VNFs/DPPD-PROX/main.c

   1 /*
   2 // Copyright (c) 2010-2017 Intel Corporation
   3 //
   4 // Licensed under the Apache License, Version 2.0 (the "License");
   5 // you may not use this file except in compliance with the License.
   6 // You may obtain a copy of the License at
   7 //
   8 //     http://www.apache.org/licenses/LICENSE-2.0
   9 //
  10 // Unless required by applicable law or agreed to in writing, software
  11 // distributed under the License is distributed on an "AS IS" BASIS,
  12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 // See the License for the specific language governing permissions and
  14 // limitations under the License.
  15 */
  16
  17 #include <string.h>
  18 #include <locale.h>
  19 #include <unistd.h>
  20 #include <signal.h>
  21
  22 #include <rte_cycles.h>
  23 #include <rte_atomic.h>
  24 #include <rte_table_hash.h>
  25 #include <rte_memzone.h>
  26 #include <rte_errno.h>
  27
  28 #include "prox_malloc.h"
  29 #include "run.h"
  30 #include "main.h"
  31 #include "log.h"
  32 #include "quit.h"
  33 #include "clock.h"
  34 #include "defines.h"
  35 #include "version.h"
  36 #include "prox_args.h"
  37 #include "prox_assert.h"
  38 #include "prox_cfg.h"
  39 #include "prox_shared.h"
  40 #include "prox_port_cfg.h"
  41 #include "toeplitz.h"
  42 #include "hash_utils.h"
  43 #include "handle_lb_net.h"
  44 #include "prox_cksum.h"
  45 #include "thread_nop.h"
  46 #include "thread_generic.h"
  47 #include "thread_pipeline.h"
  48 #include "cqm.h"
  49 #include "handle_master.h"
  50
  51 #if RTE_VERSION < RTE_VERSION_NUM(1,8,0,0)
  52 #define RTE_CACHE_LINE_SIZE CACHE_LINE_SIZE
  53 #endif
  54
  55 uint8_t lb_nb_txrings = 0xff;
  56 struct rte_ring *ctrl_rings[RTE_MAX_LCORE*MAX_TASKS_PER_CORE];
  57
  58 static void __attribute__((noreturn)) prox_usage(const char *prgname)
  59 {
  60         plog_info("\nUsage: %s [-f CONFIG_FILE] [-a|-e] [-m|-s|-i] [-w DEF] [-u] [-t]\n"
  61                   "\t-f CONFIG_FILE : configuration file to load, ./prox.cfg by default\n"
  62                   "\t-l LOG_FILE : log file name, ./prox.log by default\n"
  63                   "\t-p : include PID in log file name if default log file is used\n"
  64                   "\t-o DISPLAY: Set display to use, can be 'curses' (default), 'cli' or 'none'\n"
  65                   "\t-v verbosity : initial logging verbosity\n"
  66                   "\t-a : autostart all cores (by default)\n"
  67                   "\t-e : don't autostart\n"
  68                   "\t-n : Create NULL devices instead of using PCI devices, useful together with -i\n"
  69                   "\t-m : list supported task modes and exit\n"
  70                   "\t-s : check configuration file syntax and exit\n"
  71                   "\t-i : check initialization sequence and exit\n"
  72                   "\t-u : Listen on UDS /tmp/prox.sock\n"
  73                   "\t-t : Listen on TCP port 8474\n"
  74                   "\t-q : Pass argument to Lua interpreter, useful to define variables\n"
  75                   "\t-w : define variable using syntax varname=value\n"
  76                   "\t     takes precedence over variables defined in CONFIG_FILE\n"
  77                   "\t-k : Log statistics to file \"stats_dump\" in current directory\n"
  78                   "\t-d : Run as daemon, the parent process will block until PROX is not initialized\n"
  79                   "\t-z : Ignore CPU topology, implies -i\n"
  80                   "\t-r : Change initial screen refresh rate. If set to a lower than 0.001 seconds,\n"
  81                   "\t     screen refreshing will be disabled\n"
  82                   , prgname);
  83         exit(EXIT_FAILURE);
  84 }
  85
  86 static void check_mixed_normal_pipeline(void)
  87 {
  88         struct lcore_cfg *lconf = NULL;
  89         uint32_t lcore_id = -1;
  90
  91         while (prox_core_next(&lcore_id, 0) == 0) {
  92                 lconf = &lcore_cfg[lcore_id];
  93
  94                 int all_thread_nop = 1;
  95                 int generic = 0;
  96                 int pipeline = 0;
  97                 int l3 = 0;
  98                 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
  99                         struct task_args *targ = &lconf->targs[task_id];
 100                         l3 = !strcmp("l3", targ->sub_mode_str);
 101                         all_thread_nop = all_thread_nop && !l3 &&
 102                                 targ->task_init->thread_x == thread_nop;
 103
 104                         pipeline = pipeline || targ->task_init->thread_x == thread_pipeline;
 105                         generic = generic || targ->task_init->thread_x == thread_generic || l3;
 106                 }
 107                 PROX_PANIC(generic && pipeline, "Can't run both pipeline and normal thread on same core\n");
 108
 109                 if (all_thread_nop)
 110                         lconf->thread_x = thread_nop;
 111                 else {
 112                         lconf->thread_x = thread_generic;
 113                 }
 114         }
 115 }
 116
 117 static void check_zero_rx(void)
 118 {
 119         struct lcore_cfg *lconf = NULL;
 120         struct task_args *targ;
 121
 122         while (core_targ_next(&lconf, &targ, 0) == 0) {
 123                 if (targ->nb_rxports != 0) {
 124                         PROX_PANIC(task_init_flag_set(targ->task_init, TASK_FEATURE_NO_RX),
 125                            "\tCore %u task %u: rx_ports configured while mode %s does not use it\n", lconf->id, targ->id, targ->task_init->mode_str);
 126                 }
 127         }
 128 }
 129
 130 static void check_missing_rx(void)
 131 {
 132         struct lcore_cfg *lconf = NULL, *rx_lconf = NULL, *tx_lconf = NULL;
 133         struct task_args *targ, *rx_targ = NULL, *tx_targ = NULL;
 134         struct prox_port_cfg *port;
 135         uint8_t port_id, rx_port_id, ok;
 136
 137         while (core_targ_next(&lconf, &targ, 0) == 0) {
 138                 PROX_PANIC((targ->flags & TASK_ARG_RX_RING) && targ->rx_rings[0] == 0 && !targ->tx_opt_ring_task,
 139                            "Configuration Error - Core %u task %u Receiving from ring, but nobody xmitting to this ring\n", lconf->id, targ->id);
 140                 if (targ->nb_rxports == 0 && targ->nb_rxrings == 0) {
 141                         PROX_PANIC(!task_init_flag_set(targ->task_init, TASK_FEATURE_NO_RX),
 142                                    "\tCore %u task %u: no rx_ports and no rx_rings configured while required by mode %s\n", lconf->id, targ->id, targ->task_init->mode_str);
 143                 }
 144         }
 145
 146         lconf = NULL;
 147         while (core_targ_next(&lconf, &targ, 0) == 0) {
 148                 if (strcmp(targ->sub_mode_str, "l3") != 0)
 149                         continue;
 150
 151                 PROX_PANIC((targ->nb_rxports == 0) && (targ->nb_txports == 0), "L3 task must have a RX or a TX port\n");
 152                 // If the L3 sub_mode receives from a port, check that there is at least one core/task
 153                 // transmitting to this port in L3 sub_mode
 154                 for (uint8_t i = 0; i < targ->nb_rxports; ++i) {
 155                         rx_port_id = targ->rx_port_queue[i].port;
 156                         ok = 0;
 157                         tx_lconf = NULL;
 158                         while (core_targ_next(&tx_lconf, &tx_targ, 0) == 0) {
 159                                 if ((port_id = tx_targ->tx_port_queue[0].port) == OUT_DISCARD)
 160                                         continue;
 161                                 if ((rx_port_id == port_id) && (tx_targ->flags & TASK_ARG_L3)){
 162                                         ok = 1;
 163                                         break;
 164                                 }
 165                         }
 166                         PROX_PANIC(ok == 0, "RX L3 sub mode for port %d on core %d task %d, but no core/task transmitting on that port\n", rx_port_id, lconf->id, targ->id);
 167                 }
 168
 169                 // If the L3 sub_mode transmits to a port, check that there is at least one core/task
 170                 // receiving from that port in L3 sub_mode.
