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 <rte_atomic.h>
51 #include <malloc_heap.h>
53 #include <rte_telemetry.h>
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"
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 /* Allow the application to print its usage message too if set */
73 static rte_usage_hook_t rte_application_usage_hook = NULL;
75 /* early configuration structure, when memory config is not mmapped */
76 static struct rte_mem_config early_mem_config;
78 /* define fd variable here, because file needs to be kept open for the
79 * duration of the program, as we hold a write lock on it in the primary proc */
80 static int mem_cfg_fd = -1;
82 static struct flock wr_lock = {
85 .l_start = offsetof(struct rte_mem_config, memsegs),
86 .l_len = sizeof(early_mem_config.memsegs),
89 /* Address of global and public configuration */
90 static struct rte_config rte_config = {
91 .mem_config = &early_mem_config,
94 /* internal configuration (per-core) */
95 struct lcore_config lcore_config[RTE_MAX_LCORE];
97 /* internal configuration */
98 struct internal_config internal_config;
100 /* used by rte_rdtsc() */
101 int rte_cycles_vmware_tsc_map;
103 /* platform-specific runtime dir */
104 static char runtime_dir[PATH_MAX];
106 static const char *default_runtime_dir = "/var/run";
109 eal_create_runtime_dir(void)
111 const char *directory = default_runtime_dir;
112 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
113 const char *fallback = "/tmp";
118 /* try XDG path first, fall back to /tmp */
119 if (xdg_runtime_dir != NULL)
120 directory = xdg_runtime_dir;
122 directory = fallback;
124 /* create DPDK subdirectory under runtime dir */
125 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
126 if (ret < 0 || ret == sizeof(tmp)) {
127 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
131 /* create prefix-specific subdirectory under DPDK runtime dir */
132 ret = snprintf(runtime_dir, sizeof(runtime_dir), "%s/%s",
133 tmp, eal_get_hugefile_prefix());
134 if (ret < 0 || ret == sizeof(runtime_dir)) {
135 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
139 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
142 ret = mkdir(tmp, 0700);
143 if (ret < 0 && errno != EEXIST) {
144 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
145 tmp, strerror(errno));
149 ret = mkdir(runtime_dir, 0700);
150 if (ret < 0 && errno != EEXIST) {
151 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
152 runtime_dir, strerror(errno));
160 eal_clean_runtime_dir(void)
163 struct dirent *dirent;
164 int dir_fd, fd, lck_result;
165 static const char * const filters[] = {
171 dir = opendir(runtime_dir);
173 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
179 /* lock the directory before doing anything, to avoid races */
180 if (flock(dir_fd, LOCK_EX) < 0) {
181 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
186 dirent = readdir(dir);
188 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
193 while (dirent != NULL) {
197 /* skip files that don't match the patterns */
198 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
199 const char *filter = filters[f_idx];
201 if (fnmatch(filter, dirent->d_name, 0) == 0) {
207 dirent = readdir(dir);
211 /* try and lock the file */
212 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
214 /* skip to next file */
216 dirent = readdir(dir);
220 /* non-blocking lock */
221 lck_result = flock(fd, LOCK_EX | LOCK_NB);
223 /* if lock succeeds, remove the file */
224 if (lck_result != -1)
225 unlinkat(dir_fd, dirent->d_name, 0);
227 dirent = readdir(dir);
230 /* closedir closes dir_fd and drops the lock */
238 RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
245 rte_eal_get_runtime_dir(void)
250 /* Return user provided mbuf pool ops name */
252 rte_eal_mbuf_user_pool_ops(void)
254 return internal_config.user_mbuf_pool_ops_name;
257 /* Return a pointer to the configuration structure */
259 rte_eal_get_configuration(void)
265 rte_eal_iova_mode(void)
267 return rte_eal_get_configuration()->iova_mode;
270 /* parse a sysfs (or other) file containing one integer value */
272 eal_parse_sysfs_value(const char *filename, unsigned long *val)
278 if ((f = fopen(filename, "r")) == NULL) {
279 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
284 if (fgets(buf, sizeof(buf), f) == NULL) {
285 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
290 *val = strtoul(buf, &end, 0);
291 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
292 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
302 /* create memory configuration in shared/mmap memory. Take out
303 * a write lock on the memsegs, so we can auto-detect primary/secondary.
304 * This means we never close the file while running (auto-close on exit).
