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_branch_prediction.h>
  40 #include <rte_mempool.h>
  41 #include <rte_malloc.h>
  42 #include <rte_mbuf.h>
  43 #include <rte_mbuf_pool_ops.h>
  44 #include <rte_interrupts.h>
  45 #include <rte_pci.h>
  46 #include <rte_ether.h>
  47 #include <rte_ethdev.h>
  48 #include <rte_dev.h>
  49 #include <rte_string_fns.h>
  50 #ifdef RTE_NET_IXGBE
  51 #include <rte_pmd_ixgbe.h>
  52 #endif
  53 #ifdef RTE_LIB_PDUMP
  54 #include <rte_pdump.h>
  55 #endif
  56 #include <rte_flow.h>
  57 #ifdef RTE_LIB_METRICS
  58 #include <rte_metrics.h>
  59 #endif
  60 #ifdef RTE_LIB_BITRATESTATS
  61 #include <rte_bitrate.h>
  62 #endif
  63 #ifdef RTE_LIB_LATENCYSTATS
  64 #include <rte_latencystats.h>
  65 #endif
  66 #ifdef RTE_EXEC_ENV_WINDOWS
  67 #include <process.h>
  68 #endif
  69
  70 #include "testpmd.h"
  71
  72 #ifndef MAP_HUGETLB
  73 /* FreeBSD may not have MAP_HUGETLB (in fact, it probably doesn't) */
  74 #define HUGE_FLAG (0x40000)
  75 #else
  76 #define HUGE_FLAG MAP_HUGETLB
  77 #endif
  78
  79 #ifndef MAP_HUGE_SHIFT
  80 /* older kernels (or FreeBSD) will not have this define */
  81 #define HUGE_SHIFT (26)
  82 #else
  83 #define HUGE_SHIFT MAP_HUGE_SHIFT
  84 #endif
  85
  86 #define EXTMEM_HEAP_NAME "extmem"
  87 /*
  88  * Zone size with the malloc overhead (max of debug and release variants)
  89  * must fit into the smallest supported hugepage size (2M),
  90  * so that an IOVA-contiguous zone of this size can always be allocated
  91  * if there are free 2M hugepages.
  92  */
  93 #define EXTBUF_ZONE_SIZE (RTE_PGSIZE_2M - 4 * RTE_CACHE_LINE_SIZE)
  94
  95 uint16_t verbose_level = 0; /**< Silent by default. */
  96 int testpmd_logtype; /**< Log type for testpmd logs */
  97
  98 /* use main core for command line ? */
  99 uint8_t interactive = 0;
 100 uint8_t auto_start = 0;
 101 uint8_t tx_first;
 102 char cmdline_filename[PATH_MAX] = {0};
 103
 104 /*
 105  * NUMA support configuration.
 106  * When set, the NUMA support attempts to dispatch the allocation of the
 107  * RX and TX memory rings, and of the DMA memory buffers (mbufs) for the
 108  * probed ports among the CPU sockets 0 and 1.
 109  * Otherwise, all memory is allocated from CPU socket 0.
 110  */
 111 uint8_t numa_support = 1; /**< numa enabled by default */
 112
 113 /*
 114  * In UMA mode,all memory is allocated from socket 0 if --socket-num is
 115  * not configured.
 116  */
 117 uint8_t socket_num = UMA_NO_CONFIG;
 118
 119 /*
 120  * Select mempool allocation type:
 121  * - native: use regular DPDK memory
 122  * - anon: use regular DPDK memory to create mempool, but populate using
 123  *         anonymous memory (may not be IOVA-contiguous)
 124  * - xmem: use externally allocated hugepage memory
 125  */
 126 uint8_t mp_alloc_type = MP_ALLOC_NATIVE;
 127
 128 /*
 129  * Store specified sockets on which memory pool to be used by ports
 130  * is allocated.
 131  */
 132 uint8_t port_numa[RTE_MAX_ETHPORTS];
 133
 134 /*
 135  * Store specified sockets on which RX ring to be used by ports
 136  * is allocated.
 137  */
 138 uint8_t rxring_numa[RTE_MAX_ETHPORTS];
 139
 140 /*
 141  * Store specified sockets on which TX ring to be used by ports
 142  * is allocated.
 143  */
 144 uint8_t txring_numa[RTE_MAX_ETHPORTS];
 145
 146 /*
 147  * Record the Ethernet address of peer target ports to which packets are
 148  * forwarded.
 149  * Must be instantiated with the ethernet addresses of peer traffic generator
 150  * ports.
 151  */
 152 struct rte_ether_addr peer_eth_addrs[RTE_MAX_ETHPORTS];
 153 portid_t nb_peer_eth_addrs = 0;
 154
 155 /*
 156  * Probed Target Environment.
 157  */
 158 struct rte_port *ports;        /**< For all probed ethernet ports. */
 159 portid_t nb_ports;             /**< Number of probed ethernet ports. */
 160 struct fwd_lcore **fwd_lcores; /**< For all probed logical cores. */
 161 lcoreid_t nb_lcores;           /**< Number of probed logical cores. */
 162
 163 portid_t ports_ids[RTE_MAX_ETHPORTS]; /**< Store all port ids. */
 164
 165 /*
 166  * Test Forwarding Configuration.
 167  *    nb_fwd_lcores <= nb_cfg_lcores <= nb_lcores
 168  *    nb_fwd_ports  <= nb_cfg_ports  <= nb_ports
 169  */
 170 lcoreid_t nb_cfg_lcores; /**< Number of configured logical cores. */
 171 lcoreid_t nb_fwd_lcores; /**< Number of forwarding logical cores. */
 172 portid_t  nb_cfg_ports;  /**< Number of configured ports. */
 173 portid_t  nb_fwd_ports;  /**< Number of forwarding ports. */
 174
 175 unsigned int fwd_lcores_cpuids[RTE_MAX_LCORE]; /**< CPU ids configuration. */
 176 portid_t fwd_ports_ids[RTE_MAX_ETHPORTS];      /**< Port ids configuration. */
 177
 178 struct fwd_stream **fwd_streams; /**< For each RX queue of each port. */
 179 streamid_t nb_fwd_streams;       /**< Is equal to (nb_ports * nb_rxq). */
 180
 181 /*
 182  * Forwarding engines.
 183  */
 184 struct fwd_engine * fwd_engines[] = {
 185         &io_fwd_engine,
 186         &mac_fwd_engine,
 187         &mac_swap_engine,
 188         &flow_gen_engine,
 189         &rx_only_engine,
 190         &tx_only_engine,
 191         &csum_fwd_engine,
 192         &icmp_echo_engine,
 193         &noisy_vnf_engine,
 194         &five_tuple_swap_fwd_engine,
 195 #ifdef RTE_LIBRTE_IEEE1588
 196         &ieee1588_fwd_engine,
 197 #endif
 198         &shared_rxq_engine,
 199         NULL,
 200 };
 201
 202 struct rte_mempool *mempools[RTE_MAX_NUMA_NODES * MAX_SEGS_BUFFER_SPLIT];
 203 uint16_t mempool_flags;
 204
 205 struct fwd_config cur_fwd_config;
 206 struct fwd_engine *cur_fwd_eng = &io_fwd_engine; /**< IO mode by default. */
 207 uint32_t retry_enabled;
 208 uint32_t burst_tx_delay_time = BURST_TX_WAIT_US;
 209 uint32_t burst_tx_retry_num = BURST_TX_RETRIES;
 210
 211 uint32_t mbuf_data_size_n = 1; /* Number of specified mbuf sizes. */
 212 uint16_t mbuf_data_size[MAX_SEGS_BUFFER_SPLIT] = {
 213         DEFAULT_MBUF_DATA_SIZE
 214 }; /**< Mbuf data space size. */
 215 uint32_t param_total_num_mbufs = 0;  /**< number of mbufs in all pools - if
 216                                       * specified on command-line. */
 217 uint16_t stats_period; /**< Period to show statistics (disabled by default) */
 218
 219 /** Extended statistics to show. */
 220 struct rte_eth_xstat_name *xstats_display;
 221
 222 unsigned int xstats_display_num; /**< Size of extended statistics to show */
 223
 224 /*
 225  * In container, it cannot terminate the process which running with 'stats-period'
 226  * option. Set flag to exit stats period loop after received SIGINT/SIGTERM.
 227  */
 228 uint8_t f_quit;
 229
 230 /*
 231  * Max Rx frame size, set by '--max-pkt-len' parameter.
 232  */
 233 uint32_t max_rx_pkt_len;
 234
 235 /*
 236  * Configuration of packet segments used to scatter received packets
 237  * if some of split features is configured.
 238  */
 239 uint16_t rx_pkt_seg_lengths[MAX_SEGS_BUFFER_SPLIT];
 240 uint8_t  rx_pkt_nb_segs; /**< Number of segments to split */
 241 uint16_t rx_pkt_seg_offsets[MAX_SEGS_BUFFER_SPLIT];
 242 uint8_t  rx_pkt_nb_offs; /**< Number of specified offsets */
 243
 244 /*
 245  * Configuration of packet segments used by the "txonly" processing engine.
 246  */
 247 uint16_t tx_pkt_length = TXONLY_DEF_PACKET_LEN; /**< TXONLY packet length. */
 248 uint16_t tx_pkt_seg_lengths[RTE_MAX_SEGS_PER_PKT] = {
 249         TXONLY_DEF_PACKET_LEN,
 250 };
 251 uint8_t  tx_pkt_nb_segs = 1; /**< Number of segments in TXONLY packets */
 252
 253 enum tx_pkt_split tx_pkt_split = TX_PKT_SPLIT_OFF;
 254 /**< Split policy for packets to TX. */
 255
 256 uint8_t txonly_multi_flow;
 257 /**< Whether multiple flows are generated in TXONLY mode. */
 258
 259 uint32_t tx_pkt_times_inter;
 260 /**< Timings for send scheduling in TXONLY mode, time between bursts. */
 261
 262 uint32_t tx_pkt_times_intra;
 263 /**< Timings for send scheduling in TXONLY mode, time between packets. */
 264
 265 uint16_t nb_pkt_per_burst = DEF_PKT_BURST; /**< Number of packets per burst. */
 266 uint16_t nb_pkt_flowgen_clones; /**< Number of Tx packet clones to send in flowgen mode. */
 267 int nb_flows_flowgen = 1024; /**< Number of flows in flowgen mode. */
 268 uint16_t mb_mempool_cache = DEF_MBUF_CACHE; /**< Size of mbuf mempool cache. */
 269
 270 /* current configuration is in DCB or not,0 means it is not in DCB mode */
 271 uint8_t dcb_config = 0;
 272
 273 /*
 274  * Configurable number of RX/TX queues.
 275  */
 276 queueid_t nb_hairpinq; /**< Number of hairpin queues per port. */
 277 queueid_t nb_rxq = 1; /**< Number of RX queues per port. */
 278 queueid_t nb_txq = 1; /**< Number of TX queues per port. */
 279
 280 /*
 281  * Configurable number of RX/TX ring descriptors.
 282  * Defaults are supplied by drivers via ethdev.
 283  */
 284 #define RTE_TEST_RX_DESC_DEFAULT 0
 285 #define RTE_TEST_TX_DESC_DEFAULT 0
 286 uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; /**< Number of RX descriptors. */
 287 uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; /**< Number of TX descriptors. */
 288
 289 #define RTE_PMD_PARAM_UNSET -1
 290 /*
 291  * Configurable values of RX and TX ring threshold registers.
 292  */
 293
 294 int8_t rx_pthresh = RTE_PMD_PARAM_UNSET;
 295 int8_t rx_hthresh = RTE_PMD_PARAM_UNSET;
 296 int8_t rx_wthresh = RTE_PMD_PARAM_UNSET;
 297
 298 int8_t tx_pthresh = RTE_PMD_PARAM_UNSET;
 299 int8_t tx_hthresh = RTE_PMD_PARAM_UNSET;
 300 int8_t tx_wthresh = RTE_PMD_PARAM_UNSET;
 301
 302 /*
 303  * Configurable value of RX free threshold.
 304  */
 305 int16_t rx_free_thresh = RTE_PMD_PARAM_UNSET;
 306
 307 /*
 308  * Configurable value of RX drop enable.
 309  */
 310 int8_t rx_drop_en = RTE_PMD_PARAM_UNSET;
 311
 312 /*
 313  * Configurable value of TX free threshold.
 314  */
 315 int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
 316
 317 /*
 318  * Configurable value of TX RS bit threshold.
 319  */
 320 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
 321
 322 /*
 323  * Configurable value of buffered packets before sending.
 324  */
 325 uint16_t noisy_tx_sw_bufsz;
 326
 327 /*
 328  * Configurable value of packet buffer timeout.
 329  */
 330 uint16_t noisy_tx_sw_buf_flush_time;
 331
 332 /*
 333  * Configurable value for size of VNF internal memory area
 334  * used for simulating noisy neighbour behaviour
 335  */
 336 uint64_t noisy_lkup_mem_sz;
 337
 338 /*
 339  * Configurable value of number of random writes done in
 340  * VNF simulation memory area.
 341  */
 342 uint64_t noisy_lkup_num_writes;
 343
 344 /*
 345  * Configurable value of number of random reads done in
 346  * VNF simulation memory area.
 347  */
 348 uint64_t noisy_lkup_num_reads;
 349
 350 /*
 351  * Configurable value of number of random reads/writes done in
 352  * VNF simulation memory area.
 353  */
 354 uint64_t noisy_lkup_num_reads_writes;
 355
 356 /*
 357  * Receive Side Scaling (RSS) configuration.
 358  */
 359 uint64_t rss_hf = RTE_ETH_RSS_IP; /* RSS IP by default. */
 360
 361 /*
 362  * Port topology configuration
 363  */
 364 uint16_t port_topology = PORT_TOPOLOGY_PAIRED; /* Ports are paired by default */
 365
 366 /*
 367  * Avoids to flush all the RX streams before starts forwarding.
 368  */
 369 uint8_t no_flush_rx = 0; /* flush by default */
 370
 371 /*
 372  * Flow API isolated mode.
 373  */
 374 uint8_t flow_isolate_all;
 375
 376 /*
 377  * Avoids to check link status when starting/stopping a port.
 378  */
 379 uint8_t no_link_check = 0; /* check by default */
 380
 381 /*
 382  * Don't automatically start all ports in interactive mode.
 383  */
 384 uint8_t no_device_start = 0;
 385
 386 /*
 387  * Enable link status change notification
 388  */
 389 uint8_t lsc_interrupt = 1; /* enabled by default */
 390
 391 /*
 392  * Enable device removal notification.
 393  */
 394 uint8_t rmv_interrupt = 1; /* enabled by default */
 395
 396 uint8_t hot_plug = 0; /**< hotplug disabled by default. */
 397
 398 /* After attach, port setup is called on event or by iterator */
 399 bool setup_on_probe_event = true;
 400
 401 /* Clear ptypes on port initialization. */
 402 uint8_t clear_ptypes = true;
 403
 404 /* Hairpin ports configuration mode. */
 405 uint16_t hairpin_mode;
 406
 407 /* Pretty printing of ethdev events */
 408 static const char * const eth_event_desc[] = {
 409         [RTE_ETH_EVENT_UNKNOWN] = "unknown",
 410         [RTE_ETH_EVENT_INTR_LSC] = "link state change",
 411         [RTE_ETH_EVENT_QUEUE_STATE] = "queue state",
 412         [RTE_ETH_EVENT_INTR_RESET] = "reset",
 413         [RTE_ETH_EVENT_VF_MBOX] = "VF mbox",
 414         [RTE_ETH_EVENT_IPSEC] = "IPsec",
 415         [RTE_ETH_EVENT_MACSEC] = "MACsec",
 416         [RTE_ETH_EVENT_INTR_RMV] = "device removal",
 417         [RTE_ETH_EVENT_NEW] = "device probed",
 418         [RTE_ETH_EVENT_DESTROY] = "device released",
 419         [RTE_ETH_EVENT_FLOW_AGED] = "flow aged",
 420         [RTE_ETH_EVENT_MAX] = NULL,
 421 };
 422
 423 /*
 424  * Display or mask ether events
 425  * Default to all events except VF_MBOX
 426  */
 427 uint32_t event_print_mask = (UINT32_C(1) << RTE_ETH_EVENT_UNKNOWN) |
 428                             (UINT32_C(1) << RTE_ETH_EVENT_INTR_LSC) |
 429                             (UINT32_C(1) << RTE_ETH_EVENT_QUEUE_STATE) |
 430                             (UINT32_C(1) << RTE_ETH_EVENT_INTR_RESET) |
 431                             (UINT32_C(1) << RTE_ETH_EVENT_IPSEC) |
 432                             (UINT32_C(1) << RTE_ETH_EVENT_MACSEC) |
 433                             (UINT32_C(1) << RTE_ETH_EVENT_INTR_RMV) |
 434                             (UINT32_C(1) << RTE_ETH_EVENT_FLOW_AGED);
 435 /*
 436  * Decide if all memory are locked for performance.
 437  */
 438 int do_mlockall = 0;
 439
 440 /*
 441  * NIC bypass mode configuration options.
 442  */
 443
 444 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
 445 /* The NIC bypass watchdog timeout. */
 446 uint32_t bypass_timeout = RTE_PMD_IXGBE_BYPASS_TMT_OFF;
 447 #endif
 448
 449
 450 #ifdef RTE_LIB_LATENCYSTATS
 451
 452 /*
 453  * Set when latency stats is enabled in the commandline
 454  */
 455 uint8_t latencystats_enabled;
 456
 457 /*
 458  * Lcore ID to service latency statistics.
 459  */
 460 lcoreid_t latencystats_lcore_id = -1;
 461
 462 #endif
 463
 464 /*
 465  * Ethernet device configuration.
 466  */
 467 struct rte_eth_rxmode rx_mode;
 468
 469 struct rte_eth_txmode tx_mode = {
 470         .offloads = RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE,
 471 };
 472
 473 struct rte_eth_fdir_conf fdir_conf = {
 474         .mode = RTE_FDIR_MODE_NONE,
 475         .pballoc = RTE_ETH_FDIR_PBALLOC_64K,
 476         .status = RTE_FDIR_REPORT_STATUS,
 477         .mask = {
 478                 .vlan_tci_mask = 0xFFEF,
 479                 .ipv4_mask     = {
 480                         .src_ip = 0xFFFFFFFF,
 481                         .dst_ip = 0xFFFFFFFF,
 482                 },
 483                 .ipv6_mask     = {
 484                         .src_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
 485                         .dst_ip = {0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF},
 486                 },
 487                 .src_port_mask = 0xFFFF,
 488                 .dst_port_mask = 0xFFFF,
 489                 .mac_addr_byte_mask = 0xFF,
 490                 .tunnel_type_mask = 1,
 491                 .tunnel_id_mask = 0xFFFFFFFF,
 492         },
 493         .drop_queue = 127,
 494 };
 495
 496 volatile int test_done = 1; /* stop packet forwarding when set to 1. */
 497
 498 /*
 499  * Display zero values by default for xstats
 500  */
 501 uint8_t xstats_hide_zero;
 502
 503 /*
 504  * Measure of CPU cycles disabled by default
 505  */
 506 uint8_t record_core_cycles;
 507
 508 /*
 509  * Display of RX and TX bursts disabled by default
 510  */
 511 uint8_t record_burst_stats;
 512
 513 /*
 514  * Number of ports per shared Rx queue group, 0 disable.
 515  */
 516 uint32_t rxq_share;
 517
 518 unsigned int num_sockets = 0;
 519 unsigned int socket_ids[RTE_MAX_NUMA_NODES];
 520
 521 #ifdef RTE_LIB_BITRATESTATS
 522 /* Bitrate statistics */
 523 struct rte_stats_bitrates *bitrate_data;
 524 lcoreid_t bitrate_lcore_id;
 525 uint8_t bitrate_enabled;
 526 #endif
 527
 528 #ifdef RTE_LIB_GRO
 529 struct gro_status gro_ports[RTE_MAX_ETHPORTS];
 530 uint8_t gro_flush_cycles = GRO_DEFAULT_FLUSH_CYCLES;
 531 #endif
 532
 533 /*
 534  * hexadecimal bitmask of RX mq mode can be enabled.
 535  */
 536 enum rte_eth_rx_mq_mode rx_mq_mode = RTE_ETH_MQ_RX_VMDQ_DCB_RSS;
 537
 538 /*
 539  * Used to set forced link speed
 540  */
 541 uint32_t eth_link_speed;
 542
 543 /*
 544  * ID of the current process in multi-process, used to
 545  * configure the queues to be polled.
 546  */
 547 int proc_id;
 548
 549 /*
 550  * Number of processes in multi-process, used to
 551  * configure the queues to be polled.
