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 #ifndef RTE_EXEC_ENV_WINDOWS
  13 #include <sys/mman.h>
  14 #endif
  15 #include <sys/types.h>
  16 #include <errno.h>
  17 #include <stdbool.h>
  18
  19 #include <sys/queue.h>
  20 #include <sys/stat.h>
  21
  22 #include <stdint.h>
  23 #include <unistd.h>
  24 #include <inttypes.h>
  25
  26 #include <rte_common.h>
  27 #include <rte_errno.h>
  28 #include <rte_byteorder.h>
  29 #include <rte_log.h>
  30 #include <rte_debug.h>
  31 #include <rte_cycles.h>
  32 #include <rte_memory.h>
  33 #include <rte_memcpy.h>
  34 #include <rte_launch.h>
  35 #include <rte_eal.h>
  36 #include <rte_alarm.h>
  37 #include <rte_per_lcore.h>
  38 #include <rte_lcore.h>
  39 #include <rte_atomic.h>
  40 #include <rte_branch_prediction.h>
  41 #include <rte_mempool.h>
  42 #include <rte_malloc.h>
  43 #include <rte_mbuf.h>
  44 #include <rte_mbuf_pool_ops.h>
  45 #include <rte_interrupts.h>
  46 #include <rte_pci.h>
  47 #include <rte_ether.h>
  48 #include <rte_ethdev.h>
  49 #include <rte_dev.h>
  50 #include <rte_string_fns.h>
  51 #ifdef RTE_NET_IXGBE
  52 #include <rte_pmd_ixgbe.h>
  53 #endif
  54 #ifdef RTE_LIB_PDUMP
  55 #include <rte_pdump.h>
  56 #endif
  57 #include <rte_flow.h>
  58 #include <rte_metrics.h>
  59 #ifdef RTE_LIB_BITRATESTATS
  60 #include <rte_bitrate.h>
  61 #endif
  62 #ifdef RTE_LIB_LATENCYSTATS
  63 #include <rte_latencystats.h>
  64 #endif
  65 #ifdef RTE_EXEC_ENV_WINDOWS
  66 #include <process.h>
  67 #endif
  68
  69 #include "testpmd.h"
  70
  71 #ifndef MAP_HUGETLB
  72 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
  73 #define HUGE_FLAG (0x40000)
  74 #else
  75 #define HUGE_FLAG MAP_HUGETLB
  76 #endif
  77
  78 #ifndef MAP_HUGE_SHIFT
  79 /* older kernels (or FreeBSD) will not have this define */
  80 #define HUGE_SHIFT (26)
  81 #else
  82 #define HUGE_SHIFT MAP_HUGE_SHIFT
  83 #endif
  84
  85 #define EXTMEM_HEAP_NAME "extmem"
  86 #define EXTBUF_ZONE_SIZE RTE_PGSIZE_2M
  87
  88 uint16_t verbose_level = 0; /**< Silent by default. */
  89 int testpmd_logtype; /**< Log type for testpmd logs */
  90
  91 /* use main core for command line ? */
  92 uint8_t interactive = 0;
  93 uint8_t auto_start = 0;
  94 uint8_t tx_first;
  95 char cmdline_filename[PATH_MAX] = {0};
  96
  97 /*
  98  * NUMA support configuration.
  99  * When set, the NUMA support attempts to dispatch the allocation of the
 100  * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
 101  * probed ports among the CPU sockets 0 and 1.
 102  * Otherwise, all memory is allocated from CPU socket 0.
 103  */
 104 uint8_t numa_support = 1; /**< numa enabled by default */
 105
 106 /*
 107  * In UMA mode,all memory is allocated from socket 0 if --socket-num is
 108  * not configured.
 109  */
 110 uint8_t socket_num = UMA_NO_CONFIG;
 111
 112 /*
 113  * Select mempool allocation type:
 114  * - native: use regular DPDK memory
 115  * - anon: use regular DPDK memory to create mempool, but populate using
 116  *         anonymous memory (may not be IOVA-contiguous)
 117  * - xmem: use externally allocated hugepage memory
 118  */
 119 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
 120
 121 /*
 122  * Store specified sockets on which memory pool to be used by ports
 123  * is allocated.
 124  */
 125 uint8_t port_numa[RTE_MAX_ETHPORTS];
 126
 127 /*
 128  * Store specified sockets on which RX ring to be used by ports
 129  * is allocated.
 130  */
 131 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
 132
 133 /*
 134  * Store specified sockets on which TX ring to be used by ports
 135  * is allocated.
 136  */
 137 uint8_t txring_numa[RTE_MAX_ETHPORTS];
 138
 139 /*
 140  * Record the Ethernet address of peer target ports to which packets are
 141  * forwarded.
 142  * Must be instantiated with the ethernet addresses of peer traffic generator
 143  * ports.
 144  */
 145 struct rte_ether_addr peer_eth_addrs[RTE_MAX_ETHPORTS];
 146 portid_t nb_peer_eth_addrs = 0;
 147
 148 /*
 149  * Probed Target Environment.
 150  */
 151 struct rte_port *ports;        /**< For all probed ethernet ports. */
 152 portid_t nb_ports;             /**< Number of probed ethernet ports. */
 153 struct fwd_lcore **fwd_lcores; /**< For all probed logical cores. */
 154 lcoreid_t nb_lcores;           /**< Number of probed logical cores. */
 155
 156 portid_t ports_ids[RTE_MAX_ETHPORTS]; /**< Store all port ids. */
 157
 158 /*
 159  * Test Forwarding Configuration.
 160  *    nb_fwd_lcores <= nb_cfg_lcores <= nb_lcores
 161  *    nb_fwd_ports  <= nb_cfg_ports  <= nb_ports
 162  */
 163 lcoreid_t nb_cfg_lcores; /**< Number of configured logical cores. */
 164 lcoreid_t nb_fwd_lcores; /**< Number of forwarding logical cores. */
 165 portid_t  nb_cfg_ports;  /**< Number of configured ports. */
 166 portid_t  nb_fwd_ports;  /**< Number of forwarding ports. */
 167
 168 unsigned int fwd_lcores_cpuids[RTE_MAX_LCORE]; /**< CPU ids configuration. */
 169 portid_t fwd_ports_ids[RTE_MAX_ETHPORTS];      /**< Port ids configuration. */
 170
 171 struct fwd_stream **fwd_streams; /**< For each RX queue of each port. */
 172 streamid_t nb_fwd_streams;       /**< Is equal to (nb_ports * nb_rxq). */
 173
 174 /*
 175  * Forwarding engines.
 176  */
 177 struct fwd_engine * fwd_engines[] = {
 178         &io_fwd_engine,
 179         &mac_fwd_engine,
 180         &mac_swap_engine,
 181         &flow_gen_engine,
 182         &rx_only_engine,
 183         &tx_only_engine,
 184         &csum_fwd_engine,
 185         &icmp_echo_engine,
 186         &noisy_vnf_engine,
 187         &five_tuple_swap_fwd_engine,
 188 #ifdef RTE_LIBRTE_IEEE1588
 189         &ieee1588_fwd_engine,
 190 #endif
 191         NULL,
 192 };
 193
 194 struct rte_mempool *mempools[RTE_MAX_NUMA_NODES * MAX_SEGS_BUFFER_SPLIT];
 195 uint16_t mempool_flags;
 196
 197 struct fwd_config cur_fwd_config;
 198 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
 199 uint32_t retry_enabled;
 200 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
 201 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
 202
 203 uint32_t mbuf_data_size_n = 1; /* Number of specified mbuf sizes. */
 204 uint16_t mbuf_data_size[MAX_SEGS_BUFFER_SPLIT] = {
 205         DEFAULT_MBUF_DATA_SIZE
 206 }; /**< Mbuf data space size. */
 207 uint32_t param_total_num_mbufs = 0;  /**< number of mbufs in all pools - if
 208                                       * specified on command-line. */
 209 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
 210
 211 /*
 212  * In container, it cannot terminate the process which running with 'stats-period'
 213  * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
 214  */
 215 uint8_t f_quit;
 216
 217 /*
 218  * Configuration of packet segments used to scatter received packets
 219  * if some of split features is configured.
 220  */
 221 uint16_t rx_pkt_seg_lengths[MAX_SEGS_BUFFER_SPLIT];
 222 uint8_t  rx_pkt_nb_segs; /**< Number of segments to split */
 223 uint16_t rx_pkt_seg_offsets[MAX_SEGS_BUFFER_SPLIT];
 224 uint8_t  rx_pkt_nb_offs; /**< Number of specified offsets */
 225
 226 /*
 227  * Configuration of packet segments used by the "txonly" processing engine.
 228  */
 229 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
 230 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
 231         TXONLY_DEF_PACKET_LEN,
 232 };
 233 uint8_t  tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
 234
 235 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
 236 /**< Split policy for packets to TX. */
 237
 238 uint8_t txonly_multi_flow;
 239 /**< Whether multiple flows are generated in TXONLY mode. */
 240
 241 uint32_t tx_pkt_times_inter;
 242 /**< Timings for send scheduling in TXONLY mode, time between bursts. */
 243
 244 uint32_t tx_pkt_times_intra;
 245 /**< Timings for send scheduling in TXONLY mode, time between packets. */
 246
 247 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
 248 uint16_t nb_pkt_flowgen_clones; /**< Number of Tx packet clones to send in flowgen mode. */
 249 int nb_flows_flowgen = 1024; /**< Number of flows in flowgen mode. */
 250 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
 251
 252 /* current configuration is in DCB or not,0 means it is not in DCB mode */
 253 uint8_t dcb_config = 0;
 254
 255 /*
 256  * Configurable number of RX/TX queues.
 257  */
 258 queueid_t nb_hairpinq; /**< Number of hairpin queues per port. */
 259 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
 260 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
 261
 262 /*
 263  * Configurable number of RX/TX ring descriptors.
 264  * Defaults are supplied by drivers via ethdev.
 265  */
 266 #define RTE_TEST_RX_DESC_DEFAULT 0
 267 #define RTE_TEST_TX_DESC_DEFAULT 0
 268 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
 269 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
 270
 271 #define RTE_PMD_PARAM_UNSET -1
 272 /*
 273  * Configurable values of RX and TX ring threshold registers.
 274  */
 275
 276 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
 277 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
 278 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
 279
 280 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
 281 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
 282 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
 283
 284 /*
 285  * Configurable value of RX free threshold.
 286  */
 287 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
 288
 289 /*
 290  * Configurable value of RX drop enable.
 291  */
 292 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
 293
 294 /*
 295  * Configurable value of TX free threshold.
 296  */
 297 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
 298
 299 /*
 300  * Configurable value of TX RS bit threshold.
 301  */
 302 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
 303
 304 /*
 305  * Configurable value of buffered packets before sending.
 306  */
 307 uint16_t noisy_tx_sw_bufsz;
 308
 309 /*
 310  * Configurable value of packet buffer timeout.
 311  */
 312 uint16_t noisy_tx_sw_buf_flush_time;
 313
 314 /*
 315  * Configurable value for size of VNF internal memory area
 316  * used for simulating noisy neighbour behaviour
 317  */
 318 uint64_t noisy_lkup_mem_sz;
 319
 320 /*
 321  * Configurable value of number of random writes done in
 322  * VNF simulation memory area.
 323  */
 324 uint64_t noisy_lkup_num_writes;
 325
 326 /*
 327  * Configurable value of number of random reads done in
 328  * VNF simulation memory area.
 329  */
 330 uint64_t noisy_lkup_num_reads;
 331
 332 /*
 333  * Configurable value of number of random reads/writes done in
 334  * VNF simulation memory area.
 335  */
 336 uint64_t noisy_lkup_num_reads_writes;
 337
 338 /*
 339  * Receive Side Scaling (RSS) configuration.
 340  */
 341 uint64_t rss_hf = ETH_RSS_IP; /* RSS IP by default. */
 342
 343 /*
 344  * Port topology configuration
 345  */
 346 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
 347
 348 /*
 349  * Avoids to flush all the RX streams before starts forwarding.
 350  */
 351 uint8_t no_flush_rx = 0; /* flush by default */
 352
 353 /*
 354  * Flow API isolated mode.
 355  */
 356 uint8_t flow_isolate_all;
 357
 358 /*
 359  * Avoids to check link status when starting/stopping a port.
 360  */
 361 uint8_t no_link_check = 0; /* check by default */
 362
 363 /*
 364  * Don't automatically start all ports in interactive mode.
 365  */
 366 uint8_t no_device_start = 0;
 367
 368 /*
 369  * Enable link status change notification
 370  */
 371 uint8_t lsc_interrupt = 1; /* enabled by default */
 372
 373 /*
 374  * Enable device removal notification.
 375  */
 376 uint8_t rmv_interrupt = 1; /* enabled by default */
 377
 378 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
 379
 380 /* After attach, port setup is called on event or by iterator */
 381 bool setup_on_probe_event = true;
 382
 383 /* Clear ptypes on port initialization. */
 384 uint8_t clear_ptypes = true;
 385
 386 /* Hairpin ports configuration mode. */
 387 uint16_t hairpin_mode;
 388
 389 /* Pretty printing of ethdev events */
 390 static const char * const eth_event_desc[] = {
 391         [RTE_ETH_EVENT_UNKNOWN] = "unknown",
 392         [RTE_ETH_EVENT_INTR_LSC] = "link state change",
 393         [RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
 394         [RTE_ETH_EVENT_INTR_RESET] = "reset",
 395         [RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
 396         [RTE_ETH_EVENT_IPSEC] = "IPsec",
 397         [RTE_ETH_EVENT_MACSEC] = "MACsec",
 398         [RTE_ETH_EVENT_INTR_RMV] = "device removal",
 399         [RTE_ETH_EVENT_NEW] = "device probed",
 400         [RTE_ETH_EVENT_DESTROY] = "device released",
 401         [RTE_ETH_EVENT_FLOW_AGED] = "flow aged",
 402         [RTE_ETH_EVENT_MAX] = NULL,
 403 };
 404
 405 /*
 406  * Display or mask ether events
 407  * Default to all events except VF_MBOX
 408  */
 409 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
 410                             (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
 411                             (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
 412                             (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
 413                             (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
 414                             (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
 415                             (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV) |
 416                             (UINT32_C(1) << RTE_ETH_EVENT_FLOW_AGED);
 417 /*
 418  * Decide if all memory are locked for performance.
 419  */
 420 int do_mlockall = 0;
 421
 422 /*
 423  * NIC bypass mode configuration options.
 424  */
 425
 426 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
 427 /* The NIC bypass watchdog timeout. */
 428 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
 429 #endif
 430
 431
 432 #ifdef RTE_LIB_LATENCYSTATS
 433
 434 /*
 435  * Set when latency stats is enabled in the commandline
 436  */
 437 uint8_t latencystats_enabled;
 438
 439 /*
 440  * Lcore ID to serive latency statistics.
 441  */
 442 lcoreid_t latencystats_lcore_id = -1;
 443
 444 #endif
 445
 446 /*
 447  * Ethernet device configuration.
 448  */
 449 struct rte_eth_rxmode rx_mode = {
 450         /* Default maximum frame length.
 451          * Zero is converted to "RTE_ETHER_MTU + PMD Ethernet overhead"
 452          * in init_config().
 453          */
 454         .max_rx_pkt_len = 0,
 455 };
 456
 457 struct rte_eth_txmode tx_mode = {
 458         .offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
 459 };
 460
 461 struct rte_fdir_conf fdir_conf = {
 462         .mode = RTE_FDIR_MODE_NONE,
 463         .pballoc = RTE_FDIR_PBALLOC_64K,
 464         .status = RTE_FDIR_REPORT_STATUS,
 465         .mask = {
 466                 .vlan_tci_mask = 0xFFEF,
 467                 .ipv4_mask     = {
 468                         .src_ip = 0xFFFFFFFF,
 469                         .dst_ip = 0xFFFFFFFF,
 470                 },
 471                 .ipv6_mask     = {
 472                         .src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
 473                         .dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
 474                 },
 475                 .src_port_mask = 0xFFFF,
 476                 .dst_port_mask = 0xFFFF,
 477                 .mac_addr_byte_mask = 0xFF,
 478                 .tunnel_type_mask = 1,
 479                 .tunnel_id_mask = 0xFFFFFFFF,
 480         },
 481         .drop_queue = 127,
 482 };
 483
 484 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
 485
 486 /*
 487  * Display zero values by default for xstats
 488  */
 489 uint8_t xstats_hide_zero;
 490
 491 /*
 492  * Measure of CPU cycles disabled by default
 493  */
 494 uint8_t record_core_cycles;
 495
 496 /*
 497  * Display of RX and TX bursts disabled by default
 498  */
 499 uint8_t record_burst_stats;
 500
 501 unsigned int num_sockets = 0;
 502 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
 503
 504 #ifdef RTE_LIB_BITRATESTATS
 505 /* Bitrate statistics */
 506 struct rte_stats_bitrates *bitrate_data;
 507 lcoreid_t bitrate_lcore_id;
 508 uint8_t bitrate_enabled;
 509 #endif
 510
 511 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
 512 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
 513
 514 /*
 515  * hexadecimal bitmask of RX mq mode can be enabled.
