1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation.
3 * Copyright(c) 2012-2014 6WIND S.A.
23 #include <sys/queue.h>
25 #if defined(RTE_ARCH_X86)
28 #include <linux/version.h>
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>
36 #include <rte_errno.h>
37 #include <rte_per_lcore.h>
38 #include <rte_lcore.h>
39 #include <rte_service_component.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>
48 #include <rte_devargs.h>
49 #include <rte_version.h>
50 #include <malloc_heap.h>
52 #include <rte_telemetry.h>
54 #include "eal_private.h"
55 #include "eal_thread.h"
56 #include "eal_internal_cfg.h"
57 #include "eal_filesystem.h"
58 #include "eal_hugepages.h"
59 #include "eal_memcfg.h"
60 #include "eal_trace.h"
61 #include "eal_options.h"
63 #include "hotplug_mp.h"
65 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
67 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
69 #define KERNEL_IOMMU_GROUPS_PATH "/sys/kernel/iommu_groups"
71 /* define fd variable here, because file needs to be kept open for the
72 * duration of the program, as we hold a write lock on it in the primary proc */
73 static int mem_cfg_fd = -1;
75 static struct flock wr_lock = {
78 .l_start = offsetof(struct rte_mem_config, memsegs),
79 .l_len = RTE_SIZEOF_FIELD(struct rte_mem_config, memsegs),
82 /* internal configuration (per-core) */
83 struct lcore_config lcore_config[RTE_MAX_LCORE];
85 /* used by rte_rdtsc() */
86 int rte_cycles_vmware_tsc_map;
88 static const char *default_runtime_dir = "/var/run";
91 eal_create_runtime_dir(void)
93 const char *directory = default_runtime_dir;
94 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
95 const char *fallback = "/tmp";
96 char run_dir[PATH_MAX];
101 /* try XDG path first, fall back to /tmp */
102 if (xdg_runtime_dir != NULL)
103 directory = xdg_runtime_dir;
105 directory = fallback;
107 /* create DPDK subdirectory under runtime dir */
108 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
109 if (ret < 0 || ret == sizeof(tmp)) {
110 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
114 /* create prefix-specific subdirectory under DPDK runtime dir */
115 ret = snprintf(run_dir, sizeof(run_dir), "%s/%s",
116 tmp, eal_get_hugefile_prefix());
117 if (ret < 0 || ret == sizeof(run_dir)) {
118 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
122 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
125 ret = mkdir(tmp, 0700);
126 if (ret < 0 && errno != EEXIST) {
127 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
128 tmp, strerror(errno));
132 ret = mkdir(run_dir, 0700);
133 if (ret < 0 && errno != EEXIST) {
134 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
135 run_dir, strerror(errno));
139 if (eal_set_runtime_dir(run_dir, sizeof(run_dir)))
146 eal_clean_runtime_dir(void)
148 const char *runtime_dir = rte_eal_get_runtime_dir();
150 struct dirent *dirent;
151 int dir_fd, fd, lck_result;
152 static const char * const filters[] = {
158 dir = opendir(runtime_dir);
160 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
166 /* lock the directory before doing anything, to avoid races */
167 if (flock(dir_fd, LOCK_EX) < 0) {
168 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
173 dirent = readdir(dir);
175 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
180 while (dirent != NULL) {
184 /* skip files that don't match the patterns */
185 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
186 const char *filter = filters[f_idx];
188 if (fnmatch(filter, dirent->d_name, 0) == 0) {
194 dirent = readdir(dir);
198 /* try and lock the file */
199 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
201 /* skip to next file */
203 dirent = readdir(dir);
207 /* non-blocking lock */
208 lck_result = flock(fd, LOCK_EX | LOCK_NB);
210 /* if lock succeeds, remove the file */
211 if (lck_result != -1)
212 unlinkat(dir_fd, dirent->d_name, 0);
214 dirent = readdir(dir);
217 /* closedir closes dir_fd and drops the lock */
225 RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
231 /* parse a sysfs (or other) file containing one integer value */
233 eal_parse_sysfs_value(const char *filename, unsigned long *val)
239 if ((f = fopen(filename, "r")) == NULL) {
240 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
245 if (fgets(buf, sizeof(buf), f) == NULL) {
246 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
251 *val = strtoul(buf, &end, 0);
252 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
253 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
263 /* create memory configuration in shared/mmap memory. Take out
264 * a write lock on the memsegs, so we can auto-detect primary/secondary.
