1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation.
3 * Copyright(c) 2014 6WIND S.A.
20 #include <sys/queue.h>
23 #include <rte_compat.h>
24 #include <rte_common.h>
25 #include <rte_debug.h>
26 #include <rte_memory.h>
27 #include <rte_launch.h>
29 #include <rte_eal_memconfig.h>
30 #include <rte_errno.h>
31 #include <rte_per_lcore.h>
32 #include <rte_lcore.h>
33 #include <rte_service_component.h>
35 #include <rte_random.h>
36 #include <rte_cycles.h>
37 #include <rte_string_fns.h>
38 #include <rte_cpuflags.h>
39 #include <rte_interrupts.h>
42 #include <rte_devargs.h>
43 #include <rte_version.h>
45 #include <rte_option.h>
46 #include <rte_atomic.h>
47 #include <malloc_heap.h>
49 #include "eal_private.h"
50 #include "eal_thread.h"
51 #include "eal_internal_cfg.h"
52 #include "eal_filesystem.h"
53 #include "eal_hugepages.h"
54 #include "eal_options.h"
56 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
58 /* Allow the application to print its usage message too if set */
59 static rte_usage_hook_t rte_application_usage_hook = NULL;
60 /* early configuration structure, when memory config is not mmapped */
61 static struct rte_mem_config early_mem_config;
63 /* define fd variable here, because file needs to be kept open for the
64 * duration of the program, as we hold a write lock on it in the primary proc */
65 static int mem_cfg_fd = -1;
67 static struct flock wr_lock = {
70 .l_start = offsetof(struct rte_mem_config, memsegs),
71 .l_len = sizeof(early_mem_config.memsegs),
74 /* Address of global and public configuration */
75 static struct rte_config rte_config = {
76 .mem_config = &early_mem_config,
79 /* internal configuration (per-core) */
80 struct lcore_config lcore_config[RTE_MAX_LCORE];
82 /* internal configuration */
83 struct internal_config internal_config;
85 /* used by rte_rdtsc() */
86 int rte_cycles_vmware_tsc_map;
88 /* platform-specific runtime dir */
89 static char runtime_dir[PATH_MAX];
91 static const char *default_runtime_dir = "/var/run";
94 eal_create_runtime_dir(void)
96 const char *directory = default_runtime_dir;
97 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
98 const char *fallback = "/tmp";
103 /* try XDG path first, fall back to /tmp */
104 if (xdg_runtime_dir != NULL)
105 directory = xdg_runtime_dir;
107 directory = fallback;
109 /* create DPDK subdirectory under runtime dir */
110 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
111 if (ret < 0 || ret == sizeof(tmp)) {
112 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
116 /* create prefix-specific subdirectory under DPDK runtime dir */
117 ret = snprintf(runtime_dir, sizeof(runtime_dir), "%s/%s",
118 tmp, eal_get_hugefile_prefix());
119 if (ret < 0 || ret == sizeof(runtime_dir)) {
120 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
124 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
127 ret = mkdir(tmp, 0700);
128 if (ret < 0 && errno != EEXIST) {
129 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
130 tmp, strerror(errno));
134 ret = mkdir(runtime_dir, 0700);
135 if (ret < 0 && errno != EEXIST) {
136 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
137 runtime_dir, strerror(errno));
145 eal_clean_runtime_dir(void)
147 /* FreeBSD doesn't need this implemented for now, because, unlike Linux,
148 * FreeBSD doesn't create per-process files, so no need to clean up.
155 rte_eal_get_runtime_dir(void)
160 /* Return user provided mbuf pool ops name */
162 rte_eal_mbuf_user_pool_ops(void)
164 return internal_config.user_mbuf_pool_ops_name;
167 /* Return a pointer to the configuration structure */
169 rte_eal_get_configuration(void)
175 rte_eal_iova_mode(void)
177 return rte_eal_get_configuration()->iova_mode;
180 /* parse a sysfs (or other) file containing one integer value */
182 eal_parse_sysfs_value(const char *filename, unsigned long *val)
188 if ((f = fopen(filename, "r")) == NULL) {
189 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
194 if (fgets(buf, sizeof(buf), f) == NULL) {
195 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
200 *val = strtoul(buf, &end, 0);
201 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
202 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
212 /* create memory configuration in shared/mmap memory. Take out
213 * a write lock on the memsegs, so we can auto-detect primary/secondary.
