app/test-pmd/testpmd.c

   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         .select_tos_ttl = 0,
 473         .vni = "\x00\x00\x00",
 474         .udp_src = 0,
 475         .udp_dst = RTE_BE16(4789),
 476         .ipv4_src = IPv4(127, 0, 0, 1),
 477         .ipv4_dst = IPv4(255, 255, 255, 255),
 478         .ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
 479                 "\x00\x00\x00\x00\x00\x00\x00\x01",
 480         .ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
 481                 "\x00\x00\x00\x00\x00\x00\x11\x11",
 482         .vlan_tci = 0,
 483         .ip_tos = 0,
 484         .ip_ttl = 255,
 485         .eth_src = "\x00\x00\x00\x00\x00\x00",
 486         .eth_dst = "\xff\xff\xff\xff\xff\xff",
 487 };
 488
 489 struct nvgre_encap_conf nvgre_encap_conf = {
 490         .select_ipv4 = 1,
 491         .select_vlan = 0,
 492         .tni = "\x00\x00\x00",
 493         .ipv4_src = IPv4(127, 0, 0, 1),
 494         .ipv4_dst = IPv4(255, 255, 255, 255),
 495         .ipv6_src = "\x00\x00\x00\x00\x00\x00\x00\x00"
 496                 "\x00\x00\x00\x00\x00\x00\x00\x01",
 497         .ipv6_dst = "\x00\x00\x00\x00\x00\x00\x00\x00"
 498                 "\x00\x00\x00\x00\x00\x00\x11\x11",
 499         .vlan_tci = 0,
 500         .eth_src = "\x00\x00\x00\x00\x00\x00",
 501         .eth_dst = "\xff\xff\xff\xff\xff\xff",
 502 };
 503
 504 /* Forward function declarations */
 505 static void setup_attached_port(portid_t pi);
 506 static void map_port_queue_stats_mapping_registers(portid_t pi,
 507                                                    struct rte_port *port);
 508 static void check_all_ports_link_status(uint32_t port_mask);
 509 static int eth_event_callback(portid_t port_id,
 510                               enum rte_eth_event_type type,
 511                               void *param, void *ret_param);
 512 static void dev_event_callback(const char *device_name,
 513                                 enum rte_dev_event_type type,
 514                                 void *param);
 515
 516 /*
 517  * Check if all the ports are started.
 518  * If yes, return positive value. If not, return zero.
 519  */
 520 static int all_ports_started(void);
 521
 522 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
 523 uint16_t gso_max_segment_size = ETHER_MAX_LEN - ETHER_CRC_LEN;
 524
 525 /*
 526  * Helper function to check if socket is already discovered.
 527  * If yes, return positive value. If not, return zero.
 528  */
 529 int
 530 new_socket_id(unsigned int socket_id)
 531 {
 532         unsigned int i;
 533
 534         for (i = 0; i < num_sockets; i++) {
 535                 if (socket_ids[i] == socket_id)
 536                         return 0;
 537         }
 538         return 1;
 539 }
 540
 541 /*
 542  * Setup default configuration.
 543  */
 544 static void
 545 set_default_fwd_lcores_config(void)
 546 {
 547         unsigned int i;
 548         unsigned int nb_lc;
 549         unsigned int sock_num;
 550
 551         nb_lc = 0;
 552         for (i = 0; i < RTE_MAX_LCORE; i++) {
 553                 if (!rte_lcore_is_enabled(i))
 554                         continue;
 555                 sock_num = rte_lcore_to_socket_id(i);
 556                 if (new_socket_id(sock_num)) {
 557                         if (num_sockets >= RTE_MAX_NUMA_NODES) {
 558                                 rte_exit(EXIT_FAILURE,
 559                                          "Total sockets greater than %u\n",
 560                                          RTE_MAX_NUMA_NODES);
 561                         }
 562                         socket_ids[num_sockets++] = sock_num;
 563                 }
 564                 if (i == rte_get_master_lcore())
 565                         continue;
 566                 fwd_lcores_cpuids[nb_lc++] = i;
 567         }
 568         nb_lcores = (lcoreid_t) nb_lc;
 569         nb_cfg_lcores = nb_lcores;
 570         nb_fwd_lcores = 1;
 571 }
 572
 573 static void
 574 set_def_peer_eth_addrs(void)
 575 {
 576         portid_t i;
 577
 578         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 579                 peer_eth_addrs[i].addr_bytes[0] = ETHER_LOCAL_ADMIN_ADDR;
 580                 peer_eth_addrs[i].addr_bytes[5] = i;
 581         }
 582 }
 583
 584 static void
 585 set_default_fwd_ports_config(void)
 586 {
 587         portid_t pt_id;
 588         int i = 0;
 589
 590         RTE_ETH_FOREACH_DEV(pt_id) {
 591                 fwd_ports_ids[i++] = pt_id;
 592
 593                 /* Update sockets info according to the attached device */
 594                 int socket_id = rte_eth_dev_socket_id(pt_id);
 595                 if (socket_id >= 0 && new_socket_id(socket_id)) {
 596                         if (num_sockets >= RTE_MAX_NUMA_NODES) {
 597                                 rte_exit(EXIT_FAILURE,
 598                                          "Total sockets greater than %u\n",
 599                                          RTE_MAX_NUMA_NODES);
 600                         }
 601                         socket_ids[num_sockets++] = socket_id;
 602                 }
 603         }
 604
 605         nb_cfg_ports = nb_ports;
 606         nb_fwd_ports = nb_ports;
 607 }
 608
 609 void
 610 set_def_fwd_config(void)
 611 {
 612         set_default_fwd_lcores_config();
 613         set_def_peer_eth_addrs();
 614         set_default_fwd_ports_config();
 615 }
 616
 617 /* extremely pessimistic estimation of memory required to create a mempool */
 618 static int
 619 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
 620 {
 621         unsigned int n_pages, mbuf_per_pg, leftover;
 622         uint64_t total_mem, mbuf_mem, obj_sz;
 623
 624         /* there is no good way to predict how much space the mempool will
 625          * occupy because it will allocate chunks on the fly, and some of those
 626          * will come from default DPDK memory while some will come from our
 627          * external memory, so just assume 128MB will be enough for everyone.
 628          */
 629         uint64_t hdr_mem = 128 << 20;
 630
 631         /* account for possible non-contiguousness */
 632         obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
 633         if (obj_sz > pgsz) {
 634                 TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
 635                 return -1;
 636         }
 637
 638         mbuf_per_pg = pgsz / obj_sz;
 639         leftover = (nb_mbufs % mbuf_per_pg) > 0;
 640         n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
 641
 642         mbuf_mem = n_pages * pgsz;
 643
 644         total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
 645
 646         if (total_mem > SIZE_MAX) {
 647                 TESTPMD_LOG(ERR, "Memory size too big\n");
 648                 return -1;
 649         }
 650         *out = (size_t)total_mem;
 651
 652         return 0;
 653 }
 654
 655 static int
 656 pagesz_flags(uint64_t page_sz)
 657 {
 658         /* as per mmap() manpage, all page sizes are log2 of page size
 659          * shifted by MAP_HUGE_SHIFT
 660          */
 661         int log2 = rte_log2_u64(page_sz);
 662
 663         return (log2 << HUGE_SHIFT);
 664 }
 665
 666 static void *
 667 alloc_mem(size_t memsz, size_t pgsz, bool huge)
 668 {
 669         void *addr;
 670         int flags;
 671
 672         /* allocate anonymous hugepages */
 673         flags = MAP_ANONYMOUS | MAP_PRIVATE;
 674         if (huge)
 675                 flags |= HUGE_FLAG | pagesz_flags(pgsz);
 676
 677         addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
 678         if (addr == MAP_FAILED)
 679                 return NULL;
 680
 681         return addr;
 682 }
 683
 684 struct extmem_param {
 685         void *addr;
 686         size_t len;
 687         size_t pgsz;
 688         rte_iova_t *iova_table;
 689         unsigned int iova_table_len;
 690 };
 691
 692 static int
 693 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
 694                 bool huge)
 695 {
 696         uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
 697                         RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
 698         unsigned int cur_page, n_pages, pgsz_idx;
 699         size_t mem_sz, cur_pgsz;
 700         rte_iova_t *iovas = NULL;
 701         void *addr;
 702         int ret;
 703
 704         for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
 705                 /* skip anything that is too big */
 706                 if (pgsizes[pgsz_idx] > SIZE_MAX)
 707                         continue;
 708
 709                 cur_pgsz = pgsizes[pgsz_idx];
 710
 711                 /* if we were told not to allocate hugepages, override */
 712                 if (!huge)
 713                         cur_pgsz = sysconf(_SC_PAGESIZE);
 714
 715                 ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
 716                 if (ret < 0) {
 717                         TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
 718                         return -1;
 719                 }
 720
 721                 /* allocate our memory */
 722                 addr = alloc_mem(mem_sz, cur_pgsz, huge);
 723
 724                 /* if we couldn't allocate memory with a specified page size,
 725                  * that doesn't mean we can't do it with other page sizes, so
 726                  * try another one.
 727                  */
 728                 if (addr == NULL)
 729                         continue;
 730
 731                 /* store IOVA addresses for every page in this memory area */
 732                 n_pages = mem_sz / cur_pgsz;
 733
 734                 iovas = malloc(sizeof(*iovas) * n_pages);
 735
 736                 if (iovas == NULL) {
 737                         TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
 738                         goto fail;
 739                 }
 740                 /* lock memory if it's not huge pages */
 741                 if (!huge)
 742                         mlock(addr, mem_sz);
 743
 744                 /* populate IOVA addresses */
 745                 for (cur_page = 0; cur_page < n_pages; cur_page++) {
 746                         rte_iova_t iova;
 747                         size_t offset;
 748                         void *cur;
 749
 750                         offset = cur_pgsz * cur_page;
 751                         cur = RTE_PTR_ADD(addr, offset);
 752
 753                         /* touch the page before getting its IOVA */
 754                         *(volatile char *)cur = 0;
 755
 756                         iova = rte_mem_virt2iova(cur);
 757
 758                         iovas[cur_page] = iova;
 759                 }
 760
 761                 break;
 762         }
 763         /* if we couldn't allocate anything */
 764         if (iovas == NULL)
 765                 return -1;
 766
 767         param->addr = addr;
 768         param->len = mem_sz;
 769         param->pgsz = cur_pgsz;
 770         param->iova_table = iovas;
 771         param->iova_table_len = n_pages;
 772
 773         return 0;
 774 fail:
 775         if (iovas)
 776                 free(iovas);
 777         if (addr)
 778                 munmap(addr, mem_sz);
 779
 780         return -1;
 781 }
 782
 783 static int
 784 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
 785 {
 786         struct extmem_param param;
 787         int socket_id, ret;
 788
 789         memset(&param, 0, sizeof(param));
 790
 791         /* check if our heap exists */
 792         socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
 793         if (socket_id < 0) {
 794                 /* create our heap */
 795                 ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
 796                 if (ret < 0) {
 797                         TESTPMD_LOG(ERR, "Cannot create heap\n");
 798                         return -1;
 799                 }
 800         }
 801
 802         ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
 803         if (ret < 0) {
 804                 TESTPMD_LOG(ERR, "Cannot create memory area\n");
 805                 return -1;
 806         }
 807
 808         /* we now have a valid memory area, so add it to heap */
 809         ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
 810                         param.addr, param.len, param.iova_table,
 811                         param.iova_table_len, param.pgsz);
 812
 813         /* when using VFIO, memory is automatically mapped for DMA by EAL */
 814
 815         /* not needed any more */
 816         free(param.iova_table);
 817
 818         if (ret < 0) {
 819                 TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
 820                 munmap(param.addr, param.len);
 821                 return -1;
 822         }
 823
 824         /* success */
 825
 826         TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
 827                         param.len >> 20);
 828
 829         return 0;
 830 }
 831
 832 /*
 833  * Configuration initialisation done once at init time.
