ceph/src/dpdk/lib/librte_eal/linuxapp/eal/eal.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   Copyright(c) 2012-2014 6WIND S.A.
   6  *   All rights reserved.
   7  *
   8  *   Redistribution and use in source and binary forms, with or without
   9  *   modification, are permitted provided that the following conditions
  10  *   are met:
  11  *
  12  *     * Redistributions of source code must retain the above copyright
  13  *       notice, this list of conditions and the following disclaimer.
  14  *     * Redistributions in binary form must reproduce the above copyright
  15  *       notice, this list of conditions and the following disclaimer in
  16  *       the documentation and/or other materials provided with the
  17  *       distribution.
  18  *     * Neither the name of Intel Corporation nor the names of its
  19  *       contributors may be used to endorse or promote products derived
  20  *       from this software without specific prior written permission.
  21  *
  22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <string.h>
  39 #include <stdarg.h>
  40 #include <unistd.h>
  41 #include <pthread.h>
  42 #include <syslog.h>
  43 #include <getopt.h>
  44 #include <sys/file.h>
  45 #include <fcntl.h>
  46 #include <stddef.h>
  47 #include <errno.h>
  48 #include <limits.h>
  49 #include <errno.h>
  50 #include <sys/mman.h>
  51 #include <sys/queue.h>
  52 #include <sys/stat.h>
  53 #if defined(RTE_ARCH_X86)
  54 #include <sys/io.h>
  55 #endif
  56
  57 #include <rte_common.h>
  58 #include <rte_debug.h>
  59 #include <rte_memory.h>
  60 #include <rte_memzone.h>
  61 #include <rte_launch.h>
  62 #include <rte_eal.h>
  63 #include <rte_eal_memconfig.h>
  64 #include <rte_per_lcore.h>
  65 #include <rte_lcore.h>
  66 #include <rte_log.h>
  67 #include <rte_random.h>
  68 #include <rte_cycles.h>
  69 #include <rte_string_fns.h>
  70 #include <rte_cpuflags.h>
  71 #include <rte_interrupts.h>
  72 #include <rte_pci.h>
  73 #include <rte_dev.h>
  74 #include <rte_devargs.h>
  75 #include <rte_common.h>
  76 #include <rte_version.h>
  77 #include <rte_atomic.h>
  78 #include <malloc_heap.h>
  79
  80 #include "eal_private.h"
  81 #include "eal_thread.h"
  82 #include "eal_internal_cfg.h"
  83 #include "eal_filesystem.h"
  84 #include "eal_hugepages.h"
  85 #include "eal_options.h"
  86 #include "eal_vfio.h"
  87
  88 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
  89
  90 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
  91
  92 /* Allow the application to print its usage message too if set */
  93 static rte_usage_hook_t rte_application_usage_hook = NULL;
  94
  95 /* early configuration structure, when memory config is not mmapped */
  96 static struct rte_mem_config early_mem_config;
  97
  98 /* define fd variable here, because file needs to be kept open for the
  99  * duration of the program, as we hold a write lock on it in the primary proc */
 100 static int mem_cfg_fd = -1;
 101
 102 static struct flock wr_lock = {
 103                 .l_type = F_WRLCK,
 104                 .l_whence = SEEK_SET,
 105                 .l_start = offsetof(struct rte_mem_config, memseg),
 106                 .l_len = sizeof(early_mem_config.memseg),
 107 };
 108
 109 /* Address of global and public configuration */
 110 static struct rte_config rte_config = {
 111                 .mem_config = &early_mem_config,
 112 };
 113
 114 /* internal configuration (per-core) */
 115 struct lcore_config lcore_config[RTE_MAX_LCORE];
 116
 117 /* internal configuration */
 118 struct internal_config internal_config;
 119
 120 /* used by rte_rdtsc() */
 121 int rte_cycles_vmware_tsc_map;
 122
 123 /* Return a pointer to the configuration structure */
 124 struct rte_config *
 125 rte_eal_get_configuration(void)
 126 {
 127         return &rte_config;
 128 }
 129
 130 /* parse a sysfs (or other) file containing one integer value */
 131 int
 132 eal_parse_sysfs_value(const char *filename, unsigned long *val)
 133 {
 134         FILE *f;
 135         char buf[BUFSIZ];
 136         char *end = NULL;
 137
 138         if ((f = fopen(filename, "r")) == NULL) {
 139                 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
 140                         __func__, filename);
 141                 return -1;
 142         }
 143
 144         if (fgets(buf, sizeof(buf), f) == NULL) {
 145                 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
 146                         __func__, filename);
 147                 fclose(f);
 148                 return -1;
 149         }
 150         *val = strtoul(buf, &end, 0);
 151         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 152                 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
 153                                 __func__, filename);
 154                 fclose(f);
 155                 return -1;
 156         }
 157         fclose(f);
 158         return 0;
 159 }
 160
 161
 162 /* create memory configuration in shared/mmap memory. Take out
 163  * a write lock on the memsegs, so we can auto-detect primary/secondary.