214 * This means we never close the file while running (auto-close on exit).
215 * We also don't lock the whole file, so that in future we can use read-locks
216 * on other parts, e.g. memzones, to detect if there are running secondary
219 rte_eal_config_create(void)
221 void *rte_mem_cfg_addr;
224 const char *pathname = eal_runtime_config_path();
226 if (internal_config.no_shconf)
230 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
232 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
235 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
238 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
241 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
244 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
245 "process running?\n", pathname);
248 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
249 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
251 if (rte_mem_cfg_addr == MAP_FAILED){
252 rte_panic("Cannot mmap memory for rte_config\n");
254 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
255 rte_config.mem_config = rte_mem_cfg_addr;
258 /* attach to an existing shared memory config */
260 rte_eal_config_attach(void)
262 void *rte_mem_cfg_addr;
263 const char *pathname = eal_runtime_config_path();
265 if (internal_config.no_shconf)
269 mem_cfg_fd = open(pathname, O_RDWR);
271 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
274 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
275 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
277 if (rte_mem_cfg_addr == MAP_FAILED)
278 rte_panic("Cannot mmap memory for rte_config\n");
280 rte_config.mem_config = rte_mem_cfg_addr;
283 /* Detect if we are a primary or a secondary process */
285 eal_proc_type_detect(void)
287 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
288 const char *pathname = eal_runtime_config_path();
290 /* if there no shared config, there can be no secondary processes */
291 if (!internal_config.no_shconf) {
292 /* if we can open the file but not get a write-lock we are a
293 * secondary process. NOTE: if we get a file handle back, we
294 * keep that open and don't close it to prevent a race condition
295 * between multiple opens.
297 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
298 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
299 ptype = RTE_PROC_SECONDARY;
302 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
303 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
308 /* Sets up rte_config structure with the pointer to shared memory config.*/
310 rte_config_init(void)
312 rte_config.process_type = internal_config.process_type;
314 switch (rte_config.process_type){
315 case RTE_PROC_PRIMARY:
316 rte_eal_config_create();
318 case RTE_PROC_SECONDARY:
319 rte_eal_config_attach();
320 rte_eal_mcfg_wait_complete(rte_config.mem_config);
323 case RTE_PROC_INVALID:
324 rte_panic("Invalid process type\n");
330 eal_usage(const char *prgname)
332 printf("\nUsage: %s ", prgname);
334 /* Allow the application to print its usage message too if hook is set */
335 if ( rte_application_usage_hook ) {
336 printf("===== Application Usage =====\n\n");
337 rte_application_usage_hook(prgname);
341 /* Set a per-application usage message */
343 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
345 rte_usage_hook_t old_func;
347 /* Will be NULL on the first call to denote the last usage routine. */
348 old_func = rte_application_usage_hook;
349 rte_application_usage_hook = usage_func;
355 eal_get_hugepage_mem_size(void)
360 for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
361 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
362 if (strnlen(hpi->hugedir, sizeof(hpi->hugedir)) != 0) {
363 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
364 size += hpi->hugepage_sz * hpi->num_pages[j];
369 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
372 /* Parse the arguments for --log-level only */
374 eal_log_level_parse(int argc, char **argv)
379 const int old_optind = optind;
380 const int old_optopt = optopt;
381 const int old_optreset = optreset;
382 char * const old_optarg = optarg;
388 while ((opt = getopt_long(argc, argvopt, eal_short_options,
389 eal_long_options, &option_index)) != EOF) {
393 /* getopt is not happy, stop right now */
397 ret = (opt == OPT_LOG_LEVEL_NUM) ?
