_burst[i]);
1508 
1509 					cur_tsc = rte_rdtsc();
1510 					diff_tsc = cur_tsc - prev_tsc;
1511 					timer_tsc += diff_tsc;
1512 				} while ((nb_rx > 0) &&
1513 					(timer_tsc < timer_period));
1514 				timer_tsc = 0;
1515 			}
1516 		}
1517 		rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
1518 	}
1519 }
1520 
1521 static void
1522 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
1523 {
1524 	struct fwd_stream **fsm;
1525 	streamid_t nb_fs;
1526 	streamid_t sm_id;
1527 #ifdef RTE_LIBRTE_BITRATE
1528 	uint64_t tics_per_1sec;
1529 	uint64_t tics_datum;
1530 	uint64_t tics_current;
1531 	uint16_t i, cnt_ports;
1532 
1533 	cnt_ports = nb_ports;
1534 	tics_datum = rte_rdtsc();
1535 	tics_per_1sec = rte_get_timer_hz();
1536 #endif
1537 	fsm = &fwd_streams[fc->stream_idx];
1538 	nb_fs = fc->stream_nb;
1539 	do {
1540 		for (sm_id = 0; sm_id < nb_fs; sm_id++)
1541 			(*pkt_fwd)(fsm[sm_id]);
1542 #ifdef RTE_LIBRTE_BITRATE
1543 		if (bitrate_enabled != 0 &&
1544 				bitrate_lcore_id == rte_lcore_id()) {
1545 			tics_current = rte_rdtsc();
1546 			if (tics_current - tics_datum >= tics_per_1sec) {
1547 				/* Periodic bitrate calculation */
1548 				for (i = 0; i < cnt_ports; i++)
1549 					rte_stats_bitrate_calc(bitrate_data,
1550 						ports_ids[i]);
1551 				tics_datum = tics_current;
1552 			}
1553 		}
1554 #endif
1555 #ifdef RTE_LIBRTE_LATENCY_STATS
1556 		if (latencystats_enabled != 0 &&
1557 				latencystats_lcore_id == rte_lcore_id())
1558 			rte_latencystats_update();
1559 #endif
1560 
1561 	} while (! fc->stopped);
1562 }
1563 
1564 static int
1565 start_pkt_forward_on_core(void *fwd_arg)
1566 {
1567 	run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
1568 			     cur_fwd_config.fwd_eng->packet_fwd);
1569 	return 0;
1570 }
1571 
1572 /*
1573  * Run the TXONLY packet forwarding engine to send a single burst of packets.
1574  * Used to start communication flows in network loopback test configurations.
1575  */
1576 static int
1577 run_one_txonly_burst_on_core(void *fwd_arg)
1578 {
1579 	struct fwd_lcore *fwd_lc;
1580 	struct fwd_lcore tmp_lcore;
1581 
1582 	fwd_lc = (struct fwd_lcore *) fwd_arg;
1583 	tmp_lcore = *fwd_lc;
1584 	tmp_lcore.stopped = 1;
1585 	run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
1586 	return 0;
1587 }
1588 
1589 /*
1590  * Launch packet forwarding:
1591  *     - Setup per-port forwarding context.
1592  *     - launch logical cores with their forwarding configuration.
1593  */
1594 static void
1595 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
1596 {
1597 	port_fwd_begin_t port_fwd_begin;
1598 	unsigned int i;
1599 	unsigned int lc_id;
1600 	int diag;
1601 
1602 	port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
1603 	if (port_fwd_begin != NULL) {
1604 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1605 			(*port_fwd_begin)(fwd_ports_ids[i]);
1606 	}
1607 	for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
1608 		lc_id = fwd_lcores_cpuids[i];
1609 		if ((interactive == 0) || (lc_id != rte_lcore_id())) {
1610 			fwd_lcores[i]->stopped = 0;
1611 			diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
1612 						     fwd_lcores[i], lc_id);
1613 			if (diag != 0)
1614 				printf("launch lcore %u failed - diag=%d\n",
1615 				       lc_id, diag);
1616 		}
1617 	}
1618 }
1619 
1620 /*
1621  * Launch packet forwarding configuration.
1622  */
1623 void
1624 start_packet_forwarding(int with_tx_first)
1625 {
1626 	port_fwd_begin_t port_fwd_begin;
1627 	port_fwd_end_t  port_fwd_end;
1628 	struct rte_port *port;
1629 	unsigned int i;
1630 	portid_t   pt_id;
1631 	streamid_t sm_id;
1632 
1633 	if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
1634 		rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
1635 
1636 	if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
1637 		rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
1638 
1639 	if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
1640 		strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
1641 		(!nb_rxq || !nb_txq))
1642 		rte_exit(EXIT_FAILURE,
1643 			"Either rxq or txq are 0, cannot use %s fwd mode\n",
1644 			cur_fwd_eng->fwd_mode_name);
1645 
1646 	if (all_ports_started() == 0) {
1647 		printf("Not all ports were started\n");
1648 		return;
1649 	}
1650 	if (test_done == 0) {
1651 		printf("Packet forwarding already started\n");
1652 		return;
1653 	}
1654 
1655 
1656 	if(dcb_test) {
1657 		for (i = 0; i < nb_fwd_ports; i++) {
1658 			pt_id = fwd_ports_ids[i];
1659 			port = &ports[pt_id];
1660 			if (!port->dcb_flag) {
1661 				printf("In DCB mode, all forwarding ports must "
1662                                        "be configured in this mode.\n");
1663 				return;
1664 			}
1665 		}
1666 		if (nb_fwd_lcores == 1) {
1667 			printf("In DCB mode,the nb forwarding cores "
1668                                "should be larger than 1.\n");
1669 			return;
1670 		}
1671 	}
1672 	test_done = 0;
1673 
1674 	fwd_config_setup();
1675 
1676 	if(!no_flush_rx)
1677 		flush_fwd_rx_queues();
1678 
1679 	pkt_fwd_config_display(&cur_fwd_config);
1680 	rxtx_config_display();
1681 
1682 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1683 		pt_id = fwd_ports_ids[i];
1684 		port = &ports[pt_id];
1685 		rte_eth_stats_get(pt_id, &port->stats);
1686 		port->tx_dropped = 0;
1687 
1688 		map_port_queue_stats_mapping_registers(pt_id, port);
1689 	}
1690 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1691 		fwd_streams[sm_id]->rx_packets = 0;
1692 		fwd_streams[sm_id]->tx_packets = 0;
1693 		fwd_streams[sm_id]->fwd_dropped = 0;
1694 		fwd_streams[sm_id]->rx_bad_ip_csum = 0;
1695 		fwd_streams[sm_id]->rx_bad_l4_csum = 0;
1696 		fwd_streams[sm_id]->rx_bad_outer_l4_csum = 0;
1697 
1698 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1699 		memset(&fwd_streams[sm_id]->rx_burst_stats, 0,
1700 		       sizeof(fwd_streams[sm_id]->rx_burst_stats));
1701 		memset(&fwd_streams[sm_id]->tx_burst_stats, 0,
1702 		       sizeof(fwd_streams[sm_id]->tx_burst_stats));
1703 #endif
1704 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1705 		fwd_streams[sm_id]->core_cycles = 0;
1706 #endif
1707 	}
1708 	if (with_tx_first) {
1709 		port_fwd_begin = tx_only_engine.port_fwd_begin;
1710 		if (port_fwd_begin != NULL) {
1711 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1712 				(*port_fwd_begin)(fwd_ports_ids[i]);
1713 		}
1714 		while (with_tx_first--) {
1715 			launch_packet_forwarding(
1716 					run_one_txonly_burst_on_core);
1717 			rte_eal_mp_wait_lcore();
1718 		}
1719 		port_fwd_end = tx_only_engine.port_fwd_end;
1720 		if (port_fwd_end != NULL) {
1721 			for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1722 				(*port_fwd_end)(fwd_ports_ids[i]);
1723 		}
1724 	}
