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         unsigned int i;
2176         unsigned int lc_id;
2177         int diag;
2178
2179         for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2180                 lc_id = fwd_lcores_cpuids[i];
2181                 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2182                         fwd_lcores[i]->stopped = 0;
2183                         diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2184                                                      fwd_lcores[i], lc_id);
2185                         if (diag != 0)
2186                                 fprintf(stderr,
2187                                         "launch lcore %u failed - diag=%d\n",
2188                                         lc_id, diag);
2189                 }
2190         }
2191 }
2192
2193 /*
2194  * Launch packet forwarding configuration.
2195  */
2196 void
2197 start_packet_forwarding(int with_tx_first)
2198 {
2199         port_fwd_begin_t port_fwd_begin;
2200         port_fwd_end_t  port_fwd_end;
2201         unsigned int i;
2202
2203         if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2204                 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2205
2206         if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2207                 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2208
2209         if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2210                 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2211                 (!nb_rxq || !nb_txq))
2212                 rte_exit(EXIT_FAILURE,
2213                         "Either rxq or txq are 0, cannot use %s fwd mode\n",
2214                         cur_fwd_eng->fwd_mode_name);
2215
2216         if (all_ports_started() == 0) {
2217                 fprintf(stderr, "Not all ports were started\n");
2218                 return;
2219         }
2220         if (test_done == 0) {
2221                 fprintf(stderr, "Packet forwarding already started\n");
2222                 return;
2223         }
2224
2225         fwd_config_setup();
2226
2227         port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2228         if (port_fwd_begin != NULL) {
2229                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2230                         if (port_fwd_begin(fwd_ports_ids[i])) {
2231                                 fprintf(stderr,
2232                                         "Packet forwarding is not ready\n");
2233                                 return;
2234                         }
2235                 }
2236         }
2237
2238         if (with_tx_first) {
2239                 port_fwd_begin = tx_only_engine.port_fwd_begin;
2240                 if (port_fwd_begin != NULL) {
2241                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2242                                 if (port_fwd_begin(fwd_ports_ids[i])) {
2243                                         fprintf(stderr,
2244                                                 "Packet forwarding is not ready\n");
2245                                         return;
2246                                 }
2247                         }
2248                 }
2249         }
2250
2251         test_done = 0;
2252
2253         if(!no_flush_rx)
2254                 flush_fwd_rx_queues();
2255
2256         pkt_fwd_config_display(&cur_fwd_config);
2257         rxtx_config_display();
2258
2259         fwd_stats_reset();
2260         if (with_tx_first) {
2261                 while (with_tx_first--) {
2262                         launch_packet_forwarding(
2263                                         run_one_txonly_burst_on_core);
2264                         rte_eal_mp_wait_lcore();
2265                 }
2266                 port_fwd_end = tx_only_engine.port_fwd_end;
2267                 if (port_fwd_end != NULL) {
2268                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2269                                 (*port_fwd_end)(fwd_ports_ids[i]);
2270                 }
2271         }
2272         launch_packet_forwarding(start_pkt_forward_on_core);
2273 }
2274
2275 void
2276 stop_packet_forwarding(void)
2277 {
2278         port_fwd_end_t port_fwd_end;
2279         lcoreid_t lc_id;
2280         portid_t pt_id;
2281         int i;
2282
2283         if (test_done) {
2284                 fprintf(stderr, "Packet forwarding not started\n");
2285                 return;
2286         }
2287         printf("Telling cores to stop...");
2288         for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2289                 fwd_lcores[lc_id]->stopped = 1;
2290         printf("\nWaiting for lcores to finish...\n");
2291         rte_eal_mp_wait_lcore();
2292         port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2293         if (port_fwd_end != NULL) {
2294                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2295                         pt_id = fwd_ports_ids[i];
2296                         (*port_fwd_end)(pt_id);
2297                 }
2298         }
2299
2300         fwd_stats_display();
2301
2302         printf("\nDone.\n");
2303         test_done = 1;
2304 }
2305
2306 void
2307 dev_set_link_up(portid_t pid)
2308 {
2309         if (rte_eth_dev_set_link_up(pid) < 0)
2310                 fprintf(stderr, "\nSet link up fail.\n");
2311 }
2312
2313 void
2314 dev_set_link_down(portid_t pid)
2315 {
2316         if (rte_eth_dev_set_link_down(pid) < 0)
2317                 fprintf(stderr, "\nSet link down fail.\n");
2318 }
2319
2320 static int
2321 all_ports_started(void)
2322 {
2323         portid_t pi;
2324         struct rte_port *port;
2325
2326         RTE_ETH_FOREACH_DEV(pi) {
2327                 port = &ports[pi];
2328                 /* Check if there is a port which is not started */
2329                 if ((port->port_status != RTE_PORT_STARTED) &&
2330                         (port->slave_flag == 0))
2331                         return 0;
2332         }
2333
2334         /* No port is not started */
2335         return 1;
2336 }
2337
2338 int
2339 port_is_stopped(portid_t port_id)
2340 {
2341         struct rte_port *port = &ports[port_id];
2342
2343         if ((port->port_status != RTE_PORT_STOPPED) &&
2344             (port->slave_flag == 0))
2345                 return 0;
2346         return 1;
2347 }
2348
2349 int
2350 all_ports_stopped(void)
2351 {
2352         portid_t pi;
2353
2354         RTE_ETH_FOREACH_DEV(pi) {
2355                 if (!port_is_stopped(pi))
2356                         return 0;
2357         }
2358
2359         return 1;
2360 }
2361
2362 int
2363 port_is_started(portid_t port_id)
2364 {
2365         if (port_id_is_invalid(port_id, ENABLED_WARN))
2366                 return 0;
2367
2368         if (ports[port_id].port_status != RTE_PORT_STARTED)
2369                 return 0;
2370
2371         return 1;
2372 }
2373
2374 /* Configure the Rx and Tx hairpin queues for the selected port. */
2375 static int
2376 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2377 {
2378         queueid_t qi;
2379         struct rte_eth_hairpin_conf hairpin_conf = {
2380                 .peer_count = 1,
2381         };
2382         int i;
2383         int diag;
2384         struct rte_port *port = &ports[pi];
2385         uint16_t peer_rx_port = pi;
2386         uint16_t peer_tx_port = pi;
2387         uint32_t manual = 1;
2388         uint32_t tx_exp = hairpin_mode & 0x10;
2389
2390         if (!(hairpin_mode & 0xf)) {
2391                 peer_rx_port = pi;
2392                 peer_tx_port = pi;
2393                 manual = 0;
2394         } else if (hairpin_mode & 0x1) {
2395                 peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2396                                                        RTE_ETH_DEV_NO_OWNER);
2397                 if (peer_tx_port >= RTE_MAX_ETHPORTS)
2398                         peer_tx_port = rte_eth_find_next_owned_by(0,
2399                                                 RTE_ETH_DEV_NO_OWNER);
2400                 if (p_pi != RTE_MAX_ETHPORTS) {
2401                         peer_rx_port = p_pi;
2402                 } else {
2403                         uint16_t next_pi;
2404
2405                         /* Last port will be the peer RX port of the first. */
2406                         RTE_ETH_FOREACH_DEV(next_pi)
2407                                 peer_rx_port = next_pi;
2408                 }
2409                 manual = 1;
2410         } else if (hairpin_mode & 0x2) {
2411                 if (cnt_pi & 0x1) {
2412                         peer_rx_port = p_pi;
2413                 } else {
2414                         peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2415                                                 RTE_ETH_DEV_NO_OWNER);
2416                         if (peer_rx_port >= RTE_MAX_ETHPORTS)
2417                                 peer_rx_port = pi;
2418                 }
2419                 peer_tx_port = peer_rx_port;
2420                 manual = 1;
2421         }
2422
2423         for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2424                 hairpin_conf.peers[0].port = peer_rx_port;
2425                 hairpin_conf.peers[0].queue = i + nb_rxq;
2426                 hairpin_conf.manual_bind = !!manual;
2427                 hairpin_conf.tx_explicit = !!tx_exp;
2428                 diag = rte_eth_tx_hairpin_queue_setup
2429                         (pi, qi, nb_txd, &hairpin_conf);
2430                 i++;
2431                 if (diag == 0)
2432                         continue;
2433
2434                 /* Fail to setup rx queue, return */
2435                 if (rte_atomic16_cmpset(&(port->port_status),
2436                                         RTE_PORT_HANDLING,
2437                                         RTE_PORT_STOPPED) == 0)
2438                         fprintf(stderr,
2439                                 "Port %d can not be set back to stopped\n", pi);
2440                 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2441                         pi);
2442                 /* try to reconfigure queues next time */
2443                 port->need_reconfig_queues = 1;
2444                 return -1;
2445         }
2446         for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2447                 hairpin_conf.peers[0].port = peer_tx_port;
