examples/l2fwd-ivshmem/host/host.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <unistd.h>
  35 #include <stdlib.h>
  36 #include <stdio.h>
  37 #include <string.h>
  38 #include <limits.h>
  39 #include <inttypes.h>
  40 #include <getopt.h>
  41 #include <signal.h>
  42
  43 #include <rte_eal.h>
  44 #include <rte_config.h>
  45 #include <rte_cycles.h>
  46 #include <rte_eal_memconfig.h>
  47 #include <rte_debug.h>
  48 #include <rte_ether.h>
  49 #include <rte_ethdev.h>
  50 #include <rte_string_fns.h>
  51 #include <rte_ivshmem.h>
  52 #include <rte_ring.h>
  53 #include <rte_mempool.h>
  54 #include <rte_mbuf.h>
  55
  56 #include "../include/common.h"
  57
  58 /*
  59  * Configurable number of RX/TX ring descriptors
  60  */
  61 #define RTE_TEST_RX_DESC_DEFAULT 128
  62 #define RTE_TEST_TX_DESC_DEFAULT 512
  63 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
  64 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
  65
  66 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  67
  68 /* mask of enabled ports */
  69 static uint32_t l2fwd_ivshmem_enabled_port_mask = 0;
  70
  71 static struct ether_addr l2fwd_ivshmem_ports_eth_addr[RTE_MAX_ETHPORTS];
  72
  73 #define NB_MBUF   8192
  74 #define MBUF_DATA_SIZE (2048 + RTE_PKTMBUF_HEADROOM)
  75
  76 #define MAX_RX_QUEUE_PER_LCORE 16
  77 #define MAX_TX_QUEUE_PER_PORT 16
  78 struct lcore_queue_conf {
  79         unsigned n_rx_port;
  80         unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
  81         struct vm_port_param * port_param[MAX_RX_QUEUE_PER_LCORE];
  82         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
  83         struct mbuf_table rx_mbufs[RTE_MAX_ETHPORTS];
  84 } __rte_cache_aligned;
  85 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
  86
  87 static const struct rte_eth_conf port_conf = {
  88         .rxmode = {
  89                 .split_hdr_size = 0,
  90                 .header_split   = 0, /**< Header Split disabled */
  91                 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
  92                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
  93                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
  94                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
  95         },
  96         .txmode = {
  97                 .mq_mode = ETH_MQ_TX_NONE,
  98         },
  99 };
 100
 101 #define METADATA_NAME "l2fwd_ivshmem"
 102 #define CMDLINE_OPT_FWD_CONF "fwd-conf"
 103
 104 #define QEMU_CMD_FMT "/tmp/ivshmem_qemu_cmdline_%s"
 105
 106 struct port_statistics port_statistics[RTE_MAX_ETHPORTS];
 107
 108 struct rte_mempool * l2fwd_ivshmem_pktmbuf_pool = NULL;
 109
 110 /* Print out statistics on packets dropped */
 111 static void
 112 print_stats(void)
 113 {
 114         uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
 115         uint64_t total_vm_packets_dropped, total_vm_packets_tx, total_vm_packets_rx;
 116         unsigned portid;
 117
 118         total_packets_dropped = 0;
 119         total_packets_tx = 0;
 120         total_packets_rx = 0;
 121         total_vm_packets_tx = 0;
 122         total_vm_packets_rx = 0;
 123
 124         const char clr[] = { 27, '[', '2', 'J', '\0' };
 125         const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
 126
 127                 /* Clear screen and move to top left */
 128         printf("%s%s", clr, topLeft);
 129
 130         printf("\nPort statistics ====================================");
 131
 132         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 133                 /* skip disabled ports */
 134                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0)
 135                         continue;
 136                 printf("\nStatistics for port %u ------------------------------"
 137                            "\nPackets sent: %24"PRIu64
 138                            "\nPackets received: %20"PRIu64
 139                            "\nPackets dropped: %21"PRIu64,
 140                            portid,
 141                            port_statistics[portid].tx,
 142                            port_statistics[portid].rx,
 143                            port_statistics[portid].dropped);
 144
 145                 total_packets_dropped += port_statistics[portid].dropped;
 146                 total_packets_tx += port_statistics[portid].tx;
 147                 total_packets_rx += port_statistics[portid].rx;
 148         }
 149
 150         printf("\nVM statistics ======================================");
 151         for (portid = 0; portid < ctrl->nb_ports; portid++) {