265 * This means we never close the file while running (auto-close on exit).
266 * We also don't lock the whole file, so that in future we can use read-locks
267 * on other parts, e.g. memzones, to detect if there are running secondary
270 rte_eal_config_create(void)
272 struct rte_config *config = rte_eal_get_configuration();
273 size_t page_sz = sysconf(_SC_PAGE_SIZE);
274 size_t cfg_len = sizeof(*config->mem_config);
275 size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
276 void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
278 const struct internal_config *internal_conf =
279 eal_get_internal_configuration();
281 const char *pathname = eal_runtime_config_path();
283 if (internal_conf->no_shconf)
286 /* map the config before hugepage address so that we don't waste a page */
287 if (internal_conf->base_virtaddr != 0)
288 rte_mem_cfg_addr = (void *)
289 RTE_ALIGN_FLOOR(internal_conf->base_virtaddr -
290 sizeof(struct rte_mem_config), page_sz);
292 rte_mem_cfg_addr = NULL;
295 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
296 if (mem_cfg_fd < 0) {
297 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
303 retval = ftruncate(mem_cfg_fd, cfg_len);
307 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
312 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
316 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
317 "process running?\n", pathname);
321 /* reserve space for config */
322 rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
323 &cfg_len_aligned, page_sz, 0, 0);
324 if (rte_mem_cfg_addr == NULL) {
325 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
331 /* remap the actual file into the space we've just reserved */
332 mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
333 cfg_len_aligned, PROT_READ | PROT_WRITE,
334 MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
335 if (mapped_mem_cfg_addr == MAP_FAILED) {
336 munmap(rte_mem_cfg_addr, cfg_len);
339 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
343 memcpy(rte_mem_cfg_addr, config->mem_config, sizeof(struct rte_mem_config));
344 config->mem_config = rte_mem_cfg_addr;
346 /* store address of the config in the config itself so that secondary
347 * processes could later map the config into this exact location
349 config->mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
350 config->mem_config->dma_maskbits = 0;
355 /* attach to an existing shared memory config */
357 rte_eal_config_attach(void)
359 struct rte_config *config = rte_eal_get_configuration();
360 struct rte_mem_config *mem_config;
361 const struct internal_config *internal_conf =
362 eal_get_internal_configuration();
364 const char *pathname = eal_runtime_config_path();
366 if (internal_conf->no_shconf)
370 mem_cfg_fd = open(pathname, O_RDWR);
371 if (mem_cfg_fd < 0) {
372 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
378 /* map it as read-only first */
379 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
380 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
381 if (mem_config == MAP_FAILED) {
384 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
385 errno, strerror(errno));
389 config->mem_config = mem_config;
394 /* reattach the shared config at exact memory location primary process has it */
396 rte_eal_config_reattach(void)
398 struct rte_config *config = rte_eal_get_configuration();
399 struct rte_mem_config *mem_config;
400 void *rte_mem_cfg_addr;
401 const struct internal_config *internal_conf =
402 eal_get_internal_configuration();
404 if (internal_conf->no_shconf)
407 /* save the address primary process has mapped shared config to */
409 (void *) (uintptr_t) config->mem_config->mem_cfg_addr;
411 /* unmap original config */
412 munmap(config->mem_config, sizeof(struct rte_mem_config));
414 /* remap the config at proper address */
415 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
416 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
422 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
423 if (mem_config != MAP_FAILED) {
424 /* errno is stale, don't use */
425 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
426 " - please use '--" OPT_BASE_VIRTADDR
427 "' option\n", rte_mem_cfg_addr, mem_config);
428 munmap(mem_config, sizeof(struct rte_mem_config));
431 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
432 errno, strerror(errno));
436 config->mem_config = mem_config;
441 /* Detect if we are a primary or a secondary process */
443 eal_proc_type_detect(void)
445 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
446 const char *pathname = eal_runtime_config_path();
447 const struct internal_config *internal_conf =
448 eal_get_internal_configuration();
450 /* if there no shared config, there can be no secondary processes */
451 if (!internal_conf->no_shconf) {
452 /* if we can open the file but not get a write-lock we are a
453 * secondary process. NOTE: if we get a file handle back, we
454 * keep that open and don't close it to prevent a race condition
455 * between multiple opens.
