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                 eal_mcfg_update_internal();
 496                 break;
 497         case RTE_PROC_AUTO:
 498         case RTE_PROC_INVALID:
 499                 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
 500                         config->process_type);
 501                 return -1;
 502         }
 503
 504         return 0;
 505 }
 506
 507 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
 508 static void
 509 eal_hugedirs_unlock(void)
 510 {
 511         int i;
 512         struct internal_config *internal_conf =
 513                 eal_get_internal_configuration();
 514
 515         for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
 516         {
 517                 /* skip uninitialized */
 518                 if (internal_conf->hugepage_info[i].lock_descriptor < 0)
 519                         continue;
 520                 /* unlock hugepage file */
 521                 flock(internal_conf->hugepage_info[i].lock_descriptor, LOCK_UN);
 522                 close(internal_conf->hugepage_info[i].lock_descriptor);
 523                 /* reset the field */
 524                 internal_conf->hugepage_info[i].lock_descriptor = -1;
 525         }
 526 }
 527
 528 /* display usage */
 529 static void
 530 eal_usage(const char *prgname)
 531 {
 532         rte_usage_hook_t hook = eal_get_application_usage_hook();
 533
 534         printf("\nUsage: %s ", prgname);
 535         eal_common_usage();
 536         printf("EAL Linux options:\n"
 537                "  --"OPT_SOCKET_MEM"        Memory to allocate on sockets (comma separated values)\n"
 538                "  --"OPT_SOCKET_LIMIT"      Limit memory allocation on sockets (comma separated values)\n"
 539                "  --"OPT_HUGE_DIR"          Directory where hugetlbfs is mounted\n"
 540                "  --"OPT_FILE_PREFIX"       Prefix for hugepage filenames\n"
 541                "  --"OPT_CREATE_UIO_DEV"    Create /dev/uioX (usually done by hotplug)\n"
 542                "  --"OPT_VFIO_INTR"         Interrupt mode for VFIO (legacy|msi|msix)\n"
 543                "  --"OPT_VFIO_VF_TOKEN"     VF token (UUID) shared between SR-IOV PF and VFs\n"
 544                "  --"OPT_LEGACY_MEM"        Legacy memory mode (no dynamic allocation, contiguous segments)\n"
 545                "  --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
 546                "  --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
 547                "\n");
 548         /* Allow the application to print its usage message too if hook is set */
 549         if (hook) {
 550                 printf("===== Application Usage =====\n\n");
 551                 (hook)(prgname);
 552         }
 553 }
 554
 555 static int
 556 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
 557 {
 558         char * arg[RTE_MAX_NUMA_NODES];
 559         char *end;
 560         int arg_num, i, len;
 561         uint64_t total_mem = 0;
 562
 563         len = strnlen(strval, SOCKET_MEM_STRLEN);
 564         if (len == SOCKET_MEM_STRLEN) {
 565                 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
 566                 return -1;
 567         }
 568
 569         /* all other error cases will be caught later */
 570         if (!isdigit(strval[len-1]))
 571                 return -1;
 572
 573         /* split the optarg into separate socket values */
 574         arg_num = rte_strsplit(strval, len,
 575                         arg, RTE_MAX_NUMA_NODES, ',');
 576
 577         /* if split failed, or 0 arguments */
 578         if (arg_num <= 0)
 579                 return -1;
 580
 581         /* parse each defined socket option */
 582         errno = 0;
 583         for (i = 0; i < arg_num; i++) {
 584                 uint64_t val;
 585                 end = NULL;
 586                 val = strtoull(arg[i], &end, 10);
 587
 588                 /* check for invalid input */
 589                 if ((errno != 0)  ||
 590                                 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
 591                         return -1;
 592                 val <<= 20;
 593                 total_mem += val;
 594                 socket_arg[i] = val;
 595         }
 596
 597         return 0;
 598 }
 599
 600 static int
 601 eal_parse_vfio_intr(const char *mode)
 602 {
 603         struct internal_config *internal_conf =
 604                 eal_get_internal_configuration();
 605         unsigned i;
 606         static struct {
 607                 const char *name;
 608                 enum rte_intr_mode value;
 609         } map[] = {
 610                 { "legacy", RTE_INTR_MODE_LEGACY },
 611                 { "msi", RTE_INTR_MODE_MSI },
 612                 { "msix", RTE_INTR_MODE_MSIX },
 613         };
 614
 615         for (i = 0; i < RTE_DIM(map); i++) {
 616                 if (!strcmp(mode, map[i].name)) {
 617                         internal_conf->vfio_intr_mode = map[i].value;
 618                         return 0;
 619                 }
 620         }
 621         return -1;
 622 }
 623
 624 static int
 625 eal_parse_vfio_vf_token(const char *vf_token)
 626 {
 627         struct internal_config *cfg = eal_get_internal_configuration();
 628         rte_uuid_t uuid;
 629
 630         if (!rte_uuid_parse(vf_token, uuid)) {
 631                 rte_uuid_copy(cfg->vfio_vf_token, uuid);