 152                 printf("\nStatistics for port %u ------------------------------"
 153                            "\nPackets sent: %24"PRIu64
 154                            "\nPackets received: %20"PRIu64,
 155                            portid,
 156                            ctrl->vm_ports[portid].stats.tx,
 157                            ctrl->vm_ports[portid].stats.rx);
 158
 159                 total_vm_packets_dropped += ctrl->vm_ports[portid].stats.dropped;
 160                 total_vm_packets_tx += ctrl->vm_ports[portid].stats.tx;
 161                 total_vm_packets_rx += ctrl->vm_ports[portid].stats.rx;
 162         }
 163         printf("\nAggregate statistics ==============================="
 164                            "\nTotal packets sent: %18"PRIu64
 165                            "\nTotal packets received: %14"PRIu64
 166                            "\nTotal packets dropped: %15"PRIu64
 167                            "\nTotal VM packets sent: %15"PRIu64
 168                            "\nTotal VM packets received: %11"PRIu64,
 169                            total_packets_tx,
 170                            total_packets_rx,
 171                            total_packets_dropped,
 172                            total_vm_packets_tx,
 173                            total_vm_packets_rx);
 174         printf("\n====================================================\n");
 175 }
 176
 177 static int
 178 print_to_file(const char *cmdline, const char *config_name)
 179 {
 180         FILE *file;
 181         char path[PATH_MAX];
 182
 183         snprintf(path, sizeof(path), QEMU_CMD_FMT, config_name);
 184         file = fopen(path, "w");
 185         if (file == NULL) {
 186                 RTE_LOG(ERR, L2FWD_IVSHMEM, "Could not open '%s' \n", path);
 187                 return -1;
 188         }
 189
 190         RTE_LOG(DEBUG, L2FWD_IVSHMEM, "QEMU command line for config '%s': %s \n",
 191                         config_name, cmdline);
 192
 193         fprintf(file, "%s\n", cmdline);
 194         fclose(file);
 195         return 0;
 196 }
 197
 198 static int
 199 generate_ivshmem_cmdline(const char *config_name)
 200 {
 201         char cmdline[PATH_MAX];
 202         if (rte_ivshmem_metadata_cmdline_generate(cmdline, sizeof(cmdline),
 203                         config_name) < 0)
 204                 return -1;
 205
 206         if (print_to_file(cmdline, config_name) < 0)
 207                 return -1;
 208
 209         rte_ivshmem_metadata_dump(stdout, config_name);
 210         return 0;
 211 }
 212
 213 /* display usage */
 214 static void
 215 l2fwd_ivshmem_usage(const char *prgname)
 216 {
 217         printf("%s [EAL options] -- -p PORTMASK [-q NQ -T PERIOD]\n"
 218                    "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 219                    "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
 220                    "  -T PERIOD: statistics will be refreshed each PERIOD seconds "
 221                        "(0 to disable, 10 default, 86400 maximum)\n",
 222                prgname);
 223 }
 224
 225 static unsigned int
 226 l2fwd_ivshmem_parse_nqueue(const char *q_arg)
 227 {
 228         char *end = NULL;
 229         unsigned long n;
 230
 231         /* parse hexadecimal string */
 232         n = strtoul(q_arg, &end, 10);
 233         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 234                 return 0;
 235         if (n == 0)
 236                 return 0;
 237         if (n >= MAX_RX_QUEUE_PER_LCORE)
 238                 return 0;
 239
 240         return n;
 241 }
 242
 243 static int
 244 l2fwd_ivshmem_parse_portmask(const char *portmask)
 245 {
 246         char *end = NULL;
 247         unsigned long pm;
 248
 249         /* parse hexadecimal string */
 250         pm = strtoul(portmask, &end, 16);
 251         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 252                 return -1;
 253
 254         if (pm == 0)
 255                 return -1;
 256
 257         return pm;
 258 }
 259
 260 static int
 261 l2fwd_ivshmem_parse_timer_period(const char *q_arg)
 262 {
 263         char *end = NULL;
 264         int n;
 265
 266         /* parse number string */
 267         n = strtol(q_arg, &end, 10);
 268         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 269                 return -1;
 270         if (n >= MAX_TIMER_PERIOD)
 271                 return -1;
 272
 273         return n;
 274 }
 275
 276 /* Parse the argument given in the command line of the application */
 277 static int