2448                 hairpin_conf.peers[0].queue = i + nb_txq;
2449                 hairpin_conf.manual_bind = !!manual;
2450                 hairpin_conf.tx_explicit = !!tx_exp;
2451                 diag = rte_eth_rx_hairpin_queue_setup
2452                         (pi, qi, nb_rxd, &hairpin_conf);
2453                 i++;
2454                 if (diag == 0)
2455                         continue;
2456
2457                 /* Fail to setup rx queue, return */
2458                 if (rte_atomic16_cmpset(&(port->port_status),
2459                                         RTE_PORT_HANDLING,
2460                                         RTE_PORT_STOPPED) == 0)
2461                         fprintf(stderr,
2462                                 "Port %d can not be set back to stopped\n", pi);
2463                 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2464                         pi);
2465                 /* try to reconfigure queues next time */
2466                 port->need_reconfig_queues = 1;
2467                 return -1;
2468         }
2469         return 0;
2470 }
2471
2472 /* Configure the Rx with optional split. */
2473 int
2474 rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
2475                uint16_t nb_rx_desc, unsigned int socket_id,
2476                struct rte_eth_rxconf *rx_conf, struct rte_mempool *mp)
2477 {
2478         union rte_eth_rxseg rx_useg[MAX_SEGS_BUFFER_SPLIT] = {};
2479         unsigned int i, mp_n;
2480         int ret;
2481
2482         if (rx_pkt_nb_segs <= 1 ||
2483             (rx_conf->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) == 0) {
2484                 rx_conf->rx_seg = NULL;
2485                 rx_conf->rx_nseg = 0;
2486                 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id,
2487                                              nb_rx_desc, socket_id,
2488                                              rx_conf, mp);
2489                 return ret;
2490         }
2491         for (i = 0; i < rx_pkt_nb_segs; i++) {
2492                 struct rte_eth_rxseg_split *rx_seg = &rx_useg[i].split;
2493                 struct rte_mempool *mpx;
2494                 /*
2495                  * Use last valid pool for the segments with number
2496                  * exceeding the pool index.
2497                  */
2498                 mp_n = (i > mbuf_data_size_n) ? mbuf_data_size_n - 1 : i;
2499                 mpx = mbuf_pool_find(socket_id, mp_n);
2500                 /* Handle zero as mbuf data buffer size. */
2501                 rx_seg->length = rx_pkt_seg_lengths[i] ?
2502                                    rx_pkt_seg_lengths[i] :
2503                                    mbuf_data_size[mp_n];
2504                 rx_seg->offset = i < rx_pkt_nb_offs ?
2505                                    rx_pkt_seg_offsets[i] : 0;
2506                 rx_seg->mp = mpx ? mpx : mp;
2507         }
2508         rx_conf->rx_nseg = rx_pkt_nb_segs;
2509         rx_conf->rx_seg = rx_useg;
2510         ret = rte_eth_rx_queue_setup(port_id, rx_queue_id, nb_rx_desc,
2511                                     socket_id, rx_conf, NULL);
2512         rx_conf->rx_seg = NULL;
2513         rx_conf->rx_nseg = 0;
2514         return ret;
2515 }
2516
2517 int
2518 start_port(portid_t pid)
2519 {
2520         int diag, need_check_link_status = -1;
2521         portid_t pi;
2522         portid_t p_pi = RTE_MAX_ETHPORTS;
2523         portid_t pl[RTE_MAX_ETHPORTS];
2524         portid_t peer_pl[RTE_MAX_ETHPORTS];
2525         uint16_t cnt_pi = 0;
2526         uint16_t cfg_pi = 0;
2527         int peer_pi;
2528         queueid_t qi;
2529         struct rte_port *port;
2530         struct rte_eth_hairpin_cap cap;
2531
2532         if (port_id_is_invalid(pid, ENABLED_WARN))
2533                 return 0;
2534
2535         RTE_ETH_FOREACH_DEV(pi) {
2536                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2537                         continue;
2538
2539                 need_check_link_status = 0;
2540                 port = &ports[pi];
2541                 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STOPPED,
2542                                                  RTE_PORT_HANDLING) == 0) {
2543                         fprintf(stderr, "Port %d is now not stopped\n", pi);
2544                         continue;
2545                 }
2546
2547                 if (port->need_reconfig > 0) {
2548                         port->need_reconfig = 0;
2549
2550                         if (flow_isolate_all) {
2551                                 int ret = port_flow_isolate(pi, 1);
2552                                 if (ret) {
2553                                         fprintf(stderr,
2554                                                 "Failed to apply isolated mode on port %d\n",
2555                                                 pi);
2556                                         return -1;
2557                                 }
2558                         }
2559                         configure_rxtx_dump_callbacks(0);
2560                         printf("Configuring Port %d (socket %u)\n", pi,
2561                                         port->socket_id);
2562                         if (nb_hairpinq > 0 &&
2563                             rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2564                                 fprintf(stderr,
2565                                         "Port %d doesn't support hairpin queues\n",
2566                                         pi);
2567                                 return -1;
2568                         }
2569                         /* configure port */
2570                         diag = eth_dev_configure_mp(pi, nb_rxq + nb_hairpinq,
2571                                                      nb_txq + nb_hairpinq,
2572                                                      &(port->dev_conf));
2573                         if (diag != 0) {
2574                                 if (rte_atomic16_cmpset(&(port->port_status),
2575                                 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2576                                         fprintf(stderr,
2577                                                 "Port %d can not be set back to stopped\n",
2578                                                 pi);
2579                                 fprintf(stderr, "Fail to configure port %d\n",
2580                                         pi);
2581                                 /* try to reconfigure port next time */
2582                                 port->need_reconfig = 1;
2583                                 return -1;
2584                         }
2585                 }
2586                 if (port->need_reconfig_queues > 0 && is_proc_primary()) {
2587                         port->need_reconfig_queues = 0;
2588                         /* setup tx queues */
2589                         for (qi = 0; qi < nb_txq; qi++) {
2590                                 if ((numa_support) &&
2591                                         (txring_numa[pi] != NUMA_NO_CONFIG))
2592                                         diag = rte_eth_tx_queue_setup(pi, qi,
2593                                                 port->nb_tx_desc[qi],
2594                                                 txring_numa[pi],
2595                                                 &(port->tx_conf[qi]));
2596                                 else
2597                                         diag = rte_eth_tx_queue_setup(pi, qi,
2598                                                 port->nb_tx_desc[qi],
2599                                                 port->socket_id,
2600                                                 &(port->tx_conf[qi]));
2601
2602                                 if (diag == 0)
2603                                         continue;
2604
2605                                 /* Fail to setup tx queue, return */
2606                                 if (rte_atomic16_cmpset(&(port->port_status),
2607                                                         RTE_PORT_HANDLING,
2608                                                         RTE_PORT_STOPPED) == 0)
2609                                         fprintf(stderr,
2610                                                 "Port %d can not be set back to stopped\n",
2611                                                 pi);
2612                                 fprintf(stderr,
2613                                         "Fail to configure port %d tx queues\n",
2614                                         pi);
2615                                 /* try to reconfigure queues next time */
2616                                 port->need_reconfig_queues = 1;
2617                                 return -1;
2618                         }
2619                         for (qi = 0; qi < nb_rxq; qi++) {
2620                                 /* setup rx queues */
2621                                 if ((numa_support) &&
2622                                         (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2623                                         struct rte_mempool * mp =
2624                                                 mbuf_pool_find
2625                                                         (rxring_numa[pi], 0);
2626                                         if (mp == NULL) {
2627                                                 fprintf(stderr,
2628                                                         "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2629                                                         rxring_numa[pi]);
2630                                                 return -1;
2631                                         }
2632
2633                                         diag = rx_queue_setup(pi, qi,
2634                                              port->nb_rx_desc[qi],
2635                                              rxring_numa[pi],
2636                                              &(port->rx_conf[qi]),
2637                                              mp);
2638                                 } else {
2639                                         struct rte_mempool *mp =
2640                                                 mbuf_pool_find