 632                 return 0;
 633         }
 634
 635         return -1;
 636 }
 637
 638 /* Parse the arguments for --log-level only */
 639 static void
 640 eal_log_level_parse(int argc, char **argv)
 641 {
 642         int opt;
 643         char **argvopt;
 644         int option_index;
 645         const int old_optind = optind;
 646         const int old_optopt = optopt;
 647         char * const old_optarg = optarg;
 648         struct internal_config *internal_conf =
 649                 eal_get_internal_configuration();
 650
 651         argvopt = argv;
 652         optind = 1;
 653
 654         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 655                                   eal_long_options, &option_index)) != EOF) {
 656
 657                 int ret;
 658
 659                 /* getopt is not happy, stop right now */
 660                 if (opt == '?')
 661                         break;
 662
 663                 ret = (opt == OPT_LOG_LEVEL_NUM) ?
 664                         eal_parse_common_option(opt, optarg, internal_conf) : 0;
 665
 666                 /* common parser is not happy */
 667                 if (ret < 0)
 668                         break;
 669         }
 670
 671         /* restore getopt lib */
 672         optind = old_optind;
 673         optopt = old_optopt;
 674         optarg = old_optarg;
 675 }
 676
 677 /* Parse the argument given in the command line of the application */
 678 static int
 679 eal_parse_args(int argc, char **argv)
 680 {
 681         int opt, ret;
 682         char **argvopt;
 683         int option_index;
 684         char *prgname = argv[0];
 685         const int old_optind = optind;
 686         const int old_optopt = optopt;
 687         char * const old_optarg = optarg;
 688         struct internal_config *internal_conf =
 689                 eal_get_internal_configuration();
 690
 691         argvopt = argv;
 692         optind = 1;
 693
 694         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 695                                   eal_long_options, &option_index)) != EOF) {
 696
 697                 /* getopt didn't recognise the option */
 698                 if (opt == '?') {
 699                         eal_usage(prgname);
 700                         ret = -1;
 701                         goto out;
 702                 }
 703
 704                 ret = eal_parse_common_option(opt, optarg, internal_conf);
 705                 /* common parser is not happy */
 706                 if (ret < 0) {
 707                         eal_usage(prgname);
 708                         ret = -1;
 709                         goto out;
 710                 }
 711                 /* common parser handled this option */
 712                 if (ret == 0)
 713                         continue;
 714
 715                 switch (opt) {
 716                 case 'h':
 717                         eal_usage(prgname);
 718                         exit(EXIT_SUCCESS);
 719
 720                 case OPT_HUGE_DIR_NUM:
 721                 {
 722                         char *hdir = strdup(optarg);
 723                         if (hdir == NULL)
 724                                 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
 725                         else {
 726                                 /* free old hugepage dir */
 727                                 if (internal_conf->hugepage_dir != NULL)
 728                                         free(internal_conf->hugepage_dir);
 729                                 internal_conf->hugepage_dir = hdir;
 730                         }
 731                         break;
 732                 }
 733                 case OPT_FILE_PREFIX_NUM:
 734                 {
 735                         char *prefix = strdup(optarg);
 736                         if (prefix == NULL)
 737                                 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
 738                         else {
 739                                 /* free old prefix */
 740                                 if (internal_conf->hugefile_prefix != NULL)
 741                                         free(internal_conf->hugefile_prefix);
 742                                 internal_conf->hugefile_prefix = prefix;
 743                         }
 744                         break;
 745                 }
 746                 case OPT_SOCKET_MEM_NUM:
 747                         if (eal_parse_socket_arg(optarg,
 748                                         internal_conf->socket_mem) < 0) {
 749                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 750                                                 OPT_SOCKET_MEM "\n");
 751                                 eal_usage(prgname);
 752                                 ret = -1;
 753                                 goto out;
 754                         }
 755                         internal_conf->force_sockets = 1;
 756                         break;
 757
 758                 case OPT_SOCKET_LIMIT_NUM:
 759                         if (eal_parse_socket_arg(optarg,
 760                                         internal_conf->socket_limit) < 0) {
 761                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 762                                                 OPT_SOCKET_LIMIT "\n");
 763                                 eal_usage(prgname);