1725 	launch_packet_forwarding(start_pkt_forward_on_core);
1726 }
1727 
1728 void
1729 stop_packet_forwarding(void)
1730 {
1731 	struct rte_eth_stats stats;
1732 	struct rte_port *port;
1733 	port_fwd_end_t  port_fwd_end;
1734 	int i;
1735 	portid_t   pt_id;
1736 	streamid_t sm_id;
1737 	lcoreid_t  lc_id;
1738 	uint64_t total_recv;
1739 	uint64_t total_xmit;
1740 	uint64_t total_rx_dropped;
1741 	uint64_t total_tx_dropped;
1742 	uint64_t total_rx_nombuf;
1743 	uint64_t tx_dropped;
1744 	uint64_t rx_bad_ip_csum;
1745 	uint64_t rx_bad_l4_csum;
1746 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1747 	uint64_t fwd_cycles;
1748 #endif
1749 
1750 	static const char *acc_stats_border = "+++++++++++++++";
1751 
1752 	if (test_done) {
1753 		printf("Packet forwarding not started\n");
1754 		return;
1755 	}
1756 	printf("Telling cores to stop...");
1757 	for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
1758 		fwd_lcores[lc_id]->stopped = 1;
1759 	printf("\nWaiting for lcores to finish...\n");
1760 	rte_eal_mp_wait_lcore();
1761 	port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
1762 	if (port_fwd_end != NULL) {
1763 		for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1764 			pt_id = fwd_ports_ids[i];
1765 			(*port_fwd_end)(pt_id);
1766 		}
1767 	}
1768 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1769 	fwd_cycles = 0;
1770 #endif
1771 	for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1772 		if (cur_fwd_config.nb_fwd_streams >
1773 		    cur_fwd_config.nb_fwd_ports) {
1774 			fwd_stream_stats_display(sm_id);
1775 			ports[fwd_streams[sm_id]->tx_port].tx_stream = NULL;
1776 			ports[fwd_streams[sm_id]->rx_port].rx_stream = NULL;
1777 		} else {
1778 			ports[fwd_streams[sm_id]->tx_port].tx_stream =
1779 				fwd_streams[sm_id];
1780 			ports[fwd_streams[sm_id]->rx_port].rx_stream =
1781 				fwd_streams[sm_id];
1782 		}
1783 		tx_dropped = ports[fwd_streams[sm_id]->tx_port].tx_dropped;
1784 		tx_dropped = (uint64_t) (tx_dropped +
1785 					 fwd_streams[sm_id]->fwd_dropped);
1786 		ports[fwd_streams[sm_id]->tx_port].tx_dropped = tx_dropped;
1787 
1788 		rx_bad_ip_csum =
1789 			ports[fwd_streams[sm_id]->rx_port].rx_bad_ip_csum;
1790 		rx_bad_ip_csum = (uint64_t) (rx_bad_ip_csum +
1791 					 fwd_streams[sm_id]->rx_bad_ip_csum);
1792 		ports[fwd_streams[sm_id]->rx_port].rx_bad_ip_csum =
1793 							rx_bad_ip_csum;
1794 
1795 		rx_bad_l4_csum =
1796 			ports[fwd_streams[sm_id]->rx_port].rx_bad_l4_csum;
1797 		rx_bad_l4_csum = (uint64_t) (rx_bad_l4_csum +
1798 					 fwd_streams[sm_id]->rx_bad_l4_csum);
1799 		ports[fwd_streams[sm_id]->rx_port].rx_bad_l4_csum =
1800 							rx_bad_l4_csum;
1801 
1802 		ports[fwd_streams[sm_id]->rx_port].rx_bad_outer_l4_csum +=
1803 				fwd_streams[sm_id]->rx_bad_outer_l4_csum;
1804 
1805 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1806 		fwd_cycles = (uint64_t) (fwd_cycles +
1807 					 fwd_streams[sm_id]->core_cycles);
1808 #endif
1809 	}
1810 	total_recv = 0;
1811 	total_xmit = 0;
1812 	total_rx_dropped = 0;
1813 	total_tx_dropped = 0;
1814 	total_rx_nombuf  = 0;
1815 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1816 		pt_id = fwd_ports_ids[i];
1817 
1818 		port = &ports[pt_id];
1819 		rte_eth_stats_get(pt_id, &stats);
1820 		stats.ipackets -= port->stats.ipackets;
1821 		port->stats.ipackets = 0;
1822 		stats.opackets -= port->stats.opackets;
1823 		port->stats.opackets = 0;
1824 		stats.ibytes   -= port->stats.ibytes;
1825 		port->stats.ibytes = 0;
1826 		stats.obytes   -= port->stats.obytes;
1827 		port->stats.obytes = 0;
1828 		stats.imissed  -= port->stats.imissed;
1829 		port->stats.imissed = 0;
1830 		stats.oerrors  -= port->stats.oerrors;
1831 		port->stats.oerrors = 0;
1832 		stats.rx_nombuf -= port->stats.rx_nombuf;
1833 		port->stats.rx_nombuf = 0;
1834 
1835 		total_recv += stats.ipackets;
1836 		total_xmit += stats.opackets;
1837 		total_rx_dropped += stats.imissed;
1838 		total_tx_dropped += port->tx_dropped;
1839 		total_rx_nombuf  += stats.rx_nombuf;
1840 
1841 		fwd_port_stats_display(pt_id, &stats);
1842 	}
1843 
1844 	printf("\n  %s Accumulated forward statistics for all ports"
1845 	       "%s\n",
1846 	       acc_stats_border, acc_stats_border);
1847 	printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1848 	       "%-"PRIu64"\n"
1849 	       "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1850 	       "%-"PRIu64"\n",
1851 	       total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1852 	       total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1853 	if (total_rx_nombuf > 0)
1854 		printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1855 	printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
1856 	       "%s\n",
1857 	       acc_stats_border, acc_stats_border);
1858 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1859 	if (total_recv > 0)
1860 		printf("\n  CPU cycles/packet=%u (total cycles="
1861 		       "%"PRIu64" / total RX packets=%"PRIu64")\n",
1862 		       (unsigned int)(fwd_cycles / total_recv),
1863 		       fwd_cycles, total_recv);
1864 #endif
1865 	printf("\nDone.\n");
1866 	test_done = 1;
1867 }
1868 
1869 void
1870 dev_set_link_up(portid_t pid)
1871 {
1872 	if (rte_eth_dev_set_link_up(pid) < 0)
1873 		printf("\nSet link up fail.\n");
1874 }
1875 
1876 void
1877 dev_set_link_down(portid_t pid)
1878 {
1879 	if (rte_eth_dev_set_link_down(pid) < 0)
1880 		printf("\nSet link down fail.\n");
1881 }
1882 
1883 static int
1884 all_ports_started(void)
1885 {
1886 	portid_t pi;
1887 	struct rte_port *port;
1888 
1889 	RTE_ETH_FOREACH_DEV(pi) {
1890 		port = &ports[pi];
1891 		/* Check if there is a port which is not started */
1892 		if ((port->port_status != RTE_PORT_STARTED) &&
1893 			(port->slave_flag == 0))
1894 			return 0;
1895 	}
1896 
1897 	/* No port is not started */
1898 	return 1;
1899 }
1900 
1901 int
1902 port_is_stopped(portid_t port_id)
1903 {
1904 	struct rte_port *port = &ports[port_id];
1905 
1906 	if ((port->port_status != RTE_PORT_STOPPED) &&
1907 	    (port->slave_flag == 0))
1908 		return 0;
1909 	return 1;
1910 }
1911 
1912 int
1913 all_ports_stopped(void)
1914 {
1915 	portid_t pi;
1916 
1917 	RTE_ETH_FOREACH_DEV(pi) {
1918 		if (!port_is_stopped(pi))
1919 			return 0;
1920 	}
1921 
1922 	return 1;
1923 }
1924 
1925 int
1926 port_is_started(portid_t port_id)
1927 {
1928 	if (port_id_is_invalid(port_id, ENABLED_WARN))
1929 		return 0;
1930 
1931 	if (ports[port_id].port_status != RTE_PORT_STARTED)
1932 		return 0;
1933 
1934 	return 1;
1935 }
1936 
1937 int
1938 start_port(portid_t pid)
1939 {
1940 	int diag, need_check_link_status = -1;
1941 	portid_t pi;
1942 	queueid_t qi;
1943 	struct rte_port *port;
1944 	struct ether_addr mac_addr;
1945 
