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);
231 if (mem_cfg_fd < 0) {
232 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
238 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
242 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
247 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
251 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
252 "process running?\n", pathname);
256 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
257 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
259 if (rte_mem_cfg_addr == MAP_FAILED){
260 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
265 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
266 rte_config.mem_config = rte_mem_cfg_addr;
271 /* attach to an existing shared memory config */
273 rte_eal_config_attach(void)
275 void *rte_mem_cfg_addr;
276 const char *pathname = eal_runtime_config_path();
278 if (internal_config.no_shconf)
282 mem_cfg_fd = open(pathname, O_RDWR);
283 if (mem_cfg_fd < 0) {
284 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
290 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
291 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
294 if (rte_mem_cfg_addr == MAP_FAILED) {
295 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
296 errno, strerror(errno));
300 rte_config.mem_config = rte_mem_cfg_addr;
305 /* Detect if we are a primary or a secondary process */
307 eal_proc_type_detect(void)
309 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
310 const char *pathname = eal_runtime_config_path();
312 /* if there no shared config, there can be no secondary processes */
313 if (!internal_config.no_shconf) {
314 /* if we can open the file but not get a write-lock we are a
315 * secondary process. NOTE: if we get a file handle back, we
316 * keep that open and don't close it to prevent a race condition
317 * between multiple opens.
319 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
320 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
321 ptype = RTE_PROC_SECONDARY;
324 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
325 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
330 /* Sets up rte_config structure with the pointer to shared memory config.*/
332 rte_config_init(void)
334 rte_config.process_type = internal_config.process_type;
336 switch (rte_config.process_type){
337 case RTE_PROC_PRIMARY:
338 if (rte_eal_config_create() < 0)
341 case RTE_PROC_SECONDARY:
342 if (rte_eal_config_attach() < 0)
344 rte_eal_mcfg_wait_complete(rte_config.mem_config);
347 case RTE_PROC_INVALID:
348 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
349 rte_config.process_type);
358 eal_usage(const char *prgname)
360 printf("\nUsage: %s ", prgname);
362 /* Allow the application to print its usage message too if hook is set */
363 if ( rte_application_usage_hook ) {
364 printf("===== Application Usage =====\n\n");
365 rte_application_usage_hook(prgname);
369 /* Set a per-application usage message */
371 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
373 rte_usage_hook_t old_func;
375 /* Will be NULL on the first call to denote the last usage routine. */
376 old_func = rte_application_usage_hook;
377 rte_application_usage_hook = usage_func;
383 eal_get_hugepage_mem_size(void)
388 for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
389 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
390 if (strnlen(hpi->hugedir, sizeof(hpi->hugedir)) != 0) {
391 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
392 size += hpi->hugepage_sz * hpi->num_pages[j];
397 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
400 /* Parse the arguments for --log-level only */
402 eal_log_level_parse(int argc, char **argv)
407 const int old_optind = optind;
408 const int old_optopt = optopt;
409 const int old_optreset = optreset;
410 char * const old_optarg = optarg;
416 while ((opt = getopt_long(argc, argvopt, eal_short_options,
417 eal_long_options, &option_index)) != EOF) {
421 /* getopt is not happy, stop right now */
425 ret = (opt == OPT_LOG_LEVEL_NUM) ?