305 * We also don't lock the whole file, so that in future we can use read-locks
306 * on other parts, e.g. memzones, to detect if there are running secondary
309 rte_eal_config_create(void)
311 size_t page_sz = sysconf(_SC_PAGE_SIZE);
312 size_t cfg_len = sizeof(*rte_config.mem_config);
313 size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
314 void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
317 const char *pathname = eal_runtime_config_path();
319 if (internal_config.no_shconf)
322 /* map the config before hugepage address so that we don't waste a page */
323 if (internal_config.base_virtaddr != 0)
324 rte_mem_cfg_addr = (void *)
325 RTE_ALIGN_FLOOR(internal_config.base_virtaddr -
326 sizeof(struct rte_mem_config), page_sz);
328 rte_mem_cfg_addr = NULL;
331 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
332 if (mem_cfg_fd < 0) {
333 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
339 retval = ftruncate(mem_cfg_fd, cfg_len);
343 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
348 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
352 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
353 "process running?\n", pathname);
357 /* reserve space for config */
358 rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
359 &cfg_len_aligned, page_sz, 0, 0);
360 if (rte_mem_cfg_addr == NULL) {
361 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
367 /* remap the actual file into the space we've just reserved */
368 mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
369 cfg_len_aligned, PROT_READ | PROT_WRITE,
370 MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
371 if (mapped_mem_cfg_addr == MAP_FAILED) {
372 munmap(rte_mem_cfg_addr, cfg_len);
375 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
379 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
380 rte_config.mem_config = rte_mem_cfg_addr;
382 /* store address of the config in the config itself so that secondary
383 * processes could later map the config into this exact location */
384 rte_config.mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
386 rte_config.mem_config->dma_maskbits = 0;
391 /* attach to an existing shared memory config */
393 rte_eal_config_attach(void)
395 struct rte_mem_config *mem_config;
397 const char *pathname = eal_runtime_config_path();
399 if (internal_config.no_shconf)
403 mem_cfg_fd = open(pathname, O_RDWR);
404 if (mem_cfg_fd < 0) {
405 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
411 /* map it as read-only first */
412 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
413 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
414 if (mem_config == MAP_FAILED) {
417 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
418 errno, strerror(errno));
422 rte_config.mem_config = mem_config;
427 /* reattach the shared config at exact memory location primary process has it */
429 rte_eal_config_reattach(void)
431 struct rte_mem_config *mem_config;
432 void *rte_mem_cfg_addr;
434 if (internal_config.no_shconf)
437 /* save the address primary process has mapped shared config to */
438 rte_mem_cfg_addr = (void *) (uintptr_t) rte_config.mem_config->mem_cfg_addr;
440 /* unmap original config */
441 munmap(rte_config.mem_config, sizeof(struct rte_mem_config));
443 /* remap the config at proper address */
444 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
445 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
451 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
452 if (mem_config != MAP_FAILED) {
453 /* errno is stale, don't use */
454 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
455 " - please use '--" OPT_BASE_VIRTADDR
456 "' option\n", rte_mem_cfg_addr, mem_config);
457 munmap(mem_config, sizeof(struct rte_mem_config));
460 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
461 errno, strerror(errno));
465 rte_config.mem_config = mem_config;
470 /* Detect if we are a primary or a secondary process */
472 eal_proc_type_detect(void)
474 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
475 const char *pathname = eal_runtime_config_path();
477 /* if there no shared config, there can be no secondary processes */
478 if (!internal_config.no_shconf) {
479 /* if we can open the file but not get a write-lock we are a
480 * secondary process. NOTE: if we get a file handle back, we
481 * keep that open and don't close it to prevent a race condition
482 * between multiple opens.