398 eal_parse_common_option(opt, optarg, &internal_config) : 0;
400 /* common parser is not happy */
405 /* restore getopt lib */
408 optreset = old_optreset;
412 /* Parse the argument given in the command line of the application */
414 eal_parse_args(int argc, char **argv)
419 char *prgname = argv[0];
420 const int old_optind = optind;
421 const int old_optopt = optopt;
422 const int old_optreset = optreset;
423 char * const old_optarg = optarg;
430 while ((opt = getopt_long(argc, argvopt, eal_short_options,
431 eal_long_options, &option_index)) != EOF) {
434 * getopt didn't recognise the option, lets parse the
435 * registered options to see if the flag is valid
438 ret = rte_option_parse(argv[optind-1]);
447 ret = eal_parse_common_option(opt, optarg, &internal_config);
448 /* common parser is not happy */
454 /* common parser handled this option */
459 case OPT_MBUF_POOL_OPS_NAME_NUM:
461 char *ops_name = strdup(optarg);
462 if (ops_name == NULL)
463 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
465 /* free old ops name */
466 if (internal_config.user_mbuf_pool_ops_name !=
468 free(internal_config.user_mbuf_pool_ops_name);
470 internal_config.user_mbuf_pool_ops_name =
479 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
480 RTE_LOG(ERR, EAL, "Option %c is not supported "
481 "on FreeBSD\n", opt);
482 } else if (opt >= OPT_LONG_MIN_NUM &&
483 opt < OPT_LONG_MAX_NUM) {
484 RTE_LOG(ERR, EAL, "Option %s is not supported "
486 eal_long_options[option_index].name);
488 RTE_LOG(ERR, EAL, "Option %d is not supported "
489 "on FreeBSD\n", opt);
497 /* create runtime data directory */
498 if (internal_config.no_shconf == 0 &&
499 eal_create_runtime_dir() < 0) {
500 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
505 if (eal_adjust_config(&internal_config) != 0) {
511 if (eal_check_common_options(&internal_config) != 0) {
518 argv[optind-1] = prgname;
522 /* restore getopt lib */
525 optreset = old_optreset;
532 check_socket(const struct rte_memseg_list *msl, void *arg)
534 int *socket_id = arg;
539 if (msl->socket_id == *socket_id && msl->memseg_arr.count != 0)
546 eal_check_mem_on_local_socket(void)
550 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
552 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
553 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
558 sync_func(__attribute__((unused)) void *arg)
564 rte_eal_mcfg_complete(void)
566 /* ALL shared mem_config related INIT DONE */
567 if (rte_config.process_type == RTE_PROC_PRIMARY)
568 rte_config.mem_config->magic = RTE_MAGIC;
571 /* return non-zero if hugepages are enabled. */
572 int rte_eal_has_hugepages(void)
574 return !internal_config.no_hugetlbfs;
577 /* Abstraction for port I/0 privilege */
579 rte_eal_iopl_init(void)
584 fd = open("/dev/io", O_RDWR);
588 /* keep fd open for iopl */
592 static void rte_eal_init_alert(const char *msg)
594 fprintf(stderr, "EAL: FATAL: %s\n", msg);
595 RTE_LOG(ERR, EAL, "%s\n", msg);
598 /* Launch threads, called at application init(). */
600 rte_eal_init(int argc, char **argv)
604 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
605 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
606 char thread_name[RTE_MAX_THREAD_NAME_LEN];
608 /* checks if the machine is adequate */
609 if (!rte_cpu_is_supported()) {
610 rte_eal_init_alert("unsupported cpu type.");
615 if (!rte_atomic32_test_and_set(&run_once)) {
616 rte_eal_init_alert("already called initialization.");
617 rte_errno = EALREADY;
621 thread_id = pthread_self();
623 eal_reset_internal_config(&internal_config);
625 /* set log level as early as possible */
626 eal_log_level_parse(argc, argv);
628 if (rte_eal_cpu_init() < 0) {
629 rte_eal_init_alert("Cannot detect lcores.");
634 fctret = eal_parse_args(argc, argv);
636 rte_eal_init_alert("Invalid 'command line' arguments.");
638 rte_atomic32_clear(&run_once);
642 /* FreeBSD always uses legacy memory model */
643 internal_config.legacy_mem = true;
645 if (eal_plugins_init() < 0) {
646 rte_eal_init_alert("Cannot init plugins");
648 rte_atomic32_clear(&run_once);
652 if (eal_option_device_parse()) {
654 rte_atomic32_clear(&run_once);
660 if (rte_eal_intr_init() < 0) {
661 rte_eal_init_alert("Cannot init interrupt-handling thread");
665 if (rte_eal_alarm_init() < 0) {
666 rte_eal_init_alert("Cannot init interrupt-handling thread");
667 /* rte_eal_alarm_init sets rte_errno on failure. */
671 /* Put mp channel init before bus scan so that we can init the vdev
672 * bus through mp channel in the secondary process before the bus scan.