 516  */
 517 enum rte_eth_rx_mq_mode rx_mq_mode = ETH_MQ_RX_VMDQ_DCB_RSS;
 518
 519 /*
 520  * Used to set forced link speed
 521  */
 522 uint32_t eth_link_speed;
 523
 524 /*
 525  * ID of the current process in multi-process, used to
 526  * configure the queues to be polled.
 527  */
 528 int proc_id;
 529
 530 /*
 531  * Number of processes in multi-process, used to
 532  * configure the queues to be polled.
 533  */
 534 unsigned int num_procs = 1;
 535
 536 static int
 537 eth_dev_configure_mp(uint16_t port_id, uint16_t nb_rx_q, uint16_t nb_tx_q,
 538                       const struct rte_eth_conf *dev_conf)
 539 {
 540         if (is_proc_primary())
 541                 return rte_eth_dev_configure(port_id, nb_rx_q, nb_tx_q,
 542                                         dev_conf);
 543         return 0;
 544 }
 545
 546 static int
 547 eth_dev_start_mp(uint16_t port_id)
 548 {
 549         if (is_proc_primary())
 550                 return rte_eth_dev_start(port_id);
 551
 552         return 0;
 553 }
 554
 555 static int
 556 eth_dev_stop_mp(uint16_t port_id)
 557 {
 558         if (is_proc_primary())
 559                 return rte_eth_dev_stop(port_id);
 560
 561         return 0;
 562 }
 563
 564 static void
 565 mempool_free_mp(struct rte_mempool *mp)
 566 {
 567         if (is_proc_primary())
 568                 rte_mempool_free(mp);
 569 }
 570
 571 static int
 572 eth_dev_set_mtu_mp(uint16_t port_id, uint16_t mtu)
 573 {
 574         if (is_proc_primary())
 575                 return rte_eth_dev_set_mtu(port_id, mtu);
 576
 577         return 0;
 578 }
 579
 580 /* Forward function declarations */
 581 static void setup_attached_port(portid_t pi);
 582 static void check_all_ports_link_status(uint32_t port_mask);
 583 static int eth_event_callback(portid_t port_id,
 584                               enum rte_eth_event_type type,
 585                               void *param, void *ret_param);
 586 static void dev_event_callback(const char *device_name,
 587                                 enum rte_dev_event_type type,
 588                                 void *param);
 589
 590 /*
 591  * Check if all the ports are started.
 592  * If yes, return positive value. If not, return zero.
 593  */
 594 static int all_ports_started(void);
 595
 596 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
 597 uint16_t gso_max_segment_size = RTE_ETHER_MAX_LEN - RTE_ETHER_CRC_LEN;
 598
 599 /* Holds the registered mbuf dynamic flags names. */
 600 char dynf_names[64][RTE_MBUF_DYN_NAMESIZE];
 601
 602 /*
 603  * Helper function to check if socket is already discovered.
 604  * If yes, return positive value. If not, return zero.
 605  */
 606 int
 607 new_socket_id(unsigned int socket_id)
 608 {
 609         unsigned int i;
 610
 611         for (i = 0; i < num_sockets; i++) {
 612                 if (socket_ids[i] == socket_id)
 613                         return 0;
 614         }
 615         return 1;
 616 }
 617
 618 /*
 619  * Setup default configuration.
 620  */
 621 static void
 622 set_default_fwd_lcores_config(void)
 623 {
 624         unsigned int i;
 625         unsigned int nb_lc;
 626         unsigned int sock_num;
 627
 628         nb_lc = 0;
 629         for (i = 0; i < RTE_MAX_LCORE; i++) {
 630                 if (!rte_lcore_is_enabled(i))
 631                         continue;
 632                 sock_num = rte_lcore_to_socket_id(i);
 633                 if (new_socket_id(sock_num)) {
 634                         if (num_sockets >= RTE_MAX_NUMA_NODES) {
 635                                 rte_exit(EXIT_FAILURE,
 636                                          "Total sockets greater than %u\n",
 637                                          RTE_MAX_NUMA_NODES);
 638                         }
 639                         socket_ids[num_sockets++] = sock_num;
 640                 }
 641                 if (i == rte_get_main_lcore())
 642                         continue;
 643                 fwd_lcores_cpuids[nb_lc++] = i;
 644         }
 645         nb_lcores = (lcoreid_t) nb_lc;
 646         nb_cfg_lcores = nb_lcores;
 647         nb_fwd_lcores = 1;
 648 }
 649
 650 static void
 651 set_def_peer_eth_addrs(void)
 652 {
 653         portid_t i;
 654
 655         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 656                 peer_eth_addrs[i].addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
 657                 peer_eth_addrs[i].addr_bytes[5] = i;
 658         }
 659 }
 660
 661 static void
 662 set_default_fwd_ports_config(void)
 663 {
 664         portid_t pt_id;
 665         int i = 0;
 666
 667         RTE_ETH_FOREACH_DEV(pt_id) {
 668                 fwd_ports_ids[i++] = pt_id;
 669
 670                 /* Update sockets info according to the attached device */
 671                 int socket_id = rte_eth_dev_socket_id(pt_id);
 672                 if (socket_id >= 0 && new_socket_id(socket_id)) {
 673                         if (num_sockets >= RTE_MAX_NUMA_NODES) {
 674                                 rte_exit(EXIT_FAILURE,
 675                                          "Total sockets greater than %u\n",
 676                                          RTE_MAX_NUMA_NODES);
 677                         }
 678                         socket_ids[num_sockets++] = socket_id;
 679                 }
 680         }
 681
 682         nb_cfg_ports = nb_ports;
 683         nb_fwd_ports = nb_ports;
 684 }
 685
 686 void
 687 set_def_fwd_config(void)
 688 {
 689         set_default_fwd_lcores_config();
 690         set_def_peer_eth_addrs();
 691         set_default_fwd_ports_config();
 692 }
 693
 694 #ifndef RTE_EXEC_ENV_WINDOWS
 695 /* extremely pessimistic estimation of memory required to create a mempool */
 696 static int
 697 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
 698 {
 699         unsigned int n_pages, mbuf_per_pg, leftover;
 700         uint64_t total_mem, mbuf_mem, obj_sz;
 701
 702         /* there is no good way to predict how much space the mempool will
 703          * occupy because it will allocate chunks on the fly, and some of those
 704          * will come from default DPDK memory while some will come from our
 705          * external memory, so just assume 128MB will be enough for everyone.
 706          */
 707         uint64_t hdr_mem = 128 << 20;
 708
 709         /* account for possible non-contiguousness */
 710         obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
 711         if (obj_sz > pgsz) {
 712                 TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
 713                 return -1;
 714         }
 715
 716         mbuf_per_pg = pgsz / obj_sz;
 717         leftover = (nb_mbufs % mbuf_per_pg) > 0;
 718         n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
 719
 720         mbuf_mem = n_pages * pgsz;
 721
 722         total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
 723
 724         if (total_mem > SIZE_MAX) {
 725                 TESTPMD_LOG(ERR, "Memory size too big\n");
 726                 return -1;
 727         }
 728         *out = (size_t)total_mem;
 729
 730         return 0;
 731 }
 732
 733 static int
 734 pagesz_flags(uint64_t page_sz)
 735 {
 736         /* as per mmap() manpage, all page sizes are log2 of page size
 737          * shifted by MAP_HUGE_SHIFT
 738          */
 739         int log2 = rte_log2_u64(page_sz);
 740
 741         return (log2 << HUGE_SHIFT);
 742 }
 743
 744 static void *
 745 alloc_mem(size_t memsz, size_t pgsz, bool huge)
 746 {
 747         void *addr;
 748         int flags;
 749
 750         /* allocate anonymous hugepages */
 751         flags = MAP_ANONYMOUS | MAP_PRIVATE;
 752         if (huge)
 753                 flags |= HUGE_FLAG | pagesz_flags(pgsz);
 754
 755         addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
 756         if (addr == MAP_FAILED)
 757                 return NULL;
 758
 759         return addr;
 760 }
 761
 762 struct extmem_param {
 763         void *addr;
 764         size_t len;
 765         size_t pgsz;
 766         rte_iova_t *iova_table;
 767         unsigned int iova_table_len;
 768 };
 769
 770 static int
 771 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
 772                 bool huge)
 773 {
 774         uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
 775                         RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
 776         unsigned int cur_page, n_pages, pgsz_idx;
 777         size_t mem_sz, cur_pgsz;
 778         rte_iova_t *iovas = NULL;
 779         void *addr;
 780         int ret;
 781
 782         for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
 783                 /* skip anything that is too big */
 784                 if (pgsizes[pgsz_idx] > SIZE_MAX)
 785                         continue;
 786
 787                 cur_pgsz = pgsizes[pgsz_idx];
 788
 789                 /* if we were told not to allocate hugepages, override */
 790                 if (!huge)
 791                         cur_pgsz = sysconf(_SC_PAGESIZE);
 792
 793                 ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
 794                 if (ret < 0) {
 795                         TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
 796                         return -1;
 797                 }
 798
 799                 /* allocate our memory */
 800                 addr = alloc_mem(mem_sz, cur_pgsz, huge);
 801
 802                 /* if we couldn't allocate memory with a specified page size,
 803                  * that doesn't mean we can't do it with other page sizes, so
 804                  * try another one.
 805                  */
 806                 if (addr == NULL)
 807                         continue;
 808
 809                 /* store IOVA addresses for every page in this memory area */
 810                 n_pages = mem_sz / cur_pgsz;
 811
 812                 iovas = malloc(sizeof(*iovas) * n_pages);
 813
 814                 if (iovas == NULL) {
 815                         TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
 816                         goto fail;
 817                 }
 818                 /* lock memory if it's not huge pages */
 819                 if (!huge)
 820                         mlock(addr, mem_sz);
 821
 822                 /* populate IOVA addresses */
 823                 for (cur_page = 0; cur_page < n_pages; cur_page++) {
 824                         rte_iova_t iova;
 825                         size_t offset;
 826                         void *cur;
 827
 828                         offset = cur_pgsz * cur_page;
 829                         cur = RTE_PTR_ADD(addr, offset);
 830
 831                         /* touch the page before getting its IOVA */
 832                         *(volatile char *)cur = 0;
 833
 834                         iova = rte_mem_virt2iova(cur);
 835
 836                         iovas[cur_page] = iova;
 837                 }
 838
 839                 break;
 840         }
 841         /* if we couldn't allocate anything */
 842         if (iovas == NULL)
 843                 return -1;
 844
 845         param->addr = addr;
 846         param->len = mem_sz;
 847         param->pgsz = cur_pgsz;
 848         param->iova_table = iovas;
 849         param->iova_table_len = n_pages;
 850
 851         return 0;
 852 fail:
 853         if (iovas)
 854                 free(iovas);
 855         if (addr)
 856                 munmap(addr, mem_sz);
 857
 858         return -1;
 859 }
 860
 861 static int
 862 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
 863 {
 864         struct extmem_param param;
 865         int socket_id, ret;
 866
 867         memset(&param, 0, sizeof(param));
 868
 869         /* check if our heap exists */
 870         socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
 871         if (socket_id < 0) {
 872                 /* create our heap */
 873                 ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
 874                 if (ret < 0) {
 875                         TESTPMD_LOG(ERR, "Cannot create heap\n");
 876                         return -1;
 877                 }
 878         }
 879
 880         ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
 881         if (ret < 0) {
 882                 TESTPMD_LOG(ERR, "Cannot create memory area\n");
 883                 return -1;
 884         }
 885
 886         /* we now have a valid memory area, so add it to heap */
 887         ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
 888                         param.addr, param.len, param.iova_table,
 889                         param.iova_table_len, param.pgsz);
 890
 891         /* when using VFIO, memory is automatically mapped for DMA by EAL */
 892
 893         /* not needed any more */
 894         free(param.iova_table);
 895
 896         if (ret < 0) {
 897                 TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
 898                 munmap(param.addr, param.len);
 899                 return -1;
 900         }
 901
 902         /* success */
 903
 904         TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
 905                         param.len >> 20);
 906
 907         return 0;
 908 }
 909 static void
 910 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
 911              struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
 912 {
 913         uint16_t pid = 0;
 914         int ret;
 915
 916         RTE_ETH_FOREACH_DEV(pid) {
 917                 struct rte_eth_dev_info dev_info;
 918
 919                 ret = eth_dev_info_get_print_err(pid, &dev_info);
 920                 if (ret != 0) {
 921                         TESTPMD_LOG(DEBUG,
 922                                     "unable to get device info for port %d on addr 0x%p,"
 923                                     "mempool unmapping will not be performed\n",
 924                                     pid, memhdr->addr);
 925                         continue;
 926                 }
 927
 928                 ret = rte_dev_dma_unmap(dev_info.device, memhdr->addr, 0, memhdr->len);
 929                 if (ret) {
 930                         TESTPMD_LOG(DEBUG,
 931                                     "unable to DMA unmap addr 0x%p "
 932                                     "for device %s\n",
 933                                     memhdr->addr, dev_info.device->name);
 934                 }
 935         }
 936         ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
 937         if (ret) {
 938                 TESTPMD_LOG(DEBUG,
 939                             "unable to un-register addr 0x%p\n", memhdr->addr);
 940         }
 941 }
 942
 943 static void
 944 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
 945            struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
 946 {
 947         uint16_t pid = 0;
 948         size_t page_size = sysconf(_SC_PAGESIZE);
 949         int ret;
 950
 951         ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
 952                                   page_size);
 953         if (ret) {
 954                 TESTPMD_LOG(DEBUG,
 955                             "unable to register addr 0x%p\n", memhdr->addr);
 956                 return;
 957         }
 958         RTE_ETH_FOREACH_DEV(pid) {
 959                 struct rte_eth_dev_info dev_info;
 960
 961                 ret = eth_dev_info_get_print_err(pid, &dev_info);
 962                 if (ret != 0) {
 963                         TESTPMD_LOG(DEBUG,
 964                                     "unable to get device info for port %d on addr 0x%p,"
 965                                     "mempool mapping will not be performed\n",
 966                                     pid, memhdr->addr);
 967                         continue;
 968                 }
 969                 ret = rte_dev_dma_map(dev_info.device, memhdr->addr, 0, memhdr->len);
 970                 if (ret) {
 971                         TESTPMD_LOG(DEBUG,
 972                                     "unable to DMA map addr 0x%p "
 973                                     "for device %s\n",
 974                                     memhdr->addr, dev_info.device->name);
 975                 }
 976         }
 977 }
 978 #endif
 979
 980 static unsigned int
 981 setup_extbuf(uint32_t nb_mbufs, uint16_t mbuf_sz, unsigned int socket_id,
 982             char *pool_name, struct rte_pktmbuf_extmem **ext_mem)
 983 {
 984         struct rte_pktmbuf_extmem *xmem;
 985         unsigned int ext_num, zone_num, elt_num;
 986         uint16_t elt_size;
 987
 988         elt_size = RTE_ALIGN_CEIL(mbuf_sz, RTE_CACHE_LINE_SIZE);
 989         elt_num = EXTBUF_ZONE_SIZE / elt_size;
 990         zone_num = (nb_mbufs + elt_num - 1) / elt_num;
 991
 992         xmem = malloc(sizeof(struct rte_pktmbuf_extmem) * zone_num);
 993         if (xmem == NULL) {
 994                 TESTPMD_LOG(ERR, "Cannot allocate memory for "
 995                                  "external buffer descriptors\n");
 996                 *ext_mem = NULL;
 997                 return 0;
 998         }
 999         for (ext_num = 0; ext_num < zone_num; ext_num++) {
1000                 struct rte_pktmbuf_extmem *xseg = xmem + ext_num;
1001                 const struct rte_memzone *mz;
1002                 char mz_name[RTE_MEMZONE_NAMESIZE];
1003                 int ret;
1004
1005                 ret = snprintf(mz_name, sizeof(mz_name),
1006                         RTE_MEMPOOL_MZ_FORMAT "_xb_%u", pool_name, ext_num);
1007                 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
1008                         errno = ENAMETOOLONG;
1009                         ext_num = 0;
1010                         break;
1011                 }
1012                 mz = rte_memzone_reserve_aligned(mz_name, EXTBUF_ZONE_SIZE,
1013                                                  socket_id,
1014                                                  RTE_MEMZONE_IOVA_CONTIG |
1015                                                  RTE_MEMZONE_1GB |
1016                                                  RTE_MEMZONE_SIZE_HINT_ONLY,
1017                                                  EXTBUF_ZONE_SIZE);
1018                 if (mz == NULL) {
1019                         /*
1020                          * The caller exits on external buffer creation
1021                          * error, so there is no need to free memzones.
1022                          */
1023                         errno = ENOMEM;
1024                         ext_num = 0;
1025                         break;
1026                 }
1027                 xseg->buf_ptr = mz->addr;
1028                 xseg->buf_iova = mz->iova;
1029                 xseg->buf_len = EXTBUF_ZONE_SIZE;
1030                 xseg->elt_size = elt_size;
1031         }
1032         if (ext_num == 0 && xmem != NULL) {
1033                 free(xmem);
1034                 xmem = NULL;
1035         }
1036         *ext_mem = xmem;
1037         return ext_num;
1038 }
1039
1040 /*
1041  * Configuration initialisation done once at init time.