 834  */
 835 static void
 836 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
 837                  unsigned int socket_id)
 838 {
 839         char pool_name[RTE_MEMPOOL_NAMESIZE];
 840         struct rte_mempool *rte_mp = NULL;
 841         uint32_t mb_size;
 842
 843         mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
 844         mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name));
 845
 846         TESTPMD_LOG(INFO,
 847                 "create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
 848                 pool_name, nb_mbuf, mbuf_seg_size, socket_id);
 849
 850         switch (mp_alloc_type) {
 851         case MP_ALLOC_NATIVE:
 852                 {
 853                         /* wrapper to rte_mempool_create() */
 854                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
 855                                         rte_mbuf_best_mempool_ops());
 856                         rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
 857                                 mb_mempool_cache, 0, mbuf_seg_size, socket_id);
 858                         break;
 859                 }
 860         case MP_ALLOC_ANON:
 861                 {
 862                         rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
 863                                 mb_size, (unsigned int) mb_mempool_cache,
 864                                 sizeof(struct rte_pktmbuf_pool_private),
 865                                 socket_id, 0);
 866                         if (rte_mp == NULL)
 867                                 goto err;
 868
 869                         if (rte_mempool_populate_anon(rte_mp) == 0) {
 870                                 rte_mempool_free(rte_mp);
 871                                 rte_mp = NULL;
 872                                 goto err;
 873                         }
 874                         rte_pktmbuf_pool_init(rte_mp, NULL);
 875                         rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
 876                         break;
 877                 }
 878         case MP_ALLOC_XMEM:
 879         case MP_ALLOC_XMEM_HUGE:
 880                 {
 881                         int heap_socket;
 882                         bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
 883
 884                         if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
 885                                 rte_exit(EXIT_FAILURE, "Could not create external memory\n");
 886
 887                         heap_socket =
 888                                 rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
 889                         if (heap_socket < 0)
 890                                 rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
 891
 892                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
 893                                         rte_mbuf_best_mempool_ops());
 894                         rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
 895                                         mb_mempool_cache, 0, mbuf_seg_size,
 896                                         heap_socket);
 897                         break;
 898                 }
 899         default:
 900                 {
 901                         rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
 902                 }
 903         }
 904
 905 err:
 906         if (rte_mp == NULL) {
 907                 rte_exit(EXIT_FAILURE,
 908                         "Creation of mbuf pool for socket %u failed: %s\n",
 909                         socket_id, rte_strerror(rte_errno));
 910         } else if (verbose_level > 0) {
 911                 rte_mempool_dump(stdout, rte_mp);
 912         }
 913 }
 914
 915 /*
 916  * Check given socket id is valid or not with NUMA mode,
 917  * if valid, return 0, else return -1
 918  */
 919 static int
 920 check_socket_id(const unsigned int socket_id)
 921 {
 922         static int warning_once = 0;
 923
 924         if (new_socket_id(socket_id)) {
 925                 if (!warning_once && numa_support)
 926                         printf("Warning: NUMA should be configured manually by"
 927                                " using --port-numa-config and"
 928                                " --ring-numa-config parameters along with"
 929                                " --numa.\n");
 930                 warning_once = 1;
 931                 return -1;
 932         }
 933         return 0;
 934 }
 935
 936 /*
 937  * Get the allowed maximum number of RX queues.
 938  * *pid return the port id which has minimal value of
 939  * max_rx_queues in all ports.
 940  */
 941 queueid_t
 942 get_allowed_max_nb_rxq(portid_t *pid)
 943 {
 944         queueid_t allowed_max_rxq = MAX_QUEUE_ID;
 945         portid_t pi;
 946         struct rte_eth_dev_info dev_info;
 947
 948         RTE_ETH_FOREACH_DEV(pi) {
 949                 rte_eth_dev_info_get(pi, &dev_info);
 950                 if (dev_info.max_rx_queues < allowed_max_rxq) {
 951                         allowed_max_rxq = dev_info.max_rx_queues;
 952                         *pid = pi;
 953                 }
 954         }
 955         return allowed_max_rxq;
 956 }
 957
 958 /*
 959  * Check input rxq is valid or not.
 960  * If input rxq is not greater than any of maximum number
 961  * of RX queues of all ports, it is valid.
 962  * if valid, return 0, else return -1
 963  */
 964 int
 965 check_nb_rxq(queueid_t rxq)
 966 {
 967         queueid_t allowed_max_rxq;
 968         portid_t pid = 0;
 969
 970         allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
 971         if (rxq > allowed_max_rxq) {
 972                 printf("Fail: input rxq (%u) can't be greater "
 973                        "than max_rx_queues (%u) of port %u\n",
 974                        rxq,
 975                        allowed_max_rxq,
 976                        pid);
 977                 return -1;
 978         }
 979         return 0;
 980 }
 981
 982 /*
 983  * Get the allowed maximum number of TX queues.
 984  * *pid return the port id which has minimal value of
 985  * max_tx_queues in all ports.
 986  */
 987 queueid_t
 988 get_allowed_max_nb_txq(portid_t *pid)
 989 {
 990         queueid_t allowed_max_txq = MAX_QUEUE_ID;
 991         portid_t pi;
 992         struct rte_eth_dev_info dev_info;
 993
 994         RTE_ETH_FOREACH_DEV(pi) {
 995                 rte_eth_dev_info_get(pi, &dev_info);
 996                 if (dev_info.max_tx_queues < allowed_max_txq) {
 997                         allowed_max_txq = dev_info.max_tx_queues;
 998                         *pid = pi;
 999                 }
1000         }
1001         return allowed_max_txq;
1002 }
1003
1004 /*
1005  * Check input txq is valid or not.
1006  * If input txq is not greater than any of maximum number
1007  * of TX queues of all ports, it is valid.
1008  * if valid, return 0, else return -1
1009  */
1010 int
1011 check_nb_txq(queueid_t txq)
1012 {
1013         queueid_t allowed_max_txq;
1014         portid_t pid = 0;
1015
1016         allowed_max_txq = get_allowed_max_nb_txq(&pid);
1017         if (txq > allowed_max_txq) {
1018                 printf("Fail: input txq (%u) can't be greater "
1019                        "than max_tx_queues (%u) of port %u\n",
1020                        txq,
1021                        allowed_max_txq,
1022                        pid);
1023                 return -1;
1024         }
1025         return 0;
1026 }
1027
1028 static void
1029 init_config(void)
1030 {
1031         portid_t pid;
1032         struct rte_port *port;
1033         struct rte_mempool *mbp;
1034         unsigned int nb_mbuf_per_pool;
1035         lcoreid_t  lc_id;
1036         uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
1037         struct rte_gro_param gro_param;
1038         uint32_t gso_types;
1039         int k;
1040
1041         memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
1042
1043         /* Configuration of logical cores. */
1044         fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1045                                 sizeof(struct fwd_lcore *) * nb_lcores,
1046                                 RTE_CACHE_LINE_SIZE);
1047         if (fwd_lcores == NULL) {
1048                 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1049                                                         "failed\n", nb_lcores);
1050         }
1051         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1052                 fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1053                                                sizeof(struct fwd_lcore),
1054                                                RTE_CACHE_LINE_SIZE);
1055                 if (fwd_lcores[lc_id] == NULL) {
1056                         rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1057                                                                 "failed\n");
1058                 }
1059                 fwd_lcores[lc_id]->cpuid_idx = lc_id;
1060         }
1061
1062         RTE_ETH_FOREACH_DEV(pid) {
1063                 port = &ports[pid];
1064                 /* Apply default TxRx configuration for all ports */
1065                 port->dev_conf.txmode = tx_mode;
1066                 port->dev_conf.rxmode = rx_mode;
1067                 rte_eth_dev_info_get(pid, &port->dev_info);
1068
1069                 if (!(port->dev_info.tx_offload_capa &
1070                       DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1071                         port->dev_conf.txmode.offloads &=
1072                                 ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1073                 if (!(port->dev_info.tx_offload_capa &
1074                         DEV_TX_OFFLOAD_MATCH_METADATA))
1075                         port->dev_conf.txmode.offloads &=
1076                                 ~DEV_TX_OFFLOAD_MATCH_METADATA;
1077                 if (numa_support) {
1078                         if (port_numa[pid] != NUMA_NO_CONFIG)
1079                                 port_per_socket[port_numa[pid]]++;
1080                         else {
1081                                 uint32_t socket_id = rte_eth_dev_socket_id(pid);
1082
1083                                 /*
1084                                  * if socket_id is invalid,
1085                                  * set to the first available socket.
1086                                  */
1087                                 if (check_socket_id(socket_id) < 0)
1088                                         socket_id = socket_ids[0];
1089                                 port_per_socket[socket_id]++;
1090                         }
1091                 }
1092
1093                 /* Apply Rx offloads configuration */
1094                 for (k = 0; k < port->dev_info.max_rx_queues; k++)
1095                         port->rx_conf[k].offloads =
1096                                 port->dev_conf.rxmode.offloads;
1097                 /* Apply Tx offloads configuration */
1098                 for (k = 0; k < port->dev_info.max_tx_queues; k++)
1099                         port->tx_conf[k].offloads =
1100                                 port->dev_conf.txmode.offloads;
1101
1102                 /* set flag to initialize port/queue */
1103                 port->need_reconfig = 1;
1104                 port->need_reconfig_queues = 1;
1105                 port->tx_metadata = 0;
1106         }
1107
1108         /*
1109          * Create pools of mbuf.
1110          * If NUMA support is disabled, create a single pool of mbuf in
1111          * socket 0 memory by default.
1112          * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1113          *
1114          * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1115          * nb_txd can be configured at run time.
1116          */
1117         if (param_total_num_mbufs)
1118                 nb_mbuf_per_pool = param_total_num_mbufs;
1119         else {
1120                 nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1121                         (nb_lcores * mb_mempool_cache) +
1122                         RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1123                 nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1124         }
1125
1126         if (numa_support) {
1127                 uint8_t i;
1128
1129                 for (i = 0; i < num_sockets; i++)
1130                         mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool,
1131                                          socket_ids[i]);
1132         } else {
1133                 if (socket_num == UMA_NO_CONFIG)
1134                         mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool, 0);
1135                 else
1136                         mbuf_pool_create(mbuf_data_size, nb_mbuf_per_pool,
1137                                                  socket_num);
1138         }
1139
1140         init_port_config();
1141
1142         gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1143                 DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1144         /*
1145          * Records which Mbuf pool to use by each logical core, if needed.