 552  */
 553 unsigned int num_procs = 1;
 554
 555 static void
 556 eth_rx_metadata_negotiate_mp(uint16_t port_id)
 557 {
 558         uint64_t rx_meta_features = 0;
 559         int ret;
 560
 561         if (!is_proc_primary())
 562                 return;
 563
 564         rx_meta_features |= RTE_ETH_RX_METADATA_USER_FLAG;
 565         rx_meta_features |= RTE_ETH_RX_METADATA_USER_MARK;
 566         rx_meta_features |= RTE_ETH_RX_METADATA_TUNNEL_ID;
 567
 568         ret = rte_eth_rx_metadata_negotiate(port_id, &rx_meta_features);
 569         if (ret == 0) {
 570                 if (!(rx_meta_features & RTE_ETH_RX_METADATA_USER_FLAG)) {
 571                         TESTPMD_LOG(DEBUG, "Flow action FLAG will not affect Rx mbufs on port %u\n",
 572                                     port_id);
 573                 }
 574
 575                 if (!(rx_meta_features & RTE_ETH_RX_METADATA_USER_MARK)) {
 576                         TESTPMD_LOG(DEBUG, "Flow action MARK will not affect Rx mbufs on port %u\n",
 577                                     port_id);
 578                 }
 579
 580                 if (!(rx_meta_features & RTE_ETH_RX_METADATA_TUNNEL_ID)) {
 581                         TESTPMD_LOG(DEBUG, "Flow tunnel offload support might be limited or unavailable on port %u\n",
 582                                     port_id);
 583                 }
 584         } else if (ret != -ENOTSUP) {
 585                 rte_exit(EXIT_FAILURE, "Error when negotiating Rx meta features on port %u: %s\n",
 586                          port_id, rte_strerror(-ret));
 587         }
 588 }
 589
 590 static int
 591 eth_dev_configure_mp(uint16_t port_id, uint16_t nb_rx_q, uint16_t nb_tx_q,
 592                       const struct rte_eth_conf *dev_conf)
 593 {
 594         if (is_proc_primary())
 595                 return rte_eth_dev_configure(port_id, nb_rx_q, nb_tx_q,
 596                                         dev_conf);
 597         return 0;
 598 }
 599
 600 static int
 601 eth_dev_start_mp(uint16_t port_id)
 602 {
 603         if (is_proc_primary())
 604                 return rte_eth_dev_start(port_id);
 605
 606         return 0;
 607 }
 608
 609 static int
 610 eth_dev_stop_mp(uint16_t port_id)
 611 {
 612         if (is_proc_primary())
 613                 return rte_eth_dev_stop(port_id);
 614
 615         return 0;
 616 }
 617
 618 static void
 619 mempool_free_mp(struct rte_mempool *mp)
 620 {
 621         if (is_proc_primary())
 622                 rte_mempool_free(mp);
 623 }
 624
 625 static int
 626 eth_dev_set_mtu_mp(uint16_t port_id, uint16_t mtu)
 627 {
 628         if (is_proc_primary())
 629                 return rte_eth_dev_set_mtu(port_id, mtu);
 630
 631         return 0;
 632 }
 633
 634 /* Forward function declarations */
 635 static void setup_attached_port(portid_t pi);
 636 static void check_all_ports_link_status(uint32_t port_mask);
 637 static int eth_event_callback(portid_t port_id,
 638                               enum rte_eth_event_type type,
 639                               void *param, void *ret_param);
 640 static void dev_event_callback(const char *device_name,
 641                                 enum rte_dev_event_type type,
 642                                 void *param);
 643 static void fill_xstats_display_info(void);
 644
 645 /*
 646  * Check if all the ports are started.
 647  * If yes, return positive value. If not, return zero.
 648  */
 649 static int all_ports_started(void);
 650
 651 #ifdef RTE_LIB_GSO
 652 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
 653 uint16_t gso_max_segment_size = RTE_ETHER_MAX_LEN - RTE_ETHER_CRC_LEN;
 654 #endif
 655
 656 /* Holds the registered mbuf dynamic flags names. */
 657 char dynf_names[64][RTE_MBUF_DYN_NAMESIZE];
 658
 659
 660 /*
 661  * Helper function to check if socket is already discovered.
 662  * If yes, return positive value. If not, return zero.
 663  */
 664 int
 665 new_socket_id(unsigned int socket_id)
 666 {
 667         unsigned int i;
 668
 669         for (i = 0; i < num_sockets; i++) {
 670                 if (socket_ids[i] == socket_id)
 671                         return 0;
 672         }
 673         return 1;
 674 }
 675
 676 /*
 677  * Setup default configuration.
 678  */
 679 static void
 680 set_default_fwd_lcores_config(void)
 681 {
 682         unsigned int i;
 683         unsigned int nb_lc;
 684         unsigned int sock_num;
 685
 686         nb_lc = 0;
 687         for (i = 0; i < RTE_MAX_LCORE; i++) {
 688                 if (!rte_lcore_is_enabled(i))
 689                         continue;
 690                 sock_num = rte_lcore_to_socket_id(i);
 691                 if (new_socket_id(sock_num)) {
 692                         if (num_sockets >= RTE_MAX_NUMA_NODES) {
 693                                 rte_exit(EXIT_FAILURE,
 694                                          "Total sockets greater than %u\n",
 695                                          RTE_MAX_NUMA_NODES);
 696                         }
 697                         socket_ids[num_sockets++] = sock_num;
 698                 }
 699                 if (i == rte_get_main_lcore())
 700                         continue;
 701                 fwd_lcores_cpuids[nb_lc++] = i;
 702         }
 703         nb_lcores = (lcoreid_t) nb_lc;
 704         nb_cfg_lcores = nb_lcores;
 705         nb_fwd_lcores = 1;
 706 }
 707
 708 static void
 709 set_def_peer_eth_addrs(void)
 710 {
 711         portid_t i;
 712
 713         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 714                 peer_eth_addrs[i].addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
 715                 peer_eth_addrs[i].addr_bytes[5] = i;
 716         }
 717 }
 718
 719 static void
 720 set_default_fwd_ports_config(void)
 721 {
 722         portid_t pt_id;
 723         int i = 0;
 724
 725         RTE_ETH_FOREACH_DEV(pt_id) {
 726                 fwd_ports_ids[i++] = pt_id;
 727
 728                 /* Update sockets info according to the attached device */
 729                 int socket_id = rte_eth_dev_socket_id(pt_id);
 730                 if (socket_id >= 0 && new_socket_id(socket_id)) {
 731                         if (num_sockets >= RTE_MAX_NUMA_NODES) {
 732                                 rte_exit(EXIT_FAILURE,
 733                                          "Total sockets greater than %u\n",
 734                                          RTE_MAX_NUMA_NODES);
 735                         }
 736                         socket_ids[num_sockets++] = socket_id;
 737                 }
 738         }
 739
 740         nb_cfg_ports = nb_ports;
 741         nb_fwd_ports = nb_ports;
 742 }
 743
 744 void
 745 set_def_fwd_config(void)
 746 {
 747         set_default_fwd_lcores_config();
 748         set_def_peer_eth_addrs();
 749         set_default_fwd_ports_config();
 750 }
 751
 752 #ifndef RTE_EXEC_ENV_WINDOWS
 753 /* extremely pessimistic estimation of memory required to create a mempool */
 754 static int
 755 calc_mem_size(uint32_t nb_mbufs, uint32_t mbuf_sz, size_t pgsz, size_t *out)
 756 {
 757         unsigned int n_pages, mbuf_per_pg, leftover;
 758         uint64_t total_mem, mbuf_mem, obj_sz;
 759
 760         /* there is no good way to predict how much space the mempool will
 761          * occupy because it will allocate chunks on the fly, and some of those
 762          * will come from default DPDK memory while some will come from our
 763          * external memory, so just assume 128MB will be enough for everyone.
 764          */
 765         uint64_t hdr_mem = 128 << 20;
 766
 767         /* account for possible non-contiguousness */
 768         obj_sz = rte_mempool_calc_obj_size(mbuf_sz, 0, NULL);
 769         if (obj_sz > pgsz) {
 770                 TESTPMD_LOG(ERR, "Object size is bigger than page size\n");
 771                 return -1;
 772         }
 773
 774         mbuf_per_pg = pgsz / obj_sz;
 775         leftover = (nb_mbufs % mbuf_per_pg) > 0;
 776         n_pages = (nb_mbufs / mbuf_per_pg) + leftover;
 777
 778         mbuf_mem = n_pages * pgsz;
 779
 780         total_mem = RTE_ALIGN(hdr_mem + mbuf_mem, pgsz);
 781
 782         if (total_mem > SIZE_MAX) {
 783                 TESTPMD_LOG(ERR, "Memory size too big\n");
 784                 return -1;
 785         }
 786         *out = (size_t)total_mem;
 787
 788         return 0;
 789 }
 790
 791 static int
 792 pagesz_flags(uint64_t page_sz)
 793 {
 794         /* as per mmap() manpage, all page sizes are log2 of page size
 795          * shifted by MAP_HUGE_SHIFT
 796          */
 797         int log2 = rte_log2_u64(page_sz);
 798
 799         return (log2 << HUGE_SHIFT);
 800 }
 801
 802 static void *
 803 alloc_mem(size_t memsz, size_t pgsz, bool huge)
 804 {
 805         void *addr;
 806         int flags;
 807
 808         /* allocate anonymous hugepages */
 809         flags = MAP_ANONYMOUS | MAP_PRIVATE;
 810         if (huge)
 811                 flags |= HUGE_FLAG | pagesz_flags(pgsz);
 812
 813         addr = mmap(NULL, memsz, PROT_READ | PROT_WRITE, flags, -1, 0);
 814         if (addr == MAP_FAILED)
 815                 return NULL;
 816
 817         return addr;
 818 }
 819
 820 struct extmem_param {
 821         void *addr;
 822         size_t len;
 823         size_t pgsz;
 824         rte_iova_t *iova_table;
 825         unsigned int iova_table_len;
 826 };
 827
 828 static int
 829 create_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, struct extmem_param *param,
 830                 bool huge)
 831 {
 832         uint64_t pgsizes[] = {RTE_PGSIZE_2M, RTE_PGSIZE_1G, /* x86_64, ARM */
 833                         RTE_PGSIZE_16M, RTE_PGSIZE_16G};    /* POWER */
 834         unsigned int cur_page, n_pages, pgsz_idx;
 835         size_t mem_sz, cur_pgsz;
 836         rte_iova_t *iovas = NULL;
 837         void *addr;
 838         int ret;
 839
 840         for (pgsz_idx = 0; pgsz_idx < RTE_DIM(pgsizes); pgsz_idx++) {
 841                 /* skip anything that is too big */
 842                 if (pgsizes[pgsz_idx] > SIZE_MAX)
 843                         continue;
 844
 845                 cur_pgsz = pgsizes[pgsz_idx];
 846
 847                 /* if we were told not to allocate hugepages, override */
 848                 if (!huge)
 849                         cur_pgsz = sysconf(_SC_PAGESIZE);
 850
 851                 ret = calc_mem_size(nb_mbufs, mbuf_sz, cur_pgsz, &mem_sz);
 852                 if (ret < 0) {
 853                         TESTPMD_LOG(ERR, "Cannot calculate memory size\n");
 854                         return -1;
 855                 }
 856
 857                 /* allocate our memory */
 858                 addr = alloc_mem(mem_sz, cur_pgsz, huge);
 859
 860                 /* if we couldn't allocate memory with a specified page size,
 861                  * that doesn't mean we can't do it with other page sizes, so
 862                  * try another one.
 863                  */
 864                 if (addr == NULL)
 865                         continue;
 866
 867                 /* store IOVA addresses for every page in this memory area */
 868                 n_pages = mem_sz / cur_pgsz;
 869
 870                 iovas = malloc(sizeof(*iovas) * n_pages);
 871
 872                 if (iovas == NULL) {
 873                         TESTPMD_LOG(ERR, "Cannot allocate memory for iova addresses\n");
 874                         goto fail;
 875                 }
 876                 /* lock memory if it's not huge pages */
 877                 if (!huge)
 878                         mlock(addr, mem_sz);
 879
 880                 /* populate IOVA addresses */
 881                 for (cur_page = 0; cur_page < n_pages; cur_page++) {
 882                         rte_iova_t iova;
 883                         size_t offset;
 884                         void *cur;
 885
 886                         offset = cur_pgsz * cur_page;
 887                         cur = RTE_PTR_ADD(addr, offset);
 888
 889                         /* touch the page before getting its IOVA */
 890                         *(volatile char *)cur = 0;
 891
 892                         iova = rte_mem_virt2iova(cur);
 893
 894                         iovas[cur_page] = iova;
 895                 }
 896
 897                 break;
 898         }
 899         /* if we couldn't allocate anything */
 900         if (iovas == NULL)
 901                 return -1;
 902
 903         param->addr = addr;
 904         param->len = mem_sz;
 905         param->pgsz = cur_pgsz;
 906         param->iova_table = iovas;
 907         param->iova_table_len = n_pages;
 908
 909         return 0;
 910 fail:
 911         free(iovas);
 912         if (addr)
 913                 munmap(addr, mem_sz);
 914
 915         return -1;
 916 }
 917
 918 static int
 919 setup_extmem(uint32_t nb_mbufs, uint32_t mbuf_sz, bool huge)
 920 {
 921         struct extmem_param param;
 922         int socket_id, ret;
 923
 924         memset(&param, 0, sizeof(param));
 925
 926         /* check if our heap exists */
 927         socket_id = rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
 928         if (socket_id < 0) {
 929                 /* create our heap */
 930                 ret = rte_malloc_heap_create(EXTMEM_HEAP_NAME);
 931                 if (ret < 0) {
 932                         TESTPMD_LOG(ERR, "Cannot create heap\n");
 933                         return -1;
 934                 }
 935         }
 936
 937         ret = create_extmem(nb_mbufs, mbuf_sz, &param, huge);
 938         if (ret < 0) {
 939                 TESTPMD_LOG(ERR, "Cannot create memory area\n");
 940                 return -1;
 941         }
 942
 943         /* we now have a valid memory area, so add it to heap */
 944         ret = rte_malloc_heap_memory_add(EXTMEM_HEAP_NAME,
 945                         param.addr, param.len, param.iova_table,
 946                         param.iova_table_len, param.pgsz);
 947
 948         /* when using VFIO, memory is automatically mapped for DMA by EAL */
 949
 950         /* not needed any more */
 951         free(param.iova_table);
 952
 953         if (ret < 0) {
 954                 TESTPMD_LOG(ERR, "Cannot add memory to heap\n");
 955                 munmap(param.addr, param.len);
 956                 return -1;
 957         }
 958
 959         /* success */
 960
 961         TESTPMD_LOG(DEBUG, "Allocated %zuMB of external memory\n",
 962                         param.len >> 20);
 963
 964         return 0;
 965 }
 966 static void
 967 dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
 968              struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
 969 {
 970         uint16_t pid = 0;
 971         int ret;
 972
 973         RTE_ETH_FOREACH_DEV(pid) {
 974                 struct rte_eth_dev_info dev_info;
 975
 976                 ret = eth_dev_info_get_print_err(pid, &dev_info);
 977                 if (ret != 0) {
 978                         TESTPMD_LOG(DEBUG,
 979                                     "unable to get device info for port %d on addr 0x%p,"
 980                                     "mempool unmapping will not be performed\n",
 981                                     pid, memhdr->addr);
 982                         continue;
 983                 }
 984
 985                 ret = rte_dev_dma_unmap(dev_info.device, memhdr->addr, 0, memhdr->len);
 986                 if (ret) {
 987                         TESTPMD_LOG(DEBUG,
 988                                     "unable to DMA unmap addr 0x%p "
 989                                     "for device %s\n",
 990                                     memhdr->addr, dev_info.device->name);
 991                 }
 992         }
 993         ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
 994         if (ret) {
 995                 TESTPMD_LOG(DEBUG,
 996                             "unable to un-register addr 0x%p\n", memhdr->addr);
 997         }
 998 }
 999
1000 static void
1001 dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
1002            struct rte_mempool_memhdr *memhdr, unsigned mem_idx __rte_unused)
1003 {
1004         uint16_t pid = 0;
1005         size_t page_size = sysconf(_SC_PAGESIZE);
1006         int ret;
1007
1008         ret = rte_extmem_register(memhdr->addr, memhdr->len, NULL, 0,
1009                                   page_size);
1010         if (ret) {
1011                 TESTPMD_LOG(DEBUG,
1012                             "unable to register addr 0x%p\n", memhdr->addr);
1013                 return;
1014         }
1015         RTE_ETH_FOREACH_DEV(pid) {
1016                 struct rte_eth_dev_info dev_info;
1017
1018                 ret = eth_dev_info_get_print_err(pid, &dev_info);
1019                 if (ret != 0) {
1020                         TESTPMD_LOG(DEBUG,
1021                                     "unable to get device info for port %d on addr 0x%p,"
1022                                     "mempool mapping will not be performed\n",
1023                                     pid, memhdr->addr);
1024                         continue;
1025                 }
1026                 ret = rte_dev_dma_map(dev_info.device, memhdr->addr, 0, memhdr->len);
1027                 if (ret) {
1028                         TESTPMD_LOG(DEBUG,
1029                                     "unable to DMA map addr 0x%p "
1030                                     "for device %s\n",
1031                                     memhdr->addr, dev_info.device->name);
1032                 }
1033         }
1034 }
1035 #endif
1036
1037 static unsigned int
1038 setup_extbuf(uint32_t nb_mbufs, uint16_t mbuf_sz, unsigned int socket_id,
1039             char *pool_name, struct rte_pktmbuf_extmem **ext_mem)
1040 {
1041         struct rte_pktmbuf_extmem *xmem;
1042         unsigned int ext_num, zone_num, elt_num;
1043         uint16_t elt_size;
1044
1045         elt_size = RTE_ALIGN_CEIL(mbuf_sz, RTE_CACHE_LINE_SIZE);
1046         elt_num = EXTBUF_ZONE_SIZE / elt_size;
1047         zone_num = (nb_mbufs + elt_num - 1) / elt_num;
1048
1049         xmem = malloc(sizeof(struct rte_pktmbuf_extmem) * zone_num);
1050         if (xmem == NULL) {
1051                 TESTPMD_LOG(ERR, "Cannot allocate memory for "
1052                                  "external buffer descriptors\n");
1053                 *ext_mem = NULL;
1054                 return 0;
1055         }
1056         for (ext_num = 0; ext_num < zone_num; ext_num++) {
1057                 struct rte_pktmbuf_extmem *xseg = xmem + ext_num;
1058                 const struct rte_memzone *mz;
1059                 char mz_name[RTE_MEMZONE_NAMESIZE];
1060                 int ret;
1061
1062                 ret = snprintf(mz_name, sizeof(mz_name),
1063                         RTE_MEMPOOL_MZ_FORMAT "_xb_%u", pool_name, ext_num);
1064                 if (ret < 0 || ret >= (int)sizeof(mz_name)) {
1065                         errno = ENAMETOOLONG;
1066                         ext_num = 0;
1067                         break;
1068                 }
1069                 mz = rte_memzone_reserve(mz_name, EXTBUF_ZONE_SIZE,
1070                                          socket_id,
1071                                          RTE_MEMZONE_IOVA_CONTIG |
1072                                          RTE_MEMZONE_1GB |
1073                                          RTE_MEMZONE_SIZE_HINT_ONLY);
1074                 if (mz == NULL) {
1075                         /*
1076                          * The caller exits on external buffer creation
1077                          * error, so there is no need to free memzones.
1078                          */
1079                         errno = ENOMEM;
1080                         ext_num = 0;
1081                         break;
1082                 }
1083                 xseg->buf_ptr = mz->addr;
1084                 xseg->buf_iova = mz->iova;
1085                 xseg->buf_len = EXTBUF_ZONE_SIZE;
1086                 xseg->elt_size = elt_size;
1087         }
1088         if (ext_num == 0 && xmem != NULL) {
1089                 free(xmem);
1090                 xmem = NULL;
1091         }
1092         *ext_mem = xmem;
1093         return ext_num;
1094 }
1095
1096 /*
1097  * Configuration initialisation done once at init time.