1042  */
1043 static struct rte_mempool *
1044 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
1045                  unsigned int socket_id, uint16_t size_idx)
1046 {
1047         char pool_name[RTE_MEMPOOL_NAMESIZE];
1048         struct rte_mempool *rte_mp = NULL;
1049 #ifndef RTE_EXEC_ENV_WINDOWS
1050         uint32_t mb_size;
1051
1052         mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
1053 #endif
1054         mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name), size_idx);
1055         if (!is_proc_primary()) {
1056                 rte_mp = rte_mempool_lookup(pool_name);
1057                 if (rte_mp == NULL)
1058                         rte_exit(EXIT_FAILURE,
1059                                 "Get mbuf pool for socket %u failed: %s\n",
1060                                 socket_id, rte_strerror(rte_errno));
1061                 return rte_mp;
1062         }
1063
1064         TESTPMD_LOG(INFO,
1065                 "create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
1066                 pool_name, nb_mbuf, mbuf_seg_size, socket_id);
1067
1068         switch (mp_alloc_type) {
1069         case MP_ALLOC_NATIVE:
1070                 {
1071                         /* wrapper to rte_mempool_create() */
1072                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1073                                         rte_mbuf_best_mempool_ops());
1074                         rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1075                                 mb_mempool_cache, 0, mbuf_seg_size, socket_id);
1076                         break;
1077                 }
1078 #ifndef RTE_EXEC_ENV_WINDOWS
1079         case MP_ALLOC_ANON:
1080                 {
1081                         rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
1082                                 mb_size, (unsigned int) mb_mempool_cache,
1083                                 sizeof(struct rte_pktmbuf_pool_private),
1084                                 socket_id, mempool_flags);
1085                         if (rte_mp == NULL)
1086                                 goto err;
1087
1088                         if (rte_mempool_populate_anon(rte_mp) == 0) {
1089                                 rte_mempool_free(rte_mp);
1090                                 rte_mp = NULL;
1091                                 goto err;
1092                         }
1093                         rte_pktmbuf_pool_init(rte_mp, NULL);
1094                         rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
1095                         rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
1096                         break;
1097                 }
1098         case MP_ALLOC_XMEM:
1099         case MP_ALLOC_XMEM_HUGE:
1100                 {
1101                         int heap_socket;
1102                         bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
1103
1104                         if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
1105                                 rte_exit(EXIT_FAILURE, "Could not create external memory\n");
1106
1107                         heap_socket =
1108                                 rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
1109                         if (heap_socket < 0)
1110                                 rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
1111
1112                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1113                                         rte_mbuf_best_mempool_ops());
1114                         rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1115                                         mb_mempool_cache, 0, mbuf_seg_size,
1116                                         heap_socket);
1117                         break;
1118                 }
1119 #endif
1120         case MP_ALLOC_XBUF:
1121                 {
1122                         struct rte_pktmbuf_extmem *ext_mem;
1123                         unsigned int ext_num;
1124
1125                         ext_num = setup_extbuf(nb_mbuf, mbuf_seg_size,
1126                                                socket_id, pool_name, &ext_mem);
1127                         if (ext_num == 0)
1128                                 rte_exit(EXIT_FAILURE,
1129                                          "Can't create pinned data buffers\n");
1130
1131                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1132                                         rte_mbuf_best_mempool_ops());
1133                         rte_mp = rte_pktmbuf_pool_create_extbuf
1134                                         (pool_name, nb_mbuf, mb_mempool_cache,
1135                                          0, mbuf_seg_size, socket_id,
1136                                          ext_mem, ext_num);
1137                         free(ext_mem);
1138                         break;
1139                 }
1140         default:
1141                 {
1142                         rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
1143                 }
1144         }
1145
1146 #ifndef RTE_EXEC_ENV_WINDOWS
1147 err:
1148 #endif
1149         if (rte_mp == NULL) {
1150                 rte_exit(EXIT_FAILURE,
1151                         "Creation of mbuf pool for socket %u failed: %s\n",
1152                         socket_id, rte_strerror(rte_errno));
1153         } else if (verbose_level > 0) {
1154                 rte_mempool_dump(stdout, rte_mp);
1155         }
1156         return rte_mp;
1157 }
1158
1159 /*
1160  * Check given socket id is valid or not with NUMA mode,
1161  * if valid, return 0, else return -1
1162  */
1163 static int
1164 check_socket_id(const unsigned int socket_id)
1165 {
1166         static int warning_once = 0;
1167
1168         if (new_socket_id(socket_id)) {
1169                 if (!warning_once && numa_support)
1170                         fprintf(stderr,
1171                                 "Warning: NUMA should be configured manually by using --port-numa-config and --ring-numa-config parameters along with --numa.\n");
1172                 warning_once = 1;
1173                 return -1;
1174         }
1175         return 0;
1176 }
1177
1178 /*
1179  * Get the allowed maximum number of RX queues.
1180  * *pid return the port id which has minimal value of
1181  * max_rx_queues in all ports.
1182  */
1183 queueid_t
1184 get_allowed_max_nb_rxq(portid_t *pid)
1185 {
1186         queueid_t allowed_max_rxq = RTE_MAX_QUEUES_PER_PORT;
1187         bool max_rxq_valid = false;
1188         portid_t pi;
1189         struct rte_eth_dev_info dev_info;
1190
1191         RTE_ETH_FOREACH_DEV(pi) {
1192                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1193                         continue;
1194
1195                 max_rxq_valid = true;
1196                 if (dev_info.max_rx_queues < allowed_max_rxq) {
1197                         allowed_max_rxq = dev_info.max_rx_queues;
1198                         *pid = pi;
1199                 }
1200         }
1201         return max_rxq_valid ? allowed_max_rxq : 0;
1202 }
1203
1204 /*
1205  * Check input rxq is valid or not.
1206  * If input rxq is not greater than any of maximum number
1207  * of RX queues of all ports, it is valid.
1208  * if valid, return 0, else return -1
1209  */
1210 int
1211 check_nb_rxq(queueid_t rxq)
1212 {
1213         queueid_t allowed_max_rxq;
1214         portid_t pid = 0;
1215
1216         allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1217         if (rxq > allowed_max_rxq) {
1218                 fprintf(stderr,
1219                         "Fail: input rxq (%u) can't be greater than max_rx_queues (%u) of port %u\n",
1220                         rxq, allowed_max_rxq, pid);
1221                 return -1;
1222         }
1223         return 0;
1224 }
1225
1226 /*
1227  * Get the allowed maximum number of TX queues.
1228  * *pid return the port id which has minimal value of
1229  * max_tx_queues in all ports.
1230  */
1231 queueid_t
1232 get_allowed_max_nb_txq(portid_t *pid)
1233 {
1234         queueid_t allowed_max_txq = RTE_MAX_QUEUES_PER_PORT;
1235         bool max_txq_valid = false;
1236         portid_t pi;
1237         struct rte_eth_dev_info dev_info;
1238
1239         RTE_ETH_FOREACH_DEV(pi) {
1240                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1241                         continue;
1242
1243                 max_txq_valid = true;
1244                 if (dev_info.max_tx_queues < allowed_max_txq) {
1245                         allowed_max_txq = dev_info.max_tx_queues;
1246                         *pid = pi;
1247                 }
1248         }
1249         return max_txq_valid ? allowed_max_txq : 0;
1250 }
1251
1252 /*
1253  * Check input txq is valid or not.
1254  * If input txq is not greater than any of maximum number
1255  * of TX queues of all ports, it is valid.
1256  * if valid, return 0, else return -1
1257  */
1258 int
1259 check_nb_txq(queueid_t txq)
1260 {
1261         queueid_t allowed_max_txq;
1262         portid_t pid = 0;
1263
1264         allowed_max_txq = get_allowed_max_nb_txq(&pid);
1265         if (txq > allowed_max_txq) {
1266                 fprintf(stderr,
1267                         "Fail: input txq (%u) can't be greater than max_tx_queues (%u) of port %u\n",
1268                         txq, allowed_max_txq, pid);
1269                 return -1;
1270         }
1271         return 0;
1272 }
1273
1274 /*
1275  * Get the allowed maximum number of RXDs of every rx queue.
1276  * *pid return the port id which has minimal value of
1277  * max_rxd in all queues of all ports.
1278  */
1279 static uint16_t
1280 get_allowed_max_nb_rxd(portid_t *pid)
1281 {
1282         uint16_t allowed_max_rxd = UINT16_MAX;
1283         portid_t pi;
1284         struct rte_eth_dev_info dev_info;
1285
1286         RTE_ETH_FOREACH_DEV(pi) {
1287                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1288                         continue;
1289
1290                 if (dev_info.rx_desc_lim.nb_max < allowed_max_rxd) {
1291                         allowed_max_rxd = dev_info.rx_desc_lim.nb_max;
1292                         *pid = pi;
1293                 }
1294         }
1295         return allowed_max_rxd;
1296 }
1297
1298 /*
1299  * Get the allowed minimal number of RXDs of every rx queue.
1300  * *pid return the port id which has minimal value of
1301  * min_rxd in all queues of all ports.
1302  */
1303 static uint16_t
1304 get_allowed_min_nb_rxd(portid_t *pid)
1305 {
1306         uint16_t allowed_min_rxd = 0;
1307         portid_t pi;
1308         struct rte_eth_dev_info dev_info;
1309
1310         RTE_ETH_FOREACH_DEV(pi) {
1311                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1312                         continue;
1313
1314                 if (dev_info.rx_desc_lim.nb_min > allowed_min_rxd) {
1315                         allowed_min_rxd = dev_info.rx_desc_lim.nb_min;
1316                         *pid = pi;
1317                 }
1318         }
1319
1320         return allowed_min_rxd;
1321 }
1322
1323 /*
1324  * Check input rxd is valid or not.
1325  * If input rxd is not greater than any of maximum number
1326  * of RXDs of every Rx queues and is not less than any of
1327  * minimal number of RXDs of every Rx queues, it is valid.
1328  * if valid, return 0, else return -1
1329  */
1330 int
1331 check_nb_rxd(queueid_t rxd)
1332 {
1333         uint16_t allowed_max_rxd;
1334         uint16_t allowed_min_rxd;
1335         portid_t pid = 0;
1336
1337         allowed_max_rxd = get_allowed_max_nb_rxd(&pid);
1338         if (rxd > allowed_max_rxd) {
1339                 fprintf(stderr,
1340                         "Fail: input rxd (%u) can't be greater than max_rxds (%u) of port %u\n",
1341                         rxd, allowed_max_rxd, pid);
1342                 return -1;
1343         }
1344
1345         allowed_min_rxd = get_allowed_min_nb_rxd(&pid);
1346         if (rxd < allowed_min_rxd) {
1347                 fprintf(stderr,
1348                         "Fail: input rxd (%u) can't be less than min_rxds (%u) of port %u\n",
1349                         rxd, allowed_min_rxd, pid);
1350                 return -1;
1351         }
1352
1353         return 0;
1354 }
1355
1356 /*
1357  * Get the allowed maximum number of TXDs of every rx queues.
1358  * *pid return the port id which has minimal value of
1359  * max_txd in every tx queue.
1360  */
1361 static uint16_t
1362 get_allowed_max_nb_txd(portid_t *pid)
1363 {
1364         uint16_t allowed_max_txd = UINT16_MAX;
1365         portid_t pi;
1366         struct rte_eth_dev_info dev_info;
1367
1368         RTE_ETH_FOREACH_DEV(pi) {
1369                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1370                         continue;
1371
1372                 if (dev_info.tx_desc_lim.nb_max < allowed_max_txd) {
1373                         allowed_max_txd = dev_info.tx_desc_lim.nb_max;
1374                         *pid = pi;
1375                 }
1376         }
1377         return allowed_max_txd;
1378 }
1379
1380 /*
1381  * Get the allowed maximum number of TXDs of every tx queues.
1382  * *pid return the port id which has minimal value of
1383  * min_txd in every tx queue.
1384  */
1385 static uint16_t
1386 get_allowed_min_nb_txd(portid_t *pid)
1387 {
1388         uint16_t allowed_min_txd = 0;
1389         portid_t pi;
1390         struct rte_eth_dev_info dev_info;
1391
1392         RTE_ETH_FOREACH_DEV(pi) {
1393                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1394                         continue;
1395
1396                 if (dev_info.tx_desc_lim.nb_min > allowed_min_txd) {
1397                         allowed_min_txd = dev_info.tx_desc_lim.nb_min;
1398                         *pid = pi;
1399                 }
1400         }
1401
1402         return allowed_min_txd;
1403 }
1404
1405 /*
1406  * Check input txd is valid or not.
1407  * If input txd is not greater than any of maximum number
1408  * of TXDs of every Rx queues, it is valid.
1409  * if valid, return 0, else return -1
1410  */
1411 int
1412 check_nb_txd(queueid_t txd)
1413 {
1414         uint16_t allowed_max_txd;
1415         uint16_t allowed_min_txd;
1416         portid_t pid = 0;
1417
1418         allowed_max_txd = get_allowed_max_nb_txd(&pid);
1419         if (txd > allowed_max_txd) {
1420                 fprintf(stderr,
1421                         "Fail: input txd (%u) can't be greater than max_txds (%u) of port %u\n",
1422                         txd, allowed_max_txd, pid);
1423                 return -1;
1424         }
1425
1426         allowed_min_txd = get_allowed_min_nb_txd(&pid);
1427         if (txd < allowed_min_txd) {
1428                 fprintf(stderr,
1429                         "Fail: input txd (%u) can't be less than min_txds (%u) of port %u\n",
1430                         txd, allowed_min_txd, pid);
1431                 return -1;
1432         }
1433         return 0;
1434 }
1435
1436
1437 /*
1438  * Get the allowed maximum number of hairpin queues.
1439  * *pid return the port id which has minimal value of
1440  * max_hairpin_queues in all ports.
1441  */
1442 queueid_t
1443 get_allowed_max_nb_hairpinq(portid_t *pid)
1444 {
1445         queueid_t allowed_max_hairpinq = RTE_MAX_QUEUES_PER_PORT;
1446         portid_t pi;
1447         struct rte_eth_hairpin_cap cap;
1448
1449         RTE_ETH_FOREACH_DEV(pi) {
1450                 if (rte_eth_dev_hairpin_capability_get(pi, &cap) != 0) {
1451                         *pid = pi;
1452                         return 0;
1453                 }
1454                 if (cap.max_nb_queues < allowed_max_hairpinq) {
1455                         allowed_max_hairpinq = cap.max_nb_queues;
1456                         *pid = pi;
1457                 }
1458         }
1459         return allowed_max_hairpinq;
1460 }
1461
1462 /*
1463  * Check input hairpin is valid or not.
1464  * If input hairpin is not greater than any of maximum number
1465  * of hairpin queues of all ports, it is valid.
1466  * if valid, return 0, else return -1
1467  */
1468 int
1469 check_nb_hairpinq(queueid_t hairpinq)
1470 {
1471         queueid_t allowed_max_hairpinq;
1472         portid_t pid = 0;
1473
1474         allowed_max_hairpinq = get_allowed_max_nb_hairpinq(&pid);
1475         if (hairpinq > allowed_max_hairpinq) {
1476                 fprintf(stderr,
1477                         "Fail: input hairpin (%u) can't be greater than max_hairpin_queues (%u) of port %u\n",
1478                         hairpinq, allowed_max_hairpinq, pid);
1479                 return -1;
1480         }
1481         return 0;
1482 }
1483
1484 static void
1485 init_config_port_offloads(portid_t pid, uint32_t socket_id)
1486 {
1487         struct rte_port *port = &ports[pid];
1488         uint16_t data_size;
1489         int ret;
1490         int i;
1491
1492         port->dev_conf.txmode = tx_mode;
1493         port->dev_conf.rxmode = rx_mode;
1494
1495         ret = eth_dev_info_get_print_err(pid, &port->dev_info);
1496         if (ret != 0)
1497                 rte_exit(EXIT_FAILURE, "rte_eth_dev_info_get() failed\n");
1498
1499         ret = update_jumbo_frame_offload(pid);
1500         if (ret != 0)
1501                 fprintf(stderr,
1502                         "Updating jumbo frame offload failed for port %u\n",
1503                         pid);
1504
1505         if (!(port->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE))
1506                 port->dev_conf.txmode.offloads &=
1507                         ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1508
1509         /* Apply Rx offloads configuration */
1510         for (i = 0; i < port->dev_info.max_rx_queues; i++)
1511                 port->rx_conf[i].offloads = port->dev_conf.rxmode.offloads;
1512         /* Apply Tx offloads configuration */
1513         for (i = 0; i < port->dev_info.max_tx_queues; i++)
1514                 port->tx_conf[i].offloads = port->dev_conf.txmode.offloads;
1515
1516         if (eth_link_speed)
1517                 port->dev_conf.link_speeds = eth_link_speed;
1518
1519         /* set flag to initialize port/queue */
1520         port->need_reconfig = 1;
1521         port->need_reconfig_queues = 1;
1522         port->socket_id = socket_id;
1523         port->tx_metadata = 0;
1524
1525         /*
1526          * Check for maximum number of segments per MTU.
1527          * Accordingly update the mbuf data size.