 764                                 ret = -1;
 765                                 goto out;
 766                         }
 767                         internal_conf->force_socket_limits = 1;
 768                         break;
 769
 770                 case OPT_VFIO_INTR_NUM:
 771                         if (eal_parse_vfio_intr(optarg) < 0) {
 772                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 773                                                 OPT_VFIO_INTR "\n");
 774                                 eal_usage(prgname);
 775                                 ret = -1;
 776                                 goto out;
 777                         }
 778                         break;
 779
 780                 case OPT_VFIO_VF_TOKEN_NUM:
 781                         if (eal_parse_vfio_vf_token(optarg) < 0) {
 782                                 RTE_LOG(ERR, EAL, "invalid parameters for --"
 783                                                 OPT_VFIO_VF_TOKEN "\n");
 784                                 eal_usage(prgname);
 785                                 ret = -1;
 786                                 goto out;
 787                         }
 788                         break;
 789
 790                 case OPT_CREATE_UIO_DEV_NUM:
 791                         internal_conf->create_uio_dev = 1;
 792                         break;
 793
 794                 case OPT_MBUF_POOL_OPS_NAME_NUM:
 795                 {
 796                         char *ops_name = strdup(optarg);
 797                         if (ops_name == NULL)
 798                                 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
 799                         else {
 800                                 /* free old ops name */
 801                                 if (internal_conf->user_mbuf_pool_ops_name !=
 802                                                 NULL)
 803                                         free(internal_conf->user_mbuf_pool_ops_name);
 804
 805                                 internal_conf->user_mbuf_pool_ops_name =
 806                                                 ops_name;
 807                         }
 808                         break;
 809                 }
 810                 case OPT_MATCH_ALLOCATIONS_NUM:
 811                         internal_conf->match_allocations = 1;
 812                         break;
 813
 814                 default:
 815                         if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
 816                                 RTE_LOG(ERR, EAL, "Option %c is not supported "
 817                                         "on Linux\n", opt);
 818                         } else if (opt >= OPT_LONG_MIN_NUM &&
 819                                    opt < OPT_LONG_MAX_NUM) {
 820                                 RTE_LOG(ERR, EAL, "Option %s is not supported "
 821                                         "on Linux\n",
 822                                         eal_long_options[option_index].name);
 823                         } else {
 824                                 RTE_LOG(ERR, EAL, "Option %d is not supported "
 825                                         "on Linux\n", opt);
 826                         }
 827                         eal_usage(prgname);
 828                         ret = -1;
 829                         goto out;
 830                 }
 831         }
 832
 833         /* create runtime data directory */
 834         if (internal_conf->no_shconf == 0 &&
 835                         eal_create_runtime_dir() < 0) {
 836                 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
 837                 ret = -1;
 838                 goto out;
 839         }
 840
 841         if (eal_adjust_config(internal_conf) != 0) {
 842                 ret = -1;
 843                 goto out;
 844         }
 845
 846         /* sanity checks */
 847         if (eal_check_common_options(internal_conf) != 0) {
 848                 eal_usage(prgname);
 849                 ret = -1;
 850                 goto out;
 851         }
 852
 853         if (optind >= 0)
 854                 argv[optind-1] = prgname;
 855         ret = optind-1;
 856
 857 out:
 858         /* restore getopt lib */
 859         optind = old_optind;
 860         optopt = old_optopt;
 861         optarg = old_optarg;
 862
 863         return ret;
 864 }
 865
 866 static int
 867 check_socket(const struct rte_memseg_list *msl, void *arg)
 868 {
 869         int *socket_id = arg;
 870
 871         if (msl->external)
 872                 return 0;
 873
 874         return *socket_id == msl->socket_id;
 875 }
 876
 877 static void
 878 eal_check_mem_on_local_socket(void)
 879 {
 880         int socket_id;
 881         const struct rte_config *config = rte_eal_get_configuration();
 882
 883         socket_id = rte_lcore_to_socket_id(config->master_lcore);
 884
 885         if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
 886                 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
 887 }
 888
 889 static int
 890 sync_func(__rte_unused void *arg)
 891 {
 892         return 0;
 893 }
 894
 895 /*
 896  * Request iopl privilege for all RPL, returns 0 on success
 897  * iopl() call is mostly for the i386 architecture. For other architectures,
 898  * return -1 to indicate IO privilege can't be changed in this way.