 278 l2fwd_ivshmem_parse_args(int argc, char **argv)
 279 {
 280         int opt, ret;
 281         char **argvopt;
 282         int option_index;
 283         char *prgname = argv[0];
 284         static struct option lgopts[] = {
 285                         {CMDLINE_OPT_FWD_CONF, 1, 0, 0},
 286                 {NULL, 0, 0, 0}
 287         };
 288
 289         argvopt = argv;
 290
 291         while ((opt = getopt_long(argc, argvopt, "q:p:T:",
 292                                   lgopts, &option_index)) != EOF) {
 293
 294                 switch (opt) {
 295                 /* portmask */
 296                 case 'p':
 297                         l2fwd_ivshmem_enabled_port_mask = l2fwd_ivshmem_parse_portmask(optarg);
 298                         if (l2fwd_ivshmem_enabled_port_mask == 0) {
 299                                 printf("invalid portmask\n");
 300                                 l2fwd_ivshmem_usage(prgname);
 301                                 return -1;
 302                         }
 303                         break;
 304
 305                 /* nqueue */
 306                 case 'q':
 307                         l2fwd_ivshmem_rx_queue_per_lcore = l2fwd_ivshmem_parse_nqueue(optarg);
 308                         if (l2fwd_ivshmem_rx_queue_per_lcore == 0) {
 309                                 printf("invalid queue number\n");
 310                                 l2fwd_ivshmem_usage(prgname);
 311                                 return -1;
 312                         }
 313                         break;
 314
 315                 /* timer period */
 316                 case 'T':
 317                         timer_period = l2fwd_ivshmem_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
 318                         if (timer_period < 0) {
 319                                 printf("invalid timer period\n");
 320                                 l2fwd_ivshmem_usage(prgname);
 321                                 return -1;
 322                         }
 323                         break;
 324
 325                 /* long options */
 326                 case 0:
 327                         l2fwd_ivshmem_usage(prgname);
 328                         return -1;
 329
 330                 default:
 331                         l2fwd_ivshmem_usage(prgname);
 332                         return -1;
 333                 }
 334         }
 335
 336         if (optind >= 0)
 337                 argv[optind-1] = prgname;
 338
 339         ret = optind-1;
 340         optind = 0; /* reset getopt lib */
 341         return ret;
 342 }
 343
 344 /* Check the link status of all ports in up to 9s, and print them finally */
 345 static void
 346 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 347 {
 348 #define CHECK_INTERVAL 100 /* 100ms */
 349 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 350         uint8_t portid, count, all_ports_up, print_flag = 0;
 351         struct rte_eth_link link;
 352
 353         printf("\nChecking link status");
 354         fflush(stdout);
 355         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 356                 all_ports_up = 1;
 357                 for (portid = 0; portid < port_num; portid++) {
 358                         if ((port_mask & (1 << portid)) == 0)
 359                                 continue;
 360                         memset(&link, 0, sizeof(link));
 361                         rte_eth_link_get_nowait(portid, &link);
 362                         /* print link status if flag set */
 363                         if (print_flag == 1) {
 364                                 if (link.link_status)
 365                                         printf("Port %d Link Up - speed %u "
 366                                                 "Mbps - %s\n", (uint8_t)portid,
 367                                                 (unsigned)link.link_speed,
 368                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 369                                         ("full-duplex") : ("half-duplex\n"));
 370                                 else
 371                                         printf("Port %d Link Down\n",
 372                                                 (uint8_t)portid);
 373                                 continue;
 374                         }
 375                         /* clear all_ports_up flag if any link down */
 376                         if (link.link_status == 0) {
 377                                 all_ports_up = 0;
 378                                 break;
 379                         }
 380                 }
