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>
22 #include <rte_compat.h>
23 #include <rte_common.h>
24 #include <rte_debug.h>
25 #include <rte_memory.h>
26 #include <rte_launch.h>
28 #include <rte_eal_memconfig.h>
29 #include <rte_errno.h>
30 #include <rte_per_lcore.h>
31 #include <rte_lcore.h>
32 #include <rte_service_component.h>
34 #include <rte_random.h>
35 #include <rte_cycles.h>
36 #include <rte_string_fns.h>
37 #include <rte_cpuflags.h>
38 #include <rte_interrupts.h>
41 #include <rte_devargs.h>
42 #include <rte_version.h>
43 #include <rte_atomic.h>
44 #include <malloc_heap.h>
46 #include "eal_private.h"
47 #include "eal_thread.h"
48 #include "eal_internal_cfg.h"
49 #include "eal_filesystem.h"
50 #include "eal_hugepages.h"
51 #include "eal_options.h"
53 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
55 /* Allow the application to print its usage message too if set */
56 static rte_usage_hook_t rte_application_usage_hook = NULL;
57 /* early configuration structure, when memory config is not mmapped */
58 static struct rte_mem_config early_mem_config;
60 /* define fd variable here, because file needs to be kept open for the
61 * duration of the program, as we hold a write lock on it in the primary proc */
62 static int mem_cfg_fd = -1;
64 static struct flock wr_lock = {
67 .l_start = offsetof(struct rte_mem_config, memseg),
68 .l_len = sizeof(early_mem_config.memseg),
71 /* Address of global and public configuration */
72 static struct rte_config rte_config = {
73 .mem_config = &early_mem_config,
76 /* internal configuration (per-core) */
77 struct lcore_config lcore_config[RTE_MAX_LCORE];
79 /* internal configuration */
80 struct internal_config internal_config;
82 /* used by rte_rdtsc() */
83 int rte_cycles_vmware_tsc_map;
85 /* Return user provided mbuf pool ops name */
86 const char * __rte_experimental
87 rte_eal_mbuf_user_pool_ops(void)
89 return internal_config.user_mbuf_pool_ops_name;
92 /* Return mbuf pool ops name */
94 rte_eal_mbuf_default_mempool_ops(void)
96 if (internal_config.user_mbuf_pool_ops_name == NULL)
97 return RTE_MBUF_DEFAULT_MEMPOOL_OPS;
99 return internal_config.user_mbuf_pool_ops_name;
102 /* Return a pointer to the configuration structure */
104 rte_eal_get_configuration(void)
110 rte_eal_iova_mode(void)
112 return rte_eal_get_configuration()->iova_mode;
115 /* parse a sysfs (or other) file containing one integer value */
117 eal_parse_sysfs_value(const char *filename, unsigned long *val)
123 if ((f = fopen(filename, "r")) == NULL) {
124 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
129 if (fgets(buf, sizeof(buf), f) == NULL) {
130 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
135 *val = strtoul(buf, &end, 0);
136 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
137 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
147 /* create memory configuration in shared/mmap memory. Take out
148 * a write lock on the memsegs, so we can auto-detect primary/secondary.
149 * This means we never close the file while running (auto-close on exit).
