lib/librte_eal/linux/eal.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation.
   3  * Copyright(c) 2012-2014 6WIND S.A.
   4  */
   5
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <stdint.h>
   9 #include <string.h>
  10 #include <stdarg.h>
  11 #include <unistd.h>
  12 #include <pthread.h>
  13 #include <syslog.h>
  14 #include <getopt.h>
  15 #include <sys/file.h>
  16 #include <dirent.h>
  17 #include <fcntl.h>
  18 #include <fnmatch.h>
  19 #include <stddef.h>
  20 #include <errno.h>
  21 #include <limits.h>
  22 #include <sys/mman.h>
  23 #include <sys/queue.h>
  24 #include <sys/stat.h>
  25 #if defined(RTE_ARCH_X86)
  26 #include <sys/io.h>
  27 #endif
  28 #include <linux/version.h>
  29
  30 #include <rte_compat.h>
  31 #include <rte_common.h>
  32 #include <rte_debug.h>
  33 #include <rte_memory.h>
  34 #include <rte_launch.h>
  35 #include <rte_eal.h>
  36 #include <rte_errno.h>
  37 #include <rte_per_lcore.h>
  38 #include <rte_lcore.h>
  39 #include <rte_service_component.h>
  40 #include <rte_log.h>
  41 #include <rte_random.h>
  42 #include <rte_cycles.h>
  43 #include <rte_string_fns.h>
  44 #include <rte_cpuflags.h>
  45 #include <rte_interrupts.h>
  46 #include <rte_bus.h>
  47 #include <rte_dev.h>
  48 #include <rte_devargs.h>
  49 #include <rte_version.h>
  50 #include <rte_atomic.h>
  51 #include <malloc_heap.h>
  52 #include <rte_vfio.h>
  53 #include <rte_telemetry.h>
  54
  55 #include "eal_private.h"
  56 #include "eal_thread.h"
  57 #include "eal_internal_cfg.h"
  58 #include "eal_filesystem.h"
  59 #include "eal_hugepages.h"
  60 #include "eal_memcfg.h"
  61 #include "eal_trace.h"
  62 #include "eal_options.h"
  63 #include "eal_vfio.h"
  64 #include "hotplug_mp.h"
  65
  66 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
  67
  68 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
  69
  70 #define KERNEL_IOMMU_GROUPS_PATH "/sys/kernel/iommu_groups"
  71
  72 /* define fd variable here, because file needs to be kept open for the
  73  * duration of the program, as we hold a write lock on it in the primary proc */
  74 static int mem_cfg_fd = -1;
  75
  76 static struct flock wr_lock = {
  77                 .l_type = F_WRLCK,
  78                 .l_whence = SEEK_SET,
  79                 .l_start = offsetof(struct rte_mem_config, memsegs),
  80                 .l_len = RTE_SIZEOF_FIELD(struct rte_mem_config, memsegs),
  81 };
  82
  83 /* internal configuration (per-core) */
  84 struct lcore_config lcore_config[RTE_MAX_LCORE];
  85
  86 /* used by rte_rdtsc() */
  87 int rte_cycles_vmware_tsc_map;
  88
  89 static const char *default_runtime_dir = "/var/run";
  90
  91 int
  92 eal_create_runtime_dir(void)
  93 {
  94         const char *directory = default_runtime_dir;
  95         const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
  96         const char *fallback = "/tmp";
  97         char run_dir[PATH_MAX];
  98         char tmp[PATH_MAX];
  99         int ret;
 100
 101         if (getuid() != 0) {
 102                 /* try XDG path first, fall back to /tmp */
 103                 if (xdg_runtime_dir != NULL)
 104                         directory = xdg_runtime_dir;
 105                 else
 106                         directory = fallback;
 107         }
 108         /* create DPDK subdirectory under runtime dir */
 109         ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
 110         if (ret < 0 || ret == sizeof(tmp)) {
 111                 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
 112                 return -1;
 113         }
 114
 115         /* create prefix-specific subdirectory under DPDK runtime dir */
 116         ret = snprintf(run_dir, sizeof(run_dir), "%s/%s",
 117                         tmp, eal_get_hugefile_prefix());
 118         if (ret < 0 || ret == sizeof(run_dir)) {
 119                 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
 120                 return -1;
 121         }
 122
 123         /* create the path if it doesn't exist. no "mkdir -p" here, so do it
 124          * step by step.