 171                 if ((port_id = targ->tx_port_queue[0].port) == OUT_DISCARD)
 172                         continue;
 173                 rx_lconf = NULL;
 174                 ok = 0;
 175                 plog_info("\tCore %d task %d transmitting to port %d in L3 mode\n", lconf->id, targ->id, port_id);
 176                 while (core_targ_next(&rx_lconf, &rx_targ, 0) == 0) {
 177                         for (uint8_t i = 0; i < rx_targ->nb_rxports; ++i) {
 178                                 rx_port_id = rx_targ->rx_port_queue[i].port;
 179                                 if ((rx_port_id == port_id) && (rx_targ->flags & TASK_ARG_L3)){
 180                                         ok = 1;
 181                                         break;
 182                                 }
 183                         }
 184                         if (ok == 1) {
 185                                 plog_info("\tCore %d task %d has found core %d task %d receiving from port %d\n", lconf->id, targ->id, rx_lconf->id, rx_targ->id, port_id);
 186                                 break;
 187                         }
 188                 }
 189                 PROX_PANIC(ok == 0, "L3 sub mode for port %d on core %d task %d, but no core/task receiving on that port\n", port_id, lconf->id, targ->id);
 190         }
 191 }
 192
 193 static void check_cfg_consistent(void)
 194 {
 195         check_missing_rx();
 196         check_zero_rx();
 197         check_mixed_normal_pipeline();
 198 }
 199
 200 static void plog_all_rings(void)
 201 {
 202         struct lcore_cfg *lconf = NULL;
 203         struct task_args *targ;
 204
 205         while (core_targ_next(&lconf, &targ, 0) == 0) {
 206                 for (uint8_t ring_idx = 0; ring_idx < targ->nb_rxrings; ++ring_idx) {
 207                         plog_info("\tCore %u, task %u, rx_ring[%u] %p\n", lconf->id, targ->id, ring_idx, targ->rx_rings[ring_idx]);
 208                 }
 209         }
 210 }
 211
 212 static int chain_flag_state(struct task_args *targ, uint64_t flag, int is_set)
 213 {
 214         if (task_init_flag_set(targ->task_init, flag) == is_set)
 215                 return 1;
 216
 217         int ret = 0;
 218
 219         for (uint32_t i = 0; i < targ->n_prev_tasks; ++i) {
 220                 ret = chain_flag_state(targ->prev_tasks[i], flag, is_set);
 221                 if (ret)
 222                         return 1;
 223         }
 224         return 0;
 225 }
 226
 227 static int chain_flag_always_set(struct task_args *targ, uint64_t flag)
 228 {
 229         return (!chain_flag_state(targ, flag, 0));
 230 }
 231
 232 static int chain_flag_never_set(struct task_args *targ, uint64_t flag)
 233 {
 234         return (!chain_flag_state(targ, flag, 1));
 235 }
 236
 237 static int chain_flag_sometimes_set(struct task_args *targ, uint64_t flag)
 238 {
 239         return (chain_flag_state(targ, flag, 1));
 240 }
 241
 242 static void configure_if_tx_queues(struct task_args *targ, uint8_t socket)
 243 {
 244         uint8_t if_port;
 245
 246         for (uint8_t i = 0; i < targ->nb_txports; ++i) {
 247                 if_port = targ->tx_port_queue[i].port;
 248
 249                 PROX_PANIC(if_port == OUT_DISCARD, "port misconfigured, exiting\n");
 250
 251                 PROX_PANIC(!prox_port_cfg[if_port].active, "\tPort %u not used, skipping...\n", if_port);
 252
 253                 int dsocket = prox_port_cfg[if_port].socket;
 254                 if (dsocket != -1 && dsocket != socket) {
 255                         plog_warn("TX core on socket %d while device on socket %d\n", socket, dsocket);
 256                 }
 257
 258                 if (prox_port_cfg[if_port].tx_ring[0] == '\0') {  // Rings-backed port can use single queue
 259                         targ->tx_port_queue[i].queue = prox_port_cfg[if_port].n_txq;
 260                         prox_port_cfg[if_port].n_txq++;
 261                 } else {
 262                         prox_port_cfg[if_port].n_txq = 1;
 263                         targ->tx_port_queue[i].queue = 0;
 264                 }
 265                 /* By default OFFLOAD is enabled, but if the whole
 266                    chain has NOOFFLOADS set all the way until the
 267                    first task that receives from a port, it will be
 268                    disabled for the destination port. */
 269 #if RTE_VERSION < RTE_VERSION_NUM(18,8,0,1)
 270                 if (chain_flag_always_set(targ, TASK_FEATURE_TXQ_FLAGS_NOOFFLOADS)) {
 271                         prox_port_cfg[if_port].tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOOFFLOADS;
 272                 }
 273 #else
 274                 if (chain_flag_always_set(targ, TASK_FEATURE_TXQ_FLAGS_NOOFFLOADS)) {
 275                         prox_port_cfg[if_port].requested_tx_offload &= ~(DEV_TX_OFFLOAD_IPV4_CKSUM | DEV_TX_OFFLOAD_UDP_CKSUM);
 276                 }
 277 #endif
 278         }
 279 }
 280
 281 static void configure_if_rx_queues(struct task_args *targ, uint8_t socket)
 282 {
 283         struct prox_port_cfg *port;
 284         for (int i = 0; i < targ->nb_rxports; i++) {
 285                 uint8_t if_port = targ->rx_port_queue[i].port;
 286
 287                 if (if_port == OUT_DISCARD) {
 288                         return;
 289                 }
 290
 291                 port = &prox_port_cfg[if_port];
 292                 PROX_PANIC(!port->active, "Port %u not used, aborting...\n", if_port);
 293
 294                 if(port->rx_ring[0] != '\0') {
 295                         port->n_rxq = 0;
 296                 }
 297
 298                 // If the mbuf size (of the rx task) is not big enough, we might receive multiple segments
 299                 // This is usually the case when setting a big mtu size i.e. enabling jumbo frames.