1946 	if (port_id_is_invalid(pid, ENABLED_WARN))
1947 		return 0;
1948 
1949 	if(dcb_config)
1950 		dcb_test = 1;
1951 	RTE_ETH_FOREACH_DEV(pi) {
1952 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
1953 			continue;
1954 
1955 		need_check_link_status = 0;
1956 		port = &ports[pi];
1957 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
1958 						 RTE_PORT_HANDLING) == 0) {
1959 			printf("Port %d is now not stopped\n", pi);
1960 			continue;
1961 		}
1962 
1963 		if (port->need_reconfig > 0) {
1964 			port->need_reconfig = 0;
1965 
1966 			if (flow_isolate_all) {
1967 				int ret = port_flow_isolate(pi, 1);
1968 				if (ret) {
1969 					printf("Failed to apply isolated"
1970 					       " mode on port %d\n", pi);
1971 					return -1;
1972 				}
1973 			}
1974 			configure_rxtx_dump_callbacks(0);
1975 			printf("Configuring Port %d (socket %u)\n", pi,
1976 					port->socket_id);
1977 			/* configure port */
1978 			diag = rte_eth_dev_configure(pi, nb_rxq, nb_txq,
1979 						&(port->dev_conf));
1980 			if (diag != 0) {
1981 				if (rte_atomic16_cmpset(&(port->port_status),
1982 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
1983 					printf("Port %d can not be set back "
1984 							"to stopped\n", pi);
1985 				printf("Fail to configure port %d\n", pi);
1986 				/* try to reconfigure port next time */
1987 				port->need_reconfig = 1;
1988 				return -1;
1989 			}
1990 		}
1991 		if (port->need_reconfig_queues > 0) {
1992 			port->need_reconfig_queues = 0;
1993 			/* setup tx queues */
1994 			for (qi = 0; qi < nb_txq; qi++) {
1995 				if ((numa_support) &&
1996 					(txring_numa[pi] != NUMA_NO_CONFIG))
1997 					diag = rte_eth_tx_queue_setup(pi, qi,
1998 						port->nb_tx_desc[qi],
1999 						txring_numa[pi],
2000 						&(port->tx_conf[qi]));
2001 				else
2002 					diag = rte_eth_tx_queue_setup(pi, qi,
2003 						port->nb_tx_desc[qi],
2004 						port->socket_id,
2005 						&(port->tx_conf[qi]));
2006 
2007 				if (diag == 0)
2008 					continue;
2009 
2010 				/* Fail to setup tx queue, return */
2011 				if (rte_atomic16_cmpset(&(port->port_status),
2012 							RTE_PORT_HANDLING,
2013 							RTE_PORT_STOPPED) == 0)
2014 					printf("Port %d can not be set back "
2015 							"to stopped\n", pi);
2016 				printf("Fail to configure port %d tx queues\n",
2017 				       pi);
2018 				/* try to reconfigure queues next time */
2019 				port->need_reconfig_queues = 1;
2020 				return -1;
2021 			}
2022 			for (qi = 0; qi < nb_rxq; qi++) {
2023 				/* setup rx queues */
2024 				if ((numa_support) &&
2025 					(rxring_numa[pi] != NUMA_NO_CONFIG)) {
2026 					struct rte_mempool * mp =
2027 						mbuf_pool_find(rxring_numa[pi]);
2028 					if (mp == NULL) {
2029 						printf("Failed to setup RX queue:"
2030 							"No mempool allocation"
2031 							" on the socket %d\n",
2032 							rxring_numa[pi]);
2033 						return -1;
2034 					}
2035 
2036 					diag = rte_eth_rx_queue_setup(pi, qi,
2037 					     port->nb_rx_desc[qi],
2038 					     rxring_numa[pi],
2039 					     &(port->rx_conf[qi]),
2040 					     mp);
2041 				} else {
2042 					struct rte_mempool *mp =
2043 						mbuf_pool_find(port->socket_id);
2044 					if (mp == NULL) {
2045 						printf("Failed to setup RX queue:"
2046 							"No mempool allocation"
2047 							" on the socket %d\n",
2048 							port->socket_id);
2049 						return -1;
2050 					}
2051 					diag = rte_eth_rx_queue_setup(pi, qi,
2052 					     port->nb_rx_desc[qi],
2053 					     port->socket_id,
2054 					     &(port->rx_conf[qi]),
2055 					     mp);
2056 				}
2057 				if (diag == 0)
2058 					continue;
2059 
2060 				/* Fail to setup rx queue, return */
2061 				if (rte_atomic16_cmpset(&(port->port_status),
2062 							RTE_PORT_HANDLING,
2063 							RTE_PORT_STOPPED) == 0)
2064 					printf("Port %d can not be set back "
2065 							"to stopped\n", pi);
2066 				printf("Fail to configure port %d rx queues\n",
2067 				       pi);
2068 				/* try to reconfigure queues next time */
2069 				port->need_reconfig_queues = 1;
2070 				return -1;
2071 			}
2072 		}
2073 		configure_rxtx_dump_callbacks(verbose_level);
2074 		/* start port */
2075 		if (rte_eth_dev_start(pi) < 0) {
2076 			printf("Fail to start port %d\n", pi);
2077 
2078 			/* Fail to setup rx queue, return */
2079 			if (rte_atomic16_cmpset(&(port->port_status),
2080 				RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2081 				printf("Port %d can not be set back to "
2082 							"stopped\n", pi);
2083 			continue;
2084 		}
2085 
2086 		if (rte_atomic16_cmpset(&(port->port_status),
2087 			RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2088 			printf("Port %d can not be set into started\n", pi);
2089 
2090 		rte_eth_macaddr_get(pi, &mac_addr);
2091 		printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2092 				mac_addr.addr_bytes[0], mac_addr.addr_bytes[1],
2093 				mac_addr.addr_bytes[2], mac_addr.addr_bytes[3],
2094 				mac_addr.addr_bytes[4], mac_addr.addr_bytes[5]);
2095 
2096 		/* at least one port started, need checking link status */
2097 		need_check_link_status = 1;
2098 	}
2099 
2100 	if (need_check_link_status == 1 && !no_link_check)
2101 		check_all_ports_link_status(RTE_PORT_ALL);
2102 	else if (need_check_link_status == 0)
2103 		printf("Please stop the ports first\n");
2104 
2105 	printf("Done\n");
2106 	return 0;
2107 }
2108 
2109 void
2110 stop_port(portid_t pid)
2111 {
2112 	portid_t pi;
2113 	struct rte_port *port;
2114 	int need_check_link_status = 0;
2115 
2116 	if (dcb_test) {
2117 		dcb_test = 0;
2118 		dcb_config = 0;
2119 	}
2120 
2121 	if (port_id_is_invalid(pid, ENABLED_WARN))
2122 		return;
2123 
2124 	printf("Stopping ports...\n");
2125 
2126 	RTE_ETH_FOREACH_DEV(pi) {
2127 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2128 			continue;
2129 
2130 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2131 			printf("Please remove port %d from forwarding configuration.\n", pi);
2132 			continue;
2133 		}
2134 
2135 		if (port_is_bonding_slave(pi)) {
2136 			printf("Please remove port %d from bonded device.\n", pi);
2137 			continue;
2138 		}
2139 
2140 		port = &ports[pi];
2141 		if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2142 						RTE_PORT_HANDLING) == 0)
2143 			continue;
2144 
2145 		rte_eth_dev_stop(pi);
2146 
2147 		if (rte_atomic16_cmpset(&(port->port_status),
2148 			RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2149 			printf("Port %d can not be set into stopped\n", pi);
2150 		need_check_link_status = 1;
2151 	}
2152 	if (need_check_link_status && !no_link_check)
2153 		check_all_ports_link_status(RTE_PORT_ALL);
2154 
2155 	printf("Done\n");
2156 }
2157 
2158 static void
2159 remove_invalid_ports_in(portid_t *array, portid_t *total)