 125          */
 126         ret = mkdir(tmp, 0700);
 127         if (ret < 0 && errno != EEXIST) {
 128                 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
 129                         tmp, strerror(errno));
 130                 return -1;
 131         }
 132
 133         ret = mkdir(run_dir, 0700);
 134         if (ret < 0 && errno != EEXIST) {
 135                 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
 136                         run_dir, strerror(errno));
 137                 return -1;
 138         }
 139
 140         if (eal_set_runtime_dir(run_dir, sizeof(run_dir)))
 141                 return -1;
 142
 143         return 0;
 144 }
 145
 146 int
 147 eal_clean_runtime_dir(void)
 148 {
 149         const char *runtime_dir = rte_eal_get_runtime_dir();
 150         DIR *dir;
 151         struct dirent *dirent;
 152         int dir_fd, fd, lck_result;
 153         static const char * const filters[] = {
 154                 "fbarray_*",
 155                 "mp_socket_*"
 156         };
 157
 158         /* open directory */
 159         dir = opendir(runtime_dir);
 160         if (!dir) {
 161                 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
 162                                 runtime_dir);
 163                 goto error;
 164         }
 165         dir_fd = dirfd(dir);
 166
 167         /* lock the directory before doing anything, to avoid races */
 168         if (flock(dir_fd, LOCK_EX) < 0) {
 169                 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
 170                         runtime_dir);
 171                 goto error;
 172         }
 173
 174         dirent = readdir(dir);
 175         if (!dirent) {
 176                 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
 177                                 runtime_dir);
 178                 goto error;
 179         }
 180
 181         while (dirent != NULL) {
 182                 unsigned int f_idx;
 183                 bool skip = true;
 184
 185                 /* skip files that don't match the patterns */
 186                 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
 187                         const char *filter = filters[f_idx];
 188
 189                         if (fnmatch(filter, dirent->d_name, 0) == 0) {
 190                                 skip = false;
 191                                 break;
 192                         }
 193                 }
 194                 if (skip) {
 195                         dirent = readdir(dir);
 196                         continue;
 197                 }
 198
 199                 /* try and lock the file */
 200                 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
 201
 202                 /* skip to next file */
 203                 if (fd == -1) {
 204                         dirent = readdir(dir);
 205                         continue;
 206                 }
 207
 208                 /* non-blocking lock */
 209                 lck_result = flock(fd, LOCK_EX | LOCK_NB);
 210
 211                 /* if lock succeeds, remove the file */
 212                 if (lck_result != -1)
 213                         unlinkat(dir_fd, dirent->d_name, 0);
 214                 close(fd);
 215                 dirent = readdir(dir);
 216         }
 217
 218         /* closedir closes dir_fd and drops the lock */
 219         closedir(dir);
 220         return 0;
 221
 222 error:
 223         if (dir)
 224                 closedir(dir);
 225
 226         RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
 227                 strerror(errno));
 228
 229         return -1;
 230 }
 231
 232 /* parse a sysfs (or other) file containing one integer value */
 233 int
 234 eal_parse_sysfs_value(const char *filename, unsigned long *val)
 235 {
 236         FILE *f;
 237         char buf[BUFSIZ];
 238         char *end = NULL;
 239
 240         if ((f = fopen(filename, "r")) == NULL) {
 241                 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
 242                         __func__, filename);
 243                 return -1;
 244         }
 245
 246         if (fgets(buf, sizeof(buf), f) == NULL) {
 247                 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
 248                         __func__, filename);
 249                 fclose(f);
 250                 return -1;
 251         }
 252         *val = strtoul(buf, &end, 0);
 253         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 254                 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
 255                                 __func__, filename);
 256                 fclose(f);
 257                 return -1;
 258         }
 259         fclose(f);
 260         return 0;
 261 }
 262
 263
 264 /* create memory configuration in shared/mmap memory. Take out
 265  * a write lock on the memsegs, so we can auto-detect primary/secondary.
 266  * This means we never close the file while running (auto-close on exit).
 267  * We also don't lock the whole file, so that in future we can use read-locks
 268  * on other parts, e.g. memzones, to detect if there are running secondary
 269  * processes. */
 270 static int
 271 rte_eal_config_create(void)
 272 {
 273         struct rte_config *config = rte_eal_get_configuration();
 274         size_t page_sz = sysconf(_SC_PAGE_SIZE);
 275         size_t cfg_len = sizeof(*config->mem_config);
 276         size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
 277         void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
 278         int retval;
 279         const struct internal_config *internal_conf =
 280                 eal_get_internal_configuration();
 281
 282         const char *pathname = eal_runtime_config_path();
 283
 284         if (internal_conf->no_shconf)
 285                 return 0;
 286
 287         /* map the config before hugepage address so that we don't waste a page */
 288         if (internal_conf->base_virtaddr != 0)
 289                 rte_mem_cfg_addr = (void *)
 290                         RTE_ALIGN_FLOOR(internal_conf->base_virtaddr -
 291                         sizeof(struct rte_mem_config), page_sz);
 292         else
 293                 rte_mem_cfg_addr = NULL;
 294
 295         if (mem_cfg_fd < 0){
 296                 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
 297                 if (mem_cfg_fd < 0) {
 298                         RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
 299                                 pathname);
 300                         return -1;
 301                 }
 302         }
 303
 304         retval = ftruncate(mem_cfg_fd, cfg_len);
 305         if (retval < 0){
 306                 close(mem_cfg_fd);
 307                 mem_cfg_fd = -1;
 308                 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
 309                         pathname);
 310                 return -1;
 311         }
 312
 313         retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
 314         if (retval < 0){
