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>
53 #include <telemetry_internal.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"
62 #include "eal_options.h"
64 #include "hotplug_mp.h"
66 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
68 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
70 #define KERNEL_IOMMU_GROUPS_PATH "/sys/kernel/iommu_groups"
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;
76 static struct flock wr_lock = {
79 .l_start = offsetof(struct rte_mem_config, memsegs),
80 .l_len = RTE_SIZEOF_FIELD(struct rte_mem_config, memsegs),
83 /* internal configuration (per-core) */
84 struct lcore_config lcore_config[RTE_MAX_LCORE];
86 /* used by rte_rdtsc() */
87 int rte_cycles_vmware_tsc_map;
90 eal_create_runtime_dir(void)
92 const char *directory;
93 char run_dir[PATH_MAX];
97 /* from RuntimeDirectory= see systemd.exec */
98 directory = getenv("RUNTIME_DIRECTORY");
99 if (directory == NULL) {
101 * Used standard convention defined in
102 * XDG Base Directory Specification and
103 * Filesystem Hierarchy Standard.
106 directory = "/var/run";
108 directory = getenv("XDG_RUNTIME_DIR") ? : "/tmp";
111 /* create DPDK subdirectory under runtime dir */
112 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
113 if (ret < 0 || ret == sizeof(tmp)) {
114 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
118 /* create prefix-specific subdirectory under DPDK runtime dir */
119 ret = snprintf(run_dir, sizeof(run_dir), "%s/%s",
120 tmp, eal_get_hugefile_prefix());
121 if (ret < 0 || ret == sizeof(run_dir)) {
122 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
126 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
129 ret = mkdir(tmp, 0700);
130 if (ret < 0 && errno != EEXIST) {
131 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
132 tmp, strerror(errno));
136 ret = mkdir(run_dir, 0700);
137 if (ret < 0 && errno != EEXIST) {
138 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
139 run_dir, strerror(errno));
143 if (eal_set_runtime_dir(run_dir))
150 eal_clean_runtime_dir(void)
152 const char *runtime_dir = rte_eal_get_runtime_dir();
154 struct dirent *dirent;
155 int dir_fd, fd, lck_result;
156 static const char * const filters[] = {
162 dir = opendir(runtime_dir);
164 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
170 /* lock the directory before doing anything, to avoid races */
171 if (flock(dir_fd, LOCK_EX) < 0) {
172 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
177 dirent = readdir(dir);
179 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
184 while (dirent != NULL) {
188 /* skip files that don't match the patterns */
189 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
190 const char *filter = filters[f_idx];
192 if (fnmatch(filter, dirent->d_name, 0) == 0) {
198 dirent = readdir(dir);
202 /* try and lock the file */
203 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
205 /* skip to next file */
207 dirent = readdir(dir);
211 /* non-blocking lock */
212 lck_result = flock(fd, LOCK_EX | LOCK_NB);
214 /* if lock succeeds, remove the file */
215 if (lck_result != -1)
216 unlinkat(dir_fd, dirent->d_name, 0);
218 dirent = readdir(dir);
221 /* closedir closes dir_fd and drops the lock */
229 RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
235 /* parse a sysfs (or other) file containing one integer value */
237 eal_parse_sysfs_value(const char *filename, unsigned long *val)
243 if ((f = fopen(filename, "r")) == NULL) {
244 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
249 if (fgets(buf, sizeof(buf), f) == NULL) {
250 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
255 *val = strtoul(buf, &end, 0);
256 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
257 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
267 /* create memory configuration in shared/mmap memory. Take out
268 * a write lock on the memsegs, so we can auto-detect primary/secondary.
269 * This means we never close the file while running (auto-close on exit).