2160 {
2161 	portid_t i;
2162 	portid_t new_total = 0;
2163 
2164 	for (i = 0; i < *total; i++)
2165 		if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2166 			array[new_total] = array[i];
2167 			new_total++;
2168 		}
2169 	*total = new_total;
2170 }
2171 
2172 static void
2173 remove_invalid_ports(void)
2174 {
2175 	remove_invalid_ports_in(ports_ids, &nb_ports);
2176 	remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2177 	nb_cfg_ports = nb_fwd_ports;
2178 }
2179 
2180 void
2181 close_port(portid_t pid)
2182 {
2183 	portid_t pi;
2184 	struct rte_port *port;
2185 
2186 	if (port_id_is_invalid(pid, ENABLED_WARN))
2187 		return;
2188 
2189 	printf("Closing ports...\n");
2190 
2191 	RTE_ETH_FOREACH_DEV(pi) {
2192 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2193 			continue;
2194 
2195 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2196 			printf("Please remove port %d from forwarding configuration.\n", pi);
2197 			continue;
2198 		}
2199 
2200 		if (port_is_bonding_slave(pi)) {
2201 			printf("Please remove port %d from bonded device.\n", pi);
2202 			continue;
2203 		}
2204 
2205 		port = &ports[pi];
2206 		if (rte_atomic16_cmpset(&(port->port_status),
2207 			RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2208 			printf("Port %d is already closed\n", pi);
2209 			continue;
2210 		}
2211 
2212 		if (rte_atomic16_cmpset(&(port->port_status),
2213 			RTE_PORT_STOPPED, RTE_PORT_HANDLING) == 0) {
2214 			printf("Port %d is now not stopped\n", pi);
2215 			continue;
2216 		}
2217 
2218 		if (port->flow_list)
2219 			port_flow_flush(pi);
2220 		rte_eth_dev_close(pi);
2221 
2222 		remove_invalid_ports();
2223 
2224 		if (rte_atomic16_cmpset(&(port->port_status),
2225 			RTE_PORT_HANDLING, RTE_PORT_CLOSED) == 0)
2226 			printf("Port %d cannot be set to closed\n", pi);
2227 	}
2228 
2229 	printf("Done\n");
2230 }
2231 
2232 void
2233 reset_port(portid_t pid)
2234 {
2235 	int diag;
2236 	portid_t pi;
2237 	struct rte_port *port;
2238 
2239 	if (port_id_is_invalid(pid, ENABLED_WARN))
2240 		return;
2241 
2242 	printf("Resetting ports...\n");
2243 
2244 	RTE_ETH_FOREACH_DEV(pi) {
2245 		if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2246 			continue;
2247 
2248 		if (port_is_forwarding(pi) != 0 && test_done == 0) {
2249 			printf("Please remove port %d from forwarding "
2250 			       "configuration.\n", pi);
2251 			continue;
2252 		}
2253 
2254 		if (port_is_bonding_slave(pi)) {
2255 			printf("Please remove port %d from bonded device.\n",
2256 			       pi);
2257 			continue;
2258 		}
2259 
2260 		diag = rte_eth_dev_reset(pi);
2261 		if (diag == 0) {
2262 			port = &ports[pi];
2263 			port->need_reconfig = 1;
2264 			port->need_reconfig_queues = 1;
2265 		} else {
2266 			printf("Failed to reset port %d. diag=%d\n", pi, diag);
2267 		}
2268 	}
2269 
2270 	printf("Done\n");
2271 }
2272 
2273 void
2274 attach_port(char *identifier)
2275 {
2276 	portid_t pi;
2277 	struct rte_dev_iterator iterator;
2278 
2279 	printf("Attaching a new port...\n");
2280 
2281 	if (identifier == NULL) {
2282 		printf("Invalid parameters are specified\n");
2283 		return;
2284 	}
2285 
2286 	if (rte_dev_probe(identifier) != 0) {
2287 		TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2288 		return;
2289 	}
2290 
2291 	/* first attach mode: event */
2292 	if (setup_on_probe_event) {
2293 		/* new ports are detected on RTE_ETH_EVENT_NEW event */
2294 		for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2295 			if (ports[pi].port_status == RTE_PORT_HANDLING &&
2296 					ports[pi].need_setup != 0)
2297 				setup_attached_port(pi);
2298 		return;
2299 	}
2300 
2301 	/* second attach mode: iterator */
2302 	RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2303 		/* setup ports matching the devargs used for probing */
2304 		if (port_is_forwarding(pi))
2305 			continue; /* port was already attached before */
2306 		setup_attached_port(pi);
2307 	}
2308 }
2309 
2310 static void
2311 setup_attached_port(portid_t pi)
2312 {
2313 	unsigned int socket_id;
2314 
2315 	socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2316 	/* if socket_id is invalid, set to the first available socket. */
2317 	if (check_socket_id(socket_id) < 0)
2318 		socket_id = socket_ids[0];
2319 	reconfig(pi, socket_id);
2320 	rte_eth_promiscuous_enable(pi);
2321 
2322 	ports_ids[nb_ports++] = pi;
2323 	fwd_ports_ids[nb_fwd_ports++] = pi;
2324 	nb_cfg_ports = nb_fwd_ports;
2325 	ports[pi].need_setup = 0;
2326 	ports[pi].port_status = RTE_PORT_STOPPED;
2327 
2328 	printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2329 	printf("Done\n");
2330 }
2331 
2332 void
2333 detach_port_device(portid_t port_id)
2334 {
2335 	struct rte_device *dev;
2336 	portid_t sibling;
2337 
2338 	printf("Removing a device...\n");
2339 
2340 	dev = rte_eth_devices[port_id].device;
2341 	if (dev == NULL) {
2342 		printf("Device already removed\n");
2343 		return;
2344 	}
2345 
2346 	if (ports[port_id].port_status != RTE_PORT_CLOSED) {
2347 		if (ports[port_id].port_status != RTE_PORT_STOPPED) {
2348 			printf("Port not stopped\n");
2349 			return;
2350 		}
2351 		printf("Port was not closed\n");
2352 		if (ports[port_id].flow_list)
2353 			port_flow_flush(port_id);
2354 	}
2355 
2356 	if (rte_dev_remove(dev) != 0) {
2357 		TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
2358 		return;
2359 	}
2360 
2361 	for (sibling = 0; sibling < RTE_MAX_ETHPORTS; sibling++) {
2362 		if (rte_eth_devices[sibling].device != dev)
2363 			continue;
2364 		/* reset mapping between old ports and removed device */
2365 		rte_eth_devices[sibling].device = NULL;
2366 		if (ports[sibling].port_status != RTE_PORT_CLOSED) {
2367 			/* sibling ports are forced to be closed */
2368 			ports[sibling].port_status = RTE_PORT_CLOSED;
2369 			printf("Port %u is closed\n", sibling);
2370 		}
2371 	}
2372 
2373 	remove_invalid_ports();
2374 
2375 	printf("Device of port %u is detached\n", port_id);
2376 	printf("Now total ports is %d\n", nb_ports);
2377 	printf("Done\n");
2378 	return;
2379 }
2380 
2381 void
2382 pmd_test_exit(void)
2383 {
2384 	struct rte_device *device;
2385 	portid_t pt_id;
2386 	int ret;
2387 
2388 	if (test_done == 0)
2389 		stop_packet_forwarding();
2390 
2391 	if (ports != NULL) {
2392 		no_link_check = 1;
2393 		RTE_ETH_FOREACH_DEV(pt_id) {
2394 			printf("\nStopping port %d...\n", pt_id);
2395 			fflush(stdout);
2396 			stop_port(pt_id);
2397 		}
2398 		RTE_ETH_FOREACH_DEV(pt_id) {
2399 			printf("\nShutting down port %d...\n", pt_id);
2400 			fflush(stdout);
2401 			close_port(pt_id);
2402 
2403 			/*
2404 			 * This is a workaround to fix a virtio-user issue that
2405 			 * requires to call clean-up routine to remove existing
2406 			 * socket.
