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;
91 eal_clean_runtime_dir(void)
93 const char *runtime_dir = rte_eal_get_runtime_dir();
95 struct dirent *dirent;
96 int dir_fd, fd, lck_result;
97 static const char * const filters[] = {
103 dir = opendir(runtime_dir);
105 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
111 /* lock the directory before doing anything, to avoid races */
112 if (flock(dir_fd, LOCK_EX) < 0) {
113 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
118 dirent = readdir(dir);
120 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
125 while (dirent != NULL) {
129 /* skip files that don't match the patterns */
130 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
131 const char *filter = filters[f_idx];
133 if (fnmatch(filter, dirent->d_name, 0) == 0) {
139 dirent = readdir(dir);
143 /* try and lock the file */
144 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
146 /* skip to next file */
148 dirent = readdir(dir);
152 /* non-blocking lock */
153 lck_result = flock(fd, LOCK_EX | LOCK_NB);
155 /* if lock succeeds, remove the file */
156 if (lck_result != -1)
157 unlinkat(dir_fd, dirent->d_name, 0);
159 dirent = readdir(dir);
162 /* closedir closes dir_fd and drops the lock */
170 RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
177 /* create memory configuration in shared/mmap memory. Take out
178 * a write lock on the memsegs, so we can auto-detect primary/secondary.
179 * This means we never close the file while running (auto-close on exit).
180 * We also don't lock the whole file, so that in future we can use read-locks
181 * on other parts, e.g. memzones, to detect if there are running secondary
184 rte_eal_config_create(void)
186 struct rte_config *config = rte_eal_get_configuration();
187 size_t page_sz = sysconf(_SC_PAGE_SIZE);
188 size_t cfg_len = sizeof(*config->mem_config);
189 size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
190 void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
192 const struct internal_config *internal_conf =
193 eal_get_internal_configuration();
195 const char *pathname = eal_runtime_config_path();
197 if (internal_conf->no_shconf)
200 /* map the config before hugepage address so that we don't waste a page */
201 if (internal_conf->base_virtaddr != 0)
202 rte_mem_cfg_addr = (void *)
203 RTE_ALIGN_FLOOR(internal_conf->base_virtaddr -
204 sizeof(struct rte_mem_config), page_sz);
206 rte_mem_cfg_addr = NULL;
209 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
210 if (mem_cfg_fd < 0) {
211 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
217 retval = ftruncate(mem_cfg_fd, cfg_len);
221 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
226 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
230 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
231 "process running?\n", pathname);
235 /* reserve space for config */
236 rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
237 &cfg_len_aligned, page_sz, 0, 0);
238 if (rte_mem_cfg_addr == NULL) {
239 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
245 /* remap the actual file into the space we've just reserved */
246 mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
247 cfg_len_aligned, PROT_READ | PROT_WRITE,
248 MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
249 if (mapped_mem_cfg_addr == MAP_FAILED) {
250 munmap(rte_mem_cfg_addr, cfg_len);
253 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
257 memcpy(rte_mem_cfg_addr, config->mem_config, sizeof(struct rte_mem_config));
258 config->mem_config = rte_mem_cfg_addr;
260 /* store address of the config in the config itself so that secondary
261 * processes could later map the config into this exact location
263 config->mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
264 config->mem_config->dma_maskbits = 0;
269 /* attach to an existing shared memory config */
271 rte_eal_config_attach(void)
273 struct rte_config *config = rte_eal_get_configuration();
274 struct rte_mem_config *mem_config;
275 const struct internal_config *internal_conf =
276 eal_get_internal_configuration();
278 const char *pathname = eal_runtime_config_path();
280 if (internal_conf->no_shconf)
284 mem_cfg_fd = open(pathname, O_RDWR);
285 if (mem_cfg_fd < 0) {
286 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
292 /* map it as read-only first */
293 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
294 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
295 if (mem_config == MAP_FAILED) {
298 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
299 errno, strerror(errno));
303 config->mem_config = mem_config;
308 /* reattach the shared config at exact memory location primary process has it */
310 rte_eal_config_reattach(void)
312 struct rte_config *config = rte_eal_get_configuration();
313 struct rte_mem_config *mem_config;
314 void *rte_mem_cfg_addr;
315 const struct internal_config *internal_conf =
316 eal_get_internal_configuration();
318 if (internal_conf->no_shconf)
321 /* save the address primary process has mapped shared config to */
323 (void *) (uintptr_t) config->mem_config->mem_cfg_addr;
325 /* unmap original config */
326 munmap(config->mem_config, sizeof(struct rte_mem_config));
328 /* remap the config at proper address */
329 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
330 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
336 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
337 if (mem_config != MAP_FAILED) {
338 /* errno is stale, don't use */
339 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
340 " - please use '--" OPT_BASE_VIRTADDR
341 "' option\n", rte_mem_cfg_addr, mem_config);
342 munmap(mem_config, sizeof(struct rte_mem_config));
345 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
346 errno, strerror(errno));
350 config->mem_config = mem_config;
355 /* Detect if we are a primary or a secondary process */
357 eal_proc_type_detect(void)
359 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
360 const char *pathname = eal_runtime_config_path();
361 const struct internal_config *internal_conf =
362 eal_get_internal_configuration();
364 /* if there no shared config, there can be no secondary processes */
365 if (!internal_conf->no_shconf) {
366 /* if we can open the file but not get a write-lock we are a
367 * secondary process. NOTE: if we get a file handle back, we
368 * keep that open and don't close it to prevent a race condition
369 * between multiple opens.