 300                 // If the packets get transmitted, then multi segments will have to be enabled on the TX port
 301                 uint16_t max_frame_size = port->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * PROX_VLAN_TAG_SIZE;
 302                 if (max_frame_size + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM > targ->mbuf_size) {
 303                         targ->task_init->flag_features |= TASK_FEATURE_TXQ_FLAGS_MULTSEGS;
 304                 }
 305                 targ->rx_port_queue[i].queue = port->n_rxq;
 306                 port->pool[targ->rx_port_queue[i].queue] = targ->pool;
 307                 port->pool_size[targ->rx_port_queue[i].queue] = targ->nb_mbuf - 1;
 308                 port->n_rxq++;
 309
 310                 int dsocket = port->socket;
 311                 if (dsocket != -1 && dsocket != socket) {
 312                         plog_warn("RX core on socket %d while device on socket %d\n", socket, dsocket);
 313                 }
 314         }
 315 }
 316
 317 static void configure_if_queues(void)
 318 {
 319         struct lcore_cfg *lconf = NULL;
 320         struct task_args *targ;
 321         uint8_t socket;
 322
 323         while (core_targ_next(&lconf, &targ, 0) == 0) {
 324                 socket = rte_lcore_to_socket_id(lconf->id);
 325
 326                 configure_if_rx_queues(targ, socket);
 327                 configure_if_tx_queues(targ, socket);
 328         }
 329 }
 330
 331 static void configure_tx_queue_flags(void)
 332 {
 333         struct lcore_cfg *lconf = NULL;
 334         struct task_args *targ;
 335         uint8_t socket;
 336         uint8_t if_port;
 337
 338         while (core_targ_next(&lconf, &targ, 0) == 0) {
 339                 socket = rte_lcore_to_socket_id(lconf->id);
 340                 for (uint8_t i = 0; i < targ->nb_txports; ++i) {
 341                         if_port = targ->tx_port_queue[i].port;
 342 #if RTE_VERSION < RTE_VERSION_NUM(18,8,0,1)
 343                         /* Set the ETH_TXQ_FLAGS_NOREFCOUNT flag if none of
 344                         the tasks up to the task transmitting to the port
 345                         use refcnt. */
 346                         if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_REFCOUNT)) {
 347                                 prox_port_cfg[if_port].tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 348                         }
 349 #else
 350                         /* Set the DEV_TX_OFFLOAD_MBUF_FAST_FREE flag if none of
 351                         the tasks up to the task transmitting to the port
 352                         use refcnt and per-queue all mbufs comes from the same mempool. */
 353                         if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_REFCOUNT)) {
 354                                 if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_MULTIPLE_MEMPOOL))
 355                                         prox_port_cfg[if_port].requested_tx_offload |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 356                         }
 357 #endif
 358                 }
 359         }
 360 }
 361
 362 static void configure_multi_segments(void)
 363 {
 364         struct lcore_cfg *lconf = NULL;
 365         struct task_args *targ;
 366         uint8_t if_port;
 367
 368         while (core_targ_next(&lconf, &targ, 0) == 0) {
 369                 for (uint8_t i = 0; i < targ->nb_txports; ++i) {
 370                         if_port = targ->tx_port_queue[i].port;
 371                         // Multi segment is disabled for most tasks. It is only enabled for tasks requiring big packets.
 372 #if RTE_VERSION < RTE_VERSION_NUM(18,8,0,1)
 373                         // We can only enable "no multi segment" if no such task exists in the chain of tasks.
 374                         if (chain_flag_never_set(targ, TASK_FEATURE_TXQ_FLAGS_MULTSEGS)) {
 375                                 prox_port_cfg[if_port].tx_conf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 376                         }
 377 #else
 378                         // We enable "multi segment" if at least one task requires it in the chain of tasks.
 379                         if (chain_flag_sometimes_set(targ, TASK_FEATURE_TXQ_FLAGS_MULTSEGS)) {
 380                                 prox_port_cfg[if_port].requested_tx_offload |= DEV_TX_OFFLOAD_MULTI_SEGS;
 381                         }
 382 #endif
 383                 }
 384         }
 385 }
 386
 387 static const char *gen_ring_name(void)
 388 {
 389         static char retval[] = "XX";
 390         static const char* ring_names =
 391                 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
 392                 "abcdefghijklmnopqrstuvwxyz"
 393                 "[\\]^_`!\"#$%&'()*+,-./:;<="
 394                 ">?@{|}0123456789";
 395         static int idx2 = 0;
 396
 397         int idx = idx2;
 398
 399         retval[0] = ring_names[idx % strlen(ring_names)];
 400         idx /= strlen(ring_names);
 401         retval[1] = idx ? ring_names[(idx - 1) % strlen(ring_names)] : 0;
 402
 403         idx2++;
 404
 405         return retval;
 406 }
 407
 408 struct ring_init_stats {
 409         uint32_t n_pkt_rings;
 410         uint32_t n_ctrl_rings;
 411         uint32_t n_opt_rings;
 412 };
 413
 414 static uint32_t ring_init_stats_total(const struct ring_init_stats *ris)
 415 {
 416         return ris->n_pkt_rings + ris->n_ctrl_rings + ris->n_opt_rings;
 417 }
 418
 419 static uint32_t count_incoming_tasks(uint32_t lcore_worker, uint32_t dest_task)
 420 {
 421         struct lcore_cfg *lconf = NULL;
 422         struct task_args *targ;
 423         uint32_t ret = 0;
 424         struct core_task ct;
 425
 426         while (core_targ_next(&lconf, &targ, 0) == 0) {
 427                 for (uint8_t idxx = 0; idxx < MAX_PROTOCOLS; ++idxx) {
 428                         for (uint8_t ridx = 0; ridx < targ->core_task_set[idxx].n_elems; ++ridx) {
 429                                 ct = targ->core_task_set[idxx].core_task[ridx];
 430
 431                                 if (dest_task == ct.task && lcore_worker == ct.core)
 432                                         ret++;
 433                         }
 434                 }
 435         }
 436         return ret;
 437 }
 438
 439 static struct rte_ring *get_existing_ring(uint32_t lcore_id, uint32_t task_id)
 440 {
 441         if (!prox_core_active(lcore_id, 0))
 442                 return NULL;
 443
 444         struct lcore_cfg *lconf = &lcore_cfg[lcore_id];
 445
 446         if (task_id >= lconf->n_tasks_all)
 447                 return NULL;
 448