2641                                                         (port->socket_id, 0);
2642                                         if (mp == NULL) {
2643                                                 fprintf(stderr,
2644                                                         "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2645                                                         port->socket_id);
2646                                                 return -1;
2647                                         }
2648                                         diag = rx_queue_setup(pi, qi,
2649                                              port->nb_rx_desc[qi],
2650                                              port->socket_id,
2651                                              &(port->rx_conf[qi]),
2652                                              mp);
2653                                 }
2654                                 if (diag == 0)
2655                                         continue;
2656
2657                                 /* Fail to setup rx queue, return */
2658                                 if (rte_atomic16_cmpset(&(port->port_status),
2659                                                         RTE_PORT_HANDLING,
2660                                                         RTE_PORT_STOPPED) == 0)
2661                                         fprintf(stderr,
2662                                                 "Port %d can not be set back to stopped\n",
2663                                                 pi);
2664                                 fprintf(stderr,
2665                                         "Fail to configure port %d rx queues\n",
2666                                         pi);
2667                                 /* try to reconfigure queues next time */
2668                                 port->need_reconfig_queues = 1;
2669                                 return -1;
2670                         }
2671                         /* setup hairpin queues */
2672                         if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2673                                 return -1;
2674                 }
2675                 configure_rxtx_dump_callbacks(verbose_level);
2676                 if (clear_ptypes) {
2677                         diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2678                                         NULL, 0);
2679                         if (diag < 0)
2680                                 fprintf(stderr,
2681                                         "Port %d: Failed to disable Ptype parsing\n",
2682                                         pi);
2683                 }
2684
2685                 p_pi = pi;
2686                 cnt_pi++;
2687
2688                 /* start port */
2689                 diag = eth_dev_start_mp(pi);
2690                 if (diag < 0) {
2691                         fprintf(stderr, "Fail to start port %d: %s\n",
2692                                 pi, rte_strerror(-diag));
2693
2694                         /* Fail to setup rx queue, return */
2695                         if (rte_atomic16_cmpset(&(port->port_status),
2696                                 RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2697                                 fprintf(stderr,
2698                                         "Port %d can not be set back to stopped\n",
2699                                         pi);
2700                         continue;
2701                 }
2702
2703                 if (rte_atomic16_cmpset(&(port->port_status),
2704                         RTE_PORT_HANDLING, RTE_PORT_STARTED) == 0)
2705                         fprintf(stderr, "Port %d can not be set into started\n",
2706                                 pi);
2707
2708                 if (eth_macaddr_get_print_err(pi, &port->eth_addr) == 0)
2709                         printf("Port %d: " RTE_ETHER_ADDR_PRT_FMT "\n", pi,
2710                                         RTE_ETHER_ADDR_BYTES(&port->eth_addr));
2711
2712                 /* at least one port started, need checking link status */
2713                 need_check_link_status = 1;
2714
2715                 pl[cfg_pi++] = pi;
2716         }
2717
2718         if (need_check_link_status == 1 && !no_link_check)
2719                 check_all_ports_link_status(RTE_PORT_ALL);
2720         else if (need_check_link_status == 0)
2721                 fprintf(stderr, "Please stop the ports first\n");
2722
2723         if (hairpin_mode & 0xf) {
2724                 uint16_t i;
2725                 int j;
2726
2727                 /* bind all started hairpin ports */
2728                 for (i = 0; i < cfg_pi; i++) {
2729                         pi = pl[i];
2730                         /* bind current Tx to all peer Rx */
2731                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2732                                                         RTE_MAX_ETHPORTS, 1);
2733                         if (peer_pi < 0)
2734                                 return peer_pi;
2735                         for (j = 0; j < peer_pi; j++) {
2736                                 if (!port_is_started(peer_pl[j]))
2737                                         continue;
2738                                 diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2739                                 if (diag < 0) {
2740                                         fprintf(stderr,
2741                                                 "Error during binding hairpin Tx port %u to %u: %s\n",
2742                                                 pi, peer_pl[j],
2743                                                 rte_strerror(-diag));
2744                                         return -1;
2745                                 }
2746                         }
2747                         /* bind all peer Tx to current Rx */
2748                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2749                                                         RTE_MAX_ETHPORTS, 0);
2750                         if (peer_pi < 0)
2751                                 return peer_pi;
2752                         for (j = 0; j < peer_pi; j++) {
2753                                 if (!port_is_started(peer_pl[j]))
2754                                         continue;
2755                                 diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2756                                 if (diag < 0) {
2757                                         fprintf(stderr,
2758                                                 "Error during binding hairpin Tx port %u to %u: %s\n",
2759                                                 peer_pl[j], pi,
2760                                                 rte_strerror(-diag));
2761                                         return -1;
2762                                 }
2763                         }
2764                 }
2765         }
2766
2767         printf("Done\n");
2768         return 0;
2769 }
2770
2771 void
2772 stop_port(portid_t pid)
2773 {
2774         portid_t pi;
2775         struct rte_port *port;
2776         int need_check_link_status = 0;
2777         portid_t peer_pl[RTE_MAX_ETHPORTS];
2778         int peer_pi;
2779
2780         if (port_id_is_invalid(pid, ENABLED_WARN))
2781                 return;
2782
2783         printf("Stopping ports...\n");
2784
2785         RTE_ETH_FOREACH_DEV(pi) {
2786                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2787                         continue;
2788
2789                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2790                         fprintf(stderr,
2791                                 "Please remove port %d from forwarding configuration.\n",
2792                                 pi);
2793                         continue;
2794                 }
2795
2796                 if (port_is_bonding_slave(pi)) {
2797                         fprintf(stderr,
2798                                 "Please remove port %d from bonded device.\n",
2799                                 pi);
2800                         continue;
2801                 }
2802
2803                 port = &ports[pi];
2804                 if (rte_atomic16_cmpset(&(port->port_status), RTE_PORT_STARTED,
2805                                                 RTE_PORT_HANDLING) == 0)
2806                         continue;
2807
2808                 if (hairpin_mode & 0xf) {
2809                         int j;
2810
2811                         rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
2812                         /* unbind all peer Tx from current Rx */
2813                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2814                                                         RTE_MAX_ETHPORTS, 0);
2815                         if (peer_pi < 0)
2816                                 continue;
2817                         for (j = 0; j < peer_pi; j++) {
2818                                 if (!port_is_started(peer_pl[j]))
2819                                         continue;
2820                                 rte_eth_hairpin_unbind(peer_pl[j], pi);
2821                         }
2822                 }
2823
2824                 if (port->flow_list)
2825                         port_flow_flush(pi);
2826
2827                 if (eth_dev_stop_mp(pi) != 0)
2828                         RTE_LOG(ERR, EAL, "rte_eth_dev_stop failed for port %u\n",
2829                                 pi);
2830
2831                 if (rte_atomic16_cmpset(&(port->port_status),
2832                         RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
2833                         fprintf(stderr, "Port %d can not be set into stopped\n",
2834                                 pi);
2835                 need_check_link_status = 1;
2836         }
2837         if (need_check_link_status && !no_link_check)
2838                 check_all_ports_link_status(RTE_PORT_ALL);
2839
2840         printf("Done\n");
2841 }
2842
2843 static void
2844 remove_invalid_ports_in(portid_t *array, portid_t *total)
2845 {
2846         portid_t i;
2847         portid_t new_total = 0;
2848
2849         for (i = 0; i < *total; i++)
2850                 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