1098  */
1099 static struct rte_mempool *
1100 mbuf_pool_create(uint16_t mbuf_seg_size, unsigned nb_mbuf,
1101                  unsigned int socket_id, uint16_t size_idx)
1102 {
1103         char pool_name[RTE_MEMPOOL_NAMESIZE];
1104         struct rte_mempool *rte_mp = NULL;
1105 #ifndef RTE_EXEC_ENV_WINDOWS
1106         uint32_t mb_size;
1107
1108         mb_size = sizeof(struct rte_mbuf) + mbuf_seg_size;
1109 #endif
1110         mbuf_poolname_build(socket_id, pool_name, sizeof(pool_name), size_idx);
1111         if (!is_proc_primary()) {
1112                 rte_mp = rte_mempool_lookup(pool_name);
1113                 if (rte_mp == NULL)
1114                         rte_exit(EXIT_FAILURE,
1115                                 "Get mbuf pool for socket %u failed: %s\n",
1116                                 socket_id, rte_strerror(rte_errno));
1117                 return rte_mp;
1118         }
1119
1120         TESTPMD_LOG(INFO,
1121                 "create a new mbuf pool <%s>: n=%u, size=%u, socket=%u\n",
1122                 pool_name, nb_mbuf, mbuf_seg_size, socket_id);
1123
1124         switch (mp_alloc_type) {
1125         case MP_ALLOC_NATIVE:
1126                 {
1127                         /* wrapper to rte_mempool_create() */
1128                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1129                                         rte_mbuf_best_mempool_ops());
1130                         rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1131                                 mb_mempool_cache, 0, mbuf_seg_size, socket_id);
1132                         break;
1133                 }
1134 #ifndef RTE_EXEC_ENV_WINDOWS
1135         case MP_ALLOC_ANON:
1136                 {
1137                         rte_mp = rte_mempool_create_empty(pool_name, nb_mbuf,
1138                                 mb_size, (unsigned int) mb_mempool_cache,
1139                                 sizeof(struct rte_pktmbuf_pool_private),
1140                                 socket_id, mempool_flags);
1141                         if (rte_mp == NULL)
1142                                 goto err;
1143
1144                         if (rte_mempool_populate_anon(rte_mp) == 0) {
1145                                 rte_mempool_free(rte_mp);
1146                                 rte_mp = NULL;
1147                                 goto err;
1148                         }
1149                         rte_pktmbuf_pool_init(rte_mp, NULL);
1150                         rte_mempool_obj_iter(rte_mp, rte_pktmbuf_init, NULL);
1151                         rte_mempool_mem_iter(rte_mp, dma_map_cb, NULL);
1152                         break;
1153                 }
1154         case MP_ALLOC_XMEM:
1155         case MP_ALLOC_XMEM_HUGE:
1156                 {
1157                         int heap_socket;
1158                         bool huge = mp_alloc_type == MP_ALLOC_XMEM_HUGE;
1159
1160                         if (setup_extmem(nb_mbuf, mbuf_seg_size, huge) < 0)
1161                                 rte_exit(EXIT_FAILURE, "Could not create external memory\n");
1162
1163                         heap_socket =
1164                                 rte_malloc_heap_get_socket(EXTMEM_HEAP_NAME);
1165                         if (heap_socket < 0)
1166                                 rte_exit(EXIT_FAILURE, "Could not get external memory socket ID\n");
1167
1168                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1169                                         rte_mbuf_best_mempool_ops());
1170                         rte_mp = rte_pktmbuf_pool_create(pool_name, nb_mbuf,
1171                                         mb_mempool_cache, 0, mbuf_seg_size,
1172                                         heap_socket);
1173                         break;
1174                 }
1175 #endif
1176         case MP_ALLOC_XBUF:
1177                 {
1178                         struct rte_pktmbuf_extmem *ext_mem;
1179                         unsigned int ext_num;
1180
1181                         ext_num = setup_extbuf(nb_mbuf, mbuf_seg_size,
1182                                                socket_id, pool_name, &ext_mem);
1183                         if (ext_num == 0)
1184                                 rte_exit(EXIT_FAILURE,
1185                                          "Can't create pinned data buffers\n");
1186
1187                         TESTPMD_LOG(INFO, "preferred mempool ops selected: %s\n",
1188                                         rte_mbuf_best_mempool_ops());
1189                         rte_mp = rte_pktmbuf_pool_create_extbuf
1190                                         (pool_name, nb_mbuf, mb_mempool_cache,
1191                                          0, mbuf_seg_size, socket_id,
1192                                          ext_mem, ext_num);
1193                         free(ext_mem);
1194                         break;
1195                 }
1196         default:
1197                 {
1198                         rte_exit(EXIT_FAILURE, "Invalid mempool creation mode\n");
1199                 }
1200         }
1201
1202 #ifndef RTE_EXEC_ENV_WINDOWS
1203 err:
1204 #endif
1205         if (rte_mp == NULL) {
1206                 rte_exit(EXIT_FAILURE,
1207                         "Creation of mbuf pool for socket %u failed: %s\n",
1208                         socket_id, rte_strerror(rte_errno));
1209         } else if (verbose_level > 0) {
1210                 rte_mempool_dump(stdout, rte_mp);
1211         }
1212         return rte_mp;
1213 }
1214
1215 /*
1216  * Check given socket id is valid or not with NUMA mode,
1217  * if valid, return 0, else return -1
1218  */
1219 static int
1220 check_socket_id(const unsigned int socket_id)
1221 {
1222         static int warning_once = 0;
1223
1224         if (new_socket_id(socket_id)) {
1225                 if (!warning_once && numa_support)
1226                         fprintf(stderr,
1227                                 "Warning: NUMA should be configured manually by using --port-numa-config and --ring-numa-config parameters along with --numa.\n");
1228                 warning_once = 1;
1229                 return -1;
1230         }
1231         return 0;
1232 }
1233
1234 /*
1235  * Get the allowed maximum number of RX queues.
1236  * *pid return the port id which has minimal value of
1237  * max_rx_queues in all ports.
1238  */
1239 queueid_t
1240 get_allowed_max_nb_rxq(portid_t *pid)
1241 {
1242         queueid_t allowed_max_rxq = RTE_MAX_QUEUES_PER_PORT;
1243         bool max_rxq_valid = false;
1244         portid_t pi;
1245         struct rte_eth_dev_info dev_info;
1246
1247         RTE_ETH_FOREACH_DEV(pi) {
1248                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1249                         continue;
1250
1251                 max_rxq_valid = true;
1252                 if (dev_info.max_rx_queues < allowed_max_rxq) {
1253                         allowed_max_rxq = dev_info.max_rx_queues;
1254                         *pid = pi;
1255                 }
1256         }
1257         return max_rxq_valid ? allowed_max_rxq : 0;
1258 }
1259
1260 /*
1261  * Check input rxq is valid or not.
1262  * If input rxq is not greater than any of maximum number
1263  * of RX queues of all ports, it is valid.
1264  * if valid, return 0, else return -1
1265  */
1266 int
1267 check_nb_rxq(queueid_t rxq)
1268 {
1269         queueid_t allowed_max_rxq;
1270         portid_t pid = 0;
1271
1272         allowed_max_rxq = get_allowed_max_nb_rxq(&pid);
1273         if (rxq > allowed_max_rxq) {
1274                 fprintf(stderr,
1275                         "Fail: input rxq (%u) can't be greater than max_rx_queues (%u) of port %u\n",
1276                         rxq, allowed_max_rxq, pid);
1277                 return -1;
1278         }
1279         return 0;
1280 }
1281
1282 /*
1283  * Get the allowed maximum number of TX queues.
1284  * *pid return the port id which has minimal value of
1285  * max_tx_queues in all ports.
1286  */
1287 queueid_t
1288 get_allowed_max_nb_txq(portid_t *pid)
1289 {
1290         queueid_t allowed_max_txq = RTE_MAX_QUEUES_PER_PORT;
1291         bool max_txq_valid = false;
1292         portid_t pi;
1293         struct rte_eth_dev_info dev_info;
1294
1295         RTE_ETH_FOREACH_DEV(pi) {
1296                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1297                         continue;
1298
1299                 max_txq_valid = true;
1300                 if (dev_info.max_tx_queues < allowed_max_txq) {
1301                         allowed_max_txq = dev_info.max_tx_queues;
1302                         *pid = pi;
1303                 }
1304         }
1305         return max_txq_valid ? allowed_max_txq : 0;
1306 }
1307
1308 /*
1309  * Check input txq is valid or not.
1310  * If input txq is not greater than any of maximum number
1311  * of TX queues of all ports, it is valid.
1312  * if valid, return 0, else return -1
1313  */
1314 int
1315 check_nb_txq(queueid_t txq)
1316 {
1317         queueid_t allowed_max_txq;
1318         portid_t pid = 0;
1319
1320         allowed_max_txq = get_allowed_max_nb_txq(&pid);
1321         if (txq > allowed_max_txq) {
1322                 fprintf(stderr,
1323                         "Fail: input txq (%u) can't be greater than max_tx_queues (%u) of port %u\n",
1324                         txq, allowed_max_txq, pid);
1325                 return -1;
1326         }
1327         return 0;
1328 }
1329
1330 /*
1331  * Get the allowed maximum number of RXDs of every rx queue.
1332  * *pid return the port id which has minimal value of
1333  * max_rxd in all queues of all ports.
1334  */
1335 static uint16_t
1336 get_allowed_max_nb_rxd(portid_t *pid)
1337 {
1338         uint16_t allowed_max_rxd = UINT16_MAX;
1339         portid_t pi;
1340         struct rte_eth_dev_info dev_info;
1341
1342         RTE_ETH_FOREACH_DEV(pi) {
1343                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1344                         continue;
1345
1346                 if (dev_info.rx_desc_lim.nb_max < allowed_max_rxd) {
1347                         allowed_max_rxd = dev_info.rx_desc_lim.nb_max;
1348                         *pid = pi;
1349                 }
1350         }
1351         return allowed_max_rxd;
1352 }
1353
1354 /*
1355  * Get the allowed minimal number of RXDs of every rx queue.
1356  * *pid return the port id which has minimal value of
1357  * min_rxd in all queues of all ports.
1358  */
1359 static uint16_t
1360 get_allowed_min_nb_rxd(portid_t *pid)
1361 {
1362         uint16_t allowed_min_rxd = 0;
1363         portid_t pi;
1364         struct rte_eth_dev_info dev_info;
1365
1366         RTE_ETH_FOREACH_DEV(pi) {
1367                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1368                         continue;
1369
1370                 if (dev_info.rx_desc_lim.nb_min > allowed_min_rxd) {
1371                         allowed_min_rxd = dev_info.rx_desc_lim.nb_min;
1372                         *pid = pi;
1373                 }
1374         }
1375
1376         return allowed_min_rxd;
1377 }
1378
1379 /*
1380  * Check input rxd is valid or not.
1381  * If input rxd is not greater than any of maximum number
1382  * of RXDs of every Rx queues and is not less than any of
1383  * minimal number of RXDs of every Rx queues, it is valid.
1384  * if valid, return 0, else return -1
1385  */
1386 int
1387 check_nb_rxd(queueid_t rxd)
1388 {
1389         uint16_t allowed_max_rxd;
1390         uint16_t allowed_min_rxd;
1391         portid_t pid = 0;
1392
1393         allowed_max_rxd = get_allowed_max_nb_rxd(&pid);
1394         if (rxd > allowed_max_rxd) {
1395                 fprintf(stderr,
1396                         "Fail: input rxd (%u) can't be greater than max_rxds (%u) of port %u\n",
1397                         rxd, allowed_max_rxd, pid);
1398                 return -1;
1399         }
1400
1401         allowed_min_rxd = get_allowed_min_nb_rxd(&pid);
1402         if (rxd < allowed_min_rxd) {
1403                 fprintf(stderr,
1404                         "Fail: input rxd (%u) can't be less than min_rxds (%u) of port %u\n",
1405                         rxd, allowed_min_rxd, pid);
1406                 return -1;
1407         }
1408
1409         return 0;
1410 }
1411
1412 /*
1413  * Get the allowed maximum number of TXDs of every rx queues.
1414  * *pid return the port id which has minimal value of
1415  * max_txd in every tx queue.
1416  */
1417 static uint16_t
1418 get_allowed_max_nb_txd(portid_t *pid)
1419 {
1420         uint16_t allowed_max_txd = UINT16_MAX;
1421         portid_t pi;
1422         struct rte_eth_dev_info dev_info;
1423
1424         RTE_ETH_FOREACH_DEV(pi) {
1425                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1426                         continue;
1427
1428                 if (dev_info.tx_desc_lim.nb_max < allowed_max_txd) {
1429                         allowed_max_txd = dev_info.tx_desc_lim.nb_max;
1430                         *pid = pi;
1431                 }
1432         }
1433         return allowed_max_txd;
1434 }
1435
1436 /*
1437  * Get the allowed maximum number of TXDs of every tx queues.
1438  * *pid return the port id which has minimal value of
1439  * min_txd in every tx queue.
1440  */
1441 static uint16_t
1442 get_allowed_min_nb_txd(portid_t *pid)
1443 {
1444         uint16_t allowed_min_txd = 0;
1445         portid_t pi;
1446         struct rte_eth_dev_info dev_info;
1447
1448         RTE_ETH_FOREACH_DEV(pi) {
1449                 if (eth_dev_info_get_print_err(pi, &dev_info) != 0)
1450                         continue;
1451
1452                 if (dev_info.tx_desc_lim.nb_min > allowed_min_txd) {
1453                         allowed_min_txd = dev_info.tx_desc_lim.nb_min;
1454                         *pid = pi;
1455                 }
1456         }
1457
1458         return allowed_min_txd;
1459 }
1460
1461 /*
1462  * Check input txd is valid or not.
1463  * If input txd is not greater than any of maximum number
1464  * of TXDs of every Rx queues, it is valid.
1465  * if valid, return 0, else return -1
1466  */
1467 int
1468 check_nb_txd(queueid_t txd)
1469 {
1470         uint16_t allowed_max_txd;
1471         uint16_t allowed_min_txd;
1472         portid_t pid = 0;
1473
1474         allowed_max_txd = get_allowed_max_nb_txd(&pid);
1475         if (txd > allowed_max_txd) {
1476                 fprintf(stderr,
1477                         "Fail: input txd (%u) can't be greater than max_txds (%u) of port %u\n",
1478                         txd, allowed_max_txd, pid);
1479                 return -1;
1480         }
1481
1482         allowed_min_txd = get_allowed_min_nb_txd(&pid);
1483         if (txd < allowed_min_txd) {
1484                 fprintf(stderr,
1485                         "Fail: input txd (%u) can't be less than min_txds (%u) of port %u\n",
1486                         txd, allowed_min_txd, pid);
1487                 return -1;
1488         }
1489         return 0;
1490 }
1491
1492
1493 /*
1494  * Get the allowed maximum number of hairpin queues.
1495  * *pid return the port id which has minimal value of
1496  * max_hairpin_queues in all ports.
1497  */
1498 queueid_t
1499 get_allowed_max_nb_hairpinq(portid_t *pid)
1500 {
1501         queueid_t allowed_max_hairpinq = RTE_MAX_QUEUES_PER_PORT;
1502         portid_t pi;
1503         struct rte_eth_hairpin_cap cap;
1504
1505         RTE_ETH_FOREACH_DEV(pi) {
1506                 if (rte_eth_dev_hairpin_capability_get(pi, &cap) != 0) {
1507                         *pid = pi;
1508                         return 0;
1509                 }
1510                 if (cap.max_nb_queues < allowed_max_hairpinq) {
1511                         allowed_max_hairpinq = cap.max_nb_queues;
1512                         *pid = pi;
1513                 }
1514         }
1515         return allowed_max_hairpinq;
1516 }
1517
1518 /*
1519  * Check input hairpin is valid or not.
1520  * If input hairpin is not greater than any of maximum number
1521  * of hairpin queues of all ports, it is valid.
1522  * if valid, return 0, else return -1
1523  */
1524 int
1525 check_nb_hairpinq(queueid_t hairpinq)
1526 {
1527         queueid_t allowed_max_hairpinq;
1528         portid_t pid = 0;
1529
1530         allowed_max_hairpinq = get_allowed_max_nb_hairpinq(&pid);
1531         if (hairpinq > allowed_max_hairpinq) {
1532                 fprintf(stderr,
1533                         "Fail: input hairpin (%u) can't be greater than max_hairpin_queues (%u) of port %u\n",
1534                         hairpinq, allowed_max_hairpinq, pid);
1535                 return -1;
1536         }
1537         return 0;
1538 }
1539
1540 static int
1541 get_eth_overhead(struct rte_eth_dev_info *dev_info)
1542 {
1543         uint32_t eth_overhead;
1544
1545         if (dev_info->max_mtu != UINT16_MAX &&
1546             dev_info->max_rx_pktlen > dev_info->max_mtu)
1547                 eth_overhead = dev_info->max_rx_pktlen - dev_info->max_mtu;
1548         else
1549                 eth_overhead = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
1550
1551         return eth_overhead;
1552 }
1553
1554 static void
1555 init_config_port_offloads(portid_t pid, uint32_t socket_id)
1556 {
1557         struct rte_port *port = &ports[pid];
1558         int ret;
1559         int i;
1560
1561         eth_rx_metadata_negotiate_mp(pid);
1562
1563         port->dev_conf.txmode = tx_mode;
1564         port->dev_conf.rxmode = rx_mode;
1565
1566         ret = eth_dev_info_get_print_err(pid, &port->dev_info);
1567         if (ret != 0)
1568                 rte_exit(EXIT_FAILURE, "rte_eth_dev_info_get() failed\n");
1569
1570         if (!(port->dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE))
1571                 port->dev_conf.txmode.offloads &=
1572                         ~RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
1573
1574         /* Apply Rx offloads configuration */
1575         for (i = 0; i < port->dev_info.max_rx_queues; i++)
1576                 port->rx_conf[i].offloads = port->dev_conf.rxmode.offloads;
1577         /* Apply Tx offloads configuration */
1578         for (i = 0; i < port->dev_info.max_tx_queues; i++)
1579                 port->tx_conf[i].offloads = port->dev_conf.txmode.offloads;
1580
1581         if (eth_link_speed)
1582                 port->dev_conf.link_speeds = eth_link_speed;
1583
1584         if (max_rx_pkt_len)
1585                 port->dev_conf.rxmode.mtu = max_rx_pkt_len -
1586                         get_eth_overhead(&port->dev_info);
1587
1588         /* set flag to initialize port/queue */
1589         port->need_reconfig = 1;
1590         port->need_reconfig_queues = 1;
1591         port->socket_id = socket_id;
1592         port->tx_metadata = 0;
1593
1594         /*
1595          * Check for maximum number of segments per MTU.
1596          * Accordingly update the mbuf data size.
1597          */
1598         if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
1599             port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
1600                 uint32_t eth_overhead = get_eth_overhead(&port->dev_info);
1601                 uint16_t mtu;
1602
1603                 if (rte_eth_dev_get_mtu(pid, &mtu) == 0) {
1604                         uint16_t data_size = (mtu + eth_overhead) /
1605                                 port->dev_info.rx_desc_lim.nb_mtu_seg_max;
1606                         uint16_t buffer_size = data_size + RTE_PKTMBUF_HEADROOM;
1607
1608                         if (buffer_size > mbuf_data_size[0]) {
1609                                 mbuf_data_size[0] = buffer_size;
1610                                 TESTPMD_LOG(WARNING,
1611                                         "Configured mbuf size of the first segment %hu\n",
1612                                         mbuf_data_size[0]);
1613                         }
1614                 }
1615         }
1616 }
1617
1618 static void
1619 init_config(void)
1620 {
1621         portid_t pid;
1622         struct rte_mempool *mbp;
1623         unsigned int nb_mbuf_per_pool;
1624         lcoreid_t  lc_id;
1625 #ifdef RTE_LIB_GRO
1626         struct rte_gro_param gro_param;
1627 #endif
1628 #ifdef RTE_LIB_GSO
1629         uint32_t gso_types;
1630 #endif
1631
1632         /* Configuration of logical cores. */
1633         fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
1634                                 sizeof(struct fwd_lcore *) * nb_lcores,
1635                                 RTE_CACHE_LINE_SIZE);
1636         if (fwd_lcores == NULL) {
1637                 rte_exit(EXIT_FAILURE, "rte_zmalloc(%d (struct fwd_lcore *)) "
1638                                                         "failed\n", nb_lcores);
1639         }
1640         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1641                 fwd_lcores[lc_id] = rte_zmalloc("testpmd: struct fwd_lcore",
1642                                                sizeof(struct fwd_lcore),
1643                                                RTE_CACHE_LINE_SIZE);
1644                 if (fwd_lcores[lc_id] == NULL) {
1645                         rte_exit(EXIT_FAILURE, "rte_zmalloc(struct fwd_lcore) "
1646                                                                 "failed\n");
1647                 }
1648                 fwd_lcores[lc_id]->cpuid_idx = lc_id;
1649         }
1650
1651         RTE_ETH_FOREACH_DEV(pid) {
1652                 uint32_t socket_id;
1653
1654                 if (numa_support) {
1655                         socket_id = port_numa[pid];
1656                         if (port_numa[pid] == NUMA_NO_CONFIG) {
1657                                 socket_id = rte_eth_dev_socket_id(pid);
1658
1659                                 /*
1660                                  * if socket_id is invalid,
1661                                  * set to the first available socket.