484 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
485 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
486 ptype = RTE_PROC_SECONDARY;
489 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
490 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
495 /* Sets up rte_config structure with the pointer to shared memory config.*/
497 rte_config_init(void)
499 rte_config.process_type = internal_config.process_type;
501 switch (rte_config.process_type){
502 case RTE_PROC_PRIMARY:
503 if (rte_eal_config_create() < 0)
505 eal_mcfg_update_from_internal();
507 case RTE_PROC_SECONDARY:
508 if (rte_eal_config_attach() < 0)
510 eal_mcfg_wait_complete();
511 if (eal_mcfg_check_version() < 0) {
512 RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
515 if (rte_eal_config_reattach() < 0)
517 eal_mcfg_update_internal();
520 case RTE_PROC_INVALID:
521 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
522 rte_config.process_type);
529 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
531 eal_hugedirs_unlock(void)
535 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
537 /* skip uninitialized */
538 if (internal_config.hugepage_info[i].lock_descriptor < 0)
540 /* unlock hugepage file */
541 flock(internal_config.hugepage_info[i].lock_descriptor, LOCK_UN);
542 close(internal_config.hugepage_info[i].lock_descriptor);
543 /* reset the field */
544 internal_config.hugepage_info[i].lock_descriptor = -1;
550 eal_usage(const char *prgname)
552 printf("\nUsage: %s ", prgname);
554 printf("EAL Linux options:\n"
555 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
556 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
557 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
558 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
559 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
560 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
561 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
562 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
563 " --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
565 /* Allow the application to print its usage message too if hook is set */
566 if ( rte_application_usage_hook ) {
567 printf("===== Application Usage =====\n\n");
568 rte_application_usage_hook(prgname);
572 /* Set a per-application usage message */
574 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
576 rte_usage_hook_t old_func;
578 /* Will be NULL on the first call to denote the last usage routine. */
579 old_func = rte_application_usage_hook;
580 rte_application_usage_hook = usage_func;
586 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
588 char * arg[RTE_MAX_NUMA_NODES];
591 uint64_t total_mem = 0;
593 len = strnlen(strval, SOCKET_MEM_STRLEN);
594 if (len == SOCKET_MEM_STRLEN) {
595 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
599 /* all other error cases will be caught later */
600 if (!isdigit(strval[len-1]))
603 /* split the optarg into separate socket values */
604 arg_num = rte_strsplit(strval, len,
605 arg, RTE_MAX_NUMA_NODES, ',');
607 /* if split failed, or 0 arguments */
611 /* parse each defined socket option */
613 for (i = 0; i < arg_num; i++) {
616 val = strtoull(arg[i], &end, 10);
618 /* check for invalid input */
620 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
631 eal_parse_vfio_intr(const char *mode)
636 enum rte_intr_mode value;
638 { "legacy", RTE_INTR_MODE_LEGACY },
639 { "msi", RTE_INTR_MODE_MSI },
640 { "msix", RTE_INTR_MODE_MSIX },
643 for (i = 0; i < RTE_DIM(map); i++) {
644 if (!strcmp(mode, map[i].name)) {
645 internal_config.vfio_intr_mode = map[i].value;
652 /* Parse the arguments for --log-level only */
654 eal_log_level_parse(int argc, char **argv)
659 const int old_optind = optind;
660 const int old_optopt = optopt;
661 char * const old_optarg = optarg;
666 while ((opt = getopt_long(argc, argvopt, eal_short_options,
667 eal_long_options, &option_index)) != EOF) {
671 /* getopt is not happy, stop right now */
675 ret = (opt == OPT_LOG_LEVEL_NUM) ?
676 eal_parse_common_option(opt, optarg, &internal_config) : 0;
678 /* common parser is not happy */
683 /* restore getopt lib */
689 /* Parse the argument given in the command line of the application */
691 eal_parse_args(int argc, char **argv)
696 char *prgname = argv[0];
697 const int old_optind = optind;
698 const int old_optopt = optopt;
699 char * const old_optarg = optarg;
704 while ((opt = getopt_long(argc, argvopt, eal_short_options,
705 eal_long_options, &option_index)) != EOF) {
707 /* getopt didn't recognise the option */
714 ret = eal_parse_common_option(opt, optarg, &internal_config);
715 /* common parser is not happy */
721 /* common parser handled this option */
730 case OPT_HUGE_DIR_NUM:
732 char *hdir = strdup(optarg);
734 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
736 /* free old hugepage dir */
737 if (internal_config.hugepage_dir != NULL)
738 free(internal_config.hugepage_dir);
739 internal_config.hugepage_dir = hdir;
743 case OPT_FILE_PREFIX_NUM:
745 char *prefix = strdup(optarg);
747 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