 164  * This means we never close the file while running (auto-close on exit).
 165  * We also don't lock the whole file, so that in future we can use read-locks
 166  * on other parts, e.g. memzones, to detect if there are running secondary
 167  * processes. */
 168 static void
 169 rte_eal_config_create(void)
 170 {
 171         void *rte_mem_cfg_addr;
 172         int retval;
 173
 174         const char *pathname = eal_runtime_config_path();
 175
 176         if (internal_config.no_shconf)
 177                 return;
 178
 179         /* map the config before hugepage address so that we don't waste a page */
 180         if (internal_config.base_virtaddr != 0)
 181                 rte_mem_cfg_addr = (void *)
 182                         RTE_ALIGN_FLOOR(internal_config.base_virtaddr -
 183                         sizeof(struct rte_mem_config), sysconf(_SC_PAGE_SIZE));
 184         else
 185                 rte_mem_cfg_addr = NULL;
 186
 187         if (mem_cfg_fd < 0){
 188                 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
 189                 if (mem_cfg_fd < 0)
 190                         rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
 191         }
 192
 193         retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
 194         if (retval < 0){
 195                 close(mem_cfg_fd);
 196                 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
 197         }
 198
 199         retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
 200         if (retval < 0){
 201                 close(mem_cfg_fd);
 202                 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
 203                                 "process running?\n", pathname);
 204         }
 205
 206         rte_mem_cfg_addr = mmap(rte_mem_cfg_addr, sizeof(*rte_config.mem_config),
 207                                 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
 208
 209         if (rte_mem_cfg_addr == MAP_FAILED){
 210                 rte_panic("Cannot mmap memory for rte_config\n");
 211         }
 212         memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
 213         rte_config.mem_config = (struct rte_mem_config *) rte_mem_cfg_addr;
 214
 215         /* store address of the config in the config itself so that secondary
 216          * processes could later map the config into this exact location */
 217         rte_config.mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
 218
 219 }
 220
 221 /* attach to an existing shared memory config */
 222 static void
 223 rte_eal_config_attach(void)
 224 {
 225         struct rte_mem_config *mem_config;
 226
 227         const char *pathname = eal_runtime_config_path();
 228
 229         if (internal_config.no_shconf)
 230                 return;
 231
 232         if (mem_cfg_fd < 0){
 233                 mem_cfg_fd = open(pathname, O_RDWR);
 234                 if (mem_cfg_fd < 0)
 235                         rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
 236         }
 237
 238         /* map it as read-only first */
 239         mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
 240                         PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
 241         if (mem_config == MAP_FAILED)
 242                 rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
 243                           errno, strerror(errno));
 244
 245         rte_config.mem_config = mem_config;
 246 }
 247
 248 /* reattach the shared config at exact memory location primary process has it */
 249 static void
 250 rte_eal_config_reattach(void)
 251 {
 252         struct rte_mem_config *mem_config;
 253         void *rte_mem_cfg_addr;
 254
 255         if (internal_config.no_shconf)
 256                 return;
 257
 258         /* save the address primary process has mapped shared config to */
 259         rte_mem_cfg_addr = (void *) (uintptr_t) rte_config.mem_config->mem_cfg_addr;
 260
 261         /* unmap original config */
 262         munmap(rte_config.mem_config, sizeof(struct rte_mem_config));
 263
 264         /* remap the config at proper address */
 265         mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
 266                         sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
 267                         mem_cfg_fd, 0);
 268         if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
 269                 if (mem_config != MAP_FAILED)