426 eal_parse_common_option(opt, optarg, &internal_config) : 0;
428 /* common parser is not happy */
433 /* restore getopt lib */
436 optreset = old_optreset;
440 /* Parse the argument given in the command line of the application */
442 eal_parse_args(int argc, char **argv)
447 char *prgname = argv[0];
448 const int old_optind = optind;
449 const int old_optopt = optopt;
450 const int old_optreset = optreset;
451 char * const old_optarg = optarg;
458 while ((opt = getopt_long(argc, argvopt, eal_short_options,
459 eal_long_options, &option_index)) != EOF) {
462 * getopt didn't recognise the option, lets parse the
463 * registered options to see if the flag is valid
466 ret = rte_option_parse(argv[optind-1]);
475 ret = eal_parse_common_option(opt, optarg, &internal_config);
476 /* common parser is not happy */
482 /* common parser handled this option */
487 case OPT_MBUF_POOL_OPS_NAME_NUM:
489 char *ops_name = strdup(optarg);
490 if (ops_name == NULL)
491 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
493 /* free old ops name */
494 if (internal_config.user_mbuf_pool_ops_name !=
496 free(internal_config.user_mbuf_pool_ops_name);
498 internal_config.user_mbuf_pool_ops_name =
507 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
508 RTE_LOG(ERR, EAL, "Option %c is not supported "
509 "on FreeBSD\n", opt);
510 } else if (opt >= OPT_LONG_MIN_NUM &&
511 opt < OPT_LONG_MAX_NUM) {
512 RTE_LOG(ERR, EAL, "Option %s is not supported "
514 eal_long_options[option_index].name);
516 RTE_LOG(ERR, EAL, "Option %d is not supported "
517 "on FreeBSD\n", opt);
525 /* create runtime data directory */
526 if (internal_config.no_shconf == 0 &&
527 eal_create_runtime_dir() < 0) {
528 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
533 if (eal_adjust_config(&internal_config) != 0) {
539 if (eal_check_common_options(&internal_config) != 0) {
546 argv[optind-1] = prgname;
550 /* restore getopt lib */
553 optreset = old_optreset;
560 check_socket(const struct rte_memseg_list *msl, void *arg)
562 int *socket_id = arg;
567 if (msl->socket_id == *socket_id && msl->memseg_arr.count != 0)
574 eal_check_mem_on_local_socket(void)
578 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
580 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
581 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
586 sync_func(__attribute__((unused)) void *arg)
592 rte_eal_mcfg_complete(void)
594 /* ALL shared mem_config related INIT DONE */
595 if (rte_config.process_type == RTE_PROC_PRIMARY)
596 rte_config.mem_config->magic = RTE_MAGIC;
599 /* return non-zero if hugepages are enabled. */
600 int rte_eal_has_hugepages(void)
602 return !internal_config.no_hugetlbfs;
605 /* Abstraction for port I/0 privilege */
607 rte_eal_iopl_init(void)
612 fd = open("/dev/io", O_RDWR);
616 /* keep fd open for iopl */
620 static void rte_eal_init_alert(const char *msg)
622 fprintf(stderr, "EAL: FATAL: %s\n", msg);
623 RTE_LOG(ERR, EAL, "%s\n", msg);
626 /* Launch threads, called at application init(). */
628 rte_eal_init(int argc, char **argv)
632 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
633 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
634 char thread_name[RTE_MAX_THREAD_NAME_LEN];
636 /* checks if the machine is adequate */
637 if (!rte_cpu_is_supported()) {
638 rte_eal_init_alert("unsupported cpu type.");
643 if (!rte_atomic32_test_and_set(&run_once)) {
644 rte_eal_init_alert("already called initialization.");
645 rte_errno = EALREADY;
649 thread_id = pthread_self();
651 eal_reset_internal_config(&internal_config);
653 /* set log level as early as possible */
654 eal_log_level_parse(argc, argv);
656 if (rte_eal_cpu_init() < 0) {
657 rte_eal_init_alert("Cannot detect lcores.");
662 fctret = eal_parse_args(argc, argv);
664 rte_eal_init_alert("Invalid 'command line' arguments.");
666 rte_atomic32_clear(&run_once);
670 /* FreeBSD always uses legacy memory model */
671 internal_config.legacy_mem = true;
673 if (eal_plugins_init() < 0) {
674 rte_eal_init_alert("Cannot init plugins");
676 rte_atomic32_clear(&run_once);
680 if (eal_option_device_parse()) {
682 rte_atomic32_clear(&run_once);
686 if (rte_config_init() < 0) {
687 rte_eal_init_alert("Cannot init config");
691 if (rte_eal_intr_init() < 0) {
692 rte_eal_init_alert("Cannot init interrupt-handling thread");
696 if (rte_eal_alarm_init() < 0) {
697 rte_eal_init_alert("Cannot init alarm");
698 /* rte_eal_alarm_init sets rte_errno on failure. */
702 /* Put mp channel init before bus scan so that we can init the vdev
703 * bus through mp channel in the secondary process before the bus scan.