457 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
458 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
459 ptype = RTE_PROC_SECONDARY;
462 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
463 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
468 /* Sets up rte_config structure with the pointer to shared memory config.*/
470 rte_config_init(void)
472 struct rte_config *config = rte_eal_get_configuration();
473 const struct internal_config *internal_conf =
474 eal_get_internal_configuration();
476 config->process_type = internal_conf->process_type;
478 switch (config->process_type) {
479 case RTE_PROC_PRIMARY:
480 if (rte_eal_config_create() < 0)
482 eal_mcfg_update_from_internal();
484 case RTE_PROC_SECONDARY:
485 if (rte_eal_config_attach() < 0)
487 eal_mcfg_wait_complete();
488 if (eal_mcfg_check_version() < 0) {
489 RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
492 if (rte_eal_config_reattach() < 0)
494 if (!__rte_mp_enable()) {
495 RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
498 eal_mcfg_update_internal();
501 case RTE_PROC_INVALID:
502 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
503 config->process_type);
510 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
512 eal_hugedirs_unlock(void)
515 struct internal_config *internal_conf =
516 eal_get_internal_configuration();
518 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
520 /* skip uninitialized */
521 if (internal_conf->hugepage_info[i].lock_descriptor < 0)
523 /* unlock hugepage file */
524 flock(internal_conf->hugepage_info[i].lock_descriptor, LOCK_UN);
525 close(internal_conf->hugepage_info[i].lock_descriptor);
526 /* reset the field */
527 internal_conf->hugepage_info[i].lock_descriptor = -1;
533 eal_usage(const char *prgname)
535 rte_usage_hook_t hook = eal_get_application_usage_hook();
537 printf("\nUsage: %s ", prgname);
539 printf("EAL Linux options:\n"
540 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
541 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
542 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
543 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
544 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
545 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
546 " --"OPT_VFIO_VF_TOKEN" VF token (UUID) shared between SR-IOV PF and VFs\n"
547 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
548 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
549 " --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
551 /* Allow the application to print its usage message too if hook is set */
553 printf("===== Application Usage =====\n\n");
559 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
561 char * arg[RTE_MAX_NUMA_NODES];
564 uint64_t total_mem = 0;
566 len = strnlen(strval, SOCKET_MEM_STRLEN);
567 if (len == SOCKET_MEM_STRLEN) {
568 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
572 /* all other error cases will be caught later */
573 if (!isdigit(strval[len-1]))
576 /* split the optarg into separate socket values */
577 arg_num = rte_strsplit(strval, len,
578 arg, RTE_MAX_NUMA_NODES, ',');
580 /* if split failed, or 0 arguments */
584 /* parse each defined socket option */
586 for (i = 0; i < arg_num; i++) {
589 val = strtoull(arg[i], &end, 10);
591 /* check for invalid input */
593 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
604 eal_parse_vfio_intr(const char *mode)
606 struct internal_config *internal_conf =
607 eal_get_internal_configuration();
611 enum rte_intr_mode value;
613 { "legacy", RTE_INTR_MODE_LEGACY },
614 { "msi", RTE_INTR_MODE_MSI },
615 { "msix", RTE_INTR_MODE_MSIX },
618 for (i = 0; i < RTE_DIM(map); i++) {
619 if (!strcmp(mode, map[i].name)) {
620 internal_conf->vfio_intr_mode = map[i].value;
628 eal_parse_vfio_vf_token(const char *vf_token)
630 struct internal_config *cfg = eal_get_internal_configuration();
633 if (!rte_uuid_parse(vf_token, uuid)) {
634 rte_uuid_copy(cfg->vfio_vf_token, uuid);
641 /* Parse the arguments for --log-level only */
643 eal_log_level_parse(int argc, char **argv)
648 const int old_optind = optind;
649 const int old_optopt = optopt;
650 char * const old_optarg = optarg;
651 struct internal_config *internal_conf =
652 eal_get_internal_configuration();
657 while ((opt = getopt_long(argc, argvopt, eal_short_options,
658 eal_long_options, &option_index)) != EOF) {
662 /* getopt is not happy, stop right now */
666 ret = (opt == OPT_LOG_LEVEL_NUM) ?