 270                         /* errno is stale, don't use */
 271                         rte_panic("Cannot mmap memory for rte_config at [%p], got [%p]"
 272                                   " - please use '--base-virtaddr' option\n",
 273                                   rte_mem_cfg_addr, mem_config);
 274                 else
 275                         rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
 276                                   errno, strerror(errno));
 277         }
 278         close(mem_cfg_fd);
 279
 280         rte_config.mem_config = mem_config;
 281 }
 282
 283 /* Detect if we are a primary or a secondary process */
 284 enum rte_proc_type_t
 285 eal_proc_type_detect(void)
 286 {
 287         enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
 288         const char *pathname = eal_runtime_config_path();
 289
 290         /* if we can open the file but not get a write-lock we are a secondary
 291          * process. NOTE: if we get a file handle back, we keep that open
 292          * and don't close it to prevent a race condition between multiple opens */
 293         if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
 294                         (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
 295                 ptype = RTE_PROC_SECONDARY;
 296
 297         RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
 298                         ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
 299
 300         return ptype;
 301 }
 302
 303 /* Sets up rte_config structure with the pointer to shared memory config.*/
 304 static void
 305 rte_config_init(void)
 306 {
 307         rte_config.process_type = internal_config.process_type;
 308
 309         switch (rte_config.process_type){
 310         case RTE_PROC_PRIMARY:
 311                 rte_eal_config_create();
 312                 break;
 313         case RTE_PROC_SECONDARY:
 314                 rte_eal_config_attach();
 315                 rte_eal_mcfg_wait_complete(rte_config.mem_config);
 316                 rte_eal_config_reattach();
 317                 break;
 318         case RTE_PROC_AUTO:
 319         case RTE_PROC_INVALID:
 320                 rte_panic("Invalid process type\n");
 321         }
 322 }
 323
 324 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
 325 static void
 326 eal_hugedirs_unlock(void)
 327 {
 328         int i;
 329
 330         for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
 331         {
 332                 /* skip uninitialized */
 333                 if (internal_config.hugepage_info[i].lock_descriptor < 0)
 334                         continue;
 335                 /* unlock hugepage file */
 336                 flock(internal_config.hugepage_info[i].lock_descriptor, LOCK_UN);
 337                 close(internal_config.hugepage_info[i].lock_descriptor);
 338                 /* reset the field */
 339                 internal_config.hugepage_info[i].lock_descriptor = -1;
 340         }
 341 }
 342
 343 /* display usage */
 344 static void
 345 eal_usage(const char *prgname)
 346 {
 347         printf("\nUsage: %s ", prgname);
 348         eal_common_usage();
 349         printf("EAL Linux options:\n"
 350                "  --"OPT_SOCKET_MEM"        Memory to allocate on sockets (comma separated values)\n"
 351                "  --"OPT_HUGE_DIR"          Directory where hugetlbfs is mounted\n"
 352                "  --"OPT_FILE_PREFIX"       Prefix for hugepage filenames\n"
 353                "  --"OPT_BASE_VIRTADDR"     Base virtual address\n"
 354                "  --"OPT_CREATE_UIO_DEV"    Create /dev/uioX (usually done by hotplug)\n"
 355                "  --"OPT_VFIO_INTR"         Interrupt mode for VFIO (legacy|msi|msix)\n"
 356                "  --"OPT_XEN_DOM0"          Support running on Xen dom0 without hugetlbfs\n"
 357                "\n");
 358         /* Allow the application to print its usage message too if hook is set */
 359         if ( rte_application_usage_hook ) {
 360                 printf("===== Application Usage =====\n\n");
 361                 rte_application_usage_hook(prgname);
 362         }
 363 }
 364
 365 /* Set a per-application usage message */
 366 rte_usage_hook_t
 367 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
 368 {
 369         rte_usage_hook_t        old_func;
 370
 371         /* Will be NULL on the first call to denote the last usage routine. */