705 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
706 rte_eal_init_alert("failed to init mp channel");
707 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
713 if (rte_bus_scan()) {
714 rte_eal_init_alert("Cannot scan the buses for devices");
716 rte_atomic32_clear(&run_once);
720 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
721 if (internal_config.iova_mode == RTE_IOVA_DC) {
722 /* autodetect the IOVA mapping mode (default is RTE_IOVA_PA) */
723 rte_eal_get_configuration()->iova_mode =
724 rte_bus_get_iommu_class();
726 rte_eal_get_configuration()->iova_mode =
727 internal_config.iova_mode;
730 if (internal_config.no_hugetlbfs == 0) {
731 /* rte_config isn't initialized yet */
732 ret = internal_config.process_type == RTE_PROC_PRIMARY ?
733 eal_hugepage_info_init() :
734 eal_hugepage_info_read();
736 rte_eal_init_alert("Cannot get hugepage information.");
738 rte_atomic32_clear(&run_once);
743 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
744 if (internal_config.no_hugetlbfs)
745 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
747 internal_config.memory = eal_get_hugepage_mem_size();
750 if (internal_config.vmware_tsc_map == 1) {
751 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
752 rte_cycles_vmware_tsc_map = 1;
753 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
754 "you must have monitor_control.pseudo_perfctr = TRUE\n");
756 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
757 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
761 rte_srand(rte_rdtsc());
763 /* in secondary processes, memory init may allocate additional fbarrays
764 * not present in primary processes, so to avoid any potential issues,
765 * initialize memzones first.
767 if (rte_eal_memzone_init() < 0) {
768 rte_eal_init_alert("Cannot init memzone");
773 if (rte_eal_memory_init() < 0) {
774 rte_eal_init_alert("Cannot init memory");
779 if (rte_eal_malloc_heap_init() < 0) {
780 rte_eal_init_alert("Cannot init malloc heap");
785 if (rte_eal_tailqs_init() < 0) {
786 rte_eal_init_alert("Cannot init tail queues for objects");
791 if (rte_eal_timer_init() < 0) {
792 rte_eal_init_alert("Cannot init HPET or TSC timers");
797 eal_check_mem_on_local_socket();
799 eal_thread_init_master(rte_config.master_lcore);
801 ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
803 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
804 rte_config.master_lcore, thread_id, cpuset,
805 ret == 0 ? "" : "...");
807 RTE_LCORE_FOREACH_SLAVE(i) {
810 * create communication pipes between master thread
813 if (pipe(lcore_config[i].pipe_master2slave) < 0)
814 rte_panic("Cannot create pipe\n");
815 if (pipe(lcore_config[i].pipe_slave2master) < 0)
816 rte_panic("Cannot create pipe\n");
818 lcore_config[i].state = WAIT;
820 /* create a thread for each lcore */
821 ret = pthread_create(&lcore_config[i].thread_id, NULL,
822 eal_thread_loop, NULL);
824 rte_panic("Cannot create thread\n");
826 /* Set thread_name for aid in debugging. */
827 snprintf(thread_name, sizeof(thread_name),
828 "lcore-slave-%d", i);
829 rte_thread_setname(lcore_config[i].thread_id, thread_name);
833 * Launch a dummy function on all slave lcores, so that master lcore
834 * knows they are all ready when this function returns.
