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 mbuf pool ops name */
87 rte_eal_mbuf_default_mempool_ops(void)
89 if (internal_config.user_mbuf_pool_ops_name == NULL)
90 return RTE_MBUF_DEFAULT_MEMPOOL_OPS;
92 return internal_config.user_mbuf_pool_ops_name;
95 /* Return a pointer to the configuration structure */
97 rte_eal_get_configuration(void)
103 rte_eal_iova_mode(void)
105 return rte_eal_get_configuration()->iova_mode;
108 /* parse a sysfs (or other) file containing one integer value */
110 eal_parse_sysfs_value(const char *filename, unsigned long *val)
116 if ((f = fopen(filename, "r")) == NULL) {
117 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
122 if (fgets(buf, sizeof(buf), f) == NULL) {
123 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
128 *val = strtoul(buf, &end, 0);
129 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
130 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
140 /* create memory configuration in shared/mmap memory. Take out
141 * a write lock on the memsegs, so we can auto-detect primary/secondary.
142 * This means we never close the file while running (auto-close on exit).
143 * We also don't lock the whole file, so that in future we can use read-locks
144 * on other parts, e.g. memzones, to detect if there are running secondary
147 rte_eal_config_create(void)
149 void *rte_mem_cfg_addr;
152 const char *pathname = eal_runtime_config_path();
154 if (internal_config.no_shconf)
158 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
160 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
163 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
166 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
169 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
172 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
173 "process running?\n", pathname);
176 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
177 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
179 if (rte_mem_cfg_addr == MAP_FAILED){
180 rte_panic("Cannot mmap memory for rte_config\n");
182 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
183 rte_config.mem_config = rte_mem_cfg_addr;
186 /* attach to an existing shared memory config */
188 rte_eal_config_attach(void)
190 void *rte_mem_cfg_addr;
191 const char *pathname = eal_runtime_config_path();
193 if (internal_config.no_shconf)
197 mem_cfg_fd = open(pathname, O_RDWR);
199 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
202 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
203 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
205 if (rte_mem_cfg_addr == MAP_FAILED)
206 rte_panic("Cannot mmap memory for rte_config\n");
208 rte_config.mem_config = rte_mem_cfg_addr;
211 /* Detect if we are a primary or a secondary process */
213 eal_proc_type_detect(void)
215 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
216 const char *pathname = eal_runtime_config_path();
218 /* if we can open the file but not get a write-lock we are a secondary
219 * process. NOTE: if we get a file handle back, we keep that open
220 * and don't close it to prevent a race condition between multiple opens */
221 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
222 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
223 ptype = RTE_PROC_SECONDARY;
225 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
226 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
231 /* Sets up rte_config structure with the pointer to shared memory config.*/
233 rte_config_init(void)
235 rte_config.process_type = internal_config.process_type;
237 switch (rte_config.process_type){
238 case RTE_PROC_PRIMARY:
239 rte_eal_config_create();
241 case RTE_PROC_SECONDARY:
242 rte_eal_config_attach();
243 rte_eal_mcfg_wait_complete(rte_config.mem_config);
246 case RTE_PROC_INVALID:
247 rte_panic("Invalid process type\n");
253 eal_usage(const char *prgname)
255 printf("\nUsage: %s ", prgname);
257 /* Allow the application to print its usage message too if hook is set */
258 if ( rte_application_usage_hook ) {
259 printf("===== Application Usage =====\n\n");
260 rte_application_usage_hook(prgname);
264 /* Set a per-application usage message */
266 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
268 rte_usage_hook_t old_func;
270 /* Will be NULL on the first call to denote the last usage routine. */
271 old_func = rte_application_usage_hook;
272 rte_application_usage_hook = usage_func;
278 eal_get_hugepage_mem_size(void)
283 for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
284 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
285 if (hpi->hugedir != NULL) {
286 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
287 size += hpi->hugepage_sz * hpi->num_pages[j];
292 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
295 /* Parse the arguments for --log-level only */
297 eal_log_level_parse(int argc, char **argv)
302 const int old_optind = optind;
303 const int old_optopt = optopt;
304 const int old_optreset = optreset;
305 char * const old_optarg = optarg;
311 while ((opt = getopt_long(argc, argvopt, eal_short_options,
312 eal_long_options, &option_index)) != EOF) {
316 /* getopt is not happy, stop right now */
320 ret = (opt == OPT_LOG_LEVEL_NUM) ?