 899  */
 900 int
 901 rte_eal_iopl_init(void)
 902 {
 903 #if defined(RTE_ARCH_X86)
 904         if (iopl(3) != 0)
 905                 return -1;
 906 #endif
 907         return 0;
 908 }
 909
 910 #ifdef VFIO_PRESENT
 911 static int rte_eal_vfio_setup(void)
 912 {
 913         if (rte_vfio_enable("vfio"))
 914                 return -1;
 915
 916         return 0;
 917 }
 918 #endif
 919
 920 static void rte_eal_init_alert(const char *msg)
 921 {
 922         fprintf(stderr, "EAL: FATAL: %s\n", msg);
 923         RTE_LOG(ERR, EAL, "%s\n", msg);
 924 }
 925
 926 /*
 927  * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
 928  * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
 929  * IOMMU groups. If IOMMU is not enabled, that path would be empty.
 930  * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
 931  */
 932 static bool
 933 is_iommu_enabled(void)
 934 {
 935         DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
 936         struct dirent *d;
 937         int n = 0;
 938
 939         /* if directory doesn't exist, assume IOMMU is not enabled */
 940         if (dir == NULL)
 941                 return false;
 942
 943         while ((d = readdir(dir)) != NULL) {
 944                 /* skip dot and dot-dot */
 945                 if (++n > 2)
 946                         break;
 947         }
 948         closedir(dir);
 949
 950         return n > 2;
 951 }
 952
 953 /* Launch threads, called at application init(). */
 954 int
 955 rte_eal_init(int argc, char **argv)
 956 {
 957         int i, fctret, ret;
 958         pthread_t thread_id;
 959         static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
 960         const char *p;
 961         static char logid[PATH_MAX];
 962         char cpuset[RTE_CPU_AFFINITY_STR_LEN];
 963         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 964         bool phys_addrs;
 965         const struct rte_config *config = rte_eal_get_configuration();
 966         struct internal_config *internal_conf =
 967                 eal_get_internal_configuration();
 968
 969         /* checks if the machine is adequate */
 970         if (!rte_cpu_is_supported()) {
 971                 rte_eal_init_alert("unsupported cpu type.");
 972                 rte_errno = ENOTSUP;
 973                 return -1;
 974         }
 975
 976         if (!rte_atomic32_test_and_set(&run_once)) {
 977                 rte_eal_init_alert("already called initialization.");
 978                 rte_errno = EALREADY;
 979                 return -1;
 980         }
 981
 982         p = strrchr(argv[0], '/');
 983         strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
 984         thread_id = pthread_self();
 985
 986         eal_reset_internal_config(internal_conf);
 987
 988         /* set log level as early as possible */
 989         eal_log_level_parse(argc, argv);
 990
 991         /* clone argv to report out later in telemetry */
 992         eal_save_args(argc, argv);
 993
 994         if (rte_eal_cpu_init() < 0) {
 995                 rte_eal_init_alert("Cannot detect lcores.");
 996                 rte_errno = ENOTSUP;
 997                 return -1;
 998         }
 999
1000         fctret = eal_parse_args(argc, argv);
1001         if (fctret < 0) {
1002                 rte_eal_init_alert("Invalid 'command line' arguments.");
1003                 rte_errno = EINVAL;
1004                 rte_atomic32_clear(&run_once);
1005                 return -1;
1006         }
1007
1008         if (eal_plugins_init() < 0) {
1009                 rte_eal_init_alert("Cannot init plugins");
1010                 rte_errno = EINVAL;
1011                 rte_atomic32_clear(&run_once);
1012                 return -1;
1013         }
1014
1015         if (eal_trace_init() < 0) {
1016                 rte_eal_init_alert("Cannot init trace");
1017                 rte_errno = EFAULT;
1018                 return -1;
1019         }
1020
1021         if (eal_option_device_parse()) {
1022                 rte_errno = ENODEV;
1023                 rte_atomic32_clear(&run_once);
1024                 return -1;
1025         }
1026
1027         if (rte_config_init() < 0) {
1028                 rte_eal_init_alert("Cannot init config");
1029                 return -1;
1030         }
1031
1032         if (rte_eal_intr_init() < 0) {
1033                 rte_eal_init_alert("Cannot init interrupt-handling thread");
1034                 return -1;
1035         }
1036
1037         if (rte_eal_alarm_init() < 0) {
1038                 rte_eal_init_alert("Cannot init alarm");
1039                 /* rte_eal_alarm_init sets rte_errno on failure. */
1040                 return -1;
1041         }
1042
1043         /* Put mp channel init before bus scan so that we can init the vdev
1044          * bus through mp channel in the secondary process before the bus scan.