270 * We also don't lock the whole file, so that in future we can use read-locks
271 * on other parts, e.g. memzones, to detect if there are running secondary
274 rte_eal_config_create(void)
276 struct rte_config *config = rte_eal_get_configuration();
277 size_t page_sz = sysconf(_SC_PAGE_SIZE);
278 size_t cfg_len = sizeof(*config->mem_config);
279 size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
280 void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
282 const struct internal_config *internal_conf =
283 eal_get_internal_configuration();
285 const char *pathname = eal_runtime_config_path();
287 if (internal_conf->no_shconf)
290 /* map the config before hugepage address so that we don't waste a page */
291 if (internal_conf->base_virtaddr != 0)
292 rte_mem_cfg_addr = (void *)
293 RTE_ALIGN_FLOOR(internal_conf->base_virtaddr -
294 sizeof(struct rte_mem_config), page_sz);
296 rte_mem_cfg_addr = NULL;
299 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
300 if (mem_cfg_fd < 0) {
301 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
307 retval = ftruncate(mem_cfg_fd, cfg_len);
311 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
316 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
320 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
321 "process running?\n", pathname);
325 /* reserve space for config */
326 rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
327 &cfg_len_aligned, page_sz, 0, 0);
328 if (rte_mem_cfg_addr == NULL) {
329 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
335 /* remap the actual file into the space we've just reserved */
336 mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
337 cfg_len_aligned, PROT_READ | PROT_WRITE,
338 MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
339 if (mapped_mem_cfg_addr == MAP_FAILED) {
340 munmap(rte_mem_cfg_addr, cfg_len);
343 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
347 memcpy(rte_mem_cfg_addr, config->mem_config, sizeof(struct rte_mem_config));
348 config->mem_config = rte_mem_cfg_addr;
350 /* store address of the config in the config itself so that secondary
351 * processes could later map the config into this exact location
353 config->mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
354 config->mem_config->dma_maskbits = 0;
359 /* attach to an existing shared memory config */
361 rte_eal_config_attach(void)
363 struct rte_config *config = rte_eal_get_configuration();
364 struct rte_mem_config *mem_config;
365 const struct internal_config *internal_conf =
366 eal_get_internal_configuration();
368 const char *pathname = eal_runtime_config_path();
370 if (internal_conf->no_shconf)
374 mem_cfg_fd = open(pathname, O_RDWR);
375 if (mem_cfg_fd < 0) {
376 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
382 /* map it as read-only first */
383 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
384 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
385 if (mem_config == MAP_FAILED) {
388 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
389 errno, strerror(errno));
393 config->mem_config = mem_config;
398 /* reattach the shared config at exact memory location primary process has it */
400 rte_eal_config_reattach(void)
402 struct rte_config *config = rte_eal_get_configuration();
403 struct rte_mem_config *mem_config;
404 void *rte_mem_cfg_addr;
405 const struct internal_config *internal_conf =
406 eal_get_internal_configuration();
408 if (internal_conf->no_shconf)
411 /* save the address primary process has mapped shared config to */
413 (void *) (uintptr_t) config->mem_config->mem_cfg_addr;
415 /* unmap original config */
416 munmap(config->mem_config, sizeof(struct rte_mem_config));
418 /* remap the config at proper address */
419 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
420 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
426 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
427 if (mem_config != MAP_FAILED) {
428 /* errno is stale, don't use */
429 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
430 " - please use '--" OPT_BASE_VIRTADDR
431 "' option\n", rte_mem_cfg_addr, mem_config);
432 munmap(mem_config, sizeof(struct rte_mem_config));
435 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
436 errno, strerror(errno));
440 config->mem_config = mem_config;
445 /* Detect if we are a primary or a secondary process */
447 eal_proc_type_detect(void)
449 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
450 const char *pathname = eal_runtime_config_path();
451 const struct internal_config *internal_conf =
452 eal_get_internal_configuration();
454 /* if there no shared config, there can be no secondary processes */
455 if (!internal_conf->no_shconf) {
456 /* if we can open the file but not get a write-lock we are a
457 * secondary process. NOTE: if we get a file handle back, we
458 * keep that open and don't close it to prevent a race condition
459 * between multiple opens.