1662                                  */
1663                                 if (check_socket_id(socket_id) < 0)
1664                                         socket_id = socket_ids[0];
1665                         }
1666                 } else {
1667                         socket_id = (socket_num == UMA_NO_CONFIG) ?
1668                                     0 : socket_num;
1669                 }
1670                 /* Apply default TxRx configuration for all ports */
1671                 init_config_port_offloads(pid, socket_id);
1672         }
1673         /*
1674          * Create pools of mbuf.
1675          * If NUMA support is disabled, create a single pool of mbuf in
1676          * socket 0 memory by default.
1677          * Otherwise, create a pool of mbuf in the memory of sockets 0 and 1.
1678          *
1679          * Use the maximum value of nb_rxd and nb_txd here, then nb_rxd and
1680          * nb_txd can be configured at run time.
1681          */
1682         if (param_total_num_mbufs)
1683                 nb_mbuf_per_pool = param_total_num_mbufs;
1684         else {
1685                 nb_mbuf_per_pool = RTE_TEST_RX_DESC_MAX +
1686                         (nb_lcores * mb_mempool_cache) +
1687                         RTE_TEST_TX_DESC_MAX + MAX_PKT_BURST;
1688                 nb_mbuf_per_pool *= RTE_MAX_ETHPORTS;
1689         }
1690
1691         if (numa_support) {
1692                 uint8_t i, j;
1693
1694                 for (i = 0; i < num_sockets; i++)
1695                         for (j = 0; j < mbuf_data_size_n; j++)
1696                                 mempools[i * MAX_SEGS_BUFFER_SPLIT + j] =
1697                                         mbuf_pool_create(mbuf_data_size[j],
1698                                                           nb_mbuf_per_pool,
1699                                                           socket_ids[i], j);
1700         } else {
1701                 uint8_t i;
1702
1703                 for (i = 0; i < mbuf_data_size_n; i++)
1704                         mempools[i] = mbuf_pool_create
1705                                         (mbuf_data_size[i],
1706                                          nb_mbuf_per_pool,
1707                                          socket_num == UMA_NO_CONFIG ?
1708                                          0 : socket_num, i);
1709         }
1710
1711         init_port_config();
1712
1713 #ifdef RTE_LIB_GSO
1714         gso_types = RTE_ETH_TX_OFFLOAD_TCP_TSO | RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
1715                 RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO | RTE_ETH_TX_OFFLOAD_UDP_TSO;
1716 #endif
1717         /*
1718          * Records which Mbuf pool to use by each logical core, if needed.
1719          */
1720         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1721                 mbp = mbuf_pool_find(
1722                         rte_lcore_to_socket_id(fwd_lcores_cpuids[lc_id]), 0);
1723
1724                 if (mbp == NULL)
1725                         mbp = mbuf_pool_find(0, 0);
1726                 fwd_lcores[lc_id]->mbp = mbp;
1727 #ifdef RTE_LIB_GSO
1728                 /* initialize GSO context */
1729                 fwd_lcores[lc_id]->gso_ctx.direct_pool = mbp;
1730                 fwd_lcores[lc_id]->gso_ctx.indirect_pool = mbp;
1731                 fwd_lcores[lc_id]->gso_ctx.gso_types = gso_types;
1732                 fwd_lcores[lc_id]->gso_ctx.gso_size = RTE_ETHER_MAX_LEN -
1733                         RTE_ETHER_CRC_LEN;
1734                 fwd_lcores[lc_id]->gso_ctx.flag = 0;
1735 #endif
1736         }
1737
1738         fwd_config_setup();
1739
1740 #ifdef RTE_LIB_GRO
1741         /* create a gro context for each lcore */
1742         gro_param.gro_types = RTE_GRO_TCP_IPV4;
1743         gro_param.max_flow_num = GRO_MAX_FLUSH_CYCLES;
1744         gro_param.max_item_per_flow = MAX_PKT_BURST;
1745         for (lc_id = 0; lc_id < nb_lcores; lc_id++) {
1746                 gro_param.socket_id = rte_lcore_to_socket_id(
1747                                 fwd_lcores_cpuids[lc_id]);
1748                 fwd_lcores[lc_id]->gro_ctx = rte_gro_ctx_create(&gro_param);
1749                 if (fwd_lcores[lc_id]->gro_ctx == NULL) {
1750                         rte_exit(EXIT_FAILURE,
1751                                         "rte_gro_ctx_create() failed\n");
1752                 }
1753         }
1754 #endif
1755 }
1756
1757
1758 void
1759 reconfig(portid_t new_port_id, unsigned socket_id)
1760 {
1761         /* Reconfiguration of Ethernet ports. */
1762         init_config_port_offloads(new_port_id, socket_id);
1763         init_port_config();
1764 }
1765
1766
1767 int
1768 init_fwd_streams(void)
1769 {
1770         portid_t pid;
1771         struct rte_port *port;
1772         streamid_t sm_id, nb_fwd_streams_new;
1773         queueid_t q;
1774
1775         /* set socket id according to numa or not */
1776         RTE_ETH_FOREACH_DEV(pid) {
1777                 port = &ports[pid];
1778                 if (nb_rxq > port->dev_info.max_rx_queues) {
1779                         fprintf(stderr,
1780                                 "Fail: nb_rxq(%d) is greater than max_rx_queues(%d)\n",
1781                                 nb_rxq, port->dev_info.max_rx_queues);
1782                         return -1;
1783                 }
1784                 if (nb_txq > port->dev_info.max_tx_queues) {
1785                         fprintf(stderr,
1786                                 "Fail: nb_txq(%d) is greater than max_tx_queues(%d)\n",
1787                                 nb_txq, port->dev_info.max_tx_queues);
1788                         return -1;
1789                 }
1790                 if (numa_support) {
1791                         if (port_numa[pid] != NUMA_NO_CONFIG)
1792                                 port->socket_id = port_numa[pid];
1793                         else {
1794                                 port->socket_id = rte_eth_dev_socket_id(pid);
1795
1796                                 /*
1797                                  * if socket_id is invalid,
1798                                  * set to the first available socket.
1799                                  */
1800                                 if (check_socket_id(port->socket_id) < 0)
1801                                         port->socket_id = socket_ids[0];
1802                         }
1803                 }
1804                 else {
1805                         if (socket_num == UMA_NO_CONFIG)
1806                                 port->socket_id = 0;
1807                         else
1808                                 port->socket_id = socket_num;
1809                 }
1810         }
1811
1812         q = RTE_MAX(nb_rxq, nb_txq);
1813         if (q == 0) {
1814                 fprintf(stderr,
1815                         "Fail: Cannot allocate fwd streams as number of queues is 0\n");
1816                 return -1;
1817         }
1818         nb_fwd_streams_new = (streamid_t)(nb_ports * q);
1819         if (nb_fwd_streams_new == nb_fwd_streams)
1820                 return 0;
1821         /* clear the old */
1822         if (fwd_streams != NULL) {
1823                 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1824                         if (fwd_streams[sm_id] == NULL)
1825                                 continue;
1826                         rte_free(fwd_streams[sm_id]);
1827                         fwd_streams[sm_id] = NULL;
1828                 }
1829                 rte_free(fwd_streams);
1830                 fwd_streams = NULL;
1831         }
1832
1833         /* init new */
1834         nb_fwd_streams = nb_fwd_streams_new;
1835         if (nb_fwd_streams) {
1836                 fwd_streams = rte_zmalloc("testpmd: fwd_streams",
1837                         sizeof(struct fwd_stream *) * nb_fwd_streams,
1838                         RTE_CACHE_LINE_SIZE);
1839                 if (fwd_streams == NULL)
1840                         rte_exit(EXIT_FAILURE, "rte_zmalloc(%d"
1841                                  " (struct fwd_stream *)) failed\n",
1842                                  nb_fwd_streams);
1843
1844                 for (sm_id = 0; sm_id < nb_fwd_streams; sm_id++) {
1845                         fwd_streams[sm_id] = rte_zmalloc("testpmd:"
1846                                 " struct fwd_stream", sizeof(struct fwd_stream),
1847                                 RTE_CACHE_LINE_SIZE);
1848                         if (fwd_streams[sm_id] == NULL)
1849                                 rte_exit(EXIT_FAILURE, "rte_zmalloc"
1850                                          "(struct fwd_stream) failed\n");
1851                 }
1852         }
1853
1854         return 0;
1855 }
1856
1857 static void
1858 pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
1859 {
1860         uint64_t total_burst, sburst;
1861         uint64_t nb_burst;
1862         uint64_t burst_stats[4];
1863         uint16_t pktnb_stats[4];
1864         uint16_t nb_pkt;
1865         int burst_percent[4], sburstp;
1866         int i;
1867
1868         /*
1869          * First compute the total number of packet bursts and the
1870          * two highest numbers of bursts of the same number of packets.
1871          */
1872         memset(&burst_stats, 0x0, sizeof(burst_stats));
1873         memset(&pktnb_stats, 0x0, sizeof(pktnb_stats));
1874
1875         /* Show stats for 0 burst size always */
1876         total_burst = pbs->pkt_burst_spread[0];
1877         burst_stats[0] = pbs->pkt_burst_spread[0];
1878         pktnb_stats[0] = 0;
1879
1880         /* Find the next 2 burst sizes with highest occurrences. */
1881         for (nb_pkt = 1; nb_pkt < MAX_PKT_BURST + 1; nb_pkt++) {
1882                 nb_burst = pbs->pkt_burst_spread[nb_pkt];
1883
1884                 if (nb_burst == 0)
1885                         continue;
1886
1887                 total_burst += nb_burst;
1888
1889                 if (nb_burst > burst_stats[1]) {
1890                         burst_stats[2] = burst_stats[1];
1891                         pktnb_stats[2] = pktnb_stats[1];
1892                         burst_stats[1] = nb_burst;
1893                         pktnb_stats[1] = nb_pkt;
1894                 } else if (nb_burst > burst_stats[2]) {
1895                         burst_stats[2] = nb_burst;
1896                         pktnb_stats[2] = nb_pkt;
1897                 }
1898         }
1899         if (total_burst == 0)
1900                 return;
1901
1902         printf("  %s-bursts : %"PRIu64" [", rx_tx, total_burst);
1903         for (i = 0, sburst = 0, sburstp = 0; i < 4; i++) {
1904                 if (i == 3) {
1905                         printf("%d%% of other]\n", 100 - sburstp);
1906                         return;
1907                 }
1908
1909                 sburst += burst_stats[i];
1910                 if (sburst == total_burst) {
1911                         printf("%d%% of %d pkts]\n",
1912                                 100 - sburstp, (int) pktnb_stats[i]);
1913                         return;
1914                 }
1915
1916                 burst_percent[i] =
1917                         (double)burst_stats[i] / total_burst * 100;
1918                 printf("%d%% of %d pkts + ",
1919                         burst_percent[i], (int) pktnb_stats[i]);
1920                 sburstp += burst_percent[i];
1921         }
1922 }
1923
1924 static void
1925 fwd_stream_stats_display(streamid_t stream_id)
1926 {
1927         struct fwd_stream *fs;
1928         static const char *fwd_top_stats_border = "-------";
1929
1930         fs = fwd_streams[stream_id];
1931         if ((fs->rx_packets == 0) && (fs->tx_packets == 0) &&
1932             (fs->fwd_dropped == 0))
1933                 return;
1934         printf("\n  %s Forward Stats for RX Port=%2d/Queue=%2d -> "
1935                "TX Port=%2d/Queue=%2d %s\n",
1936                fwd_top_stats_border, fs->rx_port, fs->rx_queue,
1937                fs->tx_port, fs->tx_queue, fwd_top_stats_border);
1938         printf("  RX-packets: %-14"PRIu64" TX-packets: %-14"PRIu64
1939                " TX-dropped: %-14"PRIu64,
1940                fs->rx_packets, fs->tx_packets, fs->fwd_dropped);
1941
1942         /* if checksum mode */
1943         if (cur_fwd_eng == &csum_fwd_engine) {
1944                 printf("  RX- bad IP checksum: %-14"PRIu64
1945                        "  Rx- bad L4 checksum: %-14"PRIu64
1946                        " Rx- bad outer L4 checksum: %-14"PRIu64"\n",
1947                         fs->rx_bad_ip_csum, fs->rx_bad_l4_csum,
1948                         fs->rx_bad_outer_l4_csum);
1949                 printf(" RX- bad outer IP checksum: %-14"PRIu64"\n",
1950                         fs->rx_bad_outer_ip_csum);
1951         } else {
1952                 printf("\n");
1953         }
1954
1955         if (record_burst_stats) {
1956                 pkt_burst_stats_display("RX", &fs->rx_burst_stats);
1957                 pkt_burst_stats_display("TX", &fs->tx_burst_stats);
1958         }
1959 }
1960
1961 void
1962 fwd_stats_display(void)
1963 {
1964         static const char *fwd_stats_border = "----------------------";
1965         static const char *acc_stats_border = "+++++++++++++++";
1966         struct {
1967                 struct fwd_stream *rx_stream;
1968                 struct fwd_stream *tx_stream;
1969                 uint64_t tx_dropped;
1970                 uint64_t rx_bad_ip_csum;
1971                 uint64_t rx_bad_l4_csum;
1972                 uint64_t rx_bad_outer_l4_csum;
1973                 uint64_t rx_bad_outer_ip_csum;
1974         } ports_stats[RTE_MAX_ETHPORTS];
1975         uint64_t total_rx_dropped = 0;
1976         uint64_t total_tx_dropped = 0;
1977         uint64_t total_rx_nombuf = 0;
1978         struct rte_eth_stats stats;
1979         uint64_t fwd_cycles = 0;
1980         uint64_t total_recv = 0;
1981         uint64_t total_xmit = 0;
1982         struct rte_port *port;
1983         streamid_t sm_id;
1984         portid_t pt_id;
1985         int i;
1986
1987         memset(ports_stats, 0, sizeof(ports_stats));
1988
1989         for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
1990                 struct fwd_stream *fs = fwd_streams[sm_id];
1991
1992                 if (cur_fwd_config.nb_fwd_streams >
1993                     cur_fwd_config.nb_fwd_ports) {
1994                         fwd_stream_stats_display(sm_id);
1995                 } else {
1996                         ports_stats[fs->tx_port].tx_stream = fs;
1997                         ports_stats[fs->rx_port].rx_stream = fs;
1998                 }
1999
2000                 ports_stats[fs->tx_port].tx_dropped += fs->fwd_dropped;
2001
2002                 ports_stats[fs->rx_port].rx_bad_ip_csum += fs->rx_bad_ip_csum;
2003                 ports_stats[fs->rx_port].rx_bad_l4_csum += fs->rx_bad_l4_csum;
2004                 ports_stats[fs->rx_port].rx_bad_outer_l4_csum +=
2005                                 fs->rx_bad_outer_l4_csum;
2006                 ports_stats[fs->rx_port].rx_bad_outer_ip_csum +=
2007                                 fs->rx_bad_outer_ip_csum;
2008
2009                 if (record_core_cycles)
2010                         fwd_cycles += fs->core_cycles;
2011         }
2012         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2013                 pt_id = fwd_ports_ids[i];
2014                 port = &ports[pt_id];
2015
2016                 rte_eth_stats_get(pt_id, &stats);
2017                 stats.ipackets -= port->stats.ipackets;
2018                 stats.opackets -= port->stats.opackets;
2019                 stats.ibytes -= port->stats.ibytes;
2020                 stats.obytes -= port->stats.obytes;
2021                 stats.imissed -= port->stats.imissed;
2022                 stats.oerrors -= port->stats.oerrors;
2023                 stats.rx_nombuf -= port->stats.rx_nombuf;
2024
2025                 total_recv += stats.ipackets;
2026                 total_xmit += stats.opackets;
2027                 total_rx_dropped += stats.imissed;
2028                 total_tx_dropped += ports_stats[pt_id].tx_dropped;
2029                 total_tx_dropped += stats.oerrors;
2030                 total_rx_nombuf  += stats.rx_nombuf;
2031
2032                 printf("\n  %s Forward statistics for port %-2d %s\n",
2033                        fwd_stats_border, pt_id, fwd_stats_border);
2034
2035                 printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64
2036                        "RX-total: %-"PRIu64"\n", stats.ipackets, stats.imissed,
2037                        stats.ipackets + stats.imissed);
2038
2039                 if (cur_fwd_eng == &csum_fwd_engine) {
2040                         printf("  Bad-ipcsum: %-14"PRIu64
2041                                " Bad-l4csum: %-14"PRIu64
2042                                "Bad-outer-l4csum: %-14"PRIu64"\n",
2043                                ports_stats[pt_id].rx_bad_ip_csum,
2044                                ports_stats[pt_id].rx_bad_l4_csum,
2045                                ports_stats[pt_id].rx_bad_outer_l4_csum);
2046                         printf("  Bad-outer-ipcsum: %-14"PRIu64"\n",
2047                                ports_stats[pt_id].rx_bad_outer_ip_csum);
2048                 }
2049                 if (stats.ierrors + stats.rx_nombuf > 0) {
2050                         printf("  RX-error: %-"PRIu64"\n", stats.ierrors);
2051                         printf("  RX-nombufs: %-14"PRIu64"\n", stats.rx_nombuf);
2052                 }
2053
2054                 printf("  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64
2055                        "TX-total: %-"PRIu64"\n",
2056                        stats.opackets, ports_stats[pt_id].tx_dropped,
2057                        stats.opackets + ports_stats[pt_id].tx_dropped);
2058
2059                 if (record_burst_stats) {
2060                         if (ports_stats[pt_id].rx_stream)
2061                                 pkt_burst_stats_display("RX",
2062                                         &ports_stats[pt_id].rx_stream->rx_burst_stats);
2063                         if (ports_stats[pt_id].tx_stream)
2064                                 pkt_burst_stats_display("TX",
2065                                 &ports_stats[pt_id].tx_stream->tx_burst_stats);
2066                 }
2067
2068                 printf("  %s--------------------------------%s\n",
2069                        fwd_stats_border, fwd_stats_border);
2070         }
2071
2072         printf("\n  %s Accumulated forward statistics for all ports"
2073                "%s\n",
2074                acc_stats_border, acc_stats_border);
2075         printf("  RX-packets: %-14"PRIu64" RX-dropped: %-14"PRIu64"RX-total: "
2076                "%-"PRIu64"\n"
2077                "  TX-packets: %-14"PRIu64" TX-dropped: %-14"PRIu64"TX-total: "
2078                "%-"PRIu64"\n",
2079                total_recv, total_rx_dropped, total_recv + total_rx_dropped,
2080                total_xmit, total_tx_dropped, total_xmit + total_tx_dropped);
2081         if (total_rx_nombuf > 0)
2082                 printf("  RX-nombufs: %-14"PRIu64"\n", total_rx_nombuf);
2083         printf("  %s++++++++++++++++++++++++++++++++++++++++++++++"
2084                "%s\n",
2085                acc_stats_border, acc_stats_border);
2086         if (record_core_cycles) {
2087 #define CYC_PER_MHZ 1E6
2088                 if (total_recv > 0 || total_xmit > 0) {
2089                         uint64_t total_pkts = 0;
2090                         if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 ||
2091                             strcmp(cur_fwd_eng->fwd_mode_name, "flowgen") == 0)
2092                                 total_pkts = total_xmit;
2093                         else
2094                                 total_pkts = total_recv;
2095
2096                         printf("\n  CPU cycles/packet=%.2F (total cycles="
2097                                "%"PRIu64" / total %s packets=%"PRIu64") at %"PRIu64
2098                                " MHz Clock\n",
2099                                (double) fwd_cycles / total_pkts,
2100                                fwd_cycles, cur_fwd_eng->fwd_mode_name, total_pkts,
2101                                (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
2102                 }
2103         }
2104 }
2105
2106 void
2107 fwd_stats_reset(void)
2108 {
2109         streamid_t sm_id;
2110         portid_t pt_id;
2111         int i;
2112
2113         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2114                 pt_id = fwd_ports_ids[i];
2115                 rte_eth_stats_get(pt_id, &ports[pt_id].stats);
2116         }
2117         for (sm_id = 0; sm_id < cur_fwd_config.nb_fwd_streams; sm_id++) {
2118                 struct fwd_stream *fs = fwd_streams[sm_id];
2119
2120                 fs->rx_packets = 0;
2121                 fs->tx_packets = 0;
2122                 fs->fwd_dropped = 0;
2123                 fs->rx_bad_ip_csum = 0;
2124                 fs->rx_bad_l4_csum = 0;
2125                 fs->rx_bad_outer_l4_csum = 0;
2126                 fs->rx_bad_outer_ip_csum = 0;
2127
2128                 memset(&fs->rx_burst_stats, 0, sizeof(fs->rx_burst_stats));
2129                 memset(&fs->tx_burst_stats, 0, sizeof(fs->tx_burst_stats));
2130                 fs->core_cycles = 0;
2131         }
2132 }
2133
2134 static void
2135 flush_fwd_rx_queues(void)
2136 {
2137         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2138         portid_t  rxp;
2139         portid_t port_id;
2140         queueid_t rxq;
2141         uint16_t  nb_rx;
2142         uint16_t  i;
2143         uint8_t   j;
2144         uint64_t prev_tsc = 0, diff_tsc, cur_tsc, timer_tsc = 0;
2145         uint64_t timer_period;
2146
2147         if (num_procs > 1) {
2148                 printf("multi-process not support for flushing fwd Rx queues, skip the below lines and return.\n");
2149                 return;
2150         }
2151
2152         /* convert to number of cycles */
2153         timer_period = rte_get_timer_hz(); /* 1 second timeout */
2154
2155         for (j = 0; j < 2; j++) {
2156                 for (rxp = 0; rxp < cur_fwd_config.nb_fwd_ports; rxp++) {
2157                         for (rxq = 0; rxq < nb_rxq; rxq++) {
2158                                 port_id = fwd_ports_ids[rxp];
2159                                 /**
2160                                 * testpmd can stuck in the below do while loop
2161                                 * if rte_eth_rx_burst() always returns nonzero
2162                                 * packets. So timer is added to exit this loop
2163                                 * after 1sec timer expiry.