 372         old_func                                        = rte_application_usage_hook;
 373         rte_application_usage_hook      = usage_func;
 374
 375         return old_func;
 376 }
 377
 378 static int
 379 eal_parse_socket_mem(char *socket_mem)
 380 {
 381         char * arg[RTE_MAX_NUMA_NODES];
 382         char *end;
 383         int arg_num, i, len;
 384         uint64_t total_mem = 0;
 385
 386         len = strnlen(socket_mem, SOCKET_MEM_STRLEN);
 387         if (len == SOCKET_MEM_STRLEN) {
 388                 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
 389                 return -1;
 390         }
 391
 392         /* all other error cases will be caught later */
 393         if (!isdigit(socket_mem[len-1]))
 394                 return -1;
 395
 396         /* split the optarg into separate socket values */
 397         arg_num = rte_strsplit(socket_mem, len,
 398                         arg, RTE_MAX_NUMA_NODES, ',');
 399
 400         /* if split failed, or 0 arguments */
 401         if (arg_num <= 0)
 402                 return -1;
 403
 404         internal_config.force_sockets = 1;
 405
 406         /* parse each defined socket option */
 407         errno = 0;
 408         for (i = 0; i < arg_num; i++) {
 409                 end = NULL;
 410                 internal_config.socket_mem[i] = strtoull(arg[i], &end, 10);
 411
 412                 /* check for invalid input */
 413                 if ((errno != 0)  ||
 414                                 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
 415                         return -1;
 416                 internal_config.socket_mem[i] *= 1024ULL;
 417                 internal_config.socket_mem[i] *= 1024ULL;
 418                 total_mem += internal_config.socket_mem[i];
 419         }
 420
 421         /* check if we have a positive amount of total memory */
 422         if (total_mem == 0)
 423                 return -1;
 424
 425         return 0;
 426 }
 427
 428 static int
 429 eal_parse_base_virtaddr(const char *arg)
 430 {
 431         char *end;
 432         uint64_t addr;
 433
 434         errno = 0;
 435         addr = strtoull(arg, &end, 16);
 436
 437         /* check for errors */
 438         if ((errno != 0) || (arg[0] == '\0') || end == NULL || (*end != '\0'))
 439                 return -1;
 440
 441         /* make sure we don't exceed 32-bit boundary on 32-bit target */
 442 #ifndef RTE_ARCH_64
 443         if (addr >= UINTPTR_MAX)
 444                 return -1;
 445 #endif
 446
 447         /* align the addr on 16M boundary, 16MB is the minimum huge page
 448          * size on IBM Power architecture. If the addr is aligned to 16MB,
 449          * it can align to 2MB for x86. So this alignment can also be used
 450          * on x86 */
 451         internal_config.base_virtaddr =
 452                 RTE_PTR_ALIGN_CEIL((uintptr_t)addr, (size_t)RTE_PGSIZE_16M);
 453
 454         return 0;
 455 }
 456
 457 static int
 458 eal_parse_vfio_intr(const char *mode)
 459 {
 460         unsigned i;
 461         static struct {
 462                 const char *name;
 463                 enum rte_intr_mode value;
 464         } map[] = {
 465                 { "legacy", RTE_INTR_MODE_LEGACY },
 466                 { "msi", RTE_INTR_MODE_MSI },
 467                 { "msix", RTE_INTR_MODE_MSIX },
 468         };
 469
 470         for (i = 0; i < RTE_DIM(map); i++) {
 471                 if (!strcmp(mode, map[i].name)) {
 472                         internal_config.vfio_intr_mode = map[i].value;
 473                         return 0;
 474                 }
 475         }
 476         return -1;
 477 }
 478
 479 /* Parse the arguments for --log-level only */
 480 static void
 481 eal_log_level_parse(int argc, char **argv)
 482 {
 483         int opt;
 484         char **argvopt;
 485         int option_index;
 486         const int old_optind = optind;
 487         const int old_optopt = optopt;
 488         char * const old_optarg = optarg;
 489
 490         argvopt = argv;
 491         optind = 1;
 492
 493         eal_reset_internal_config(&internal_config);
 494
 495         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 496                                   eal_long_options, &option_index)) != EOF) {
 497
 498                 int ret;
 499
 500                 /* getopt is not happy, stop right now */
 501                 if (opt == '?')