667 eal_parse_common_option(opt, optarg, internal_conf) : 0;
669 /* common parser is not happy */
674 /* restore getopt lib */
680 /* Parse the argument given in the command line of the application */
682 eal_parse_args(int argc, char **argv)
687 char *prgname = argv[0];
688 const int old_optind = optind;
689 const int old_optopt = optopt;
690 char * const old_optarg = optarg;
691 struct internal_config *internal_conf =
692 eal_get_internal_configuration();
697 while ((opt = getopt_long(argc, argvopt, eal_short_options,
698 eal_long_options, &option_index)) != EOF) {
700 /* getopt didn't recognise the option */
707 ret = eal_parse_common_option(opt, optarg, internal_conf);
708 /* common parser is not happy */
714 /* common parser handled this option */
723 case OPT_HUGE_DIR_NUM:
725 char *hdir = strdup(optarg);
727 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
729 /* free old hugepage dir */
730 if (internal_conf->hugepage_dir != NULL)
731 free(internal_conf->hugepage_dir);
732 internal_conf->hugepage_dir = hdir;
736 case OPT_FILE_PREFIX_NUM:
738 char *prefix = strdup(optarg);
740 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
742 /* free old prefix */
743 if (internal_conf->hugefile_prefix != NULL)
744 free(internal_conf->hugefile_prefix);
745 internal_conf->hugefile_prefix = prefix;
749 case OPT_SOCKET_MEM_NUM:
750 if (eal_parse_socket_arg(optarg,
751 internal_conf->socket_mem) < 0) {
752 RTE_LOG(ERR, EAL, "invalid parameters for --"
753 OPT_SOCKET_MEM "\n");
758 internal_conf->force_sockets = 1;
761 case OPT_SOCKET_LIMIT_NUM:
762 if (eal_parse_socket_arg(optarg,
763 internal_conf->socket_limit) < 0) {
764 RTE_LOG(ERR, EAL, "invalid parameters for --"
765 OPT_SOCKET_LIMIT "\n");
770 internal_conf->force_socket_limits = 1;
773 case OPT_VFIO_INTR_NUM:
774 if (eal_parse_vfio_intr(optarg) < 0) {
775 RTE_LOG(ERR, EAL, "invalid parameters for --"
783 case OPT_VFIO_VF_TOKEN_NUM:
784 if (eal_parse_vfio_vf_token(optarg) < 0) {
785 RTE_LOG(ERR, EAL, "invalid parameters for --"
786 OPT_VFIO_VF_TOKEN "\n");
793 case OPT_CREATE_UIO_DEV_NUM:
794 internal_conf->create_uio_dev = 1;
797 case OPT_MBUF_POOL_OPS_NAME_NUM:
799 char *ops_name = strdup(optarg);
800 if (ops_name == NULL)
801 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
803 /* free old ops name */
804 if (internal_conf->user_mbuf_pool_ops_name !=
806 free(internal_conf->user_mbuf_pool_ops_name);
808 internal_conf->user_mbuf_pool_ops_name =
813 case OPT_MATCH_ALLOCATIONS_NUM:
814 internal_conf->match_allocations = 1;
818 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
819 RTE_LOG(ERR, EAL, "Option %c is not supported "
821 } else if (opt >= OPT_LONG_MIN_NUM &&
822 opt < OPT_LONG_MAX_NUM) {
823 RTE_LOG(ERR, EAL, "Option %s is not supported "
825 eal_long_options[option_index].name);
827 RTE_LOG(ERR, EAL, "Option %d is not supported "
836 /* create runtime data directory */
837 if (internal_conf->no_shconf == 0 &&
838 eal_create_runtime_dir() < 0) {
839 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
844 if (eal_adjust_config(internal_conf) != 0) {
850 if (eal_check_common_options(internal_conf) != 0) {
857 argv[optind-1] = prgname;
861 /* restore getopt lib */
870 check_socket(const struct rte_memseg_list *msl, void *arg)
872 int *socket_id = arg;
877 return *socket_id == msl->socket_id;
881 eal_check_mem_on_local_socket(void)
884 const struct rte_config *config = rte_eal_get_configuration();
886 socket_id = rte_lcore_to_socket_id(config->main_lcore);
888 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
889 RTE_LOG(WARNING, EAL, "WARNING: Main core has no memory on local socket!\n");
893 sync_func(__rte_unused void *arg)
899 * Request iopl privilege for all RPL, returns 0 on success
900 * iopl() call is mostly for the i386 architecture. For other architectures,
901 * return -1 to indicate IO privilege can't be changed in this way.