1146          */
1147         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1148                 mbp = mbuf_pool_find(
1149                         rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]));
1150
1151                 if (mbp == NULL)
1152                         mbp = mbuf_pool_find(0);
1153                 fwd_lcores[lc_id]->mbp = mbp;
1154                 /* initialize GSO context */
1155                 fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1156                 fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1157                 fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1158                 fwd_lcores[lc_id]->gso_ctx.gso_size = ETHER_MAX_LEN -
1159                         ETHER_CRC_LEN;
1160                 fwd_lcores[lc_id]->gso_ctx.flag = 0;
1161         }
1162
1163         /* Configuration of packet forwarding streams. */
1164         if (init_fwd_streams() < 0)
1165                 rte_exit(EXIT_FAILURE, "FAIL from init_fwd_streams()\n");
1166
1167         fwd_config_setup();
1168
1169         /* create a gro context for each lcore */
1170         gro_param.gro_types = RTE_GRO_TCP_IPV4;
1171         gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1172         gro_param.max_item_per_flow = MAX_PKT_BURST;
1173         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1174                 gro_param.socket_id = rte_lcore_to_socket_id(
1175                                 fwd_lcores_cpuids[lc_id]);
1176                 fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1177                 if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1178                         rte_exit(EXIT_FAILURE,
1179                                         "rte_gro_ctx_create() failed\n");
1180                 }
1181         }
1182
1183 #if defined RTE_LIBRTE_PMD_SOFTNIC
1184         if (strcmp(cur_fwd_eng->fwd_mode_name, "softnic") == 0) {
1185                 RTE_ETH_FOREACH_DEV(pid) {
1186                         port = &ports[pid];
1187                         const char *driver = port->dev_info.driver_name;
1188
1189                         if (strcmp(driver, "net_softnic") == 0)
1190                                 port->softport.fwd_lcore_arg = fwd_lcores;
1191                 }
1192         }
1193 #endif
1194
1195 }
1196
1197
1198 void
1199 reconfig(portid_t new_port_id, unsigned socket_id)
1200 {
1201         struct rte_port *port;
1202
1203         /* Reconfiguration of Ethernet ports. */
1204         port = &ports[new_port_id];
1205         rte_eth_dev_info_get(new_port_id, &port->dev_info);
1206
1207         /* set flag to initialize port/queue */
1208         port->need_reconfig = 1;
1209         port->need_reconfig_queues = 1;
1210         port->socket_id = socket_id;
1211
1212         init_port_config();
1213 }
1214
1215
1216 int
1217 init_fwd_streams(void)
1218 {
1219         portid_t pid;
1220         struct rte_port *port;
1221         streamid_t sm_id, nb_fwd_streams_new;
1222         queueid_t q;
1223
1224         /* set socket id according to numa or not */
1225         RTE_ETH_FOREACH_DEV(pid) {
1226                 port = &ports[pid];
1227                 if (nb_rxq > port->dev_info.max_rx_queues) {
1228                         printf("Fail: nb_rxq(%d) is greater than "
1229                                 "max_rx_queues(%d)\n", nb_rxq,
1230                                 port->dev_info.max_rx_queues);
1231                         return -1;
1232                 }
1233                 if (nb_txq > port->dev_info.max_tx_queues) {
1234                         printf("Fail: nb_txq(%d) is greater than "
1235                                 "max_tx_queues(%d)\n", nb_txq,
1236                                 port->dev_info.max_tx_queues);
1237                         return -1;
1238                 }
1239                 if (numa_support) {
1240                         if (port_numa[pid] != NUMA_NO_CONFIG)
1241                                 port->socket_id = port_numa[pid];
1242                         else {
1243                                 port->socket_id = rte_eth_dev_socket_id(pid);
1244
1245                                 /*
1246                                  * if socket_id is invalid,
1247                                  * set to the first available socket.
1248                                  */
1249                                 if (check_socket_id(port->socket_id) < 0)
1250                                         port->socket_id = socket_ids[0];
1251                         }
1252                 }
1253                 else {
1254                         if (socket_num == UMA_NO_CONFIG)
1255                                 port->socket_id = 0;
1256                         else
1257                                 port->socket_id = socket_num;
1258                 }
1259         }
1260
1261         q = RTE_MAX(nb_rxq, nb_txq);
1262         if (q == 0) {
1263                 printf("Fail: Cannot allocate fwd streams as number of queues is 0\n");
1264                 return -1;
1265         }
1266         nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1267         if (nb_fwd_streams_new == nb_fwd_streams)
1268                 return 0;
1269         /* clear the old */
1270         if (fwd_streams != NULL) {
1271                 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1272                         if (fwd_streams[sm_id] == NULL)
1273                                 continue;
1274                         rte_free(fwd_streams[sm_id]);
1275                         fwd_streams[sm_id] = NULL;
1276                 }
1277                 rte_free(fwd_streams);
1278                 fwd_streams = NULL;
1279         }
1280
1281         /* init new */
1282         nb_fwd_streams = nb_fwd_streams_new;
1283         if (nb_fwd_streams) {
1284                 fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1285                         sizeof(struct fwd_stream *) * nb_fwd_streams,
1286                         RTE_CACHE_LINE_SIZE);
1287                 if (fwd_streams == NULL)
1288                         rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1289                                  " (struct fwd_stream *)) failed\n",
1290                                  nb_fwd_streams);
1291
1292                 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1293                         fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1294                                 " struct fwd_stream", sizeof(struct fwd_stream),
1295                                 RTE_CACHE_LINE_SIZE);
1296                         if (fwd_streams[sm_id] == NULL)
1297                                 rte_exit(EXIT_FAILURE, "rte_zmalloc"
1298                                          "(struct fwd_stream) failed\n");
1299                 }
1300         }
1301
1302         return 0;
1303 }
1304
1305 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1306 static void
1307 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1308 {
1309         unsigned int total_burst;
1310         unsigned int nb_burst;
1311         unsigned int burst_stats[3];
1312         uint16_t pktnb_stats[3];
1313         uint16_t nb_pkt;
1314         int burst_percent[3];
1315
1316         /*
1317          * First compute the total number of packet bursts and the
1318          * two highest numbers of bursts of the same number of packets.
1319          */
1320         total_burst = 0;
1321         burst_stats[0] = burst_stats[1] = burst_stats[2] = 0;
1322         pktnb_stats[0] = pktnb_stats[1] = pktnb_stats[2] = 0;
1323         for (nb_pkt = 0; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1324                 nb_burst = pbs->pkt_burst_spread[nb_pkt];
1325                 if (nb_burst == 0)
1326                         continue;
1327                 total_burst += nb_burst;
1328                 if (nb_burst > burst_stats[0]) {
1329                         burst_stats[1] = burst_stats[0];
1330                         pktnb_stats[1] = pktnb_stats[0];
1331                         burst_stats[0] = nb_burst;
1332                         pktnb_stats[0] = nb_pkt;
1333                 } else if (nb_burst > burst_stats[1]) {
1334                         burst_stats[1] = nb_burst;
1335                         pktnb_stats[1] = nb_pkt;
1336                 }
1337         }
1338         if (total_burst == 0)
1339                 return;
1340         burst_percent[0] = (burst_stats[0] * 100) / total_burst;
1341         printf("  %s-bursts : %u [%d%% of %d pkts", rx_tx, total_burst,
1342                burst_percent[0], (int) pktnb_stats[0]);
1343         if (burst_stats[0] == total_burst) {
1344                 printf("]\n");
1345                 return;
1346         }
1347         if (burst_stats[0] + burst_stats[1] == total_burst) {
1348                 printf(" + %d%% of %d pkts]\n",
1349                        100 - burst_percent[0], pktnb_stats[1]);
1350                 return;
1351         }
1352         burst_percent[1] = (burst_stats[1] * 100) / total_burst;
1353         burst_percent[2] = 100 - (burst_percent[0] + burst_percent[1]);
1354         if ((burst_percent[1] == 0) || (burst_percent[2] == 0)) {
1355                 printf(" + %d%% of others]\n", 100 - burst_percent[0]);
1356                 return;
1357         }
1358         printf(" + %d%% of %d pkts + %d%% of others]\n",
1359                burst_percent[1], (int) pktnb_stats[1], burst_percent[2]);
1360 }
1361 #endif /* RTE_TEST_PMD_RECORD_BURST_STATS */
1362
1363 static void
1364 fwd_stream_stats_display(streamid_t stream_id)
1365 {
1366         struct fwd_stream *fs;
1367         static const char *fwd_top_stats_border = "-------";
1368
1369         fs = fwd_streams[stream_id];
1370         if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1371             (fs->fwd_dropped == 0))
1372                 return;
1373         printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1374                "TX Port=%2d/Queue=%2d %s\n",
1375                fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1376                fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1377         printf("  RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1378                " TX-dropped: %-14"PRIu64,
1379                fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1380
1381         /* if checksum mode */
1382         if (cur_fwd_eng == &csum_fwd_engine) {
1383                 printf("  RX- bad IP checksum: %-14"PRIu64
1384                        "  Rx- bad L4 checksum: %-14"PRIu64
1385                        " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1386                         fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1387                         fs->rx_bad_outer_l4_csum);
1388         } else {
1389                 printf("\n");
1390         }
1391
1392 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1393         pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1394         pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1395 #endif
1396 }
1397
1398 void
1399 fwd_stats_display(void)
1400 {
1401         static const char *fwd_stats_border = "----------------------";
1402         static const char *acc_stats_border = "+++++++++++++++";
1403         struct {
1404                 struct fwd_stream *rx_stream;
1405                 struct fwd_stream *tx_stream;
1406                 uint64_t tx_dropped;
1407                 uint64_t rx_bad_ip_csum;
1408                 uint64_t rx_bad_l4_csum;
1409                 uint64_t rx_bad_outer_l4_csum;
1410         } ports_stats[RTE_MAX_ETHPORTS];
1411         uint64_t total_rx_dropped = 0;
1412         uint64_t total_tx_dropped = 0;
1413         uint64_t total_rx_nombuf = 0;
1414         struct rte_eth_stats stats;
1415 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1416         uint64_t fwd_cycles = 0;
1417 #endif
1418         uint64_t total_recv = 0;
1419         uint64_t total_xmit = 0;
1420         struct rte_port *port;
1421         streamid_t sm_id;
1422         portid_t pt_id;
1423         int i;
1424
1425         memset(ports_stats, 0, sizeof(ports_stats));
1426
1427         for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1428                 struct fwd_stream *fs = fwd_streams[sm_id];
1429
1430                 if (cur_fwd_config.nb_fwd_streams >
1431                     cur_fwd_config.nb_fwd_ports) {
1432                         fwd_stream_stats_display(sm_id);
1433                 } else {
1434                         ports_stats[fs->tx_port].tx_stream = fs;
1435                         ports_stats[fs->rx_port].rx_stream = fs;
1436                 }
1437
1438                 ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1439
1440                 ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1441                 ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1442                 ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1443                                 fs->rx_bad_outer_l4_csum;
1444
1445 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1446                 fwd_cycles += fs->core_cycles;
1447 #endif
1448         }
1449         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1450                 uint8_t j;
1451
1452                 pt_id = fwd_ports_ids[i];
1453                 port = &ports[pt_id];
1454
1455                 rte_eth_stats_get(pt_id, &stats);
1456                 stats.ipackets -= port->stats.ipackets;
1457                 stats.opackets -= port->stats.opackets;
1458                 stats.ibytes -= port->stats.ibytes;
1459                 stats.obytes -= port->stats.obytes;
1460                 stats.imissed -= port->stats.imissed;
1461                 stats.oerrors -= port->stats.oerrors;
1462                 stats.rx_nombuf -= port->stats.rx_nombuf;
1463
1464                 total_recv += stats.ipackets;
1465                 total_xmit += stats.opackets;