150 * We also don't lock the whole file, so that in future we can use read-locks
151 * on other parts, e.g. memzones, to detect if there are running secondary
154 rte_eal_config_create(void)
156 void *rte_mem_cfg_addr;
159 const char *pathname = eal_runtime_config_path();
161 if (internal_config.no_shconf)
165 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
167 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
170 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
173 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
176 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
179 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
180 "process running?\n", pathname);
183 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
184 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
186 if (rte_mem_cfg_addr == MAP_FAILED){
187 rte_panic("Cannot mmap memory for rte_config\n");
189 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
190 rte_config.mem_config = rte_mem_cfg_addr;
193 /* attach to an existing shared memory config */
195 rte_eal_config_attach(void)
197 void *rte_mem_cfg_addr;
198 const char *pathname = eal_runtime_config_path();
200 if (internal_config.no_shconf)
204 mem_cfg_fd = open(pathname, O_RDWR);
206 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
209 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
210 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
212 if (rte_mem_cfg_addr == MAP_FAILED)
213 rte_panic("Cannot mmap memory for rte_config\n");
215 rte_config.mem_config = rte_mem_cfg_addr;
218 /* Detect if we are a primary or a secondary process */
220 eal_proc_type_detect(void)
222 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
223 const char *pathname = eal_runtime_config_path();
225 /* if we can open the file but not get a write-lock we are a secondary
226 * process. NOTE: if we get a file handle back, we keep that open
227 * and don't close it to prevent a race condition between multiple opens */
228 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
229 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
230 ptype = RTE_PROC_SECONDARY;
232 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
233 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
238 /* Sets up rte_config structure with the pointer to shared memory config.*/
240 rte_config_init(void)
242 rte_config.process_type = internal_config.process_type;
244 switch (rte_config.process_type){
245 case RTE_PROC_PRIMARY:
246 rte_eal_config_create();
248 case RTE_PROC_SECONDARY:
249 rte_eal_config_attach();
250 rte_eal_mcfg_wait_complete(rte_config.mem_config);
253 case RTE_PROC_INVALID:
254 rte_panic("Invalid process type\n");
260 eal_usage(const char *prgname)
262 printf("\nUsage: %s ", prgname);
264 /* Allow the application to print its usage message too if hook is set */
265 if ( rte_application_usage_hook ) {
266 printf("===== Application Usage =====\n\n");
267 rte_application_usage_hook(prgname);
271 /* Set a per-application usage message */
273 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
275 rte_usage_hook_t old_func;
277 /* Will be NULL on the first call to denote the last usage routine. */
278 old_func = rte_application_usage_hook;
279 rte_application_usage_hook = usage_func;
285 eal_get_hugepage_mem_size(void)
290 for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
291 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
292 if (hpi->hugedir != NULL) {
293 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
294 size += hpi->hugepage_sz * hpi->num_pages[j];
299 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
302 /* Parse the arguments for --log-level only */
304 eal_log_level_parse(int argc, char **argv)
309 const int old_optind = optind;
310 const int old_optopt = optopt;
311 const int old_optreset = optreset;
312 char * const old_optarg = optarg;
318 while ((opt = getopt_long(argc, argvopt, eal_short_options,
319 eal_long_options, &option_index)) != EOF) {
323 /* getopt is not happy, stop right now */
327 ret = (opt == OPT_LOG_LEVEL_NUM) ?
328 eal_parse_common_option(opt, optarg, &internal_config) : 0;
330 /* common parser is not happy */
335 /* restore getopt lib */