2164                                 */
2165                                 prev_tsc = rte_rdtsc();
2166                                 do {
2167                                         nb_rx = rte_eth_rx_burst(port_id, rxq,
2168                                                 pkts_burst, MAX_PKT_BURST);
2169                                         for (i = 0; i < nb_rx; i++)
2170                                                 rte_pktmbuf_free(pkts_burst[i]);
2171
2172                                         cur_tsc = rte_rdtsc();
2173                                         diff_tsc = cur_tsc - prev_tsc;
2174                                         timer_tsc += diff_tsc;
2175                                 } while ((nb_rx > 0) &&
2176                                         (timer_tsc < timer_period));
2177                                 timer_tsc = 0;
2178                         }
2179                 }
2180                 rte_delay_ms(10); /* wait 10 milli-seconds before retrying */
2181         }
2182 }
2183
2184 static void
2185 run_pkt_fwd_on_lcore(struct fwd_lcore *fc, packet_fwd_t pkt_fwd)
2186 {
2187         struct fwd_stream **fsm;
2188         streamid_t nb_fs;
2189         streamid_t sm_id;
2190 #ifdef RTE_LIB_BITRATESTATS
2191         uint64_t tics_per_1sec;
2192         uint64_t tics_datum;
2193         uint64_t tics_current;
2194         uint16_t i, cnt_ports;
2195
2196         cnt_ports = nb_ports;
2197         tics_datum = rte_rdtsc();
2198         tics_per_1sec = rte_get_timer_hz();
2199 #endif
2200         fsm = &fwd_streams[fc->stream_idx];
2201         nb_fs = fc->stream_nb;
2202         do {
2203                 for (sm_id = 0; sm_id < nb_fs; sm_id++)
2204                         (*pkt_fwd)(fsm[sm_id]);
2205 #ifdef RTE_LIB_BITRATESTATS
2206                 if (bitrate_enabled != 0 &&
2207                                 bitrate_lcore_id == rte_lcore_id()) {
2208                         tics_current = rte_rdtsc();
2209                         if (tics_current - tics_datum >= tics_per_1sec) {
2210                                 /* Periodic bitrate calculation */
2211                                 for (i = 0; i < cnt_ports; i++)
2212                                         rte_stats_bitrate_calc(bitrate_data,
2213                                                 ports_ids[i]);
2214                                 tics_datum = tics_current;
2215                         }
2216                 }
2217 #endif
2218 #ifdef RTE_LIB_LATENCYSTATS
2219                 if (latencystats_enabled != 0 &&
2220                                 latencystats_lcore_id == rte_lcore_id())
2221                         rte_latencystats_update();
2222 #endif
2223
2224         } while (! fc->stopped);
2225 }
2226
2227 static int
2228 start_pkt_forward_on_core(void *fwd_arg)
2229 {
2230         run_pkt_fwd_on_lcore((struct fwd_lcore *) fwd_arg,
2231                              cur_fwd_config.fwd_eng->packet_fwd);
2232         return 0;
2233 }
2234
2235 /*
2236  * Run the TXONLY packet forwarding engine to send a single burst of packets.
2237  * Used to start communication flows in network loopback test configurations.
2238  */
2239 static int
2240 run_one_txonly_burst_on_core(void *fwd_arg)
2241 {
2242         struct fwd_lcore *fwd_lc;
2243         struct fwd_lcore tmp_lcore;
2244
2245         fwd_lc = (struct fwd_lcore *) fwd_arg;
2246         tmp_lcore = *fwd_lc;
2247         tmp_lcore.stopped = 1;
2248         run_pkt_fwd_on_lcore(&tmp_lcore, tx_only_engine.packet_fwd);
2249         return 0;
2250 }
2251
2252 /*
2253  * Launch packet forwarding:
2254  *     - Setup per-port forwarding context.
2255  *     - launch logical cores with their forwarding configuration.
2256  */
2257 static void
2258 launch_packet_forwarding(lcore_function_t *pkt_fwd_on_lcore)
2259 {
2260         unsigned int i;
2261         unsigned int lc_id;
2262         int diag;
2263
2264         for (i = 0; i < cur_fwd_config.nb_fwd_lcores; i++) {
2265                 lc_id = fwd_lcores_cpuids[i];
2266                 if ((interactive == 0) || (lc_id != rte_lcore_id())) {
2267                         fwd_lcores[i]->stopped = 0;
2268                         diag = rte_eal_remote_launch(pkt_fwd_on_lcore,
2269                                                      fwd_lcores[i], lc_id);
2270                         if (diag != 0)
2271                                 fprintf(stderr,
2272                                         "launch lcore %u failed - diag=%d\n",
2273                                         lc_id, diag);
2274                 }
2275         }
2276 }
2277
2278 /*
2279  * Launch packet forwarding configuration.
2280  */
2281 void
2282 start_packet_forwarding(int with_tx_first)
2283 {
2284         port_fwd_begin_t port_fwd_begin;
2285         port_fwd_end_t  port_fwd_end;
2286         unsigned int i;
2287
2288         if (strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") == 0 && !nb_rxq)
2289                 rte_exit(EXIT_FAILURE, "rxq are 0, cannot use rxonly fwd mode\n");
2290
2291         if (strcmp(cur_fwd_eng->fwd_mode_name, "txonly") == 0 && !nb_txq)
2292                 rte_exit(EXIT_FAILURE, "txq are 0, cannot use txonly fwd mode\n");
2293
2294         if ((strcmp(cur_fwd_eng->fwd_mode_name, "rxonly") != 0 &&
2295                 strcmp(cur_fwd_eng->fwd_mode_name, "txonly") != 0) &&
2296                 (!nb_rxq || !nb_txq))
2297                 rte_exit(EXIT_FAILURE,
2298                         "Either rxq or txq are 0, cannot use %s fwd mode\n",
2299                         cur_fwd_eng->fwd_mode_name);
2300
2301         if (all_ports_started() == 0) {
2302                 fprintf(stderr, "Not all ports were started\n");
2303                 return;
2304         }
2305         if (test_done == 0) {
2306                 fprintf(stderr, "Packet forwarding already started\n");
2307                 return;
2308         }
2309
2310         fwd_config_setup();
2311
2312         pkt_fwd_config_display(&cur_fwd_config);
2313         if (!pkt_fwd_shared_rxq_check())
2314                 return;
2315
2316         port_fwd_begin = cur_fwd_config.fwd_eng->port_fwd_begin;
2317         if (port_fwd_begin != NULL) {
2318                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2319                         if (port_fwd_begin(fwd_ports_ids[i])) {
2320                                 fprintf(stderr,
2321                                         "Packet forwarding is not ready\n");
2322                                 return;
2323                         }
2324                 }
2325         }
2326
2327         if (with_tx_first) {
2328                 port_fwd_begin = tx_only_engine.port_fwd_begin;
2329                 if (port_fwd_begin != NULL) {
2330                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2331                                 if (port_fwd_begin(fwd_ports_ids[i])) {
2332                                         fprintf(stderr,
2333                                                 "Packet forwarding is not ready\n");
2334                                         return;
2335                                 }
2336                         }
2337                 }
2338         }
2339
2340         test_done = 0;
2341
2342         if(!no_flush_rx)
2343                 flush_fwd_rx_queues();
2344
2345         rxtx_config_display();
2346
2347         fwd_stats_reset();
2348         if (with_tx_first) {
2349                 while (with_tx_first--) {
2350                         launch_packet_forwarding(
2351                                         run_one_txonly_burst_on_core);
2352                         rte_eal_mp_wait_lcore();
2353                 }
2354                 port_fwd_end = tx_only_engine.port_fwd_end;
2355                 if (port_fwd_end != NULL) {
2356                         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
2357                                 (*port_fwd_end)(fwd_ports_ids[i]);
2358                 }
2359         }
2360         launch_packet_forwarding(start_pkt_forward_on_core);
2361 }
2362
2363 void
2364 stop_packet_forwarding(void)
2365 {
2366         port_fwd_end_t port_fwd_end;
2367         lcoreid_t lc_id;
2368         portid_t pt_id;
2369         int i;
2370
2371         if (test_done) {
2372                 fprintf(stderr, "Packet forwarding not started\n");
2373                 return;
2374         }
2375         printf("Telling cores to stop...");
2376         for (lc_id = 0; lc_id < cur_fwd_config.nb_fwd_lcores; lc_id++)
2377                 fwd_lcores[lc_id]->stopped = 1;
2378         printf("\nWaiting for lcores to finish...\n");
2379         rte_eal_mp_wait_lcore();
2380         port_fwd_end = cur_fwd_config.fwd_eng->port_fwd_end;
2381         if (port_fwd_end != NULL) {
2382                 for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
2383                         pt_id = fwd_ports_ids[i];
2384                         (*port_fwd_end)(pt_id);
2385                 }
2386         }
2387
2388         fwd_stats_display();
2389
2390         printf("\nDone.\n");
2391         test_done = 1;
2392 }
2393
2394 void
2395 dev_set_link_up(portid_t pid)
2396 {
2397         if (rte_eth_dev_set_link_up(pid) < 0)
2398                 fprintf(stderr, "\nSet link up fail.\n");
2399 }
2400
2401 void
2402 dev_set_link_down(portid_t pid)
2403 {
2404         if (rte_eth_dev_set_link_down(pid) < 0)
2405                 fprintf(stderr, "\nSet link down fail.\n");
2406 }
2407
2408 static int
2409 all_ports_started(void)
2410 {
2411         portid_t pi;
2412         struct rte_port *port;
2413
2414         RTE_ETH_FOREACH_DEV(pi) {
2415                 port = &ports[pi];
2416                 /* Check if there is a port which is not started */
2417                 if ((port->port_status != RTE_PORT_STARTED) &&
2418                         (port->slave_flag == 0))
2419                         return 0;
2420         }
2421
2422         /* No port is not started */
2423         return 1;
2424 }
2425
2426 int
2427 port_is_stopped(portid_t port_id)
2428 {
2429         struct rte_port *port = &ports[port_id];
2430
2431         if ((port->port_status != RTE_PORT_STOPPED) &&
2432             (port->slave_flag == 0))
2433                 return 0;
2434         return 1;
2435 }
2436
2437 int
2438 all_ports_stopped(void)
2439 {
2440         portid_t pi;
2441
2442         RTE_ETH_FOREACH_DEV(pi) {
2443                 if (!port_is_stopped(pi))
2444                         return 0;
2445         }
2446
2447         return 1;
2448 }
2449
2450 int
2451 port_is_started(portid_t port_id)
2452 {
2453         if (port_id_is_invalid(port_id, ENABLED_WARN))
2454                 return 0;
2455
2456         if (ports[port_id].port_status != RTE_PORT_STARTED)
2457                 return 0;
2458
2459         return 1;
2460 }
2461
2462 /* Configure the Rx and Tx hairpin queues for the selected port. */
2463 static int
2464 setup_hairpin_queues(portid_t pi, portid_t p_pi, uint16_t cnt_pi)
2465 {
2466         queueid_t qi;
2467         struct rte_eth_hairpin_conf hairpin_conf = {
2468                 .peer_count = 1,
2469         };
2470         int i;
2471         int diag;
2472         struct rte_port *port = &ports[pi];
2473         uint16_t peer_rx_port = pi;
2474         uint16_t peer_tx_port = pi;
2475         uint32_t manual = 1;
2476         uint32_t tx_exp = hairpin_mode & 0x10;
2477
2478         if (!(hairpin_mode & 0xf)) {
2479                 peer_rx_port = pi;
2480                 peer_tx_port = pi;
2481                 manual = 0;
2482         } else if (hairpin_mode & 0x1) {
2483                 peer_tx_port = rte_eth_find_next_owned_by(pi + 1,
2484                                                        RTE_ETH_DEV_NO_OWNER);
2485                 if (peer_tx_port >= RTE_MAX_ETHPORTS)
2486                         peer_tx_port = rte_eth_find_next_owned_by(0,
2487                                                 RTE_ETH_DEV_NO_OWNER);
2488                 if (p_pi != RTE_MAX_ETHPORTS) {
2489                         peer_rx_port = p_pi;
2490                 } else {
2491                         uint16_t next_pi;
2492
2493                         /* Last port will be the peer RX port of the first. */
2494                         RTE_ETH_FOREACH_DEV(next_pi)
2495                                 peer_rx_port = next_pi;
2496                 }
2497                 manual = 1;
2498         } else if (hairpin_mode & 0x2) {
2499                 if (cnt_pi & 0x1) {
2500                         peer_rx_port = p_pi;
2501                 } else {
2502                         peer_rx_port = rte_eth_find_next_owned_by(pi + 1,
2503                                                 RTE_ETH_DEV_NO_OWNER);
2504                         if (peer_rx_port >= RTE_MAX_ETHPORTS)
2505                                 peer_rx_port = pi;
2506                 }
2507                 peer_tx_port = peer_rx_port;
2508                 manual = 1;
2509         }
2510
2511         for (qi = nb_txq, i = 0; qi < nb_hairpinq + nb_txq; qi++) {
2512                 hairpin_conf.peers[0].port = peer_rx_port;
2513                 hairpin_conf.peers[0].queue = i + nb_rxq;
2514                 hairpin_conf.manual_bind = !!manual;
2515                 hairpin_conf.tx_explicit = !!tx_exp;
2516                 diag = rte_eth_tx_hairpin_queue_setup
2517                         (pi, qi, nb_txd, &hairpin_conf);
2518                 i++;
2519                 if (diag == 0)
2520                         continue;
2521
2522                 /* Fail to setup rx queue, return */
2523                 if (port->port_status == RTE_PORT_HANDLING)
2524                         port->port_status = RTE_PORT_STOPPED;
2525                 else
2526                         fprintf(stderr,
2527                                 "Port %d can not be set back to stopped\n", pi);
2528                 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2529                         pi);
2530                 /* try to reconfigure queues next time */
2531                 port->need_reconfig_queues = 1;
2532                 return -1;
2533         }
2534         for (qi = nb_rxq, i = 0; qi < nb_hairpinq + nb_rxq; qi++) {
2535                 hairpin_conf.peers[0].port = peer_tx_port;
2536                 hairpin_conf.peers[0].queue = i + nb_txq;
2537                 hairpin_conf.manual_bind = !!manual;
2538                 hairpin_conf.tx_explicit = !!tx_exp;
2539                 diag = rte_eth_rx_hairpin_queue_setup
2540                         (pi, qi, nb_rxd, &hairpin_conf);
2541                 i++;
2542                 if (diag == 0)
2543                         continue;
2544
2545                 /* Fail to setup rx queue, return */
2546                 if (port->port_status == RTE_PORT_HANDLING)
2547                         port->port_status = RTE_PORT_STOPPED;
2548                 else
2549                         fprintf(stderr,
2550                                 "Port %d can not be set back to stopped\n", pi);
2551                 fprintf(stderr, "Fail to configure port %d hairpin queues\n",
2552                         pi);
2553                 /* try to reconfigure queues next time */
2554                 port->need_reconfig_queues = 1;
2555                 return -1;
2556         }
2557         return 0;
2558 }
2559
2560 /* Configure the Rx with optional split. */
2561 int
2562 rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
2563                uint16_t nb_rx_desc, unsigned int socket_id,
2564                struct rte_eth_rxconf *rx_conf, struct rte_mempool *mp)
2565 {
2566         union rte_eth_rxseg rx_useg[MAX_SEGS_BUFFER_SPLIT] = {};
2567         unsigned int i, mp_n;
2568         int ret;
2569
2570         if (rx_pkt_nb_segs <= 1 ||
2571             (rx_conf->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) == 0) {
2572                 rx_conf->rx_seg = NULL;
2573                 rx_conf->rx_nseg = 0;
2574                 ret = rte_eth_rx_queue_setup(port_id, rx_queue_id,
2575                                              nb_rx_desc, socket_id,
2576                                              rx_conf, mp);
2577                 return ret;
2578         }
2579         for (i = 0; i < rx_pkt_nb_segs; i++) {
2580                 struct rte_eth_rxseg_split *rx_seg = &rx_useg[i].split;
2581                 struct rte_mempool *mpx;
2582                 /*
2583                  * Use last valid pool for the segments with number
2584                  * exceeding the pool index.
2585                  */
2586                 mp_n = (i > mbuf_data_size_n) ? mbuf_data_size_n - 1 : i;
2587                 mpx = mbuf_pool_find(socket_id, mp_n);
2588                 /* Handle zero as mbuf data buffer size. */
2589                 rx_seg->length = rx_pkt_seg_lengths[i] ?
2590                                    rx_pkt_seg_lengths[i] :
2591                                    mbuf_data_size[mp_n];
2592                 rx_seg->offset = i < rx_pkt_nb_offs ?