1045          */
1046         if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
1047                 rte_eal_init_alert("failed to init mp channel");
1048                 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1049                         rte_errno = EFAULT;
1050                         return -1;
1051                 }
1052         }
1053
1054         /* register multi-process action callbacks for hotplug */
1055         if (eal_mp_dev_hotplug_init() < 0) {
1056                 rte_eal_init_alert("failed to register mp callback for hotplug");
1057                 return -1;
1058         }
1059
1060         if (rte_bus_scan()) {
1061                 rte_eal_init_alert("Cannot scan the buses for devices");
1062                 rte_errno = ENODEV;
1063                 rte_atomic32_clear(&run_once);
1064                 return -1;
1065         }
1066
1067         phys_addrs = rte_eal_using_phys_addrs() != 0;
1068
1069         /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
1070         if (internal_conf->iova_mode == RTE_IOVA_DC) {
1071                 /* autodetect the IOVA mapping mode */
1072                 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
1073
1074                 if (iova_mode == RTE_IOVA_DC) {
1075                         RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
1076
1077                         if (!phys_addrs) {
1078                                 /* if we have no access to physical addresses,
1079                                  * pick IOVA as VA mode.
1080                                  */
1081                                 iova_mode = RTE_IOVA_VA;
1082                                 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
1083 #if defined(RTE_LIBRTE_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
1084                         } else if (rte_eal_check_module("rte_kni") == 1) {
1085                                 iova_mode = RTE_IOVA_PA;
1086                                 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
1087 #endif
1088                         } else if (is_iommu_enabled()) {
1089                                 /* we have an IOMMU, pick IOVA as VA mode */
1090                                 iova_mode = RTE_IOVA_VA;
1091                                 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
1092                         } else {
1093                                 /* physical addresses available, and no IOMMU
1094                                  * found, so pick IOVA as PA.
1095                                  */
1096                                 iova_mode = RTE_IOVA_PA;
1097                                 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1098                         }
1099                 }
1100 #if defined(RTE_LIBRTE_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1101                 /* Workaround for KNI which requires physical address to work
1102                  * in kernels < 4.10
1103                  */
1104                 if (iova_mode == RTE_IOVA_VA &&
1105                                 rte_eal_check_module("rte_kni") == 1) {
1106                         if (phys_addrs) {
1107                                 iova_mode = RTE_IOVA_PA;
1108                                 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1109                         } else {
1110                                 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1111                         }
1112                 }
1113 #endif
1114                 rte_eal_get_configuration()->iova_mode = iova_mode;
1115         } else {
1116                 rte_eal_get_configuration()->iova_mode =
1117                         internal_conf->iova_mode;
1118         }
1119
1120         if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1121                 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1122                 rte_errno = EINVAL;
1123                 return -1;
1124         }
1125
1126         RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1127                 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1128
1129         if (internal_conf->no_hugetlbfs == 0) {
1130                 /* rte_config isn't initialized yet */
1131                 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
1132                                 eal_hugepage_info_init() :
1133                                 eal_hugepage_info_read();
1134                 if (ret < 0) {
1135                         rte_eal_init_alert("Cannot get hugepage information.");
1136                         rte_errno = EACCES;
1137                         rte_atomic32_clear(&run_once);
1138                         return -1;
1139                 }
1140         }
1141
1142         if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
1143                 if (internal_conf->no_hugetlbfs)
1144                         internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
1145         }
1146
1147         if (internal_conf->vmware_tsc_map == 1) {
1148 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1149                 rte_cycles_vmware_tsc_map = 1;
1150                 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1151                                 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1152 #else
1153                 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1154                                 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1155 #endif
1156         }
1157
1158         if (rte_eal_log_init(logid, internal_conf->syslog_facility) < 0) {
1159                 rte_eal_init_alert("Cannot init logging.");
1160                 rte_errno = ENOMEM;
1161                 rte_atomic32_clear(&run_once);
1162                 return -1;
1163         }
1164
1165 #ifdef VFIO_PRESENT
1166         if (rte_eal_vfio_setup() < 0) {
1167                 rte_eal_init_alert("Cannot init VFIO");
1168                 rte_errno = EAGAIN;
1169                 rte_atomic32_clear(&run_once);
1170                 return -1;
1171         }
1172 #endif
1173         /* in secondary processes, memory init may allocate additional fbarrays
1174          * not present in primary processes, so to avoid any potential issues,
1175          * initialize memzones first.
