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_atomic.h>
46 #include <malloc_heap.h>
48 #include "eal_private.h"
49 #include "eal_thread.h"
50 #include "eal_internal_cfg.h"
51 #include "eal_filesystem.h"
52 #include "eal_hugepages.h"
53 #include "eal_options.h"
55 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
57 /* Allow the application to print its usage message too if set */
58 static rte_usage_hook_t rte_application_usage_hook = NULL;
59 /* early configuration structure, when memory config is not mmapped */
60 static struct rte_mem_config early_mem_config;
62 /* define fd variable here, because file needs to be kept open for the
63 * duration of the program, as we hold a write lock on it in the primary proc */
64 static int mem_cfg_fd = -1;
66 static struct flock wr_lock = {
69 .l_start = offsetof(struct rte_mem_config, memsegs),
70 .l_len = sizeof(early_mem_config.memsegs),
73 /* Address of global and public configuration */
74 static struct rte_config rte_config = {
75 .mem_config = &early_mem_config,
78 /* internal configuration (per-core) */
79 struct lcore_config lcore_config[RTE_MAX_LCORE];
81 /* internal configuration */
82 struct internal_config internal_config;
84 /* used by rte_rdtsc() */
85 int rte_cycles_vmware_tsc_map;
87 /* platform-specific runtime dir */
88 static char runtime_dir[PATH_MAX];
91 eal_create_runtime_dir(void)
93 const char *directory = default_config_dir;
94 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
95 const char *fallback = "/tmp";
100 /* try XDG path first, fall back to /tmp */
101 if (xdg_runtime_dir != NULL)
102 directory = xdg_runtime_dir;
104 directory = fallback;
106 /* create DPDK subdirectory under runtime dir */
107 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
108 if (ret < 0 || ret == sizeof(tmp)) {
109 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
113 /* create prefix-specific subdirectory under DPDK runtime dir */
114 ret = snprintf(runtime_dir, sizeof(runtime_dir), "%s/%s",
115 tmp, internal_config.hugefile_prefix);
116 if (ret < 0 || ret == sizeof(runtime_dir)) {
117 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
121 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
124 ret = mkdir(tmp, 0600);
125 if (ret < 0 && errno != EEXIST) {
126 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
127 tmp, strerror(errno));
131 ret = mkdir(runtime_dir, 0600);
132 if (ret < 0 && errno != EEXIST) {
133 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
134 runtime_dir, strerror(errno));
142 eal_get_runtime_dir(void)
147 /* Return user provided mbuf pool ops name */
148 const char * __rte_experimental
149 rte_eal_mbuf_user_pool_ops(void)
151 return internal_config.user_mbuf_pool_ops_name;
154 /* Return mbuf pool ops name */
156 rte_eal_mbuf_default_mempool_ops(void)
158 if (internal_config.user_mbuf_pool_ops_name == NULL)
159 return RTE_MBUF_DEFAULT_MEMPOOL_OPS;
161 return internal_config.user_mbuf_pool_ops_name;
164 /* Return a pointer to the configuration structure */
166 rte_eal_get_configuration(void)
172 rte_eal_iova_mode(void)
174 return rte_eal_get_configuration()->iova_mode;
177 /* parse a sysfs (or other) file containing one integer value */
179 eal_parse_sysfs_value(const char *filename, unsigned long *val)
185 if ((f = fopen(filename, "r")) == NULL) {
186 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
191 if (fgets(buf, sizeof(buf), f) == NULL) {
192 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
197 *val = strtoul(buf, &end, 0);
198 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
199 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
209 /* create memory configuration in shared/mmap memory. Take out
210 * a write lock on the memsegs, so we can auto-detect primary/secondary.
211 * This means we never close the file while running (auto-close on exit).