 381                 /* after finally printing all link status, get out */
 382                 if (print_flag == 1)
 383                         break;
 384
 385                 if (all_ports_up == 0) {
 386                         printf(".");
 387                         fflush(stdout);
 388                         rte_delay_ms(CHECK_INTERVAL);
 389                 }
 390
 391                 /* set the print_flag if all ports up or timeout */
 392                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 393                         print_flag = 1;
 394                         printf("done\n");
 395                 }
 396         }
 397 }
 398
 399 /* Send the burst of packets on an output interface */
 400 static int
 401 l2fwd_ivshmem_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
 402 {
 403         struct rte_mbuf **m_table;
 404         unsigned ret;
 405         unsigned queueid =0;
 406
 407         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 408
 409         ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
 410         port_statistics[port].tx += ret;
 411         if (unlikely(ret < n)) {
 412                 port_statistics[port].dropped += (n - ret);
 413                 do {
 414                         rte_pktmbuf_free(m_table[ret]);
 415                 } while (++ret < n);
 416         }
 417
 418         return 0;
 419 }
 420
 421 /* Enqueue packets for TX and prepare them to be sent on the network */
 422 static int
 423 l2fwd_ivshmem_send_packet(struct rte_mbuf *m, uint8_t port)
 424 {
 425         unsigned lcore_id, len;
 426         struct lcore_queue_conf *qconf;
 427
 428         lcore_id = rte_lcore_id();
 429
 430         qconf = &lcore_queue_conf[lcore_id];
 431         len = qconf->tx_mbufs[port].len;
 432         qconf->tx_mbufs[port].m_table[len] = m;
 433         len++;
 434
 435         /* enough pkts to be sent */
 436         if (unlikely(len == MAX_PKT_BURST)) {
 437                 l2fwd_ivshmem_send_burst(qconf, MAX_PKT_BURST, port);
 438                 len = 0;
 439         }
 440
 441         qconf->tx_mbufs[port].len = len;
 442         return 0;
 443 }
 444
 445 static int
 446 l2fwd_ivshmem_receive_burst(struct lcore_queue_conf *qconf, unsigned portid,
 447                 unsigned vm_port)
 448 {
 449         struct rte_mbuf ** m;
 450         struct rte_ring * rx;
 451         unsigned len, pkt_idx;
 452
 453         m = qconf->rx_mbufs[portid].m_table;
 454         len = qconf->rx_mbufs[portid].len;
 455         rx = qconf->port_param[vm_port]->rx_ring;
 456
 457         /* if enqueueing failed, ring is probably full, so drop the packets */
 458         if (rte_ring_enqueue_bulk(rx, (void**) m, len) < 0) {
 459                 port_statistics[portid].dropped += len;
 460
 461                 pkt_idx = 0;
 462                 do {
 463                         rte_pktmbuf_free(m[pkt_idx]);
 464                 } while (++pkt_idx < len);
 465         }
 466         else
 467                 /* increment rx stats by however many packets we managed to receive */
 468                 port_statistics[portid].rx += len;
 469
 470         return 0;
 471 }
 472
 473 /* Enqueue packets for RX and prepare them to be sent to VM */
 474 static int
 475 l2fwd_ivshmem_receive_packets(struct rte_mbuf ** m, unsigned n, unsigned portid,
 476                 unsigned vm_port)
 477 {
 478         unsigned lcore_id, len, pkt_idx;
 479         struct lcore_queue_conf *qconf;
 480
 481         lcore_id = rte_lcore_id();
 482
 483         qconf = &lcore_queue_conf[lcore_id];
 484
 485         len = qconf->rx_mbufs[portid].len;
 486         pkt_idx = 0;
 487
 488         /* enqueue packets */
 489         while (pkt_idx < n && len < MAX_PKT_BURST * 2) {
 490                 qconf->rx_mbufs[portid].m_table[len++] = m[pkt_idx++];
 491         }
 492
 493         /* increment queue len by however many packets we managed to receive */
 494         qconf->rx_mbufs[portid].len += pkt_idx;
 495
 496         /* drop the unreceived packets */
 497         if (unlikely(pkt_idx < n)) {
 498                 port_statistics[portid].dropped += n - pkt_idx;
 499                 do {
 500                         rte_pktmbuf_free(m[pkt_idx]);
 501                 } while (++pkt_idx < n);
 502         }
 503
 504         /* drain the queue halfway through the maximum capacity */
 505         if (unlikely(qconf->rx_mbufs[portid].len >= MAX_PKT_BURST))
 506                 l2fwd_ivshmem_receive_burst(qconf, portid, vm_port);
 507
 508         return 0;
 509 }
 510
 511 /* loop for host forwarding mode.