1176          */
1177         if (rte_eal_memzone_init() < 0) {
1178                 rte_eal_init_alert("Cannot init memzone");
1179                 rte_errno = ENODEV;
1180                 return -1;
1181         }
1182
1183         if (rte_eal_memory_init() < 0) {
1184                 rte_eal_init_alert("Cannot init memory");
1185                 rte_errno = ENOMEM;
1186                 return -1;
1187         }
1188
1189         /* the directories are locked during eal_hugepage_info_init */
1190         eal_hugedirs_unlock();
1191
1192         if (rte_eal_malloc_heap_init() < 0) {
1193                 rte_eal_init_alert("Cannot init malloc heap");
1194                 rte_errno = ENODEV;
1195                 return -1;
1196         }
1197
1198         if (rte_eal_tailqs_init() < 0) {
1199                 rte_eal_init_alert("Cannot init tail queues for objects");
1200                 rte_errno = EFAULT;
1201                 return -1;
1202         }
1203
1204         if (rte_eal_timer_init() < 0) {
1205                 rte_eal_init_alert("Cannot init HPET or TSC timers");
1206                 rte_errno = ENOTSUP;
1207                 return -1;
1208         }
1209
1210         eal_check_mem_on_local_socket();
1211
1212         if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
1213                         &lcore_config[config->master_lcore].cpuset) != 0) {
1214                 rte_eal_init_alert("Cannot set affinity");
1215                 rte_errno = EINVAL;
1216                 return -1;
1217         }
1218         __rte_thread_init(config->master_lcore,
1219                 &lcore_config[config->master_lcore].cpuset);
1220
1221         ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
1222         RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1223                 config->master_lcore, (uintptr_t)thread_id, cpuset,
1224                 ret == 0 ? "" : "...");
1225
1226         RTE_LCORE_FOREACH_SLAVE(i) {
1227
1228                 /*
1229                  * create communication pipes between master thread
1230                  * and children
1231                  */
1232                 if (pipe(lcore_config[i].pipe_master2slave) < 0)
1233                         rte_panic("Cannot create pipe\n");
1234                 if (pipe(lcore_config[i].pipe_slave2master) < 0)
1235                         rte_panic("Cannot create pipe\n");
1236
1237                 lcore_config[i].state = WAIT;
1238
1239                 /* create a thread for each lcore */
1240                 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1241                                      eal_thread_loop, NULL);
1242                 if (ret != 0)
1243                         rte_panic("Cannot create thread\n");
1244
1245                 /* Set thread_name for aid in debugging. */
1246                 snprintf(thread_name, sizeof(thread_name),
1247                         "lcore-slave-%d", i);
1248                 ret = rte_thread_setname(lcore_config[i].thread_id,
1249                                                 thread_name);
1250                 if (ret != 0)
1251                         RTE_LOG(DEBUG, EAL,
1252                                 "Cannot set name for lcore thread\n");
1253
1254                 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
1255                         sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
1256                 if (ret != 0)
1257                         rte_panic("Cannot set affinity\n");
1258         }
1259
1260         /*
1261          * Launch a dummy function on all slave lcores, so that master lcore
1262          * knows they are all ready when this function returns.