749 /* free old prefix */
750 if (internal_config.hugefile_prefix != NULL)
751 free(internal_config.hugefile_prefix);
752 internal_config.hugefile_prefix = prefix;
756 case OPT_SOCKET_MEM_NUM:
757 if (eal_parse_socket_arg(optarg,
758 internal_config.socket_mem) < 0) {
759 RTE_LOG(ERR, EAL, "invalid parameters for --"
760 OPT_SOCKET_MEM "\n");
765 internal_config.force_sockets = 1;
768 case OPT_SOCKET_LIMIT_NUM:
769 if (eal_parse_socket_arg(optarg,
770 internal_config.socket_limit) < 0) {
771 RTE_LOG(ERR, EAL, "invalid parameters for --"
772 OPT_SOCKET_LIMIT "\n");
777 internal_config.force_socket_limits = 1;
780 case OPT_VFIO_INTR_NUM:
781 if (eal_parse_vfio_intr(optarg) < 0) {
782 RTE_LOG(ERR, EAL, "invalid parameters for --"
790 case OPT_CREATE_UIO_DEV_NUM:
791 internal_config.create_uio_dev = 1;
794 case OPT_MBUF_POOL_OPS_NAME_NUM:
796 char *ops_name = strdup(optarg);
797 if (ops_name == NULL)
798 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
800 /* free old ops name */
801 if (internal_config.user_mbuf_pool_ops_name !=
803 free(internal_config.user_mbuf_pool_ops_name);
805 internal_config.user_mbuf_pool_ops_name =
810 case OPT_MATCH_ALLOCATIONS_NUM:
811 internal_config.match_allocations = 1;
815 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
816 RTE_LOG(ERR, EAL, "Option %c is not supported "
818 } else if (opt >= OPT_LONG_MIN_NUM &&
819 opt < OPT_LONG_MAX_NUM) {
820 RTE_LOG(ERR, EAL, "Option %s is not supported "
822 eal_long_options[option_index].name);
824 RTE_LOG(ERR, EAL, "Option %d is not supported "
833 /* create runtime data directory */
834 if (internal_config.no_shconf == 0 &&
835 eal_create_runtime_dir() < 0) {
836 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
841 if (eal_adjust_config(&internal_config) != 0) {
847 if (eal_check_common_options(&internal_config) != 0) {
854 argv[optind-1] = prgname;
858 /* restore getopt lib */
867 check_socket(const struct rte_memseg_list *msl, void *arg)
869 int *socket_id = arg;
874 return *socket_id == msl->socket_id;
878 eal_check_mem_on_local_socket(void)
882 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
884 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
885 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
889 sync_func(__rte_unused void *arg)
895 * Request iopl privilege for all RPL, returns 0 on success
896 * iopl() call is mostly for the i386 architecture. For other architectures,
897 * return -1 to indicate IO privilege can't be changed in this way.
900 rte_eal_iopl_init(void)
902 #if defined(RTE_ARCH_X86)
910 static int rte_eal_vfio_setup(void)
912 if (rte_vfio_enable("vfio"))
919 static void rte_eal_init_alert(const char *msg)
921 fprintf(stderr, "EAL: FATAL: %s\n", msg);
922 RTE_LOG(ERR, EAL, "%s\n", msg);
926 * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
927 * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
928 * IOMMU groups. If IOMMU is not enabled, that path would be empty.
929 * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
932 is_iommu_enabled(void)
934 DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
938 /* if directory doesn't exist, assume IOMMU is not enabled */
942 while ((d = readdir(dir)) != NULL) {
943 /* skip dot and dot-dot */
952 /* Launch threads, called at application init(). */
954 rte_eal_init(int argc, char **argv)
958 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
960 static char logid[PATH_MAX];
961 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
962 char thread_name[RTE_MAX_THREAD_NAME_LEN];
965 /* checks if the machine is adequate */
966 if (!rte_cpu_is_supported()) {
967 rte_eal_init_alert("unsupported cpu type.");
972 if (!rte_atomic32_test_and_set(&run_once)) {
973 rte_eal_init_alert("already called initialization.");
974 rte_errno = EALREADY;
978 p = strrchr(argv[0], '/');
979 strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
980 thread_id = pthread_self();
982 eal_reset_internal_config(&internal_config);
984 /* set log level as early as possible */
985 eal_log_level_parse(argc, argv);
987 /* clone argv to report out later in telemetry */
988 eal_save_args(argc, argv);
990 if (rte_eal_cpu_init() < 0) {
991 rte_eal_init_alert("Cannot detect lcores.");
996 fctret = eal_parse_args(argc, argv);
998 rte_eal_init_alert("Invalid 'command line' arguments.");
1000 rte_atomic32_clear(&run_once);
1004 if (eal_plugins_init() < 0) {
1005 rte_eal_init_alert("Cannot init plugins");
1007 rte_atomic32_clear(&run_once);
1011 if (eal_trace_init() < 0) {
1012 rte_eal_init_alert("Cannot init trace");
1017 if (eal_option_device_parse()) {
1019 rte_atomic32_clear(&run_once);
1023 if (rte_config_init() < 0) {
1024 rte_eal_init_alert("Cannot init config");
1028 if (rte_eal_intr_init() < 0) {
1029 rte_eal_init_alert("Cannot init interrupt-handling thread");
1033 if (rte_eal_alarm_init() < 0) {
1034 rte_eal_init_alert("Cannot init alarm");
1035 /* rte_eal_alarm_init sets rte_errno on failure. */
1039 /* Put mp channel init before bus scan so that we can init the vdev
1040 * bus through mp channel in the secondary process before the bus scan.