2407 			 * This workaround valid only for testpmd, needs a fix
2408 			 * valid for all applications.
2409 			 * TODO: Implement proper resource cleanup
2410 			 */
2411 			device = rte_eth_devices[pt_id].device;
2412 			if (device && !strcmp(device->driver->name, "net_virtio_user"))
2413 				detach_port_device(pt_id);
2414 		}
2415 	}
2416 
2417 	if (hot_plug) {
2418 		ret = rte_dev_event_monitor_stop();
2419 		if (ret) {
2420 			RTE_LOG(ERR, EAL,
2421 				"fail to stop device event monitor.");
2422 			return;
2423 		}
2424 
2425 		ret = rte_dev_event_callback_unregister(NULL,
2426 			dev_event_callback, NULL);
2427 		if (ret < 0) {
2428 			RTE_LOG(ERR, EAL,
2429 				"fail to unregister device event callback.\n");
2430 			return;
2431 		}
2432 
2433 		ret = rte_dev_hotplug_handle_disable();
2434 		if (ret) {
2435 			RTE_LOG(ERR, EAL,
2436 				"fail to disable hotplug handling.\n");
2437 			return;
2438 		}
2439 	}
2440 
2441 	printf("\nBye...\n");
2442 }
2443 
2444 typedef void (*cmd_func_t)(void);
2445 struct pmd_test_command {
2446 	const char *cmd_name;
2447 	cmd_func_t cmd_func;
2448 };
2449 
2450 #define PMD_TEST_CMD_NB (sizeof(pmd_test_menu) / sizeof(pmd_test_menu[0]))
2451 
2452 /* Check the link status of all ports in up to 9s, and print them finally */
2453 static void
2454 check_all_ports_link_status(uint32_t port_mask)
2455 {
2456 #define CHECK_INTERVAL 100 /* 100ms */
2457 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2458 	portid_t portid;
2459 	uint8_t count, all_ports_up, print_flag = 0;
2460 	struct rte_eth_link link;
2461 
2462 	printf("Checking link statuses...\n");
2463 	fflush(stdout);
2464 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
2465 		all_ports_up = 1;
2466 		RTE_ETH_FOREACH_DEV(portid) {
2467 			if ((port_mask & (1 << portid)) == 0)
2468 				continue;
2469 			memset(&link, 0, sizeof(link));
2470 			rte_eth_link_get_nowait(portid, &link);
2471 			/* print link status if flag set */
2472 			if (print_flag == 1) {
2473 				if (link.link_status)
2474 					printf(
2475 					"Port%d Link Up. speed %u Mbps- %s\n",
2476 					portid, link.link_speed,
2477 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2478 					("full-duplex") : ("half-duplex\n"));
2479 				else
2480 					printf("Port %d Link Down\n", portid);
2481 				continue;
2482 			}
2483 			/* clear all_ports_up flag if any link down */
2484 			if (link.link_status == ETH_LINK_DOWN) {
2485 				all_ports_up = 0;
2486 				break;
2487 			}
2488 		}
2489 		/* after finally printing all link status, get out */
2490 		if (print_flag == 1)
2491 			break;
2492 
2493 		if (all_ports_up == 0) {
2494 			fflush(stdout);
2495 			rte_delay_ms(CHECK_INTERVAL);
2496 		}
2497 
2498 		/* set the print_flag if all ports up or timeout */
2499 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2500 			print_flag = 1;
2501 		}
2502 
2503 		if (lsc_interrupt)
2504 			break;
2505 	}
2506 }
2507 
2508 /*
2509  * This callback is for remove a port for a device. It has limitation because
2510  * it is not for multiple port removal for a device.
2511  * TODO: the device detach invoke will plan to be removed from user side to
2512  * eal. And convert all PMDs to free port resources on ether device closing.
2513  */
2514 static void
2515 rmv_port_callback(void *arg)
2516 {
2517 	int need_to_start = 0;
2518 	int org_no_link_check = no_link_check;
2519 	portid_t port_id = (intptr_t)arg;
2520 
2521 	RTE_ETH_VALID_PORTID_OR_RET(port_id);
2522 
2523 	if (!test_done && port_is_forwarding(port_id)) {
2524 		need_to_start = 1;
2525 		stop_packet_forwarding();
2526 	}
2527 	no_link_check = 1;
2528 	stop_port(port_id);
2529 	no_link_check = org_no_link_check;
2530 	close_port(port_id);
2531 	detach_port_device(port_id);
2532 	if (need_to_start)
2533 		start_packet_forwarding(0);
2534 }
2535 
2536 /* This function is used by the interrupt thread */
2537 static int
2538 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
2539 		  void *ret_param)
2540 {
2541 	RTE_SET_USED(param);
2542 	RTE_SET_USED(ret_param);
2543 
2544 	if (type >= RTE_ETH_EVENT_MAX) {
2545 		fprintf(stderr, "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
2546 			port_id, __func__, type);
2547 		fflush(stderr);
2548 	} else if (event_print_mask & (UINT32_C(1) << type)) {
2549 		printf("\nPort %" PRIu16 ": %s event\n", port_id,
2550 			eth_event_desc[type]);
2551 		fflush(stdout);
2552 	}
2553 
2554 	switch (type) {
2555 	case RTE_ETH_EVENT_NEW:
2556 		ports[port_id].need_setup = 1;
2557 		ports[port_id].port_status = RTE_PORT_HANDLING;
2558 		break;
2559 	case RTE_ETH_EVENT_INTR_RMV:
2560 		if (port_id_is_invalid(port_id, DISABLED_WARN))
2561 			break;
2562 		if (rte_eal_alarm_set(100000,
2563 				rmv_port_callback, (void *)(intptr_t)port_id))
2564 			fprintf(stderr, "Could not set up deferred device removal\n");
2565 		break;
2566 	default:
2567 		break;
2568 	}
2569 	return 0;
2570 }
2571 
2572 static int
2573 register_eth_event_callback(void)
2574 {
2575 	int ret;
2576 	enum rte_eth_event_type event;
2577 
2578 	for (event = RTE_ETH_EVENT_UNKNOWN;
2579 			event < RTE_ETH_EVENT_MAX; event++) {
2580 		ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
2581 				event,
2582 				eth_event_callback,
2583 				NULL);
2584 		if (ret != 0) {
2585 			TESTPMD_LOG(ERR, "Failed to register callback for "
2586 					"%s event\n", eth_event_desc[event]);
2587 			return -1;
2588 		}
2589 	}
2590 
2591 	return 0;
2592 }
2593 
2594 /* This function is used by the interrupt thread */
2595 static void
2596 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
2597 			     __rte_unused void *arg)
2598 {
2599 	uint16_t port_id;
2600 	int ret;
2601 
2602 	if (type >= RTE_DEV_EVENT_MAX) {
2603 		fprintf(stderr, "%s called upon invalid event %d\n",
2604 			__func__, type);
2605 		fflush(stderr);
2606 	}
2607 
2608 	switch (type) {
2609 	case RTE_DEV_EVENT_REMOVE:
2610 		RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
2611 			device_name);
2612 		ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
2613 		if (ret) {
2614 			RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
2615 				device_name);
2616 			return;
2617 		}
2618 		/*
2619 		 * Because the user's callback is invoked in eal interrupt
2620 		 * callback, the interrupt callback need to be finished before
2621 		 * it can be unregistered when detaching device. So finish
2622 		 * callback soon and use a deferred removal to detach device
2623 		 * is need. It is a workaround, once the device detaching be
2624 		 * moved into the eal in the future, the deferred removal could
2625 		 * be deleted.