461 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
462 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
463 ptype = RTE_PROC_SECONDARY;
466 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
467 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
472 /* Sets up rte_config structure with the pointer to shared memory config.*/
474 rte_config_init(void)
476 struct rte_config *config = rte_eal_get_configuration();
477 const struct internal_config *internal_conf =
478 eal_get_internal_configuration();
480 config->process_type = internal_conf->process_type;
482 switch (config->process_type) {
483 case RTE_PROC_PRIMARY:
484 if (rte_eal_config_create() < 0)
486 eal_mcfg_update_from_internal();
488 case RTE_PROC_SECONDARY:
489 if (rte_eal_config_attach() < 0)
491 eal_mcfg_wait_complete();
492 if (eal_mcfg_check_version() < 0) {
493 RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
496 if (rte_eal_config_reattach() < 0)
498 if (!__rte_mp_enable()) {
499 RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
502 eal_mcfg_update_internal();
505 case RTE_PROC_INVALID:
506 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
507 config->process_type);
514 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
516 eal_hugedirs_unlock(void)
519 struct internal_config *internal_conf =
520 eal_get_internal_configuration();
522 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
524 /* skip uninitialized */
525 if (internal_conf->hugepage_info[i].lock_descriptor < 0)
527 /* unlock hugepage file */
528 flock(internal_conf->hugepage_info[i].lock_descriptor, LOCK_UN);
529 close(internal_conf->hugepage_info[i].lock_descriptor);
530 /* reset the field */
531 internal_conf->hugepage_info[i].lock_descriptor = -1;
537 eal_usage(const char *prgname)
539 rte_usage_hook_t hook = eal_get_application_usage_hook();
541 printf("\nUsage: %s ", prgname);
543 printf("EAL Linux options:\n"
544 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
545 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
546 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
547 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
548 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
549 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
550 " --"OPT_VFIO_VF_TOKEN" VF token (UUID) shared between SR-IOV PF and VFs\n"
551 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
552 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
553 " --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
555 /* Allow the application to print its usage message too if hook is set */
557 printf("===== Application Usage =====\n\n");
563 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
565 char * arg[RTE_MAX_NUMA_NODES];
569 len = strnlen(strval, SOCKET_MEM_STRLEN);
570 if (len == SOCKET_MEM_STRLEN) {
571 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
575 /* all other error cases will be caught later */
576 if (!isdigit(strval[len-1]))
579 /* split the optarg into separate socket values */
580 arg_num = rte_strsplit(strval, len,
581 arg, RTE_MAX_NUMA_NODES, ',');
583 /* if split failed, or 0 arguments */
587 /* parse each defined socket option */
589 for (i = 0; i < arg_num; i++) {
592 val = strtoull(arg[i], &end, 10);
594 /* check for invalid input */
596 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
606 eal_parse_vfio_intr(const char *mode)
608 struct internal_config *internal_conf =
609 eal_get_internal_configuration();
613 enum rte_intr_mode value;
615 { "legacy", RTE_INTR_MODE_LEGACY },
616 { "msi", RTE_INTR_MODE_MSI },
617 { "msix", RTE_INTR_MODE_MSIX },
620 for (i = 0; i < RTE_DIM(map); i++) {
621 if (!strcmp(mode, map[i].name)) {
622 internal_conf->vfio_intr_mode = map[i].value;
630 eal_parse_vfio_vf_token(const char *vf_token)
632 struct internal_config *cfg = eal_get_internal_configuration();
635 if (!rte_uuid_parse(vf_token, uuid)) {
636 rte_uuid_copy(cfg->vfio_vf_token, uuid);
643 /* Parse the arguments for --log-level only */
645 eal_log_level_parse(int argc, char **argv)
650 const int old_optind = optind;
651 const int old_optopt = optopt;
652 char * const old_optarg = optarg;
653 struct internal_config *internal_conf =
654 eal_get_internal_configuration();
659 while ((opt = getopt_long(argc, argvopt, eal_short_options,
660 eal_long_options, &option_index)) != EOF) {
664 /* getopt is not happy, stop right now */
668 ret = (opt == OPT_LOG_LEVEL_NUM) ?