904 rte_eal_iopl_init(void)
906 #if defined(RTE_ARCH_X86)
914 static int rte_eal_vfio_setup(void)
916 if (rte_vfio_enable("vfio"))
923 static void rte_eal_init_alert(const char *msg)
925 fprintf(stderr, "EAL: FATAL: %s\n", msg);
926 RTE_LOG(ERR, EAL, "%s\n", msg);
930 * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
931 * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
932 * IOMMU groups. If IOMMU is not enabled, that path would be empty.
933 * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
936 is_iommu_enabled(void)
938 DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
942 /* if directory doesn't exist, assume IOMMU is not enabled */
946 while ((d = readdir(dir)) != NULL) {
947 /* skip dot and dot-dot */
956 /* Launch threads, called at application init(). */
958 rte_eal_init(int argc, char **argv)
962 static uint32_t run_once;
963 uint32_t has_run = 0;
965 static char logid[PATH_MAX];
966 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
967 char thread_name[RTE_MAX_THREAD_NAME_LEN];
969 const struct rte_config *config = rte_eal_get_configuration();
970 struct internal_config *internal_conf =
971 eal_get_internal_configuration();
973 /* checks if the machine is adequate */
974 if (!rte_cpu_is_supported()) {
975 rte_eal_init_alert("unsupported cpu type.");
980 if (!__atomic_compare_exchange_n(&run_once, &has_run, 1, 0,
981 __ATOMIC_RELAXED, __ATOMIC_RELAXED)) {
982 rte_eal_init_alert("already called initialization.");
983 rte_errno = EALREADY;
987 p = strrchr(argv[0], '/');
988 strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
989 thread_id = pthread_self();
991 eal_reset_internal_config(internal_conf);
993 /* set log level as early as possible */
994 eal_log_level_parse(argc, argv);
996 /* clone argv to report out later in telemetry */
997 eal_save_args(argc, argv);
999 if (rte_eal_cpu_init() < 0) {
1000 rte_eal_init_alert("Cannot detect lcores.");
1001 rte_errno = ENOTSUP;
1005 fctret = eal_parse_args(argc, argv);
1007 rte_eal_init_alert("Invalid 'command line' arguments.");
1009 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1013 if (eal_plugins_init() < 0) {
1014 rte_eal_init_alert("Cannot init plugins");
1016 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1020 if (eal_trace_init() < 0) {
1021 rte_eal_init_alert("Cannot init trace");
1026 if (eal_option_device_parse()) {
1028 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1032 if (rte_config_init() < 0) {
1033 rte_eal_init_alert("Cannot init config");
1037 if (rte_eal_intr_init() < 0) {
1038 rte_eal_init_alert("Cannot init interrupt-handling thread");
1042 if (rte_eal_alarm_init() < 0) {
1043 rte_eal_init_alert("Cannot init alarm");
1044 /* rte_eal_alarm_init sets rte_errno on failure. */
1048 /* Put mp channel init before bus scan so that we can init the vdev
1049 * bus through mp channel in the secondary process before the bus scan.