371 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
372 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
373 ptype = RTE_PROC_SECONDARY;
376 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
377 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
382 /* Sets up rte_config structure with the pointer to shared memory config.*/
384 rte_config_init(void)
386 struct rte_config *config = rte_eal_get_configuration();
387 const struct internal_config *internal_conf =
388 eal_get_internal_configuration();
390 config->process_type = internal_conf->process_type;
392 switch (config->process_type) {
393 case RTE_PROC_PRIMARY:
394 if (rte_eal_config_create() < 0)
396 eal_mcfg_update_from_internal();
398 case RTE_PROC_SECONDARY:
399 if (rte_eal_config_attach() < 0)
401 eal_mcfg_wait_complete();
402 if (eal_mcfg_check_version() < 0) {
403 RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
406 if (rte_eal_config_reattach() < 0)
408 if (!__rte_mp_enable()) {
409 RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
412 eal_mcfg_update_internal();
415 case RTE_PROC_INVALID:
416 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
417 config->process_type);
424 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
426 eal_hugedirs_unlock(void)
429 struct internal_config *internal_conf =
430 eal_get_internal_configuration();
432 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
434 /* skip uninitialized */
435 if (internal_conf->hugepage_info[i].lock_descriptor < 0)
437 /* unlock hugepage file */
438 flock(internal_conf->hugepage_info[i].lock_descriptor, LOCK_UN);
439 close(internal_conf->hugepage_info[i].lock_descriptor);
440 /* reset the field */
441 internal_conf->hugepage_info[i].lock_descriptor = -1;
447 eal_usage(const char *prgname)
449 rte_usage_hook_t hook = eal_get_application_usage_hook();
451 printf("\nUsage: %s ", prgname);
453 printf("EAL Linux options:\n"
454 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
455 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
456 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
457 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
458 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
459 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
460 " --"OPT_VFIO_VF_TOKEN" VF token (UUID) shared between SR-IOV PF and VFs\n"
461 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
462 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
463 " --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
465 /* Allow the application to print its usage message too if hook is set */
467 printf("===== Application Usage =====\n\n");
473 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
475 char * arg[RTE_MAX_NUMA_NODES];
479 len = strnlen(strval, SOCKET_MEM_STRLEN);
480 if (len == SOCKET_MEM_STRLEN) {
481 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
485 /* all other error cases will be caught later */
486 if (!isdigit(strval[len-1]))
489 /* split the optarg into separate socket values */
490 arg_num = rte_strsplit(strval, len,
491 arg, RTE_MAX_NUMA_NODES, ',');
493 /* if split failed, or 0 arguments */
497 /* parse each defined socket option */
499 for (i = 0; i < arg_num; i++) {
502 val = strtoull(arg[i], &end, 10);
504 /* check for invalid input */
506 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
516 eal_parse_vfio_intr(const char *mode)
518 struct internal_config *internal_conf =
519 eal_get_internal_configuration();
523 enum rte_intr_mode value;
525 { "legacy", RTE_INTR_MODE_LEGACY },
526 { "msi", RTE_INTR_MODE_MSI },
527 { "msix", RTE_INTR_MODE_MSIX },
530 for (i = 0; i < RTE_DIM(map); i++) {
531 if (!strcmp(mode, map[i].name)) {
532 internal_conf->vfio_intr_mode = map[i].value;
540 eal_parse_vfio_vf_token(const char *vf_token)
542 struct internal_config *cfg = eal_get_internal_configuration();
545 if (!rte_uuid_parse(vf_token, uuid)) {
546 rte_uuid_copy(cfg->vfio_vf_token, uuid);
553 /* Parse the arguments for --log-level only */
555 eal_log_level_parse(int argc, char **argv)
560 const int old_optind = optind;
561 const int old_optopt = optopt;
562 char * const old_optarg = optarg;
563 struct internal_config *internal_conf =
564 eal_get_internal_configuration();
569 while ((opt = getopt_long(argc, argvopt, eal_short_options,
570 eal_long_options, &option_index)) != EOF) {
574 /* getopt is not happy, stop right now */
578 ret = (opt == OPT_LOG_LEVEL_NUM) ?