 502                         break;
 503
 504                 ret = (opt == OPT_LOG_LEVEL_NUM) ?
 505                         eal_parse_common_option(opt, optarg, &internal_config) : 0;
 506
 507                 /* common parser is not happy */
 508                 if (ret < 0)
 509                         break;
 510         }
 511
 512         /* restore getopt lib */
 513         optind = old_optind;
 514         optopt = old_optopt;
 515         optarg = old_optarg;
 516 }
 517
 518 /* Parse the argument given in the command line of the application */
 519 static int
 520 eal_parse_args(int argc, char **argv)
 521 {
 522         int opt, ret;
 523         char **argvopt;
 524         int option_index;
 525         char *prgname = argv[0];
 526         const int old_optind = optind;
 527         const int old_optopt = optopt;
 528         char * const old_optarg = optarg;
 529
 530         argvopt = argv;
 531         optind = 1;
 532
 533         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 534                                   eal_long_options, &option_index)) != EOF) {
 535
 536                 /* getopt is not happy, stop right now */
 537                 if (opt == '?') {
 538                         eal_usage(prgname);
 539                         ret = -1;
 540                         goto out;
 541                 }
 542
 543                 ret = eal_parse_common_option(opt, optarg, &internal_config);
 544                 /* common parser is not happy */
 545                 if (ret < 0) {
 546                         eal_usage(prgname);
 547                         ret = -1;
 548                         goto out;
 549                 }
 550                 /* common parser handled this option */
 551                 if (ret == 0)
 552                         continue;
 553
 554                 switch (opt) {
 555                 case 'h':
 556                         eal_usage(prgname);
 557                         exit(EXIT_SUCCESS);
 558
 559                 /* long options */
 560                 case OPT_XEN_DOM0_NUM:
 561 #ifdef RTE_LIBRTE_XEN_DOM0
 562                         internal_config.xen_dom0_support = 1;
 563 #else
 564                         RTE_LOG(ERR, EAL, "Can't support DPDK app "
 565                                 "running on Dom0, please configure"
 566                                 " RTE_LIBRTE_XEN_DOM0=y\n");
 567                         ret = -1;
 568                         goto out;
 569 #endif
 570                         break;
 571
 572                 case OPT_HUGE_DIR_NUM:
 573                         internal_config.hugepage_dir = optarg;
 574                         break;
 575
 576                 case OPT_FILE_PREFIX_NUM:
 577                         internal_config.hugefile_prefix = optarg;
 578                         break;
 579
 580                 case OPT_SOCKET_MEM_NUM:
 581                         if (eal_parse_socket_mem(optarg) < 0) {
 582                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 583                                                 OPT_SOCKET_MEM "\n");
 584                                 eal_usage(prgname);
 585                                 ret = -1;
 586                                 goto out;
 587                         }
 588                         break;
 589
 590                 case OPT_BASE_VIRTADDR_NUM:
 591                         if (eal_parse_base_virtaddr(optarg) < 0) {
 592                                 RTE_LOG(ERR, EAL, "invalid parameter for --"
 593                                                 OPT_BASE_VIRTADDR "\n");
 594                                 eal_usage(prgname);
 595                                 ret = -1;
 596                                 goto out;
 597                         }
 598                         break;
 599
 600                 case OPT_VFIO_INTR_NUM:
 601                         if (eal_parse_vfio_intr(optarg) < 0) {
 602                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 603                                                 OPT_VFIO_INTR "\n");
 604                                 eal_usage(prgname);
 605                                 ret = -1;
 606                                 goto out;
 607                         }
 608                         break;
 609
 610                 case OPT_CREATE_UIO_DEV_NUM:
 611                         internal_config.create_uio_dev = 1;
 612                         break;
 613
 614                 default:
 615                         if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
 616                                 RTE_LOG(ERR, EAL, "Option %c is not supported "
 617                                         "on Linux\n", opt);
 618                         } else if (opt >= OPT_LONG_MIN_NUM &&
 619                                    opt < OPT_LONG_MAX_NUM) {
 620                                 RTE_LOG(ERR, EAL, "Option %s is not supported "
 621                                         "on Linux\n",
 622                                         eal_long_options[option_index].name);
 623                         } else {
 624                                 RTE_LOG(ERR, EAL, "Option %d is not supported "
 625                                         "on Linux\n", opt);
 626                         }
 627                         eal_usage(prgname);
 628                         ret = -1;
 629                         goto out;
 630                 }
 631         }
 632
 633         if (eal_adjust_config(&internal_config) != 0) {
 634                 ret = -1;
 635                 goto out;
 636         }
 637
 638         /* sanity checks */
 639         if (eal_check_common_options(&internal_config) != 0) {
 640                 eal_usage(prgname);
 641                 ret = -1;
 642                 goto out;
 643         }
 644
 645         /* --xen-dom0 doesn't make sense with --socket-mem */
 646         if (internal_config.xen_dom0_support && internal_config.force_sockets == 1) {
 647                 RTE_LOG(ERR, EAL, "Options --"OPT_SOCKET_MEM" cannot be specified "
 648                         "together with --"OPT_XEN_DOM0"\n");
 649                 eal_usage(prgname);
 650                 ret = -1;
 651                 goto out;
 652         }
 653
 654         if (optind >= 0)
 655                 argv[optind-1] = prgname;
 656         ret = optind-1;
 657
 658 out:
 659         /* restore getopt lib */
 660         optind = old_optind;
 661         optopt = old_optopt;
 662         optarg = old_optarg;
 663
 664         return ret;
 665 }
 666
 667 static void
 668 eal_check_mem_on_local_socket(void)
 669 {
 670         const struct rte_memseg *ms;
 671         int i, socket_id;
 672
 673         socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
 674
 675         ms = rte_eal_get_physmem_layout();
 676
 677         for (i = 0; i < RTE_MAX_MEMSEG; i++)
 678                 if (ms[i].socket_id == socket_id &&
 679                                 ms[i].len > 0)
 680                         return;
 681
 682         RTE_LOG(WARNING, EAL, "WARNING: Master core has no "
 683                         "memory on local socket!\n");
 684 }
 685
 686 static int
 687 sync_func(__attribute__((unused)) void *arg)
 688 {
 689         return 0;
 690 }
 691
 692 inline static void
 693 rte_eal_mcfg_complete(void)
 694 {
 695         /* ALL shared mem_config related INIT DONE */
 696         if (rte_config.process_type == RTE_PROC_PRIMARY)
 697                 rte_config.mem_config->magic = RTE_MAGIC;
 698 }
 699
 700 /*
 701  * Request iopl privilege for all RPL, returns 0 on success
 702  * iopl() call is mostly for the i386 architecture. For other architectures,
 703  * return -1 to indicate IO privilege can't be changed in this way.