 315                 close(mem_cfg_fd);
 316                 mem_cfg_fd = -1;
 317                 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
 318                         "process running?\n", pathname);
 319                 return -1;
 320         }
 321
 322         /* reserve space for config */
 323         rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
 324                         &cfg_len_aligned, page_sz, 0, 0);
 325         if (rte_mem_cfg_addr == NULL) {
 326                 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
 327                 close(mem_cfg_fd);
 328                 mem_cfg_fd = -1;
 329                 return -1;
 330         }
 331
 332         /* remap the actual file into the space we've just reserved */
 333         mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
 334                         cfg_len_aligned, PROT_READ | PROT_WRITE,
 335                         MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
 336         if (mapped_mem_cfg_addr == MAP_FAILED) {
 337                 munmap(rte_mem_cfg_addr, cfg_len);
 338                 close(mem_cfg_fd);
 339                 mem_cfg_fd = -1;
 340                 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
 341                 return -1;
 342         }
 343
 344         memcpy(rte_mem_cfg_addr, config->mem_config, sizeof(struct rte_mem_config));
 345         config->mem_config = rte_mem_cfg_addr;
 346
 347         /* store address of the config in the config itself so that secondary
 348          * processes could later map the config into this exact location
 349          */
 350         config->mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
 351         config->mem_config->dma_maskbits = 0;
 352
 353         return 0;
 354 }
 355
 356 /* attach to an existing shared memory config */
 357 static int
 358 rte_eal_config_attach(void)
 359 {
 360         struct rte_config *config = rte_eal_get_configuration();
 361         struct rte_mem_config *mem_config;
 362         const struct internal_config *internal_conf =
 363                 eal_get_internal_configuration();
 364
 365         const char *pathname = eal_runtime_config_path();
 366
 367         if (internal_conf->no_shconf)
 368                 return 0;
 369
 370         if (mem_cfg_fd < 0){
 371                 mem_cfg_fd = open(pathname, O_RDWR);
 372                 if (mem_cfg_fd < 0) {
 373                         RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
 374                                 pathname);
 375                         return -1;
 376                 }
 377         }
 378
 379         /* map it as read-only first */
 380         mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
 381                         PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
 382         if (mem_config == MAP_FAILED) {
 383                 close(mem_cfg_fd);
 384                 mem_cfg_fd = -1;
 385                 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
 386                         errno, strerror(errno));
 387                 return -1;
 388         }
 389
 390         config->mem_config = mem_config;
 391
 392         return 0;
 393 }
 394
 395 /* reattach the shared config at exact memory location primary process has it */
 396 static int
 397 rte_eal_config_reattach(void)
 398 {
 399         struct rte_config *config = rte_eal_get_configuration();
 400         struct rte_mem_config *mem_config;
 401         void *rte_mem_cfg_addr;
 402         const struct internal_config *internal_conf =
 403                 eal_get_internal_configuration();
 404
 405         if (internal_conf->no_shconf)
 406                 return 0;
 407
 408         /* save the address primary process has mapped shared config to */
 409         rte_mem_cfg_addr =
 410                 (void *) (uintptr_t) config->mem_config->mem_cfg_addr;
 411
 412         /* unmap original config */
 413         munmap(config->mem_config, sizeof(struct rte_mem_config));
 414
 415         /* remap the config at proper address */
 416         mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
 417                         sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
 418                         mem_cfg_fd, 0);
 419
 420         close(mem_cfg_fd);
 421         mem_cfg_fd = -1;
 422
 423         if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
 424                 if (mem_config != MAP_FAILED) {
 425                         /* errno is stale, don't use */
 426                         RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
 427                                 " - please use '--" OPT_BASE_VIRTADDR
 428                                 "' option\n", rte_mem_cfg_addr, mem_config);
 429                         munmap(mem_config, sizeof(struct rte_mem_config));
 430                         return -1;
 431                 }
 432                 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
 433                         errno, strerror(errno));
 434                 return -1;
 435         }
 436
 437         config->mem_config = mem_config;
 438
 439         return 0;
 440 }
 441
 442 /* Detect if we are a primary or a secondary process */
 443 enum rte_proc_type_t
 444 eal_proc_type_detect(void)
 445 {
 446         enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
 447         const char *pathname = eal_runtime_config_path();
 448         const struct internal_config *internal_conf =
 449                 eal_get_internal_configuration();
 450
 451         /* if there no shared config, there can be no secondary processes */
 452         if (!internal_conf->no_shconf) {
 453                 /* if we can open the file but not get a write-lock we are a
 454                  * secondary process. NOTE: if we get a file handle back, we
 455                  * keep that open and don't close it to prevent a race condition
 456                  * between multiple opens.
 457                  */
 458                 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
 459                                 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
 460                         ptype = RTE_PROC_SECONDARY;
 461         }
 462
 463         RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
 464                         ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
 465
 466         return ptype;
 467 }
 468
 469 /* Sets up rte_config structure with the pointer to shared memory config.*/
 470 static int
 471 rte_config_init(void)
 472 {
 473         struct rte_config *config = rte_eal_get_configuration();
 474         const struct internal_config *internal_conf =
 475                 eal_get_internal_configuration();
 476
 477         config->process_type = internal_conf->process_type;
 478
 479         switch (config->process_type) {
 480         case RTE_PROC_PRIMARY:
 481                 if (rte_eal_config_create() < 0)
 482                         return -1;
 483                 eal_mcfg_update_from_internal();