579 eal_parse_common_option(opt, optarg, internal_conf) : 0;
581 /* common parser is not happy */
586 /* restore getopt lib */
592 /* Parse the argument given in the command line of the application */
594 eal_parse_args(int argc, char **argv)
599 char *prgname = argv[0];
600 const int old_optind = optind;
601 const int old_optopt = optopt;
602 char * const old_optarg = optarg;
603 struct internal_config *internal_conf =
604 eal_get_internal_configuration();
609 while ((opt = getopt_long(argc, argvopt, eal_short_options,
610 eal_long_options, &option_index)) != EOF) {
612 /* getopt didn't recognise the option */
619 /* eal_log_level_parse() already handled this option */
620 if (opt == OPT_LOG_LEVEL_NUM)
623 ret = eal_parse_common_option(opt, optarg, internal_conf);
624 /* common parser is not happy */
630 /* common parser handled this option */
639 case OPT_HUGE_DIR_NUM:
641 char *hdir = strdup(optarg);
643 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
645 /* free old hugepage dir */
646 if (internal_conf->hugepage_dir != NULL)
647 free(internal_conf->hugepage_dir);
648 internal_conf->hugepage_dir = hdir;
652 case OPT_FILE_PREFIX_NUM:
654 char *prefix = strdup(optarg);
656 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
658 /* free old prefix */
659 if (internal_conf->hugefile_prefix != NULL)
660 free(internal_conf->hugefile_prefix);
661 internal_conf->hugefile_prefix = prefix;
665 case OPT_SOCKET_MEM_NUM:
666 if (eal_parse_socket_arg(optarg,
667 internal_conf->socket_mem) < 0) {
668 RTE_LOG(ERR, EAL, "invalid parameters for --"
669 OPT_SOCKET_MEM "\n");
674 internal_conf->force_sockets = 1;
677 case OPT_SOCKET_LIMIT_NUM:
678 if (eal_parse_socket_arg(optarg,
679 internal_conf->socket_limit) < 0) {
680 RTE_LOG(ERR, EAL, "invalid parameters for --"
681 OPT_SOCKET_LIMIT "\n");
686 internal_conf->force_socket_limits = 1;
689 case OPT_VFIO_INTR_NUM:
690 if (eal_parse_vfio_intr(optarg) < 0) {
691 RTE_LOG(ERR, EAL, "invalid parameters for --"
699 case OPT_VFIO_VF_TOKEN_NUM:
700 if (eal_parse_vfio_vf_token(optarg) < 0) {
701 RTE_LOG(ERR, EAL, "invalid parameters for --"
702 OPT_VFIO_VF_TOKEN "\n");
709 case OPT_CREATE_UIO_DEV_NUM:
710 internal_conf->create_uio_dev = 1;
713 case OPT_MBUF_POOL_OPS_NAME_NUM:
715 char *ops_name = strdup(optarg);
716 if (ops_name == NULL)
717 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
719 /* free old ops name */
720 if (internal_conf->user_mbuf_pool_ops_name !=
722 free(internal_conf->user_mbuf_pool_ops_name);
724 internal_conf->user_mbuf_pool_ops_name =
729 case OPT_MATCH_ALLOCATIONS_NUM:
730 internal_conf->match_allocations = 1;
734 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
735 RTE_LOG(ERR, EAL, "Option %c is not supported "
737 } else if (opt >= OPT_LONG_MIN_NUM &&
738 opt < OPT_LONG_MAX_NUM) {
739 RTE_LOG(ERR, EAL, "Option %s is not supported "
741 eal_long_options[option_index].name);
743 RTE_LOG(ERR, EAL, "Option %d is not supported "
752 /* create runtime data directory. In no_shconf mode, skip any errors */
753 if (eal_create_runtime_dir() < 0) {
754 if (internal_conf->no_shconf == 0) {
755 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
759 RTE_LOG(WARNING, EAL, "No DPDK runtime directory created\n");
762 if (eal_adjust_config(internal_conf) != 0) {
768 if (eal_check_common_options(internal_conf) != 0) {
775 argv[optind-1] = prgname;
779 /* restore getopt lib */
788 check_socket(const struct rte_memseg_list *msl, void *arg)
790 int *socket_id = arg;
795 return *socket_id == msl->socket_id;
799 eal_check_mem_on_local_socket(void)
802 const struct rte_config *config = rte_eal_get_configuration();
804 socket_id = rte_lcore_to_socket_id(config->main_lcore);
806 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
807 RTE_LOG(WARNING, EAL, "WARNING: Main core has no memory on local socket!\n");
811 sync_func(__rte_unused void *arg)
817 * Request iopl privilege for all RPL, returns 0 on success
818 * iopl() call is mostly for the i386 architecture. For other architectures,
819 * return -1 to indicate IO privilege can't be changed in this way.