 449         if (lconf->targs[task_id].nb_rxrings == 0)
 450                 return NULL;
 451
 452         return lconf->targs[task_id].rx_rings[0];
 453 }
 454
 455 static struct rte_ring *init_ring_between_tasks(struct lcore_cfg *lconf, struct task_args *starg,
 456                                     const struct core_task ct, uint8_t ring_idx, int idx,
 457                                     struct ring_init_stats *ris)
 458 {
 459         uint8_t socket;
 460         struct rte_ring *ring = NULL;
 461         struct lcore_cfg *lworker;
 462         struct task_args *dtarg;
 463
 464         PROX_ASSERT(prox_core_active(ct.core, 0));
 465         lworker = &lcore_cfg[ct.core];
 466
 467         /* socket used is the one that the sending core resides on */
 468         socket = rte_lcore_to_socket_id(lconf->id);
 469
 470         plog_info("\t\tCreating ring on socket %u with size %u\n"
 471                   "\t\t\tsource core, task and socket = %u, %u, %u\n"
 472                   "\t\t\tdestination core, task and socket = %u, %u, %u\n"
 473                   "\t\t\tdestination worker id = %u\n",
 474                   socket, starg->ring_size,
 475                   lconf->id, starg->id, socket,
 476                   ct.core, ct.task, rte_lcore_to_socket_id(ct.core),
 477                   ring_idx);
 478
 479         if (ct.type) {
 480                 struct rte_ring **dring = NULL;
 481
 482                 if (ct.type == CTRL_TYPE_MSG)
 483                         dring = &lworker->ctrl_rings_m[ct.task];
 484                 else if (ct.type == CTRL_TYPE_PKT) {
 485                         dring = &lworker->ctrl_rings_p[ct.task];
 486                         starg->flags |= TASK_ARG_CTRL_RINGS_P;
 487                 }
 488
 489                 if (*dring == NULL)
 490                         ring = rte_ring_create(gen_ring_name(), starg->ring_size, socket, RING_F_SC_DEQ);
 491                 else
 492                         ring = *dring;
 493                 PROX_PANIC(ring == NULL, "Cannot create ring to connect I/O core %u with worker core %u\n", lconf->id, ct.core);
 494
 495                 starg->tx_rings[starg->tot_n_txrings_inited] = ring;
 496                 starg->tot_n_txrings_inited++;
 497                 *dring = ring;
 498                 if (lconf->id == prox_cfg.master) {
 499                         ctrl_rings[ct.core*MAX_TASKS_PER_CORE + ct.task] = ring;
 500                 } else if (ct.core == prox_cfg.master) {
 501                         starg->ctrl_plane_ring = ring;
 502                 }
 503
 504                 plog_info("\t\tCore %u task %u to -> core %u task %u ctrl_ring %s %p %s\n",
 505                           lconf->id, starg->id, ct.core, ct.task, ct.type == CTRL_TYPE_PKT?
 506                           "pkt" : "msg", ring, ring->name);
 507                 ris->n_ctrl_rings++;
 508                 return ring;
 509         }
 510
 511         dtarg = &lworker->targs[ct.task];
 512         lworker->targs[ct.task].worker_thread_id = ring_idx;
 513         PROX_ASSERT(dtarg->flags & TASK_ARG_RX_RING);
 514         PROX_ASSERT(ct.task < lworker->n_tasks_all);
 515
 516         /* If all the following conditions are met, the ring can be
 517            optimized away. */
 518         if (!task_is_master(starg) && !task_is_master(dtarg) && starg->lconf->id == dtarg->lconf->id &&
 519             starg->nb_txrings == 1 && idx == 0 && dtarg->task &&
 520             dtarg->tot_rxrings == 1 && starg->task == dtarg->task - 1) {
 521                 plog_info("\t\tOptimizing away ring on core %u from task %u to task %u\n",
 522                           dtarg->lconf->id, starg->task, dtarg->task);
 523                 /* No need to set up ws_mbuf. */
 524                 starg->tx_opt_ring = 1;
 525                 /* During init of destination task, the buffer in the
 526                    source task will be initialized. */
 527                 dtarg->tx_opt_ring_task = starg;
 528                 ris->n_opt_rings++;
 529                 ++dtarg->nb_rxrings;
 530                 return NULL;
 531         }
 532
 533         int ring_created = 1;
 534         /* Only create multi-producer rings if configured to do so AND
 535            there is only one task sending to the task */
 536         if ((prox_cfg.flags & DSF_MP_RINGS && count_incoming_tasks(ct.core, ct.task) > 1)
 537                 || (prox_cfg.flags & DSF_ENABLE_BYPASS)) {
 538                 ring = get_existing_ring(ct.core, ct.task);
 539
 540                 if (ring) {
 541                         plog_info("\t\tCore %u task %u creatign MP ring %p to core %u task %u\n",
 542                                   lconf->id, starg->id, ring, ct.core, ct.task);
 543                         ring_created = 0;
 544                 }
 545                 else {
 546                         ring = rte_ring_create(gen_ring_name(), starg->ring_size, socket, RING_F_SC_DEQ);
 547                         plog_info("\t\tCore %u task %u using MP ring %p from core %u task %u\n",
 548                                   lconf->id, starg->id, ring, ct.core, ct.task);
 549                 }
 550         }
 551         else
 552                 ring = rte_ring_create(gen_ring_name(), starg->ring_size, socket, RING_F_SP_ENQ | RING_F_SC_DEQ);
 553
 554         PROX_PANIC(ring == NULL, "Cannot create ring to connect I/O core %u with worker core %u\n", lconf->id, ct.core);
 555
 556         starg->tx_rings[starg->tot_n_txrings_inited] = ring;
 557         starg->tot_n_txrings_inited++;
 558
 559         if (ring_created) {
 560                 PROX_ASSERT(dtarg->nb_rxrings < MAX_RINGS_PER_TASK);
 561                 dtarg->rx_rings[dtarg->nb_rxrings] = ring;
 562                 ++dtarg->nb_rxrings;
 563                 if (dtarg->nb_rxrings > 1)
 564                         dtarg->task_init->flag_features |= TASK_FEATURE_TXQ_FLAGS_MULTIPLE_MEMPOOL;
 565         }
 566         dtarg->nb_slave_threads = starg->core_task_set[idx].n_elems;
 567         dtarg->lb_friend_core = lconf->id;
 568         dtarg->lb_friend_task = starg->id;
 569         plog_info("\t\tWorker thread %d has core %d, task %d as a lb friend\n", ct.core, lconf->id, starg->id);
 570         plog_info("\t\tCore %u task %u tx_ring[%u] -> core %u task %u rx_ring[%u] %p %s %u WT\n",
 571                   lconf->id, starg->id, ring_idx, ct.core, ct.task, dtarg->nb_rxrings, ring, ring->name,
 572                   dtarg->nb_slave_threads);
 573         ++ris->n_pkt_rings;
 574         return ring;
 575 }
 576
 577 static void init_rings(void)
 578 {
 579         struct lcore_cfg *lconf = NULL;