 484                 break;
 485         case RTE_PROC_SECONDARY:
 486                 if (rte_eal_config_attach() < 0)
 487                         return -1;
 488                 eal_mcfg_wait_complete();
 489                 if (eal_mcfg_check_version() < 0) {
 490                         RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
 491                         return -1;
 492                 }
 493                 if (rte_eal_config_reattach() < 0)
 494                         return -1;
 495                 if (!__rte_mp_enable()) {
 496                         RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
 497                         return -1;
 498                 }
 499                 eal_mcfg_update_internal();
 500                 break;
 501         case RTE_PROC_AUTO:
 502         case RTE_PROC_INVALID:
 503                 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
 504                         config->process_type);
 505                 return -1;
 506         }
 507
 508         return 0;
 509 }
 510
 511 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
 512 static void
 513 eal_hugedirs_unlock(void)
 514 {
 515         int i;
 516         struct internal_config *internal_conf =
 517                 eal_get_internal_configuration();
 518
 519         for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
 520         {
 521                 /* skip uninitialized */
 522                 if (internal_conf->hugepage_info[i].lock_descriptor < 0)
 523                         continue;
 524                 /* unlock hugepage file */
 525                 flock(internal_conf->hugepage_info[i].lock_descriptor, LOCK_UN);
 526                 close(internal_conf->hugepage_info[i].lock_descriptor);
 527                 /* reset the field */
 528                 internal_conf->hugepage_info[i].lock_descriptor = -1;
 529         }
 530 }
 531
 532 /* display usage */
 533 static void
 534 eal_usage(const char *prgname)
 535 {
 536         rte_usage_hook_t hook = eal_get_application_usage_hook();
 537
 538         printf("\nUsage: %s ", prgname);
 539         eal_common_usage();
 540         printf("EAL Linux options:\n"
 541                "  --"OPT_SOCKET_MEM"        Memory to allocate on sockets (comma separated values)\n"
 542                "  --"OPT_SOCKET_LIMIT"      Limit memory allocation on sockets (comma separated values)\n"
 543                "  --"OPT_HUGE_DIR"          Directory where hugetlbfs is mounted\n"
 544                "  --"OPT_FILE_PREFIX"       Prefix for hugepage filenames\n"
 545                "  --"OPT_CREATE_UIO_DEV"    Create /dev/uioX (usually done by hotplug)\n"
 546                "  --"OPT_VFIO_INTR"         Interrupt mode for VFIO (legacy|msi|msix)\n"
 547                "  --"OPT_VFIO_VF_TOKEN"     VF token (UUID) shared between SR-IOV PF and VFs\n"
 548                "  --"OPT_LEGACY_MEM"        Legacy memory mode (no dynamic allocation, contiguous segments)\n"
 549                "  --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
 550                "  --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
 551                "\n");
 552         /* Allow the application to print its usage message too if hook is set */
 553         if (hook) {
 554                 printf("===== Application Usage =====\n\n");
 555                 (hook)(prgname);
 556         }
 557 }
 558
 559 static int
 560 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
 561 {
 562         char * arg[RTE_MAX_NUMA_NODES];
 563         char *end;
 564         int arg_num, i, len;
 565         uint64_t total_mem = 0;
 566
 567         len = strnlen(strval, SOCKET_MEM_STRLEN);
 568         if (len == SOCKET_MEM_STRLEN) {
 569                 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
 570                 return -1;
 571         }
 572
 573         /* all other error cases will be caught later */
 574         if (!isdigit(strval[len-1]))
 575                 return -1;
 576
 577         /* split the optarg into separate socket values */
 578         arg_num = rte_strsplit(strval, len,
 579                         arg, RTE_MAX_NUMA_NODES, ',');
 580
 581         /* if split failed, or 0 arguments */
 582         if (arg_num <= 0)
 583                 return -1;
 584
 585         /* parse each defined socket option */
 586         errno = 0;
 587         for (i = 0; i < arg_num; i++) {
 588                 uint64_t val;
 589                 end = NULL;
 590                 val = strtoull(arg[i], &end, 10);
 591
 592                 /* check for invalid input */
 593                 if ((errno != 0)  ||
 594                                 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
 595                         return -1;
 596                 val <<= 20;
 597                 total_mem += val;
 598                 socket_arg[i] = val;
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int
 605 eal_parse_vfio_intr(const char *mode)
 606 {
 607         struct internal_config *internal_conf =
 608                 eal_get_internal_configuration();
 609         unsigned i;
 610         static struct {
 611                 const char *name;
 612                 enum rte_intr_mode value;
 613         } map[] = {
 614                 { "legacy", RTE_INTR_MODE_LEGACY },
 615                 { "msi", RTE_INTR_MODE_MSI },
 616                 { "msix", RTE_INTR_MODE_MSIX },
 617         };
 618
 619         for (i = 0; i < RTE_DIM(map); i++) {
 620                 if (!strcmp(mode, map[i].name)) {
 621                         internal_conf->vfio_intr_mode = map[i].value;
 622                         return 0;
 623                 }
 624         }
 625         return -1;
 626 }
 627
 628 static int
 629 eal_parse_vfio_vf_token(const char *vf_token)
 630 {
 631         struct internal_config *cfg = eal_get_internal_configuration();
 632         rte_uuid_t uuid;
 633
 634         if (!rte_uuid_parse(vf_token, uuid)) {
 635                 rte_uuid_copy(cfg->vfio_vf_token, uuid);
 636                 return 0;
 637         }
 638
 639         return -1;
 640 }
 641
 642 /* Parse the arguments for --log-level only */
 643 static void
 644 eal_log_level_parse(int argc, char **argv)
 645 {
 646         int opt;
 647         char **argvopt;
 648         int option_index;
 649         const int old_optind = optind;
 650         const int old_optopt = optopt;
 651         char * const old_optarg = optarg;
 652         struct internal_config *internal_conf =
 653                 eal_get_internal_configuration();
 654
 655         argvopt = argv;
 656         optind = 1;
 657
 658         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 659                                   eal_long_options, &option_index)) != EOF) {
 660
 661                 int ret;
 662
 663                 /* getopt is not happy, stop right now */
 664                 if (opt == '?')
 665                         break;
 666
 667                 ret = (opt == OPT_LOG_LEVEL_NUM) ?