 512  * the data flow is as follows:
 513  *  1) get packets from TX queue and send it out from a given port
 514  *  2) RX packets from given port and enqueue them on RX ring
 515  *  3) dequeue packets from TX ring and put them on TX queue for a given port
 516  */
 517 static void
 518 fwd_loop(void)
 519 {
 520         struct rte_mbuf *pkts_burst[MAX_PKT_BURST * 2];
 521         struct rte_mbuf *m;
 522         unsigned lcore_id;
 523         uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
 524         unsigned i, j, portid, nb_rx;
 525         struct lcore_queue_conf *qconf;
 526         struct rte_ring *tx;
 527         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 528
 529         prev_tsc = 0;
 530         timer_tsc = 0;
 531
 532         lcore_id = rte_lcore_id();
 533         qconf = &lcore_queue_conf[lcore_id];
 534
 535         if (qconf->n_rx_port == 0) {
 536                 RTE_LOG(INFO, L2FWD_IVSHMEM, "lcore %u has nothing to do\n", lcore_id);
 537                 return;
 538         }
 539
 540         RTE_LOG(INFO, L2FWD_IVSHMEM, "entering main loop on lcore %u\n", lcore_id);
 541
 542         for (i = 0; i < qconf->n_rx_port; i++) {
 543
 544                 portid = qconf->rx_port_list[i];
 545                 RTE_LOG(INFO, L2FWD_IVSHMEM, " -- lcoreid=%u portid=%u\n", lcore_id,
 546                         portid);
 547         }
 548
 549         while (ctrl->state == STATE_FWD) {
 550
 551                 cur_tsc = rte_rdtsc();
 552
 553                 /*
 554                  * Burst queue drain
 555                  */
 556                 diff_tsc = cur_tsc - prev_tsc;
 557                 if (unlikely(diff_tsc > drain_tsc)) {
 558
 559                         /*
 560                          * TX
 561                          */
 562                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 563                                 if (qconf->tx_mbufs[portid].len == 0)
 564                                         continue;
 565                                 l2fwd_ivshmem_send_burst(qconf,
 566                                                  qconf->tx_mbufs[portid].len,
 567                                                  (uint8_t) portid);
 568                                 qconf->tx_mbufs[portid].len = 0;
 569                         }
 570
 571                         /*
 572                          * RX
 573                          */
 574                         for (i = 0; i < qconf->n_rx_port; i++) {
 575                                 portid = qconf->rx_port_list[i];
 576                                 if (qconf->rx_mbufs[portid].len == 0)
 577                                         continue;
 578                                 l2fwd_ivshmem_receive_burst(qconf, portid, i);
 579                                 qconf->rx_mbufs[portid].len = 0;
 580                         }
 581
 582                         /* if timer is enabled */
 583                         if (timer_period > 0) {
 584
 585                                 /* advance the timer */
 586                                 timer_tsc += diff_tsc;
 587
 588                                 /* if timer has reached its timeout */
 589                                 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
 590
 591                                         /* do this only on master core */
 592                                         if (lcore_id == rte_get_master_lcore()) {
 593                                                 print_stats();
 594                                                 /* reset the timer */
 595                                                 timer_tsc = 0;
 596                                         }
 597                                 }
 598                         }
 599
 600                         prev_tsc = cur_tsc;
 601                 }
 602
 603                 /*