1528          */
1529         if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
1530             port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
1531                 data_size = rx_mode.max_rx_pkt_len /
1532                         port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1533
1534                 if ((data_size + RTE_PKTMBUF_HEADROOM) > mbuf_data_size[0]) {
1535                         mbuf_data_size[0] = data_size + RTE_PKTMBUF_HEADROOM;
1536                         TESTPMD_LOG(WARNING,
1537                                     "Configured mbuf size of the first segment %hu\n",
1538                                     mbuf_data_size[0]);
1539                 }
1540         }
1541 }
1542
1543 static void
1544 init_config(void)
1545 {
1546         portid_t pid;
1547         struct rte_mempool *mbp;
1548         unsigned int nb_mbuf_per_pool;
1549         lcoreid_t  lc_id;
1550         struct rte_gro_param gro_param;
1551         uint32_t gso_types;
1552
1553         /* Configuration of logical cores. */
1554         fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1555                                 sizeof(struct fwd_lcore *) * nb_lcores,
1556                                 RTE_CACHE_LINE_SIZE);
1557         if (fwd_lcores == NULL) {
1558                 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1559                                                         "failed\n", nb_lcores);
1560         }
1561         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1562                 fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1563                                                sizeof(struct fwd_lcore),
1564                                                RTE_CACHE_LINE_SIZE);
1565                 if (fwd_lcores[lc_id] == NULL) {
1566                         rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1567                                                                 "failed\n");
1568                 }
1569                 fwd_lcores[lc_id]->cpuid_idx = lc_id;
1570         }
1571
1572         RTE_ETH_FOREACH_DEV(pid) {
1573                 uint32_t socket_id;
1574
1575                 if (numa_support) {
1576                         socket_id = port_numa[pid];
1577                         if (port_numa[pid] == NUMA_NO_CONFIG) {
1578                                 socket_id = rte_eth_dev_socket_id(pid);
1579
1580                                 /*
1581                                  * if socket_id is invalid,
1582                                  * set to the first available socket.
1583                                  */
1584                                 if (check_socket_id(socket_id) < 0)
1585                                         socket_id = socket_ids[0];
1586                         }
1587                 } else {
1588                         socket_id = (socket_num == UMA_NO_CONFIG) ?
1589                                     0 : socket_num;
1590                 }
1591                 /* Apply default TxRx configuration for all ports */
1592                 init_config_port_offloads(pid, socket_id);
1593         }
1594         /*
1595          * Create pools of mbuf.
1596          * If NUMA support is disabled, create a single pool of mbuf in
1597          * socket 0 memory by default.
1598          * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1599          *
1600          * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1601          * nb_txd can be configured at run time.
1602          */
1603         if (param_total_num_mbufs)
1604                 nb_mbuf_per_pool = param_total_num_mbufs;
1605         else {
1606                 nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1607                         (nb_lcores * mb_mempool_cache) +
1608                         RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1609                 nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1610         }
1611
1612         if (numa_support) {
1613                 uint8_t i, j;
1614
1615                 for (i = 0; i < num_sockets; i++)
1616                         for (j = 0; j < mbuf_data_size_n; j++)
1617                                 mempools[i * MAX_SEGS_BUFFER_SPLIT + j] =
1618                                         mbuf_pool_create(mbuf_data_size[j],
1619                                                           nb_mbuf_per_pool,
1620                                                           socket_ids[i], j);
1621         } else {
1622                 uint8_t i;
1623
1624                 for (i = 0; i < mbuf_data_size_n; i++)
1625                         mempools[i] = mbuf_pool_create
1626                                         (mbuf_data_size[i],
1627                                          nb_mbuf_per_pool,
1628                                          socket_num == UMA_NO_CONFIG ?
1629                                          0 : socket_num, i);
1630         }
1631
1632         init_port_config();
1633
1634         gso_types = DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1635                 DEV_TX_OFFLOAD_GRE_TNL_TSO | DEV_TX_OFFLOAD_UDP_TSO;
1636         /*
1637          * Records which Mbuf pool to use by each logical core, if needed.
1638          */
1639         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1640                 mbp = mbuf_pool_find(
1641                         rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]), 0);
1642
1643                 if (mbp == NULL)
1644                         mbp = mbuf_pool_find(0, 0);
1645                 fwd_lcores[lc_id]->mbp = mbp;
1646                 /* initialize GSO context */
1647                 fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1648                 fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1649                 fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1650                 fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1651                         RTE_ETHER_CRC_LEN;
1652                 fwd_lcores[lc_id]->gso_ctx.flag = 0;
1653         }
1654
1655         fwd_config_setup();
1656
1657         /* create a gro context for each lcore */
1658         gro_param.gro_types = RTE_GRO_TCP_IPV4;
1659         gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1660         gro_param.max_item_per_flow = MAX_PKT_BURST;
1661         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1662                 gro_param.socket_id = rte_lcore_to_socket_id(
1663                                 fwd_lcores_cpuids[lc_id]);
1664                 fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1665                 if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1666                         rte_exit(EXIT_FAILURE,
1667                                         "rte_gro_ctx_create() failed\n");
1668                 }
1669         }
1670 }
1671
1672
1673 void
1674 reconfig(portid_t new_port_id, unsigned socket_id)
1675 {
1676         /* Reconfiguration of Ethernet ports. */
1677         init_config_port_offloads(new_port_id, socket_id);
1678         init_port_config();
1679 }
1680
1681
1682 int
1683 init_fwd_streams(void)
1684 {
1685         portid_t pid;
1686         struct rte_port *port;
1687         streamid_t sm_id, nb_fwd_streams_new;
1688         queueid_t q;
1689
1690         /* set socket id according to numa or not */
1691         RTE_ETH_FOREACH_DEV(pid) {
1692                 port = &ports[pid];
1693                 if (nb_rxq > port->dev_info.max_rx_queues) {
1694                         fprintf(stderr,
1695                                 "Fail: nb_rxq(%d) is greater than max_rx_queues(%d)\n",
1696                                 nb_rxq, port->dev_info.max_rx_queues);
1697                         return -1;
1698                 }
1699                 if (nb_txq > port->dev_info.max_tx_queues) {
1700                         fprintf(stderr,
1701                                 "Fail: nb_txq(%d) is greater than max_tx_queues(%d)\n",
1702                                 nb_txq, port->dev_info.max_tx_queues);
1703                         return -1;
1704                 }
1705                 if (numa_support) {
1706                         if (port_numa[pid] != NUMA_NO_CONFIG)
1707                                 port->socket_id = port_numa[pid];
1708                         else {
1709                                 port->socket_id = rte_eth_dev_socket_id(pid);
1710
1711                                 /*
1712                                  * if socket_id is invalid,
1713                                  * set to the first available socket.
1714                                  */
1715                                 if (check_socket_id(port->socket_id) < 0)
1716                                         port->socket_id = socket_ids[0];
1717                         }
1718                 }
1719                 else {
1720                         if (socket_num == UMA_NO_CONFIG)
1721                                 port->socket_id = 0;
1722                         else
1723                                 port->socket_id = socket_num;
1724                 }
1725         }
1726
1727         q = RTE_MAX(nb_rxq, nb_txq);
1728         if (q == 0) {
1729                 fprintf(stderr,
1730                         "Fail: Cannot allocate fwd streams as number of queues is 0\n");
1731                 return -1;
1732         }
1733         nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1734         if (nb_fwd_streams_new == nb_fwd_streams)
1735                 return 0;
1736         /* clear the old */
1737         if (fwd_streams != NULL) {
1738                 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1739                         if (fwd_streams[sm_id] == NULL)
1740                                 continue;
1741                         rte_free(fwd_streams[sm_id]);
1742                         fwd_streams[sm_id] = NULL;
1743                 }
1744                 rte_free(fwd_streams);
1745                 fwd_streams = NULL;
1746         }
1747
1748         /* init new */
1749         nb_fwd_streams = nb_fwd_streams_new;
1750         if (nb_fwd_streams) {
1751                 fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1752                         sizeof(struct fwd_stream *) * nb_fwd_streams,
1753                         RTE_CACHE_LINE_SIZE);
1754                 if (fwd_streams == NULL)
1755                         rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1756                                  " (struct fwd_stream *)) failed\n",
1757                                  nb_fwd_streams);
1758
1759                 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1760                         fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1761                                 " struct fwd_stream", sizeof(struct fwd_stream),
1762                                 RTE_CACHE_LINE_SIZE);
1763                         if (fwd_streams[sm_id] == NULL)
1764                                 rte_exit(EXIT_FAILURE, "rte_zmalloc"
1765                                          "(struct fwd_stream) failed\n");
1766                 }
1767         }
1768
1769         return 0;
1770 }
1771
1772 static void
1773 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1774 {
1775         uint64_t total_burst, sburst;
1776         uint64_t nb_burst;
1777         uint64_t burst_stats[4];
1778         uint16_t pktnb_stats[4];
1779         uint16_t nb_pkt;
1780         int burst_percent[4], sburstp;
1781         int i;
1782
1783         /*
1784          * First compute the total number of packet bursts and the
1785          * two highest numbers of bursts of the same number of packets.
1786          */
1787         memset(&burst_stats, 0x0, sizeof(burst_stats));
1788         memset(&pktnb_stats, 0x0, sizeof(pktnb_stats));
1789
1790         /* Show stats for 0 burst size always */
1791         total_burst = pbs->pkt_burst_spread[0];
1792         burst_stats[0] = pbs->pkt_burst_spread[0];
1793         pktnb_stats[0] = 0;
1794
1795         /* Find the next 2 burst sizes with highest occurrences. */
1796         for (nb_pkt = 1; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
1797                 nb_burst = pbs->pkt_burst_spread[nb_pkt];
1798
1799                 if (nb_burst == 0)
1800                         continue;
1801
1802                 total_burst += nb_burst;
1803
1804                 if (nb_burst > burst_stats[1]) {
1805                         burst_stats[2] = burst_stats[1];
1806                         pktnb_stats[2] = pktnb_stats[1];
1807                         burst_stats[1] = nb_burst;
1808                         pktnb_stats[1] = nb_pkt;
1809                 } else if (nb_burst > burst_stats[2]) {
1810                         burst_stats[2] = nb_burst;
1811                         pktnb_stats[2] = nb_pkt;
1812                 }
1813         }
1814         if (total_burst == 0)
1815                 return;
1816
1817         printf("  %s-bursts : %"PRIu64" [", rx_tx, total_burst);
1818         for (i = 0, sburst = 0, sburstp = 0; i < 4; i++) {
1819                 if (i == 3) {
1820                         printf("%d%% of other]\n", 100 - sburstp);
1821                         return;
1822                 }
1823
1824                 sburst += burst_stats[i];
1825                 if (sburst == total_burst) {
1826                         printf("%d%% of %d pkts]\n",
1827                                 100 - sburstp, (int) pktnb_stats[i]);
1828                         return;
1829                 }
1830
1831                 burst_percent[i] =
1832                         (double)burst_stats[i] / total_burst * 100;
1833                 printf("%d%% of %d pkts + ",
1834                         burst_percent[i], (int) pktnb_stats[i]);
1835                 sburstp += burst_percent[i];
1836         }
1837 }
1838
1839 static void
1840 fwd_stream_stats_display(streamid_t stream_id)
1841 {
1842         struct fwd_stream *fs;
1843         static const char *fwd_top_stats_border = "-------";
1844
1845         fs = fwd_streams[stream_id];
1846         if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1847             (fs->fwd_dropped == 0))
1848                 return;
1849         printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1850                "TX Port=%2d/Queue=%2d %s\n",
1851                fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1852                fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1853         printf("  RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1854                " TX-dropped: %-14"PRIu64,
1855                fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1856
1857         /* if checksum mode */
1858         if (cur_fwd_eng == &csum_fwd_engine) {
1859                 printf("  RX- bad IP checksum: %-14"PRIu64
1860                        "  Rx- bad L4 checksum: %-14"PRIu64
1861                        " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1862                         fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1863                         fs->rx_bad_outer_l4_csum);
1864                 printf(" RX- bad outer IP checksum: %-14"PRIu64"\n",
1865                         fs->rx_bad_outer_ip_csum);
1866         } else {
1867                 printf("\n");
1868         }
1869
1870         if (record_burst_stats) {
1871                 pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1872                 pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1873         }
1874 }
1875
1876 void
1877 fwd_stats_display(void)
1878 {
1879         static const char *fwd_stats_border = "----------------------";
1880         static const char *acc_stats_border = "+++++++++++++++";
1881         struct {
1882                 struct fwd_stream *rx_stream;
1883                 struct fwd_stream *tx_stream;
1884                 uint64_t tx_dropped;
1885                 uint64_t rx_bad_ip_csum;
1886                 uint64_t rx_bad_l4_csum;
1887                 uint64_t rx_bad_outer_l4_csum;
1888                 uint64_t rx_bad_outer_ip_csum;
1889         } ports_stats[RTE_MAX_ETHPORTS];
1890         uint64_t total_rx_dropped = 0;
1891         uint64_t total_tx_dropped = 0;
1892         uint64_t total_rx_nombuf = 0;
1893         struct rte_eth_stats stats;
1894         uint64_t fwd_cycles = 0;
1895         uint64_t total_recv = 0;
1896         uint64_t total_xmit = 0;
1897         struct rte_port *port;
1898         streamid_t sm_id;
1899         portid_t pt_id;
1900         int i;
1901
1902         memset(ports_stats, 0, sizeof(ports_stats));
1903
1904         for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1905                 struct fwd_stream *fs = fwd_streams[sm_id];
1906
1907                 if (cur_fwd_config.nb_fwd_streams >
1908                     cur_fwd_config.nb_fwd_ports) {
1909                         fwd_stream_stats_display(sm_id);
1910                 } else {
1911                         ports_stats[fs->tx_port].tx_stream = fs;
1912                         ports_stats[fs->rx_port].rx_stream = fs;
1913                 }
1914
1915                 ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
1916
1917                 ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
1918                 ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
1919                 ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
1920                                 fs->rx_bad_outer_l4_csum;
1921                 ports_stats[fs->rx_port].rx_bad_outer_ip_csum +=
1922                                 fs->rx_bad_outer_ip_csum;
1923
1924                 if (record_core_cycles)
1925                         fwd_cycles += fs->core_cycles;
1926         }
1927         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
1928                 pt_id = fwd_ports_ids[i];
1929                 port = &ports[pt_id];
1930
1931                 rte_eth_stats_get(pt_id, &stats);
1932                 stats.ipackets -= port->stats.ipackets;
1933                 stats.opackets -= port->stats.opackets;
1934                 stats.ibytes -= port->stats.ibytes;
1935                 stats.obytes -= port->stats.obytes;
1936                 stats.imissed -= port->stats.imissed;
1937                 stats.oerrors -= port->stats.oerrors;
1938                 stats.rx_nombuf -= port->stats.rx_nombuf;
1939
1940                 total_recv += stats.ipackets;
1941                 total_xmit += stats.opackets;
1942                 total_rx_dropped += stats.imissed;
1943                 total_tx_dropped += ports_stats[pt_id].tx_dropped;
1944                 total_tx_dropped += stats.oerrors;
1945                 total_rx_nombuf  += stats.rx_nombuf;
1946
1947                 printf("\n  %s Forward statistics for port %-2d %s\n",
1948                        fwd_stats_border, pt_id, fwd_stats_border);
1949
1950                 printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64
1951                        "RX-total: %-"PRIu64"\n", stats.ipackets, stats.imissed,
1952                        stats.ipackets + stats.imissed);
1953
1954                 if (cur_fwd_eng == &csum_fwd_engine) {
1955                         printf("  Bad-ipcsum: %-14"PRIu64
1956                                " Bad-l4csum: %-14"PRIu64
1957                                "Bad-outer-l4csum: %-14"PRIu64"\n",
1958                                ports_stats[pt_id].rx_bad_ip_csum,
1959                                ports_stats[pt_id].rx_bad_l4_csum,
1960                                ports_stats[pt_id].rx_bad_outer_l4_csum);
1961                         printf("  Bad-outer-ipcsum: %-14"PRIu64"\n",
1962                                ports_stats[pt_id].rx_bad_outer_ip_csum);
1963                 }
1964                 if (stats.ierrors + stats.rx_nombuf > 0) {
1965                         printf("  RX-error: %-"PRIu64"\n", stats.ierrors);
1966                         printf("  RX-nombufs: %-14"PRIu64"\n", stats.rx_nombuf);
1967                 }
1968
1969                 printf("  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64
1970                        "TX-total: %-"PRIu64"\n",
1971                        stats.opackets, ports_stats[pt_id].tx_dropped,
1972                        stats.opackets + ports_stats[pt_id].tx_dropped);
1973
1974                 if (record_burst_stats) {
1975                         if (ports_stats[pt_id].rx_stream)
1976                                 pkt_burst_stats_display("RX",
1977                                         &ports_stats[pt_id].rx_stream->rx_burst_stats);
1978                         if (ports_stats[pt_id].tx_stream)
1979                                 pkt_burst_stats_display("TX",
1980                                 &ports_stats[pt_id].tx_stream->tx_burst_stats);
1981                 }
1982
1983                 printf("  %s--------------------------------%s\n",
1984                        fwd_stats_border, fwd_stats_border);
1985         }
1986
1987         printf("\n  %s Accumulated forward statistics for all ports"
1988                "%s\n",
1989                acc_stats_border, acc_stats_border);
1990         printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
1991                "%-"PRIu64"\n"
1992                "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
1993                "%-"PRIu64"\n",
1994                total_recv, total_rx_dropped, total_recv + total_rx_dropped,
1995                total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
1996         if (total_rx_nombuf > 0)
1997                 printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
1998         printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
1999                "%s\n",
2000                acc_stats_border, acc_stats_border);
2001         if (record_core_cycles) {
2002 #define CYC_PER_MHZ 1E6
2003                 if (total_recv > 0 || total_xmit > 0) {
2004                         uint64_t total_pkts = 0;
2005                         if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 ||
2006                             strcmp(cur_fwd_eng->fwd_mode_name, "flowgen") == 0)
2007                                 total_pkts = total_xmit;
2008                         else
2009                                 total_pkts = total_recv;
2010
2011                         printf("\n  CPU cycles/packet=%.2F (total cycles="
2012                                "%"PRIu64" / total %s packets=%"PRIu64") at %"PRIu64
2013                                " MHz Clock\n",
2014                                (double) fwd_cycles / total_pkts,
2015                                fwd_cycles, cur_fwd_eng->fwd_mode_name, total_pkts,
2016                                (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
2017                 }
2018         }
2019 }
2020
2021 void
2022 fwd_stats_reset(void)
2023 {
2024         streamid_t sm_id;
2025         portid_t pt_id;
2026         int i;
2027
2028         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2029                 pt_id = fwd_ports_ids[i];
2030                 rte_eth_stats_get(pt_id, &ports[pt_id].stats);
2031         }
2032         for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
2033                 struct fwd_stream *fs = fwd_streams[sm_id];
2034
2035                 fs->rx_packets = 0;
2036                 fs->tx_packets = 0;
2037                 fs->fwd_dropped = 0;
2038                 fs->rx_bad_ip_csum = 0;
2039                 fs->rx_bad_l4_csum = 0;
2040                 fs->rx_bad_outer_l4_csum = 0;
2041                 fs->rx_bad_outer_ip_csum = 0;
2042
2043                 memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
2044                 memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
2045                 fs->core_cycles = 0;
2046         }
2047 }
2048
2049 static void
2050 flush_fwd_rx_queues(void)
2051 {
2052         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2053         portid_t  rxp;
2054         portid_t port_id;
2055         queueid_t rxq;
2056         uint16_t  nb_rx;
2057         uint16_t  i;
2058         uint8_t   j;
2059         uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
2060         uint64_t timer_period;
2061
2062         if (num_procs > 1) {
2063                 printf("multi-process not support for flushing fwd Rx queues, skip the below lines and return.\n");
2064                 return;
2065         }
2066
2067         /* convert to number of cycles */
2068         timer_period = rte_get_timer_hz(); /* 1 second timeout */
2069
2070         for (j = 0; j < 2; j++) {
2071                 for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
2072                         for (rxq = 0; rxq < nb_rxq; rxq++) {
2073                                 port_id = fwd_ports_ids[rxp];
2074                                 /**
2075                                 * testpmd can stuck in the below do while loop
2076                                 * if rte_eth_rx_burst() always returns nonzero
2077                                 * packets. So timer is added to exit this loop
2078                                 * after 1sec timer expiry.