2593                                    rx_pkt_seg_offsets[i] : 0;
2594                 rx_seg->mp = mpx ? mpx : mp;
2595         }
2596         rx_conf->rx_nseg = rx_pkt_nb_segs;
2597         rx_conf->rx_seg = rx_useg;
2598         ret = rte_eth_rx_queue_setup(port_id, rx_queue_id, nb_rx_desc,
2599                                     socket_id, rx_conf, NULL);
2600         rx_conf->rx_seg = NULL;
2601         rx_conf->rx_nseg = 0;
2602         return ret;
2603 }
2604
2605 static int
2606 alloc_xstats_display_info(portid_t pi)
2607 {
2608         uint64_t **ids_supp = &ports[pi].xstats_info.ids_supp;
2609         uint64_t **prev_values = &ports[pi].xstats_info.prev_values;
2610         uint64_t **curr_values = &ports[pi].xstats_info.curr_values;
2611
2612         if (xstats_display_num == 0)
2613                 return 0;
2614
2615         *ids_supp = calloc(xstats_display_num, sizeof(**ids_supp));
2616         if (*ids_supp == NULL)
2617                 goto fail_ids_supp;
2618
2619         *prev_values = calloc(xstats_display_num,
2620                               sizeof(**prev_values));
2621         if (*prev_values == NULL)
2622                 goto fail_prev_values;
2623
2624         *curr_values = calloc(xstats_display_num,
2625                               sizeof(**curr_values));
2626         if (*curr_values == NULL)
2627                 goto fail_curr_values;
2628
2629         ports[pi].xstats_info.allocated = true;
2630
2631         return 0;
2632
2633 fail_curr_values:
2634         free(*prev_values);
2635 fail_prev_values:
2636         free(*ids_supp);
2637 fail_ids_supp:
2638         return -ENOMEM;
2639 }
2640
2641 static void
2642 free_xstats_display_info(portid_t pi)
2643 {
2644         if (!ports[pi].xstats_info.allocated)
2645                 return;
2646         free(ports[pi].xstats_info.ids_supp);
2647         free(ports[pi].xstats_info.prev_values);
2648         free(ports[pi].xstats_info.curr_values);
2649         ports[pi].xstats_info.allocated = false;
2650 }
2651
2652 /** Fill helper structures for specified port to show extended statistics. */
2653 static void
2654 fill_xstats_display_info_for_port(portid_t pi)
2655 {
2656         unsigned int stat, stat_supp;
2657         const char *xstat_name;
2658         struct rte_port *port;
2659         uint64_t *ids_supp;
2660         int rc;
2661
2662         if (xstats_display_num == 0)
2663                 return;
2664
2665         if (pi == (portid_t)RTE_PORT_ALL) {
2666                 fill_xstats_display_info();
2667                 return;
2668         }
2669
2670         port = &ports[pi];
2671         if (port->port_status != RTE_PORT_STARTED)
2672                 return;
2673
2674         if (!port->xstats_info.allocated && alloc_xstats_display_info(pi) != 0)
2675                 rte_exit(EXIT_FAILURE,
2676                          "Failed to allocate xstats display memory\n");
2677
2678         ids_supp = port->xstats_info.ids_supp;
2679         for (stat = stat_supp = 0; stat < xstats_display_num; stat++) {
2680                 xstat_name = xstats_display[stat].name;
2681                 rc = rte_eth_xstats_get_id_by_name(pi, xstat_name,
2682                                                    ids_supp + stat_supp);
2683                 if (rc != 0) {
2684                         fprintf(stderr, "No xstat '%s' on port %u - skip it %u\n",
2685                                 xstat_name, pi, stat);
2686                         continue;
2687                 }
2688                 stat_supp++;
2689         }
2690
2691         port->xstats_info.ids_supp_sz = stat_supp;
2692 }
2693
2694 /** Fill helper structures for all ports to show extended statistics. */
2695 static void
2696 fill_xstats_display_info(void)
2697 {
2698         portid_t pi;
2699
2700         if (xstats_display_num == 0)
2701                 return;
2702
2703         RTE_ETH_FOREACH_DEV(pi)
2704                 fill_xstats_display_info_for_port(pi);
2705 }
2706
2707 int
2708 start_port(portid_t pid)
2709 {
2710         int diag, need_check_link_status = -1;
2711         portid_t pi;
2712         portid_t p_pi = RTE_MAX_ETHPORTS;
2713         portid_t pl[RTE_MAX_ETHPORTS];
2714         portid_t peer_pl[RTE_MAX_ETHPORTS];
2715         uint16_t cnt_pi = 0;
2716         uint16_t cfg_pi = 0;
2717         int peer_pi;
2718         queueid_t qi;
2719         struct rte_port *port;
2720         struct rte_eth_hairpin_cap cap;
2721
2722         if (port_id_is_invalid(pid, ENABLED_WARN))
2723                 return 0;
2724
2725         RTE_ETH_FOREACH_DEV(pi) {
2726                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
2727                         continue;
2728
2729                 need_check_link_status = 0;
2730                 port = &ports[pi];
2731                 if (port->port_status == RTE_PORT_STOPPED)
2732                         port->port_status = RTE_PORT_HANDLING;
2733                 else {
2734                         fprintf(stderr, "Port %d is now not stopped\n", pi);
2735                         continue;
2736                 }
2737
2738                 if (port->need_reconfig > 0) {
2739                         struct rte_eth_conf dev_conf;
2740                         int k;
2741
2742                         port->need_reconfig = 0;
2743
2744                         if (flow_isolate_all) {
2745                                 int ret = port_flow_isolate(pi, 1);
2746                                 if (ret) {
2747                                         fprintf(stderr,
2748                                                 "Failed to apply isolated mode on port %d\n",
2749                                                 pi);
2750                                         return -1;
2751                                 }
2752                         }
2753                         configure_rxtx_dump_callbacks(0);
2754                         printf("Configuring Port %d (socket %u)\n", pi,
2755                                         port->socket_id);
2756                         if (nb_hairpinq > 0 &&
2757                             rte_eth_dev_hairpin_capability_get(pi, &cap)) {
2758                                 fprintf(stderr,
2759                                         "Port %d doesn't support hairpin queues\n",
2760                                         pi);
2761                                 return -1;
2762                         }
2763
2764                         /* configure port */
2765                         diag = eth_dev_configure_mp(pi, nb_rxq + nb_hairpinq,
2766                                                      nb_txq + nb_hairpinq,
2767                                                      &(port->dev_conf));
2768                         if (diag != 0) {
2769                                 if (port->port_status == RTE_PORT_HANDLING)
2770                                         port->port_status = RTE_PORT_STOPPED;
2771                                 else
2772                                         fprintf(stderr,
2773                                                 "Port %d can not be set back to stopped\n",
2774                                                 pi);
2775                                 fprintf(stderr, "Fail to configure port %d\n",
2776                                         pi);
2777                                 /* try to reconfigure port next time */
2778                                 port->need_reconfig = 1;
2779                                 return -1;
2780                         }
2781                         /* get device configuration*/
2782                         if (0 !=
2783                                 eth_dev_conf_get_print_err(pi, &dev_conf)) {
2784                                 fprintf(stderr,
2785                                         "port %d can not get device configuration\n",
2786                                         pi);
2787                                 return -1;
2788                         }
2789                         /* Apply Rx offloads configuration */
2790                         if (dev_conf.rxmode.offloads !=
2791                             port->dev_conf.rxmode.offloads) {
2792                                 port->dev_conf.rxmode.offloads |=
2793                                         dev_conf.rxmode.offloads;
2794                                 for (k = 0;
2795                                      k < port->dev_info.max_rx_queues;
2796                                      k++)
2797                                         port->rx_conf[k].offloads |=
2798                                                 dev_conf.rxmode.offloads;
2799                         }
2800                         /* Apply Tx offloads configuration */
2801                         if (dev_conf.txmode.offloads !=
2802                             port->dev_conf.txmode.offloads) {
2803                                 port->dev_conf.txmode.offloads |=
2804                                         dev_conf.txmode.offloads;
2805                                 for (k = 0;
2806                                      k < port->dev_info.max_tx_queues;
2807                                      k++)
2808                                         port->tx_conf[k].offloads |=
2809                                                 dev_conf.txmode.offloads;
2810                         }
2811                 }
2812                 if (port->need_reconfig_queues > 0 && is_proc_primary()) {
2813                         port->need_reconfig_queues = 0;
2814                         /* setup tx queues */
2815                         for (qi = 0; qi < nb_txq; qi++) {
2816                                 if ((numa_support) &&
2817                                         (txring_numa[pi] != NUMA_NO_CONFIG))
2818                                         diag = rte_eth_tx_queue_setup(pi, qi,
2819                                                 port->nb_tx_desc[qi],
2820                                                 txring_numa[pi],
2821                                                 &(port->tx_conf[qi]));
2822                                 else
2823                                         diag = rte_eth_tx_queue_setup(pi, qi,
2824                                                 port->nb_tx_desc[qi],
2825                                                 port->socket_id,
2826                                                 &(port->tx_conf[qi]));
2827
2828                                 if (diag == 0)
2829                                         continue;
2830
2831                                 /* Fail to setup tx queue, return */
2832                                 if (port->port_status == RTE_PORT_HANDLING)
2833                                         port->port_status = RTE_PORT_STOPPED;
2834                                 else
2835                                         fprintf(stderr,
2836                                                 "Port %d can not be set back to stopped\n",
2837                                                 pi);
2838                                 fprintf(stderr,
2839                                         "Fail to configure port %d tx queues\n",
2840                                         pi);
2841                                 /* try to reconfigure queues next time */
2842                                 port->need_reconfig_queues = 1;
2843                                 return -1;
2844                         }
2845                         for (qi = 0; qi < nb_rxq; qi++) {
2846                                 /* setup rx queues */
2847                                 if ((numa_support) &&
2848                                         (rxring_numa[pi] != NUMA_NO_CONFIG)) {
2849                                         struct rte_mempool * mp =
2850                                                 mbuf_pool_find
2851                                                         (rxring_numa[pi], 0);
2852                                         if (mp == NULL) {
2853                                                 fprintf(stderr,
2854                                                         "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2855                                                         rxring_numa[pi]);
2856                                                 return -1;
2857                                         }
2858
2859                                         diag = rx_queue_setup(pi, qi,
2860                                              port->nb_rx_desc[qi],
2861                                              rxring_numa[pi],
2862                                              &(port->rx_conf[qi]),
2863                                              mp);
2864                                 } else {
2865                                         struct rte_mempool *mp =
2866                                                 mbuf_pool_find
2867                                                         (port->socket_id, 0);
2868                                         if (mp == NULL) {
2869                                                 fprintf(stderr,
2870                                                         "Failed to setup RX queue: No mempool allocation on the socket %d\n",
2871                                                         port->socket_id);
2872                                                 return -1;
2873                                         }
2874                                         diag = rx_queue_setup(pi, qi,
2875                                              port->nb_rx_desc[qi],
2876                                              port->socket_id,
2877                                              &(port->rx_conf[qi]),
2878                                              mp);
2879                                 }
2880                                 if (diag == 0)
2881                                         continue;
2882
2883                                 /* Fail to setup rx queue, return */
2884                                 if (port->port_status == RTE_PORT_HANDLING)
2885                                         port->port_status = RTE_PORT_STOPPED;
2886                                 else
2887                                         fprintf(stderr,
2888                                                 "Port %d can not be set back to stopped\n",
2889                                                 pi);
2890                                 fprintf(stderr,
2891                                         "Fail to configure port %d rx queues\n",
2892                                         pi);
2893                                 /* try to reconfigure queues next time */
2894                                 port->need_reconfig_queues = 1;
2895                                 return -1;
2896                         }
2897                         /* setup hairpin queues */
2898                         if (setup_hairpin_queues(pi, p_pi, cnt_pi) != 0)
2899                                 return -1;
2900                 }
2901                 configure_rxtx_dump_callbacks(verbose_level);
2902                 if (clear_ptypes) {
2903                         diag = rte_eth_dev_set_ptypes(pi, RTE_PTYPE_UNKNOWN,
2904                                         NULL, 0);
2905                         if (diag < 0)
2906                                 fprintf(stderr,
2907                                         "Port %d: Failed to disable Ptype parsing\n",
2908                                         pi);
2909                 }
2910
2911                 p_pi = pi;
2912                 cnt_pi++;
2913
2914                 /* start port */
2915                 diag = eth_dev_start_mp(pi);
2916                 if (diag < 0) {
2917                         fprintf(stderr, "Fail to start port %d: %s\n",
2918                                 pi, rte_strerror(-diag));
2919
2920                         /* Fail to setup rx queue, return */
2921                         if (port->port_status == RTE_PORT_HANDLING)
2922                                 port->port_status = RTE_PORT_STOPPED;
2923                         else
2924                                 fprintf(stderr,
2925                                         "Port %d can not be set back to stopped\n",
2926                                         pi);
2927                         continue;
2928                 }
2929
2930                 if (port->port_status == RTE_PORT_HANDLING)
2931                         port->port_status = RTE_PORT_STARTED;
2932                 else
2933                         fprintf(stderr, "Port %d can not be set into started\n",
2934                                 pi);
2935
2936                 if (eth_macaddr_get_print_err(pi, &port->eth_addr) == 0)
2937                         printf("Port %d: " RTE_ETHER_ADDR_PRT_FMT "\n", pi,
2938                                         RTE_ETHER_ADDR_BYTES(&port->eth_addr));
2939
2940                 /* at least one port started, need checking link status */
2941                 need_check_link_status = 1;
2942
2943                 pl[cfg_pi++] = pi;
2944         }
2945
2946         if (need_check_link_status == 1 && !no_link_check)
2947                 check_all_ports_link_status(RTE_PORT_ALL);
2948         else if (need_check_link_status == 0)
2949                 fprintf(stderr, "Please stop the ports first\n");
2950
2951         if (hairpin_mode & 0xf) {
2952                 uint16_t i;
2953                 int j;
2954
2955                 /* bind all started hairpin ports */
2956                 for (i = 0; i < cfg_pi; i++) {
2957                         pi = pl[i];
2958                         /* bind current Tx to all peer Rx */
2959                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2960                                                         RTE_MAX_ETHPORTS, 1);
2961                         if (peer_pi < 0)
2962                                 return peer_pi;
2963                         for (j = 0; j < peer_pi; j++) {
2964                                 if (!port_is_started(peer_pl[j]))
2965                                         continue;
2966                                 diag = rte_eth_hairpin_bind(pi, peer_pl[j]);
2967                                 if (diag < 0) {
2968                                         fprintf(stderr,
2969                                                 "Error during binding hairpin Tx port %u to %u: %s\n",
2970                                                 pi, peer_pl[j],
2971                                                 rte_strerror(-diag));
2972                                         return -1;
2973                                 }
2974                         }
2975                         /* bind all peer Tx to current Rx */
2976                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
2977                                                         RTE_MAX_ETHPORTS, 0);
2978                         if (peer_pi < 0)
2979                                 return peer_pi;
2980                         for (j = 0; j < peer_pi; j++) {
2981                                 if (!port_is_started(peer_pl[j]))
2982                                         continue;
2983                                 diag = rte_eth_hairpin_bind(peer_pl[j], pi);
2984                                 if (diag < 0) {
2985                                         fprintf(stderr,
2986                                                 "Error during binding hairpin Tx port %u to %u: %s\n",
2987                                                 peer_pl[j], pi,
2988                                                 rte_strerror(-diag));
2989                                         return -1;
2990                                 }
2991                         }
2992                 }
2993         }
2994
2995         fill_xstats_display_info_for_port(pid);
2996
2997         printf("Done\n");
2998         return 0;
2999 }
3000
3001 void
3002 stop_port(portid_t pid)
3003 {
3004         portid_t pi;
3005         struct rte_port *port;
3006         int need_check_link_status = 0;
3007         portid_t peer_pl[RTE_MAX_ETHPORTS];
3008         int peer_pi;
3009
3010         if (port_id_is_invalid(pid, ENABLED_WARN))
3011                 return;
3012
3013         printf("Stopping ports...\n");
3014
3015         RTE_ETH_FOREACH_DEV(pi) {
3016                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
3017                         continue;
3018
3019                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
3020                         fprintf(stderr,
3021                                 "Please remove port %d from forwarding configuration.\n",
3022                                 pi);
3023                         continue;
3024                 }
3025
3026                 if (port_is_bonding_slave(pi)) {
3027                         fprintf(stderr,
3028                                 "Please remove port %d from bonded device.\n",
3029                                 pi);
3030                         continue;
3031                 }
3032
3033                 port = &ports[pi];
3034                 if (port->port_status == RTE_PORT_STARTED)
3035                         port->port_status = RTE_PORT_HANDLING;
3036                 else
3037                         continue;
3038
3039                 if (hairpin_mode & 0xf) {
3040                         int j;
3041
3042                         rte_eth_hairpin_unbind(pi, RTE_MAX_ETHPORTS);
3043                         /* unbind all peer Tx from current Rx */
3044                         peer_pi = rte_eth_hairpin_get_peer_ports(pi, peer_pl,
3045                                                         RTE_MAX_ETHPORTS, 0);
3046                         if (peer_pi < 0)
3047                                 continue;
3048                         for (j = 0; j < peer_pi; j++) {
3049                                 if (!port_is_started(peer_pl[j]))
3050                                         continue;
3051                                 rte_eth_hairpin_unbind(peer_pl[j], pi);
3052                         }
3053                 }
3054
3055                 if (port->flow_list)
3056                         port_flow_flush(pi);
3057
3058                 if (eth_dev_stop_mp(pi) != 0)
3059                         RTE_LOG(ERR, EAL, "rte_eth_dev_stop failed for port %u\n",
3060                                 pi);
3061
3062                 if (port->port_status == RTE_PORT_HANDLING)
3063                         port->port_status = RTE_PORT_STOPPED;
3064                 else
3065                         fprintf(stderr, "Port %d can not be set into stopped\n",
3066                                 pi);
3067                 need_check_link_status = 1;
3068         }
3069         if (need_check_link_status && !no_link_check)
3070                 check_all_ports_link_status(RTE_PORT_ALL);
3071
3072         printf("Done\n");
3073 }
3074
3075 static void
3076 remove_invalid_ports_in(portid_t *array, portid_t *total)
3077 {
3078         portid_t i;
3079         portid_t new_total = 0;
3080
3081         for (i = 0; i < *total; i++)
3082                 if (!port_id_is_invalid(array[i], DISABLED_WARN)) {
3083                         array[new_total] = array[i];
3084                         new_total++;
3085                 }
3086         *total = new_total;
3087 }
3088
3089 static void
3090 remove_invalid_ports(void)
3091 {
3092         remove_invalid_ports_in(ports_ids, &nb_ports);
3093         remove_invalid_ports_in(fwd_ports_ids, &nb_fwd_ports);
3094         nb_cfg_ports = nb_fwd_ports;
3095 }
3096
3097 void
3098 close_port(portid_t pid)
3099 {
3100         portid_t pi;
3101         struct rte_port *port;
3102
3103         if (port_id_is_invalid(pid, ENABLED_WARN))
3104                 return;
3105
3106         printf("Closing ports...\n");
3107
3108         RTE_ETH_FOREACH_DEV(pi) {
3109                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
3110                         continue;
3111
3112                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
3113                         fprintf(stderr,
3114                                 "Please remove port %d from forwarding configuration.\n",
3115                                 pi);
3116                         continue;
3117                 }
3118
3119                 if (port_is_bonding_slave(pi)) {
3120                         fprintf(stderr,
3121                                 "Please remove port %d from bonded device.\n",
3122                                 pi);
3123                         continue;
3124                 }
3125
3126                 port = &ports[pi];
3127                 if (port->port_status == RTE_PORT_CLOSED) {
3128                         fprintf(stderr, "Port %d is already closed\n", pi);
3129                         continue;
3130                 }
3131
3132                 if (is_proc_primary()) {
3133                         port_flow_flush(pi);
3134                         port_flex_item_flush(pi);
3135                         rte_eth_dev_close(pi);
3136                 }
3137
3138                 free_xstats_display_info(pi);
3139         }
3140
3141         remove_invalid_ports();
3142         printf("Done\n");
3143 }
3144
3145 void
3146 reset_port(portid_t pid)
3147 {
3148         int diag;
3149         portid_t pi;
3150         struct rte_port *port;
3151
3152         if (port_id_is_invalid(pid, ENABLED_WARN))
3153                 return;
3154
3155         if ((pid == (portid_t)RTE_PORT_ALL && !all_ports_stopped()) ||
3156                 (pid != (portid_t)RTE_PORT_ALL && !port_is_stopped(pid))) {
3157                 fprintf(stderr,
3158                         "Can not reset port(s), please stop port(s) first.\n");
3159                 return;
3160         }
3161
3162         printf("Resetting ports...\n");
3163
3164         RTE_ETH_FOREACH_DEV(pi) {
3165                 if (pid != pi && pid != (portid_t)RTE_PORT_ALL)
3166                         continue;
3167
3168                 if (port_is_forwarding(pi) != 0 && test_done == 0) {
3169                         fprintf(stderr,
3170                                 "Please remove port %d from forwarding configuration.\n",
3171                                 pi);
3172                         continue;
3173                 }
3174
3175                 if (port_is_bonding_slave(pi)) {
3176                         fprintf(stderr,
3177                                 "Please remove port %d from bonded device.\n",
3178                                 pi);
3179                         continue;