674 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
675 rte_eal_init_alert("failed to init mp channel");
676 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
682 if (rte_bus_scan()) {
683 rte_eal_init_alert("Cannot scan the buses for devices");
685 rte_atomic32_clear(&run_once);
689 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
690 if (internal_config.iova_mode == RTE_IOVA_DC) {
691 /* autodetect the IOVA mapping mode (default is RTE_IOVA_PA) */
692 rte_eal_get_configuration()->iova_mode =
693 rte_bus_get_iommu_class();
695 rte_eal_get_configuration()->iova_mode =
696 internal_config.iova_mode;
699 if (internal_config.no_hugetlbfs == 0) {
700 /* rte_config isn't initialized yet */
701 ret = internal_config.process_type == RTE_PROC_PRIMARY ?
702 eal_hugepage_info_init() :
703 eal_hugepage_info_read();
705 rte_eal_init_alert("Cannot get hugepage information.");
707 rte_atomic32_clear(&run_once);
712 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
713 if (internal_config.no_hugetlbfs)
714 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
716 internal_config.memory = eal_get_hugepage_mem_size();
719 if (internal_config.vmware_tsc_map == 1) {
720 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
721 rte_cycles_vmware_tsc_map = 1;
722 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
723 "you must have monitor_control.pseudo_perfctr = TRUE\n");
725 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
726 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
730 rte_srand(rte_rdtsc());
732 /* in secondary processes, memory init may allocate additional fbarrays
733 * not present in primary processes, so to avoid any potential issues,
734 * initialize memzones first.
736 if (rte_eal_memzone_init() < 0) {
737 rte_eal_init_alert("Cannot init memzone");
742 if (rte_eal_memory_init() < 0) {
743 rte_eal_init_alert("Cannot init memory");
748 if (rte_eal_malloc_heap_init() < 0) {
749 rte_eal_init_alert("Cannot init malloc heap");
754 if (rte_eal_tailqs_init() < 0) {
755 rte_eal_init_alert("Cannot init tail queues for objects");
760 if (rte_eal_timer_init() < 0) {
761 rte_eal_init_alert("Cannot init HPET or TSC timers");
766 eal_check_mem_on_local_socket();
768 eal_thread_init_master(rte_config.master_lcore);
770 ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
772 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
773 rte_config.master_lcore, thread_id, cpuset,
774 ret == 0 ? "" : "...");
776 RTE_LCORE_FOREACH_SLAVE(i) {
779 * create communication pipes between master thread
782 if (pipe(lcore_config[i].pipe_master2slave) < 0)
783 rte_panic("Cannot create pipe\n");
784 if (pipe(lcore_config[i].pipe_slave2master) < 0)
785 rte_panic("Cannot create pipe\n");
787 lcore_config[i].state = WAIT;
789 /* create a thread for each lcore */
790 ret = pthread_create(&lcore_config[i].thread_id, NULL,
791 eal_thread_loop, NULL);
793 rte_panic("Cannot create thread\n");
795 /* Set thread_name for aid in debugging. */
796 snprintf(thread_name, sizeof(thread_name),
797 "lcore-slave-%d", i);
798 rte_thread_setname(lcore_config[i].thread_id, thread_name);
802 * Launch a dummy function on all slave lcores, so that master lcore
803 * knows they are all ready when this function returns.