669 eal_parse_common_option(opt, optarg, internal_conf) : 0;
671 /* common parser is not happy */
676 /* restore getopt lib */
682 /* Parse the argument given in the command line of the application */
684 eal_parse_args(int argc, char **argv)
689 char *prgname = argv[0];
690 const int old_optind = optind;
691 const int old_optopt = optopt;
692 char * const old_optarg = optarg;
693 struct internal_config *internal_conf =
694 eal_get_internal_configuration();
699 while ((opt = getopt_long(argc, argvopt, eal_short_options,
700 eal_long_options, &option_index)) != EOF) {
702 /* getopt didn't recognise the option */
709 /* eal_log_level_parse() already handled this option */
710 if (opt == OPT_LOG_LEVEL_NUM)
713 ret = eal_parse_common_option(opt, optarg, internal_conf);
714 /* common parser is not happy */
720 /* common parser handled this option */
729 case OPT_HUGE_DIR_NUM:
731 char *hdir = strdup(optarg);
733 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
735 /* free old hugepage dir */
736 if (internal_conf->hugepage_dir != NULL)
737 free(internal_conf->hugepage_dir);
738 internal_conf->hugepage_dir = hdir;
742 case OPT_FILE_PREFIX_NUM:
744 char *prefix = strdup(optarg);
746 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
748 /* free old prefix */
749 if (internal_conf->hugefile_prefix != NULL)
750 free(internal_conf->hugefile_prefix);
751 internal_conf->hugefile_prefix = prefix;
755 case OPT_SOCKET_MEM_NUM:
756 if (eal_parse_socket_arg(optarg,
757 internal_conf->socket_mem) < 0) {
758 RTE_LOG(ERR, EAL, "invalid parameters for --"
759 OPT_SOCKET_MEM "\n");
764 internal_conf->force_sockets = 1;
767 case OPT_SOCKET_LIMIT_NUM:
768 if (eal_parse_socket_arg(optarg,
769 internal_conf->socket_limit) < 0) {
770 RTE_LOG(ERR, EAL, "invalid parameters for --"
771 OPT_SOCKET_LIMIT "\n");
776 internal_conf->force_socket_limits = 1;
779 case OPT_VFIO_INTR_NUM:
780 if (eal_parse_vfio_intr(optarg) < 0) {
781 RTE_LOG(ERR, EAL, "invalid parameters for --"
789 case OPT_VFIO_VF_TOKEN_NUM:
790 if (eal_parse_vfio_vf_token(optarg) < 0) {
791 RTE_LOG(ERR, EAL, "invalid parameters for --"
792 OPT_VFIO_VF_TOKEN "\n");
799 case OPT_CREATE_UIO_DEV_NUM:
800 internal_conf->create_uio_dev = 1;
803 case OPT_MBUF_POOL_OPS_NAME_NUM:
805 char *ops_name = strdup(optarg);
806 if (ops_name == NULL)
807 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
809 /* free old ops name */
810 if (internal_conf->user_mbuf_pool_ops_name !=
812 free(internal_conf->user_mbuf_pool_ops_name);
814 internal_conf->user_mbuf_pool_ops_name =
819 case OPT_MATCH_ALLOCATIONS_NUM:
820 internal_conf->match_allocations = 1;
824 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
825 RTE_LOG(ERR, EAL, "Option %c is not supported "
827 } else if (opt >= OPT_LONG_MIN_NUM &&
828 opt < OPT_LONG_MAX_NUM) {
829 RTE_LOG(ERR, EAL, "Option %s is not supported "
831 eal_long_options[option_index].name);
833 RTE_LOG(ERR, EAL, "Option %d is not supported "
842 /* create runtime data directory. In no_shconf mode, skip any errors */
843 if (eal_create_runtime_dir() < 0) {
844 if (internal_conf->no_shconf == 0) {
845 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
849 RTE_LOG(WARNING, EAL, "No DPDK runtime directory created\n");
852 if (eal_adjust_config(internal_conf) != 0) {
858 if (eal_check_common_options(internal_conf) != 0) {
865 argv[optind-1] = prgname;
869 /* restore getopt lib */
878 check_socket(const struct rte_memseg_list *msl, void *arg)
880 int *socket_id = arg;
885 return *socket_id == msl->socket_id;
889 eal_check_mem_on_local_socket(void)
892 const struct rte_config *config = rte_eal_get_configuration();
894 socket_id = rte_lcore_to_socket_id(config->main_lcore);
896 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
897 RTE_LOG(WARNING, EAL, "WARNING: Main core has no memory on local socket!\n");
901 sync_func(__rte_unused void *arg)
907 * Request iopl privilege for all RPL, returns 0 on success
908 * iopl() call is mostly for the i386 architecture. For other architectures,
909 * return -1 to indicate IO privilege can't be changed in this way.