2851                         array[new_total] = array[i];
2852                         new_total++;
2853                 }
2854         *total = new_total;
2855 }
2856
2857 static void
2858 remove_invalid_ports(void)
2859 {
2860         remove_invalid_ports_in(ports_ids, &nb_ports);
2861         remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
2862         nb_cfg_ports = nb_fwd_ports;
2863 }
2864
2865 void
2866 close_port(portid_t pid)
2867 {
2868         portid_t pi;
2869         struct rte_port *port;
2870
2871         if (port_id_is_invalid(pid, ENABLED_WARN))
2872                 return;
2873
2874         printf("Closing ports...\n");
2875
2876         RTE_ETH_FOREACH_DEV(pi) {
2877                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2878                         continue;
2879
2880                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2881                         fprintf(stderr,
2882                                 "Please remove port %d from forwarding configuration.\n",
2883                                 pi);
2884                         continue;
2885                 }
2886
2887                 if (port_is_bonding_slave(pi)) {
2888                         fprintf(stderr,
2889                                 "Please remove port %d from bonded device.\n",
2890                                 pi);
2891                         continue;
2892                 }
2893
2894                 port = &ports[pi];
2895                 if (rte_atomic16_cmpset(&(port->port_status),
2896                         RTE_PORT_CLOSED, RTE_PORT_CLOSED) == 1) {
2897                         fprintf(stderr, "Port %d is already closed\n", pi);
2898                         continue;
2899                 }
2900
2901                 if (is_proc_primary()) {
2902                         port_flow_flush(pi);
2903                         rte_eth_dev_close(pi);
2904                 }
2905         }
2906
2907         remove_invalid_ports();
2908         printf("Done\n");
2909 }
2910
2911 void
2912 reset_port(portid_t pid)
2913 {
2914         int diag;
2915         portid_t pi;
2916         struct rte_port *port;
2917
2918         if (port_id_is_invalid(pid, ENABLED_WARN))
2919                 return;
2920
2921         if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
2922                 (pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
2923                 fprintf(stderr,
2924                         "Can not reset port(s), please stop port(s) first.\n");
2925                 return;
2926         }
2927
2928         printf("Resetting ports...\n");
2929
2930         RTE_ETH_FOREACH_DEV(pi) {
2931                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2932                         continue;
2933
2934                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
2935                         fprintf(stderr,
2936                                 "Please remove port %d from forwarding configuration.\n",
2937                                 pi);
2938                         continue;
2939                 }
2940
2941                 if (port_is_bonding_slave(pi)) {
2942                         fprintf(stderr,
2943                                 "Please remove port %d from bonded device.\n",
2944                                 pi);
2945                         continue;
2946                 }
2947
2948                 diag = rte_eth_dev_reset(pi);
2949                 if (diag == 0) {
2950                         port = &ports[pi];
2951                         port->need_reconfig = 1;
2952                         port->need_reconfig_queues = 1;
2953                 } else {
2954                         fprintf(stderr, "Failed to reset port %d. diag=%d\n",
2955                                 pi, diag);
2956                 }
2957         }
2958
2959         printf("Done\n");
2960 }
2961
2962 void
2963 attach_port(char *identifier)
2964 {
2965         portid_t pi;
2966         struct rte_dev_iterator iterator;
2967
2968         printf("Attaching a new port...\n");
2969
2970         if (identifier == NULL) {
2971                 fprintf(stderr, "Invalid parameters are specified\n");
2972                 return;
2973         }
2974
2975         if (rte_dev_probe(identifier) < 0) {
2976                 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
2977                 return;
2978         }
2979
2980         /* first attach mode: event */
2981         if (setup_on_probe_event) {
2982                 /* new ports are detected on RTE_ETH_EVENT_NEW event */
2983                 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
2984                         if (ports[pi].port_status == RTE_PORT_HANDLING &&
2985                                         ports[pi].need_setup != 0)
2986                                 setup_attached_port(pi);
2987                 return;
2988         }
2989
2990         /* second attach mode: iterator */
2991         RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
2992                 /* setup ports matching the devargs used for probing */
2993                 if (port_is_forwarding(pi))
2994                         continue; /* port was already attached before */
2995                 setup_attached_port(pi);
2996         }
2997 }
2998
2999 static void
3000 setup_attached_port(portid_t pi)
3001 {
3002         unsigned int socket_id;
3003         int ret;
3004
3005         socket_id = (unsigned)rte_eth_dev_socket_id(pi);
3006         /* if socket_id is invalid, set to the first available socket. */
3007         if (check_socket_id(socket_id) < 0)
3008                 socket_id = socket_ids[0];
3009         reconfig(pi, socket_id);
3010         ret = rte_eth_promiscuous_enable(pi);
3011         if (ret != 0)
3012                 fprintf(stderr,
3013                         "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3014                         pi, rte_strerror(-ret));
3015
3016         ports_ids[nb_ports++] = pi;
3017         fwd_ports_ids[nb_fwd_ports++] = pi;
3018         nb_cfg_ports = nb_fwd_ports;
3019         ports[pi].need_setup = 0;
3020         ports[pi].port_status = RTE_PORT_STOPPED;
3021
3022         printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
3023         printf("Done\n");
3024 }
3025
3026 static void
3027 detach_device(struct rte_device *dev)
3028 {
3029         portid_t sibling;
3030
3031         if (dev == NULL) {
3032                 fprintf(stderr, "Device already removed\n");
3033                 return;
3034         }
3035
3036         printf("Removing a device...\n");
3037
3038         RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
3039                 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
3040                         if (ports[sibling].port_status != RTE_PORT_STOPPED) {
3041                                 fprintf(stderr, "Port %u not stopped\n",
3042                                         sibling);
3043                                 return;
3044                         }
3045                         port_flow_flush(sibling);
3046                 }
3047         }
3048
3049         if (rte_dev_remove(dev) < 0) {
3050                 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
3051                 return;
3052         }
3053         remove_invalid_ports();
3054
3055         printf("Device is detached\n");
3056         printf("Now total ports is %d\n", nb_ports);
3057         printf("Done\n");
3058         return;
3059 }
3060
3061 void
3062 detach_port_device(portid_t port_id)
3063 {
3064         int ret;
3065         struct rte_eth_dev_info dev_info;
3066
3067         if (port_id_is_invalid(port_id, ENABLED_WARN))
3068                 return;
3069
3070         if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3071                 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3072                         fprintf(stderr, "Port not stopped\n");
3073                         return;
3074                 }
3075                 fprintf(stderr, "Port was not closed\n");
3076         }
3077
3078         ret = eth_dev_info_get_print_err(port_id, &dev_info);
3079         if (ret != 0) {
3080                 TESTPMD_LOG(ERR,
3081                         "Failed to get device info for port %d, not detaching\n",
3082                         port_id);
3083                 return;
3084         }
3085         detach_device(dev_info.device);
3086 }
3087
3088 void
3089 detach_devargs(char *identifier)
3090 {
3091         struct rte_dev_iterator iterator;
3092         struct rte_devargs da;
3093         portid_t port_id;
3094
3095         printf("Removing a device...\n");
3096
3097         memset(&da, 0, sizeof(da));
3098         if (rte_devargs_parsef(&da, "%s", identifier)) {
3099                 fprintf(stderr, "cannot parse identifier\n");
3100                 return;
3101         }
3102
3103         RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3104                 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3105                         if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3106                                 fprintf(stderr, "Port %u not stopped\n",
3107                                         port_id);
3108                                 rte_eth_iterator_cleanup(&iterator);
3109                                 rte_devargs_reset(&da);
3110                                 return;
3111                         }
3112                         port_flow_flush(port_id);
3113                 }
3114         }
3115
3116         if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3117                 TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3118                             da.name, da.bus->name);
3119                 rte_devargs_reset(&da);
3120                 return;
3121         }
3122
3123         remove_invalid_ports();
3124
3125         printf("Device %s is detached\n", identifier);
3126         printf("Now total ports is %d\n", nb_ports);
3127         printf("Done\n");
3128         rte_devargs_reset(&da);
3129 }
3130
3131 void
3132 pmd_test_exit(void)
3133 {
3134         portid_t pt_id;
3135         unsigned int i;
3136         int ret;
3137
3138         if (test_done == 0)
3139                 stop_packet_forwarding();
3140