 604                  * packet RX and forwarding
 605                  */
 606                 for (i = 0; i < qconf->n_rx_port; i++) {
 607
 608                         /* RX packets from port and put them on RX ring */
 609                         portid = qconf->rx_port_list[i];
 610                         nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
 611                                                  pkts_burst, MAX_PKT_BURST);
 612
 613                         if (nb_rx != 0)
 614                                 l2fwd_ivshmem_receive_packets(pkts_burst, nb_rx, portid, i);
 615
 616                         /* dequeue packets from TX ring and send them to TX queue */
 617                         tx = qconf->port_param[i]->tx_ring;
 618
 619                         nb_rx = rte_ring_count(tx);
 620
 621                         nb_rx = RTE_MIN(nb_rx, (unsigned) MAX_PKT_BURST);
 622
 623                         if (nb_rx == 0)
 624                                 continue;
 625
 626                         /* should not happen */
 627                         if (unlikely(rte_ring_dequeue_bulk(tx, (void**) pkts_burst, nb_rx) < 0)) {
 628                                 ctrl->state = STATE_FAIL;
 629                                 return;
 630                         }
 631
 632                         for (j = 0; j < nb_rx; j++) {
 633                                 m = pkts_burst[j];
 634                                 l2fwd_ivshmem_send_packet(m, portid);
 635                         }
 636                 }
 637         }
 638 }
 639
 640 static int
 641 l2fwd_ivshmem_launch_one_lcore(__attribute__((unused)) void *dummy)
 642 {
 643         fwd_loop();
 644         return 0;
 645 }
 646
 647 int main(int argc, char **argv)
 648 {
 649         char name[RTE_RING_NAMESIZE];
 650         struct rte_ring *r;
 651         struct lcore_queue_conf *qconf;
 652         struct rte_eth_dev_info dev_info;
 653         uint8_t portid, port_nr;
 654         uint8_t nb_ports, nb_ports_available;
 655         uint8_t nb_ports_in_mask;
 656         int ret;
 657         unsigned lcore_id, rx_lcore_id;
 658
 659         /* init EAL */
 660         ret = rte_eal_init(argc, argv);
 661         if (ret < 0)
 662                 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
 663         argc -= ret;
 664         argv += ret;
 665
 666         /* parse application arguments (after the EAL ones) */
 667         ret = l2fwd_ivshmem_parse_args(argc, argv);
 668         if (ret < 0)
 669                 rte_exit(EXIT_FAILURE, "Invalid l2fwd-ivshmem arguments\n");
 670
 671         /* create a shared mbuf pool */
 672         l2fwd_ivshmem_pktmbuf_pool =
 673                 rte_pktmbuf_pool_create(MBUF_MP_NAME, NB_MBUF, 32,
 674                         0, MBUF_DATA_SIZE, rte_socket_id());
 675         if (l2fwd_ivshmem_pktmbuf_pool == NULL)
 676                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
 677
 678         nb_ports = rte_eth_dev_count();
 679         if (nb_ports == 0)
 680                 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
 681
 682         if (nb_ports > RTE_MAX_ETHPORTS)
 683                 nb_ports = RTE_MAX_ETHPORTS;
 684
 685         /*
 686          * reserve memzone to communicate with VMs - we cannot use rte_malloc here
 687          * because while it is technically possible, it is a very bad idea to share
 688          * the heap between two primary processes.
 689          */
 690         ctrl_mz = rte_memzone_reserve(CTRL_MZ_NAME, sizeof(struct ivshmem_ctrl),
 691                         SOCKET_ID_ANY, 0);
 692         if (ctrl_mz == NULL)
 693                 rte_exit(EXIT_FAILURE, "Cannot reserve control memzone\n");
 694         ctrl = (struct ivshmem_ctrl*) ctrl_mz->addr;
 695
 696         memset(ctrl, 0, sizeof(struct ivshmem_ctrl));
 697
 698         /*
 699          * Each port is assigned an output port.