1263          */
1264         rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
1265         rte_eal_mp_wait_lcore();
1266
1267         /* initialize services so vdevs register service during bus_probe. */
1268         ret = rte_service_init();
1269         if (ret) {
1270                 rte_eal_init_alert("rte_service_init() failed");
1271                 rte_errno = ENOEXEC;
1272                 return -1;
1273         }
1274
1275         /* Probe all the buses and devices/drivers on them */
1276         if (rte_bus_probe()) {
1277                 rte_eal_init_alert("Cannot probe devices");
1278                 rte_errno = ENOTSUP;
1279                 return -1;
1280         }
1281
1282 #ifdef VFIO_PRESENT
1283         /* Register mp action after probe() so that we got enough info */
1284         if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1285                 return -1;
1286 #endif
1287
1288         /* initialize default service/lcore mappings and start running. Ignore
1289          * -ENOTSUP, as it indicates no service coremask passed to EAL.
1290          */
1291         ret = rte_service_start_with_defaults();
1292         if (ret < 0 && ret != -ENOTSUP) {
1293                 rte_errno = ENOEXEC;
1294                 return -1;
1295         }
1296
1297         /*
1298          * Clean up unused files in runtime directory. We do this at the end of
1299          * init and not at the beginning because we want to clean stuff up
1300          * whether we are primary or secondary process, but we cannot remove
1301          * primary process' files because secondary should be able to run even
1302          * if primary process is dead.
1303          *
1304          * In no_shconf mode, no runtime directory is created in the first
1305          * place, so no cleanup needed.
1306          */
1307         if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
1308                 rte_eal_init_alert("Cannot clear runtime directory");
1309                 return -1;
1310         }
1311         if (!internal_conf->no_telemetry) {
1312                 const char *error_str = NULL;
1313                 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1314                                 &internal_conf->ctrl_cpuset, &error_str)
1315                                 != 0) {
1316                         rte_eal_init_alert(error_str);
1317                         return -1;
1318                 }
1319                 if (error_str != NULL)
1320                         RTE_LOG(NOTICE, EAL, "%s\n", error_str);
1321         }
1322
1323         eal_mcfg_complete();
1324
1325         return fctret;
1326 }
1327
1328 static int
1329 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1330                 void *arg __rte_unused)
1331 {
1332         /* ms is const, so find this memseg */
1333         struct rte_memseg *found;
1334
1335         if (msl->external)
1336                 return 0;
1337
1338         found = rte_mem_virt2memseg(ms->addr, msl);
1339
1340         found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1341
1342         return 0;
1343 }
1344
1345 int
1346 rte_eal_cleanup(void)
1347 {
1348         /* if we're in a primary process, we need to mark hugepages as freeable
1349          * so that finalization can release them back to the system.
1350          */
1351         struct internal_config *internal_conf =
1352                 eal_get_internal_configuration();
1353
1354         if (rte_eal_process_type() == RTE_PROC_PRIMARY)
1355                 rte_memseg_walk(mark_freeable, NULL);
1356         rte_service_finalize();
1357         rte_mp_channel_cleanup();
1358         rte_trace_save();
1359         eal_trace_fini();
1360         eal_cleanup_config(internal_conf);
1361         return 0;
1362 }
1363
1364 int rte_eal_create_uio_dev(void)
1365 {
1366         const struct internal_config *internal_conf =
1367                 eal_get_internal_configuration();
1368
1369         return internal_conf->create_uio_dev;
1370 }
1371
1372 enum rte_intr_mode
1373 rte_eal_vfio_intr_mode(void)
1374 {
1375         const struct internal_config *internal_conf =
1376                 eal_get_internal_configuration();
1377
1378         return internal_conf->vfio_intr_mode;
1379 }
1380
1381 void
1382 rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)
1383 {
1384         struct internal_config *cfg = eal_get_internal_configuration();
1385
1386         rte_uuid_copy(vf_token, cfg->vfio_vf_token);
1387 }
1388
1389 int
1390 rte_eal_check_module(const char *module_name)
1391 {
1392         char sysfs_mod_name[PATH_MAX];
1393         struct stat st;
1394         int n;
1395
1396         if (NULL == module_name)
1397                 return -1;
1398
1399         /* Check if there is sysfs mounted */
1400         if (stat("/sys/module", &st) != 0) {
1401                 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1402                         errno, strerror(errno));
1403                 return -1;
1404         }
1405
1406         /* A module might be built-in, therefore try sysfs */
1407         n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1408         if (n < 0 || n > PATH_MAX) {
1409                 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1410                 return -1;
1411         }
1412
1413         if (stat(sysfs_mod_name, &st) != 0) {
1414                 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1415                         sysfs_mod_name, errno, strerror(errno));
1416                 return 0;
1417         }
1418
1419         /* Module has been found */
1420         return 1;
1421 }