1042 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
1043 rte_eal_init_alert("failed to init mp channel");
1044 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1050 /* register multi-process action callbacks for hotplug */
1051 if (eal_mp_dev_hotplug_init() < 0) {
1052 rte_eal_init_alert("failed to register mp callback for hotplug");
1056 if (rte_bus_scan()) {
1057 rte_eal_init_alert("Cannot scan the buses for devices");
1059 rte_atomic32_clear(&run_once);
1063 phys_addrs = rte_eal_using_phys_addrs() != 0;
1065 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
1066 if (internal_config.iova_mode == RTE_IOVA_DC) {
1067 /* autodetect the IOVA mapping mode */
1068 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
1070 if (iova_mode == RTE_IOVA_DC) {
1071 RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
1074 /* if we have no access to physical addresses,
1075 * pick IOVA as VA mode.
1077 iova_mode = RTE_IOVA_VA;
1078 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
1079 #if defined(RTE_LIBRTE_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
1080 } else if (rte_eal_check_module("rte_kni") == 1) {
1081 iova_mode = RTE_IOVA_PA;
1082 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
1084 } else if (is_iommu_enabled()) {
1085 /* we have an IOMMU, pick IOVA as VA mode */
1086 iova_mode = RTE_IOVA_VA;
1087 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
1089 /* physical addresses available, and no IOMMU
1090 * found, so pick IOVA as PA.
1092 iova_mode = RTE_IOVA_PA;
1093 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1096 #if defined(RTE_LIBRTE_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1097 /* Workaround for KNI which requires physical address to work
1100 if (iova_mode == RTE_IOVA_VA &&
1101 rte_eal_check_module("rte_kni") == 1) {
1103 iova_mode = RTE_IOVA_PA;
1104 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1106 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1110 rte_eal_get_configuration()->iova_mode = iova_mode;
1112 rte_eal_get_configuration()->iova_mode =
1113 internal_config.iova_mode;
1116 if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1117 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1122 RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1123 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1125 if (internal_config.no_hugetlbfs == 0) {
1126 /* rte_config isn't initialized yet */
1127 ret = internal_config.process_type == RTE_PROC_PRIMARY ?
1128 eal_hugepage_info_init() :
1129 eal_hugepage_info_read();
1131 rte_eal_init_alert("Cannot get hugepage information.");
1133 rte_atomic32_clear(&run_once);
1138 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
1139 if (internal_config.no_hugetlbfs)
1140 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
1143 if (internal_config.vmware_tsc_map == 1) {
1144 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1145 rte_cycles_vmware_tsc_map = 1;
1146 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1147 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1149 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1150 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1154 if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0) {
1155 rte_eal_init_alert("Cannot init logging.");
1157 rte_atomic32_clear(&run_once);
1162 if (rte_eal_vfio_setup() < 0) {
1163 rte_eal_init_alert("Cannot init VFIO");
1165 rte_atomic32_clear(&run_once);
1169 /* in secondary processes, memory init may allocate additional fbarrays
1170 * not present in primary processes, so to avoid any potential issues,
1171 * initialize memzones first.