1466                 total_rx_dropped += stats.imissed;
1467                 total_tx_dropped += ports_stats[pt_id].tx_dropped;
1468                 total_tx_dropped += stats.oerrors;
1469                 total_rx_nombuf  += stats.rx_nombuf;
1470
1471                 printf("\n  %s Forward statistics for port %-2d %s\n",
1472                        fwd_stats_border, pt_id, fwd_stats_border);
1473
1474                 if (!port->rx_queue_stats_mapping_enabled &&
1475                     !port->tx_queue_stats_mapping_enabled) {
1476                         printf("  RX-packets: %-14"PRIu64
1477                                " RX-dropped: %-14"PRIu64
1478                                "RX-total: %-"PRIu64"\n",
1479                                stats.ipackets, stats.imissed,
1480                                stats.ipackets + stats.imissed);
1481
1482                         if (cur_fwd_eng == &csum_fwd_engine)
1483                                 printf("  Bad-ipcsum: %-14"PRIu64
1484                                        " Bad-l4csum: %-14"PRIu64
1485                                        "Bad-outer-l4csum: %-14"PRIu64"\n",
1486                                        ports_stats[pt_id].rx_bad_ip_csum,
1487                                        ports_stats[pt_id].rx_bad_l4_csum,
1488                                        ports_stats[pt_id].rx_bad_outer_l4_csum);
1489                         if (stats.ierrors + stats.rx_nombuf > 0) {
1490                                 printf("  RX-error: %-"PRIu64"\n",
1491                                        stats.ierrors);
1492                                 printf("  RX-nombufs: %-14"PRIu64"\n",
1493                                        stats.rx_nombuf);
1494                         }
1495
1496                         printf("  TX-packets: %-14"PRIu64
1497                                " TX-dropped: %-14"PRIu64
1498                                "TX-total: %-"PRIu64"\n",
1499                                stats.opackets, ports_stats[pt_id].tx_dropped,
1500                                stats.opackets + ports_stats[pt_id].tx_dropped);
1501                 } else {
1502                         printf("  RX-packets:             %14"PRIu64
1503                                "    RX-dropped:%14"PRIu64
1504                                "    RX-total:%14"PRIu64"\n",
1505                                stats.ipackets, stats.imissed,
1506                                stats.ipackets + stats.imissed);
1507
1508                         if (cur_fwd_eng == &csum_fwd_engine)
1509                                 printf("  Bad-ipcsum:%14"PRIu64
1510                                        "    Bad-l4csum:%14"PRIu64
1511                                        "    Bad-outer-l4csum: %-14"PRIu64"\n",
1512                                        ports_stats[pt_id].rx_bad_ip_csum,
1513                                        ports_stats[pt_id].rx_bad_l4_csum,
1514                                        ports_stats[pt_id].rx_bad_outer_l4_csum);
1515                         if ((stats.ierrors + stats.rx_nombuf) > 0) {
1516                                 printf("  RX-error:%"PRIu64"\n", stats.ierrors);
1517                                 printf("  RX-nombufs:             %14"PRIu64"\n",
1518                                        stats.rx_nombuf);
1519                         }
1520
1521                         printf("  TX-packets:             %14"PRIu64
1522                                "    TX-dropped:%14"PRIu64
1523                                "    TX-total:%14"PRIu64"\n",
1524                                stats.opackets, ports_stats[pt_id].tx_dropped,
1525                                stats.opackets + ports_stats[pt_id].tx_dropped);
1526                 }
1527
1528 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1529                 if (ports_stats[pt_id].rx_stream)
1530                         pkt_burst_stats_display("RX",
1531                                 &ports_stats[pt_id].rx_stream->rx_burst_stats);
1532                 if (ports_stats[pt_id].tx_stream)
1533                         pkt_burst_stats_display("TX",
1534                                 &ports_stats[pt_id].tx_stream->tx_burst_stats);
1535 #endif
1536
1537                 if (port->rx_queue_stats_mapping_enabled) {
1538                         printf("\n");
1539                         for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1540                                 printf("  Stats reg %2d RX-packets:%14"PRIu64
1541                                        "     RX-errors:%14"PRIu64
1542                                        "    RX-bytes:%14"PRIu64"\n",
1543                                        j, stats.q_ipackets[j],
1544                                        stats.q_errors[j], stats.q_ibytes[j]);
1545                         }
1546                         printf("\n");
1547                 }
1548                 if (port->tx_queue_stats_mapping_enabled) {
1549                         for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1550                                 printf("  Stats reg %2d TX-packets:%14"PRIu64
1551                                        "                                 TX-bytes:%14"
1552                                        PRIu64"\n",
1553                                        j, stats.q_opackets[j],
1554                                        stats.q_obytes[j]);
1555                         }
1556                 }
1557
1558                 printf("  %s--------------------------------%s\n",
1559                        fwd_stats_border, fwd_stats_border);
1560         }
1561
1562         printf("\n  %s Accumulated forward statistics for all ports"
1563                "%s\n",
1564                acc_stats_border, acc_stats_border);
1565         printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1566                "%-"PRIu64"\n"
1567                "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1568                "%-"PRIu64"\n",
1569                total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1570                total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1571         if (total_rx_nombuf > 0)
1572                 printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1573         printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
1574                "%s\n",
1575                acc_stats_border, acc_stats_border);
1576 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1577         if (total_recv > 0)
1578                 printf("\n  CPU cycles/packet=%u (total cycles="
1579                        "%"PRIu64" / total RX packets=%"PRIu64")\n",
1580                        (unsigned int)(fwd_cycles / total_recv),
1581                        fwd_cycles, total_recv);
1582 #endif
1583 }
1584
1585 void
1586 fwd_stats_reset(void)
1587 {
1588         streamid_t sm_id;
1589         portid_t pt_id;
1590         int i;
1591
1592         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1593                 pt_id = fwd_ports_ids[i];
1594                 rte_eth_stats_get(pt_id, &ports[pt_id].stats);
1595         }
1596         for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1597                 struct fwd_stream *fs = fwd_streams[sm_id];
1598
1599                 fs->rx_packets = 0;
1600                 fs->tx_packets = 0;
1601                 fs->fwd_dropped = 0;
1602                 fs->rx_bad_ip_csum = 0;
1603                 fs->rx_bad_l4_csum = 0;
1604                 fs->rx_bad_outer_l4_csum = 0;
1605
1606 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1607                 memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
1608                 memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
1609 #endif
1610 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1611                 fs->core_cycles = 0;
1612 #endif
1613         }
1614 }
1615
1616 static void
1617 flush_fwd_rx_queues(void)
1618 {
1619         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1620         portid_t  rxp;
1621         portid_t port_id;
1622         queueid_t rxq;
1623         uint16_t  nb_rx;
1624         uint16_t  i;
1625         uint8_t   j;
1626         uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
1627         uint64_t timer_period;
1628
1629         /* convert to number of cycles */
1630         timer_period = rte_get_timer_hz(); /* 1 second timeout */
1631
1632         for (j = 0; j < 2; j++) {
1633                 for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
1634                         for (rxq = 0; rxq < nb_rxq; rxq++) {
1635                                 port_id = fwd_ports_ids[rxp];
1636                                 /**
1637                                 * testpmd can stuck in the below do while loop
1638                                 * if rte_eth_rx_burst() always returns nonzero
1639                                 * packets. So timer is added to exit this loop
1640                                 * after 1sec timer expiry.
1641                                 */
1642                                 prev_tsc = rte_rdtsc();
1643                                 do {
1644                                         nb_rx = rte_eth_rx_burst(port_id, rxq,
1645                                                 pkts_burst, MAX_PKT_BURST);
1646                                         for (i = 0; i < nb_rx; i++)
1647                                                 rte_pktmbuf_free(pkts_burst[i]);
1648
1649                                         cur_tsc = rte_rdtsc();
1650                                         diff_tsc = cur_tsc - prev_tsc;
1651                                         timer_tsc += diff_tsc;
1652                                 } while ((nb_rx > 0) &&
1653                                         (timer_tsc < timer_period));
1654                                 timer_tsc = 0;
1655                         }
1656                 }
1657                 rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
1658         }
1659 }
1660
1661 static void
1662 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
1663 {
1664         struct fwd_stream **fsm;
1665         streamid_t nb_fs;
1666         streamid_t sm_id;
1667 #ifdef RTE_LIBRTE_BITRATE
1668         uint64_t tics_per_1sec;
1669         uint64_t tics_datum;
1670         uint64_t tics_current;
1671         uint16_t i, cnt_ports;
1672
1673         cnt_ports = nb_ports;
1674         tics_datum = rte_rdtsc();
1675         tics_per_1sec = rte_get_timer_hz();
1676 #endif
1677         fsm = &fwd_streams[fc->stream_idx];
1678         nb_fs = fc->stream_nb;
1679         do {
1680                 for (sm_id = 0; sm_id < nb_fs; sm_id++)
1681                         (*pkt_fwd)(fsm[sm_id]);
1682 #ifdef RTE_LIBRTE_BITRATE
1683                 if (bitrate_enabled != 0 &&
1684                                 bitrate_lcore_id == rte_lcore_id()) {
1685                         tics_current = rte_rdtsc();
1686                         if (tics_current - tics_datum >= tics_per_1sec) {
1687                                 /* Periodic bitrate calculation */
1688                                 for (i = 0; i < cnt_ports; i++)
1689                                         rte_stats_bitrate_calc(bitrate_data,
1690                                                 ports_ids[i]);
1691                                 tics_datum = tics_current;
1692                         }
1693                 }
1694 #endif
1695 #ifdef RTE_LIBRTE_LATENCY_STATS
1696                 if (latencystats_enabled != 0 &&
1697                                 latencystats_lcore_id == rte_lcore_id())
1698                         rte_latencystats_update();
1699 #endif
1700
1701         } while (! fc->stopped);
1702 }
1703
1704 static int
1705 start_pkt_forward_on_core(void *fwd_arg)
1706 {
1707         run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
1708                              cur_fwd_config.fwd_eng->packet_fwd);
1709         return 0;
1710 }
1711
1712 /*
1713  * Run the TXONLY packet forwarding engine to send a single burst of packets.
1714  * Used to start communication flows in network loopback test configurations.
1715  */
1716 static int
1717 run_one_txonly_burst_on_core(void *fwd_arg)
1718 {
1719         struct fwd_lcore *fwd_lc;
1720         struct fwd_lcore tmp_lcore;
1721
1722         fwd_lc = (struct fwd_lcore *) fwd_arg;
1723         tmp_lcore = *fwd_lc;
1724         tmp_lcore.stopped = 1;
1725         run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
1726         return 0;
1727 }
1728
1729 /*
1730  * Launch packet forwarding:
1731  *     - Setup per-port forwarding context.
1732  *     - launch logical cores with their forwarding configuration.
1733  */
1734 static void
1735 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
1736 {
1737         port_fwd_begin_t port_fwd_begin;
1738         unsigned int i;
1739         unsigned int lc_id;
1740         int diag;
1741
1742         port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
1743         if (port_fwd_begin != NULL) {
1744                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1745                         (*port_fwd_begin)(fwd_ports_ids[i]);
1746         }
1747         for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
1748                 lc_id = fwd_lcores_cpuids[i];
1749                 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
1750                         fwd_lcores[i]->stopped = 0;
1751                         diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
1752                                                      fwd_lcores[i], lc_id);
1753                         if (diag != 0)
1754                                 printf("launch lcore %u failed - diag=%d\n",
1755                                        lc_id, diag);
1756                 }
1757         }
1758 }
1759
1760 /*
1761  * Launch packet forwarding configuration.