2626 		 */
2627 		if (rte_eal_alarm_set(100000,
2628 				rmv_port_callback, (void *)(intptr_t)port_id))
2629 			RTE_LOG(ERR, EAL,
2630 				"Could not set up deferred device removal\n");
2631 		break;
2632 	case RTE_DEV_EVENT_ADD:
2633 		RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
2634 			device_name);
2635 		/* TODO: After finish kernel driver binding,
2636 		 * begin to attach port.
2637 		 */
2638 		break;
2639 	default:
2640 		break;
2641 	}
2642 }
2643 
2644 static int
2645 set_tx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2646 {
2647 	uint16_t i;
2648 	int diag;
2649 	uint8_t mapping_found = 0;
2650 
2651 	for (i = 0; i < nb_tx_queue_stats_mappings; i++) {
2652 		if ((tx_queue_stats_mappings[i].port_id == port_id) &&
2653 				(tx_queue_stats_mappings[i].queue_id < nb_txq )) {
2654 			diag = rte_eth_dev_set_tx_queue_stats_mapping(port_id,
2655 					tx_queue_stats_mappings[i].queue_id,
2656 					tx_queue_stats_mappings[i].stats_counter_id);
2657 			if (diag != 0)
2658 				return diag;
2659 			mapping_found = 1;
2660 		}
2661 	}
2662 	if (mapping_found)
2663 		port->tx_queue_stats_mapping_enabled = 1;
2664 	return 0;
2665 }
2666 
2667 static int
2668 set_rx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2669 {
2670 	uint16_t i;
2671 	int diag;
2672 	uint8_t mapping_found = 0;
2673 
2674 	for (i = 0; i < nb_rx_queue_stats_mappings; i++) {
2675 		if ((rx_queue_stats_mappings[i].port_id == port_id) &&
2676 				(rx_queue_stats_mappings[i].queue_id < nb_rxq )) {
2677 			diag = rte_eth_dev_set_rx_queue_stats_mapping(port_id,
2678 					rx_queue_stats_mappings[i].queue_id,
2679 					rx_queue_stats_mappings[i].stats_counter_id);
2680 			if (diag != 0)
2681 				return diag;
2682 			mapping_found = 1;
2683 		}
2684 	}
2685 	if (mapping_found)
2686 		port->rx_queue_stats_mapping_enabled = 1;
2687 	return 0;
2688 }
2689 
2690 static void
2691 map_port_queue_stats_mapping_registers(portid_t pi, struct rte_port *port)
2692 {
2693 	int diag = 0;
2694 
2695 	diag = set_tx_queue_stats_mapping_registers(pi, port);
2696 	if (diag != 0) {
2697 		if (diag == -ENOTSUP) {
2698 			port->tx_queue_stats_mapping_enabled = 0;
2699 			printf("TX queue stats mapping not supported port id=%d\n", pi);
2700 		}
2701 		else
2702 			rte_exit(EXIT_FAILURE,
2703 					"set_tx_queue_stats_mapping_registers "
2704 					"failed for port id=%d diag=%d\n",
2705 					pi, diag);
2706 	}
2707 
2708 	diag = set_rx_queue_stats_mapping_registers(pi, port);
2709 	if (diag != 0) {
2710 		if (diag == -ENOTSUP) {
2711 			port->rx_queue_stats_mapping_enabled = 0;
2712 			printf("RX queue stats mapping not supported port id=%d\n", pi);
2713 		}
2714 		else
2715 			rte_exit(EXIT_FAILURE,
2716 					"set_rx_queue_stats_mapping_registers "
2717 					"failed for port id=%d diag=%d\n",
2718 					pi, diag);
2719 	}
2720 }
2721 
2722 static void
2723 rxtx_port_config(struct rte_port *port)
2724 {
2725 	uint16_t qid;
2726 
2727 	for (qid = 0; qid < nb_rxq; qid++) {
2728 		port->rx_conf[qid] = port->dev_info.default_rxconf;
2729 
2730 		/* Check if any Rx parameters have been passed */
2731 		if (rx_pthresh != RTE_PMD_PARAM_UNSET)
2732 			port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
2733 
2734 		if (rx_hthresh != RTE_PMD_PARAM_UNSET)
2735 			port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
2736 
2737 		if (rx_wthresh != RTE_PMD_PARAM_UNSET)
2738 			port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
2739 
2740 		if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
2741 			port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
2742 
2743 		if (rx_drop_en != RTE_PMD_PARAM_UNSET)
2744 			port->rx_conf[qid].rx_drop_en = rx_drop_en;
2745 
2746 		port->nb_rx_desc[qid] = nb_rxd;
2747 	}
2748 
2749 	for (qid = 0; qid < nb_txq; qid++) {
2750 		port->tx_conf[qid] = port->dev_info.default_txconf;
2751 
2752 		/* Check if any Tx parameters have been passed */
2753 		if (tx_pthresh != RTE_PMD_PARAM_UNSET)
2754 			port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
2755 
2756 		if (tx_hthresh != RTE_PMD_PARAM_UNSET)
2757 			port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
2758 
2759 		if (tx_wthresh != RTE_PMD_PARAM_UNSET)
2760 			port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
2761 
2762 		if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
2763 			port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
2764 
2765 		if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
2766 			port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
2767 
2768 		port->nb_tx_desc[qid] = nb_txd;
2769 	}
2770 }
2771 
2772 void
2773 init_port_config(void)
2774 {
2775 	portid_t pid;
2776 	struct rte_port *port;
2777 
2778 	RTE_ETH_FOREACH_DEV(pid) {
2779 		port = &ports[pid];
2780 		port->dev_conf.fdir_conf = fdir_conf;
2781 		rte_eth_dev_info_get(pid, &port->dev_info);
2782 		if (nb_rxq > 1) {
2783 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2784 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
2785 				rss_hf & port->dev_info.flow_type_rss_offloads;
2786 		} else {
2787 			port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2788 			port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
2789 		}
2790 
2791 		if (port->dcb_flag == 0) {
2792 			if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
2793 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
2794 			else
2795 				port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
2796 		}
2797 
2798 		rxtx_port_config(port);
2799 
2800 		rte_eth_macaddr_get(pid, &port->eth_addr);
2801 
2802 		map_port_queue_stats_mapping_registers(pid, port);
2803 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
2804 		rte_pmd_ixgbe_bypass_init(pid);
2805 #endif
2806 
2807 		if (lsc_interrupt &&
2808 		    (rte_eth_devices[pid].data->dev_flags &
2809 		     RTE_ETH_DEV_INTR_LSC))
2810 			port->dev_conf.intr_conf.lsc = 1;
2811 		if (rmv_interrupt &&
2812 		    (rte_eth_devices[pid].data->dev_flags &
2813 		     RTE_ETH_DEV_INTR_RMV))
2814 			port->dev_conf.intr_conf.rmv = 1;
2815 	}
2816 }
2817 
2818 void set_port_slave_flag(portid_t slave_pid)
2819 {
2820 	struct rte_port *port;
2821 
2822 	port = &ports[slave_pid];
2823 	port->slave_flag = 1;
2824 }
2825 
2826 void clear_port_slave_flag(portid_t slave_pid)
2827 {
2828 	struct rte_port *port;
2829 
2830 	port = &ports[slave_pid];
2831 	port->slave_flag = 0;
2832 }
2833 
2834 uint8_t port_is_bonding_slave(portid_t slave_pid)
2835 {
2836 	struct rte_port *port;
2837 
2838 	port = &ports[slave_pid];
2839 	if ((rte_eth_devices[slave_pid].data->dev_flags &
2840 	    RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
2841 		return 1;
2842 	return 0;
2843 }
2844 
2845 const uint16_t vlan_tags[] = {
2846 		0,  1,  2,  3,  4,  5,  6,  7,
2847 		8,  9, 10, 11,  12, 13, 14, 15,
2848 		16, 17, 18, 19, 20, 21, 22, 23,
2849 		24, 25, 26, 27, 28, 29, 30, 31
2850 };
2851 
2852 static  int
2853 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
2854 		 enum dcb_mode_enable dcb_mode,
2855 		 enum rte_eth_nb_tcs num_tcs,
2856 		 uint8_t pfc_en)
2857 {
2858 	uint8_t i;
2859 	int32_t rc;
2860 	struct rte_eth_rss_conf rss_conf;
2861 
2862 	/*
2863 	 * Builds up the correct configuration for dcb+vt based on the vlan tags array
2864 	 * given above, and the number of traffic classes available for use.