3141 #ifndef RTE_EXEC_ENV_WINDOWS
3142         for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3143                 if (mempools[i]) {
3144                         if (mp_alloc_type == MP_ALLOC_ANON)
3145                                 rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3146                                                      NULL);
3147                 }
3148         }
3149 #endif
3150         if (ports != NULL) {
3151                 no_link_check = 1;
3152                 RTE_ETH_FOREACH_DEV(pt_id) {
3153                         printf("\nStopping port %d...\n", pt_id);
3154                         fflush(stdout);
3155                         stop_port(pt_id);
3156                 }
3157                 RTE_ETH_FOREACH_DEV(pt_id) {
3158                         printf("\nShutting down port %d...\n", pt_id);
3159                         fflush(stdout);
3160                         close_port(pt_id);
3161                 }
3162         }
3163
3164         if (hot_plug) {
3165                 ret = rte_dev_event_monitor_stop();
3166                 if (ret) {
3167                         RTE_LOG(ERR, EAL,
3168                                 "fail to stop device event monitor.");
3169                         return;
3170                 }
3171
3172                 ret = rte_dev_event_callback_unregister(NULL,
3173                         dev_event_callback, NULL);
3174                 if (ret < 0) {
3175                         RTE_LOG(ERR, EAL,
3176                                 "fail to unregister device event callback.\n");
3177                         return;
3178                 }
3179
3180                 ret = rte_dev_hotplug_handle_disable();
3181                 if (ret) {
3182                         RTE_LOG(ERR, EAL,
3183                                 "fail to disable hotplug handling.\n");
3184                         return;
3185                 }
3186         }
3187         for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3188                 if (mempools[i])
3189                         mempool_free_mp(mempools[i]);
3190         }
3191
3192         printf("\nBye...\n");
3193 }
3194
3195 typedef void (*cmd_func_t)(void);
3196 struct pmd_test_command {
3197         const char *cmd_name;
3198         cmd_func_t cmd_func;
3199 };
3200
3201 /* Check the link status of all ports in up to 9s, and print them finally */
3202 static void
3203 check_all_ports_link_status(uint32_t port_mask)
3204 {
3205 #define CHECK_INTERVAL 100 /* 100ms */
3206 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3207         portid_t portid;
3208         uint8_t count, all_ports_up, print_flag = 0;
3209         struct rte_eth_link link;
3210         int ret;
3211         char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3212
3213         printf("Checking link statuses...\n");
3214         fflush(stdout);
3215         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3216                 all_ports_up = 1;
3217                 RTE_ETH_FOREACH_DEV(portid) {
3218                         if ((port_mask & (1 << portid)) == 0)
3219                                 continue;
3220                         memset(&link, 0, sizeof(link));
3221                         ret = rte_eth_link_get_nowait(portid, &link);
3222                         if (ret < 0) {
3223                                 all_ports_up = 0;
3224                                 if (print_flag == 1)
3225                                         fprintf(stderr,
3226                                                 "Port %u link get failed: %s\n",
3227                                                 portid, rte_strerror(-ret));
3228                                 continue;
3229                         }
3230                         /* print link status if flag set */
3231                         if (print_flag == 1) {
3232                                 rte_eth_link_to_str(link_status,
3233                                         sizeof(link_status), &link);
3234                                 printf("Port %d %s\n", portid, link_status);
3235                                 continue;
3236                         }
3237                         /* clear all_ports_up flag if any link down */
3238                         if (link.link_status == ETH_LINK_DOWN) {
3239                                 all_ports_up = 0;
3240                                 break;
3241                         }
3242                 }
3243                 /* after finally printing all link status, get out */
3244                 if (print_flag == 1)
3245                         break;
3246
3247                 if (all_ports_up == 0) {
3248                         fflush(stdout);
3249                         rte_delay_ms(CHECK_INTERVAL);
3250                 }
3251
3252                 /* set the print_flag if all ports up or timeout */
3253                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3254                         print_flag = 1;
3255                 }
3256
3257                 if (lsc_interrupt)
3258                         break;
3259         }
3260 }
3261
3262 static void
3263 rmv_port_callback(void *arg)
3264 {
3265         int need_to_start = 0;
3266         int org_no_link_check = no_link_check;
3267         portid_t port_id = (intptr_t)arg;
3268         struct rte_eth_dev_info dev_info;
3269         int ret;
3270
3271         RTE_ETH_VALID_PORTID_OR_RET(port_id);
3272
3273         if (!test_done && port_is_forwarding(port_id)) {
3274                 need_to_start = 1;
3275                 stop_packet_forwarding();
3276         }
3277         no_link_check = 1;
3278         stop_port(port_id);
3279         no_link_check = org_no_link_check;
3280
3281         ret = eth_dev_info_get_print_err(port_id, &dev_info);
3282         if (ret != 0)
3283                 TESTPMD_LOG(ERR,
3284                         "Failed to get device info for port %d, not detaching\n",
3285                         port_id);
3286         else {
3287                 struct rte_device *device = dev_info.device;
3288                 close_port(port_id);
3289                 detach_device(device); /* might be already removed or have more ports */
3290         }
3291         if (need_to_start)
3292                 start_packet_forwarding(0);
3293 }
3294
3295 /* This function is used by the interrupt thread */
3296 static int
3297 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3298                   void *ret_param)
3299 {
3300         RTE_SET_USED(param);
3301         RTE_SET_USED(ret_param);
3302
3303         if (type >= RTE_ETH_EVENT_MAX) {
3304                 fprintf(stderr,
3305                         "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3306                         port_id, __func__, type);
3307                 fflush(stderr);
3308         } else if (event_print_mask & (UINT32_C(1) << type)) {
3309                 printf("\nPort %" PRIu16 ": %s event\n", port_id,
3310                         eth_event_desc[type]);
3311                 fflush(stdout);
3312         }
3313
3314         switch (type) {
3315         case RTE_ETH_EVENT_NEW:
3316                 ports[port_id].need_setup = 1;
3317                 ports[port_id].port_status = RTE_PORT_HANDLING;
3318                 break;
3319         case RTE_ETH_EVENT_INTR_RMV:
3320                 if (port_id_is_invalid(port_id, DISABLED_WARN))
3321                         break;
3322                 if (rte_eal_alarm_set(100000,
3323                                 rmv_port_callback, (void *)(intptr_t)port_id))
3324                         fprintf(stderr,
3325                                 "Could not set up deferred device removal\n");
3326                 break;
3327         case RTE_ETH_EVENT_DESTROY:
3328                 ports[port_id].port_status = RTE_PORT_CLOSED;
3329                 printf("Port %u is closed\n", port_id);
3330                 break;
3331         default:
3332                 break;
3333         }
3334         return 0;
3335 }
3336
3337 static int
3338 register_eth_event_callback(void)
3339 {
3340         int ret;
3341         enum rte_eth_event_type event;
3342
3343         for (event = RTE_ETH_EVENT_UNKNOWN;
3344                         event < RTE_ETH_EVENT_MAX; event++) {
3345                 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3346                                 event,
3347                                 eth_event_callback,
3348                                 NULL);
3349                 if (ret != 0) {
3350                         TESTPMD_LOG(ERR, "Failed to register callback for "
3351                                         "%s event\n", eth_event_desc[event]);
3352                         return -1;
3353                 }
3354         }
3355
3356         return 0;
3357 }
3358
3359 /* This function is used by the interrupt thread */
3360 static void
3361 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3362                              __rte_unused void *arg)
3363 {
3364         uint16_t port_id;
3365         int ret;
3366
3367         if (type >= RTE_DEV_EVENT_MAX) {
3368                 fprintf(stderr, "%s called upon invalid event %d\n",
3369                         __func__, type);
3370                 fflush(stderr);
3371         }
3372
3373         switch (type) {
3374         case RTE_DEV_EVENT_REMOVE:
3375                 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3376                         device_name);
3377                 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3378                 if (ret) {
3379                         RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3380                                 device_name);
3381                         return;
3382                 }
3383                 /*
3384                  * Because the user's callback is invoked in eal interrupt
3385                  * callback, the interrupt callback need to be finished before
3386                  * it can be unregistered when detaching device. So finish
3387                  * callback soon and use a deferred removal to detach device
3388                  * is need. It is a workaround, once the device detaching be
3389                  * moved into the eal in the future, the deferred removal could
3390                  * be deleted.