1762  */
1763 void
1764 start_packet_forwarding(int with_tx_first)
1765 {
1766         port_fwd_begin_t port_fwd_begin;
1767         port_fwd_end_t  port_fwd_end;
1768         struct rte_port *port;
1769         unsigned int i;
1770         portid_t   pt_id;
1771
1772         if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
1773                 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
1774
1775         if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
1776                 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
1777
1778         if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
1779                 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
1780                 (!nb_rxq || !nb_txq))
1781                 rte_exit(EXIT_FAILURE,
1782                         "Either rxq or txq are 0, cannot use %s fwd mode\n",
1783                         cur_fwd_eng->fwd_mode_name);
1784
1785         if (all_ports_started() == 0) {
1786                 printf("Not all ports were started\n");
1787                 return;
1788         }
1789         if (test_done == 0) {
1790                 printf("Packet forwarding already started\n");
1791                 return;
1792         }
1793
1794
1795         if(dcb_test) {
1796                 for (i = 0; i < nb_fwd_ports; i++) {
1797                         pt_id = fwd_ports_ids[i];
1798                         port = &ports[pt_id];
1799                         if (!port->dcb_flag) {
1800                                 printf("In DCB mode, all forwarding ports must "
1801                                        "be configured in this mode.\n");
1802                                 return;
1803                         }
1804                 }
1805                 if (nb_fwd_lcores == 1) {
1806                         printf("In DCB mode,the nb forwarding cores "
1807                                "should be larger than 1.\n");
1808                         return;
1809                 }
1810         }
1811         test_done = 0;
1812
1813         fwd_config_setup();
1814
1815         if(!no_flush_rx)
1816                 flush_fwd_rx_queues();
1817
1818         pkt_fwd_config_display(&cur_fwd_config);
1819         rxtx_config_display();
1820
1821         fwd_stats_reset();
1822         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1823                 pt_id = fwd_ports_ids[i];
1824                 port = &ports[pt_id];
1825                 map_port_queue_stats_mapping_registers(pt_id, port);
1826         }
1827         if (with_tx_first) {
1828                 port_fwd_begin = tx_only_engine.port_fwd_begin;
1829                 if (port_fwd_begin != NULL) {
1830                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1831                                 (*port_fwd_begin)(fwd_ports_ids[i]);
1832                 }
1833                 while (with_tx_first--) {
1834                         launch_packet_forwarding(
1835                                         run_one_txonly_burst_on_core);
1836                         rte_eal_mp_wait_lcore();
1837                 }
1838                 port_fwd_end = tx_only_engine.port_fwd_end;
1839                 if (port_fwd_end != NULL) {
1840                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
1841                                 (*port_fwd_end)(fwd_ports_ids[i]);
1842                 }
1843         }
1844         launch_packet_forwarding(start_pkt_forward_on_core);
1845 }
1846
1847 void
1848 stop_packet_forwarding(void)
1849 {
1850         port_fwd_end_t port_fwd_end;
1851         lcoreid_t lc_id;
1852         portid_t pt_id;
1853         int i;
1854
1855         if (test_done) {
1856                 printf("Packet forwarding not started\n");
1857                 return;
1858         }
1859         printf("Telling cores to stop...");
1860         for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
1861                 fwd_lcores[lc_id]->stopped = 1;
1862         printf("\nWaiting for lcores to finish...\n");
1863         rte_eal_mp_wait_lcore();
1864         port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
1865         if (port_fwd_end != NULL) {
1866                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1867                         pt_id = fwd_ports_ids[i];
1868                         (*port_fwd_end)(pt_id);
1869                 }
1870         }
1871
1872         fwd_stats_display();
1873
1874         printf("\nDone.\n");
1875         test_done = 1;
1876 }
1877
1878 void
1879 dev_set_link_up(portid_t pid)
1880 {
1881         if (rte_eth_dev_set_link_up(pid) < 0)
1882                 printf("\nSet link up fail.\n");
1883 }
1884
1885 void
1886 dev_set_link_down(portid_t pid)
1887 {
1888         if (rte_eth_dev_set_link_down(pid) < 0)
1889                 printf("\nSet link down fail.\n");
1890 }
1891
1892 static int
1893 all_ports_started(void)
1894 {
1895         portid_t pi;
1896         struct rte_port *port;
1897
1898         RTE_ETH_FOREACH_DEV(pi) {
1899                 port = &ports[pi];
1900                 /* Check if there is a port which is not started */
1901                 if ((port->port_status != RTE_PORT_STARTED) &&
1902                         (port->slave_flag == 0))
1903                         return 0;
1904         }
1905
1906         /* No port is not started */
1907         return 1;
1908 }
1909
1910 int
1911 port_is_stopped(portid_t port_id)
1912 {
1913         struct rte_port *port = &ports[port_id];
1914
1915         if ((port->port_status != RTE_PORT_STOPPED) &&
1916             (port->slave_flag == 0))
1917                 return 0;
1918         return 1;
1919 }
1920
1921 int
1922 all_ports_stopped(void)
1923 {
1924         portid_t pi;
1925
1926         RTE_ETH_FOREACH_DEV(pi) {
1927                 if (!port_is_stopped(pi))
1928                         return 0;
1929         }
1930
1931         return 1;
1932 }
1933
1934 int
1935 port_is_started(portid_t port_id)
1936 {
1937         if (port_id_is_invalid(port_id, ENABLED_WARN))
1938                 return 0;
1939
1940         if (ports[port_id].port_status != RTE_PORT_STARTED)
1941                 return 0;
1942
1943         return 1;
1944 }
1945
1946 int
1947 start_port(portid_t pid)
1948 {
1949         int diag, need_check_link_status = -1;
1950         portid_t pi;
1951         queueid_t qi;
1952         struct rte_port *port;
1953         struct ether_addr mac_addr;
1954
1955         if (port_id_is_invalid(pid, ENABLED_WARN))
1956                 return 0;
1957
1958         if(dcb_config)
1959                 dcb_test = 1;
1960         RTE_ETH_FOREACH_DEV(pi) {
1961                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
1962                         continue;
1963
1964                 need_check_link_status = 0;
1965                 port = &ports[pi];
1966                 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
1967                                                  RTE_PORT_HANDLING) == 0) {
1968                         printf("Port %d is now not stopped\n", pi);
1969                         continue;
1970                 }
1971
1972                 if (port->need_reconfig > 0) {
1973                         port->need_reconfig = 0;
1974
1975                         if (flow_isolate_all) {
1976                                 int ret = port_flow_isolate(pi, 1);
1977                                 if (ret) {
1978                                         printf("Failed to apply isolated"
1979                                                " mode on port %d\n", pi);
1980                                         return -1;
1981                                 }
1982                         }
1983                         configure_rxtx_dump_callbacks(0);
1984                         printf("Configuring Port %d (socket %u)\n", pi,
1985                                         port->socket_id);
1986                         /* configure port */
1987                         diag = rte_eth_dev_configure(pi, nb_rxq, nb_txq,
1988                                                 &(port->dev_conf));
1989                         if (diag != 0) {
1990                                 if (rte_atomic16_cmpset(&(port->port_status),
1991                                 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
1992                                         printf("Port %d can not be set back "
1993                                                         "to stopped\n", pi);
1994                                 printf("Fail to configure port %d\n", pi);
1995                                 /* try to reconfigure port next time */
1996                                 port->need_reconfig = 1;
1997                                 return -1;
1998                         }
1999                 }
2000                 if (port->need_reconfig_queues > 0) {
2001                         port->need_reconfig_queues = 0;
2002                         /* setup tx queues */
2003                         for (qi = 0; qi < nb_txq; qi++) {
2004                                 if ((numa_support) &&
2005                                         (txring_numa[pi] != NUMA_NO_CONFIG))
2006                                         diag = rte_eth_tx_queue_setup(pi, qi,
2007                                                 port->nb_tx_desc[qi],
2008                                                 txring_numa[pi],
2009                                                 &(port->tx_conf[qi]));
2010                                 else
2011                                         diag = rte_eth_tx_queue_setup(pi, qi,
2012                                                 port->nb_tx_desc[qi],
2013                                                 port->socket_id,
2014                                                 &(port->tx_conf[qi]));
2015
2016                                 if (diag == 0)
2017                                         continue;
2018
2019                                 /* Fail to setup tx queue, return */
2020                                 if (rte_atomic16_cmpset(&(port->port_status),
2021                                                         RTE_PORT_HANDLING,
2022                                                         RTE_PORT_STOPPED) == 0)
2023                                         printf("Port %d can not be set back "
2024                                                         "to stopped\n", pi);
2025                                 printf("Fail to configure port %d tx queues\n",
2026                                        pi);
2027                                 /* try to reconfigure queues next time */
2028                                 port->need_reconfig_queues = 1;
2029                                 return -1;
2030                         }
2031                         for (qi = 0; qi < nb_rxq; qi++) {
2032                                 /* setup rx queues */
2033                                 if ((numa_support) &&
2034                                         (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2035                                         struct rte_mempool * mp =
2036                                                 mbuf_pool_find(rxring_numa[pi]);
2037                                         if (mp == NULL) {
2038                                                 printf("Failed to setup RX queue:"
2039                                                         "No mempool allocation"
2040                                                         " on the socket %d\n",
2041                                                         rxring_numa[pi]);
2042                                                 return -1;
2043                                         }
2044
2045                                         diag = rte_eth_rx_queue_setup(pi, qi,
2046                                              port->nb_rx_desc[qi],
2047                                              rxring_numa[pi],
2048                                              &(port->rx_conf[qi]),
2049                                              mp);
2050                                 } else {
2051                                         struct rte_mempool *mp =
2052                                                 mbuf_pool_find(port->socket_id);
2053                                         if (mp == NULL) {
2054                                                 printf("Failed to setup RX queue:"
2055                                                         "No mempool allocation"
2056                                                         " on the socket %d\n",
2057                                                         port->socket_id);
2058                                                 return -1;
2059                                         }
2060                                         diag = rte_eth_rx_queue_setup(pi, qi,
2061                                              port->nb_rx_desc[qi],
2062                                              port->socket_id,
2063                                              &(port->rx_conf[qi]),
2064                                              mp);
2065                                 }
2066                                 if (diag == 0)
2067                                         continue;
2068
2069                                 /* Fail to setup rx queue, return */
2070                                 if (rte_atomic16_cmpset(&(port->port_status),
2071                                                         RTE_PORT_HANDLING,
2072                                                         RTE_PORT_STOPPED) == 0)
2073                                         printf("Port %d can not be set back "
2074                                                         "to stopped\n", pi);
2075                                 printf("Fail to configure port %d rx queues\n",
2076                                        pi);
2077                                 /* try to reconfigure queues next time */
2078                                 port->need_reconfig_queues = 1;
2079                                 return -1;
2080                         }
2081                 }
2082                 configure_rxtx_dump_callbacks(verbose_level);
2083                 /* start port */
2084                 if (rte_eth_dev_start(pi) < 0) {
2085                         printf("Fail to start port %d\n", pi);
2086
2087                         /* Fail to setup rx queue, return */
2088                         if (rte_atomic16_cmpset(&(port->port_status),
2089                                 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2090                                 printf("Port %d can not be set back to "
2091                                                         "stopped\n", pi);
2092                         continue;
2093                 }
2094
2095                 if (rte_atomic16_cmpset(&(port->port_status),
2096                         RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2097                         printf("Port %d can not be set into started\n", pi);
2098
2099                 rte_eth_macaddr_get(pi, &mac_addr);
2100                 printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2101                                 mac_addr.addr_bytes[0], mac_addr.addr_bytes[1],
2102                                 mac_addr.addr_bytes[2], mac_addr.addr_bytes[3],
2103                                 mac_addr.addr_bytes[4], mac_addr.addr_bytes[5]);
2104
2105                 /* at least one port started, need checking link status */
2106                 need_check_link_status = 1;
2107         }
2108
2109         if (need_check_link_status == 1 && !no_link_check)