3180                 }
3181
3182                 diag = rte_eth_dev_reset(pi);
3183                 if (diag == 0) {
3184                         port = &ports[pi];
3185                         port->need_reconfig = 1;
3186                         port->need_reconfig_queues = 1;
3187                 } else {
3188                         fprintf(stderr, "Failed to reset port %d. diag=%d\n",
3189                                 pi, diag);
3190                 }
3191         }
3192
3193         printf("Done\n");
3194 }
3195
3196 void
3197 attach_port(char *identifier)
3198 {
3199         portid_t pi;
3200         struct rte_dev_iterator iterator;
3201
3202         printf("Attaching a new port...\n");
3203
3204         if (identifier == NULL) {
3205                 fprintf(stderr, "Invalid parameters are specified\n");
3206                 return;
3207         }
3208
3209         if (rte_dev_probe(identifier) < 0) {
3210                 TESTPMD_LOG(ERR, "Failed to attach port %s\n", identifier);
3211                 return;
3212         }
3213
3214         /* first attach mode: event */
3215         if (setup_on_probe_event) {
3216                 /* new ports are detected on RTE_ETH_EVENT_NEW event */
3217                 for (pi = 0; pi < RTE_MAX_ETHPORTS; pi++)
3218                         if (ports[pi].port_status == RTE_PORT_HANDLING &&
3219                                         ports[pi].need_setup != 0)
3220                                 setup_attached_port(pi);
3221                 return;
3222         }
3223
3224         /* second attach mode: iterator */
3225         RTE_ETH_FOREACH_MATCHING_DEV(pi, identifier, &iterator) {
3226                 /* setup ports matching the devargs used for probing */
3227                 if (port_is_forwarding(pi))
3228                         continue; /* port was already attached before */
3229                 setup_attached_port(pi);
3230         }
3231 }
3232
3233 static void
3234 setup_attached_port(portid_t pi)
3235 {
3236         unsigned int socket_id;
3237         int ret;
3238
3239         socket_id = (unsigned)rte_eth_dev_socket_id(pi);
3240         /* if socket_id is invalid, set to the first available socket. */
3241         if (check_socket_id(socket_id) < 0)
3242                 socket_id = socket_ids[0];
3243         reconfig(pi, socket_id);
3244         ret = rte_eth_promiscuous_enable(pi);
3245         if (ret != 0)
3246                 fprintf(stderr,
3247                         "Error during enabling promiscuous mode for port %u: %s - ignore\n",
3248                         pi, rte_strerror(-ret));
3249
3250         ports_ids[nb_ports++] = pi;
3251         fwd_ports_ids[nb_fwd_ports++] = pi;
3252         nb_cfg_ports = nb_fwd_ports;
3253         ports[pi].need_setup = 0;
3254         ports[pi].port_status = RTE_PORT_STOPPED;
3255
3256         printf("Port %d is attached. Now total ports is %d\n", pi, nb_ports);
3257         printf("Done\n");
3258 }
3259
3260 static void
3261 detach_device(struct rte_device *dev)
3262 {
3263         portid_t sibling;
3264
3265         if (dev == NULL) {
3266                 fprintf(stderr, "Device already removed\n");
3267                 return;
3268         }
3269
3270         printf("Removing a device...\n");
3271
3272         RTE_ETH_FOREACH_DEV_OF(sibling, dev) {
3273                 if (ports[sibling].port_status != RTE_PORT_CLOSED) {
3274                         if (ports[sibling].port_status != RTE_PORT_STOPPED) {
3275                                 fprintf(stderr, "Port %u not stopped\n",
3276                                         sibling);
3277                                 return;
3278                         }
3279                         port_flow_flush(sibling);
3280                 }
3281         }
3282
3283         if (rte_dev_remove(dev) < 0) {
3284                 TESTPMD_LOG(ERR, "Failed to detach device %s\n", dev->name);
3285                 return;
3286         }
3287         remove_invalid_ports();
3288
3289         printf("Device is detached\n");
3290         printf("Now total ports is %d\n", nb_ports);
3291         printf("Done\n");
3292         return;
3293 }
3294
3295 void
3296 detach_port_device(portid_t port_id)
3297 {
3298         int ret;
3299         struct rte_eth_dev_info dev_info;
3300
3301         if (port_id_is_invalid(port_id, ENABLED_WARN))
3302                 return;
3303
3304         if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3305                 if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3306                         fprintf(stderr, "Port not stopped\n");
3307                         return;
3308                 }
3309                 fprintf(stderr, "Port was not closed\n");
3310         }
3311
3312         ret = eth_dev_info_get_print_err(port_id, &dev_info);
3313         if (ret != 0) {
3314                 TESTPMD_LOG(ERR,
3315                         "Failed to get device info for port %d, not detaching\n",
3316                         port_id);
3317                 return;
3318         }
3319         detach_device(dev_info.device);
3320 }
3321
3322 void
3323 detach_devargs(char *identifier)
3324 {
3325         struct rte_dev_iterator iterator;
3326         struct rte_devargs da;
3327         portid_t port_id;
3328
3329         printf("Removing a device...\n");
3330
3331         memset(&da, 0, sizeof(da));
3332         if (rte_devargs_parsef(&da, "%s", identifier)) {
3333                 fprintf(stderr, "cannot parse identifier\n");
3334                 return;
3335         }
3336
3337         RTE_ETH_FOREACH_MATCHING_DEV(port_id, identifier, &iterator) {
3338                 if (ports[port_id].port_status != RTE_PORT_CLOSED) {
3339                         if (ports[port_id].port_status != RTE_PORT_STOPPED) {
3340                                 fprintf(stderr, "Port %u not stopped\n",
3341                                         port_id);
3342                                 rte_eth_iterator_cleanup(&iterator);
3343                                 rte_devargs_reset(&da);
3344                                 return;
3345                         }
3346                         port_flow_flush(port_id);
3347                 }
3348         }
3349
3350         if (rte_eal_hotplug_remove(da.bus->name, da.name) != 0) {
3351                 TESTPMD_LOG(ERR, "Failed to detach device %s(%s)\n",
3352                             da.name, da.bus->name);
3353                 rte_devargs_reset(&da);
3354                 return;
3355         }
3356
3357         remove_invalid_ports();
3358
3359         printf("Device %s is detached\n", identifier);
3360         printf("Now total ports is %d\n", nb_ports);
3361         printf("Done\n");
3362         rte_devargs_reset(&da);
3363 }
3364
3365 void
3366 pmd_test_exit(void)
3367 {
3368         portid_t pt_id;
3369         unsigned int i;
3370         int ret;
3371
3372         if (test_done == 0)
3373                 stop_packet_forwarding();
3374
3375 #ifndef RTE_EXEC_ENV_WINDOWS
3376         for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3377                 if (mempools[i]) {
3378                         if (mp_alloc_type == MP_ALLOC_ANON)
3379                                 rte_mempool_mem_iter(mempools[i], dma_unmap_cb,
3380                                                      NULL);
3381                 }
3382         }
3383 #endif
3384         if (ports != NULL) {
3385                 no_link_check = 1;
3386                 RTE_ETH_FOREACH_DEV(pt_id) {
3387                         printf("\nStopping port %d...\n", pt_id);
3388                         fflush(stdout);
3389                         stop_port(pt_id);
3390                 }
3391                 RTE_ETH_FOREACH_DEV(pt_id) {
3392                         printf("\nShutting down port %d...\n", pt_id);
3393                         fflush(stdout);
3394                         close_port(pt_id);
3395                 }
3396         }
3397
3398         if (hot_plug) {
3399                 ret = rte_dev_event_monitor_stop();
3400                 if (ret) {
3401                         RTE_LOG(ERR, EAL,
3402                                 "fail to stop device event monitor.");
3403                         return;
3404                 }
3405
3406                 ret = rte_dev_event_callback_unregister(NULL,
3407                         dev_event_callback, NULL);
3408                 if (ret < 0) {
3409                         RTE_LOG(ERR, EAL,
3410                                 "fail to unregister device event callback.\n");
3411                         return;
3412                 }
3413
3414                 ret = rte_dev_hotplug_handle_disable();
3415                 if (ret) {
3416                         RTE_LOG(ERR, EAL,
3417                                 "fail to disable hotplug handling.\n");
3418                         return;
3419                 }
3420         }
3421         for (i = 0 ; i < RTE_DIM(mempools) ; i++) {
3422                 if (mempools[i])
3423                         mempool_free_mp(mempools[i]);
3424         }
3425         free(xstats_display);
3426
3427         printf("\nBye...\n");
3428 }
3429
3430 typedef void (*cmd_func_t)(void);
3431 struct pmd_test_command {
3432         const char *cmd_name;
3433         cmd_func_t cmd_func;
3434 };
3435
3436 /* Check the link status of all ports in up to 9s, and print them finally */
3437 static void
3438 check_all_ports_link_status(uint32_t port_mask)
3439 {
3440 #define CHECK_INTERVAL 100 /* 100ms */
3441 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3442         portid_t portid;
3443         uint8_t count, all_ports_up, print_flag = 0;
3444         struct rte_eth_link link;
3445         int ret;
3446         char link_status[RTE_ETH_LINK_MAX_STR_LEN];
3447
3448         printf("Checking link statuses...\n");
3449         fflush(stdout);
3450         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3451                 all_ports_up = 1;
3452                 RTE_ETH_FOREACH_DEV(portid) {
3453                         if ((port_mask & (1 << portid)) == 0)
3454                                 continue;
3455                         memset(&link, 0, sizeof(link));
3456                         ret = rte_eth_link_get_nowait(portid, &link);
3457                         if (ret < 0) {
3458                                 all_ports_up = 0;
3459                                 if (print_flag == 1)
3460                                         fprintf(stderr,
3461                                                 "Port %u link get failed: %s\n",
3462                                                 portid, rte_strerror(-ret));
3463                                 continue;
3464                         }
3465                         /* print link status if flag set */
3466                         if (print_flag == 1) {
3467                                 rte_eth_link_to_str(link_status,
3468                                         sizeof(link_status), &link);
3469                                 printf("Port %d %s\n", portid, link_status);
3470                                 continue;
3471                         }
3472                         /* clear all_ports_up flag if any link down */
3473                         if (link.link_status == RTE_ETH_LINK_DOWN) {
3474                                 all_ports_up = 0;
3475                                 break;
3476                         }
3477                 }
3478                 /* after finally printing all link status, get out */
3479                 if (print_flag == 1)
3480                         break;
3481
3482                 if (all_ports_up == 0) {
3483                         fflush(stdout);
3484                         rte_delay_ms(CHECK_INTERVAL);
3485                 }
3486
3487                 /* set the print_flag if all ports up or timeout */
3488                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3489                         print_flag = 1;
3490                 }
3491
3492                 if (lsc_interrupt)
3493                         break;
3494         }
3495 }
3496
3497 static void
3498 rmv_port_callback(void *arg)
3499 {
3500         int need_to_start = 0;
3501         int org_no_link_check = no_link_check;
3502         portid_t port_id = (intptr_t)arg;
3503         struct rte_eth_dev_info dev_info;
3504         int ret;
3505
3506         RTE_ETH_VALID_PORTID_OR_RET(port_id);
3507
3508         if (!test_done && port_is_forwarding(port_id)) {
3509                 need_to_start = 1;
3510                 stop_packet_forwarding();
3511         }
3512         no_link_check = 1;
3513         stop_port(port_id);
3514         no_link_check = org_no_link_check;
3515
3516         ret = eth_dev_info_get_print_err(port_id, &dev_info);
3517         if (ret != 0)
3518                 TESTPMD_LOG(ERR,
3519                         "Failed to get device info for port %d, not detaching\n",
3520                         port_id);
3521         else {
3522                 struct rte_device *device = dev_info.device;
3523                 close_port(port_id);
3524                 detach_device(device); /* might be already removed or have more ports */
3525         }
3526         if (need_to_start)
3527                 start_packet_forwarding(0);
3528 }
3529
3530 /* This function is used by the interrupt thread */
3531 static int
3532 eth_event_callback(portid_t port_id, enum rte_eth_event_type type, void *param,
3533                   void *ret_param)
3534 {
3535         RTE_SET_USED(param);
3536         RTE_SET_USED(ret_param);
3537
3538         if (type >= RTE_ETH_EVENT_MAX) {
3539                 fprintf(stderr,
3540                         "\nPort %" PRIu16 ": %s called upon invalid event %d\n",
3541                         port_id, __func__, type);
3542                 fflush(stderr);
3543         } else if (event_print_mask & (UINT32_C(1) << type)) {
3544                 printf("\nPort %" PRIu16 ": %s event\n", port_id,
3545                         eth_event_desc[type]);
3546                 fflush(stdout);
3547         }
3548
3549         switch (type) {
3550         case RTE_ETH_EVENT_NEW:
3551                 ports[port_id].need_setup = 1;
3552                 ports[port_id].port_status = RTE_PORT_HANDLING;
3553                 break;
3554         case RTE_ETH_EVENT_INTR_RMV:
3555                 if (port_id_is_invalid(port_id, DISABLED_WARN))
3556                         break;
3557                 if (rte_eal_alarm_set(100000,
3558                                 rmv_port_callback, (void *)(intptr_t)port_id))
3559                         fprintf(stderr,
3560                                 "Could not set up deferred device removal\n");
3561                 break;
3562         case RTE_ETH_EVENT_DESTROY:
3563                 ports[port_id].port_status = RTE_PORT_CLOSED;
3564                 printf("Port %u is closed\n", port_id);
3565                 break;
3566         default:
3567                 break;
3568         }
3569         return 0;
3570 }
3571
3572 static int
3573 register_eth_event_callback(void)
3574 {
3575         int ret;
3576         enum rte_eth_event_type event;
3577
3578         for (event = RTE_ETH_EVENT_UNKNOWN;
3579                         event < RTE_ETH_EVENT_MAX; event++) {
3580                 ret = rte_eth_dev_callback_register(RTE_ETH_ALL,
3581                                 event,
3582                                 eth_event_callback,
3583                                 NULL);
3584                 if (ret != 0) {
3585                         TESTPMD_LOG(ERR, "Failed to register callback for "
3586                                         "%s event\n", eth_event_desc[event]);
3587                         return -1;
3588                 }
3589         }
3590
3591         return 0;
3592 }
3593
3594 /* This function is used by the interrupt thread */
3595 static void
3596 dev_event_callback(const char *device_name, enum rte_dev_event_type type,
3597                              __rte_unused void *arg)
3598 {
3599         uint16_t port_id;
3600         int ret;
3601
3602         if (type >= RTE_DEV_EVENT_MAX) {
3603                 fprintf(stderr, "%s called upon invalid event %d\n",
3604                         __func__, type);
3605                 fflush(stderr);
3606         }
3607
3608         switch (type) {
3609         case RTE_DEV_EVENT_REMOVE:
3610                 RTE_LOG(DEBUG, EAL, "The device: %s has been removed!\n",
3611                         device_name);
3612                 ret = rte_eth_dev_get_port_by_name(device_name, &port_id);
3613                 if (ret) {
3614                         RTE_LOG(ERR, EAL, "can not get port by device %s!\n",
3615                                 device_name);
3616                         return;
3617                 }
3618                 /*
3619                  * Because the user's callback is invoked in eal interrupt
3620                  * callback, the interrupt callback need to be finished before
3621                  * it can be unregistered when detaching device. So finish
3622                  * callback soon and use a deferred removal to detach device
3623                  * is need. It is a workaround, once the device detaching be
3624                  * moved into the eal in the future, the deferred removal could
3625                  * be deleted.
3626                  */
3627                 if (rte_eal_alarm_set(100000,
3628                                 rmv_port_callback, (void *)(intptr_t)port_id))
3629                         RTE_LOG(ERR, EAL,
3630                                 "Could not set up deferred device removal\n");
3631                 break;
3632         case RTE_DEV_EVENT_ADD:
3633                 RTE_LOG(ERR, EAL, "The device: %s has been added!\n",
3634                         device_name);
3635                 /* TODO: After finish kernel driver binding,
3636                  * begin to attach port.
3637                  */
3638                 break;
3639         default:
3640                 break;
3641         }
3642 }
3643
3644 static void
3645 rxtx_port_config(portid_t pid)
3646 {
3647         uint16_t qid;
3648         uint64_t offloads;
3649         struct rte_port *port = &ports[pid];
3650
3651         for (qid = 0; qid < nb_rxq; qid++) {
3652                 offloads = port->rx_conf[qid].offloads;
3653                 port->rx_conf[qid] = port->dev_info.default_rxconf;
3654
3655                 if (rxq_share > 0 &&
3656                     (port->dev_info.dev_capa & RTE_ETH_DEV_CAPA_RXQ_SHARE)) {
3657                         /* Non-zero share group to enable RxQ share. */
3658                         port->rx_conf[qid].share_group = pid / rxq_share + 1;
3659                         port->rx_conf[qid].share_qid = qid; /* Equal mapping. */
3660                 }
3661
3662                 if (offloads != 0)
3663                         port->rx_conf[qid].offloads = offloads;
3664
3665                 /* Check if any Rx parameters have been passed */
3666                 if (rx_pthresh != RTE_PMD_PARAM_UNSET)
3667                         port->rx_conf[qid].rx_thresh.pthresh = rx_pthresh;
3668
3669                 if (rx_hthresh != RTE_PMD_PARAM_UNSET)
3670                         port->rx_conf[qid].rx_thresh.hthresh = rx_hthresh;
3671
3672                 if (rx_wthresh != RTE_PMD_PARAM_UNSET)
3673                         port->rx_conf[qid].rx_thresh.wthresh = rx_wthresh;
3674
3675                 if (rx_free_thresh != RTE_PMD_PARAM_UNSET)
3676                         port->rx_conf[qid].rx_free_thresh = rx_free_thresh;
3677
3678                 if (rx_drop_en != RTE_PMD_PARAM_UNSET)
3679                         port->rx_conf[qid].rx_drop_en = rx_drop_en;
3680
3681                 port->nb_rx_desc[qid] = nb_rxd;
3682         }
3683
3684         for (qid = 0; qid < nb_txq; qid++) {
3685                 offloads = port->tx_conf[qid].offloads;
3686                 port->tx_conf[qid] = port->dev_info.default_txconf;
3687                 if (offloads != 0)
3688                         port->tx_conf[qid].offloads = offloads;
3689
3690                 /* Check if any Tx parameters have been passed */
3691                 if (tx_pthresh != RTE_PMD_PARAM_UNSET)
3692                         port->tx_conf[qid].tx_thresh.pthresh = tx_pthresh;
3693
3694                 if (tx_hthresh != RTE_PMD_PARAM_UNSET)
3695                         port->tx_conf[qid].tx_thresh.hthresh = tx_hthresh;
3696
3697                 if (tx_wthresh != RTE_PMD_PARAM_UNSET)
3698                         port->tx_conf[qid].tx_thresh.wthresh = tx_wthresh;
3699
3700                 if (tx_rs_thresh != RTE_PMD_PARAM_UNSET)
3701                         port->tx_conf[qid].tx_rs_thresh = tx_rs_thresh;
3702
3703                 if (tx_free_thresh != RTE_PMD_PARAM_UNSET)
3704                         port->tx_conf[qid].tx_free_thresh = tx_free_thresh;
3705
3706                 port->nb_tx_desc[qid] = nb_txd;
3707         }
3708 }
3709
3710 /*
3711  * Helper function to set MTU from frame size
3712  *
3713  * port->dev_info should be set before calling this function.
3714  *
3715  * return 0 on success, negative on error
3716  */
3717 int
3718 update_mtu_from_frame_size(portid_t portid, uint32_t max_rx_pktlen)
3719 {
3720         struct rte_port *port = &ports[portid];
3721         uint32_t eth_overhead;
3722         uint16_t mtu, new_mtu;
3723
3724         eth_overhead = get_eth_overhead(&port->dev_info);
3725
3726         if (rte_eth_dev_get_mtu(portid, &mtu) != 0) {
3727                 printf("Failed to get MTU for port %u\n", portid);
3728                 return -1;
3729         }
3730
3731         new_mtu = max_rx_pktlen - eth_overhead;
3732
3733         if (mtu == new_mtu)
3734                 return 0;
3735
3736         if (eth_dev_set_mtu_mp(portid, new_mtu) != 0) {
3737                 fprintf(stderr,
3738                         "Failed to set MTU to %u for port %u\n",
3739                         new_mtu, portid);
3740                 return -1;
3741         }
3742
3743         port->dev_conf.rxmode.mtu = new_mtu;
3744
3745         return 0;
3746 }
3747
3748 void
3749 init_port_config(void)
3750 {
3751         portid_t pid;
3752         struct rte_port *port;
3753         int ret, i;
3754
3755         RTE_ETH_FOREACH_DEV(pid) {
3756                 port = &ports[pid];
3757                 port->dev_conf.fdir_conf = fdir_conf;
3758
3759                 ret = eth_dev_info_get_print_err(pid, &port->dev_info);
3760                 if (ret != 0)
3761                         return;
3762
3763                 if (nb_rxq > 1) {
3764                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3765                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf =
3766                                 rss_hf & port->dev_info.flow_type_rss_offloads;
3767                 } else {
3768                         port->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
3769                         port->dev_conf.rx_adv_conf.rss_conf.rss_hf = 0;
3770                 }
3771
3772                 if (port->dcb_flag == 0) {
3773                         if (port->dev_conf.rx_adv_conf.rss_conf.rss_hf != 0) {
3774                                 port->dev_conf.rxmode.mq_mode =
3775                                         (enum rte_eth_rx_mq_mode)
3776                                                 (rx_mq_mode & RTE_ETH_MQ_RX_RSS);
3777                         } else {
3778                                 port->dev_conf.rxmode.mq_mode = RTE_ETH_MQ_RX_NONE;
3779                                 port->dev_conf.rxmode.offloads &=
3780                                                 ~RTE_ETH_RX_OFFLOAD_RSS_HASH;
3781
3782                                 for (i = 0;
3783                                      i < port->dev_info.nb_rx_queues;
3784                                      i++)
3785                                         port->rx_conf[i].offloads &=
3786                                                 ~RTE_ETH_RX_OFFLOAD_RSS_HASH;
3787                         }
3788                 }
3789
3790                 rxtx_port_config(pid);
3791
3792                 ret = eth_macaddr_get_print_err(pid, &port->eth_addr);
3793                 if (ret != 0)
3794                         return;
3795
3796 #if defined RTE_NET_IXGBE && defined RTE_LIBRTE_IXGBE_BYPASS
3797                 rte_pmd_ixgbe_bypass_init(pid);
3798 #endif
3799
3800                 if (lsc_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_LSC))
3801                         port->dev_conf.intr_conf.lsc = 1;
3802                 if (rmv_interrupt && (*port->dev_info.dev_flags & RTE_ETH_DEV_INTR_RMV))
3803                         port->dev_conf.intr_conf.rmv = 1;
3804         }
3805 }
3806
3807 void set_port_slave_flag(portid_t slave_pid)
3808 {
3809         struct rte_port *port;
3810
3811         port = &ports[slave_pid];
3812         port->slave_flag = 1;
3813 }
3814
3815 void clear_port_slave_flag(portid_t slave_pid)
3816 {
3817         struct rte_port *port;
3818
3819         port = &ports[slave_pid];
3820         port->slave_flag = 0;
3821 }
3822
3823 uint8_t port_is_bonding_slave(portid_t slave_pid)
3824 {
3825         struct rte_port *port;
3826         struct rte_eth_dev_info dev_info;
3827         int ret;
3828
3829         port = &ports[slave_pid];
3830         ret = eth_dev_info_get_print_err(slave_pid, &dev_info);
3831         if (ret != 0) {
3832                 TESTPMD_LOG(ERR,
3833                         "Failed to get device info for port id %d,"
3834                         "cannot determine if the port is a bonded slave",
3835                         slave_pid);
3836                 return 0;
3837         }
3838         if ((*dev_info.dev_flags & RTE_ETH_DEV_BONDED_SLAVE) || (port->slave_flag == 1))
3839                 return 1;
3840         return 0;
3841 }
3842
3843 const uint16_t vlan_tags[] = {
3844                 0,  1,  2,  3,  4,  5,  6,  7,
3845                 8,  9, 10, 11,  12, 13, 14, 15,
3846                 16, 17, 18, 19, 20, 21, 22, 23,
3847                 24, 25, 26, 27, 28, 29, 30, 31
3848 };
3849
3850 static  int
3851 get_eth_dcb_conf(portid_t pid, struct rte_eth_conf *eth_conf,
3852                  enum dcb_mode_enable dcb_mode,
3853                  enum rte_eth_nb_tcs num_tcs,
3854                  uint8_t pfc_en)
3855 {
3856         uint8_t i;
3857         int32_t rc;
3858         struct rte_eth_rss_conf rss_conf;
3859
3860         /*
3861          * Builds up the correct configuration for dcb+vt based on the vlan tags array
3862          * given above, and the number of traffic classes available for use.