 580         struct task_args *starg;
 581         struct ring_init_stats ris = {0};
 582
 583         while (core_targ_next(&lconf, &starg, 1) == 0) {
 584                 plog_info("\t*** Initializing rings on core %u, task %u ***\n", lconf->id, starg->id);
 585                 for (uint8_t idx = 0; idx < MAX_PROTOCOLS; ++idx) {
 586                         for (uint8_t ring_idx = 0; ring_idx < starg->core_task_set[idx].n_elems; ++ring_idx) {
 587                                 PROX_ASSERT(ring_idx < MAX_WT_PER_LB);
 588                                 PROX_ASSERT(starg->tot_n_txrings_inited < MAX_RINGS_PER_TASK);
 589
 590                                 struct core_task ct = starg->core_task_set[idx].core_task[ring_idx];
 591                                 init_ring_between_tasks(lconf, starg, ct, ring_idx, idx, &ris);
 592                         }
 593                 }
 594         }
 595
 596         plog_info("\tInitialized %d rings:\n"
 597                   "\t\tNumber of packet rings: %u\n"
 598                   "\t\tNumber of control rings: %u\n"
 599                   "\t\tNumber of optimized rings: %u\n",
 600                   ring_init_stats_total(&ris),
 601                   ris.n_pkt_rings,
 602                   ris.n_ctrl_rings,
 603                   ris.n_opt_rings);
 604
 605         lconf = NULL;
 606         struct prox_port_cfg *port;
 607         while (core_targ_next(&lconf, &starg, 1) == 0) {
 608                 if ((starg->task_init) && (starg->flags & TASK_ARG_L3)) {
 609                         struct core_task ct;
 610                         ct.core = prox_cfg.master;
 611                         ct.task = 0;
 612                         ct.type = CTRL_TYPE_PKT;
 613                         struct rte_ring *rx_ring = init_ring_between_tasks(lconf, starg, ct, 0, 0, &ris);
 614
 615                         ct.core = lconf->id;
 616                         ct.task = starg->id;;
 617                         struct rte_ring *tx_ring = init_ring_between_tasks(lcore_cfg, lcore_cfg[prox_cfg.master].targs, ct, 0, 0, &ris);
 618                 }
 619         }
 620 }
 621
 622 static void shuffle_mempool(struct rte_mempool* mempool, uint32_t nb_mbuf)
 623 {
 624         struct rte_mbuf** pkts = prox_zmalloc(nb_mbuf * sizeof(*pkts), rte_socket_id());
 625         uint64_t got = 0;
 626
 627         while (rte_mempool_get_bulk(mempool, (void**)(pkts + got), 1) == 0)
 628                 ++got;
 629
 630         while (got) {
 631                 int idx;
 632                 do {
 633                         idx = rand() % nb_mbuf - 1;
 634                 } while (pkts[idx] == 0);
 635
 636                 rte_mempool_put_bulk(mempool, (void**)&pkts[idx], 1);
 637                 pkts[idx] = 0;
 638                 --got;
 639         };
 640         prox_free(pkts);
 641 }
 642
 643 static void set_mbuf_size(struct task_args *targ)
 644 {
 645         /* mbuf size can be set
 646          *  - from config file (highest priority, overwriting any other config) - should only be used as workaround
 647          *  - defaulted to MBUF_SIZE.
 648          * Except if set explicitely, ensure that size is big enough for vmxnet3 driver
 649          */
 650         if (targ->mbuf_size)
 651                 return;
 652
 653         targ->mbuf_size = MBUF_SIZE;
 654         struct prox_port_cfg *port;
 655         uint16_t max_frame_size = 0, min_buffer_size = 0;
 656         int i40e = 0;
 657         for (int i = 0; i < targ->nb_rxports; i++) {
 658                 uint8_t if_port = targ->rx_port_queue[i].port;
 659
 660                 if (if_port == OUT_DISCARD) {
 661                         continue;
 662                 }
 663                 port = &prox_port_cfg[if_port];
 664                 if (max_frame_size < port->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * PROX_VLAN_TAG_SIZE)
 665                         max_frame_size = port->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * PROX_VLAN_TAG_SIZE;
 666                 if (min_buffer_size < port->min_rx_bufsize)
 667                         min_buffer_size = port->min_rx_bufsize;
 668
 669                 // Check whether we receive from i40e. This driver have extra mbuf size requirements
 670                 if (strcmp(port->short_name, "i40e") == 0)
 671                         i40e = 1;
 672         }
 673         if (i40e) {
 674                 // i40e supports a maximum of 5 descriptors chained
 675                 uint16_t required_mbuf_size = RTE_ALIGN(max_frame_size / 5, 128) + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM;
 676                 if (required_mbuf_size > targ->mbuf_size) {
 677                         targ->mbuf_size = required_mbuf_size;
 678                         plog_info("\t\tSetting mbuf_size to %u to support frame_size %u\n", targ->mbuf_size, max_frame_size);
 679                 }
 680         }
 681         if (min_buffer_size > targ->mbuf_size) {
 682                 plog_warn("Mbuf size might be too small. This might result in packet segmentation and memory leak\n");
 683         }
 684
 685 }
 686
 687 static void setup_mempools_unique_per_socket(void)
 688 {
 689         uint32_t flags = 0;
 690         char name[64];
 691         struct lcore_cfg *lconf = NULL;
 692         struct task_args *targ;
 693
 694         struct rte_mempool     *pool[MAX_SOCKETS];
 695         uint32_t mbuf_count[MAX_SOCKETS] = {0};
 696         uint32_t nb_cache_mbuf[MAX_SOCKETS] = {0};
 697         uint32_t mbuf_size[MAX_SOCKETS] = {0};
 698
 699         while (core_targ_next_early(&lconf, &targ, 0) == 0) {
 700                 PROX_PANIC(targ->task_init == NULL, "task_init = NULL, is mode specified for core %d, task %d ?\n", lconf->id, targ->id);
 701                 uint8_t socket = rte_lcore_to_socket_id(lconf->id);
 702                 PROX_ASSERT(socket < MAX_SOCKETS);
 703
 704                 set_mbuf_size(targ);
 705                 if (targ->rx_port_queue[0].port != OUT_DISCARD) {
 706                         struct prox_port_cfg* port_cfg = &prox_port_cfg[targ->rx_port_queue[0].port];
 707                         PROX_ASSERT(targ->nb_mbuf != 0);
 708                         mbuf_count[socket] += targ->nb_mbuf;
 709                         if (nb_cache_mbuf[socket] == 0)
 710                                 nb_cache_mbuf[socket] = targ->nb_cache_mbuf;
 711                         else {
 712                                 PROX_PANIC(nb_cache_mbuf[socket] != targ->nb_cache_mbuf,
 713                                            "all mbuf_cache must have the same size if using a unique mempool per socket\n");