2110                 check_all_ports_link_status(RTE_PORT_ALL);
2111         else if (need_check_link_status == 0)
2112                 printf("Please stop the ports first\n");
2113
2114         printf("Done\n");
2115         return 0;
2116 }
2117
2118 void
2119 stop_port(portid_t pid)
2120 {
2121         portid_t pi;
2122         struct rte_port *port;
2123         int need_check_link_status = 0;
2124
2125         if (dcb_test) {
2126                 dcb_test = 0;
2127                 dcb_config = 0;
2128         }
2129
2130         if (port_id_is_invalid(pid, ENABLED_WARN))
2131                 return;
2132
2133         printf("Stopping ports...\n");
2134
2135         RTE_ETH_FOREACH_DEV(pi) {
2136                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2137                         continue;
2138
2139                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2140                         printf("Please remove port %d from forwarding configuration.\n", pi);
2141                         continue;
2142                 }
2143
2144                 if (port_is_bonding_slave(pi)) {
2145                         printf("Please remove port %d from bonded device.\n", pi);
2146                         continue;
2147                 }
2148
2149                 port = &ports[pi];
2150                 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2151                                                 RTE_PORT_HANDLING) == 0)
2152                         continue;
2153
2154                 rte_eth_dev_stop(pi);
2155
2156                 if (rte_atomic16_cmpset(&(port->port_status),
2157                         RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2158                         printf("Port %d can not be set into stopped\n", pi);
2159                 need_check_link_status = 1;
2160         }
2161         if (need_check_link_status && !no_link_check)
2162                 check_all_ports_link_status(RTE_PORT_ALL);
2163
2164         printf("Done\n");
2165 }
2166
2167 static void
2168 remove_invalid_ports_in(portid_t *array, portid_t *total)
2169 {
2170         portid_t i;
2171         portid_t new_total = 0;
2172
2173         for (i = 0; i < *total; i++)
2174                 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2175                         array[new_total] = array[i];
2176                         new_total++;
2177                 }
2178         *total = new_total;
2179 }
2180
2181 static void
2182 remove_invalid_ports(void)
2183 {
2184         remove_invalid_ports_in(ports_ids, &nb_ports);
2185         remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2186         nb_cfg_ports = nb_fwd_ports;
2187 }
2188
2189 void
2190 close_port(portid_t pid)
2191 {
2192         portid_t pi;
2193         struct rte_port *port;
2194
2195         if (port_id_is_invalid(pid, ENABLED_WARN))
2196                 return;
2197
2198         printf("Closing ports...\n");
2199
2200         RTE_ETH_FOREACH_DEV(pi) {
2201                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2202                         continue;
2203
2204                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2205                         printf("Please remove port %d from forwarding configuration.\n", pi);
2206                         continue;
2207                 }
2208
2209                 if (port_is_bonding_slave(pi)) {
2210                         printf("Please remove port %d from bonded device.\n", pi);
2211                         continue;
2212                 }
2213
2214                 port = &ports[pi];
2215                 if (rte_atomic16_cmpset(&(port->port_status),
2216                         RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2217                         printf("Port %d is already closed\n", pi);
2218                         continue;
2219                 }
2220
2221                 if (rte_atomic16_cmpset(&(port->port_status),
2222                         RTE_PORT_STOPPED, RTE_PORT_HANDLING) == 0) {
2223                         printf("Port %d is now not stopped\n", pi);
2224                         continue;
2225                 }
2226
2227                 if (port->flow_list)
2228                         port_flow_flush(pi);
2229                 rte_eth_dev_close(pi);
2230
2231                 remove_invalid_ports();
2232
2233                 if (rte_atomic16_cmpset(&(port->port_status),
2234                         RTE_PORT_HANDLING, RTE_PORT_CLOSED) == 0)
2235                         printf("Port %d cannot be set to closed\n", pi);
2236         }
2237
2238         printf("Done\n");
2239 }
2240
2241 void
2242 reset_port(portid_t pid)
2243 {
2244         int diag;
2245         portid_t pi;
2246         struct rte_port *port;
2247
2248         if (port_id_is_invalid(pid, ENABLED_WARN))
2249                 return;
2250
2251         printf("Resetting ports...\n");
2252
2253         RTE_ETH_FOREACH_DEV(pi) {
2254                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2255                         continue;
2256
2257                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2258                         printf("Please remove port %d from forwarding "
2259                                "configuration.\n", pi);
2260                         continue;
2261                 }
2262
2263                 if (port_is_bonding_slave(pi)) {
2264                         printf("Please remove port %d from bonded device.\n",
2265                                pi);
2266                         continue;
2267                 }
2268
2269                 diag = rte_eth_dev_reset(pi);
2270                 if (diag == 0) {
2271                         port = &ports[pi];
2272                         port->need_reconfig = 1;
2273                         port->need_reconfig_queues = 1;
2274                 } else {
2275                         printf("Failed to reset port %d. diag=%d\n", pi, diag);
2276                 }
2277         }
2278
2279         printf("Done\n");
2280 }
2281
2282 void
2283 attach_port(char *identifier)
2284 {
2285         portid_t pi;
2286         struct rte_dev_iterator iterator;
2287
2288         printf("Attaching a new port...\n");
2289
2290         if (identifier == NULL) {
2291                 printf("Invalid parameters are specified\n");
2292                 return;
2293         }
2294
2295         if (rte_dev_probe(identifier) != 0) {
2296                 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2297                 return;
2298         }
2299
2300         /* first attach mode: event */
2301         if (setup_on_probe_event) {
2302                 /* new ports are detected on RTE_ETH_EVENT_NEW event */
2303                 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2304                         if (ports[pi].port_status == RTE_PORT_HANDLING &&
2305                                         ports[pi].need_setup != 0)
2306                                 setup_attached_port(pi);
2307                 return;
2308         }
2309
2310         /* second attach mode: iterator */
2311         RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2312                 /* setup ports matching the devargs used for probing */
2313                 if (port_is_forwarding(pi))
2314                         continue; /* port was already attached before */
2315                 setup_attached_port(pi);
2316         }
2317 }
2318
2319 static void
2320 setup_attached_port(portid_t pi)
2321 {
2322         unsigned int socket_id;
2323
2324         socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2325         /* if socket_id is invalid, set to the first available socket. */
2326         if (check_socket_id(socket_id) < 0)
2327                 socket_id = socket_ids[0];
2328         reconfig(pi, socket_id);
2329         rte_eth_promiscuous_enable(pi);
2330
2331         ports_ids[nb_ports++] = pi;
2332         fwd_ports_ids[nb_fwd_ports++] = pi;
2333         nb_cfg_ports = nb_fwd_ports;
2334         ports[pi].need_setup = 0;
2335         ports[pi].port_status = RTE_PORT_STOPPED;
2336
2337         printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
2338         printf("Done\n");
2339 }
2340
2341 void
2342 detach_port_device(portid_t port_id)
2343 {
2344         struct rte_device *dev;
2345         portid_t sibling;
2346
2347         printf("Removing a device...\n");
2348
2349         dev = rte_eth_devices[port_id].device;
2350         if (dev == NULL) {
2351                 printf("Device already removed\n");
2352                 return;
2353         }
2354
2355         if (ports[port_id].port_status != RTE_PORT_CLOSED) {
2356                 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
2357                         printf("Port not stopped\n");
2358                         return;
2359                 }
2360                 printf("Port was not closed\n");
2361                 if (ports[port_id].flow_list)
2362                         port_flow_flush(port_id);
2363         }
2364
2365         if (rte_dev_remove(dev) != 0) {
2366                 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
2367                 return;
2368         }
2369
2370         for (sibling = 0; sibling < RTE_MAX_ETHPORTS; sibling++) {
2371                 if (rte_eth_devices[sibling].device != dev)
2372                         continue;
2373                 /* reset mapping between old ports and removed device */
2374                 rte_eth_devices[sibling].device = NULL;
2375                 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
2376                         /* sibling ports are forced to be closed */
2377                         ports[sibling].port_status = RTE_PORT_CLOSED;
2378                         printf("Port %u is closed\n", sibling);
2379                 }
2380         }
2381
2382         remove_invalid_ports();
2383
2384         printf("Device of port %u is detached\n", port_id);
2385         printf("Now total ports is %d\n", nb_ports);
2386         printf("Done\n");
2387         return;
2388 }
2389
2390 void
2391 pmd_test_exit(void)
2392 {
2393         struct rte_device *device;
2394         portid_t pt_id;
2395         int ret;
2396
2397         if (test_done == 0)
2398                 stop_packet_forwarding();
2399
2400         if (ports != NULL) {
2401                 no_link_check = 1;
2402                 RTE_ETH_FOREACH_DEV(pt_id) {
2403                         printf("\nStopping port %d...\n", pt_id);
2404                         fflush(stdout);
2405                         stop_port(pt_id);
2406                 }
2407                 RTE_ETH_FOREACH_DEV(pt_id) {
2408                         printf("\nShutting down port %d...\n", pt_id);
2409                         fflush(stdout);
2410                         close_port(pt_id);
2411
2412                         /*
2413                          * This is a workaround to fix a virtio-user issue that
2414                          * requires to call clean-up routine to remove existing
2415                          * socket.
2416                          * This workaround valid only for testpmd, needs a fix
2417                          * valid for all applications.
2418                          * TODO: Implement proper resource cleanup
2419                          */
2420                         device = rte_eth_devices[pt_id].device;
2421                         if (device && !strcmp(device->driver->name, "net_virtio_user"))
2422                                 detach_port_device(pt_id);
2423                 }
2424         }
2425
2426         if (hot_plug) {
2427                 ret = rte_dev_event_monitor_stop();
2428                 if (ret) {
2429                         RTE_LOG(ERR, EAL,
2430                                 "fail to stop device event monitor.");
2431                         return;
2432                 }
2433
2434                 ret = rte_dev_event_callback_unregister(NULL,
2435                         dev_event_callback, NULL);
2436                 if (ret < 0) {
2437                         RTE_LOG(ERR, EAL,
2438                                 "fail to unregister device event callback.\n");
2439                         return;
2440                 }
2441
2442                 ret = rte_dev_hotplug_handle_disable();
2443                 if (ret) {
2444                         RTE_LOG(ERR, EAL,
2445                                 "fail to disable hotplug handling.\n");
2446                         return;
2447                 }
2448         }
2449
2450         printf("\nBye...\n");
2451 }
2452
2453 typedef void (*cmd_func_t)(void);
2454 struct pmd_test_command {
2455         const char *cmd_name;
2456         cmd_func_t cmd_func;
2457 };
2458
2459 #define PMD_TEST_CMD_NB (sizeof(pmd_test_menu) / sizeof(pmd_test_menu[0]))
2460
2461 /* Check the link status of all ports in up to 9s, and print them finally */
2462 static void
2463 check_all_ports_link_status(uint32_t port_mask)
2464 {
2465 #define CHECK_INTERVAL 100 /* 100ms */
2466 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2467         portid_t portid;
2468         uint8_t count, all_ports_up, print_flag = 0;
2469         struct rte_eth_link link;
2470
2471         printf("Checking link statuses...\n");
2472         fflush(stdout);
2473         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2474                 all_ports_up = 1;
2475                 RTE_ETH_FOREACH_DEV(portid) {
2476                         if ((port_mask & (1 << portid)) == 0)
2477                                 continue;
2478                         memset(&link, 0, sizeof(link));
2479                         rte_eth_link_get_nowait(portid, &link);
2480                         /* print link status if flag set */
2481                         if (print_flag == 1) {
2482                                 if (link.link_status)
2483                                         printf(
2484                                         "Port%d Link Up. speed %u Mbps- %s\n",
2485                                         portid, link.link_speed,
2486                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2487                                         ("full-duplex") : ("half-duplex\n"));
2488                                 else
2489                                         printf("Port %d Link Down\n", portid);
2490                                 continue;
2491                         }
2492                         /* clear all_ports_up flag if any link down */
2493                         if (link.link_status == ETH_LINK_DOWN) {
2494                                 all_ports_up = 0;
2495                                 break;
2496                         }
2497                 }
2498                 /* after finally printing all link status, get out */
2499                 if (print_flag == 1)
2500                         break;
2501
2502                 if (all_ports_up == 0) {
2503                         fflush(stdout);
2504                         rte_delay_ms(CHECK_INTERVAL);
2505                 }
2506
2507                 /* set the print_flag if all ports up or timeout */
2508                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2509                         print_flag = 1;
2510                 }
2511
2512                 if (lsc_interrupt)
2513                         break;
2514         }
2515 }
2516
2517 /*
2518  * This callback is for remove a port for a device. It has limitation because
2519  * it is not for multiple port removal for a device.
2520  * TODO: the device detach invoke will plan to be removed from user side to
2521  * eal. And convert all PMDs to free port resources on ether device closing.