3863          */
3864         if (dcb_mode == DCB_VT_ENABLED) {
3865                 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3866                                 &eth_conf->rx_adv_conf.vmdq_dcb_conf;
3867                 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3868                                 &eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
3869
3870                 /* VMDQ+DCB RX and TX configurations */
3871                 vmdq_rx_conf->enable_default_pool = 0;
3872                 vmdq_rx_conf->default_pool = 0;
3873                 vmdq_rx_conf->nb_queue_pools =
3874                         (num_tcs ==  RTE_ETH_4_TCS ? RTE_ETH_32_POOLS : RTE_ETH_16_POOLS);
3875                 vmdq_tx_conf->nb_queue_pools =
3876                         (num_tcs ==  RTE_ETH_4_TCS ? RTE_ETH_32_POOLS : RTE_ETH_16_POOLS);
3877
3878                 vmdq_rx_conf->nb_pool_maps = vmdq_rx_conf->nb_queue_pools;
3879                 for (i = 0; i < vmdq_rx_conf->nb_pool_maps; i++) {
3880                         vmdq_rx_conf->pool_map[i].vlan_id = vlan_tags[i];
3881                         vmdq_rx_conf->pool_map[i].pools =
3882                                 1 << (i % vmdq_rx_conf->nb_queue_pools);
3883                 }
3884                 for (i = 0; i < RTE_ETH_DCB_NUM_USER_PRIORITIES; i++) {
3885                         vmdq_rx_conf->dcb_tc[i] = i % num_tcs;
3886                         vmdq_tx_conf->dcb_tc[i] = i % num_tcs;
3887                 }
3888
3889                 /* set DCB mode of RX and TX of multiple queues */
3890                 eth_conf->rxmode.mq_mode =
3891                                 (enum rte_eth_rx_mq_mode)
3892                                         (rx_mq_mode & RTE_ETH_MQ_RX_VMDQ_DCB);
3893                 eth_conf->txmode.mq_mode = RTE_ETH_MQ_TX_VMDQ_DCB;
3894         } else {
3895                 struct rte_eth_dcb_rx_conf *rx_conf =
3896                                 &eth_conf->rx_adv_conf.dcb_rx_conf;
3897                 struct rte_eth_dcb_tx_conf *tx_conf =
3898                                 &eth_conf->tx_adv_conf.dcb_tx_conf;
3899
3900                 memset(&rss_conf, 0, sizeof(struct rte_eth_rss_conf));
3901
3902                 rc = rte_eth_dev_rss_hash_conf_get(pid, &rss_conf);
3903                 if (rc != 0)
3904                         return rc;
3905
3906                 rx_conf->nb_tcs = num_tcs;
3907                 tx_conf->nb_tcs = num_tcs;
3908
3909                 for (i = 0; i < RTE_ETH_DCB_NUM_USER_PRIORITIES; i++) {
3910                         rx_conf->dcb_tc[i] = i % num_tcs;
3911                         tx_conf->dcb_tc[i] = i % num_tcs;
3912                 }
3913
3914                 eth_conf->rxmode.mq_mode =
3915                                 (enum rte_eth_rx_mq_mode)
3916                                         (rx_mq_mode & RTE_ETH_MQ_RX_DCB_RSS);
3917                 eth_conf->rx_adv_conf.rss_conf = rss_conf;
3918                 eth_conf->txmode.mq_mode = RTE_ETH_MQ_TX_DCB;
3919         }
3920
3921         if (pfc_en)
3922                 eth_conf->dcb_capability_en =
3923                                 RTE_ETH_DCB_PG_SUPPORT | RTE_ETH_DCB_PFC_SUPPORT;
3924         else
3925                 eth_conf->dcb_capability_en = RTE_ETH_DCB_PG_SUPPORT;
3926
3927         return 0;
3928 }
3929
3930 int
3931 init_port_dcb_config(portid_t pid,
3932                      enum dcb_mode_enable dcb_mode,
3933                      enum rte_eth_nb_tcs num_tcs,
3934                      uint8_t pfc_en)
3935 {
3936         struct rte_eth_conf port_conf;
3937         struct rte_port *rte_port;
3938         int retval;
3939         uint16_t i;
3940
3941         if (num_procs > 1) {
3942                 printf("The multi-process feature doesn't support dcb.\n");
3943                 return -ENOTSUP;
3944         }
3945         rte_port = &ports[pid];
3946
3947         /* retain the original device configuration. */
3948         memcpy(&port_conf, &rte_port->dev_conf, sizeof(struct rte_eth_conf));
3949
3950         /*set configuration of DCB in vt mode and DCB in non-vt mode*/
3951         retval = get_eth_dcb_conf(pid, &port_conf, dcb_mode, num_tcs, pfc_en);
3952         if (retval < 0)
3953                 return retval;
3954         port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_VLAN_FILTER;
3955         /* remove RSS HASH offload for DCB in vt mode */
3956         if (port_conf.rxmode.mq_mode == RTE_ETH_MQ_RX_VMDQ_DCB) {
3957                 port_conf.rxmode.offloads &= ~RTE_ETH_RX_OFFLOAD_RSS_HASH;
3958                 for (i = 0; i < nb_rxq; i++)
3959                         rte_port->rx_conf[i].offloads &=
3960                                 ~RTE_ETH_RX_OFFLOAD_RSS_HASH;
3961         }
3962
3963         /* re-configure the device . */
3964         retval = rte_eth_dev_configure(pid, nb_rxq, nb_rxq, &port_conf);
3965         if (retval < 0)
3966                 return retval;
3967
3968         retval = eth_dev_info_get_print_err(pid, &rte_port->dev_info);
3969         if (retval != 0)
3970                 return retval;
3971
3972         /* If dev_info.vmdq_pool_base is greater than 0,
3973          * the queue id of vmdq pools is started after pf queues.
3974          */
3975         if (dcb_mode == DCB_VT_ENABLED &&
3976             rte_port->dev_info.vmdq_pool_base > 0) {
3977                 fprintf(stderr,
3978                         "VMDQ_DCB multi-queue mode is nonsensical for port %d.\n",
3979                         pid);
3980                 return -1;
3981         }
3982
3983         /* Assume the ports in testpmd have the same dcb capability
3984          * and has the same number of rxq and txq in dcb mode
3985          */
3986         if (dcb_mode == DCB_VT_ENABLED) {
3987                 if (rte_port->dev_info.max_vfs > 0) {
3988                         nb_rxq = rte_port->dev_info.nb_rx_queues;
3989                         nb_txq = rte_port->dev_info.nb_tx_queues;
3990                 } else {
3991                         nb_rxq = rte_port->dev_info.max_rx_queues;
3992                         nb_txq = rte_port->dev_info.max_tx_queues;
3993                 }
3994         } else {
3995                 /*if vt is disabled, use all pf queues */
3996                 if (rte_port->dev_info.vmdq_pool_base == 0) {
3997                         nb_rxq = rte_port->dev_info.max_rx_queues;
3998                         nb_txq = rte_port->dev_info.max_tx_queues;
3999                 } else {
4000                         nb_rxq = (queueid_t)num_tcs;
4001                         nb_txq = (queueid_t)num_tcs;
4002
4003                 }
4004         }
4005         rx_free_thresh = 64;
4006
4007         memcpy(&rte_port->dev_conf, &port_conf, sizeof(struct rte_eth_conf));
4008
4009         rxtx_port_config(pid);
4010         /* VLAN filter */
4011         rte_port->dev_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_VLAN_FILTER;
4012         for (i = 0; i < RTE_DIM(vlan_tags); i++)
4013                 rx_vft_set(pid, vlan_tags[i], 1);
4014
4015         retval = eth_macaddr_get_print_err(pid, &rte_port->eth_addr);
4016         if (retval != 0)
4017                 return retval;
4018
4019         rte_port->dcb_flag = 1;
4020
4021         /* Enter DCB configuration status */
4022         dcb_config = 1;
4023
4024         return 0;
4025 }
4026
4027 static void
4028 init_port(void)
4029 {
4030         int i;
4031
4032         /* Configuration of Ethernet ports. */
4033         ports = rte_zmalloc("testpmd: ports",
4034                             sizeof(struct rte_port) * RTE_MAX_ETHPORTS,
4035                             RTE_CACHE_LINE_SIZE);
4036         if (ports == NULL) {
4037                 rte_exit(EXIT_FAILURE,
4038                                 "rte_zmalloc(%d struct rte_port) failed\n",
4039                                 RTE_MAX_ETHPORTS);
4040         }
4041         for (i = 0; i < RTE_MAX_ETHPORTS; i++)
4042                 ports[i].xstats_info.allocated = false;
4043         for (i = 0; i < RTE_MAX_ETHPORTS; i++)
4044                 LIST_INIT(&ports[i].flow_tunnel_list);
4045         /* Initialize ports NUMA structures */
4046         memset(port_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
4047         memset(rxring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
4048         memset(txring_numa, NUMA_NO_CONFIG, RTE_MAX_ETHPORTS);
4049 }
4050
4051 static void
4052 force_quit(void)
4053 {
4054         pmd_test_exit();
4055         prompt_exit();
4056 }
4057
4058 static void
4059 print_stats(void)
4060 {
4061         uint8_t i;
4062         const char clr[] = { 27, '[', '2', 'J', '\0' };
4063         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
4064
4065         /* Clear screen and move to top left */
4066         printf("%s%s", clr, top_left);
4067
4068         printf("\nPort statistics ====================================");
4069         for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++)
4070                 nic_stats_display(fwd_ports_ids[i]);
4071
4072         fflush(stdout);
4073 }
4074
4075 static void
4076 signal_handler(int signum)
4077 {
4078         if (signum == SIGINT || signum == SIGTERM) {
4079                 fprintf(stderr, "\nSignal %d received, preparing to exit...\n",
4080                         signum);
4081 #ifdef RTE_LIB_PDUMP
4082                 /* uninitialize packet capture framework */
4083                 rte_pdump_uninit();
4084 #endif
4085 #ifdef RTE_LIB_LATENCYSTATS
4086                 if (latencystats_enabled != 0)
4087                         rte_latencystats_uninit();
4088 #endif
4089                 force_quit();
4090                 /* Set flag to indicate the force termination. */
4091                 f_quit = 1;
4092                 /* exit with the expected status */
4093 #ifndef RTE_EXEC_ENV_WINDOWS
4094                 signal(signum, SIG_DFL);
4095                 kill(getpid(), signum);
4096 #endif
4097         }
4098 }
4099
4100 int
4101 main(int argc, char** argv)
4102 {
4103         int diag;
4104         portid_t port_id;
4105         uint16_t count;
4106         int ret;
4107
4108         signal(SIGINT, signal_handler);
4109         signal(SIGTERM, signal_handler);
4110
4111         testpmd_logtype = rte_log_register("testpmd");
4112         if (testpmd_logtype < 0)
4113                 rte_exit(EXIT_FAILURE, "Cannot register log type");
4114         rte_log_set_level(testpmd_logtype, RTE_LOG_DEBUG);
4115
4116         diag = rte_eal_init(argc, argv);
4117         if (diag < 0)
4118                 rte_exit(EXIT_FAILURE, "Cannot init EAL: %s\n",
4119                          rte_strerror(rte_errno));
4120
4121         ret = register_eth_event_callback();
4122         if (ret != 0)
4123                 rte_exit(EXIT_FAILURE, "Cannot register for ethdev events");
4124
4125 #ifdef RTE_LIB_PDUMP
4126         /* initialize packet capture framework */
4127         rte_pdump_init();
4128 #endif
4129
4130         count = 0;
4131         RTE_ETH_FOREACH_DEV(port_id) {
4132                 ports_ids[count] = port_id;
4133                 count++;
4134         }
4135         nb_ports = (portid_t) count;
4136         if (nb_ports == 0)
4137                 TESTPMD_LOG(WARNING, "No probed ethernet devices\n");
4138
4139         /* allocate port structures, and init them */
4140         init_port();
4141
4142         set_def_fwd_config();
4143         if (nb_lcores == 0)
4144                 rte_exit(EXIT_FAILURE, "No cores defined for forwarding\n"
4145                          "Check the core mask argument\n");
4146
4147         /* Bitrate/latency stats disabled by default */
4148 #ifdef RTE_LIB_BITRATESTATS
4149         bitrate_enabled = 0;
4150 #endif
4151 #ifdef RTE_LIB_LATENCYSTATS
4152         latencystats_enabled = 0;
4153 #endif
4154
4155         /* on FreeBSD, mlockall() is disabled by default */
4156 #ifdef RTE_EXEC_ENV_FREEBSD
4157         do_mlockall = 0;
4158 #else
4159         do_mlockall = 1;
4160 #endif
4161
4162         argc -= diag;
4163         argv += diag;
4164         if (argc > 1)
4165                 launch_args_parse(argc, argv);
4166
4167 #ifndef RTE_EXEC_ENV_WINDOWS
4168         if (do_mlockall && mlockall(MCL_CURRENT | MCL_FUTURE)) {
4169                 TESTPMD_LOG(NOTICE, "mlockall() failed with error \"%s\"\n",
4170                         strerror(errno));
4171         }
4172 #endif
4173
4174         if (tx_first && interactive)
4175                 rte_exit(EXIT_FAILURE, "--tx-first cannot be used on "
4176                                 "interactive mode.\n");
4177
4178         if (tx_first && lsc_interrupt) {
4179                 fprintf(stderr,
4180                         "Warning: lsc_interrupt needs to be off when using tx_first. Disabling.\n");
4181                 lsc_interrupt = 0;
4182         }
4183
4184         if (!nb_rxq && !nb_txq)
4185                 fprintf(stderr,
4186                         "Warning: Either rx or tx queues should be non-zero\n");
4187
4188         if (nb_rxq > 1 && nb_rxq > nb_txq)
4189                 fprintf(stderr,
4190                         "Warning: nb_rxq=%d enables RSS configuration, but nb_txq=%d will prevent to fully test it.\n",
4191                         nb_rxq, nb_txq);
4192
4193         init_config();
4194
4195         if (hot_plug) {
4196                 ret = rte_dev_hotplug_handle_enable();
4197                 if (ret) {
4198                         RTE_LOG(ERR, EAL,
4199                                 "fail to enable hotplug handling.");
4200                         return -1;
4201                 }
4202
4203                 ret = rte_dev_event_monitor_start();
4204                 if (ret) {
4205                         RTE_LOG(ERR, EAL,
4206                                 "fail to start device event monitoring.");
4207                         return -1;
4208                 }
4209
4210                 ret = rte_dev_event_callback_register(NULL,
4211                         dev_event_callback, NULL);
4212                 if (ret) {
4213                         RTE_LOG(ERR, EAL,
4214                                 "fail  to register device event callback\n");
4215                         return -1;
4216                 }
4217         }
4218
4219         if (!no_device_start && start_port(RTE_PORT_ALL) != 0)
4220                 rte_exit(EXIT_FAILURE, "Start ports failed\n");
4221
4222         /* set all ports to promiscuous mode by default */
4223         RTE_ETH_FOREACH_DEV(port_id) {
4224                 ret = rte_eth_promiscuous_enable(port_id);
4225                 if (ret != 0)
4226                         fprintf(stderr,
4227                                 "Error during enabling promiscuous mode for port %u: %s - ignore\n",
4228                                 port_id, rte_strerror(-ret));
4229         }
4230
4231 #ifdef RTE_LIB_METRICS
4232         /* Init metrics library */
4233         rte_metrics_init(rte_socket_id());
4234 #endif
4235
4236 #ifdef RTE_LIB_LATENCYSTATS
4237         if (latencystats_enabled != 0) {
4238                 int ret = rte_latencystats_init(1, NULL);
4239                 if (ret)
4240                         fprintf(stderr,
4241                                 "Warning: latencystats init() returned error %d\n",
4242                                 ret);
4243                 fprintf(stderr, "Latencystats running on lcore %d\n",
4244                         latencystats_lcore_id);
4245         }
4246 #endif
4247
4248         /* Setup bitrate stats */
4249 #ifdef RTE_LIB_BITRATESTATS
4250         if (bitrate_enabled != 0) {
4251                 bitrate_data = rte_stats_bitrate_create();
4252                 if (bitrate_data == NULL)
4253                         rte_exit(EXIT_FAILURE,
4254                                 "Could not allocate bitrate data.\n");
4255                 rte_stats_bitrate_reg(bitrate_data);
4256         }
4257 #endif
4258 #ifdef RTE_LIB_CMDLINE
4259         if (strlen(cmdline_filename) != 0)
4260                 cmdline_read_from_file(cmdline_filename);
4261
4262         if (interactive == 1) {
4263                 if (auto_start) {
4264                         printf("Start automatic packet forwarding\n");
4265                         start_packet_forwarding(0);
4266                 }
4267                 prompt();
4268                 pmd_test_exit();
4269         } else
4270 #endif
4271         {
4272                 char c;
4273                 int rc;
4274
4275                 f_quit = 0;
4276
4277                 printf("No commandline core given, start packet forwarding\n");
4278                 start_packet_forwarding(tx_first);
4279                 if (stats_period != 0) {
4280                         uint64_t prev_time = 0, cur_time, diff_time = 0;
4281                         uint64_t timer_period;
4282
4283                         /* Convert to number of cycles */
4284                         timer_period = stats_period * rte_get_timer_hz();
4285
4286                         while (f_quit == 0) {
4287                                 cur_time = rte_get_timer_cycles();
4288                                 diff_time += cur_time - prev_time;
4289
4290                                 if (diff_time >= timer_period) {
4291                                         print_stats();
4292                                         /* Reset the timer */
4293                                         diff_time = 0;
4294                                 }
4295                                 /* Sleep to avoid unnecessary checks */
4296                                 prev_time = cur_time;
4297                                 rte_delay_us_sleep(US_PER_S);
4298                         }
4299                 }
4300
4301                 printf("Press enter to exit\n");
4302                 rc = read(0, &c, 1);
4303                 pmd_test_exit();
4304                 if (rc < 0)
4305                         return 1;
4306         }
4307
4308         ret = rte_eal_cleanup();
4309         if (ret != 0)
4310                 rte_exit(EXIT_FAILURE,
4311                          "EAL cleanup failed: %s\n", strerror(-ret));
4312
4313         return EXIT_SUCCESS;
4314 }