 714                         }
 715                         if (mbuf_size[socket] == 0)
 716                                 mbuf_size[socket] = targ->mbuf_size;
 717                         else {
 718                                 PROX_PANIC(mbuf_size[socket] != targ->mbuf_size,
 719                                            "all mbuf_size must have the same size if using a unique mempool per socket\n");
 720                         }
 721                 }
 722         }
 723         for (int i = 0 ; i < MAX_SOCKETS; i++) {
 724                 if (mbuf_count[i] != 0) {
 725                         sprintf(name, "socket_%u_pool", i);
 726                         pool[i] = rte_mempool_create(name,
 727                                                      mbuf_count[i] - 1, mbuf_size[i],
 728                                                      nb_cache_mbuf[i],
 729                                                      sizeof(struct rte_pktmbuf_pool_private),
 730                                                      rte_pktmbuf_pool_init, NULL,
 731                                                      prox_pktmbuf_init, NULL,
 732                                                      i, flags);
 733                         PROX_PANIC(pool[i] == NULL, "\t\tError: cannot create mempool for socket %u\n", i);
 734                         plog_info("\t\tMempool %p size = %u * %u cache %u, socket %d\n", pool[i],
 735                                   mbuf_count[i], mbuf_size[i], nb_cache_mbuf[i], i);
 736
 737                         if (prox_cfg.flags & DSF_SHUFFLE) {
 738                                 shuffle_mempool(pool[i], mbuf_count[i]);
 739                         }
 740                 }
 741         }
 742
 743         lconf = NULL;
 744         while (core_targ_next_early(&lconf, &targ, 0) == 0) {
 745                 uint8_t socket = rte_lcore_to_socket_id(lconf->id);
 746
 747                 if (targ->rx_port_queue[0].port != OUT_DISCARD) {
 748                         /* use this pool for the interface that the core is receiving from */
 749                         /* If one core receives from multiple ports, all the ports use the same mempool */
 750                         targ->pool = pool[socket];
 751                         /* Set the number of mbuf to the number of the unique mempool, so that the used and free work */
 752                         targ->nb_mbuf = mbuf_count[socket];
 753                         plog_info("\t\tMempool %p size = %u * %u cache %u, socket %d\n", targ->pool,
 754                                   targ->nb_mbuf, mbuf_size[socket], targ->nb_cache_mbuf, socket);
 755                 }
 756         }
 757 }
 758
 759 static void setup_mempool_for_rx_task(struct lcore_cfg *lconf, struct task_args *targ)
 760 {
 761         const uint8_t socket = rte_lcore_to_socket_id(lconf->id);
 762         struct prox_port_cfg *port_cfg = &prox_port_cfg[targ->rx_port_queue[0].port];
 763         const struct rte_memzone *mz;
 764         struct rte_mempool *mp = NULL;
 765         uint32_t flags = 0;
 766         char memzone_name[64];
 767         char name[64];
 768
 769         set_mbuf_size(targ);
 770
 771         /* allocate memory pool for packets */
 772         PROX_ASSERT(targ->nb_mbuf != 0);
 773
 774         if (targ->pool_name[0] == '\0') {
 775                 sprintf(name, "core_%u_port_%u_pool", lconf->id, targ->id);
 776         }
 777
 778         snprintf(memzone_name, sizeof(memzone_name)-1, "MP_%s", targ->pool_name);
 779         mz = rte_memzone_lookup(memzone_name);
 780
 781         if (mz != NULL) {
 782                 mp = (struct rte_mempool*)mz->addr;
 783
 784                 targ->nb_mbuf = mp->size;
 785                 targ->pool = mp;
 786         }
 787
 788 #ifdef RTE_LIBRTE_IVSHMEM_FALSE
 789         if (mz != NULL && mp != NULL && mp->phys_addr != mz->ioremap_addr) {
 790                 /* Init mbufs with ioremap_addr for dma */
 791                 mp->phys_addr = mz->ioremap_addr;
 792                 mp->elt_pa[0] = mp->phys_addr + (mp->elt_va_start - (uintptr_t)mp);
 793
 794                 struct prox_pktmbuf_reinit_args init_args;
 795                 init_args.mp = mp;
 796                 init_args.lconf = lconf;
 797
 798                 uint32_t elt_sz = mp->elt_size + mp->header_size + mp->trailer_size;
 799                 rte_mempool_obj_iter((void*)mp->elt_va_start, mp->size, elt_sz, 1,
 800                                      mp->elt_pa, mp->pg_num, mp->pg_shift, prox_pktmbuf_reinit, &init_args);
 801         }
 802 #endif
 803
 804         /* Use this pool for the interface that the core is
 805            receiving from if one core receives from multiple
 806            ports, all the ports use the same mempool */
 807         if (targ->pool == NULL) {
 808                 plog_info("\t\tCreating mempool with name '%s'\n", name);
 809                 targ->pool = rte_mempool_create(name,
 810                                                 targ->nb_mbuf - 1, targ->mbuf_size,
 811                                                 targ->nb_cache_mbuf,
 812                                                 sizeof(struct rte_pktmbuf_pool_private),
 813                                                 rte_pktmbuf_pool_init, NULL,
 814                                                 prox_pktmbuf_init, lconf,
 815                                                 socket, flags);
 816         }
 817
 818         PROX_PANIC(targ->pool == NULL,
 819                    "\t\tError: cannot create mempool for core %u port %u: %s\n", lconf->id, targ->id, rte_strerror(rte_errno));
 820
 821         plog_info("\t\tMempool %p size = %u * %u cache %u, socket %d\n", targ->pool,
 822                   targ->nb_mbuf, targ->mbuf_size, targ->nb_cache_mbuf, socket);
 823         if (prox_cfg.flags & DSF_SHUFFLE) {
 824                 shuffle_mempool(targ->pool, targ->nb_mbuf);
 825         }
 826 }
 827
 828 static void setup_mempools_multiple_per_socket(void)
 829 {
 830         struct lcore_cfg *lconf = NULL;
 831         struct task_args *targ;
 832
 833         while (core_targ_next_early(&lconf, &targ, 0) == 0) {
 834                 PROX_PANIC(targ->task_init == NULL, "task_init = NULL, is mode specified for core %d, task %d ?\n", lconf->id, targ->id);
 835                 if (targ->rx_port_queue[0].port == OUT_DISCARD)
 836                         continue;
 837                 setup_mempool_for_rx_task(lconf, targ);
 838         }
 839 }
 840
 841 static void setup_mempools(void)
 842 {
 843         if (prox_cfg.flags & UNIQUE_MEMPOOL_PER_SOCKET)
 844                 setup_mempools_unique_per_socket();