 700          */
 701         nb_ports_in_mask = 0;
 702         for (portid = 0; portid < nb_ports; portid++) {
 703                 /* skip ports that are not enabled */
 704                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0)
 705                         continue;
 706                 if (portid % 2) {
 707                         ctrl->vm_ports[nb_ports_in_mask].dst = &ctrl->vm_ports[nb_ports_in_mask-1];
 708                         ctrl->vm_ports[nb_ports_in_mask-1].dst = &ctrl->vm_ports[nb_ports_in_mask];
 709                 }
 710
 711                 nb_ports_in_mask++;
 712
 713                 rte_eth_dev_info_get(portid, &dev_info);
 714         }
 715         if (nb_ports_in_mask % 2) {
 716                 printf("Notice: odd number of ports in portmask.\n");
 717                 ctrl->vm_ports[nb_ports_in_mask-1].dst =
 718                                 &ctrl->vm_ports[nb_ports_in_mask-1];
 719         }
 720
 721         rx_lcore_id = 0;
 722         qconf = NULL;
 723
 724         printf("Initializing ports configuration...\n");
 725
 726         nb_ports_available = nb_ports;
 727
 728         /* Initialise each port */
 729         for (portid = 0; portid < nb_ports; portid++) {
 730
 731                 /* skip ports that are not enabled */
 732                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0) {
 733                         printf("Skipping disabled port %u\n", (unsigned) portid);
 734                         nb_ports_available--;
 735                         continue;
 736                 }
 737
 738                 /* init port */
 739                 printf("Initializing port %u... ", (unsigned) portid);
 740                 fflush(stdout);
 741                 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
 742                 if (ret < 0)
 743                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
 744                                   ret, (unsigned) portid);
 745
 746                 rte_eth_macaddr_get(portid,&l2fwd_ivshmem_ports_eth_addr[portid]);
 747
 748                 /* init one RX queue */
 749                 fflush(stdout);
 750                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 751                                                  rte_eth_dev_socket_id(portid),
 752                                                  NULL,
 753                                                  l2fwd_ivshmem_pktmbuf_pool);
 754                 if (ret < 0)
 755                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
 756                                   ret, (unsigned) portid);
 757
 758                 /* init one TX queue on each port */
 759                 fflush(stdout);
 760                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
 761                                 rte_eth_dev_socket_id(portid),
 762                                 NULL);
 763                 if (ret < 0)
 764                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
 765                                 ret, (unsigned) portid);
 766
 767                 /* Start device */
 768                 ret = rte_eth_dev_start(portid);
 769                 if (ret < 0)
 770                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
 771                                   ret, (unsigned) portid);
 772
 773                 printf("done: \n");
 774
 775                 rte_eth_promiscuous_enable(portid);
 776
 777                 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
 778                                 (unsigned) portid,
 779                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[0],
 780                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[1],
 781                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[2],
 782                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[3],
 783                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[4],
 784                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[5]);
 785
 786                 /* initialize port stats */
 787                 memset(&port_statistics, 0, sizeof(port_statistics));
 788         }
 789
 790         if (!nb_ports_available) {
 791                 rte_exit(EXIT_FAILURE,
 792                         "All available ports are disabled. Please set portmask.\n");
 793         }
 794         port_nr = 0;
 795
 796         /* Initialize the port/queue configuration of each logical core */
 797         for (portid = 0; portid < nb_ports; portid++) {
 798                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0)
 799                         continue;
 800
 801                 /* get the lcore_id for this port */
 802                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 803                            lcore_queue_conf[rx_lcore_id].n_rx_port ==
 804                                            l2fwd_ivshmem_rx_queue_per_lcore) {
 805                         rx_lcore_id++;
 806                         if (rx_lcore_id >= RTE_MAX_LCORE)
 807                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 808                 }
 809
 810                 if (qconf != &lcore_queue_conf[rx_lcore_id])
 811                         /* Assigned a new logical core in the loop above. */
 812                         qconf = &lcore_queue_conf[rx_lcore_id];
 813
 814
 815                 rte_eth_macaddr_get(portid, &ctrl->vm_ports[port_nr].ethaddr);
 816
 817                 qconf->rx_port_list[qconf->n_rx_port] = portid;
 818                 qconf->port_param[qconf->n_rx_port] = &ctrl->vm_ports[port_nr];
 819                 qconf->n_rx_port++;
 820                 port_nr++;
 821                 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
 822         }
 823
 824         check_all_ports_link_status(nb_ports_available, l2fwd_ivshmem_enabled_port_mask);
 825
 826         /* create rings for each VM port (several ports can be on the same VM).