2079                                 */
2080                                 prev_tsc = rte_rdtsc();
2081                                 do {
2082                                         nb_rx = rte_eth_rx_burst(port_id, rxq,
2083                                                 pkts_burst, MAX_PKT_BURST);
2084                                         for (i = 0; i < nb_rx; i++)
2085                                                 rte_pktmbuf_free(pkts_burst[i]);
2086
2087                                         cur_tsc = rte_rdtsc();
2088                                         diff_tsc = cur_tsc - prev_tsc;
2089                                         timer_tsc += diff_tsc;
2090                                 } while ((nb_rx > 0) &&
2091                                         (timer_tsc < timer_period));
2092                                 timer_tsc = 0;
2093                         }
2094                 }
2095                 rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
2096         }
2097 }
2098
2099 static void
2100 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
2101 {
2102         struct fwd_stream **fsm;
2103         streamid_t nb_fs;
2104         streamid_t sm_id;
2105 #ifdef RTE_LIB_BITRATESTATS
2106         uint64_t tics_per_1sec;
2107         uint64_t tics_datum;
2108         uint64_t tics_current;
2109         uint16_t i, cnt_ports;
2110
2111         cnt_ports = nb_ports;
2112         tics_datum = rte_rdtsc();
2113         tics_per_1sec = rte_get_timer_hz();
2114 #endif
2115         fsm = &fwd_streams[fc->stream_idx];
2116         nb_fs = fc->stream_nb;
2117         do {
2118                 for (sm_id = 0; sm_id < nb_fs; sm_id++)
2119                         (*pkt_fwd)(fsm[sm_id]);
2120 #ifdef RTE_LIB_BITRATESTATS
2121                 if (bitrate_enabled != 0 &&
2122                                 bitrate_lcore_id == rte_lcore_id()) {
2123                         tics_current = rte_rdtsc();
2124                         if (tics_current - tics_datum >= tics_per_1sec) {
2125                                 /* Periodic bitrate calculation */
2126                                 for (i = 0; i < cnt_ports; i++)
2127                                         rte_stats_bitrate_calc(bitrate_data,
2128                                                 ports_ids[i]);
2129                                 tics_datum = tics_current;
2130                         }
2131                 }
2132 #endif
2133 #ifdef RTE_LIB_LATENCYSTATS
2134                 if (latencystats_enabled != 0 &&
2135                                 latencystats_lcore_id == rte_lcore_id())
2136                         rte_latencystats_update();
2137 #endif
2138
2139         } while (! fc->stopped);
2140 }
2141
2142 static int
2143 start_pkt_forward_on_core(void *fwd_arg)
2144 {
2145         run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
2146                              cur_fwd_config.fwd_eng->packet_fwd);
2147         return 0;
2148 }
2149
2150 /*
2151  * Run the TXONLY packet forwarding engine to send a single burst of packets.
2152  * Used to start communication flows in network loopback test configurations.
2153  */
2154 static int
2155 run_one_txonly_burst_on_core(void *fwd_arg)
2156 {
2157         struct fwd_lcore *fwd_lc;
2158         struct fwd_lcore tmp_lcore;
2159
2160         fwd_lc = (struct fwd_lcore *) fwd_arg;
2161         tmp_lcore = *fwd_lc;
2162         tmp_lcore.stopped = 1;
2163         run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
2164         return 0;
2165 }
2166
2167 /*
2168  * Launch packet forwarding:
2169  *     - Setup per-port forwarding context.
2170  *     - launch logical cores with their forwarding configuration.
2171  */
2172 static void
2173 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
2174 {
2175         port_fwd_begin_t port_fwd_begin;
2176         unsigned int i;
2177         unsigned int lc_id;
2178         int diag;
2179
2180         port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2181         if (port_fwd_begin != NULL) {
2182                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2183                         (*port_fwd_begin)(fwd_ports_ids[i]);
2184         }
2185         for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2186                 lc_id = fwd_lcores_cpuids[i];
2187                 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2188                         fwd_lcores[i]->stopped = 0;
2189                         diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2190                                                      fwd_lcores[i], lc_id);
2191                         if (diag != 0)
2192                                 fprintf(stderr,
2193                                         "launch lcore %u failed - diag=%d\n",
2194                                         lc_id, diag);
2195                 }
2196         }
2197 }
2198
2199 /*
2200  * Launch packet forwarding configuration.
2201  */
2202 void
2203 start_packet_forwarding(int with_tx_first)
2204 {
2205         port_fwd_begin_t port_fwd_begin;
2206         port_fwd_end_t  port_fwd_end;
2207         unsigned int i;
2208
2209         if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2210                 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2211
2212         if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2213                 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2214
2215         if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2216                 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2217                 (!nb_rxq || !nb_txq))
2218                 rte_exit(EXIT_FAILURE,
2219                         "Either rxq or txq are 0, cannot use %s fwd mode\n",
2220                         cur_fwd_eng->fwd_mode_name);
2221
2222         if (all_ports_started() == 0) {
2223                 fprintf(stderr, "Not all ports were started\n");
2224                 return;
2225         }
2226         if (test_done == 0) {
2227                 fprintf(stderr, "Packet forwarding already started\n");
2228                 return;
2229         }
2230         test_done = 0;
2231
2232         fwd_config_setup();
2233
2234         if(!no_flush_rx)
2235                 flush_fwd_rx_queues();
2236
2237         pkt_fwd_config_display(&cur_fwd_config);
2238         rxtx_config_display();
2239
2240         fwd_stats_reset();
2241         if (with_tx_first) {
2242                 port_fwd_begin = tx_only_engine.port_fwd_begin;
2243                 if (port_fwd_begin != NULL) {
2244                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2245                                 (*port_fwd_begin)(fwd_ports_ids[i]);
2246                 }
2247                 while (with_tx_first--) {
2248                         launch_packet_forwarding(
2249                                         run_one_txonly_burst_on_core);
2250                         rte_eal_mp_wait_lcore();
2251                 }
2252                 port_fwd_end = tx_only_engine.port_fwd_end;
2253                 if (port_fwd_end != NULL) {
2254                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2255                                 (*port_fwd_end)(fwd_ports_ids[i]);
2256                 }
2257         }
2258         launch_packet_forwarding(start_pkt_forward_on_core);
2259 }
2260
2261 void
2262 stop_packet_forwarding(void)
2263 {
2264         port_fwd_end_t port_fwd_end;
2265         lcoreid_t lc_id;
2266         portid_t pt_id;
2267         int i;
2268
2269         if (test_done) {
2270                 fprintf(stderr, "Packet forwarding not started\n");
2271                 return;
2272         }
2273         printf("Telling cores to stop...");
2274         for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2275                 fwd_lcores[lc_id]->stopped = 1;
2276         printf("\nWaiting for lcores to finish...\n");
2277         rte_eal_mp_wait_lcore();
2278         port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2279         if (port_fwd_end != NULL) {
2280                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2281                         pt_id = fwd_ports_ids[i];
2282                         (*port_fwd_end)(pt_id);
2283                 }
2284         }
2285
2286         fwd_stats_display();
2287
2288         printf("\nDone.\n");
2289         test_done = 1;
2290 }
2291
2292 void
2293 dev_set_link_up(portid_t pid)
2294 {
2295         if (rte_eth_dev_set_link_up(pid) < 0)
2296                 fprintf(stderr, "\nSet link up fail.\n");
2297 }
2298
2299 void
2300 dev_set_link_down(portid_t pid)
2301 {
2302         if (rte_eth_dev_set_link_down(pid) < 0)
2303                 fprintf(stderr, "\nSet link down fail.\n");
2304 }
2305
2306 static int
2307 all_ports_started(void)
2308 {
2309         portid_t pi;
2310         struct rte_port *port;
2311
2312         RTE_ETH_FOREACH_DEV(pi) {
2313                 port = &ports[pi];
2314                 /* Check if there is a port which is not started */
2315                 if ((port->port_status != RTE_PORT_STARTED) &&
2316                         (port->slave_flag == 0))
2317                         return 0;
2318         }
2319
2320         /* No port is not started */
2321         return 1;
2322 }
2323
2324 int
2325 port_is_stopped(portid_t port_id)
2326 {
2327         struct rte_port *port = &ports[port_id];
2328
2329         if ((port->port_status != RTE_PORT_STOPPED) &&
2330             (port->slave_flag == 0))
2331                 return 0;
2332         return 1;
2333 }
2334
2335 int
2336 all_ports_stopped(void)
2337 {
2338         portid_t pi;
2339
2340         RTE_ETH_FOREACH_DEV(pi) {
2341                 if (!port_is_stopped(pi))
2342                         return 0;
2343         }
2344
2345         return 1;
2346 }
2347
2348 int
2349 port_is_started(portid_t port_id)
2350 {
2351         if (port_id_is_invalid(port_id, ENABLED_WARN))
2352                 return 0;
2353
2354         if (ports[port_id].port_status != RTE_PORT_STARTED)
2355                 return 0;
2356
2357         return 1;
2358 }
2359
2360 /* Configure the Rx and Tx hairpin queues for the selected port. */
2361 static int
2362 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2363 {
2364         queueid_t qi;
2365         struct rte_eth_hairpin_conf hairpin_conf = {
2366                 .peer_count = 1,
2367         };
2368         int i;
2369         int diag;
2370         struct rte_port *port = &ports[pi];
2371         uint16_t peer_rx_port = pi;
2372         uint16_t peer_tx_port = pi;
2373         uint32_t manual = 1;
2374         uint32_t tx_exp = hairpin_mode & 0x10;
2375
2376         if (!(hairpin_mode & 0xf)) {
2377                 peer_rx_port = pi;
2378                 peer_tx_port = pi;
2379                 manual = 0;
2380         } else if (hairpin_mode & 0x1) {
2381                 peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2382                                                        RTE_ETH_DEV_NO_OWNER);
2383                 if (peer_tx_port >= RTE_MAX_ETHPORTS)
2384                         peer_tx_port = rte_eth_find_next_owned_by(0,
2385                                                 RTE_ETH_DEV_NO_OWNER);
2386                 if (p_pi != RTE_MAX_ETHPORTS) {
2387                         peer_rx_port = p_pi;
2388                 } else {
2389                         uint16_t next_pi;
2390
2391                         /* Last port will be the peer RX port of the first. */
2392                         RTE_ETH_FOREACH_DEV(next_pi)
2393                                 peer_rx_port = next_pi;
2394                 }
2395                 manual = 1;
2396         } else if (hairpin_mode & 0x2) {
2397                 if (cnt_pi & 0x1) {
2398                         peer_rx_port = p_pi;
2399                 } else {
2400                         peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2401                                                 RTE_ETH_DEV_NO_OWNER);
2402                         if (peer_rx_port >= RTE_MAX_ETHPORTS)
2403                                 peer_rx_port = pi;
2404                 }
2405                 peer_tx_port = peer_rx_port;
2406                 manual = 1;
2407         }
2408
2409         for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2410                 hairpin_conf.peers[0].port = peer_rx_port;
2411                 hairpin_conf.peers[0].queue = i + nb_rxq;
2412                 hairpin_conf.manual_bind = !!manual;
2413                 hairpin_conf.tx_explicit = !!tx_exp;
2414                 diag = rte_eth_tx_hairpin_queue_setup
2415                         (pi, qi, nb_txd, &hairpin_conf);
2416                 i++;
2417                 if (diag == 0)
2418                         continue;
2419
2420                 /* Fail to setup rx queue, return */
2421                 if (rte_atomic16_cmpset(&(port->port_status),
2422                                         RTE_PORT_HANDLING,
2423                                         RTE_PORT_STOPPED) == 0)
2424                         fprintf(stderr,
2425                                 "Port %d can not be set back to stopped\n", pi);
2426                 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2427                         pi);
2428                 /* try to reconfigure queues next time */
2429                 port->need_reconfig_queues = 1;
2430                 return -1;
2431         }
2432         for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2433                 hairpin_conf.peers[0].port = peer_tx_port;
2434                 hairpin_conf.peers[0].queue = i + nb_txq;
2435                 hairpin_conf.manual_bind = !!manual;
2436                 hairpin_conf.tx_explicit = !!tx_exp;
2437                 diag = rte_eth_rx_hairpin_queue_setup
2438                         (pi, qi, nb_rxd, &hairpin_conf);
2439                 i++;
2440                 if (diag == 0)
2441                         continue;
2442
2443                 /* Fail to setup rx queue, return */
2444                 if (rte_atomic16_cmpset(&(port->port_status),
2445                                         RTE_PORT_HANDLING,
2446                                         RTE_PORT_STOPPED) == 0)
2447                         fprintf(stderr,
2448                                 "Port %d can not be set back to stopped\n", pi);
2449                 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2450                         pi);
2451                 /* try to reconfigure queues next time */
2452                 port->need_reconfig_queues = 1;
2453                 return -1;
2454         }
2455         return 0;
2456 }
2457
2458 /* Configure the Rx with optional split. */
2459 int
2460 rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
2461                uint16_t nb_rx_desc, unsigned int socket_id,
2462                struct rte_eth_rxconf *rx_conf, struct rte_mempool *mp)
2463 {
2464         union rte_eth_rxseg rx_useg[MAX_SEGS_BUFFER_SPLIT] = {};
2465         unsigned int i, mp_n;
2466         int ret;
2467
2468         if (rx_pkt_nb_segs <= 1 ||
2469             (rx_conf->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) == 0) {
2470                 rx_conf->rx_seg = NULL;
2471                 rx_conf->rx_nseg = 0;
2472                 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id,
2473                                              nb_rx_desc, socket_id,
2474                                              rx_conf, mp);
2475                 return ret;
2476         }
2477         for (i = 0; i < rx_pkt_nb_segs; i++) {
2478                 struct rte_eth_rxseg_split *rx_seg = &rx_useg[i].split;
2479                 struct rte_mempool *mpx;
2480                 /*
2481                  * Use last valid pool for the segments with number
2482                  * exceeding the pool index.
2483                  */
2484                 mp_n = (i > mbuf_data_size_n) ? mbuf_data_size_n - 1 : i;
2485                 mpx = mbuf_pool_find(socket_id, mp_n);
2486                 /* Handle zero as mbuf data buffer size. */
2487                 rx_seg->length = rx_pkt_seg_lengths[i] ?
2488                                    rx_pkt_seg_lengths[i] :
2489                                    mbuf_data_size[mp_n];
2490                 rx_seg->offset = i < rx_pkt_nb_offs ?