212 * We also don't lock the whole file, so that in future we can use read-locks
213 * on other parts, e.g. memzones, to detect if there are running secondary
216 rte_eal_config_create(void)
218 void *rte_mem_cfg_addr;
221 const char *pathname = eal_runtime_config_path();
223 if (internal_config.no_shconf)
227 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
229 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
232 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
235 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
238 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
241 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
242 "process running?\n", pathname);
245 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
246 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
248 if (rte_mem_cfg_addr == MAP_FAILED){
249 rte_panic("Cannot mmap memory for rte_config\n");
251 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
252 rte_config.mem_config = rte_mem_cfg_addr;
255 /* attach to an existing shared memory config */
257 rte_eal_config_attach(void)
259 void *rte_mem_cfg_addr;
260 const char *pathname = eal_runtime_config_path();
262 if (internal_config.no_shconf)
266 mem_cfg_fd = open(pathname, O_RDWR);
268 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
271 rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
272 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
274 if (rte_mem_cfg_addr == MAP_FAILED)
275 rte_panic("Cannot mmap memory for rte_config\n");
277 rte_config.mem_config = rte_mem_cfg_addr;
280 /* Detect if we are a primary or a secondary process */
282 eal_proc_type_detect(void)
284 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
285 const char *pathname = eal_runtime_config_path();
287 /* if we can open the file but not get a write-lock we are a secondary
288 * process. NOTE: if we get a file handle back, we keep that open
289 * and don't close it to prevent a race condition between multiple opens */
290 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
291 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
292 ptype = RTE_PROC_SECONDARY;
294 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
295 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
300 /* Sets up rte_config structure with the pointer to shared memory config.*/
302 rte_config_init(void)
304 rte_config.process_type = internal_config.process_type;
306 switch (rte_config.process_type){
307 case RTE_PROC_PRIMARY:
308 rte_eal_config_create();
310 case RTE_PROC_SECONDARY:
311 rte_eal_config_attach();
312 rte_eal_mcfg_wait_complete(rte_config.mem_config);
315 case RTE_PROC_INVALID:
316 rte_panic("Invalid process type\n");
322 eal_usage(const char *prgname)
324 printf("\nUsage: %s ", prgname);
326 /* Allow the application to print its usage message too if hook is set */
327 if ( rte_application_usage_hook ) {
328 printf("===== Application Usage =====\n\n");
329 rte_application_usage_hook(prgname);
333 /* Set a per-application usage message */
335 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
337 rte_usage_hook_t old_func;
339 /* Will be NULL on the first call to denote the last usage routine. */
340 old_func = rte_application_usage_hook;
341 rte_application_usage_hook = usage_func;
347 eal_get_hugepage_mem_size(void)
352 for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
353 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
354 if (strnlen(hpi->hugedir, sizeof(hpi->hugedir)) != 0) {
355 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
356 size += hpi->hugepage_sz * hpi->num_pages[j];
361 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
364 /* Parse the arguments for --log-level only */
366 eal_log_level_parse(int argc, char **argv)
371 const int old_optind = optind;
372 const int old_optopt = optopt;
373 const int old_optreset = optreset;
374 char * const old_optarg = optarg;
380 while ((opt = getopt_long(argc, argvopt, eal_short_options,
381 eal_long_options, &option_index)) != EOF) {
385 /* getopt is not happy, stop right now */
389 ret = (opt == OPT_LOG_LEVEL_NUM) ?
390 eal_parse_common_option(opt, optarg, &internal_config) : 0;
392 /* common parser is not happy */
397 /* restore getopt lib */
400 optreset = old_optreset;
404 /* Parse the argument given in the command line of the application */
406 eal_parse_args(int argc, char **argv)
411 char *prgname = argv[0];
412 const int old_optind = optind;
413 const int old_optopt = optopt;
414 const int old_optreset = optreset;
415 char * const old_optarg = optarg;
421 while ((opt = getopt_long(argc, argvopt, eal_short_options,
422 eal_long_options, &option_index)) != EOF) {
424 /* getopt is not happy, stop right now */
431 ret = eal_parse_common_option(opt, optarg, &internal_config);