2522  */
2523 static void
2524 rmv_port_callback(void *arg)
2525 {
2526         int need_to_start = 0;
2527         int org_no_link_check = no_link_check;
2528         portid_t port_id = (intptr_t)arg;
2529
2530         RTE_ETH_VALID_PORTID_OR_RET(port_id);
2531
2532         if (!test_done && port_is_forwarding(port_id)) {
2533                 need_to_start = 1;
2534                 stop_packet_forwarding();
2535         }
2536         no_link_check = 1;
2537         stop_port(port_id);
2538         no_link_check = org_no_link_check;
2539         close_port(port_id);
2540         detach_port_device(port_id);
2541         if (need_to_start)
2542                 start_packet_forwarding(0);
2543 }
2544
2545 /* This function is used by the interrupt thread */
2546 static int
2547 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
2548                   void *ret_param)
2549 {
2550         RTE_SET_USED(param);
2551         RTE_SET_USED(ret_param);
2552
2553         if (type >= RTE_ETH_EVENT_MAX) {
2554                 fprintf(stderr, "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
2555                         port_id, __func__, type);
2556                 fflush(stderr);
2557         } else if (event_print_mask & (UINT32_C(1) << type)) {
2558                 printf("\nPort %" PRIu16 ": %s event\n", port_id,
2559                         eth_event_desc[type]);
2560                 fflush(stdout);
2561         }
2562
2563         switch (type) {
2564         case RTE_ETH_EVENT_NEW:
2565                 ports[port_id].need_setup = 1;
2566                 ports[port_id].port_status = RTE_PORT_HANDLING;
2567                 break;
2568         case RTE_ETH_EVENT_INTR_RMV:
2569                 if (port_id_is_invalid(port_id, DISABLED_WARN))
2570                         break;
2571                 if (rte_eal_alarm_set(100000,
2572                                 rmv_port_callback, (void *)(intptr_t)port_id))
2573                         fprintf(stderr, "Could not set up deferred device removal\n");
2574                 break;
2575         default:
2576                 break;
2577         }
2578         return 0;
2579 }
2580
2581 static int
2582 register_eth_event_callback(void)
2583 {
2584         int ret;
2585         enum rte_eth_event_type event;
2586
2587         for (event = RTE_ETH_EVENT_UNKNOWN;
2588                         event < RTE_ETH_EVENT_MAX; event++) {
2589                 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
2590                                 event,
2591                                 eth_event_callback,
2592                                 NULL);
2593                 if (ret != 0) {
2594                         TESTPMD_LOG(ERR, "Failed to register callback for "
2595                                         "%s event\n", eth_event_desc[event]);
2596                         return -1;
2597                 }
2598         }
2599
2600         return 0;
2601 }
2602
2603 /* This function is used by the interrupt thread */
2604 static void
2605 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
2606                              __rte_unused void *arg)
2607 {
2608         uint16_t port_id;
2609         int ret;
2610
2611         if (type >= RTE_DEV_EVENT_MAX) {
2612                 fprintf(stderr, "%s called upon invalid event %d\n",
2613                         __func__, type);
2614                 fflush(stderr);
2615         }
2616
2617         switch (type) {
2618         case RTE_DEV_EVENT_REMOVE:
2619                 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
2620                         device_name);
2621                 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
2622                 if (ret) {
2623                         RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
2624                                 device_name);
2625                         return;
2626                 }
2627                 /*
2628                  * Because the user's callback is invoked in eal interrupt
2629                  * callback, the interrupt callback need to be finished before
2630                  * it can be unregistered when detaching device. So finish
2631                  * callback soon and use a deferred removal to detach device
2632                  * is need. It is a workaround, once the device detaching be
2633                  * moved into the eal in the future, the deferred removal could
2634                  * be deleted.
2635                  */
2636                 if (rte_eal_alarm_set(100000,
2637                                 rmv_port_callback, (void *)(intptr_t)port_id))
2638                         RTE_LOG(ERR, EAL,
2639                                 "Could not set up deferred device removal\n");
2640                 break;
2641         case RTE_DEV_EVENT_ADD:
2642                 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
2643                         device_name);
2644                 /* TODO: After finish kernel driver binding,
2645                  * begin to attach port.
2646                  */
2647                 break;
2648         default:
2649                 break;
2650         }
2651 }
2652
2653 static int
2654 set_tx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2655 {
2656         uint16_t i;
2657         int diag;
2658         uint8_t mapping_found = 0;
2659
2660         for (i = 0; i < nb_tx_queue_stats_mappings; i++) {
2661                 if ((tx_queue_stats_mappings[i].port_id == port_id) &&
2662                                 (tx_queue_stats_mappings[i].queue_id < nb_txq )) {
2663                         diag = rte_eth_dev_set_tx_queue_stats_mapping(port_id,
2664                                         tx_queue_stats_mappings[i].queue_id,
2665                                         tx_queue_stats_mappings[i].stats_counter_id);
2666                         if (diag != 0)
2667                                 return diag;
2668                         mapping_found = 1;
2669                 }
2670         }
2671         if (mapping_found)
2672                 port->tx_queue_stats_mapping_enabled = 1;
2673         return 0;
2674 }
2675
2676 static int
2677 set_rx_queue_stats_mapping_registers(portid_t port_id, struct rte_port *port)
2678 {
2679         uint16_t i;
2680         int diag;
2681         uint8_t mapping_found = 0;
2682
2683         for (i = 0; i < nb_rx_queue_stats_mappings; i++) {
2684                 if ((rx_queue_stats_mappings[i].port_id == port_id) &&
2685                                 (rx_queue_stats_mappings[i].queue_id < nb_rxq )) {
2686                         diag = rte_eth_dev_set_rx_queue_stats_mapping(port_id,
2687                                         rx_queue_stats_mappings[i].queue_id,
2688                                         rx_queue_stats_mappings[i].stats_counter_id);
2689                         if (diag != 0)
2690                                 return diag;
2691                         mapping_found = 1;
2692                 }
2693         }
2694         if (mapping_found)
2695                 port->rx_queue_stats_mapping_enabled = 1;
2696         return 0;
2697 }
2698
2699 static void
2700 map_port_queue_stats_mapping_registers(portid_t pi, struct rte_port *port)
2701 {
2702         int diag = 0;
2703
2704         diag = set_tx_queue_stats_mapping_registers(pi, port);
2705         if (diag != 0) {
2706                 if (diag == -ENOTSUP) {
2707                         port->tx_queue_stats_mapping_enabled = 0;
2708                         printf("TX queue stats mapping not supported port id=%d\n", pi);
2709                 }
2710                 else
2711                         rte_exit(EXIT_FAILURE,
2712                                         "set_tx_queue_stats_mapping_registers "
2713                                         "failed for port id=%d diag=%d\n",
2714                                         pi, diag);
2715         }
2716
2717         diag = set_rx_queue_stats_mapping_registers(pi, port);
2718         if (diag != 0) {
2719                 if (diag == -ENOTSUP) {
2720                         port->rx_queue_stats_mapping_enabled = 0;
2721                         printf("RX queue stats mapping not supported port id=%d\n", pi);
2722                 }
2723                 else
2724                         rte_exit(EXIT_FAILURE,
2725                                         "set_rx_queue_stats_mapping_registers "
2726                                         "failed for port id=%d diag=%d\n",
2727                                         pi, diag);
2728         }
2729 }
2730
2731 static void
2732 rxtx_port_config(struct rte_port *port)
2733 {
2734         uint16_t qid;
2735
2736         for (qid = 0; qid < nb_rxq; qid++) {
2737                 port->rx_conf[qid] = port->dev_info.default_rxconf;
2738
2739                 /* Check if any Rx parameters have been passed */
2740                 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
2741                         port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
2742
2743                 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
2744                         port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
2745
2746                 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
2747                         port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
2748
2749                 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
2750                         port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
2751
2752                 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
2753                         port->rx_conf[qid].rx_drop_en = rx_drop_en;
2754
2755                 port->nb_rx_desc[qid] = nb_rxd;
2756         }
2757
2758         for (qid = 0; qid < nb_txq; qid++) {
2759                 port->tx_conf[qid] = port->dev_info.default_txconf;
2760
2761                 /* Check if any Tx parameters have been passed */
2762                 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
2763                         port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
2764
2765                 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
2766                         port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
2767
2768                 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
2769                         port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
2770
2771                 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
2772                         port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
2773
2774                 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
2775                         port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
2776
2777                 port->nb_tx_desc[qid] = nb_txd;
2778         }
2779 }
2780
2781 void
2782 init_port_config(void)
2783 {
2784         portid_t pid;
2785         struct rte_port *port;
2786
2787         RTE_ETH_FOREACH_DEV(pid) {
2788                 port = &ports[pid];
2789                 port->dev_conf.fdir_conf = fdir_conf;
2790                 rte_eth_dev_info_get(pid, &port->dev_info);
2791                 if (nb_rxq > 1) {
2792                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2793                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
2794                                 rss_hf & port->dev_info.flow_type_rss_offloads;
2795                 } else {
2796                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
2797                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
2798                 }
2799
2800                 if (port->dcb_flag == 0) {
2801                         if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
2802                                 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
2803                         else
2804                                 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
2805                 }
2806
2807                 rxtx_port_config(port);
2808
2809                 rte_eth_macaddr_get(pid, &port->eth_addr);
2810
2811                 map_port_queue_stats_mapping_registers(pid, port);
2812 #if defined RTE_LIBRTE_IXGBE_PMD && defined RTE_LIBRTE_IXGBE_BYPASS
2813                 rte_pmd_ixgbe_bypass_init(pid);
2814 #endif
2815
2816                 if (lsc_interrupt &&
2817                     (rte_eth_devices[pid].data->dev_flags &
2818                      RTE_ETH_DEV_INTR_LSC))
2819                         port->dev_conf.intr_conf.lsc = 1;
2820                 if (rmv_interrupt &&
2821                     (rte_eth_devices[pid].data->dev_flags &
2822                      RTE_ETH_DEV_INTR_RMV))
2823                         port->dev_conf.intr_conf.rmv = 1;
2824         }
2825 }
2826
2827 void set_port_slave_flag(portid_t slave_pid)
2828 {
2829         struct rte_port *port;
2830
2831         port = &ports[slave_pid];
2832         port->slave_flag = 1;
2833 }
2834
2835 void clear_port_slave_flag(portid_t slave_pid)
2836 {
2837         struct rte_port *port;
2838
2839         port = &ports[slave_pid];
2840         port->slave_flag = 0;
2841 }
2842
2843 uint8_t port_is_bonding_slave(portid_t slave_pid)
2844 {
2845         struct rte_port *port;
2846
2847         port = &ports[slave_pid];
2848         if ((rte_eth_devices[slave_pid].data->dev_flags &
2849             RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
2850                 return 1;
2851         return 0;
2852 }
2853
2854 const uint16_t vlan_tags[] = {
2855                 0,  1,  2,  3,  4,  5,  6,  7,
2856                 8,  9, 10, 11,  12, 13, 14, 15,
2857                 16, 17, 18, 19, 20, 21, 22, 23,
2858                 24, 25, 26, 27, 28, 29, 30, 31
2859 };
2860
2861 static  int
2862 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
2863                  enum dcb_mode_enable dcb_mode,
2864                  enum rte_eth_nb_tcs num_tcs,
2865                  uint8_t pfc_en)
2866 {
2867         uint8_t i;
2868         int32_t rc;
2869         struct rte_eth_rss_conf rss_conf;
2870
2871         /*
2872          * Builds up the correct configuration for dcb+vt based on the vlan tags array
2873          * given above, and the number of traffic classes available for use.
2874          */
2875         if (dcb_mode == DCB_VT_ENABLED) {
2876                 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
2877                                 &eth_conf->rx_adv_conf.vmdq_dcb_conf;
2878                 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
2879                                 &eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
2880
2881                 /* VMDQ+DCB RX and TX configurations */
2882                 vmdq_rx_conf->enable_default_pool = 0;
2883                 vmdq_rx_conf->default_pool = 0;
2884                 vmdq_rx_conf->nb_queue_pools =
2885                         (num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2886                 vmdq_tx_conf->nb_queue_pools =
2887                         (num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
2888
2889                 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
2890                 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
2891                         vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
2892                         vmdq_rx_conf->pool_map[i].pools =
2893                                 1 << (i % vmdq_rx_conf->nb_queue_pools);
2894                 }
2895                 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2896                         vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
2897                         vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
2898                 }
2899
2900                 /* set DCB mode of RX and TX of multiple queues */
2901                 eth_conf->rxmode.mq_mode = ETH_MQ_RX_VMDQ_DCB;
2902                 eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
2903         } else {
2904                 struct rte_eth_dcb_rx_conf *rx_conf =
2905                                 &eth_conf->rx_adv_conf.dcb_rx_conf;
2906                 struct rte_eth_dcb_tx_conf *tx_conf =
2907                                 &eth_conf->tx_adv_conf.dcb_tx_conf;
2908
2909                 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
2910                 if (rc != 0)
2911                         return rc;
2912
2913                 rx_conf->nb_tcs = num_tcs;
2914                 tx_conf->nb_tcs = num_tcs;
2915
2916                 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2917                         rx_conf->dcb_tc[i] = i % num_tcs;
2918                         tx_conf->dcb_tc[i] = i % num_tcs;
2919                 }
2920
2921                 eth_conf->rxmode.mq_mode = ETH_MQ_RX_DCB_RSS;
2922                 eth_conf->rx_adv_conf.rss_conf = rss_conf;
2923                 eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
2924         }
2925
2926         if (pfc_en)
2927                 eth_conf->dcb_capability_en =
2928                                 ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
2929         else
2930                 eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
2931
2932         return 0;
2933 }
2934
2935 int
2936 init_port_dcb_config(portid_t pid,
2937                      enum dcb_mode_enable dcb_mode,
2938                      enum rte_eth_nb_tcs num_tcs,
2939                      uint8_t pfc_en)
2940 {
2941         struct rte_eth_conf port_conf;
2942         struct rte_port *rte_port;
2943         int retval;
2944         uint16_t i;
2945
2946         rte_port = &ports[pid];
2947
2948         memset(&port_conf, 0, sizeof(struct rte_eth_conf));
2949         /* Enter DCB configuration status */
2950         dcb_config = 1;
2951
2952         port_conf.rxmode = rte_port->dev_conf.rxmode;
2953         port_conf.txmode = rte_port->dev_conf.txmode;
2954
2955         /*set configuration of DCB in vt mode and DCB in non-vt mode*/
2956         retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
2957         if (retval < 0)
2958                 return retval;
2959         port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
2960
2961         /* re-configure the device . */
2962         rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
2963
2964         rte_eth_dev_info_get(pid, &rte_port->dev_info);
2965
2966         /* If dev_info.vmdq_pool_base is greater than 0,
2967          * the queue id of vmdq pools is started after pf queues.