1051 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
1052 rte_eal_init_alert("failed to init mp channel");
1053 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1059 /* register multi-process action callbacks for hotplug */
1060 if (eal_mp_dev_hotplug_init() < 0) {
1061 rte_eal_init_alert("failed to register mp callback for hotplug");
1065 if (rte_bus_scan()) {
1066 rte_eal_init_alert("Cannot scan the buses for devices");
1068 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1072 phys_addrs = rte_eal_using_phys_addrs() != 0;
1074 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
1075 if (internal_conf->iova_mode == RTE_IOVA_DC) {
1076 /* autodetect the IOVA mapping mode */
1077 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
1079 if (iova_mode == RTE_IOVA_DC) {
1080 RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
1083 /* if we have no access to physical addresses,
1084 * pick IOVA as VA mode.
1086 iova_mode = RTE_IOVA_VA;
1087 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
1088 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
1089 } else if (rte_eal_check_module("rte_kni") == 1) {
1090 iova_mode = RTE_IOVA_PA;
1091 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
1093 } else if (is_iommu_enabled()) {
1094 /* we have an IOMMU, pick IOVA as VA mode */
1095 iova_mode = RTE_IOVA_VA;
1096 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
1098 /* physical addresses available, and no IOMMU
1099 * found, so pick IOVA as PA.
1101 iova_mode = RTE_IOVA_PA;
1102 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1105 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1106 /* Workaround for KNI which requires physical address to work
1109 if (iova_mode == RTE_IOVA_VA &&
1110 rte_eal_check_module("rte_kni") == 1) {
1112 iova_mode = RTE_IOVA_PA;
1113 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1115 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1119 rte_eal_get_configuration()->iova_mode = iova_mode;
1121 rte_eal_get_configuration()->iova_mode =
1122 internal_conf->iova_mode;
1125 if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1126 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1131 RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1132 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1134 if (internal_conf->no_hugetlbfs == 0) {
1135 /* rte_config isn't initialized yet */
1136 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
1137 eal_hugepage_info_init() :
1138 eal_hugepage_info_read();
1140 rte_eal_init_alert("Cannot get hugepage information.");
1142 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1147 if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
1148 if (internal_conf->no_hugetlbfs)
1149 internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
1152 if (internal_conf->vmware_tsc_map == 1) {
1153 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1154 rte_cycles_vmware_tsc_map = 1;
1155 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1156 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1158 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1159 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1163 if (rte_eal_log_init(logid, internal_conf->syslog_facility) < 0) {
1164 rte_eal_init_alert("Cannot init logging.");
1166 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1171 if (rte_eal_vfio_setup() < 0) {
1172 rte_eal_init_alert("Cannot init VFIO");
1174 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1178 /* in secondary processes, memory init may allocate additional fbarrays
1179 * not present in primary processes, so to avoid any potential issues,
1180 * initialize memzones first.
1182 if (rte_eal_memzone_init() < 0) {
1183 rte_eal_init_alert("Cannot init memzone");
1188 if (rte_eal_memory_init() < 0) {
1189 rte_eal_init_alert("Cannot init memory");
1194 /* the directories are locked during eal_hugepage_info_init */
1195 eal_hugedirs_unlock();
1197 if (rte_eal_malloc_heap_init() < 0) {
1198 rte_eal_init_alert("Cannot init malloc heap");
1203 if (rte_eal_tailqs_init() < 0) {
1204 rte_eal_init_alert("Cannot init tail queues for objects");
1209 if (rte_eal_timer_init() < 0) {
1210 rte_eal_init_alert("Cannot init HPET or TSC timers");
1211 rte_errno = ENOTSUP;
1215 eal_check_mem_on_local_socket();
1217 if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
1218 &lcore_config[config->main_lcore].cpuset) != 0) {
1219 rte_eal_init_alert("Cannot set affinity");
1223 __rte_thread_init(config->main_lcore,
1224 &lcore_config[config->main_lcore].cpuset);
1226 ret = eal_thread_dump_current_affinity(cpuset, sizeof(cpuset));
1227 RTE_LOG(DEBUG, EAL, "Main lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1228 config->main_lcore, (uintptr_t)thread_id, cpuset,
1229 ret == 0 ? "" : "...");
1231 RTE_LCORE_FOREACH_WORKER(i) {
1234 * create communication pipes between main thread
1237 if (pipe(lcore_config[i].pipe_main2worker) < 0)
1238 rte_panic("Cannot create pipe\n");
1239 if (pipe(lcore_config[i].pipe_worker2main) < 0)
1240 rte_panic("Cannot create pipe\n");
1242 lcore_config[i].state = WAIT;
1244 /* create a thread for each lcore */
1245 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1246 eal_thread_loop, NULL);
1248 rte_panic("Cannot create thread\n");
1250 /* Set thread_name for aid in debugging. */
1251 snprintf(thread_name, sizeof(thread_name),
1252 "lcore-worker-%d", i);
1253 ret = rte_thread_setname(lcore_config[i].thread_id,
1257 "Cannot set name for lcore thread\n");
1259 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
1260 sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
1262 rte_panic("Cannot set affinity\n");
1266 * Launch a dummy function on all worker lcores, so that main lcore
1267 * knows they are all ready when this function returns.