321 eal_parse_common_option(opt, optarg, &internal_config) : 0;
323 /* common parser is not happy */
328 /* restore getopt lib */
331 optreset = old_optreset;
335 /* Parse the argument given in the command line of the application */
337 eal_parse_args(int argc, char **argv)
342 char *prgname = argv[0];
343 const int old_optind = optind;
344 const int old_optopt = optopt;
345 const int old_optreset = optreset;
346 char * const old_optarg = optarg;
352 while ((opt = getopt_long(argc, argvopt, eal_short_options,
353 eal_long_options, &option_index)) != EOF) {
355 /* getopt is not happy, stop right now */
362 ret = eal_parse_common_option(opt, optarg, &internal_config);
363 /* common parser is not happy */
369 /* common parser handled this option */
374 case OPT_MBUF_POOL_OPS_NAME_NUM:
375 internal_config.user_mbuf_pool_ops_name = optarg;
381 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
382 RTE_LOG(ERR, EAL, "Option %c is not supported "
383 "on FreeBSD\n", opt);
384 } else if (opt >= OPT_LONG_MIN_NUM &&
385 opt < OPT_LONG_MAX_NUM) {
386 RTE_LOG(ERR, EAL, "Option %s is not supported "
388 eal_long_options[option_index].name);
390 RTE_LOG(ERR, EAL, "Option %d is not supported "
391 "on FreeBSD\n", opt);
399 if (eal_adjust_config(&internal_config) != 0) {
405 if (eal_check_common_options(&internal_config) != 0) {
412 argv[optind-1] = prgname;
416 /* restore getopt lib */
419 optreset = old_optreset;
426 eal_check_mem_on_local_socket(void)
428 const struct rte_memseg *ms;
431 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
433 ms = rte_eal_get_physmem_layout();
435 for (i = 0; i < RTE_MAX_MEMSEG; i++)
436 if (ms[i].socket_id == socket_id &&
440 RTE_LOG(WARNING, EAL, "WARNING: Master core has no "
441 "memory on local socket!\n");
445 sync_func(__attribute__((unused)) void *arg)
451 rte_eal_mcfg_complete(void)
453 /* ALL shared mem_config related INIT DONE */
454 if (rte_config.process_type == RTE_PROC_PRIMARY)
455 rte_config.mem_config->magic = RTE_MAGIC;
458 /* return non-zero if hugepages are enabled. */
459 int rte_eal_has_hugepages(void)
461 return !internal_config.no_hugetlbfs;
464 /* Abstraction for port I/0 privilege */
466 rte_eal_iopl_init(void)
470 fd = open("/dev/io", O_RDWR);
473 /* keep fd open for iopl */
477 static void rte_eal_init_alert(const char *msg)
479 fprintf(stderr, "EAL: FATAL: %s\n", msg);
480 RTE_LOG(ERR, EAL, "%s\n", msg);
483 /* Launch threads, called at application init(). */
485 rte_eal_init(int argc, char **argv)
489 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
490 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
491 char thread_name[RTE_MAX_THREAD_NAME_LEN];
493 /* checks if the machine is adequate */
494 if (!rte_cpu_is_supported()) {
495 rte_eal_init_alert("unsupported cpu type.");
500 if (!rte_atomic32_test_and_set(&run_once)) {
501 rte_eal_init_alert("already called initialization.");
502 rte_errno = EALREADY;
506 thread_id = pthread_self();
508 eal_reset_internal_config(&internal_config);
510 /* set log level as early as possible */
511 eal_log_level_parse(argc, argv);
513 if (rte_eal_cpu_init() < 0) {
514 rte_eal_init_alert("Cannot detect lcores.");
519 fctret = eal_parse_args(argc, argv);
521 rte_eal_init_alert("Invalid 'command line' arguments.");
523 rte_atomic32_clear(&run_once);
527 if (eal_plugins_init() < 0) {
528 rte_eal_init_alert("Cannot init plugins\n");
530 rte_atomic32_clear(&run_once);
534 if (eal_option_device_parse()) {
536 rte_atomic32_clear(&run_once);
540 if (rte_bus_scan()) {
541 rte_eal_init_alert("Cannot scan the buses for devices\n");
543 rte_atomic32_clear(&run_once);
547 /* autodetect the iova mapping mode (default is iova_pa) */
548 rte_eal_get_configuration()->iova_mode = rte_bus_get_iommu_class();
550 if (internal_config.no_hugetlbfs == 0 &&
551 internal_config.process_type != RTE_PROC_SECONDARY &&
552 eal_hugepage_info_init() < 0) {
553 rte_eal_init_alert("Cannot get hugepage information.");
555 rte_atomic32_clear(&run_once);
559 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
560 if (internal_config.no_hugetlbfs)
561 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
563 internal_config.memory = eal_get_hugepage_mem_size();
566 if (internal_config.vmware_tsc_map == 1) {
567 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