2865 	 */
2866 	if (dcb_mode == DCB_VT_ENABLED) {
2867 		struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
2868 				&eth_conf->rx_adv_conf.vmdq_dcb_conf;
2869 		struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
2870 				&eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
2871 
2872 		/* VMDQ+DCB RX and TX configurations */
2873 		vmdq_rx_conf->enable_default_pool = 0;
2874 		vmdq_rx_conf->default_pool = 0;
2875 		vmdq_rx_conf->nb_queue_pools =
2876 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2877 		vmdq_tx_conf->nb_queue_pools =
2878 			(num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2879 
2880 		vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
2881 		for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
2882 			vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
2883 			vmdq_rx_conf->pool_map[i].pools =
2884 				1 << (i % vmdq_rx_conf->nb_queue_pools);
2885 		}
2886 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2887 			vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
2888 			vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
2889 		}
2890 
2891 		/* set DCB mode of RX and TX of multiple queues */
2892 		eth_conf->rxmode.mq_mode = ETH_MQ_RX_VMDQ_DCB;
2893 		eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
2894 	} else {
2895 		struct rte_eth_dcb_rx_conf *rx_conf =
2896 				&eth_conf->rx_adv_conf.dcb_rx_conf;
2897 		struct rte_eth_dcb_tx_conf *tx_conf =
2898 				&eth_conf->tx_adv_conf.dcb_tx_conf;
2899 
2900 		rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
2901 		if (rc != 0)
2902 			return rc;
2903 
2904 		rx_conf->nb_tcs = num_tcs;
2905 		tx_conf->nb_tcs = num_tcs;
2906 
2907 		for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2908 			rx_conf->dcb_tc[i] = i % num_tcs;
2909 			tx_conf->dcb_tc[i] = i % num_tcs;
2910 		}
2911 
2912 		eth_conf->rxmode.mq_mode = ETH_MQ_RX_DCB_RSS;
2913 		eth_conf->rx_adv_conf.rss_conf = rss_conf;
2914 		eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
2915 	}
2916 
2917 	if (pfc_en)
2918 		eth_conf->dcb_capability_en =
2919 				ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
2920 	else
2921 		eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
2922 
2923 	return 0;
2924 }
2925 
2926 int
2927 init_port_dcb_config(portid_t pid,
2928 		     enum dcb_mode_enable dcb_mode,
2929 		     enum rte_eth_nb_tcs num_tcs,
2930 		     uint8_t pfc_en)
2931 {
2932 	struct rte_eth_conf port_conf;
2933 	struct rte_port *rte_port;
2934 	int retval;
2935 	uint16_t i;
2936 
2937 	rte_port = &ports[pid];
2938 
2939 	memset(&port_conf, 0, sizeof(struct rte_eth_conf));
2940 	/* Enter DCB configuration status */
2941 	dcb_config = 1;
2942 
2943 	port_conf.rxmode = rte_port->dev_conf.rxmode;
2944 	port_conf.txmode = rte_port->dev_conf.txmode;
2945 
2946 	/*set configuration of DCB in vt mode and DCB in non-vt mode*/
2947 	retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
2948 	if (retval < 0)
2949 		return retval;
2950 	port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
2951 
2952 	/* re-configure the device . */
2953 	rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
2954 
2955 	rte_eth_dev_info_get(pid, &rte_port->dev_info);
2956 
2957 	/* If dev_info.vmdq_pool_base is greater than 0,
2958 	 * the queue id of vmdq pools is started after pf queues.
2959 	 */
2960 	if (dcb_mode == DCB_VT_ENABLED &&
2961 	    rte_port->dev_info.vmdq_pool_base > 0) {
2962 		printf("VMDQ_DCB multi-queue mode is nonsensical"
2963 			" for port %d.", pid);
2964 		return -1;
2965 	}
2966 
2967 	/* Assume the ports in testpmd have the same dcb capability
2968 	 * and has the same number of rxq and txq in dcb mode
2969 	 */
2970 	if (dcb_mode == DCB_VT_ENABLED) {
2971 		if (rte_port->dev_info.max_vfs > 0) {
2972 			nb_rxq = rte_port->dev_info.nb_rx_queues;
2973 			nb_txq = rte_port->dev_info.nb_tx_queues;
2974 		} else {
2975 			nb_rxq = rte_port->dev_info.max_rx_queues;
2976 			nb_txq = rte_port->dev_info.max_tx_queues;
2977 		}
2978 	} else {
2979 		/*if vt is disabled, use all pf queues */
2980 		if (rte_port->dev_info.vmdq_pool_base == 0) {
2981 			nb_rxq = rte_port->dev_info.max_rx_queues;
2982 			nb_txq = rte_port->dev_info.max_tx_queues;
2983 		} else {
2984 			nb_rxq = (queueid_t)num_tcs;
2985 			nb_txq = (queueid_t)num_tcs;
2986 
2987 		}
2988 	}
2989 	rx_free_thresh = 64;
2990 
2991 	memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
2992 
2993 	rxtx_port_config(rte_port);
2994 	/* VLAN filter */
2995 	rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
2996 	for (i = 0; i < RTE_DIM(vlan_tags); i++)
2997 		rx_vft_set(pid, vlan_tags[i], 1);
2998 
2999 	rte_eth_macaddr_get(pid, &rte_port->eth_addr);
3000 	map_port_queue_stats_mapping_registers(pid, rte_port);
3001 
3002 	rte_port->dcb_flag = 1;
3003 
3004 	return 0;
3005 }
3006 
3007 static void
3008 init_port(void)
3009 {
3010 	/* Configuration of Ethernet ports. */
3011 	ports = rte_zmalloc("testpmd: ports",
3012 			    sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3013 			    RTE_CACHE_LINE_SIZE);
3014 	if (ports == NULL) {
3015 		rte_exit(EXIT_FAILURE,
3016 				"rte_zmalloc(%d struct rte_port) failed\n",
3017 				RTE_MAX_ETHPORTS);
3018 	}
3019 
3020 	/* Initialize ports NUMA structures */
3021 	memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3022 	memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3023 	memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3024 }
3025 
3026 static void
3027 force_quit(void)
3028 {
3029 	pmd_test_exit();
3030 	prompt_exit();
3031 }
3032 
3033 static void
3034 print_stats(void)
3035 {
3036 	uint8_t i;
3037 	const char clr[] = { 27, '[', '2', 'J', '\0' };
3038 	const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3039 
3040 	/* Clear screen and move to top left */
3041 	printf("%s%s", clr, top_left);
3042 
3043 	printf("\nPort statistics ====================================");
3044 	for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3045 		nic_stats_display(fwd_ports_ids[i]);
3046 }
3047 
3048 static void