3391                  */
3392                 if (rte_eal_alarm_set(100000,
3393                                 rmv_port_callback, (void *)(intptr_t)port_id))
3394                         RTE_LOG(ERR, EAL,
3395                                 "Could not set up deferred device removal\n");
3396                 break;
3397         case RTE_DEV_EVENT_ADD:
3398                 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3399                         device_name);
3400                 /* TODO: After finish kernel driver binding,
3401                  * begin to attach port.
3402                  */
3403                 break;
3404         default:
3405                 break;
3406         }
3407 }
3408
3409 static void
3410 rxtx_port_config(struct rte_port *port)
3411 {
3412         uint16_t qid;
3413         uint64_t offloads;
3414
3415         for (qid = 0; qid < nb_rxq; qid++) {
3416                 offloads = port->rx_conf[qid].offloads;
3417                 port->rx_conf[qid] = port->dev_info.default_rxconf;
3418                 if (offloads != 0)
3419                         port->rx_conf[qid].offloads = offloads;
3420
3421                 /* Check if any Rx parameters have been passed */
3422                 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3423                         port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3424
3425                 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3426                         port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3427
3428                 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3429                         port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3430
3431                 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3432                         port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3433
3434                 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3435                         port->rx_conf[qid].rx_drop_en = rx_drop_en;
3436
3437                 port->nb_rx_desc[qid] = nb_rxd;
3438         }
3439
3440         for (qid = 0; qid < nb_txq; qid++) {
3441                 offloads = port->tx_conf[qid].offloads;
3442                 port->tx_conf[qid] = port->dev_info.default_txconf;
3443                 if (offloads != 0)
3444                         port->tx_conf[qid].offloads = offloads;
3445
3446                 /* Check if any Tx parameters have been passed */
3447                 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3448                         port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3449
3450                 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3451                         port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3452
3453                 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3454                         port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3455
3456                 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3457                         port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3458
3459                 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3460                         port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3461
3462                 port->nb_tx_desc[qid] = nb_txd;
3463         }
3464 }
3465
3466 /*
3467  * Helper function to arrange max_rx_pktlen value and JUMBO_FRAME offload,
3468  * MTU is also aligned if JUMBO_FRAME offload is not set.
3469  *
3470  * port->dev_info should be set before calling this function.
3471  *
3472  * return 0 on success, negative on error
3473  */
3474 int
3475 update_jumbo_frame_offload(portid_t portid)
3476 {
3477         struct rte_port *port = &ports[portid];
3478         uint32_t eth_overhead;
3479         uint64_t rx_offloads;
3480         int ret;
3481         bool on;
3482
3483         /* Update the max_rx_pkt_len to have MTU as RTE_ETHER_MTU */
3484         if (port->dev_info.max_mtu != UINT16_MAX &&
3485             port->dev_info.max_rx_pktlen > port->dev_info.max_mtu)
3486                 eth_overhead = port->dev_info.max_rx_pktlen -
3487                                 port->dev_info.max_mtu;
3488         else
3489                 eth_overhead = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
3490
3491         rx_offloads = port->dev_conf.rxmode.offloads;
3492
3493         /* Default config value is 0 to use PMD specific overhead */
3494         if (port->dev_conf.rxmode.max_rx_pkt_len == 0)
3495                 port->dev_conf.rxmode.max_rx_pkt_len = RTE_ETHER_MTU + eth_overhead;
3496
3497         if (port->dev_conf.rxmode.max_rx_pkt_len <= RTE_ETHER_MTU + eth_overhead) {
3498                 rx_offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3499                 on = false;
3500         } else {
3501                 if ((port->dev_info.rx_offload_capa & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3502                         fprintf(stderr,
3503                                 "Frame size (%u) is not supported by port %u\n",
3504                                 port->dev_conf.rxmode.max_rx_pkt_len,
3505                                 portid);
3506                         return -1;
3507                 }
3508                 rx_offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3509                 on = true;
3510         }
3511
3512         if (rx_offloads != port->dev_conf.rxmode.offloads) {
3513                 uint16_t qid;
3514
3515                 port->dev_conf.rxmode.offloads = rx_offloads;
3516
3517                 /* Apply JUMBO_FRAME offload configuration to Rx queue(s) */
3518                 for (qid = 0; qid < port->dev_info.nb_rx_queues; qid++) {
3519                         if (on)
3520                                 port->rx_conf[qid].offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
3521                         else
3522                                 port->rx_conf[qid].offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3523                 }
3524         }
3525
3526         /* If JUMBO_FRAME is set MTU conversion done by ethdev layer,
3527          * if unset do it here
3528          */
3529         if ((rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) == 0) {
3530                 ret = eth_dev_set_mtu_mp(portid,
3531                                 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead);
3532                 if (ret)
3533                         fprintf(stderr,
3534                                 "Failed to set MTU to %u for port %u\n",
3535                                 port->dev_conf.rxmode.max_rx_pkt_len - eth_overhead,
3536                                 portid);
3537         }
3538
3539         return 0;
3540 }
3541
3542 void
3543 init_port_config(void)
3544 {
3545         portid_t pid;
3546         struct rte_port *port;
3547         int ret;
3548
3549         RTE_ETH_FOREACH_DEV(pid) {
3550                 port = &ports[pid];
3551                 port->dev_conf.fdir_conf = fdir_conf;
3552
3553                 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3554                 if (ret != 0)
3555                         return;
3556
3557                 if (nb_rxq > 1) {
3558                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3559                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3560                                 rss_hf & port->dev_info.flow_type_rss_offloads;
3561                 } else {
3562                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3563                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3564                 }
3565
3566                 if (port->dcb_flag == 0) {
3567                         if( port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0)
3568                                 port->dev_conf.rxmode.mq_mode =
3569                                         (enum rte_eth_rx_mq_mode)
3570                                                 (rx_mq_mode & ETH_MQ_RX_RSS);
3571                         else
3572                                 port->dev_conf.rxmode.mq_mode = ETH_MQ_RX_NONE;
3573                 }
3574
3575                 rxtx_port_config(port);
3576
3577                 ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3578                 if (ret != 0)
3579                         return;
3580
3581 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
3582                 rte_pmd_ixgbe_bypass_init(pid);
3583 #endif
3584
3585                 if (lsc_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_LSC))
3586                         port->dev_conf.intr_conf.lsc = 1;
3587                 if (rmv_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_RMV))
3588                         port->dev_conf.intr_conf.rmv = 1;
3589         }
3590 }
3591
3592 void set_port_slave_flag(portid_t slave_pid)
3593 {
3594         struct rte_port *port;
3595
3596         port = &ports[slave_pid];
3597         port->slave_flag = 1;
3598 }
3599
3600 void clear_port_slave_flag(portid_t slave_pid)
3601 {
3602         struct rte_port *port;
3603
3604         port = &ports[slave_pid];
3605         port->slave_flag = 0;
3606 }
3607
3608 uint8_t port_is_bonding_slave(portid_t slave_pid)
3609 {
3610         struct rte_port *port;
3611         struct rte_eth_dev_info dev_info;
3612         int ret;
3613
3614         port = &ports[slave_pid];
3615         ret = eth_dev_info_get_print_err(slave_pid, &dev_info);
3616         if (ret != 0) {
3617                 TESTPMD_LOG(ERR,
3618                         "Failed to get device info for port id %d,"
3619                         "cannot determine if the port is a bonded slave",
3620                         slave_pid);
3621                 return 0;
3622         }
3623         if ((*dev_info.dev_flags & RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3624                 return 1;
3625         return 0;
3626 }
3627
3628 const uint16_t vlan_tags[] = {
3629                 0,  1,  2,  3,  4,  5,  6,  7,
3630                 8,  9, 10, 11,  12, 13, 14, 15,
3631                 16, 17, 18, 19, 20, 21, 22, 23,
3632                 24, 25, 26, 27, 28, 29, 30, 31
3633 };
3634
3635 static  int
3636 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3637                  enum dcb_mode_enable dcb_mode,
3638                  enum rte_eth_nb_tcs num_tcs,
3639                  uint8_t pfc_en)
3640 {
3641         uint8_t i;
3642         int32_t rc;
3643         struct rte_eth_rss_conf rss_conf;
3644
3645         /*
3646          * Builds up the correct configuration for dcb+vt based on the vlan tags array
3647          * given above, and the number of traffic classes available for use.