 704  */
 705 int
 706 rte_eal_iopl_init(void)
 707 {
 708 #if defined(RTE_ARCH_X86)
 709         if (iopl(3) != 0)
 710                 return -1;
 711 #endif
 712         return 0;
 713 }
 714
 715 #ifdef VFIO_PRESENT
 716 static int rte_eal_vfio_setup(void)
 717 {
 718         int vfio_enabled = 0;
 719
 720         if (!internal_config.no_pci) {
 721                 pci_vfio_enable();
 722                 vfio_enabled |= pci_vfio_is_enabled();
 723         }
 724
 725         if (vfio_enabled) {
 726
 727                 /* if we are primary process, create a thread to communicate with
 728                  * secondary processes. the thread will use a socket to wait for
 729                  * requests from secondary process to send open file descriptors,
 730                  * because VFIO does not allow multiple open descriptors on a group or
 731                  * VFIO container.
 732                  */
 733                 if (internal_config.process_type == RTE_PROC_PRIMARY &&
 734                                 vfio_mp_sync_setup() < 0)
 735                         return -1;
 736         }
 737
 738         return 0;
 739 }
 740 #endif
 741
 742 /* Launch threads, called at application init(). */
 743 int
 744 rte_eal_init(int argc, char **argv)
 745 {
 746         int i, fctret, ret;
 747         pthread_t thread_id;
 748         static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
 749         const char *logid;
 750         char cpuset[RTE_CPU_AFFINITY_STR_LEN];
 751         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 752
 753         /* checks if the machine is adequate */
 754         rte_cpu_check_supported();
 755
 756         if (!rte_atomic32_test_and_set(&run_once))
 757                 return -1;
 758
 759         logid = strrchr(argv[0], '/');
 760         logid = strdup(logid ? logid + 1: argv[0]);
 761
 762         thread_id = pthread_self();
 763
 764         eal_log_level_parse(argc, argv);
 765
 766         /* set log level as early as possible */
 767         rte_set_log_level(internal_config.log_level);
 768
 769         if (rte_eal_cpu_init() < 0)
 770                 rte_panic("Cannot detect lcores\n");
 771
 772         fctret = eal_parse_args(argc, argv);
 773         if (fctret < 0)
 774                 exit(1);
 775
 776         if (internal_config.no_hugetlbfs == 0 &&
 777                         internal_config.process_type != RTE_PROC_SECONDARY &&
 778                         internal_config.xen_dom0_support == 0 &&
 779                         eal_hugepage_info_init() < 0)
 780                 rte_panic("Cannot get hugepage information\n");
 781
 782         if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
 783                 if (internal_config.no_hugetlbfs)
 784                         internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
 785         }
 786
 787         if (internal_config.vmware_tsc_map == 1) {
 788 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
 789                 rte_cycles_vmware_tsc_map = 1;
 790                 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
 791                                 "you must have monitor_control.pseudo_perfctr = TRUE\n");
 792 #else
 793                 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
 794                                 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
 795 #endif
 796         }
 797
 798         rte_srand(rte_rdtsc());
 799
 800         rte_config_init();
 801
 802         if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0)
 803                 rte_panic("Cannot init logs\n");
 804
 805         if (rte_eal_pci_init() < 0)
 806                 rte_panic("Cannot init PCI\n");
 807
 808 #ifdef VFIO_PRESENT
 809         if (rte_eal_vfio_setup() < 0)
 810                 rte_panic("Cannot init VFIO\n");
 811 #endif
 812
 813         if (rte_eal_memory_init() < 0)
 814                 rte_panic("Cannot init memory\n");
 815
 816         /* the directories are locked during eal_hugepage_info_init */
 817         eal_hugedirs_unlock();
 818
 819         if (rte_eal_memzone_init() < 0)
 820                 rte_panic("Cannot init memzone\n");
 821
 822         if (rte_eal_tailqs_init() < 0)
 823                 rte_panic("Cannot init tail queues for objects\n");
 824
 825         if (rte_eal_alarm_init() < 0)