338 optreset = old_optreset;
342 /* Parse the argument given in the command line of the application */
344 eal_parse_args(int argc, char **argv)
349 char *prgname = argv[0];
350 const int old_optind = optind;
351 const int old_optopt = optopt;
352 const int old_optreset = optreset;
353 char * const old_optarg = optarg;
359 while ((opt = getopt_long(argc, argvopt, eal_short_options,
360 eal_long_options, &option_index)) != EOF) {
362 /* getopt is not happy, stop right now */
369 ret = eal_parse_common_option(opt, optarg, &internal_config);
370 /* common parser is not happy */
376 /* common parser handled this option */
381 case OPT_MBUF_POOL_OPS_NAME_NUM:
382 internal_config.user_mbuf_pool_ops_name = optarg;
388 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
389 RTE_LOG(ERR, EAL, "Option %c is not supported "
390 "on FreeBSD\n", opt);
391 } else if (opt >= OPT_LONG_MIN_NUM &&
392 opt < OPT_LONG_MAX_NUM) {
393 RTE_LOG(ERR, EAL, "Option %s is not supported "
395 eal_long_options[option_index].name);
397 RTE_LOG(ERR, EAL, "Option %d is not supported "
398 "on FreeBSD\n", opt);
406 if (eal_adjust_config(&internal_config) != 0) {
412 if (eal_check_common_options(&internal_config) != 0) {
419 argv[optind-1] = prgname;
423 /* restore getopt lib */
426 optreset = old_optreset;
433 check_socket(const struct rte_memseg *ms, void *arg)
435 int *socket_id = arg;
437 if (ms->socket_id == *socket_id)
444 eal_check_mem_on_local_socket(void)
448 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
450 if (rte_memseg_walk(check_socket, &socket_id) == 0)
451 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
455 sync_func(__attribute__((unused)) void *arg)
461 rte_eal_mcfg_complete(void)
463 /* ALL shared mem_config related INIT DONE */
464 if (rte_config.process_type == RTE_PROC_PRIMARY)
465 rte_config.mem_config->magic = RTE_MAGIC;
468 /* return non-zero if hugepages are enabled. */
469 int rte_eal_has_hugepages(void)
471 return !internal_config.no_hugetlbfs;
474 /* Abstraction for port I/0 privilege */
476 rte_eal_iopl_init(void)
480 fd = open("/dev/io", O_RDWR);
483 /* keep fd open for iopl */
487 static void rte_eal_init_alert(const char *msg)
489 fprintf(stderr, "EAL: FATAL: %s\n", msg);
490 RTE_LOG(ERR, EAL, "%s\n", msg);
493 /* Launch threads, called at application init(). */
495 rte_eal_init(int argc, char **argv)
499 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
500 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
501 char thread_name[RTE_MAX_THREAD_NAME_LEN];
503 /* checks if the machine is adequate */
504 if (!rte_cpu_is_supported()) {
505 rte_eal_init_alert("unsupported cpu type.");
510 if (!rte_atomic32_test_and_set(&run_once)) {
511 rte_eal_init_alert("already called initialization.");
512 rte_errno = EALREADY;
516 thread_id = pthread_self();
518 eal_reset_internal_config(&internal_config);
520 /* set log level as early as possible */
521 eal_log_level_parse(argc, argv);
523 if (rte_eal_cpu_init() < 0) {
524 rte_eal_init_alert("Cannot detect lcores.");
529 fctret = eal_parse_args(argc, argv);
531 rte_eal_init_alert("Invalid 'command line' arguments.");
533 rte_atomic32_clear(&run_once);
537 if (eal_plugins_init() < 0) {
538 rte_eal_init_alert("Cannot init plugins\n");
540 rte_atomic32_clear(&run_once);
544 if (eal_option_device_parse()) {
546 rte_atomic32_clear(&run_once);
550 if (rte_bus_scan()) {
551 rte_eal_init_alert("Cannot scan the buses for devices\n");
553 rte_atomic32_clear(&run_once);
557 /* autodetect the iova mapping mode (default is iova_pa) */
558 rte_eal_get_configuration()->iova_mode = rte_bus_get_iommu_class();
560 if (internal_config.no_hugetlbfs == 0 &&
561 internal_config.process_type != RTE_PROC_SECONDARY &&
562 eal_hugepage_info_init() < 0) {
563 rte_eal_init_alert("Cannot get hugepage information.");
565 rte_atomic32_clear(&run_once);
569 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
570 if (internal_config.no_hugetlbfs)
571 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
573 internal_config.memory = eal_get_hugepage_mem_size();
576 if (internal_config.vmware_tsc_map == 1) {
577 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
578 rte_cycles_vmware_tsc_map = 1;