912 rte_eal_iopl_init(void)
914 #if defined(RTE_ARCH_X86)
922 static int rte_eal_vfio_setup(void)
924 if (rte_vfio_enable("vfio"))
931 static void rte_eal_init_alert(const char *msg)
933 fprintf(stderr, "EAL: FATAL: %s\n", msg);
934 RTE_LOG(ERR, EAL, "%s\n", msg);
938 * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
939 * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
940 * IOMMU groups. If IOMMU is not enabled, that path would be empty.
941 * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
944 is_iommu_enabled(void)
946 DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
950 /* if directory doesn't exist, assume IOMMU is not enabled */
954 while ((d = readdir(dir)) != NULL) {
955 /* skip dot and dot-dot */
964 /* Launch threads, called at application init(). */
966 rte_eal_init(int argc, char **argv)
970 static uint32_t run_once;
971 uint32_t has_run = 0;
973 static char logid[PATH_MAX];
974 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
975 char thread_name[RTE_MAX_THREAD_NAME_LEN];
977 const struct rte_config *config = rte_eal_get_configuration();
978 struct internal_config *internal_conf =
979 eal_get_internal_configuration();
981 /* checks if the machine is adequate */
982 if (!rte_cpu_is_supported()) {
983 rte_eal_init_alert("unsupported cpu type.");
988 if (!__atomic_compare_exchange_n(&run_once, &has_run, 1, 0,
989 __ATOMIC_RELAXED, __ATOMIC_RELAXED)) {
990 rte_eal_init_alert("already called initialization.");
991 rte_errno = EALREADY;
995 p = strrchr(argv[0], '/');
996 strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
997 thread_id = pthread_self();
999 eal_reset_internal_config(internal_conf);
1001 /* set log level as early as possible */
1002 eal_log_level_parse(argc, argv);
1004 /* clone argv to report out later in telemetry */
1005 eal_save_args(argc, argv);
1007 if (rte_eal_cpu_init() < 0) {
1008 rte_eal_init_alert("Cannot detect lcores.");
1009 rte_errno = ENOTSUP;
1013 fctret = eal_parse_args(argc, argv);
1015 rte_eal_init_alert("Invalid 'command line' arguments.");
1017 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1021 if (eal_plugins_init() < 0) {
1022 rte_eal_init_alert("Cannot init plugins");
1024 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1028 if (eal_trace_init() < 0) {
1029 rte_eal_init_alert("Cannot init trace");
1034 if (eal_option_device_parse()) {
1036 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1040 if (rte_config_init() < 0) {
1041 rte_eal_init_alert("Cannot init config");
1045 if (rte_eal_intr_init() < 0) {
1046 rte_eal_init_alert("Cannot init interrupt-handling thread");
1050 if (rte_eal_alarm_init() < 0) {
1051 rte_eal_init_alert("Cannot init alarm");
1052 /* rte_eal_alarm_init sets rte_errno on failure. */
1056 /* Put mp channel init before bus scan so that we can init the vdev
1057 * bus through mp channel in the secondary process before the bus scan.
1059 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
1060 rte_eal_init_alert("failed to init mp channel");
1061 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1067 /* register multi-process action callbacks for hotplug */
1068 if (eal_mp_dev_hotplug_init() < 0) {
1069 rte_eal_init_alert("failed to register mp callback for hotplug");
1073 if (rte_bus_scan()) {
1074 rte_eal_init_alert("Cannot scan the buses for devices");
1076 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1080 phys_addrs = rte_eal_using_phys_addrs() != 0;
1082 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
1083 if (internal_conf->iova_mode == RTE_IOVA_DC) {
1084 /* autodetect the IOVA mapping mode */
1085 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
1087 if (iova_mode == RTE_IOVA_DC) {
1088 RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
1091 /* if we have no access to physical addresses,
1092 * pick IOVA as VA mode.