2491                                    rx_pkt_seg_offsets[i] : 0;
2492                 rx_seg->mp = mpx ? mpx : mp;
2493         }
2494         rx_conf->rx_nseg = rx_pkt_nb_segs;
2495         rx_conf->rx_seg = rx_useg;
2496         ret = rte_eth_rx_queue_setup(port_id, rx_queue_id, nb_rx_desc,
2497                                     socket_id, rx_conf, NULL);
2498         rx_conf->rx_seg = NULL;
2499         rx_conf->rx_nseg = 0;
2500         return ret;
2501 }
2502
2503 int
2504 start_port(portid_t pid)
2505 {
2506         int diag, need_check_link_status = -1;
2507         portid_t pi;
2508         portid_t p_pi = RTE_MAX_ETHPORTS;
2509         portid_t pl[RTE_MAX_ETHPORTS];
2510         portid_t peer_pl[RTE_MAX_ETHPORTS];
2511         uint16_t cnt_pi = 0;
2512         uint16_t cfg_pi = 0;
2513         int peer_pi;
2514         queueid_t qi;
2515         struct rte_port *port;
2516         struct rte_eth_hairpin_cap cap;
2517
2518         if (port_id_is_invalid(pid, ENABLED_WARN))
2519                 return 0;
2520
2521         RTE_ETH_FOREACH_DEV(pi) {
2522                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2523                         continue;
2524
2525                 need_check_link_status = 0;
2526                 port = &ports[pi];
2527                 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2528                                                  RTE_PORT_HANDLING) == 0) {
2529                         fprintf(stderr, "Port %d is now not stopped\n", pi);
2530                         continue;
2531                 }
2532
2533                 if (port->need_reconfig > 0) {
2534                         port->need_reconfig = 0;
2535
2536                         if (flow_isolate_all) {
2537                                 int ret = port_flow_isolate(pi, 1);
2538                                 if (ret) {
2539                                         fprintf(stderr,
2540                                                 "Failed to apply isolated mode on port %d\n",
2541                                                 pi);
2542                                         return -1;
2543                                 }
2544                         }
2545                         configure_rxtx_dump_callbacks(0);
2546                         printf("Configuring Port %d (socket %u)\n", pi,
2547                                         port->socket_id);
2548                         if (nb_hairpinq > 0 &&
2549                             rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2550                                 fprintf(stderr,
2551                                         "Port %d doesn't support hairpin queues\n",
2552                                         pi);
2553                                 return -1;
2554                         }
2555                         /* configure port */
2556                         diag = eth_dev_configure_mp(pi, nb_rxq + nb_hairpinq,
2557                                                      nb_txq + nb_hairpinq,
2558                                                      &(port->dev_conf));
2559                         if (diag != 0) {
2560                                 if (rte_atomic16_cmpset(&(port->port_status),
2561                                 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2562                                         fprintf(stderr,
2563                                                 "Port %d can not be set back to stopped\n",
2564                                                 pi);
2565                                 fprintf(stderr, "Fail to configure port %d\n",
2566                                         pi);
2567                                 /* try to reconfigure port next time */
2568                                 port->need_reconfig = 1;
2569                                 return -1;
2570                         }
2571                 }
2572                 if (port->need_reconfig_queues > 0 && is_proc_primary()) {
2573                         port->need_reconfig_queues = 0;
2574                         /* setup tx queues */
2575                         for (qi = 0; qi < nb_txq; qi++) {
2576                                 if ((numa_support) &&
2577                                         (txring_numa[pi] != NUMA_NO_CONFIG))
2578                                         diag = rte_eth_tx_queue_setup(pi, qi,
2579                                                 port->nb_tx_desc[qi],
2580                                                 txring_numa[pi],
2581                                                 &(port->tx_conf[qi]));
2582                                 else
2583                                         diag = rte_eth_tx_queue_setup(pi, qi,
2584                                                 port->nb_tx_desc[qi],
2585                                                 port->socket_id,
2586                                                 &(port->tx_conf[qi]));
2587
2588                                 if (diag == 0)
2589                                         continue;
2590
2591                                 /* Fail to setup tx queue, return */
2592                                 if (rte_atomic16_cmpset(&(port->port_status),
2593                                                         RTE_PORT_HANDLING,
2594                                                         RTE_PORT_STOPPED) == 0)
2595                                         fprintf(stderr,
2596                                                 "Port %d can not be set back to stopped\n",
2597                                                 pi);
2598                                 fprintf(stderr,
2599                                         "Fail to configure port %d tx queues\n",
2600                                         pi);
2601                                 /* try to reconfigure queues next time */
2602                                 port->need_reconfig_queues = 1;
2603                                 return -1;
2604                         }
2605                         for (qi = 0; qi < nb_rxq; qi++) {
2606                                 /* setup rx queues */
2607                                 if ((numa_support) &&
2608                                         (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2609                                         struct rte_mempool * mp =
2610                                                 mbuf_pool_find
2611                                                         (rxring_numa[pi], 0);
2612                                         if (mp == NULL) {
2613                                                 fprintf(stderr,
2614                                                         "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2615                                                         rxring_numa[pi]);
2616                                                 return -1;
2617                                         }
2618
2619                                         diag = rx_queue_setup(pi, qi,
2620                                              port->nb_rx_desc[qi],
2621                                              rxring_numa[pi],
2622                                              &(port->rx_conf[qi]),
2623                                              mp);
2624                                 } else {
2625                                         struct rte_mempool *mp =
2626                                                 mbuf_pool_find
2627                                                         (port->socket_id, 0);
2628                                         if (mp == NULL) {
2629                                                 fprintf(stderr,
2630                                                         "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2631                                                         port->socket_id);
2632                                                 return -1;
2633                                         }
2634                                         diag = rx_queue_setup(pi, qi,
2635                                              port->nb_rx_desc[qi],
2636                                              port->socket_id,
2637                                              &(port->rx_conf[qi]),
2638                                              mp);
2639                                 }
2640                                 if (diag == 0)
2641                                         continue;
2642
2643                                 /* Fail to setup rx queue, return */
2644                                 if (rte_atomic16_cmpset(&(port->port_status),
2645                                                         RTE_PORT_HANDLING,
2646                                                         RTE_PORT_STOPPED) == 0)
2647                                         fprintf(stderr,
2648                                                 "Port %d can not be set back to stopped\n",
2649                                                 pi);
2650                                 fprintf(stderr,
2651                                         "Fail to configure port %d rx queues\n",
2652                                         pi);
2653                                 /* try to reconfigure queues next time */
2654                                 port->need_reconfig_queues = 1;
2655                                 return -1;
2656                         }
2657                         /* setup hairpin queues */
2658                         if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2659                                 return -1;
2660                 }
2661                 configure_rxtx_dump_callbacks(verbose_level);
2662                 if (clear_ptypes) {
2663                         diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2664                                         NULL, 0);
2665                         if (diag < 0)
2666                                 fprintf(stderr,
2667                                         "Port %d: Failed to disable Ptype parsing\n",
2668                                         pi);
2669                 }
2670
2671                 p_pi = pi;
2672                 cnt_pi++;
2673
2674                 /* start port */
2675                 diag = eth_dev_start_mp(pi);
2676                 if (diag < 0) {
2677                         fprintf(stderr, "Fail to start port %d: %s\n",
2678                                 pi, rte_strerror(-diag));
2679
2680                         /* Fail to setup rx queue, return */
2681                         if (rte_atomic16_cmpset(&(port->port_status),
2682                                 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2683                                 fprintf(stderr,
2684                                         "Port %d can not be set back to stopped\n",
2685                                         pi);
2686                         continue;
2687                 }
2688
2689                 if (rte_atomic16_cmpset(&(port->port_status),
2690                         RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2691                         fprintf(stderr, "Port %d can not be set into started\n",
2692                                 pi);
2693
2694                 if (eth_macaddr_get_print_err(pi, &port->eth_addr) == 0)
2695                         printf("Port %d: %02X:%02X:%02X:%02X:%02X:%02X\n", pi,
2696                                 port->eth_addr.addr_bytes[0],
2697                                 port->eth_addr.addr_bytes[1],
2698                                 port->eth_addr.addr_bytes[2],
2699                                 port->eth_addr.addr_bytes[3],
2700                                 port->eth_addr.addr_bytes[4],
2701                                 port->eth_addr.addr_bytes[5]);
2702
2703                 /* at least one port started, need checking link status */
2704                 need_check_link_status = 1;
2705
2706                 pl[cfg_pi++] = pi;
2707         }
2708
2709         if (need_check_link_status == 1 && !no_link_check)
2710                 check_all_ports_link_status(RTE_PORT_ALL);
2711         else if (need_check_link_status == 0)
2712                 fprintf(stderr, "Please stop the ports first\n");
2713
2714         if (hairpin_mode & 0xf) {
2715                 uint16_t i;
2716                 int j;
2717
2718                 /* bind all started hairpin ports */
2719                 for (i = 0; i < cfg_pi; i++) {
2720                         pi = pl[i];
2721                         /* bind current Tx to all peer Rx */
2722                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2723                                                         RTE_MAX_ETHPORTS, 1);
2724                         if (peer_pi < 0)
2725                                 return peer_pi;
2726                         for (j = 0; j < peer_pi; j++) {
2727                                 if (!port_is_started(peer_pl[j]))
2728                                         continue;
2729                                 diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2730                                 if (diag < 0) {
2731                                         fprintf(stderr,
2732                                                 "Error during binding hairpin Tx port %u to %u: %s\n",
2733                                                 pi, peer_pl[j],
2734                                                 rte_strerror(-diag));
2735                                         return -1;
2736                                 }
2737                         }
2738                         /* bind all peer Tx to current Rx */
2739                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2740                                                         RTE_MAX_ETHPORTS, 0);
2741                         if (peer_pi < 0)
2742                                 return peer_pi;
2743                         for (j = 0; j < peer_pi; j++) {
2744                                 if (!port_is_started(peer_pl[j]))
2745                                         continue;
2746                                 diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2747                                 if (diag < 0) {
2748                                         fprintf(stderr,
2749                                                 "Error during binding hairpin Tx port %u to %u: %s\n",
2750                                                 peer_pl[j], pi,
2751                                                 rte_strerror(-diag));
2752                                         return -1;
2753                                 }
2754                         }
2755                 }
2756         }
2757
2758         printf("Done\n");
2759         return 0;
2760 }
2761
2762 void
2763 stop_port(portid_t pid)
2764 {
2765         portid_t pi;
2766         struct rte_port *port;
2767         int need_check_link_status = 0;
2768         portid_t peer_pl[RTE_MAX_ETHPORTS];
2769         int peer_pi;
2770
2771         if (port_id_is_invalid(pid, ENABLED_WARN))
2772                 return;
2773
2774         printf("Stopping ports...\n");
2775
2776         RTE_ETH_FOREACH_DEV(pi) {
2777                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2778                         continue;
2779
2780                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2781                         fprintf(stderr,
2782                                 "Please remove port %d from forwarding configuration.\n",
2783                                 pi);
2784                         continue;
2785                 }
2786
2787                 if (port_is_bonding_slave(pi)) {
2788                         fprintf(stderr,
2789                                 "Please remove port %d from bonded device.\n",
2790                                 pi);
2791                         continue;
2792                 }
2793
2794                 port = &ports[pi];
2795                 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2796                                                 RTE_PORT_HANDLING) == 0)
2797                         continue;
2798
2799                 if (hairpin_mode & 0xf) {
2800                         int j;
2801
2802                         rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
2803                         /* unbind all peer Tx from current Rx */
2804                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2805                                                         RTE_MAX_ETHPORTS, 0);
2806                         if (peer_pi < 0)
2807                                 continue;
2808                         for (j = 0; j < peer_pi; j++) {
2809                                 if (!port_is_started(peer_pl[j]))
2810                                         continue;
2811                                 rte_eth_hairpin_unbind(peer_pl[j], pi);
2812                         }
2813                 }
2814
2815                 if (port->flow_list)
2816                         port_flow_flush(pi);
2817
2818                 if (eth_dev_stop_mp(pi) != 0)
2819                         RTE_LOG(ERR, EAL, "rte_eth_dev_stop failed for port %u\n",
2820                                 pi);
2821
2822                 if (rte_atomic16_cmpset(&(port->port_status),
2823                         RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2824                         fprintf(stderr, "Port %d can not be set into stopped\n",
2825                                 pi);
2826                 need_check_link_status = 1;
2827         }
2828         if (need_check_link_status && !no_link_check)
2829                 check_all_ports_link_status(RTE_PORT_ALL);
2830
2831         printf("Done\n");
2832 }
2833
2834 static void
2835 remove_invalid_ports_in(portid_t *array, portid_t *total)
2836 {
2837         portid_t i;
2838         portid_t new_total = 0;
2839
2840         for (i = 0; i < *total; i++)
2841                 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2842                         array[new_total] = array[i];
2843                         new_total++;
2844                 }
2845         *total = new_total;
2846 }
2847
2848 static void
2849 remove_invalid_ports(void)
2850 {
2851         remove_invalid_ports_in(ports_ids, &nb_ports);
2852         remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2853         nb_cfg_ports = nb_fwd_ports;
2854 }
2855
2856 void
2857 close_port(portid_t pid)
2858 {
2859         portid_t pi;
2860         struct rte_port *port;
2861
2862         if (port_id_is_invalid(pid, ENABLED_WARN))
2863                 return;
2864
2865         printf("Closing ports...\n");
2866
2867         RTE_ETH_FOREACH_DEV(pi) {
2868                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2869                         continue;
2870
2871                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2872                         fprintf(stderr,
2873                                 "Please remove port %d from forwarding configuration.\n",
2874                                 pi);
2875                         continue;
2876                 }
2877
2878                 if (port_is_bonding_slave(pi)) {
2879                         fprintf(stderr,
2880                                 "Please remove port %d from bonded device.\n",
2881                                 pi);
2882                         continue;
2883                 }
2884
2885                 port = &ports[pi];
2886                 if (rte_atomic16_cmpset(&(port->port_status),
2887                         RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2888                         fprintf(stderr, "Port %d is already closed\n", pi);
2889                         continue;
2890                 }
2891
2892                 if (is_proc_primary()) {
2893                         port_flow_flush(pi);
2894                         rte_eth_dev_close(pi);
2895                 }
2896         }
2897
2898         remove_invalid_ports();
2899         printf("Done\n");
2900 }
2901
2902 void
2903 reset_port(portid_t pid)
2904 {
2905         int diag;
2906         portid_t pi;
2907         struct rte_port *port;
2908
2909         if (port_id_is_invalid(pid, ENABLED_WARN))
2910                 return;
2911
2912         if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
2913                 (pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
2914                 fprintf(stderr,
2915                         "Can not reset port(s), please stop port(s) first.\n");
2916                 return;
2917         }
2918
2919         printf("Resetting ports...\n");
2920
2921         RTE_ETH_FOREACH_DEV(pi) {
2922                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2923                         continue;
2924
2925                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2926                         fprintf(stderr,
2927                                 "Please remove port %d from forwarding configuration.\n",
2928                                 pi);
2929                         continue;
2930                 }
2931
2932                 if (port_is_bonding_slave(pi)) {
2933                         fprintf(stderr,
2934                                 "Please remove port %d from bonded device.\n",
2935                                 pi);
2936                         continue;
2937                 }
2938
2939                 diag = rte_eth_dev_reset(pi);
2940                 if (diag == 0) {
2941                         port = &ports[pi];
2942                         port->need_reconfig = 1;
2943                         port->need_reconfig_queues = 1;
2944                 } else {
2945                         fprintf(stderr, "Failed to reset port %d. diag=%d\n",
2946                                 pi, diag);
2947                 }
2948         }
2949
2950         printf("Done\n");
2951 }
2952
2953 void
2954 attach_port(char *identifier)
2955 {
2956         portid_t pi;
2957         struct rte_dev_iterator iterator;
2958
2959         printf("Attaching a new port...\n");
2960
2961         if (identifier == NULL) {
2962                 fprintf(stderr, "Invalid parameters are specified\n");
2963                 return;
2964         }
2965
2966         if (rte_dev_probe(identifier) < 0) {
2967                 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2968                 return;
2969         }
2970
2971         /* first attach mode: event */
2972         if (setup_on_probe_event) {
2973                 /* new ports are detected on RTE_ETH_EVENT_NEW event */
2974                 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2975                         if (ports[pi].port_status == RTE_PORT_HANDLING &&
2976                                         ports[pi].need_setup != 0)
2977                                 setup_attached_port(pi);
2978                 return;
2979         }
2980
2981         /* second attach mode: iterator */
2982         RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2983                 /* setup ports matching the devargs used for probing */
2984                 if (port_is_forwarding(pi))
2985                         continue; /* port was already attached before */
2986                 setup_attached_port(pi);
2987         }
2988 }
2989
2990 static void
2991 setup_attached_port(portid_t pi)
2992 {
2993         unsigned int socket_id;
2994         int ret;
2995
2996         socket_id = (unsigned)rte_eth_dev_socket_id(pi);
2997         /* if socket_id is invalid, set to the first available socket. */
2998         if (check_socket_id(socket_id) < 0)
2999                 socket_id = socket_ids[0];
3000         reconfig(pi, socket_id);
3001         ret = rte_eth_promiscuous_enable(pi);
3002         if (ret != 0)
3003                 fprintf(stderr,
3004                         "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3005                         pi, rte_strerror(-ret));
3006
3007         ports_ids[nb_ports++] = pi;
3008         fwd_ports_ids[nb_fwd_ports++] = pi;
3009         nb_cfg_ports = nb_fwd_ports;
3010         ports[pi].need_setup = 0;
3011         ports[pi].port_status = RTE_PORT_STOPPED;
3012
3013         printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
3014         printf("Done\n");
3015 }
3016
3017 static void
3018 detach_device(struct rte_device *dev)
3019 {
3020         portid_t sibling;
3021
3022         if (dev == NULL) {
3023                 fprintf(stderr, "Device already removed\n");
3024                 return;
3025         }
3026
3027         printf("Removing a device...\n");
3028