1269 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MAIN);
1270 rte_eal_mp_wait_lcore();
1272 /* initialize services so vdevs register service during bus_probe. */
1273 ret = rte_service_init();
1275 rte_eal_init_alert("rte_service_init() failed");
1276 rte_errno = ENOEXEC;
1280 /* Probe all the buses and devices/drivers on them */
1281 if (rte_bus_probe()) {
1282 rte_eal_init_alert("Cannot probe devices");
1283 rte_errno = ENOTSUP;
1288 /* Register mp action after probe() so that we got enough info */
1289 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1293 /* initialize default service/lcore mappings and start running. Ignore
1294 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1296 ret = rte_service_start_with_defaults();
1297 if (ret < 0 && ret != -ENOTSUP) {
1298 rte_errno = ENOEXEC;
1303 * Clean up unused files in runtime directory. We do this at the end of
1304 * init and not at the beginning because we want to clean stuff up
1305 * whether we are primary or secondary process, but we cannot remove
1306 * primary process' files because secondary should be able to run even
1307 * if primary process is dead.
1309 * In no_shconf mode, no runtime directory is created in the first
1310 * place, so no cleanup needed.
1312 if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
1313 rte_eal_init_alert("Cannot clear runtime directory");
1316 if (!internal_conf->no_telemetry) {
1317 const char *error_str = NULL;
1318 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1319 &internal_conf->ctrl_cpuset, &error_str)
1321 rte_eal_init_alert(error_str);
1324 if (error_str != NULL)
1325 RTE_LOG(NOTICE, EAL, "%s\n", error_str);
1328 eal_mcfg_complete();
1334 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1335 void *arg __rte_unused)
1337 /* ms is const, so find this memseg */
1338 struct rte_memseg *found;
1343 found = rte_mem_virt2memseg(ms->addr, msl);
1345 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1351 rte_eal_cleanup(void)
1353 /* if we're in a primary process, we need to mark hugepages as freeable
1354 * so that finalization can release them back to the system.
1356 struct internal_config *internal_conf =
1357 eal_get_internal_configuration();
1359 if (rte_eal_process_type() == RTE_PROC_PRIMARY)
1360 rte_memseg_walk(mark_freeable, NULL);
1361 rte_service_finalize();
1362 rte_mp_channel_cleanup();
1365 eal_cleanup_config(internal_conf);
1369 int rte_eal_create_uio_dev(void)
1371 const struct internal_config *internal_conf =
1372 eal_get_internal_configuration();
1374 return internal_conf->create_uio_dev;
1378 rte_eal_vfio_intr_mode(void)
1380 const struct internal_config *internal_conf =
1381 eal_get_internal_configuration();
1383 return internal_conf->vfio_intr_mode;
1387 rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)
1389 struct internal_config *cfg = eal_get_internal_configuration();
1391 rte_uuid_copy(vf_token, cfg->vfio_vf_token);
1395 rte_eal_check_module(const char *module_name)
1397 char sysfs_mod_name[PATH_MAX];
1401 if (NULL == module_name)
1404 /* Check if there is sysfs mounted */
1405 if (stat("/sys/module", &st) != 0) {
1406 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1407 errno, strerror(errno));
1411 /* A module might be built-in, therefore try sysfs */
1412 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1413 if (n < 0 || n > PATH_MAX) {
1414 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1418 if (stat(sysfs_mod_name, &st) != 0) {
1419 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1420 sysfs_mod_name, errno, strerror(errno));
1424 /* Module has been found */