 668                         eal_parse_common_option(opt, optarg, internal_conf) : 0;
 669
 670                 /* common parser is not happy */
 671                 if (ret < 0)
 672                         break;
 673         }
 674
 675         /* restore getopt lib */
 676         optind = old_optind;
 677         optopt = old_optopt;
 678         optarg = old_optarg;
 679 }
 680
 681 /* Parse the argument given in the command line of the application */
 682 static int
 683 eal_parse_args(int argc, char **argv)
 684 {
 685         int opt, ret;
 686         char **argvopt;
 687         int option_index;
 688         char *prgname = argv[0];
 689         const int old_optind = optind;
 690         const int old_optopt = optopt;
 691         char * const old_optarg = optarg;
 692         struct internal_config *internal_conf =
 693                 eal_get_internal_configuration();
 694
 695         argvopt = argv;
 696         optind = 1;
 697
 698         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 699                                   eal_long_options, &option_index)) != EOF) {
 700
 701                 /* getopt didn't recognise the option */
 702                 if (opt == '?') {
 703                         eal_usage(prgname);
 704                         ret = -1;
 705                         goto out;
 706                 }
 707
 708                 ret = eal_parse_common_option(opt, optarg, internal_conf);
 709                 /* common parser is not happy */
 710                 if (ret < 0) {
 711                         eal_usage(prgname);
 712                         ret = -1;
 713                         goto out;
 714                 }
 715                 /* common parser handled this option */
 716                 if (ret == 0)
 717                         continue;
 718
 719                 switch (opt) {
 720                 case 'h':
 721                         eal_usage(prgname);
 722                         exit(EXIT_SUCCESS);
 723
 724                 case OPT_HUGE_DIR_NUM:
 725                 {
 726                         char *hdir = strdup(optarg);
 727                         if (hdir == NULL)
 728                                 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
 729                         else {
 730                                 /* free old hugepage dir */
 731                                 if (internal_conf->hugepage_dir != NULL)
 732                                         free(internal_conf->hugepage_dir);
 733                                 internal_conf->hugepage_dir = hdir;
 734                         }
 735                         break;
 736                 }
 737                 case OPT_FILE_PREFIX_NUM:
 738                 {
 739                         char *prefix = strdup(optarg);
 740                         if (prefix == NULL)
 741                                 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
 742                         else {
 743                                 /* free old prefix */
 744                                 if (internal_conf->hugefile_prefix != NULL)
 745                                         free(internal_conf->hugefile_prefix);
 746                                 internal_conf->hugefile_prefix = prefix;
 747                         }
 748                         break;
 749                 }
 750                 case OPT_SOCKET_MEM_NUM:
 751                         if (eal_parse_socket_arg(optarg,
 752                                         internal_conf->socket_mem) < 0) {
 753                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 754                                                 OPT_SOCKET_MEM "\n");
 755                                 eal_usage(prgname);
 756                                 ret = -1;
 757                                 goto out;
 758                         }
 759                         internal_conf->force_sockets = 1;
 760                         break;
 761
 762                 case OPT_SOCKET_LIMIT_NUM:
 763                         if (eal_parse_socket_arg(optarg,
 764                                         internal_conf->socket_limit) < 0) {
 765                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 766                                                 OPT_SOCKET_LIMIT "\n");
 767                                 eal_usage(prgname);
 768                                 ret = -1;
 769                                 goto out;
 770                         }
 771                         internal_conf->force_socket_limits = 1;
 772                         break;
 773
 774                 case OPT_VFIO_INTR_NUM:
 775                         if (eal_parse_vfio_intr(optarg) < 0) {
 776                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 777                                                 OPT_VFIO_INTR "\n");
 778                                 eal_usage(prgname);
 779                                 ret = -1;
 780                                 goto out;
 781                         }
 782                         break;
 783
 784                 case OPT_VFIO_VF_TOKEN_NUM:
 785                         if (eal_parse_vfio_vf_token(optarg) < 0) {
 786                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 787                                                 OPT_VFIO_VF_TOKEN "\n");
 788                                 eal_usage(prgname);
 789                                 ret = -1;
 790                                 goto out;
 791                         }
 792                         break;
 793
 794                 case OPT_CREATE_UIO_DEV_NUM:
 795                         internal_conf->create_uio_dev = 1;
 796                         break;
 797
 798                 case OPT_MBUF_POOL_OPS_NAME_NUM:
 799                 {
 800                         char *ops_name = strdup(optarg);
 801                         if (ops_name == NULL)
 802                                 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
 803                         else {
 804                                 /* free old ops name */
 805                                 if (internal_conf->user_mbuf_pool_ops_name !=
 806                                                 NULL)
 807                                         free(internal_conf->user_mbuf_pool_ops_name);
 808
 809                                 internal_conf->user_mbuf_pool_ops_name =
 810                                                 ops_name;
 811                         }
 812                         break;
 813                 }
 814                 case OPT_MATCH_ALLOCATIONS_NUM:
 815                         internal_conf->match_allocations = 1;
 816                         break;
 817
 818                 default:
 819                         if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
 820                                 RTE_LOG(ERR, EAL, "Option %c is not supported "
 821                                         "on Linux\n", opt);
 822                         } else if (opt >= OPT_LONG_MIN_NUM &&
 823                                    opt < OPT_LONG_MAX_NUM) {
 824                                 RTE_LOG(ERR, EAL, "Option %s is not supported "
 825                                         "on Linux\n",
 826                                         eal_long_options[option_index].name);
 827                         } else {
 828                                 RTE_LOG(ERR, EAL, "Option %d is not supported "
 829                                         "on Linux\n", opt);
 830                         }
 831                         eal_usage(prgname);
 832                         ret = -1;
 833                         goto out;
 834                 }
 835         }
 836
 837         /* create runtime data directory */
 838         if (internal_conf->no_shconf == 0 &&
 839                         eal_create_runtime_dir() < 0) {
 840                 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
 841                 ret = -1;
 842                 goto out;
 843         }
 844
 845         if (eal_adjust_config(internal_conf) != 0) {
 846                 ret = -1;
 847                 goto out;
 848         }
 849
 850         /* sanity checks */
 851         if (eal_check_common_options(internal_conf) != 0) {
 852                 eal_usage(prgname);
 853                 ret = -1;
 854                 goto out;
 855         }
 856
 857         if (optind >= 0)
 858                 argv[optind-1] = prgname;
 859         ret = optind-1;
 860
 861 out:
 862         /* restore getopt lib */
 863         optind = old_optind;
 864         optopt = old_optopt;
 865         optarg = old_optarg;
 866
 867         return ret;
 868 }
 869
 870 static int
 871 check_socket(const struct rte_memseg_list *msl, void *arg)
 872 {
 873         int *socket_id = arg;
 874
 875         if (msl->external)
 876                 return 0;
 877
 878         return *socket_id == msl->socket_id;
 879 }
 880
 881 static void
 882 eal_check_mem_on_local_socket(void)
 883 {
 884         int socket_id;
 885         const struct rte_config *config = rte_eal_get_configuration();
 886
 887         socket_id = rte_lcore_to_socket_id(config->master_lcore);
 888
 889         if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
 890                 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
 891 }
 892
 893 static int
 894 sync_func(__rte_unused void *arg)
 895 {
 896         return 0;
 897 }
 898
 899 /*
 900  * Request iopl privilege for all RPL, returns 0 on success
 901  * iopl() call is mostly for the i386 architecture. For other architectures,
 902  * return -1 to indicate IO privilege can't be changed in this way.