3029         RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
3030                 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
3031                         if (ports[sibling].port_status != RTE_PORT_STOPPED) {
3032                                 fprintf(stderr, "Port %u not stopped\n",
3033                                         sibling);
3034                                 return;
3035                         }
3036                         port_flow_flush(sibling);
3037                 }
3038         }
3039
3040         if (rte_dev_remove(dev) < 0) {
3041                 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
3042                 return;
3043         }
3044         remove_invalid_ports();
3045
3046         printf("Device is detached\n");
3047         printf("Now total ports is %d\n", nb_ports);
3048         printf("Done\n");
3049         return;
3050 }
3051
3052 void
3053 detach_port_device(portid_t port_id)
3054 {
3055         int ret;
3056         struct rte_eth_dev_info dev_info;
3057
3058         if (port_id_is_invalid(port_id, ENABLED_WARN))
3059                 return;
3060
3061         if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3062                 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3063                         fprintf(stderr, "Port not stopped\n");
3064                         return;
3065                 }
3066                 fprintf(stderr, "Port was not closed\n");
3067         }
3068
3069         ret = eth_dev_info_get_print_err(port_id, &dev_info);
3070         if (ret != 0) {
3071                 TESTPMD_LOG(ERR,
3072                         "Failed to get device info for port %d, not detaching\n",
3073                         port_id);
3074                 return;
3075         }
3076         detach_device(dev_info.device);
3077 }
3078
3079 void
3080 detach_devargs(char *identifier)
3081 {
3082         struct rte_dev_iterator iterator;
3083         struct rte_devargs da;
3084         portid_t port_id;
3085
3086         printf("Removing a device...\n");
3087
3088         memset(&da, 0, sizeof(da));
3089         if (rte_devargs_parsef(&da, "%s", identifier)) {
3090                 fprintf(stderr, "cannot parse identifier\n");
3091                 return;
3092         }
3093
3094         RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3095                 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3096                         if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3097                                 fprintf(stderr, "Port %u not stopped\n",
3098                                         port_id);
3099                                 rte_eth_iterator_cleanup(&iterator);
3100                                 rte_devargs_reset(&da);
3101                                 return;
3102                         }
3103                         port_flow_flush(port_id);
3104                 }
3105         }
3106
3107         if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3108                 TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3109                             da.name, da.bus->name);
3110                 rte_devargs_reset(&da);
3111                 return;
3112         }
3113
3114         remove_invalid_ports();
3115
3116         printf("Device %s is detached\n", identifier);
3117         printf("Now total ports is %d\n", nb_ports);
3118         printf("Done\n");
3119         rte_devargs_reset(&da);
3120 }
3121
3122 void
3123 pmd_test_exit(void)
3124 {
3125         portid_t pt_id;
3126         unsigned int i;
3127         int ret;
3128
3129         if (test_done == 0)
3130                 stop_packet_forwarding();
3131
3132 #ifndef RTE_EXEC_ENV_WINDOWS
3133         for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3134                 if (mempools[i]) {
3135                         if (mp_alloc_type == MP_ALLOC_ANON)
3136                                 rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3137                                                      NULL);
3138                 }
3139         }
3140 #endif
3141         if (ports != NULL) {
3142                 no_link_check = 1;
3143                 RTE_ETH_FOREACH_DEV(pt_id) {
3144                         printf("\nStopping port %d...\n", pt_id);
3145                         fflush(stdout);
3146                         stop_port(pt_id);
3147                 }
3148                 RTE_ETH_FOREACH_DEV(pt_id) {
3149                         printf("\nShutting down port %d...\n", pt_id);
3150                         fflush(stdout);
3151                         close_port(pt_id);
3152                 }
3153         }
3154
3155         if (hot_plug) {
3156                 ret = rte_dev_event_monitor_stop();
3157                 if (ret) {
3158                         RTE_LOG(ERR, EAL,
3159                                 "fail to stop device event monitor.");
3160                         return;
3161                 }
3162
3163                 ret = rte_dev_event_callback_unregister(NULL,
3164                         dev_event_callback, NULL);
3165                 if (ret < 0) {
3166                         RTE_LOG(ERR, EAL,
3167                                 "fail to unregister device event callback.\n");
3168                         return;
3169                 }
3170
3171                 ret = rte_dev_hotplug_handle_disable();
3172                 if (ret) {
3173                         RTE_LOG(ERR, EAL,
3174                                 "fail to disable hotplug handling.\n");
3175                         return;
3176                 }
3177         }
3178         for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3179                 if (mempools[i])
3180                         mempool_free_mp(mempools[i]);
3181         }
3182
3183         printf("\nBye...\n");
3184 }
3185
3186 typedef void (*cmd_func_t)(void);
3187 struct pmd_test_command {
3188         const char *cmd_name;
3189         cmd_func_t cmd_func;
3190 };
3191
3192 /* Check the link status of all ports in up to 9s, and print them finally */
3193 static void
3194 check_all_ports_link_status(uint32_t port_mask)
3195 {
3196 #define CHECK_INTERVAL 100 /* 100ms */
3197 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3198         portid_t portid;
3199         uint8_t count, all_ports_up, print_flag = 0;
3200         struct rte_eth_link link;
3201         int ret;
3202         char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3203
3204         printf("Checking link statuses...\n");
3205         fflush(stdout);
3206         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3207                 all_ports_up = 1;
3208                 RTE_ETH_FOREACH_DEV(portid) {
3209                         if ((port_mask & (1 << portid)) == 0)
3210                                 continue;
3211                         memset(&link, 0, sizeof(link));
3212                         ret = rte_eth_link_get_nowait(portid, &link);
3213                         if (ret < 0) {
3214                                 all_ports_up = 0;
3215                                 if (print_flag == 1)
3216                                         fprintf(stderr,
3217                                                 "Port %u link get failed: %s\n",
3218                                                 portid, rte_strerror(-ret));
3219                                 continue;
3220                         }
3221                         /* print link status if flag set */
3222                         if (print_flag == 1) {
3223                                 rte_eth_link_to_str(link_status,
3224                                         sizeof(link_status), &link);
3225                                 printf("Port %d %s\n", portid, link_status);
3226                                 continue;
3227                         }
3228                         /* clear all_ports_up flag if any link down */
3229                         if (link.link_status == ETH_LINK_DOWN) {
3230                                 all_ports_up = 0;
3231                                 break;
3232                         }
3233                 }
3234                 /* after finally printing all link status, get out */
3235                 if (print_flag == 1)
3236                         break;
3237
3238                 if (all_ports_up == 0) {
3239                         fflush(stdout);
3240                         rte_delay_ms(CHECK_INTERVAL);
3241                 }
3242
3243                 /* set the print_flag if all ports up or timeout */
3244                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3245                         print_flag = 1;
3246                 }
3247
3248                 if (lsc_interrupt)
3249                         break;
3250         }
3251 }
3252
3253 static void
3254 rmv_port_callback(void *arg)
3255 {
3256         int need_to_start = 0;
3257         int org_no_link_check = no_link_check;
3258         portid_t port_id = (intptr_t)arg;
3259         struct rte_eth_dev_info dev_info;
3260         int ret;
3261
3262         RTE_ETH_VALID_PORTID_OR_RET(port_id);
3263
3264         if (!test_done && port_is_forwarding(port_id)) {
3265                 need_to_start = 1;
3266                 stop_packet_forwarding();
3267         }
3268         no_link_check = 1;
3269         stop_port(port_id);
3270         no_link_check = org_no_link_check;
3271
3272         ret = eth_dev_info_get_print_err(port_id, &dev_info);
3273         if (ret != 0)
3274                 TESTPMD_LOG(ERR,
3275                         "Failed to get device info for port %d, not detaching\n",
3276                         port_id);
3277         else {
3278                 struct rte_device *device = dev_info.device;
3279                 close_port(port_id);
3280                 detach_device(device); /* might be already removed or have more ports */
3281         }
3282         if (need_to_start)
3283                 start_packet_forwarding(0);
3284 }
3285
3286 /* This function is used by the interrupt thread */
3287 static int
3288 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3289                   void *ret_param)
3290 {
3291         RTE_SET_USED(param);
3292         RTE_SET_USED(ret_param);
3293
3294         if (type >= RTE_ETH_EVENT_MAX) {
3295                 fprintf(stderr,
3296                         "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3297                         port_id, __func__, type);
3298                 fflush(stderr);
3299         } else if (event_print_mask & (UINT32_C(1) << type)) {
3300                 printf("\nPort %" PRIu16 ": %s event\n", port_id,
3301                         eth_event_desc[type]);
3302                 fflush(stdout);
3303         }
3304
3305         switch (type) {
3306         case RTE_ETH_EVENT_NEW:
3307                 ports[port_id].need_setup = 1;
3308                 ports[port_id].port_status = RTE_PORT_HANDLING;
3309                 break;
3310         case RTE_ETH_EVENT_INTR_RMV:
3311                 if (port_id_is_invalid(port_id, DISABLED_WARN))
3312                         break;
3313                 if (rte_eal_alarm_set(100000,
3314                                 rmv_port_callback, (void *)(intptr_t)port_id))
3315                         fprintf(stderr,
3316                                 "Could not set up deferred device removal\n");
3317                 break;
3318         case RTE_ETH_EVENT_DESTROY:
3319                 ports[port_id].port_status = RTE_PORT_CLOSED;
3320                 printf("Port %u is closed\n", port_id);
3321                 break;
3322         default:
3323                 break;
3324         }
3325         return 0;
3326 }
3327
3328 static int
3329 register_eth_event_callback(void)
3330 {
3331         int ret;
3332         enum rte_eth_event_type event;
3333
3334         for (event = RTE_ETH_EVENT_UNKNOWN;
3335                         event < RTE_ETH_EVENT_MAX; event++) {
3336                 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3337                                 event,
3338                                 eth_event_callback,
3339                                 NULL);
3340                 if (ret != 0) {
3341                         TESTPMD_LOG(ERR, "Failed to register callback for "
3342                                         "%s event\n", eth_event_desc[event]);
3343                         return -1;
3344                 }
3345         }
3346
3347         return 0;
3348 }
3349
3350 /* This function is used by the interrupt thread */
3351 static void
3352 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3353                              __rte_unused void *arg)
3354 {
3355         uint16_t port_id;
3356         int ret;
3357
3358         if (type >= RTE_DEV_EVENT_MAX) {
3359                 fprintf(stderr, "%s called upon invalid event %d\n",
3360                         __func__, type);
3361                 fflush(stderr);
3362         }
3363
3364         switch (type) {
3365         case RTE_DEV_EVENT_REMOVE:
3366                 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3367                         device_name);
3368                 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3369                 if (ret) {
3370                         RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3371                                 device_name);
3372                         return;
3373                 }
3374                 /*
3375                  * Because the user's callback is invoked in eal interrupt
3376                  * callback, the interrupt callback need to be finished before
3377                  * it can be unregistered when detaching device. So finish
3378                  * callback soon and use a deferred removal to detach device
3379                  * is need. It is a workaround, once the device detaching be
3380                  * moved into the eal in the future, the deferred removal could
3381                  * be deleted.
3382                  */
3383                 if (rte_eal_alarm_set(100000,
3384                                 rmv_port_callback, (void *)(intptr_t)port_id))
3385                         RTE_LOG(ERR, EAL,
3386                                 "Could not set up deferred device removal\n");
3387                 break;
3388         case RTE_DEV_EVENT_ADD:
3389                 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3390                         device_name);
3391                 /* TODO: After finish kernel driver binding,
3392                  * begin to attach port.
3393                  */
3394                 break;
3395         default:
3396                 break;
3397         }
3398 }
3399
3400 static void
3401 rxtx_port_config(struct rte_port *port)
3402 {
3403         uint16_t qid;
3404         uint64_t offloads;
3405
3406         for (qid = 0; qid < nb_rxq; qid++) {
3407                 offloads = port->rx_conf[qid].offloads;
3408                 port->rx_conf[qid] = port->dev_info.default_rxconf;
3409                 if (offloads != 0)
3410                         port->rx_conf[qid].offloads = offloads;
3411
3412                 /* Check if any Rx parameters have been passed */
3413                 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3414                         port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3415
3416                 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3417                         port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3418
3419                 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3420                         port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3421
3422                 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3423                         port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3424
3425                 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3426                         port->rx_conf[qid].rx_drop_en = rx_drop_en;
3427
3428                 port->nb_rx_desc[qid] = nb_rxd;
3429         }
3430
3431         for (qid = 0; qid < nb_txq; qid++) {
3432                 offloads = port->tx_conf[qid].offloads;
3433                 port->tx_conf[qid] = port->dev_info.default_txconf;
3434                 if (offloads != 0)
3435                         port->tx_conf[qid].offloads = offloads;
3436
3437                 /* Check if any Tx parameters have been passed */
3438                 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3439                         port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3440
3441                 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3442                         port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3443
3444                 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3445                         port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3446
3447                 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3448                         port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3449
3450                 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3451                         port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3452
3453                 port->nb_tx_desc[qid] = nb_txd;
3454         }
3455 }
3456
3457 /*
3458  * Helper function to arrange max_rx_pktlen value and JUMBO_FRAME offload,
3459  * MTU is also aligned if JUMBO_FRAME offload is not set.
3460  *
3461  * port->dev_info should be set before calling this function.
3462  *
3463  * return 0 on success, negative on error
3464  */
3465 int
3466 update_jumbo_frame_offload(portid_t portid)
3467 {
3468         struct rte_port *port = &ports[portid];
3469         uint32_t eth_overhead;
3470         uint64_t rx_offloads;
3471         int ret;
3472         bool on;
3473
3474         /* Update the max_rx_pkt_len to have MTU as RTE_ETHER_MTU */
3475         if (port->dev_info.max_mtu != UINT16_MAX &&
3476             port->dev_info.max_rx_pktlen > port->dev_info.max_mtu)
3477                 eth_overhead = port->dev_info.max_rx_pktlen -
3478                                 port->dev_info.max_mtu;
3479         else
3480                 eth_overhead = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
3481
3482         rx_offloads = port->dev_conf.rxmode.offloads;
3483
3484         /* Default config value is 0 to use PMD specific overhead */
3485         if (port->dev_conf.rxmode.max_rx_pkt_len == 0)
3486                 port->dev_conf.rxmode.max_rx_pkt_len = RTE_ETHER_MTU + eth_overhead;
3487
3488         if (port->dev_conf.rxmode.max_rx_pkt_len <= RTE_ETHER_MTU + eth_overhead) {
3489                 rx_offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3490                 on = false;
3491         } else {
3492                 if ((port->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3493                         fprintf(stderr,
3494                                 "Frame size (%u) is not supported by port %u\n",
3495                                 port->dev_conf.rxmode.max_rx_pkt_len,
3496                                 portid);
3497                         return -1;
3498                 }
3499                 rx_offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3500                 on = true;
3501         }
3502
3503         if (rx_offloads != port->dev_conf.rxmode.offloads) {
3504                 uint16_t qid;
3505
3506                 port->dev_conf.rxmode.offloads = rx_offloads;
3507
3508                 /* Apply JUMBO_FRAME offload configuration to Rx queue(s) */
3509                 for (qid = 0; qid < port->dev_info.nb_rx_queues; qid++) {
3510                         if (on)
3511                                 port->rx_conf[qid].offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3512                         else
3513                                 port->rx_conf[qid].offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3514                 }
3515         }
3516
3517         /* If JUMBO_FRAME is set MTU conversion done by ethdev layer,
3518          * if unset do it here
3519          */
3520         if ((rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3521                 ret = eth_dev_set_mtu_mp(portid,
3522                                 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead);
3523                 if (ret)
3524                         fprintf(stderr,
3525                                 "Failed to set MTU to %u for port %u\n",
3526                                 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead,
3527                                 portid);
3528         }
3529
3530         return 0;
3531 }
3532
3533 void
3534 init_port_config(void)
3535 {
3536         portid_t pid;
3537         struct rte_port *port;
3538         int ret;
3539
3540         RTE_ETH_FOREACH_DEV(pid) {
3541                 port = &ports[pid];
3542                 port->dev_conf.fdir_conf = fdir_conf;
3543
3544                 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3545                 if (ret != 0)
3546                         return;
3547
3548                 if (nb_rxq > 1) {
3549                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3550                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3551                                 rss_hf & port->dev_info.flow_type_rss_offloads;
3552                 } else {
3553                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3554                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3555                 }
3556
3557                 if (port->dcb_flag == 0) {
3558                         if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
3559                                 port->dev_conf.rxmode.mq_mode =
3560                                         (enum rte_eth_rx_mq_mode)
3561                                                 (rx_mq_mode & ETH_MQ_RX_RSS);
3562                         else
3563                                 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
3564                 }
3565
3566                 rxtx_port_config(port);
3567
3568                 ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3569                 if (ret != 0)
3570                         return;
3571
3572 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
3573                 rte_pmd_ixgbe_bypass_init(pid);
3574 #endif
3575
3576                 if (lsc_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_LSC))
3577                         port->dev_conf.intr_conf.lsc = 1;
3578                 if (rmv_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_RMV))
3579                         port->dev_conf.intr_conf.rmv = 1;
3580         }
3581 }
3582
3583 void set_port_slave_flag(portid_t slave_pid)
3584 {
3585         struct rte_port *port;
3586
3587         port = &ports[slave_pid];
3588         port->slave_flag = 1;
3589 }
3590
3591 void clear_port_slave_flag(portid_t slave_pid)
3592 {
3593         struct rte_port *port;
3594
3595         port = &ports[slave_pid];
3596         port->slave_flag = 0;
3597 }
3598
3599 uint8_t port_is_bonding_slave(portid_t slave_pid)
3600 {
3601         struct rte_port *port;
3602         struct rte_eth_dev_info dev_info;
3603         int ret;
3604
3605         port = &ports[slave_pid];
3606         ret = eth_dev_info_get_print_err(slave_pid, &dev_info);
3607         if (ret != 0) {
3608                 TESTPMD_LOG(ERR,
3609                         "Failed to get device info for port id %d,"
3610                         "cannot determine if the port is a bonded slave",
3611                         slave_pid);
3612                 return 0;
3613         }
3614         if ((*dev_info.dev_flags & RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3615                 return 1;
3616         return 0;
3617 }
3618
3619 const uint16_t vlan_tags[] = {
3620                 0,  1,  2,  3,  4,  5,  6,  7,
3621                 8,  9, 10, 11,  12, 13, 14, 15,
3622                 16, 17, 18, 19, 20, 21, 22, 23,
3623                 24, 25, 26, 27, 28, 29, 30, 31
3624 };
3625
3626 static  int
3627 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3628                  enum dcb_mode_enable dcb_mode,
3629                  enum rte_eth_nb_tcs num_tcs,
3630                  uint8_t pfc_en)
3631 {
3632         uint8_t i;
3633         int32_t rc;
3634         struct rte_eth_rss_conf rss_conf;
3635
3636         /*
3637          * Builds up the correct configuration for dcb+vt based on the vlan tags array
3638          * given above, and the number of traffic classes available for use.