432 /* common parser is not happy */
438 /* common parser handled this option */
443 case OPT_MBUF_POOL_OPS_NAME_NUM:
444 internal_config.user_mbuf_pool_ops_name =
451 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
452 RTE_LOG(ERR, EAL, "Option %c is not supported "
453 "on FreeBSD\n", opt);
454 } else if (opt >= OPT_LONG_MIN_NUM &&
455 opt < OPT_LONG_MAX_NUM) {
456 RTE_LOG(ERR, EAL, "Option %s is not supported "
458 eal_long_options[option_index].name);
460 RTE_LOG(ERR, EAL, "Option %d is not supported "
461 "on FreeBSD\n", opt);
469 if (eal_adjust_config(&internal_config) != 0) {
475 if (eal_check_common_options(&internal_config) != 0) {
482 argv[optind-1] = prgname;
486 /* restore getopt lib */
489 optreset = old_optreset;
496 check_socket(const struct rte_memseg_list *msl, void *arg)
498 int *socket_id = arg;
500 if (msl->socket_id == *socket_id && msl->memseg_arr.count != 0)
507 eal_check_mem_on_local_socket(void)
511 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
513 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
514 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
519 sync_func(__attribute__((unused)) void *arg)
525 rte_eal_mcfg_complete(void)
527 /* ALL shared mem_config related INIT DONE */
528 if (rte_config.process_type == RTE_PROC_PRIMARY)
529 rte_config.mem_config->magic = RTE_MAGIC;
532 /* return non-zero if hugepages are enabled. */
533 int rte_eal_has_hugepages(void)
535 return !internal_config.no_hugetlbfs;
538 /* Abstraction for port I/0 privilege */
540 rte_eal_iopl_init(void)
544 fd = open("/dev/io", O_RDWR);
547 /* keep fd open for iopl */
551 static void rte_eal_init_alert(const char *msg)
553 fprintf(stderr, "EAL: FATAL: %s\n", msg);
554 RTE_LOG(ERR, EAL, "%s\n", msg);
557 /* Launch threads, called at application init(). */
559 rte_eal_init(int argc, char **argv)
563 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
564 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
565 char thread_name[RTE_MAX_THREAD_NAME_LEN];
567 /* checks if the machine is adequate */
568 if (!rte_cpu_is_supported()) {
569 rte_eal_init_alert("unsupported cpu type.");
574 if (!rte_atomic32_test_and_set(&run_once)) {
575 rte_eal_init_alert("already called initialization.");
576 rte_errno = EALREADY;
580 thread_id = pthread_self();
582 eal_reset_internal_config(&internal_config);
584 /* set log level as early as possible */
585 eal_log_level_parse(argc, argv);
587 /* create runtime data directory */
588 if (eal_create_runtime_dir() < 0) {
589 rte_eal_init_alert("Cannot create runtime directory\n");
594 if (rte_eal_cpu_init() < 0) {
595 rte_eal_init_alert("Cannot detect lcores.");
600 fctret = eal_parse_args(argc, argv);
602 rte_eal_init_alert("Invalid 'command line' arguments.");
604 rte_atomic32_clear(&run_once);
608 /* FreeBSD always uses legacy memory model */
609 internal_config.legacy_mem = true;
611 if (eal_plugins_init() < 0) {
612 rte_eal_init_alert("Cannot init plugins\n");
614 rte_atomic32_clear(&run_once);
618 if (eal_option_device_parse()) {
620 rte_atomic32_clear(&run_once);
626 /* Put mp channel init before bus scan so that we can init the vdev
627 * bus through mp channel in the secondary process before the bus scan.
629 if (rte_mp_channel_init() < 0) {
630 rte_eal_init_alert("failed to init mp channel\n");
631 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
637 if (rte_bus_scan()) {
638 rte_eal_init_alert("Cannot scan the buses for devices\n");
640 rte_atomic32_clear(&run_once);
644 /* autodetect the iova mapping mode (default is iova_pa) */
645 rte_eal_get_configuration()->iova_mode = rte_bus_get_iommu_class();
647 if (internal_config.no_hugetlbfs == 0) {
648 /* rte_config isn't initialized yet */
649 ret = internal_config.process_type == RTE_PROC_PRIMARY ?
650 eal_hugepage_info_init() :
651 eal_hugepage_info_read();
653 rte_eal_init_alert("Cannot get hugepage information.");
655 rte_atomic32_clear(&run_once);
660 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
661 if (internal_config.no_hugetlbfs)
662 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
664 internal_config.memory = eal_get_hugepage_mem_size();
667 if (internal_config.vmware_tsc_map == 1) {
668 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
669 rte_cycles_vmware_tsc_map = 1;
670 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
671 "you must have monitor_control.pseudo_perfctr = TRUE\n");
673 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
674 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
678 rte_srand(rte_rdtsc());
680 /* in secondary processes, memory init may allocate additional fbarrays
681 * not present in primary processes, so to avoid any potential issues,
682 * initialize memzones first.