579 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
580 "you must have monitor_control.pseudo_perfctr = TRUE\n");
582 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
583 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
587 rte_srand(rte_rdtsc());
591 if (rte_mp_channel_init() < 0) {
592 rte_eal_init_alert("failed to init mp channel\n");
593 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
599 if (rte_eal_memory_init() < 0) {
600 rte_eal_init_alert("Cannot init memory\n");
605 if (rte_eal_memzone_init() < 0) {
606 rte_eal_init_alert("Cannot init memzone\n");
611 if (rte_eal_tailqs_init() < 0) {
612 rte_eal_init_alert("Cannot init tail queues for objects\n");
617 if (rte_eal_alarm_init() < 0) {
618 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
619 /* rte_eal_alarm_init sets rte_errno on failure. */
623 if (rte_eal_intr_init() < 0) {
624 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
628 if (rte_eal_timer_init() < 0) {
629 rte_eal_init_alert("Cannot init HPET or TSC timers\n");
634 eal_check_mem_on_local_socket();
636 eal_thread_init_master(rte_config.master_lcore);
638 ret = eal_thread_dump_affinity(cpuset, RTE_CPU_AFFINITY_STR_LEN);
640 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
641 rte_config.master_lcore, thread_id, cpuset,
642 ret == 0 ? "" : "...");
644 RTE_LCORE_FOREACH_SLAVE(i) {
647 * create communication pipes between master thread
650 if (pipe(lcore_config[i].pipe_master2slave) < 0)
651 rte_panic("Cannot create pipe\n");
652 if (pipe(lcore_config[i].pipe_slave2master) < 0)
653 rte_panic("Cannot create pipe\n");
655 lcore_config[i].state = WAIT;
657 /* create a thread for each lcore */
658 ret = pthread_create(&lcore_config[i].thread_id, NULL,
659 eal_thread_loop, NULL);
661 rte_panic("Cannot create thread\n");
663 /* Set thread_name for aid in debugging. */
664 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
665 "lcore-slave-%d", i);
666 rte_thread_setname(lcore_config[i].thread_id, thread_name);
670 * Launch a dummy function on all slave lcores, so that master lcore
671 * knows they are all ready when this function returns.
673 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
674 rte_eal_mp_wait_lcore();
676 /* initialize services so vdevs register service during bus_probe. */
677 ret = rte_service_init();
679 rte_eal_init_alert("rte_service_init() failed\n");
684 /* Probe all the buses and devices/drivers on them */
685 if (rte_bus_probe()) {
686 rte_eal_init_alert("Cannot probe devices\n");
691 /* initialize default service/lcore mappings and start running. Ignore
692 * -ENOTSUP, as it indicates no service coremask passed to EAL.
694 ret = rte_service_start_with_defaults();
695 if (ret < 0 && ret != -ENOTSUP) {
700 rte_eal_mcfg_complete();
705 int __rte_experimental
706 rte_eal_cleanup(void)
708 rte_service_finalize();
713 enum rte_lcore_role_t
714 rte_eal_lcore_role(unsigned lcore_id)
716 return rte_config.lcore_role[lcore_id];
720 rte_eal_process_type(void)
722 return rte_config.process_type;
725 int rte_eal_has_pci(void)
727 return !internal_config.no_pci;
730 int rte_eal_create_uio_dev(void)
732 return internal_config.create_uio_dev;
736 rte_eal_vfio_intr_mode(void)
738 return RTE_INTR_MODE_NONE;
741 /* dummy forward declaration. */
742 struct vfio_device_info;
744 /* dummy prototypes. */
745 int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr,
746 int *vfio_dev_fd, struct vfio_device_info *device_info);
747 int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd);
748 int rte_vfio_enable(const char *modname);
749 int rte_vfio_is_enabled(const char *modname);
750 int rte_vfio_noiommu_is_enabled(void);
751 int rte_vfio_clear_group(int vfio_group_fd);
753 int rte_vfio_setup_device(__rte_unused const char *sysfs_base,
754 __rte_unused const char *dev_addr,
755 __rte_unused int *vfio_dev_fd,
756 __rte_unused struct vfio_device_info *device_info)
761 int rte_vfio_release_device(__rte_unused const char *sysfs_base,
762 __rte_unused const char *dev_addr,
768 int rte_vfio_enable(__rte_unused const char *modname)
773 int rte_vfio_is_enabled(__rte_unused const char *modname)
778 int rte_vfio_noiommu_is_enabled(void)
783 int rte_vfio_clear_group(__rte_unused int vfio_group_fd)