 827          * note that we store the pointers in ctrl - that way, they are the same
 828          * and valid across all VMs because ctrl is also in DPDK memory */
 829         for (portid = 0; portid < nb_ports_available; portid++) {
 830
 831                 /* RX ring. SP/SC because it's only used by host and a single VM */
 832                 snprintf(name, sizeof(name), "%s%i", RX_RING_PREFIX, portid);
 833                 r = rte_ring_create(name, NB_MBUF,
 834                                 SOCKET_ID_ANY, RING_F_SP_ENQ | RING_F_SC_DEQ);
 835                 if (r == NULL)
 836                         rte_exit(EXIT_FAILURE, "Cannot create ring %s\n", name);
 837
 838                 ctrl->vm_ports[portid].rx_ring = r;
 839
 840                 /* TX ring. SP/SC because it's only used by host and a single VM */
 841                 snprintf(name, sizeof(name), "%s%i", TX_RING_PREFIX, portid);
 842                 r = rte_ring_create(name, NB_MBUF,
 843                                 SOCKET_ID_ANY, RING_F_SP_ENQ | RING_F_SC_DEQ);
 844                 if (r == NULL)
 845                         rte_exit(EXIT_FAILURE, "Cannot create ring %s\n", name);
 846
 847                 ctrl->vm_ports[portid].tx_ring = r;
 848         }
 849
 850         /* create metadata, output cmdline */
 851         if (rte_ivshmem_metadata_create(METADATA_NAME) < 0)
 852                 rte_exit(EXIT_FAILURE, "Cannot create IVSHMEM metadata\n");
 853
 854         if (rte_ivshmem_metadata_add_memzone(ctrl_mz, METADATA_NAME))
 855                 rte_exit(EXIT_FAILURE, "Cannot add memzone to IVSHMEM metadata\n");
 856
 857         if (rte_ivshmem_metadata_add_mempool(l2fwd_ivshmem_pktmbuf_pool, METADATA_NAME))
 858                 rte_exit(EXIT_FAILURE, "Cannot add mbuf mempool to IVSHMEM metadata\n");
 859
 860         for (portid = 0; portid < nb_ports_available; portid++) {
 861                 if (rte_ivshmem_metadata_add_ring(ctrl->vm_ports[portid].rx_ring,
 862                                 METADATA_NAME) < 0)
 863                         rte_exit(EXIT_FAILURE, "Cannot add ring %s to IVSHMEM metadata\n",
 864                                         ctrl->vm_ports[portid].rx_ring->name);
 865                 if (rte_ivshmem_metadata_add_ring(ctrl->vm_ports[portid].tx_ring,
 866                                 METADATA_NAME) < 0)
 867                         rte_exit(EXIT_FAILURE, "Cannot add ring %s to IVSHMEM metadata\n",
 868                                         ctrl->vm_ports[portid].tx_ring->name);
 869         }
 870         generate_ivshmem_cmdline(METADATA_NAME);
 871
 872         ctrl->nb_ports = nb_ports_available;
 873
 874         printf("Waiting for VM to initialize...\n");
 875
 876         /* wait for VM to initialize */
 877         while (ctrl->state != STATE_FWD) {
 878                 if (ctrl->state == STATE_FAIL)
 879                         rte_exit(EXIT_FAILURE, "VM reported failure\n");
 880
 881                 sleep(1);
 882         }
 883
 884         printf("Done!\n");
 885
 886         sigsetup();
 887
 888         /* launch per-lcore init on every lcore */
 889         rte_eal_mp_remote_launch(l2fwd_ivshmem_launch_one_lcore, NULL, CALL_MASTER);
 890         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 891                 if (rte_eal_wait_lcore(lcore_id) < 0)
 892                         return -1;
 893         }
 894
 895         if (ctrl->state == STATE_FAIL)
 896                 rte_exit(EXIT_FAILURE, "VM reported failure\n");
 897
 898         return 0;
 899 }