3049 signal_handler(int signum)
3050 {
3051 	if (signum == SIGINT || signum == SIGTERM) {
3052 		printf("\nSignal %d received, preparing to exit...\n",
3053 				signum);
3054 #ifdef RTE_LIBRTE_PDUMP
3055 		/* uninitialize packet capture framework */
3056 		rte_pdump_uninit();
3057 #endif
3058 #ifdef RTE_LIBRTE_LATENCY_STATS
3059 		rte_latencystats_uninit();
3060 #endif
3061 		force_quit();
3062 		/* Set flag to indicate the force termination. */
3063 		f_quit = 1;
3064 		/* exit with the expected status */
3065 		signal(signum, SIG_DFL);
3066 		kill(getpid(), signum);
3067 	}
3068 }
3069 
3070 int
3071 main(int argc, char** argv)
3072 {
3073 	int diag;
3074 	portid_t port_id;
3075 	uint16_t count;
3076 	int ret;
3077 
3078 	signal(SIGINT, signal_handler);
3079 	signal(SIGTERM, signal_handler);
3080 
3081 	diag = rte_eal_init(argc, argv);
3082 	if (diag < 0)
3083 		rte_panic("Cannot init EAL\n");
3084 
3085 	testpmd_logtype = rte_log_register("testpmd");
3086 	if (testpmd_logtype < 0)
3087 		rte_panic("Cannot register log type");
3088 	rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3089 
3090 	ret = register_eth_event_callback();
3091 	if (ret != 0)
3092 		rte_panic("Cannot register for ethdev events");
3093 
3094 #ifdef RTE_LIBRTE_PDUMP
3095 	/* initialize packet capture framework */
3096 	rte_pdump_init();
3097 #endif
3098 
3099 	count = 0;
3100 	RTE_ETH_FOREACH_DEV(port_id) {
3101 		ports_ids[count] = port_id;
3102 		count++;
3103 	}
3104 	nb_ports = (portid_t) count;
3105 	if (nb_ports == 0)
3106 		TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3107 
3108 	/* allocate port structures, and init them */
3109 	init_port();
3110 
3111 	set_def_fwd_config();
3112 	if (nb_lcores == 0)
3113 		rte_panic("Empty set of forwarding logical cores - check the "
3114 			  "core mask supplied in the command parameters\n");
3115 
3116 	/* Bitrate/latency stats disabled by default */
3117 #ifdef RTE_LIBRTE_BITRATE
3118 	bitrate_enabled = 0;
3119 #endif
3120 #ifdef RTE_LIBRTE_LATENCY_STATS
3121 	latencystats_enabled = 0;
3122 #endif
3123 
3124 	/* on FreeBSD, mlockall() is disabled by default */
3125 #ifdef RTE_EXEC_ENV_BSDAPP
3126 	do_mlockall = 0;
3127 #else
3128 	do_mlockall = 1;
3129 #endif
3130 
3131 	argc -= diag;
3132 	argv += diag;
3133 	if (argc > 1)
3134 		launch_args_parse(argc, argv);
3135 
3136 	if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3137 		TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3138 			strerror(errno));
3139 	}
3140 
3141 	if (tx_first && interactive)
3142 		rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3143 				"interactive mode.\n");
3144 
3145 	if (tx_first && lsc_interrupt) {
3146 		printf("Warning: lsc_interrupt needs to be off when "
3147 				" using tx_first. Disabling.\n");
3148 		lsc_interrupt = 0;
3149 	}
3150 
3151 	if (!nb_rxq && !nb_txq)
3152 		printf("Warning: Either rx or tx queues should be non-zero\n");
3153 
3154 	if (nb_rxq > 1 && nb_rxq > nb_txq)
3155 		printf("Warning: nb_rxq=%d enables RSS configuration, "
3156 		       "but nb_txq=%d will prevent to fully test it.\n",
3157 		       nb_rxq, nb_txq);
3158 
3159 	init_config();
3160 
3161 	if (hot_plug) {
3162 		ret = rte_dev_hotplug_handle_enable();
3163 		if (ret) {
3164 			RTE_LOG(ERR, EAL,
3165 				"fail to enable hotplug handling.");
3166 			return -1;
3167 		}
3168 
3169 		ret = rte_dev_event_monitor_start();
3170 		if (ret) {
3171 			RTE_LOG(ERR, EAL,
3172 				"fail to start device event monitoring.");
3173 			return -1;
3174 		}
3175 
3176 		ret = rte_dev_event_callback_register(NULL,
3177 			dev_event_callback, NULL);
3178 		if (ret) {
3179 			RTE_LOG(ERR, EAL,
3180 				"fail  to register device event callback\n");
3181 			return -1;
3182 		}
3183 	}
3184 
3185 	if (start_port(RTE_PORT_ALL) != 0)
3186 		rte_exit(EXIT_FAILURE, "Start ports failed\n");
3187 
3188 	/* set all ports to promiscuous mode by default */
3189 	RTE_ETH_FOREACH_DEV(port_id)
3190 		rte_eth_promiscuous_enable(port_id);
3191 
3192 	/* Init metrics library */
3193 	rte_metrics_init(rte_socket_id());
3194 
3195 #ifdef RTE_LIBRTE_LATENCY_STATS
3196 	if (latencystats_enabled != 0) {
3197 		int ret = rte_latencystats_init(1, NULL);
3198 		if (ret)
3199 			printf("Warning: latencystats init()"
3200 				" returned error %d\n",	ret);
3201 		printf("Latencystats running on lcore %d\n",
3202 			latencystats_lcore_id);
3203 	}
3204 #endif
3205 
3206 	/* Setup bitrate stats */
3207 #ifdef RTE_LIBRTE_BITRATE
3208 	if (bitrate_enabled != 0) {
3209 		bitrate_data = rte_stats_bitrate_create();
3210 		if (bitrate_data == NULL)
3211 			rte_exit(EXIT_FAILURE,
3212 				"Could not allocate bitrate data.\n");
3213 		rte_stats_bitrate_reg(bitrate_data);
3214 	}
3215 #endif
3216 
3217 #ifdef RTE_LIBRTE_CMDLINE
3218 	if (strlen(cmdline_filename) != 0)
3219 		cmdline_read_from_file(cmdline_filename);
3220 
3221 	if (interactive == 1) {
3222 		if (auto_start) {
3223 			printf("Start automatic packet forwarding\n");
3224 			start_packet_forwarding(0);
3225 		}
3226 		prompt();
3227 		pmd_test_exit();
3228 	} else
3229 #endif
3230 	{
3231 		char c;
3232 		int rc;
3233 
3234 		f_quit = 0;
3235 
3236 		printf("No commandline core given, start packet forwarding\n");
3237 		start_packet_forwarding(tx_first);
3238 		if (stats_period != 0) {
3239 			uint64_t prev_time = 0, cur_time, diff_time = 0;
3240 			uint64_t timer_period;
3241 
3242 			/* Convert to number of cycles */
3243 			timer_period = stats_period * rte_get_timer_hz();
3244 
3245 			while (f_quit == 0) {
3246 				cur_time = rte_get_timer_cycles();
3247 				diff_time += cur_time - prev_time;
3248 
3249 				if (diff_time >= timer_period) {
3250 					print_stats();
3251 					/* Reset the timer */
3252 					diff_time = 0;
3253 				}
3254 				/* Sleep to avoid unnecessary checks */
3255 				prev_time = cur_time;
3256 				sleep(1);
3257 			}
3258 		}
3259 
3260 		printf("Press enter to exit\n");
3261 		rc = read(0, &c, 1);
3262 		pmd_test_exit();
3263 		if (rc < 0)
3264 			return 1;
3265 	}
3266 
3267 	return 0;
3268 }
3269