805 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
806 rte_eal_mp_wait_lcore();
808 /* initialize services so vdevs register service during bus_probe. */
809 ret = rte_service_init();
811 rte_eal_init_alert("rte_service_init() failed");
816 /* Probe all the buses and devices/drivers on them */
817 if (rte_bus_probe()) {
818 rte_eal_init_alert("Cannot probe devices");
823 /* initialize default service/lcore mappings and start running. Ignore
824 * -ENOTSUP, as it indicates no service coremask passed to EAL.
826 ret = rte_service_start_with_defaults();
827 if (ret < 0 && ret != -ENOTSUP) {
833 * Clean up unused files in runtime directory. We do this at the end of
834 * init and not at the beginning because we want to clean stuff up
835 * whether we are primary or secondary process, but we cannot remove
836 * primary process' files because secondary should be able to run even
837 * if primary process is dead.
839 * In no_shconf mode, no runtime directory is created in the first
840 * place, so no cleanup needed.
842 if (!internal_config.no_shconf && eal_clean_runtime_dir() < 0) {
843 rte_eal_init_alert("Cannot clear runtime directory\n");
847 rte_eal_mcfg_complete();
849 /* Call each registered callback, if enabled */
855 int __rte_experimental
856 rte_eal_cleanup(void)
858 rte_service_finalize();
859 rte_mp_channel_cleanup();
860 eal_cleanup_config(&internal_config);
865 enum rte_lcore_role_t
866 rte_eal_lcore_role(unsigned lcore_id)
868 return rte_config.lcore_role[lcore_id];
872 rte_eal_process_type(void)
874 return rte_config.process_type;
877 int rte_eal_has_pci(void)
879 return !internal_config.no_pci;
882 int rte_eal_create_uio_dev(void)
884 return internal_config.create_uio_dev;
888 rte_eal_vfio_intr_mode(void)
890 return RTE_INTR_MODE_NONE;
893 int rte_vfio_setup_device(__rte_unused const char *sysfs_base,
894 __rte_unused const char *dev_addr,
895 __rte_unused int *vfio_dev_fd,
896 __rte_unused struct vfio_device_info *device_info)
901 int rte_vfio_release_device(__rte_unused const char *sysfs_base,
902 __rte_unused const char *dev_addr,
908 int rte_vfio_enable(__rte_unused const char *modname)
913 int rte_vfio_is_enabled(__rte_unused const char *modname)
918 int rte_vfio_noiommu_is_enabled(void)
923 int rte_vfio_clear_group(__rte_unused int vfio_group_fd)
929 rte_vfio_dma_map(uint64_t __rte_unused vaddr, __rte_unused uint64_t iova,
930 __rte_unused uint64_t len)
936 rte_vfio_dma_unmap(uint64_t __rte_unused vaddr, uint64_t __rte_unused iova,
937 __rte_unused uint64_t len)
943 rte_vfio_get_group_num(__rte_unused const char *sysfs_base,
944 __rte_unused const char *dev_addr,
945 __rte_unused int *iommu_group_num)
951 rte_vfio_get_container_fd(void)
957 rte_vfio_get_group_fd(__rte_unused int iommu_group_num)
963 rte_vfio_container_create(void)
969 rte_vfio_container_destroy(__rte_unused int container_fd)
975 rte_vfio_container_group_bind(__rte_unused int container_fd,
976 __rte_unused int iommu_group_num)
982 rte_vfio_container_group_unbind(__rte_unused int container_fd,
983 __rte_unused int iommu_group_num)
989 rte_vfio_container_dma_map(__rte_unused int container_fd,
990 __rte_unused uint64_t vaddr,
991 __rte_unused uint64_t iova,
992 __rte_unused uint64_t len)
998 rte_vfio_container_dma_unmap(__rte_unused int container_fd,
999 __rte_unused uint64_t vaddr,
1000 __rte_unused uint64_t iova,
1001 __rte_unused uint64_t len)