568 rte_cycles_vmware_tsc_map = 1;
569 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
570 "you must have monitor_control.pseudo_perfctr = TRUE\n");
572 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
573 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
577 rte_srand(rte_rdtsc());
581 if (rte_mp_channel_init() < 0) {
582 rte_eal_init_alert("failed to init mp channel\n");
583 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
589 if (rte_eal_memory_init() < 0) {
590 rte_eal_init_alert("Cannot init memory\n");
595 if (rte_eal_memzone_init() < 0) {
596 rte_eal_init_alert("Cannot init memzone\n");
601 if (rte_eal_tailqs_init() < 0) {
602 rte_eal_init_alert("Cannot init tail queues for objects\n");
607 if (rte_eal_alarm_init() < 0) {
608 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
609 /* rte_eal_alarm_init sets rte_errno on failure. */
613 if (rte_eal_intr_init() < 0) {
614 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
618 if (rte_eal_timer_init() < 0) {
619 rte_eal_init_alert("Cannot init HPET or TSC timers\n");
624 eal_check_mem_on_local_socket();
626 eal_thread_init_master(rte_config.master_lcore);
628 ret = eal_thread_dump_affinity(cpuset, RTE_CPU_AFFINITY_STR_LEN);
630 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
631 rte_config.master_lcore, thread_id, cpuset,
632 ret == 0 ? "" : "...");
634 RTE_LCORE_FOREACH_SLAVE(i) {
637 * create communication pipes between master thread
640 if (pipe(lcore_config[i].pipe_master2slave) < 0)
641 rte_panic("Cannot create pipe\n");
642 if (pipe(lcore_config[i].pipe_slave2master) < 0)
643 rte_panic("Cannot create pipe\n");
645 lcore_config[i].state = WAIT;
647 /* create a thread for each lcore */
648 ret = pthread_create(&lcore_config[i].thread_id, NULL,
649 eal_thread_loop, NULL);
651 rte_panic("Cannot create thread\n");
653 /* Set thread_name for aid in debugging. */
654 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
655 "lcore-slave-%d", i);
656 rte_thread_setname(lcore_config[i].thread_id, thread_name);
660 * Launch a dummy function on all slave lcores, so that master lcore
661 * knows they are all ready when this function returns.
663 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
664 rte_eal_mp_wait_lcore();
666 /* initialize services so vdevs register service during bus_probe. */
667 ret = rte_service_init();
669 rte_eal_init_alert("rte_service_init() failed\n");
674 /* Probe all the buses and devices/drivers on them */
675 if (rte_bus_probe()) {
676 rte_eal_init_alert("Cannot probe devices\n");
681 /* initialize default service/lcore mappings and start running. Ignore
682 * -ENOTSUP, as it indicates no service coremask passed to EAL.
684 ret = rte_service_start_with_defaults();
685 if (ret < 0 && ret != -ENOTSUP) {
690 rte_eal_mcfg_complete();
695 int __rte_experimental
696 rte_eal_cleanup(void)
698 rte_service_finalize();
703 enum rte_lcore_role_t
704 rte_eal_lcore_role(unsigned lcore_id)
706 return rte_config.lcore_role[lcore_id];
710 rte_eal_process_type(void)
712 return rte_config.process_type;
715 int rte_eal_has_pci(void)
717 return !internal_config.no_pci;
720 int rte_eal_create_uio_dev(void)
722 return internal_config.create_uio_dev;
726 rte_eal_vfio_intr_mode(void)
728 return RTE_INTR_MODE_NONE;
731 /* dummy forward declaration. */
732 struct vfio_device_info;
734 /* dummy prototypes. */
735 int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr,
736 int *vfio_dev_fd, struct vfio_device_info *device_info);
737 int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd);
738 int rte_vfio_enable(const char *modname);
739 int rte_vfio_is_enabled(const char *modname);
740 int rte_vfio_noiommu_is_enabled(void);
741 int rte_vfio_clear_group(int vfio_group_fd);
743 int rte_vfio_setup_device(__rte_unused const char *sysfs_base,
744 __rte_unused const char *dev_addr,
745 __rte_unused int *vfio_dev_fd,
746 __rte_unused struct vfio_device_info *device_info)
751 int rte_vfio_release_device(__rte_unused const char *sysfs_base,
752 __rte_unused const char *dev_addr,
758 int rte_vfio_enable(__rte_unused const char *modname)
763 int rte_vfio_is_enabled(__rte_unused const char *modname)
768 int rte_vfio_noiommu_is_enabled(void)
773 int rte_vfio_clear_group(__rte_unused int vfio_group_fd)