822 rte_eal_iopl_init(void)
824 #if defined(RTE_ARCH_X86)
832 static int rte_eal_vfio_setup(void)
834 if (rte_vfio_enable("vfio"))
841 static void rte_eal_init_alert(const char *msg)
843 fprintf(stderr, "EAL: FATAL: %s\n", msg);
844 RTE_LOG(ERR, EAL, "%s\n", msg);
848 * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
849 * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
850 * IOMMU groups. If IOMMU is not enabled, that path would be empty.
851 * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
854 is_iommu_enabled(void)
856 DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
860 /* if directory doesn't exist, assume IOMMU is not enabled */
864 while ((d = readdir(dir)) != NULL) {
865 /* skip dot and dot-dot */
874 /* Launch threads, called at application init(). */
876 rte_eal_init(int argc, char **argv)
880 static uint32_t run_once;
881 uint32_t has_run = 0;
883 static char logid[PATH_MAX];
884 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
885 char thread_name[RTE_MAX_THREAD_NAME_LEN];
887 const struct rte_config *config = rte_eal_get_configuration();
888 struct internal_config *internal_conf =
889 eal_get_internal_configuration();
891 /* checks if the machine is adequate */
892 if (!rte_cpu_is_supported()) {
893 rte_eal_init_alert("unsupported cpu type.");
898 if (!__atomic_compare_exchange_n(&run_once, &has_run, 1, 0,
899 __ATOMIC_RELAXED, __ATOMIC_RELAXED)) {
900 rte_eal_init_alert("already called initialization.");
901 rte_errno = EALREADY;
905 p = strrchr(argv[0], '/');
906 strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
907 thread_id = pthread_self();
909 eal_reset_internal_config(internal_conf);
911 /* set log level as early as possible */
912 eal_log_level_parse(argc, argv);
914 /* clone argv to report out later in telemetry */
915 eal_save_args(argc, argv);
917 if (rte_eal_cpu_init() < 0) {
918 rte_eal_init_alert("Cannot detect lcores.");
923 fctret = eal_parse_args(argc, argv);
925 rte_eal_init_alert("Invalid 'command line' arguments.");
927 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
931 if (eal_plugins_init() < 0) {
932 rte_eal_init_alert("Cannot init plugins");
934 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
938 if (eal_trace_init() < 0) {
939 rte_eal_init_alert("Cannot init trace");
944 if (eal_option_device_parse()) {
946 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
950 if (rte_config_init() < 0) {
951 rte_eal_init_alert("Cannot init config");
955 if (rte_eal_intr_init() < 0) {
956 rte_eal_init_alert("Cannot init interrupt-handling thread");
960 if (rte_eal_alarm_init() < 0) {
961 rte_eal_init_alert("Cannot init alarm");
962 /* rte_eal_alarm_init sets rte_errno on failure. */
966 /* Put mp channel init before bus scan so that we can init the vdev
967 * bus through mp channel in the secondary process before the bus scan.