 826                 rte_panic("Cannot init interrupt-handling thread\n");
 827
 828         if (rte_eal_timer_init() < 0)
 829                 rte_panic("Cannot init HPET or TSC timers\n");
 830
 831         eal_check_mem_on_local_socket();
 832
 833         if (eal_plugins_init() < 0)
 834                 rte_panic("Cannot init plugins\n");
 835
 836         eal_thread_init_master(rte_config.master_lcore);
 837
 838         ret = eal_thread_dump_affinity(cpuset, RTE_CPU_AFFINITY_STR_LEN);
 839
 840         RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%x;cpuset=[%s%s])\n",
 841                 rte_config.master_lcore, (int)thread_id, cpuset,
 842                 ret == 0 ? "" : "...");
 843
 844         if (rte_eal_dev_init() < 0)
 845                 rte_panic("Cannot init pmd devices\n");
 846
 847         if (rte_eal_intr_init() < 0)
 848                 rte_panic("Cannot init interrupt-handling thread\n");
 849
 850         RTE_LCORE_FOREACH_SLAVE(i) {
 851
 852                 /*
 853                  * create communication pipes between master thread
 854                  * and children
 855                  */
 856                 if (pipe(lcore_config[i].pipe_master2slave) < 0)
 857                         rte_panic("Cannot create pipe\n");
 858                 if (pipe(lcore_config[i].pipe_slave2master) < 0)
 859                         rte_panic("Cannot create pipe\n");
 860
 861                 lcore_config[i].state = WAIT;
 862
 863                 /* create a thread for each lcore */
 864                 ret = pthread_create(&lcore_config[i].thread_id, NULL,
 865                                      eal_thread_loop, NULL);
 866                 if (ret != 0)
 867                         rte_panic("Cannot create thread\n");
 868
 869                 /* Set thread_name for aid in debugging. */
 870                 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
 871                         "lcore-slave-%d", i);
 872                 ret = rte_thread_setname(lcore_config[i].thread_id,
 873                                                 thread_name);
 874                 if (ret != 0)
 875                         RTE_LOG(DEBUG, EAL,
 876                                 "Cannot set name for lcore thread\n");
 877         }
 878
 879         /*
 880          * Launch a dummy function on all slave lcores, so that master lcore
 881          * knows they are all ready when this function returns.
 882          */
 883         rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
 884         rte_eal_mp_wait_lcore();
 885
 886         /* Probe & Initialize PCI devices */
 887         if (rte_eal_pci_probe())
 888                 rte_panic("Cannot probe PCI\n");
 889
 890         rte_eal_mcfg_complete();
 891
 892         return fctret;
 893 }
 894
 895 /* get core role */
 896 enum rte_lcore_role_t
 897 rte_eal_lcore_role(unsigned lcore_id)
 898 {
 899         return rte_config.lcore_role[lcore_id];
 900 }
 901
 902 enum rte_proc_type_t
 903 rte_eal_process_type(void)
 904 {
 905         return rte_config.process_type;
 906 }
 907
 908 int rte_eal_has_hugepages(void)
 909 {
 910         return ! internal_config.no_hugetlbfs;
 911 }
 912
 913 int
 914 rte_eal_check_module(const char *module_name)
 915 {
 916         char sysfs_mod_name[PATH_MAX];
 917         struct stat st;
 918         int n;
 919
 920         if (NULL == module_name)
 921                 return -1;
 922
 923         /* Check if there is sysfs mounted */
 924         if (stat("/sys/module", &st) != 0) {
 925                 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
 926                         errno, strerror(errno));
 927                 return -1;
 928         }
 929
 930         /* A module might be built-in, therefore try sysfs */
 931         n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
 932         if (n < 0 || n > PATH_MAX) {
 933                 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
 934                 return -1;
 935         }
 936
 937         if (stat(sysfs_mod_name, &st) != 0) {
 938                 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
 939                         sysfs_mod_name, errno, strerror(errno));
 940                 return 0;
 941         }
 942
 943         /* Module has been found */
 944         return 1;
 945 }