 903  */
 904 int
 905 rte_eal_iopl_init(void)
 906 {
 907 #if defined(RTE_ARCH_X86)
 908         if (iopl(3) != 0)
 909                 return -1;
 910 #endif
 911         return 0;
 912 }
 913
 914 #ifdef VFIO_PRESENT
 915 static int rte_eal_vfio_setup(void)
 916 {
 917         if (rte_vfio_enable("vfio"))
 918                 return -1;
 919
 920         return 0;
 921 }
 922 #endif
 923
 924 static void rte_eal_init_alert(const char *msg)
 925 {
 926         fprintf(stderr, "EAL: FATAL: %s\n", msg);
 927         RTE_LOG(ERR, EAL, "%s\n", msg);
 928 }
 929
 930 /*
 931  * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
 932  * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
 933  * IOMMU groups. If IOMMU is not enabled, that path would be empty.
 934  * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
 935  */
 936 static bool
 937 is_iommu_enabled(void)
 938 {
 939         DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
 940         struct dirent *d;
 941         int n = 0;
 942
 943         /* if directory doesn't exist, assume IOMMU is not enabled */
 944         if (dir == NULL)
 945                 return false;
 946
 947         while ((d = readdir(dir)) != NULL) {
 948                 /* skip dot and dot-dot */
 949                 if (++n > 2)
 950                         break;
 951         }
 952         closedir(dir);
 953
 954         return n > 2;
 955 }
 956
 957 /* Launch threads, called at application init(). */
 958 int
 959 rte_eal_init(int argc, char **argv)
 960 {
 961         int i, fctret, ret;
 962         pthread_t thread_id;
 963         static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
 964         const char *p;
 965         static char logid[PATH_MAX];
 966         char cpuset[RTE_CPU_AFFINITY_STR_LEN];
 967         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 968         bool phys_addrs;
 969         const struct rte_config *config = rte_eal_get_configuration();
 970         struct internal_config *internal_conf =
 971                 eal_get_internal_configuration();
 972
 973         /* checks if the machine is adequate */
 974         if (!rte_cpu_is_supported()) {
 975                 rte_eal_init_alert("unsupported cpu type.");
 976                 rte_errno = ENOTSUP;
 977                 return -1;
 978         }
 979
 980         if (!rte_atomic32_test_and_set(&run_once)) {
 981                 rte_eal_init_alert("already called initialization.");
 982                 rte_errno = EALREADY;
 983                 return -1;
 984         }
 985
 986         p = strrchr(argv[0], '/');
 987         strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
 988         thread_id = pthread_self();
 989
 990         eal_reset_internal_config(internal_conf);
 991
 992         /* set log level as early as possible */
 993         eal_log_level_parse(argc, argv);
 994
 995         /* clone argv to report out later in telemetry */
 996         eal_save_args(argc, argv);
 997
 998         if (rte_eal_cpu_init() < 0) {
 999                 rte_eal_init_alert("Cannot detect lcores.");
1000                 rte_errno = ENOTSUP;
1001                 return -1;
1002         }
1003
1004         fctret = eal_parse_args(argc, argv);
1005         if (fctret < 0) {
1006                 rte_eal_init_alert("Invalid 'command line' arguments.");
1007                 rte_errno = EINVAL;
1008                 rte_atomic32_clear(&run_once);
1009                 return -1;
1010         }
1011
1012         if (eal_plugins_init() < 0) {
1013                 rte_eal_init_alert("Cannot init plugins");
1014                 rte_errno = EINVAL;
1015                 rte_atomic32_clear(&run_once);
1016                 return -1;
1017         }
1018
1019         if (eal_trace_init() < 0) {
1020                 rte_eal_init_alert("Cannot init trace");
1021                 rte_errno = EFAULT;
1022                 return -1;
1023         }
1024
1025         if (eal_option_device_parse()) {
1026                 rte_errno = ENODEV;
1027                 rte_atomic32_clear(&run_once);
1028                 return -1;
1029         }
1030
1031         if (rte_config_init() < 0) {
1032                 rte_eal_init_alert("Cannot init config");
1033                 return -1;
1034         }
1035
1036         if (rte_eal_intr_init() < 0) {
1037                 rte_eal_init_alert("Cannot init interrupt-handling thread");
1038                 return -1;
1039         }
1040
1041         if (rte_eal_alarm_init() < 0) {
1042                 rte_eal_init_alert("Cannot init alarm");
1043                 /* rte_eal_alarm_init sets rte_errno on failure. */
1044                 return -1;
1045         }
1046
1047         /* Put mp channel init before bus scan so that we can init the vdev
1048          * bus through mp channel in the secondary process before the bus scan.
1049          */
1050         if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
1051                 rte_eal_init_alert("failed to init mp channel");
1052                 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1053                         rte_errno = EFAULT;
1054                         return -1;
1055                 }
1056         }
1057
1058         /* register multi-process action callbacks for hotplug */
1059         if (eal_mp_dev_hotplug_init() < 0) {
1060                 rte_eal_init_alert("failed to register mp callback for hotplug");
1061                 return -1;
1062         }
1063
1064         if (rte_bus_scan()) {
1065                 rte_eal_init_alert("Cannot scan the buses for devices");
1066                 rte_errno = ENODEV;
1067                 rte_atomic32_clear(&run_once);
1068                 return -1;
1069         }
1070
1071         phys_addrs = rte_eal_using_phys_addrs() != 0;
1072
1073         /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
1074         if (internal_conf->iova_mode == RTE_IOVA_DC) {
1075                 /* autodetect the IOVA mapping mode */
1076                 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
1077
1078                 if (iova_mode == RTE_IOVA_DC) {
1079                         RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
1080
1081                         if (!phys_addrs) {
1082                                 /* if we have no access to physical addresses,
1083                                  * pick IOVA as VA mode.