3648          */
3649         if (dcb_mode == DCB_VT_ENABLED) {
3650                 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3651                                 &eth_conf->rx_adv_conf.vmdq_dcb_conf;
3652                 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3653                                 &eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3654
3655                 /* VMDQ+DCB RX and TX configurations */
3656                 vmdq_rx_conf->enable_default_pool = 0;
3657                 vmdq_rx_conf->default_pool = 0;
3658                 vmdq_rx_conf->nb_queue_pools =
3659                         (num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3660                 vmdq_tx_conf->nb_queue_pools =
3661                         (num_tcs ==  ETH_4_TCS ? ETH_32_POOLS : ETH_16_POOLS);
3662
3663                 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3664                 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3665                         vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3666                         vmdq_rx_conf->pool_map[i].pools =
3667                                 1 << (i % vmdq_rx_conf->nb_queue_pools);
3668                 }
3669                 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3670                         vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3671                         vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3672                 }
3673
3674                 /* set DCB mode of RX and TX of multiple queues */
3675                 eth_conf->rxmode.mq_mode =
3676                                 (enum rte_eth_rx_mq_mode)
3677                                         (rx_mq_mode & ETH_MQ_RX_VMDQ_DCB);
3678                 eth_conf->txmode.mq_mode = ETH_MQ_TX_VMDQ_DCB;
3679         } else {
3680                 struct rte_eth_dcb_rx_conf *rx_conf =
3681                                 &eth_conf->rx_adv_conf.dcb_rx_conf;
3682                 struct rte_eth_dcb_tx_conf *tx_conf =
3683                                 &eth_conf->tx_adv_conf.dcb_tx_conf;
3684
3685                 memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3686
3687                 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3688                 if (rc != 0)
3689                         return rc;
3690
3691                 rx_conf->nb_tcs = num_tcs;
3692                 tx_conf->nb_tcs = num_tcs;
3693
3694                 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3695                         rx_conf->dcb_tc[i] = i % num_tcs;
3696                         tx_conf->dcb_tc[i] = i % num_tcs;
3697                 }
3698
3699                 eth_conf->rxmode.mq_mode =
3700                                 (enum rte_eth_rx_mq_mode)
3701                                         (rx_mq_mode & ETH_MQ_RX_DCB_RSS);
3702                 eth_conf->rx_adv_conf.rss_conf = rss_conf;
3703                 eth_conf->txmode.mq_mode = ETH_MQ_TX_DCB;
3704         }
3705
3706         if (pfc_en)
3707                 eth_conf->dcb_capability_en =
3708                                 ETH_DCB_PG_SUPPORT | ETH_DCB_PFC_SUPPORT;
3709         else
3710                 eth_conf->dcb_capability_en = ETH_DCB_PG_SUPPORT;
3711
3712         return 0;
3713 }
3714
3715 int
3716 init_port_dcb_config(portid_t pid,
3717                      enum dcb_mode_enable dcb_mode,
3718                      enum rte_eth_nb_tcs num_tcs,
3719                      uint8_t pfc_en)
3720 {
3721         struct rte_eth_conf port_conf;
3722         struct rte_port *rte_port;
3723         int retval;
3724         uint16_t i;
3725
3726         if (num_procs > 1) {
3727                 printf("The multi-process feature doesn't support dcb.\n");
3728                 return -ENOTSUP;
3729         }
3730         rte_port = &ports[pid];
3731
3732         /* retain the original device configuration. */
3733         memcpy(&port_conf, &rte_port->dev_conf, sizeof(struct rte_eth_conf));
3734
3735         /*set configuration of DCB in vt mode and DCB in non-vt mode*/
3736         retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3737         if (retval < 0)
3738                 return retval;
3739         port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3740
3741         /* re-configure the device . */
3742         retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3743         if (retval < 0)
3744                 return retval;
3745
3746         retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3747         if (retval != 0)
3748                 return retval;
3749
3750         /* If dev_info.vmdq_pool_base is greater than 0,
3751          * the queue id of vmdq pools is started after pf queues.
3752          */
3753         if (dcb_mode == DCB_VT_ENABLED &&
3754             rte_port->dev_info.vmdq_pool_base > 0) {
3755                 fprintf(stderr,
3756                         "VMDQ_DCB multi-queue mode is nonsensical for port %d.\n",
3757                         pid);
3758                 return -1;
3759         }
3760
3761         /* Assume the ports in testpmd have the same dcb capability
3762          * and has the same number of rxq and txq in dcb mode
3763          */
3764         if (dcb_mode == DCB_VT_ENABLED) {
3765                 if (rte_port->dev_info.max_vfs > 0) {
3766                         nb_rxq = rte_port->dev_info.nb_rx_queues;
3767                         nb_txq = rte_port->dev_info.nb_tx_queues;
3768                 } else {
3769                         nb_rxq = rte_port->dev_info.max_rx_queues;
3770                         nb_txq = rte_port->dev_info.max_tx_queues;
3771                 }
3772         } else {
3773                 /*if vt is disabled, use all pf queues */
3774                 if (rte_port->dev_info.vmdq_pool_base == 0) {
3775                         nb_rxq = rte_port->dev_info.max_rx_queues;
3776                         nb_txq = rte_port->dev_info.max_tx_queues;
3777                 } else {
3778                         nb_rxq = (queueid_t)num_tcs;
3779                         nb_txq = (queueid_t)num_tcs;
3780
3781                 }
3782         }
3783         rx_free_thresh = 64;
3784
3785         memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
3786
3787         rxtx_port_config(rte_port);
3788         /* VLAN filter */
3789         rte_port->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_FILTER;
3790         for (i = 0; i < RTE_DIM(vlan_tags); i++)
3791                 rx_vft_set(pid, vlan_tags[i], 1);
3792
3793         retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
3794         if (retval != 0)
3795                 return retval;
3796
3797         rte_port->dcb_flag = 1;
3798
3799         /* Enter DCB configuration status */
3800         dcb_config = 1;
3801
3802         return 0;
3803 }
3804
3805 static void
3806 init_port(void)
3807 {
3808         int i;
3809
3810         /* Configuration of Ethernet ports. */
3811         ports = rte_zmalloc("testpmd: ports",
3812                             sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
3813                             RTE_CACHE_LINE_SIZE);
3814         if (ports == NULL) {
3815                 rte_exit(EXIT_FAILURE,
3816                                 "rte_zmalloc(%d struct rte_port) failed\n",
3817                                 RTE_MAX_ETHPORTS);
3818         }
3819         for (i = 0; i < RTE_MAX_ETHPORTS; i++)
3820                 LIST_INIT(&ports[i].flow_tunnel_list);
3821         /* Initialize ports NUMA structures */
3822         memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3823         memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3824         memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
3825 }
3826
3827 static void
3828 force_quit(void)
3829 {
3830         pmd_test_exit();
3831         prompt_exit();
3832 }
3833
3834 static void
3835 print_stats(void)
3836 {
3837         uint8_t i;
3838         const char clr[] = { 27, '[', '2', 'J', '\0' };
3839         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
3840
3841         /* Clear screen and move to top left */
3842         printf("%s%s", clr, top_left);
3843
3844         printf("\nPort statistics ====================================");
3845         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
3846                 nic_stats_display(fwd_ports_ids[i]);
3847
3848         fflush(stdout);
3849 }
3850
3851 static void
3852 signal_handler(int signum)
3853 {
3854         if (signum == SIGINT || signum == SIGTERM) {
3855                 fprintf(stderr, "\nSignal %d received, preparing to exit...\n",
3856                         signum);
3857 #ifdef RTE_LIB_PDUMP
3858                 /* uninitialize packet capture framework */