1173 if (rte_eal_memzone_init() < 0) {
1174 rte_eal_init_alert("Cannot init memzone");
1179 if (rte_eal_memory_init() < 0) {
1180 rte_eal_init_alert("Cannot init memory");
1185 /* the directories are locked during eal_hugepage_info_init */
1186 eal_hugedirs_unlock();
1188 if (rte_eal_malloc_heap_init() < 0) {
1189 rte_eal_init_alert("Cannot init malloc heap");
1194 if (rte_eal_tailqs_init() < 0) {
1195 rte_eal_init_alert("Cannot init tail queues for objects");
1200 if (rte_eal_timer_init() < 0) {
1201 rte_eal_init_alert("Cannot init HPET or TSC timers");
1202 rte_errno = ENOTSUP;
1206 eal_check_mem_on_local_socket();
1208 eal_thread_init_master(rte_config.master_lcore);
1210 ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
1212 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1213 rte_config.master_lcore, (uintptr_t)thread_id, cpuset,
1214 ret == 0 ? "" : "...");
1216 RTE_LCORE_FOREACH_SLAVE(i) {
1219 * create communication pipes between master thread
1222 if (pipe(lcore_config[i].pipe_master2slave) < 0)
1223 rte_panic("Cannot create pipe\n");
1224 if (pipe(lcore_config[i].pipe_slave2master) < 0)
1225 rte_panic("Cannot create pipe\n");
1227 lcore_config[i].state = WAIT;
1229 /* create a thread for each lcore */
1230 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1231 eal_thread_loop, NULL);
1233 rte_panic("Cannot create thread\n");
1235 /* Set thread_name for aid in debugging. */
1236 snprintf(thread_name, sizeof(thread_name),
1237 "lcore-slave-%d", i);
1238 ret = rte_thread_setname(lcore_config[i].thread_id,
1242 "Cannot set name for lcore thread\n");
1246 * Launch a dummy function on all slave lcores, so that master lcore
1247 * knows they are all ready when this function returns.
1249 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
1250 rte_eal_mp_wait_lcore();
1252 /* initialize services so vdevs register service during bus_probe. */
1253 ret = rte_service_init();
1255 rte_eal_init_alert("rte_service_init() failed");
1256 rte_errno = ENOEXEC;
1260 /* Probe all the buses and devices/drivers on them */
1261 if (rte_bus_probe()) {
1262 rte_eal_init_alert("Cannot probe devices");
1263 rte_errno = ENOTSUP;
1268 /* Register mp action after probe() so that we got enough info */
1269 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1273 /* initialize default service/lcore mappings and start running. Ignore
1274 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1276 ret = rte_service_start_with_defaults();
1277 if (ret < 0 && ret != -ENOTSUP) {
1278 rte_errno = ENOEXEC;
1283 * Clean up unused files in runtime directory. We do this at the end of
1284 * init and not at the beginning because we want to clean stuff up
1285 * whether we are primary or secondary process, but we cannot remove
1286 * primary process' files because secondary should be able to run even
1287 * if primary process is dead.
1289 * In no_shconf mode, no runtime directory is created in the first
1290 * place, so no cleanup needed.
1292 if (!internal_config.no_shconf && eal_clean_runtime_dir() < 0) {
1293 rte_eal_init_alert("Cannot clear runtime directory\n");
1296 if (!internal_config.no_telemetry) {
1297 const char *error_str;
1298 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1300 rte_eal_init_alert(error_str);
1305 eal_mcfg_complete();
1311 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1312 void *arg __rte_unused)
1314 /* ms is const, so find this memseg */
1315 struct rte_memseg *found;
1320 found = rte_mem_virt2memseg(ms->addr, msl);
1322 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1328 rte_eal_cleanup(void)
1330 /* if we're in a primary process, we need to mark hugepages as freeable
1331 * so that finalization can release them back to the system.
1333 if (rte_eal_process_type() == RTE_PROC_PRIMARY)
1334 rte_memseg_walk(mark_freeable, NULL);
1335 rte_service_finalize();
1336 rte_mp_channel_cleanup();
1339 eal_cleanup_config(&internal_config);
1343 enum rte_proc_type_t
1344 rte_eal_process_type(void)
1346 return rte_config.process_type;
1349 int rte_eal_has_hugepages(void)
1351 return ! internal_config.no_hugetlbfs;
1354 int rte_eal_has_pci(void)
1356 return !internal_config.no_pci;
1359 int rte_eal_create_uio_dev(void)
1361 return internal_config.create_uio_dev;
1365 rte_eal_vfio_intr_mode(void)
1367 return internal_config.vfio_intr_mode;
1371 rte_eal_check_module(const char *module_name)
1373 char sysfs_mod_name[PATH_MAX];
1377 if (NULL == module_name)
1380 /* Check if there is sysfs mounted */
1381 if (stat("/sys/module", &st) != 0) {
1382 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1383 errno, strerror(errno));
1387 /* A module might be built-in, therefore try sysfs */
1388 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1389 if (n < 0 || n > PATH_MAX) {
1390 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1394 if (stat(sysfs_mod_name, &st) != 0) {
1395 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1396 sysfs_mod_name, errno, strerror(errno));
1400 /* Module has been found */