1094 iova_mode = RTE_IOVA_VA;
1095 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
1096 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
1097 } else if (rte_eal_check_module("rte_kni") == 1) {
1098 iova_mode = RTE_IOVA_PA;
1099 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
1101 } else if (is_iommu_enabled()) {
1102 /* we have an IOMMU, pick IOVA as VA mode */
1103 iova_mode = RTE_IOVA_VA;
1104 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
1106 /* physical addresses available, and no IOMMU
1107 * found, so pick IOVA as PA.
1109 iova_mode = RTE_IOVA_PA;
1110 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1113 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1114 /* Workaround for KNI which requires physical address to work
1117 if (iova_mode == RTE_IOVA_VA &&
1118 rte_eal_check_module("rte_kni") == 1) {
1120 iova_mode = RTE_IOVA_PA;
1121 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1123 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1127 rte_eal_get_configuration()->iova_mode = iova_mode;
1129 rte_eal_get_configuration()->iova_mode =
1130 internal_conf->iova_mode;
1133 if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1134 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1139 RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1140 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1142 if (internal_conf->no_hugetlbfs == 0) {
1143 /* rte_config isn't initialized yet */
1144 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
1145 eal_hugepage_info_init() :
1146 eal_hugepage_info_read();
1148 rte_eal_init_alert("Cannot get hugepage information.");
1150 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1155 if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
1156 if (internal_conf->no_hugetlbfs)
1157 internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
1160 if (internal_conf->vmware_tsc_map == 1) {
1161 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1162 rte_cycles_vmware_tsc_map = 1;
1163 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1164 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1166 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1167 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1171 if (eal_log_init(logid, internal_conf->syslog_facility) < 0) {
1172 rte_eal_init_alert("Cannot init logging.");
1174 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1179 if (rte_eal_vfio_setup() < 0) {
1180 rte_eal_init_alert("Cannot init VFIO");
1182 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1186 /* in secondary processes, memory init may allocate additional fbarrays
1187 * not present in primary processes, so to avoid any potential issues,
1188 * initialize memzones first.
1190 if (rte_eal_memzone_init() < 0) {
1191 rte_eal_init_alert("Cannot init memzone");
1196 if (rte_eal_memory_init() < 0) {
1197 rte_eal_init_alert("Cannot init memory");
1202 /* the directories are locked during eal_hugepage_info_init */
1203 eal_hugedirs_unlock();
1205 if (rte_eal_malloc_heap_init() < 0) {
1206 rte_eal_init_alert("Cannot init malloc heap");
1211 if (rte_eal_tailqs_init() < 0) {
1212 rte_eal_init_alert("Cannot init tail queues for objects");
1217 if (rte_eal_timer_init() < 0) {
1218 rte_eal_init_alert("Cannot init HPET or TSC timers");
1219 rte_errno = ENOTSUP;
1223 eal_check_mem_on_local_socket();
1225 if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
1226 &lcore_config[config->main_lcore].cpuset) != 0) {
1227 rte_eal_init_alert("Cannot set affinity");
1231 __rte_thread_init(config->main_lcore,
1232 &lcore_config[config->main_lcore].cpuset);
1234 ret = eal_thread_dump_current_affinity(cpuset, sizeof(cpuset));
1235 RTE_LOG(DEBUG, EAL, "Main lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1236 config->main_lcore, (uintptr_t)thread_id, cpuset,
1237 ret == 0 ? "" : "...");
1239 RTE_LCORE_FOREACH_WORKER(i) {
1242 * create communication pipes between main thread
1245 if (pipe(lcore_config[i].pipe_main2worker) < 0)
1246 rte_panic("Cannot create pipe\n");
1247 if (pipe(lcore_config[i].pipe_worker2main) < 0)
1248 rte_panic("Cannot create pipe\n");
1250 lcore_config[i].state = WAIT;
1252 /* create a thread for each lcore */
1253 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1254 eal_thread_loop, NULL);
1256 rte_panic("Cannot create thread\n");
1258 /* Set thread_name for aid in debugging. */
1259 snprintf(thread_name, sizeof(thread_name),
1260 "lcore-worker-%d", i);
1261 ret = rte_thread_setname(lcore_config[i].thread_id,
1265 "Cannot set name for lcore thread\n");
1267 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
1268 sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
1270 rte_panic("Cannot set affinity\n");
1274 * Launch a dummy function on all worker lcores, so that main lcore
1275 * knows they are all ready when this function returns.