969 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
970 rte_eal_init_alert("failed to init mp channel");
971 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
977 /* register multi-process action callbacks for hotplug */
978 if (eal_mp_dev_hotplug_init() < 0) {
979 rte_eal_init_alert("failed to register mp callback for hotplug");
983 if (rte_bus_scan()) {
984 rte_eal_init_alert("Cannot scan the buses for devices");
986 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
990 phys_addrs = rte_eal_using_phys_addrs() != 0;
992 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
993 if (internal_conf->iova_mode == RTE_IOVA_DC) {
994 /* autodetect the IOVA mapping mode */
995 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
997 if (iova_mode == RTE_IOVA_DC) {
998 RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
1001 /* if we have no access to physical addresses,
1002 * pick IOVA as VA mode.
1004 iova_mode = RTE_IOVA_VA;
1005 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
1006 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
1007 } else if (rte_eal_check_module("rte_kni") == 1) {
1008 iova_mode = RTE_IOVA_PA;
1009 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
1011 } else if (is_iommu_enabled()) {
1012 /* we have an IOMMU, pick IOVA as VA mode */
1013 iova_mode = RTE_IOVA_VA;
1014 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
1016 /* physical addresses available, and no IOMMU
1017 * found, so pick IOVA as PA.
1019 iova_mode = RTE_IOVA_PA;
1020 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1023 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1024 /* Workaround for KNI which requires physical address to work
1027 if (iova_mode == RTE_IOVA_VA &&
1028 rte_eal_check_module("rte_kni") == 1) {
1030 iova_mode = RTE_IOVA_PA;
1031 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1033 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1037 rte_eal_get_configuration()->iova_mode = iova_mode;
1039 rte_eal_get_configuration()->iova_mode =
1040 internal_conf->iova_mode;
1043 if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1044 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1049 RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1050 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1052 if (internal_conf->no_hugetlbfs == 0) {
1053 /* rte_config isn't initialized yet */
1054 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
1055 eal_hugepage_info_init() :
1056 eal_hugepage_info_read();
1058 rte_eal_init_alert("Cannot get hugepage information.");
1060 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1065 if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
1066 if (internal_conf->no_hugetlbfs)
1067 internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
1070 if (internal_conf->vmware_tsc_map == 1) {
1071 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1072 rte_cycles_vmware_tsc_map = 1;
1073 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1074 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1076 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1077 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1081 if (eal_log_init(logid, internal_conf->syslog_facility) < 0) {
1082 rte_eal_init_alert("Cannot init logging.");
1084 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1089 if (rte_eal_vfio_setup() < 0) {
1090 rte_eal_init_alert("Cannot init VFIO");
1092 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1096 /* in secondary processes, memory init may allocate additional fbarrays
1097 * not present in primary processes, so to avoid any potential issues,
1098 * initialize memzones first.
1100 if (rte_eal_memzone_init() < 0) {
1101 rte_eal_init_alert("Cannot init memzone");
1106 if (rte_eal_memory_init() < 0) {
1107 rte_eal_init_alert("Cannot init memory");
1112 /* the directories are locked during eal_hugepage_info_init */
1113 eal_hugedirs_unlock();
1115 if (rte_eal_malloc_heap_init() < 0) {
1116 rte_eal_init_alert("Cannot init malloc heap");
1121 if (rte_eal_tailqs_init() < 0) {
1122 rte_eal_init_alert("Cannot init tail queues for objects");
1127 if (rte_eal_timer_init() < 0) {
1128 rte_eal_init_alert("Cannot init HPET or TSC timers");
1129 rte_errno = ENOTSUP;
1133 eal_check_mem_on_local_socket();
1135 if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
1136 &lcore_config[config->main_lcore].cpuset) != 0) {
1137 rte_eal_init_alert("Cannot set affinity");
1141 __rte_thread_init(config->main_lcore,
1142 &lcore_config[config->main_lcore].cpuset);
1144 ret = eal_thread_dump_current_affinity(cpuset, sizeof(cpuset));
1145 RTE_LOG(DEBUG, EAL, "Main lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1146 config->main_lcore, (uintptr_t)thread_id, cpuset,
1147 ret == 0 ? "" : "...");
1149 RTE_LCORE_FOREACH_WORKER(i) {
1152 * create communication pipes between main thread
1155 if (pipe(lcore_config[i].pipe_main2worker) < 0)
1156 rte_panic("Cannot create pipe\n");
1157 if (pipe(lcore_config[i].pipe_worker2main) < 0)
1158 rte_panic("Cannot create pipe\n");
1160 lcore_config[i].state = WAIT;
1162 /* create a thread for each lcore */
1163 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1164 eal_thread_loop, NULL);
1166 rte_panic("Cannot create thread\n");
1168 /* Set thread_name for aid in debugging. */
1169 snprintf(thread_name, sizeof(thread_name),
1170 "lcore-worker-%d", i);
1171 ret = rte_thread_setname(lcore_config[i].thread_id,
1175 "Cannot set name for lcore thread\n");
1177 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
1178 sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
1180 rte_panic("Cannot set affinity\n");
1184 * Launch a dummy function on all worker lcores, so that main lcore
1185 * knows they are all ready when this function returns.