3859                 rte_pdump_uninit();
3860 #endif
3861 #ifdef RTE_LIB_LATENCYSTATS
3862                 if (latencystats_enabled != 0)
3863                         rte_latencystats_uninit();
3864 #endif
3865                 force_quit();
3866                 /* Set flag to indicate the force termination. */
3867                 f_quit = 1;
3868                 /* exit with the expected status */
3869 #ifndef RTE_EXEC_ENV_WINDOWS
3870                 signal(signum, SIG_DFL);
3871                 kill(getpid(), signum);
3872 #endif
3873         }
3874 }
3875
3876 int
3877 main(int argc, char** argv)
3878 {
3879         int diag;
3880         portid_t port_id;
3881         uint16_t count;
3882         int ret;
3883
3884         signal(SIGINT, signal_handler);
3885         signal(SIGTERM, signal_handler);
3886
3887         testpmd_logtype = rte_log_register("testpmd");
3888         if (testpmd_logtype < 0)
3889                 rte_exit(EXIT_FAILURE, "Cannot register log type");
3890         rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
3891
3892         diag = rte_eal_init(argc, argv);
3893         if (diag < 0)
3894                 rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
3895                          rte_strerror(rte_errno));
3896
3897         ret = register_eth_event_callback();
3898         if (ret != 0)
3899                 rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
3900
3901 #ifdef RTE_LIB_PDUMP
3902         /* initialize packet capture framework */
3903         rte_pdump_init();
3904 #endif
3905
3906         count = 0;
3907         RTE_ETH_FOREACH_DEV(port_id) {
3908                 ports_ids[count] = port_id;
3909                 count++;
3910         }
3911         nb_ports = (portid_t) count;
3912         if (nb_ports == 0)
3913                 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
3914
3915         /* allocate port structures, and init them */
3916         init_port();
3917
3918         set_def_fwd_config();
3919         if (nb_lcores == 0)
3920                 rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
3921                          "Check the core mask argument\n");
3922
3923         /* Bitrate/latency stats disabled by default */
3924 #ifdef RTE_LIB_BITRATESTATS
3925         bitrate_enabled = 0;
3926 #endif
3927 #ifdef RTE_LIB_LATENCYSTATS
3928         latencystats_enabled = 0;
3929 #endif
3930
3931         /* on FreeBSD, mlockall() is disabled by default */
3932 #ifdef RTE_EXEC_ENV_FREEBSD
3933         do_mlockall = 0;
3934 #else
3935         do_mlockall = 1;
3936 #endif
3937
3938         argc -= diag;
3939         argv += diag;
3940         if (argc > 1)
3941                 launch_args_parse(argc, argv);
3942
3943 #ifndef RTE_EXEC_ENV_WINDOWS
3944         if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
3945                 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
3946                         strerror(errno));
3947         }
3948 #endif
3949
3950         if (tx_first && interactive)
3951                 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
3952                                 "interactive mode.\n");
3953
3954         if (tx_first && lsc_interrupt) {
3955                 fprintf(stderr,
3956                         "Warning: lsc_interrupt needs to be off when using tx_first. Disabling.\n");
3957                 lsc_interrupt = 0;
3958         }
3959
3960         if (!nb_rxq && !nb_txq)
3961                 fprintf(stderr,
3962                         "Warning: Either rx or tx queues should be non-zero\n");
3963
3964         if (nb_rxq > 1 && nb_rxq > nb_txq)
3965                 fprintf(stderr,
3966                         "Warning: nb_rxq=%d enables RSS configuration, but nb_txq=%d will prevent to fully test it.\n",
3967                         nb_rxq, nb_txq);
3968
3969         init_config();
3970
3971         if (hot_plug) {
3972                 ret = rte_dev_hotplug_handle_enable();
3973                 if (ret) {
3974                         RTE_LOG(ERR, EAL,
3975                                 "fail to enable hotplug handling.");
3976                         return -1;
3977                 }
3978
3979                 ret = rte_dev_event_monitor_start();
3980                 if (ret) {
3981                         RTE_LOG(ERR, EAL,
3982                                 "fail to start device event monitoring.");
3983                         return -1;
3984                 }
3985
3986                 ret = rte_dev_event_callback_register(NULL,
3987                         dev_event_callback, NULL);
3988                 if (ret) {
3989                         RTE_LOG(ERR, EAL,
3990                                 "fail  to register device event callback\n");
3991                         return -1;
3992                 }
3993         }
3994
3995         if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
3996                 rte_exit(EXIT_FAILURE, "Start ports failed\n");
3997
3998         /* set all ports to promiscuous mode by default */
3999         RTE_ETH_FOREACH_DEV(port_id) {
4000                 ret = rte_eth_promiscuous_enable(port_id);
4001                 if (ret != 0)
4002                         fprintf(stderr,
4003                                 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
4004                                 port_id, rte_strerror(-ret));
4005         }
4006
4007         /* Init metrics library */
4008         rte_metrics_init(rte_socket_id());
4009
4010 #ifdef RTE_LIB_LATENCYSTATS
4011         if (latencystats_enabled != 0) {
4012                 int ret = rte_latencystats_init(1, NULL);
4013                 if (ret)
4014                         fprintf(stderr,
4015                                 "Warning: latencystats init() returned error %d\n",
4016                                 ret);
4017                 fprintf(stderr, "Latencystats running on lcore %d\n",
4018                         latencystats_lcore_id);
4019         }
4020 #endif
4021
4022         /* Setup bitrate stats */
4023 #ifdef RTE_LIB_BITRATESTATS
4024         if (bitrate_enabled != 0) {
4025                 bitrate_data = rte_stats_bitrate_create();
4026                 if (bitrate_data == NULL)
4027                         rte_exit(EXIT_FAILURE,
4028                                 "Could not allocate bitrate data.\n");
4029                 rte_stats_bitrate_reg(bitrate_data);
4030         }
4031 #endif
4032
4033 #ifdef RTE_LIB_CMDLINE
4034         if (strlen(cmdline_filename) != 0)
4035                 cmdline_read_from_file(cmdline_filename);
4036
4037         if (interactive == 1) {
4038                 if (auto_start) {
4039                         printf("Start automatic packet forwarding\n");
4040                         start_packet_forwarding(0);
4041                 }
4042                 prompt();
4043                 pmd_test_exit();
4044         } else
4045 #endif
4046         {
4047                 char c;
4048                 int rc;
4049
4050                 f_quit = 0;
4051
4052                 printf("No commandline core given, start packet forwarding\n");
4053                 start_packet_forwarding(tx_first);
4054                 if (stats_period != 0) {
4055                         uint64_t prev_time = 0, cur_time, diff_time = 0;
4056                         uint64_t timer_period;
4057
4058                         /* Convert to number of cycles */
4059                         timer_period = stats_period * rte_get_timer_hz();
4060
4061                         while (f_quit == 0) {
4062                                 cur_time = rte_get_timer_cycles();
4063                                 diff_time += cur_time - prev_time;
4064
4065                                 if (diff_time >= timer_period) {
4066                                         print_stats();
4067                                         /* Reset the timer */
4068                                         diff_time = 0;
4069                                 }
4070                                 /* Sleep to avoid unnecessary checks */
4071                                 prev_time = cur_time;
4072                                 rte_delay_us_sleep(US_PER_S);
4073                         }
4074                 }
4075
4076                 printf("Press enter to exit\n");
4077                 rc = read(0, &c, 1);
4078                 pmd_test_exit();
4079                 if (rc < 0)
4080                         return 1;
4081         }
4082
4083         ret = rte_eal_cleanup();
4084         if (ret != 0)
4085                 rte_exit(EXIT_FAILURE,
4086                          "EAL cleanup failed: %s\n", strerror(-ret));
4087
4088         return EXIT_SUCCESS;
4089 }