684 if (rte_eal_memzone_init() < 0) {
685 rte_eal_init_alert("Cannot init memzone\n");
690 if (rte_eal_memory_init() < 0) {
691 rte_eal_init_alert("Cannot init memory\n");
696 if (rte_eal_malloc_heap_init() < 0) {
697 rte_eal_init_alert("Cannot init malloc heap\n");
702 if (rte_eal_tailqs_init() < 0) {
703 rte_eal_init_alert("Cannot init tail queues for objects\n");
708 if (rte_eal_alarm_init() < 0) {
709 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
710 /* rte_eal_alarm_init sets rte_errno on failure. */
714 if (rte_eal_intr_init() < 0) {
715 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
719 if (rte_eal_timer_init() < 0) {
720 rte_eal_init_alert("Cannot init HPET or TSC timers\n");
725 eal_check_mem_on_local_socket();
727 eal_thread_init_master(rte_config.master_lcore);
729 ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
731 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
732 rte_config.master_lcore, thread_id, cpuset,
733 ret == 0 ? "" : "...");
735 RTE_LCORE_FOREACH_SLAVE(i) {
738 * create communication pipes between master thread
741 if (pipe(lcore_config[i].pipe_master2slave) < 0)
742 rte_panic("Cannot create pipe\n");
743 if (pipe(lcore_config[i].pipe_slave2master) < 0)
744 rte_panic("Cannot create pipe\n");
746 lcore_config[i].state = WAIT;
748 /* create a thread for each lcore */
749 ret = pthread_create(&lcore_config[i].thread_id, NULL,
750 eal_thread_loop, NULL);
752 rte_panic("Cannot create thread\n");
754 /* Set thread_name for aid in debugging. */
755 snprintf(thread_name, sizeof(thread_name),
756 "lcore-slave-%d", i);
757 rte_thread_setname(lcore_config[i].thread_id, thread_name);
761 * Launch a dummy function on all slave lcores, so that master lcore
762 * knows they are all ready when this function returns.
764 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
765 rte_eal_mp_wait_lcore();
767 /* initialize services so vdevs register service during bus_probe. */
768 ret = rte_service_init();
770 rte_eal_init_alert("rte_service_init() failed\n");
775 /* Probe all the buses and devices/drivers on them */
776 if (rte_bus_probe()) {
777 rte_eal_init_alert("Cannot probe devices\n");
782 /* initialize default service/lcore mappings and start running. Ignore
783 * -ENOTSUP, as it indicates no service coremask passed to EAL.
785 ret = rte_service_start_with_defaults();
786 if (ret < 0 && ret != -ENOTSUP) {
791 rte_eal_mcfg_complete();
796 int __rte_experimental
797 rte_eal_cleanup(void)
799 rte_service_finalize();
804 enum rte_lcore_role_t
805 rte_eal_lcore_role(unsigned lcore_id)
807 return rte_config.lcore_role[lcore_id];
811 rte_eal_process_type(void)
813 return rte_config.process_type;
816 int rte_eal_has_pci(void)
818 return !internal_config.no_pci;
821 int rte_eal_create_uio_dev(void)
823 return internal_config.create_uio_dev;
827 rte_eal_vfio_intr_mode(void)
829 return RTE_INTR_MODE_NONE;
832 int rte_vfio_setup_device(__rte_unused const char *sysfs_base,
833 __rte_unused const char *dev_addr,
834 __rte_unused int *vfio_dev_fd,
835 __rte_unused struct vfio_device_info *device_info)
840 int rte_vfio_release_device(__rte_unused const char *sysfs_base,
841 __rte_unused const char *dev_addr,
847 int rte_vfio_enable(__rte_unused const char *modname)
852 int rte_vfio_is_enabled(__rte_unused const char *modname)
857 int rte_vfio_noiommu_is_enabled(void)
862 int rte_vfio_clear_group(__rte_unused int vfio_group_fd)
867 int __rte_experimental
868 rte_vfio_dma_map(uint64_t __rte_unused vaddr, __rte_unused uint64_t iova,
869 __rte_unused uint64_t len)
874 int __rte_experimental
875 rte_vfio_dma_unmap(uint64_t __rte_unused vaddr, uint64_t __rte_unused iova,
876 __rte_unused uint64_t len)
881 int __rte_experimental
882 rte_vfio_get_group_num(__rte_unused const char *sysfs_base,
883 __rte_unused const char *dev_addr,
884 __rte_unused int *iommu_group_num)
889 int __rte_experimental
890 rte_vfio_get_container_fd(void)
895 int __rte_experimental
896 rte_vfio_get_group_fd(__rte_unused int iommu_group_num)
901 int __rte_experimental
902 rte_vfio_container_create(void)
907 int __rte_experimental
908 rte_vfio_container_destroy(__rte_unused int container_fd)
913 int __rte_experimental
914 rte_vfio_container_group_bind(__rte_unused int container_fd,
915 __rte_unused int iommu_group_num)
920 int __rte_experimental
921 rte_vfio_container_group_unbind(__rte_unused int container_fd,
922 __rte_unused int iommu_group_num)
927 int __rte_experimental
928 rte_vfio_container_dma_map(__rte_unused int container_fd,
929 __rte_unused uint64_t vaddr,
930 __rte_unused uint64_t iova,
931 __rte_unused uint64_t len)
936 int __rte_experimental
937 rte_vfio_container_dma_unmap(__rte_unused int container_fd,
938 __rte_unused uint64_t vaddr,
939 __rte_unused uint64_t iova,
940 __rte_unused uint64_t len)