 845         else
 846                 setup_mempools_multiple_per_socket();
 847 }
 848
 849 static void set_task_lconf(void)
 850 {
 851         struct lcore_cfg *lconf;
 852         uint32_t lcore_id = -1;
 853
 854         while(prox_core_next(&lcore_id, 1) == 0) {
 855                 lconf = &lcore_cfg[lcore_id];
 856                 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
 857                         lconf->targs[task_id].lconf = lconf;
 858                 }
 859         }
 860 }
 861
 862 static void set_dest_threads(void)
 863 {
 864         struct lcore_cfg *lconf = NULL;
 865         struct task_args *targ;
 866
 867         while (core_targ_next(&lconf, &targ, 0) == 0) {
 868                 for (uint8_t idx = 0; idx < MAX_PROTOCOLS; ++idx) {
 869                         for (uint8_t ring_idx = 0; ring_idx < targ->core_task_set[idx].n_elems; ++ring_idx) {
 870                                 struct core_task ct = targ->core_task_set[idx].core_task[ring_idx];
 871
 872                                 struct task_args *dest_task = core_targ_get(ct.core, ct.task);
 873                                 dest_task->prev_tasks[dest_task->n_prev_tasks++] = targ;
 874                         }
 875                 }
 876         }
 877 }
 878
 879 static void setup_all_task_structs_early_init(void)
 880 {
 881         struct lcore_cfg *lconf = NULL;
 882         struct task_args *targ;
 883
 884         plog_info("\t*** Calling early init on all tasks ***\n");
 885         while (core_targ_next(&lconf, &targ, 0) == 0) {
 886                 if (targ->task_init->early_init) {
 887                         targ->task_init->early_init(targ);
 888                 }
 889         }
 890 }
 891
 892 static void setup_all_task_structs(void)
 893 {
 894         struct lcore_cfg *lconf;
 895         uint32_t lcore_id = -1;
 896         struct task_base *tmaster = NULL;
 897
 898         while(prox_core_next(&lcore_id, 1) == 0) {
 899                 lconf = &lcore_cfg[lcore_id];
 900                 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
 901                         if (task_is_master(&lconf->targs[task_id])) {
 902                                 plog_info("\tInitializing MASTER struct for core %d task %d\n", lcore_id, task_id);
 903                                 lconf->tasks_all[task_id] = init_task_struct(&lconf->targs[task_id]);
 904                                 tmaster = lconf->tasks_all[task_id];
 905                         }
 906                 }
 907         }
 908         PROX_PANIC(tmaster == NULL, "Can't initialize master task\n");
 909         lcore_id = -1;
 910
 911         while(prox_core_next(&lcore_id, 1) == 0) {
 912                 lconf = &lcore_cfg[lcore_id];
 913                 plog_info("\tInitializing struct for core %d with %d task\n", lcore_id, lconf->n_tasks_all);
 914                 for (uint8_t task_id = 0; task_id < lconf->n_tasks_all; ++task_id) {
 915                         if (!task_is_master(&lconf->targs[task_id])) {
 916                                 plog_info("\tInitializing struct for core %d task %d\n", lcore_id, task_id);
 917                                 lconf->targs[task_id].tmaster = tmaster;
 918                                 lconf->tasks_all[task_id] = init_task_struct(&lconf->targs[task_id]);
 919                         }
 920                 }
 921         }
 922 }
 923
 924 static void init_port_activate(void)
 925 {
 926         struct lcore_cfg *lconf = NULL;
 927         struct task_args *targ;
 928         uint8_t port_id = 0;
 929
 930         while (core_targ_next_early(&lconf, &targ, 0) == 0) {
 931                 for (int i = 0; i < targ->nb_rxports; i++) {
 932                         port_id = targ->rx_port_queue[i].port;
 933                         prox_port_cfg[port_id].active = 1;
 934                 }
 935
 936                 for (int i = 0; i < targ->nb_txports; i++) {
 937                         port_id = targ->tx_port_queue[i].port;
 938                         prox_port_cfg[port_id].active = 1;
 939                 }
 940         }
 941 }
 942
 943 /* Initialize cores and allocate mempools */
 944 static void init_lcores(void)
 945 {
 946         struct lcore_cfg *lconf = 0;
 947         uint32_t lcore_id = -1;
 948
 949         while(prox_core_next(&lcore_id, 0) == 0) {
 950                 uint8_t socket = rte_lcore_to_socket_id(lcore_id);
 951                 PROX_PANIC(socket + 1 > MAX_SOCKETS, "Can't configure core %u (on socket %u). MAX_SOCKET is set to %d\n", lcore_id, socket, MAX_SOCKETS);
 952         }
 953
 954         /* need to allocate mempools as the first thing to use the lowest possible address range */
 955         plog_info("=== Initializing mempools ===\n");
 956         setup_mempools();
 957
 958         lcore_cfg_alloc_hp();
 959
 960         set_dest_threads();
 961         set_task_lconf();
 962
 963         plog_info("=== Initializing port addresses ===\n");
 964         init_port_addr();
 965
 966         plog_info("=== Initializing queue numbers on cores ===\n");
 967         configure_if_queues();
 968
 969         plog_info("=== Initializing rings on cores ===\n");
 970         init_rings();
 971
 972         configure_multi_segments();
 973         configure_tx_queue_flags();
 974
 975         plog_info("=== Checking configuration consistency ===\n");
 976         check_cfg_consistent();
 977
 978         plog_all_rings();
 979 }
 980
 981 static int setup_prox(int argc, char **argv)
 982 {
 983         if (prox_read_config_file() != 0 ||
 984             prox_setup_rte(argv[0]) != 0) {
 985                 return -1;
 986         }
 987
 988         if (prox_cfg.flags & DSF_CHECK_SYNTAX) {
 989                 plog_info("=== Configuration file syntax has been checked ===\n\n");
 990                 exit(EXIT_SUCCESS);
 991         }
 992
 993         init_port_activate();
 994         plog_info("=== Initializing rte devices ===\n");
 995         if (!(prox_cfg.flags & DSF_USE_DUMMY_DEVICES))
 996                 init_rte_ring_dev();
 997         init_rte_dev(prox_cfg.flags & DSF_USE_DUMMY_DEVICES);
 998         plog_info("=== Calibrating TSC overhead ===\n");
 999         clock_init();
1000         plog_info("\tTSC running at %"PRIu64" Hz\n", rte_get_tsc_hz());
1001
1002         init_lcores();
1003         plog_info("=== Initializing ports ===\n");
1004         init_port_all();
1005
1006         setup_all_task_structs_early_init();
1007         plog_info("=== Initializing tasks ===\n");
1008         setup_all_task_structs();
1009
1010         if (prox_cfg.logbuf_size) {