2968          */
2969         if (dcb_mode == DCB_VT_ENABLED &&
2970             rte_port->dev_info.vmdq_pool_base > 0) {
2971                 printf("VMDQ_DCB multi-queue mode is nonsensical"
2972                         " for port %d.", pid);
2973                 return -1;
2974         }
2975
2976         /* Assume the ports in testpmd have the same dcb capability
2977          * and has the same number of rxq and txq in dcb mode
2978          */
2979         if (dcb_mode == DCB_VT_ENABLED) {
2980                 if (rte_port->dev_info.max_vfs > 0) {
2981                         nb_rxq = rte_port->dev_info.nb_rx_queues;
2982                         nb_txq = rte_port->dev_info.nb_tx_queues;
2983                 } else {
2984                         nb_rxq = rte_port->dev_info.max_rx_queues;
2985                         nb_txq = rte_port->dev_info.max_tx_queues;
2986                 }
2987         } else {
2988                 /*if vt is disabled, use all pf queues */
2989                 if (rte_port->dev_info.vmdq_pool_base == 0) {
2990                         nb_rxq = rte_port->dev_info.max_rx_queues;
2991                         nb_txq = rte_port->dev_info.max_tx_queues;
2992                 } else {
2993                         nb_rxq = (queueid_t)num_tcs;
2994                         nb_txq = (queueid_t)num_tcs;
2995
2996                 }
2997         }
2998         rx_free_thresh = 64;
2999
3000         memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3001
3002         rxtx_port_config(rte_port);
3003         /* VLAN filter */
3004         rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3005         for (i = 0; i < RTE_DIM(vlan_tags); i++)
3006                 rx_vft_set(pid, vlan_tags[i], 1);
3007
3008         rte_eth_macaddr_get(pid, &rte_port->eth_addr);
3009         map_port_queue_stats_mapping_registers(pid, rte_port);
3010
3011         rte_port->dcb_flag = 1;
3012
3013         return 0;
3014 }
3015
3016 static void
3017 init_port(void)
3018 {
3019         /* Configuration of Ethernet ports. */
3020         ports = rte_zmalloc("testpmd: ports",
3021                             sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3022                             RTE_CACHE_LINE_SIZE);
3023         if (ports == NULL) {
3024                 rte_exit(EXIT_FAILURE,
3025                                 "rte_zmalloc(%d struct rte_port) failed\n",
3026                                 RTE_MAX_ETHPORTS);
3027         }
3028
3029         /* Initialize ports NUMA structures */
3030         memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3031         memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3032         memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3033 }
3034
3035 static void
3036 force_quit(void)
3037 {
3038         pmd_test_exit();
3039         prompt_exit();
3040 }
3041
3042 static void
3043 print_stats(void)
3044 {
3045         uint8_t i;
3046         const char clr[] = { 27, '[', '2', 'J', '\0' };
3047         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3048
3049         /* Clear screen and move to top left */
3050         printf("%s%s", clr, top_left);
3051
3052         printf("\nPort statistics ====================================");
3053         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3054                 nic_stats_display(fwd_ports_ids[i]);
3055
3056         fflush(stdout);
3057 }
3058
3059 static void
3060 signal_handler(int signum)
3061 {
3062         if (signum == SIGINT || signum == SIGTERM) {
3063                 printf("\nSignal %d received, preparing to exit...\n",
3064                                 signum);
3065 #ifdef RTE_LIBRTE_PDUMP
3066                 /* uninitialize packet capture framework */
3067                 rte_pdump_uninit();
3068 #endif
3069 #ifdef RTE_LIBRTE_LATENCY_STATS
3070                 rte_latencystats_uninit();
3071 #endif
3072                 force_quit();
3073                 /* Set flag to indicate the force termination. */
3074                 f_quit = 1;
3075                 /* exit with the expected status */
3076                 signal(signum, SIG_DFL);
3077                 kill(getpid(), signum);
3078         }
3079 }
3080
3081 int
3082 main(int argc, char** argv)
3083 {
3084         int diag;
3085         portid_t port_id;
3086         uint16_t count;
3087         int ret;
3088
3089         signal(SIGINT, signal_handler);
3090         signal(SIGTERM, signal_handler);
3091
3092         diag = rte_eal_init(argc, argv);
3093         if (diag < 0)
3094                 rte_panic("Cannot init EAL\n");
3095
3096         testpmd_logtype = rte_log_register("testpmd");
3097         if (testpmd_logtype < 0)
3098                 rte_panic("Cannot register log type");
3099         rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3100
3101         ret = register_eth_event_callback();
3102         if (ret != 0)
3103                 rte_panic("Cannot register for ethdev events");
3104
3105 #ifdef RTE_LIBRTE_PDUMP
3106         /* initialize packet capture framework */
3107         rte_pdump_init();
3108 #endif
3109
3110         count = 0;
3111         RTE_ETH_FOREACH_DEV(port_id) {
3112                 ports_ids[count] = port_id;
3113                 count++;
3114         }
3115         nb_ports = (portid_t) count;
3116         if (nb_ports == 0)
3117                 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3118
3119         /* allocate port structures, and init them */
3120         init_port();
3121
3122         set_def_fwd_config();
3123         if (nb_lcores == 0)
3124                 rte_panic("Empty set of forwarding logical cores - check the "
3125                           "core mask supplied in the command parameters\n");
3126
3127         /* Bitrate/latency stats disabled by default */
3128 #ifdef RTE_LIBRTE_BITRATE
3129         bitrate_enabled = 0;
3130 #endif
3131 #ifdef RTE_LIBRTE_LATENCY_STATS
3132         latencystats_enabled = 0;
3133 #endif
3134
3135         /* on FreeBSD, mlockall() is disabled by default */
3136 #ifdef RTE_EXEC_ENV_FREEBSD
3137         do_mlockall = 0;
3138 #else
3139         do_mlockall = 1;
3140 #endif
3141
3142         argc -= diag;
3143         argv += diag;
3144         if (argc > 1)
3145                 launch_args_parse(argc, argv);
3146
3147         if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3148                 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3149                         strerror(errno));
3150         }
3151
3152         if (tx_first && interactive)
3153                 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3154                                 "interactive mode.\n");
3155
3156         if (tx_first && lsc_interrupt) {
3157                 printf("Warning: lsc_interrupt needs to be off when "
3158                                 " using tx_first. Disabling.\n");
3159                 lsc_interrupt = 0;
3160         }
3161
3162         if (!nb_rxq && !nb_txq)
3163                 printf("Warning: Either rx or tx queues should be non-zero\n");
3164
3165         if (nb_rxq > 1 && nb_rxq > nb_txq)
3166                 printf("Warning: nb_rxq=%d enables RSS configuration, "
3167                        "but nb_txq=%d will prevent to fully test it.\n",
3168                        nb_rxq, nb_txq);
3169
3170         init_config();
3171
3172         if (hot_plug) {
3173                 ret = rte_dev_hotplug_handle_enable();
3174                 if (ret) {
3175                         RTE_LOG(ERR, EAL,
3176                                 "fail to enable hotplug handling.");
3177                         return -1;
3178                 }
3179
3180                 ret = rte_dev_event_monitor_start();
3181                 if (ret) {
3182                         RTE_LOG(ERR, EAL,
3183                                 "fail to start device event monitoring.");
3184                         return -1;
3185                 }
3186
3187                 ret = rte_dev_event_callback_register(NULL,
3188                         dev_event_callback, NULL);
3189                 if (ret) {
3190                         RTE_LOG(ERR, EAL,
3191                                 "fail  to register device event callback\n");
3192                         return -1;
3193                 }
3194         }
3195
3196         if (start_port(RTE_PORT_ALL) != 0)
3197                 rte_exit(EXIT_FAILURE, "Start ports failed\n");
3198
3199         /* set all ports to promiscuous mode by default */
3200         RTE_ETH_FOREACH_DEV(port_id)
3201                 rte_eth_promiscuous_enable(port_id);
3202
3203         /* Init metrics library */
3204         rte_metrics_init(rte_socket_id());
3205
3206 #ifdef RTE_LIBRTE_LATENCY_STATS
3207         if (latencystats_enabled != 0) {
3208                 int ret = rte_latencystats_init(1, NULL);
3209                 if (ret)
3210                         printf("Warning: latencystats init()"
3211                                 " returned error %d\n", ret);
3212                 printf("Latencystats running on lcore %d\n",
3213                         latencystats_lcore_id);
3214         }
3215 #endif
3216
3217         /* Setup bitrate stats */
3218 #ifdef RTE_LIBRTE_BITRATE
3219         if (bitrate_enabled != 0) {
3220                 bitrate_data = rte_stats_bitrate_create();
3221                 if (bitrate_data == NULL)
3222                         rte_exit(EXIT_FAILURE,
3223                                 "Could not allocate bitrate data.\n");
3224                 rte_stats_bitrate_reg(bitrate_data);
3225         }
3226 #endif
3227
3228 #ifdef RTE_LIBRTE_CMDLINE
3229         if (strlen(cmdline_filename) != 0)
3230                 cmdline_read_from_file(cmdline_filename);
3231
3232         if (interactive == 1) {
3233                 if (auto_start) {
3234                         printf("Start automatic packet forwarding\n");
3235                         start_packet_forwarding(0);
3236                 }
3237                 prompt();
3238                 pmd_test_exit();
3239         } else
3240 #endif
3241         {
3242                 char c;
3243                 int rc;
3244
3245                 f_quit = 0;
3246
3247                 printf("No commandline core given, start packet forwarding\n");
3248                 start_packet_forwarding(tx_first);
3249                 if (stats_period != 0) {
3250                         uint64_t prev_time = 0, cur_time, diff_time = 0;
3251                         uint64_t timer_period;
3252
3253                         /* Convert to number of cycles */
3254                         timer_period = stats_period * rte_get_timer_hz();
3255
3256                         while (f_quit == 0) {
3257                                 cur_time = rte_get_timer_cycles();
3258                                 diff_time += cur_time - prev_time;
3259
3260                                 if (diff_time >= timer_period) {
3261                                         print_stats();
3262                                         /* Reset the timer */
3263                                         diff_time = 0;
3264                                 }
3265                                 /* Sleep to avoid unnecessary checks */
3266                                 prev_time = cur_time;
3267                                 sleep(1);
3268                         }
3269                 }
3270
3271                 printf("Press enter to exit\n");
3272                 rc = read(0, &c, 1);
3273                 pmd_test_exit();
3274                 if (rc < 0)
3275                         return 1;
3276         }
3277
3278         return 0;
3279 }