1187 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MAIN);
1188 rte_eal_mp_wait_lcore();
1190 /* initialize services so vdevs register service during bus_probe. */
1191 ret = rte_service_init();
1193 rte_eal_init_alert("rte_service_init() failed");
1198 /* Probe all the buses and devices/drivers on them */
1199 if (rte_bus_probe()) {
1200 rte_eal_init_alert("Cannot probe devices");
1201 rte_errno = ENOTSUP;
1206 /* Register mp action after probe() so that we got enough info */
1207 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1211 /* initialize default service/lcore mappings and start running. Ignore
1212 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1214 ret = rte_service_start_with_defaults();
1215 if (ret < 0 && ret != -ENOTSUP) {
1221 * Clean up unused files in runtime directory. We do this at the end of
1222 * init and not at the beginning because we want to clean stuff up
1223 * whether we are primary or secondary process, but we cannot remove
1224 * primary process' files because secondary should be able to run even
1225 * if primary process is dead.
1227 * In no_shconf mode, no runtime directory is created in the first
1228 * place, so no cleanup needed.
1230 if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
1231 rte_eal_init_alert("Cannot clear runtime directory");
1234 if (rte_eal_process_type() == RTE_PROC_PRIMARY && !internal_conf->no_telemetry) {
1235 int tlog = rte_log_register_type_and_pick_level(
1236 "lib.telemetry", RTE_LOG_WARNING);
1238 tlog = RTE_LOGTYPE_EAL;
1239 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1241 &internal_conf->ctrl_cpuset, rte_log, tlog) != 0)
1245 eal_mcfg_complete();
1251 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1252 void *arg __rte_unused)
1254 /* ms is const, so find this memseg */
1255 struct rte_memseg *found;
1260 found = rte_mem_virt2memseg(ms->addr, msl);
1262 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1268 rte_eal_cleanup(void)
1270 /* if we're in a primary process, we need to mark hugepages as freeable
1271 * so that finalization can release them back to the system.
1273 struct internal_config *internal_conf =
1274 eal_get_internal_configuration();
1276 if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
1277 internal_conf->hugepage_file.unlink_existing)
1278 rte_memseg_walk(mark_freeable, NULL);
1280 rte_service_finalize();
1282 vfio_mp_sync_cleanup();
1284 rte_mp_channel_cleanup();
1285 /* after this point, any DPDK pointers will become dangling */
1286 rte_eal_memory_detach();
1287 eal_mp_dev_hotplug_cleanup();
1288 rte_eal_alarm_cleanup();
1291 eal_cleanup_config(internal_conf);
1292 rte_eal_log_cleanup();
1296 int rte_eal_create_uio_dev(void)
1298 const struct internal_config *internal_conf =
1299 eal_get_internal_configuration();
1301 return internal_conf->create_uio_dev;
1305 rte_eal_vfio_intr_mode(void)
1307 const struct internal_config *internal_conf =
1308 eal_get_internal_configuration();
1310 return internal_conf->vfio_intr_mode;
1314 rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)
1316 struct internal_config *cfg = eal_get_internal_configuration();
1318 rte_uuid_copy(vf_token, cfg->vfio_vf_token);
1322 rte_eal_check_module(const char *module_name)
1324 char sysfs_mod_name[PATH_MAX];
1328 if (NULL == module_name)
1331 /* Check if there is sysfs mounted */
1332 if (stat("/sys/module", &st) != 0) {
1333 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1334 errno, strerror(errno));
1338 /* A module might be built-in, therefore try sysfs */
1339 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1340 if (n < 0 || n > PATH_MAX) {
1341 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1345 if (stat(sysfs_mod_name, &st) != 0) {
1346 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1347 sysfs_mod_name, errno, strerror(errno));
1351 /* Module has been found */