1084                                  */
1085                                 iova_mode = RTE_IOVA_VA;
1086                                 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
1087 #if defined(RTE_LIBRTE_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
1088                         } else if (rte_eal_check_module("rte_kni") == 1) {
1089                                 iova_mode = RTE_IOVA_PA;
1090                                 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
1091 #endif
1092                         } else if (is_iommu_enabled()) {
1093                                 /* we have an IOMMU, pick IOVA as VA mode */
1094                                 iova_mode = RTE_IOVA_VA;
1095                                 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
1096                         } else {
1097                                 /* physical addresses available, and no IOMMU
1098                                  * found, so pick IOVA as PA.
1099                                  */
1100                                 iova_mode = RTE_IOVA_PA;
1101                                 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1102                         }
1103                 }
1104 #if defined(RTE_LIBRTE_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1105                 /* Workaround for KNI which requires physical address to work
1106                  * in kernels < 4.10
1107                  */
1108                 if (iova_mode == RTE_IOVA_VA &&
1109                                 rte_eal_check_module("rte_kni") == 1) {
1110                         if (phys_addrs) {
1111                                 iova_mode = RTE_IOVA_PA;
1112                                 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1113                         } else {
1114                                 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1115                         }
1116                 }
1117 #endif
1118                 rte_eal_get_configuration()->iova_mode = iova_mode;
1119         } else {
1120                 rte_eal_get_configuration()->iova_mode =
1121                         internal_conf->iova_mode;
1122         }
1123
1124         if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1125                 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1126                 rte_errno = EINVAL;
1127                 return -1;
1128         }
1129
1130         RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1131                 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1132
1133         if (internal_conf->no_hugetlbfs == 0) {
1134                 /* rte_config isn't initialized yet */
1135                 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
1136                                 eal_hugepage_info_init() :
1137                                 eal_hugepage_info_read();
1138                 if (ret < 0) {
1139                         rte_eal_init_alert("Cannot get hugepage information.");
1140                         rte_errno = EACCES;
1141                         rte_atomic32_clear(&run_once);
1142                         return -1;
1143                 }
1144         }
1145
1146         if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
1147                 if (internal_conf->no_hugetlbfs)
1148                         internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
1149         }
1150
1151         if (internal_conf->vmware_tsc_map == 1) {
1152 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1153                 rte_cycles_vmware_tsc_map = 1;
1154                 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1155                                 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1156 #else
1157                 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1158                                 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1159 #endif
1160         }
1161
1162         if (rte_eal_log_init(logid, internal_conf->syslog_facility) < 0) {
1163                 rte_eal_init_alert("Cannot init logging.");
1164                 rte_errno = ENOMEM;
1165                 rte_atomic32_clear(&run_once);
1166                 return -1;
1167         }
1168
1169 #ifdef VFIO_PRESENT
1170         if (rte_eal_vfio_setup() < 0) {
1171                 rte_eal_init_alert("Cannot init VFIO");
1172                 rte_errno = EAGAIN;
1173                 rte_atomic32_clear(&run_once);
1174                 return -1;
1175         }
1176 #endif
1177         /* in secondary processes, memory init may allocate additional fbarrays
1178          * not present in primary processes, so to avoid any potential issues,
1179          * initialize memzones first.
1180          */
1181         if (rte_eal_memzone_init() < 0) {
1182                 rte_eal_init_alert("Cannot init memzone");
1183                 rte_errno = ENODEV;
1184                 return -1;
1185         }
1186
1187         if (rte_eal_memory_init() < 0) {
1188                 rte_eal_init_alert("Cannot init memory");
1189                 rte_errno = ENOMEM;
1190                 return -1;
1191         }
1192
1193         /* the directories are locked during eal_hugepage_info_init */
1194         eal_hugedirs_unlock();
1195
1196         if (rte_eal_malloc_heap_init() < 0) {
1197                 rte_eal_init_alert("Cannot init malloc heap");
1198                 rte_errno = ENODEV;
1199                 return -1;
1200         }
1201
1202         if (rte_eal_tailqs_init() < 0) {
1203                 rte_eal_init_alert("Cannot init tail queues for objects");
1204                 rte_errno = EFAULT;
1205                 return -1;
1206         }
1207
1208         if (rte_eal_timer_init() < 0) {
1209                 rte_eal_init_alert("Cannot init HPET or TSC timers");
1210                 rte_errno = ENOTSUP;
1211                 return -1;
1212         }
1213
1214         eal_check_mem_on_local_socket();
1215
1216         if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
1217                         &lcore_config[config->master_lcore].cpuset) != 0) {
1218                 rte_eal_init_alert("Cannot set affinity");
1219                 rte_errno = EINVAL;
1220                 return -1;
1221         }
1222         __rte_thread_init(config->master_lcore,
1223                 &lcore_config[config->master_lcore].cpuset);
1224
1225         ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
1226         RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1227                 config->master_lcore, (uintptr_t)thread_id, cpuset,
1228                 ret == 0 ? "" : "...");
1229
1230         RTE_LCORE_FOREACH_SLAVE(i) {
1231
1232                 /*
1233                  * create communication pipes between master thread
1234                  * and children
1235                  */
1236                 if (pipe(lcore_config[i].pipe_master2slave) < 0)
1237                         rte_panic("Cannot create pipe\n");
1238                 if (pipe(lcore_config[i].pipe_slave2master) < 0)
1239                         rte_panic("Cannot create pipe\n");
1240
1241                 lcore_config[i].state = WAIT;
1242
1243                 /* create a thread for each lcore */
1244                 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1245                                      eal_thread_loop, NULL);
1246                 if (ret != 0)
1247                         rte_panic("Cannot create thread\n");
1248
1249                 /* Set thread_name for aid in debugging. */
1250                 snprintf(thread_name, sizeof(thread_name),
1251                         "lcore-slave-%d", i);
1252                 ret = rte_thread_setname(lcore_config[i].thread_id,
1253                                                 thread_name);
1254                 if (ret != 0)
1255                         RTE_LOG(DEBUG, EAL,
1256                                 "Cannot set name for lcore thread\n");
1257
1258                 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
1259                         sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
1260                 if (ret != 0)
1261                         rte_panic("Cannot set affinity\n");
1262         }
1263
1264         /*
1265          * Launch a dummy function on all slave lcores, so that master lcore
1266          * knows they are all ready when this function returns.