836 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
837 rte_eal_mp_wait_lcore();
839 /* initialize services so vdevs register service during bus_probe. */
840 ret = rte_service_init();
842 rte_eal_init_alert("rte_service_init() failed");
847 /* Probe all the buses and devices/drivers on them */
848 if (rte_bus_probe()) {
849 rte_eal_init_alert("Cannot probe devices");
854 /* initialize default service/lcore mappings and start running. Ignore
855 * -ENOTSUP, as it indicates no service coremask passed to EAL.
857 ret = rte_service_start_with_defaults();
858 if (ret < 0 && ret != -ENOTSUP) {
864 * Clean up unused files in runtime directory. We do this at the end of
865 * init and not at the beginning because we want to clean stuff up
866 * whether we are primary or secondary process, but we cannot remove
867 * primary process' files because secondary should be able to run even
868 * if primary process is dead.
870 * In no_shconf mode, no runtime directory is created in the first
871 * place, so no cleanup needed.
873 if (!internal_config.no_shconf && eal_clean_runtime_dir() < 0) {
874 rte_eal_init_alert("Cannot clear runtime directory\n");
878 rte_eal_mcfg_complete();
880 /* Call each registered callback, if enabled */
886 int __rte_experimental
887 rte_eal_cleanup(void)
889 rte_service_finalize();
890 rte_mp_channel_cleanup();
891 eal_cleanup_config(&internal_config);
896 enum rte_lcore_role_t
897 rte_eal_lcore_role(unsigned lcore_id)
899 return rte_config.lcore_role[lcore_id];
903 rte_eal_process_type(void)
905 return rte_config.process_type;
908 int rte_eal_has_pci(void)
910 return !internal_config.no_pci;
913 int rte_eal_create_uio_dev(void)
915 return internal_config.create_uio_dev;
919 rte_eal_vfio_intr_mode(void)
921 return RTE_INTR_MODE_NONE;
924 int rte_vfio_setup_device(__rte_unused const char *sysfs_base,
925 __rte_unused const char *dev_addr,
926 __rte_unused int *vfio_dev_fd,
927 __rte_unused struct vfio_device_info *device_info)
932 int rte_vfio_release_device(__rte_unused const char *sysfs_base,
933 __rte_unused const char *dev_addr,
939 int rte_vfio_enable(__rte_unused const char *modname)
944 int rte_vfio_is_enabled(__rte_unused const char *modname)
949 int rte_vfio_noiommu_is_enabled(void)
954 int rte_vfio_clear_group(__rte_unused int vfio_group_fd)
960 rte_vfio_dma_map(uint64_t __rte_unused vaddr, __rte_unused uint64_t iova,
961 __rte_unused uint64_t len)
967 rte_vfio_dma_unmap(uint64_t __rte_unused vaddr, uint64_t __rte_unused iova,
968 __rte_unused uint64_t len)
974 rte_vfio_get_group_num(__rte_unused const char *sysfs_base,
975 __rte_unused const char *dev_addr,
976 __rte_unused int *iommu_group_num)
982 rte_vfio_get_container_fd(void)
988 rte_vfio_get_group_fd(__rte_unused int iommu_group_num)
994 rte_vfio_container_create(void)
1000 rte_vfio_container_destroy(__rte_unused int container_fd)
1006 rte_vfio_container_group_bind(__rte_unused int container_fd,
1007 __rte_unused int iommu_group_num)
1013 rte_vfio_container_group_unbind(__rte_unused int container_fd,
1014 __rte_unused int iommu_group_num)
1020 rte_vfio_container_dma_map(__rte_unused int container_fd,
1021 __rte_unused uint64_t vaddr,
1022 __rte_unused uint64_t iova,
1023 __rte_unused uint64_t len)
1029 rte_vfio_container_dma_unmap(__rte_unused int container_fd,
1030 __rte_unused uint64_t vaddr,
1031 __rte_unused uint64_t iova,
1032 __rte_unused uint64_t len)