1011                 prox_cfg.logbuf = prox_zmalloc(prox_cfg.logbuf_size, rte_socket_id());
1012                 PROX_PANIC(prox_cfg.logbuf == NULL, "Failed to allocate memory for logbuf with size = %d\n", prox_cfg.logbuf_size);
1013         }
1014
1015         if (prox_cfg.flags & DSF_CHECK_INIT) {
1016                 plog_info("=== Initialization sequence completed ===\n\n");
1017                 exit(EXIT_SUCCESS);
1018         }
1019
1020         /* Current way that works to disable DPDK logging */
1021         FILE *f = fopen("/dev/null", "r");
1022         rte_openlog_stream(f);
1023         plog_info("=== PROX started ===\n");
1024         return 0;
1025 }
1026
1027 static int success = 0;
1028 static void siguser_handler(int signal)
1029 {
1030         if (signal == SIGUSR1)
1031                 success = 1;
1032         else
1033                 success = 0;
1034 }
1035
1036 static void sigabrt_handler(__attribute__((unused)) int signum)
1037 {
1038         /* restore default disposition for SIGABRT and SIGPIPE */
1039         signal(SIGABRT, SIG_DFL);
1040         signal(SIGPIPE, SIG_DFL);
1041
1042         /* ignore further Ctrl-C */
1043         signal(SIGINT, SIG_IGN);
1044
1045         /* more drastic exit on tedious termination signal */
1046         plog_info("Aborting...\n");
1047         if (lcore_cfg != NULL) {
1048                 uint32_t lcore_id;
1049                 pthread_t thread_id, tid0, tid = pthread_self();
1050                 memset(&tid0, 0, sizeof(tid0));
1051
1052                 /* cancel all threads except current one */
1053                 lcore_id = -1;
1054                 while (prox_core_next(&lcore_id, 1) == 0) {
1055                         thread_id = lcore_cfg[lcore_id].thread_id;
1056                         if (pthread_equal(thread_id, tid0))
1057                                 continue;
1058                         if (pthread_equal(thread_id, tid))
1059                                 continue;
1060                         pthread_cancel(thread_id);
1061                 }
1062
1063                 /* wait for cancelled threads to terminate */
1064                 lcore_id = -1;
1065                 while (prox_core_next(&lcore_id, 1) == 0) {
1066                         thread_id = lcore_cfg[lcore_id].thread_id;
1067                         if (pthread_equal(thread_id, tid0))
1068                                 continue;
1069                         if (pthread_equal(thread_id, tid))
1070                                 continue;
1071                         pthread_join(thread_id, NULL);
1072                 }
1073         }
1074
1075         /* close ncurses */
1076         display_end();
1077
1078         /* close ports on termination signal */
1079         close_ports_atexit();
1080
1081         /* terminate now */
1082         abort();
1083 }
1084
1085 static void sigterm_handler(int signum)
1086 {
1087         /* abort on second Ctrl-C */
1088         if (signum == SIGINT)
1089                 signal(SIGINT, sigabrt_handler);
1090
1091         /* gracefully quit on harmless termination signal */
1092         /* ports will subsequently get closed at resulting exit */
1093         quit();
1094 }
1095
1096 int main(int argc, char **argv)
1097 {
1098         /* set en_US locale to print big numbers with ',' */
1099         setlocale(LC_NUMERIC, "en_US.utf-8");
1100
1101         if (prox_parse_args(argc, argv) != 0){
1102                 prox_usage(argv[0]);
1103         }
1104
1105         plog_init(prox_cfg.log_name, prox_cfg.log_name_pid);
1106         plog_info("=== " PROGRAM_NAME " " VERSION_STR " ===\n");
1107         plog_info("\tUsing DPDK %s\n", rte_version() + sizeof(RTE_VER_PREFIX));
1108         read_rdt_info();
1109
1110         if (prox_cfg.flags & DSF_LIST_TASK_MODES) {
1111                 /* list supported task modes and exit */
1112                 tasks_list();
1113                 return EXIT_SUCCESS;
1114         }
1115
1116         /* close ports at normal exit */
1117         atexit(close_ports_atexit);
1118         /* gracefully quit on harmless termination signals */
1119         signal(SIGHUP, sigterm_handler);
1120         signal(SIGINT, sigterm_handler);
1121         signal(SIGQUIT, sigterm_handler);
1122         signal(SIGTERM, sigterm_handler);
1123         signal(SIGUSR1, sigterm_handler);
1124         signal(SIGUSR2, sigterm_handler);
1125         /* more drastic exit on tedious termination signals */
1126         signal(SIGABRT, sigabrt_handler);
1127         signal(SIGPIPE, sigabrt_handler);
1128
1129         if (prox_cfg.flags & DSF_DAEMON) {
1130                 signal(SIGUSR1, siguser_handler);
1131                 signal(SIGUSR2, siguser_handler);
1132                 plog_info("=== Running in Daemon mode ===\n");
1133                 plog_info("\tForking child and waiting for setup completion\n");
1134
1135                 pid_t ppid = getpid();
1136                 pid_t pid = fork();
1137                 if (pid < 0) {
1138                         plog_err("Failed to fork process to run in daemon mode\n");
1139                         return EXIT_FAILURE;
1140                 }
1141
1142                 if (pid == 0) {
1143                         fclose(stdin);
1144                         fclose(stdout);
1145                         fclose(stderr);
1146                         if (setsid() < 0) {
1147                                 kill(ppid, SIGUSR2);
1148                                 return EXIT_FAILURE;
1149                         }
1150                         if (setup_prox(argc, argv) != 0) {
1151                                 kill(ppid, SIGUSR2);
1152                                 return EXIT_FAILURE;
1153                         }
1154                         else {
1155                                 kill(ppid, SIGUSR1);
1156                                 run(prox_cfg.flags);
1157                                 return EXIT_SUCCESS;
1158                         }
1159                 }
1160                 else {
1161                         /* Before exiting the parent, wait until the
1162                            child process has finished setting up */
1163                         pause();
1164                         if (prox_cfg.logbuf) {
1165                                 file_print(prox_cfg.logbuf);
1166                         }
1167                         return success? EXIT_SUCCESS : EXIT_FAILURE;
1168                 }
1169         }
1170
1171         if (setup_prox(argc, argv) != 0)
1172                 return EXIT_FAILURE;
1173         run(prox_cfg.flags);
1174         return EXIT_SUCCESS;
1175 }