1267          */
1268         rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
1269         rte_eal_mp_wait_lcore();
1270
1271         /* initialize services so vdevs register service during bus_probe. */
1272         ret = rte_service_init();
1273         if (ret) {
1274                 rte_eal_init_alert("rte_service_init() failed");
1275                 rte_errno = ENOEXEC;
1276                 return -1;
1277         }
1278
1279         /* Probe all the buses and devices/drivers on them */
1280         if (rte_bus_probe()) {
1281                 rte_eal_init_alert("Cannot probe devices");
1282                 rte_errno = ENOTSUP;
1283                 return -1;
1284         }
1285
1286 #ifdef VFIO_PRESENT
1287         /* Register mp action after probe() so that we got enough info */
1288         if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1289                 return -1;
1290 #endif
1291
1292         /* initialize default service/lcore mappings and start running. Ignore
1293          * -ENOTSUP, as it indicates no service coremask passed to EAL.
1294          */
1295         ret = rte_service_start_with_defaults();
1296         if (ret < 0 && ret != -ENOTSUP) {
1297                 rte_errno = ENOEXEC;
1298                 return -1;
1299         }
1300
1301         /*
1302          * Clean up unused files in runtime directory. We do this at the end of
1303          * init and not at the beginning because we want to clean stuff up
1304          * whether we are primary or secondary process, but we cannot remove
1305          * primary process' files because secondary should be able to run even
1306          * if primary process is dead.
1307          *
1308          * In no_shconf mode, no runtime directory is created in the first
1309          * place, so no cleanup needed.
1310          */
1311         if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
1312                 rte_eal_init_alert("Cannot clear runtime directory");
1313                 return -1;
1314         }
1315         if (!internal_conf->no_telemetry) {
1316                 const char *error_str = NULL;
1317                 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1318                                 &internal_conf->ctrl_cpuset, &error_str)
1319                                 != 0) {
1320                         rte_eal_init_alert(error_str);
1321                         return -1;
1322                 }
1323                 if (error_str != NULL)
1324                         RTE_LOG(NOTICE, EAL, "%s\n", error_str);
1325         }
1326
1327         eal_mcfg_complete();
1328
1329         return fctret;
1330 }
1331
1332 static int
1333 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1334                 void *arg __rte_unused)
1335 {
1336         /* ms is const, so find this memseg */
1337         struct rte_memseg *found;
1338
1339         if (msl->external)
1340                 return 0;
1341
1342         found = rte_mem_virt2memseg(ms->addr, msl);
1343
1344         found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1345
1346         return 0;
1347 }
1348
1349 int
1350 rte_eal_cleanup(void)
1351 {
1352         /* if we're in a primary process, we need to mark hugepages as freeable
1353          * so that finalization can release them back to the system.
1354          */
1355         struct internal_config *internal_conf =
1356                 eal_get_internal_configuration();
1357
1358         if (rte_eal_process_type() == RTE_PROC_PRIMARY)
1359                 rte_memseg_walk(mark_freeable, NULL);
1360         rte_service_finalize();
1361         rte_mp_channel_cleanup();
1362         rte_trace_save();
1363         eal_trace_fini();
1364         eal_cleanup_config(internal_conf);
1365         return 0;
1366 }
1367
1368 int rte_eal_create_uio_dev(void)
1369 {
1370         const struct internal_config *internal_conf =
1371                 eal_get_internal_configuration();
1372
1373         return internal_conf->create_uio_dev;
1374 }
1375
1376 enum rte_intr_mode
1377 rte_eal_vfio_intr_mode(void)
1378 {
1379         const struct internal_config *internal_conf =
1380                 eal_get_internal_configuration();
1381
1382         return internal_conf->vfio_intr_mode;
1383 }
1384
1385 void
1386 rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)
1387 {
1388         struct internal_config *cfg = eal_get_internal_configuration();
1389
1390         rte_uuid_copy(vf_token, cfg->vfio_vf_token);
1391 }
1392
1393 int
1394 rte_eal_check_module(const char *module_name)
1395 {
1396         char sysfs_mod_name[PATH_MAX];
1397         struct stat st;
1398         int n;
1399
1400         if (NULL == module_name)
1401                 return -1;
1402
1403         /* Check if there is sysfs mounted */
1404         if (stat("/sys/module", &st) != 0) {
1405                 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1406                         errno, strerror(errno));
1407                 return -1;
1408         }
1409
1410         /* A module might be built-in, therefore try sysfs */
1411         n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1412         if (n < 0 || n > PATH_MAX) {
1413                 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1414                 return -1;
1415         }
1416
1417         if (stat(sysfs_mod_name, &st) != 0) {
1418                 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1419                         sysfs_mod_name, errno, strerror(errno));
1420                 return 0;
1421         }
1422
1423         /* Module has been found */
1424         return 1;
1425 }