3639          */
3640         if (dcb_mode == DCB_VT_ENABLED) {
3641                 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3642                                 &eth_conf->rx_adv_conf.vmdq_dcb_conf;
3643                 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3644                                 &eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3645
3646                 /* VMDQ+DCB RX and TX configurations */
3647                 vmdq_rx_conf->enable_default_pool = 0;
3648                 vmdq_rx_conf->default_pool = 0;
3649                 vmdq_rx_conf->nb_queue_pools =
3650                         (num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3651                 vmdq_tx_conf->nb_queue_pools =
3652                         (num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3653
3654                 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3655                 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3656                         vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3657                         vmdq_rx_conf->pool_map[i].pools =
3658                                 1 << (i % vmdq_rx_conf->nb_queue_pools);
3659                 }
3660                 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3661                         vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3662                         vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3663                 }
3664
3665                 /* set DCB mode of RX and TX of multiple queues */
3666                 eth_conf->rxmode.mq_mode =
3667                                 (enum rte_eth_rx_mq_mode)
3668                                         (rx_mq_mode & ETH_MQ_RX_VMDQ_DCB);
3669                 eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3670         } else {
3671                 struct rte_eth_dcb_rx_conf *rx_conf =
3672                                 &eth_conf->rx_adv_conf.dcb_rx_conf;
3673                 struct rte_eth_dcb_tx_conf *tx_conf =
3674                                 &eth_conf->tx_adv_conf.dcb_tx_conf;
3675
3676                 memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3677
3678                 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3679                 if (rc != 0)
3680                         return rc;
3681
3682                 rx_conf->nb_tcs = num_tcs;
3683                 tx_conf->nb_tcs = num_tcs;
3684
3685                 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3686                         rx_conf->dcb_tc[i] = i % num_tcs;
3687                         tx_conf->dcb_tc[i] = i % num_tcs;
3688                 }
3689
3690                 eth_conf->rxmode.mq_mode =
3691                                 (enum rte_eth_rx_mq_mode)
3692                                         (rx_mq_mode & ETH_MQ_RX_DCB_RSS);
3693                 eth_conf->rx_adv_conf.rss_conf = rss_conf;
3694                 eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3695         }
3696
3697         if (pfc_en)
3698                 eth_conf->dcb_capability_en =
3699                                 ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3700         else
3701                 eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3702
3703         return 0;
3704 }
3705
3706 int
3707 init_port_dcb_config(portid_t pid,
3708                      enum dcb_mode_enable dcb_mode,
3709                      enum rte_eth_nb_tcs num_tcs,
3710                      uint8_t pfc_en)
3711 {
3712         struct rte_eth_conf port_conf;
3713         struct rte_port *rte_port;
3714         int retval;
3715         uint16_t i;
3716
3717         if (num_procs > 1) {
3718                 printf("The multi-process feature doesn't support dcb.\n");
3719                 return -ENOTSUP;
3720         }
3721         rte_port = &ports[pid];
3722
3723         memset(&port_conf, 0, sizeof(struct rte_eth_conf));
3724
3725         port_conf.rxmode = rte_port->dev_conf.rxmode;
3726         port_conf.txmode = rte_port->dev_conf.txmode;
3727
3728         /*set configuration of DCB in vt mode and DCB in non-vt mode*/
3729         retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3730         if (retval < 0)
3731                 return retval;
3732         port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3733
3734         /* re-configure the device . */
3735         retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3736         if (retval < 0)
3737                 return retval;
3738
3739         retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3740         if (retval != 0)
3741                 return retval;
3742
3743         /* If dev_info.vmdq_pool_base is greater than 0,
3744          * the queue id of vmdq pools is started after pf queues.
3745          */
3746         if (dcb_mode == DCB_VT_ENABLED &&
3747             rte_port->dev_info.vmdq_pool_base > 0) {
3748                 fprintf(stderr,
3749                         "VMDQ_DCB multi-queue mode is nonsensical for port %d.\n",
3750                         pid);
3751                 return -1;
3752         }
3753
3754         /* Assume the ports in testpmd have the same dcb capability
3755          * and has the same number of rxq and txq in dcb mode
3756          */
3757         if (dcb_mode == DCB_VT_ENABLED) {
3758                 if (rte_port->dev_info.max_vfs > 0) {
3759                         nb_rxq = rte_port->dev_info.nb_rx_queues;
3760                         nb_txq = rte_port->dev_info.nb_tx_queues;
3761                 } else {
3762                         nb_rxq = rte_port->dev_info.max_rx_queues;
3763                         nb_txq = rte_port->dev_info.max_tx_queues;
3764                 }
3765         } else {
3766                 /*if vt is disabled, use all pf queues */
3767                 if (rte_port->dev_info.vmdq_pool_base == 0) {
3768                         nb_rxq = rte_port->dev_info.max_rx_queues;
3769                         nb_txq = rte_port->dev_info.max_tx_queues;
3770                 } else {
3771                         nb_rxq = (queueid_t)num_tcs;
3772                         nb_txq = (queueid_t)num_tcs;
3773
3774                 }
3775         }
3776         rx_free_thresh = 64;
3777
3778         memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3779
3780         rxtx_port_config(rte_port);
3781         /* VLAN filter */
3782         rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3783         for (i = 0; i < RTE_DIM(vlan_tags); i++)
3784                 rx_vft_set(pid, vlan_tags[i], 1);
3785
3786         retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
3787         if (retval != 0)
3788                 return retval;
3789
3790         rte_port->dcb_flag = 1;
3791
3792         /* Enter DCB configuration status */
3793         dcb_config = 1;
3794
3795         return 0;
3796 }
3797
3798 static void
3799 init_port(void)
3800 {
3801         int i;
3802
3803         /* Configuration of Ethernet ports. */
3804         ports = rte_zmalloc("testpmd: ports",
3805                             sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3806                             RTE_CACHE_LINE_SIZE);
3807         if (ports == NULL) {
3808                 rte_exit(EXIT_FAILURE,
3809                                 "rte_zmalloc(%d struct rte_port) failed\n",
3810                                 RTE_MAX_ETHPORTS);
3811         }
3812         for (i = 0; i < RTE_MAX_ETHPORTS; i++)
3813                 LIST_INIT(&ports[i].flow_tunnel_list);
3814         /* Initialize ports NUMA structures */
3815         memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3816         memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3817         memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3818 }
3819
3820 static void
3821 force_quit(void)
3822 {
3823         pmd_test_exit();
3824         prompt_exit();
3825 }
3826
3827 static void
3828 print_stats(void)
3829 {
3830         uint8_t i;
3831         const char clr[] = { 27, '[', '2', 'J', '\0' };
3832         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3833
3834         /* Clear screen and move to top left */
3835         printf("%s%s", clr, top_left);
3836
3837         printf("\nPort statistics ====================================");
3838         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3839                 nic_stats_display(fwd_ports_ids[i]);
3840
3841         fflush(stdout);
3842 }
3843
3844 static void
3845 signal_handler(int signum)
3846 {
3847         if (signum == SIGINT || signum == SIGTERM) {
3848                 fprintf(stderr, "\nSignal %d received, preparing to exit...\n",
3849                         signum);
3850 #ifdef RTE_LIB_PDUMP
3851                 /* uninitialize packet capture framework */
3852                 rte_pdump_uninit();
3853 #endif
3854 #ifdef RTE_LIB_LATENCYSTATS
3855                 if (latencystats_enabled != 0)
3856                         rte_latencystats_uninit();
3857 #endif
3858                 force_quit();
3859                 /* Set flag to indicate the force termination. */
3860                 f_quit = 1;
3861                 /* exit with the expected status */
3862 #ifndef RTE_EXEC_ENV_WINDOWS
3863                 signal(signum, SIG_DFL);
3864                 kill(getpid(), signum);
3865 #endif
3866         }
3867 }
3868
3869 int
3870 main(int argc, char** argv)
3871 {
3872         int diag;
3873         portid_t port_id;
3874         uint16_t count;
3875         int ret;
3876
3877         signal(SIGINT, signal_handler);
3878         signal(SIGTERM, signal_handler);
3879
3880         testpmd_logtype = rte_log_register("testpmd");
3881         if (testpmd_logtype < 0)
3882                 rte_exit(EXIT_FAILURE, "Cannot register log type");
3883         rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3884
3885         diag = rte_eal_init(argc, argv);
3886         if (diag < 0)
3887                 rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
3888                          rte_strerror(rte_errno));
3889
3890         ret = register_eth_event_callback();
3891         if (ret != 0)
3892                 rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
3893
3894 #ifdef RTE_LIB_PDUMP
3895         /* initialize packet capture framework */
3896         rte_pdump_init();
3897 #endif
3898
3899         count = 0;
3900         RTE_ETH_FOREACH_DEV(port_id) {
3901                 ports_ids[count] = port_id;
3902                 count++;
3903         }
3904         nb_ports = (portid_t) count;
3905         if (nb_ports == 0)
3906                 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3907
3908         /* allocate port structures, and init them */
3909         init_port();
3910
3911         set_def_fwd_config();
3912         if (nb_lcores == 0)
3913                 rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
3914                          "Check the core mask argument\n");
3915
3916         /* Bitrate/latency stats disabled by default */
3917 #ifdef RTE_LIB_BITRATESTATS
3918         bitrate_enabled = 0;
3919 #endif
3920 #ifdef RTE_LIB_LATENCYSTATS
3921         latencystats_enabled = 0;
3922 #endif
3923
3924         /* on FreeBSD, mlockall() is disabled by default */
3925 #ifdef RTE_EXEC_ENV_FREEBSD
3926         do_mlockall = 0;
3927 #else
3928         do_mlockall = 1;
3929 #endif
3930
3931         argc -= diag;
3932         argv += diag;
3933         if (argc > 1)
3934                 launch_args_parse(argc, argv);
3935
3936 #ifndef RTE_EXEC_ENV_WINDOWS
3937         if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3938                 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3939                         strerror(errno));
3940         }
3941 #endif
3942
3943         if (tx_first && interactive)
3944                 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3945                                 "interactive mode.\n");
3946
3947         if (tx_first && lsc_interrupt) {
3948                 fprintf(stderr,
3949                         "Warning: lsc_interrupt needs to be off when using tx_first. Disabling.\n");
3950                 lsc_interrupt = 0;
3951         }
3952
3953         if (!nb_rxq && !nb_txq)
3954                 fprintf(stderr,
3955                         "Warning: Either rx or tx queues should be non-zero\n");
3956
3957         if (nb_rxq > 1 && nb_rxq > nb_txq)
3958                 fprintf(stderr,
3959                         "Warning: nb_rxq=%d enables RSS configuration, but nb_txq=%d will prevent to fully test it.\n",
3960                         nb_rxq, nb_txq);
3961
3962         init_config();
3963
3964         if (hot_plug) {
3965                 ret = rte_dev_hotplug_handle_enable();
3966                 if (ret) {
3967                         RTE_LOG(ERR, EAL,
3968                                 "fail to enable hotplug handling.");
3969                         return -1;
3970                 }
3971
3972                 ret = rte_dev_event_monitor_start();
3973                 if (ret) {
3974                         RTE_LOG(ERR, EAL,
3975                                 "fail to start device event monitoring.");
3976                         return -1;
3977                 }
3978
3979                 ret = rte_dev_event_callback_register(NULL,
3980                         dev_event_callback, NULL);
3981                 if (ret) {
3982                         RTE_LOG(ERR, EAL,
3983                                 "fail  to register device event callback\n");
3984                         return -1;
3985                 }
3986         }
3987
3988         if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
3989                 rte_exit(EXIT_FAILURE, "Start ports failed\n");
3990
3991         /* set all ports to promiscuous mode by default */
3992         RTE_ETH_FOREACH_DEV(port_id) {
3993                 ret = rte_eth_promiscuous_enable(port_id);
3994                 if (ret != 0)
3995                         fprintf(stderr,
3996                                 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3997                                 port_id, rte_strerror(-ret));
3998         }
3999
4000         /* Init metrics library */
4001         rte_metrics_init(rte_socket_id());
4002
4003 #ifdef RTE_LIB_LATENCYSTATS
4004         if (latencystats_enabled != 0) {
4005                 int ret = rte_latencystats_init(1, NULL);
4006                 if (ret)
4007                         fprintf(stderr,
4008                                 "Warning: latencystats init() returned error %d\n",
4009                                 ret);
4010                 fprintf(stderr, "Latencystats running on lcore %d\n",
4011                         latencystats_lcore_id);
4012         }
4013 #endif
4014
4015         /* Setup bitrate stats */
4016 #ifdef RTE_LIB_BITRATESTATS
4017         if (bitrate_enabled != 0) {
4018                 bitrate_data = rte_stats_bitrate_create();
4019                 if (bitrate_data == NULL)
4020                         rte_exit(EXIT_FAILURE,
4021                                 "Could not allocate bitrate data.\n");
4022                 rte_stats_bitrate_reg(bitrate_data);
4023         }
4024 #endif
4025
4026 #ifdef RTE_LIB_CMDLINE
4027         if (strlen(cmdline_filename) != 0)
4028                 cmdline_read_from_file(cmdline_filename);
4029
4030         if (interactive == 1) {
4031                 if (auto_start) {
4032                         printf("Start automatic packet forwarding\n");
4033                         start_packet_forwarding(0);
4034                 }
4035                 prompt();
4036                 pmd_test_exit();
4037         } else
4038 #endif
4039         {
4040                 char c;
4041                 int rc;
4042
4043                 f_quit = 0;
4044
4045                 printf("No commandline core given, start packet forwarding\n");
4046                 start_packet_forwarding(tx_first);
4047                 if (stats_period != 0) {
4048                         uint64_t prev_time = 0, cur_time, diff_time = 0;
4049                         uint64_t timer_period;
4050
4051                         /* Convert to number of cycles */
4052                         timer_period = stats_period * rte_get_timer_hz();
4053
4054                         while (f_quit == 0) {
4055                                 cur_time = rte_get_timer_cycles();
4056                                 diff_time += cur_time - prev_time;
4057
4058                                 if (diff_time >= timer_period) {
4059                                         print_stats();
4060                                         /* Reset the timer */
4061                                         diff_time = 0;
4062                                 }
4063                                 /* Sleep to avoid unnecessary checks */
4064                                 prev_time = cur_time;
4065                                 rte_delay_us_sleep(US_PER_S);
4066                         }
4067                 }
4068
4069                 printf("Press enter to exit\n");
4070                 rc = read(0, &c, 1);
4071                 pmd_test_exit();
4072                 if (rc < 0)
4073                         return 1;
4074         }
4075
4076         ret = rte_eal_cleanup();
4077         if (ret != 0)
4078                 rte_exit(EXIT_FAILURE,
4079                          "EAL cleanup failed: %s\n", strerror(-ret));
4080
4081         return EXIT_SUCCESS;
4082 }