1277 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MAIN);
1278 rte_eal_mp_wait_lcore();
1280 /* initialize services so vdevs register service during bus_probe. */
1281 ret = rte_service_init();
1283 rte_eal_init_alert("rte_service_init() failed");
1288 /* Probe all the buses and devices/drivers on them */
1289 if (rte_bus_probe()) {
1290 rte_eal_init_alert("Cannot probe devices");
1291 rte_errno = ENOTSUP;
1296 /* Register mp action after probe() so that we got enough info */
1297 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1301 /* initialize default service/lcore mappings and start running. Ignore
1302 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1304 ret = rte_service_start_with_defaults();
1305 if (ret < 0 && ret != -ENOTSUP) {
1311 * Clean up unused files in runtime directory. We do this at the end of
1312 * init and not at the beginning because we want to clean stuff up
1313 * whether we are primary or secondary process, but we cannot remove
1314 * primary process' files because secondary should be able to run even
1315 * if primary process is dead.
1317 * In no_shconf mode, no runtime directory is created in the first
1318 * place, so no cleanup needed.
1320 if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
1321 rte_eal_init_alert("Cannot clear runtime directory");
1324 if (rte_eal_process_type() == RTE_PROC_PRIMARY && !internal_conf->no_telemetry) {
1325 int tlog = rte_log_register_type_and_pick_level(
1326 "lib.telemetry", RTE_LOG_WARNING);
1328 tlog = RTE_LOGTYPE_EAL;
1329 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1331 &internal_conf->ctrl_cpuset, rte_log, tlog) != 0)
1335 eal_mcfg_complete();
1341 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1342 void *arg __rte_unused)
1344 /* ms is const, so find this memseg */
1345 struct rte_memseg *found;
1350 found = rte_mem_virt2memseg(ms->addr, msl);
1352 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1358 rte_eal_cleanup(void)
1360 /* if we're in a primary process, we need to mark hugepages as freeable
1361 * so that finalization can release them back to the system.
1363 struct internal_config *internal_conf =
1364 eal_get_internal_configuration();
1366 if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
1367 internal_conf->hugepage_file.unlink_existing)
1368 rte_memseg_walk(mark_freeable, NULL);
1369 rte_service_finalize();
1370 rte_mp_channel_cleanup();
1371 /* after this point, any DPDK pointers will become dangling */
1372 rte_eal_memory_detach();
1373 rte_eal_alarm_cleanup();
1376 eal_cleanup_config(internal_conf);
1380 int rte_eal_create_uio_dev(void)
1382 const struct internal_config *internal_conf =
1383 eal_get_internal_configuration();
1385 return internal_conf->create_uio_dev;
1389 rte_eal_vfio_intr_mode(void)
1391 const struct internal_config *internal_conf =
1392 eal_get_internal_configuration();
1394 return internal_conf->vfio_intr_mode;
1398 rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)
1400 struct internal_config *cfg = eal_get_internal_configuration();
1402 rte_uuid_copy(vf_token, cfg->vfio_vf_token);
1406 rte_eal_check_module(const char *module_name)
1408 char sysfs_mod_name[PATH_MAX];
1412 if (NULL == module_name)
1415 /* Check if there is sysfs mounted */
1416 if (stat("/sys/module", &st) != 0) {
1417 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1418 errno, strerror(errno));
1422 /* A module might be built-in, therefore try sysfs */
1423 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1424 if (n < 0 || n > PATH_MAX) {
1425 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1429 if (stat(sysfs_mod_name, &st) != 0) {
1430 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1431 sysfs_mod_name, errno, strerror(errno));
1435 /* Module has been found */