1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation.
3 * Copyright(c) 2014 6WIND S.A.
4 */
5
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <stdint.h>
9 #include <string.h>
10 #include <stdarg.h>
11 #include <unistd.h>
12 #include <pthread.h>
13 #include <syslog.h>
14 #include <getopt.h>
15 #include <sys/file.h>
16 #include <stddef.h>
17 #include <errno.h>
18 #include <limits.h>
19 #include <sys/mman.h>
20 #include <sys/queue.h>
21 #include <sys/stat.h>
22
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>
28 #include <rte_eal.h>
29 #include <rte_errno.h>
30 #include <rte_per_lcore.h>
31 #include <rte_lcore.h>
32 #include <rte_service_component.h>
33 #include <rte_log.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>
39 #include <rte_bus.h>
40 #include <rte_dev.h>
41 #include <rte_devargs.h>
42 #include <rte_version.h>
43 #include <rte_vfio.h>
44 #include <malloc_heap.h>
45 #include <rte_telemetry.h>
46
47 #include "eal_private.h"
48 #include "eal_thread.h"
49 #include "eal_internal_cfg.h"
50 #include "eal_filesystem.h"
51 #include "eal_hugepages.h"
52 #include "eal_options.h"
53 #include "eal_memcfg.h"
54 #include "eal_trace.h"
55
56 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
57
58 /* define fd variable here, because file needs to be kept open for the
59 * duration of the program, as we hold a write lock on it in the primary proc */
60 static int mem_cfg_fd = -1;
61
62 static struct flock wr_lock = {
63 .l_type = F_WRLCK,
64 .l_whence = SEEK_SET,
65 .l_start = offsetof(struct rte_mem_config, memsegs),
66 .l_len = RTE_SIZEOF_FIELD(struct rte_mem_config, memsegs),
67 };
68
69 /* internal configuration (per-core) */
70 struct lcore_config lcore_config[RTE_MAX_LCORE];
71
72 /* used by rte_rdtsc() */
73 int rte_cycles_vmware_tsc_map;
74
75 static const char *default_runtime_dir = "/var/run";
76
77 int
eal_create_runtime_dir(void)78 eal_create_runtime_dir(void)
79 {
80 const char *directory = default_runtime_dir;
81 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
82 const char *fallback = "/tmp";
83 char run_dir[PATH_MAX];
84 char tmp[PATH_MAX];
85 int ret;
86
87 if (getuid() != 0) {
88 /* try XDG path first, fall back to /tmp */
89 if (xdg_runtime_dir != NULL)
90 directory = xdg_runtime_dir;
91 else
92 directory = fallback;
93 }
94 /* create DPDK subdirectory under runtime dir */
95 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
96 if (ret < 0 || ret == sizeof(tmp)) {
97 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
98 return -1;
99 }
100
101 /* create prefix-specific subdirectory under DPDK runtime dir */
102 ret = snprintf(run_dir, sizeof(run_dir), "%s/%s",
103 tmp, eal_get_hugefile_prefix());
104 if (ret < 0 || ret == sizeof(run_dir)) {
105 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
106 return -1;
107 }
108
109 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
110 * step by step.
111 */
112 ret = mkdir(tmp, 0700);
113 if (ret < 0 && errno != EEXIST) {
114 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
115 tmp, strerror(errno));
116 return -1;
117 }
118
119 ret = mkdir(run_dir, 0700);
120 if (ret < 0 && errno != EEXIST) {
121 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
122 run_dir, strerror(errno));
123 return -1;
124 }
125
126 if (eal_set_runtime_dir(run_dir, sizeof(run_dir)))
127 return -1;
128
129 return 0;
130 }
131
132 int
eal_clean_runtime_dir(void)133 eal_clean_runtime_dir(void)
134 {
135 /* FreeBSD doesn't need this implemented for now, because, unlike Linux,
136 * FreeBSD doesn't create per-process files, so no need to clean up.
137 */
138 return 0;
139 }
140
141 /* parse a sysfs (or other) file containing one integer value */
142 int
eal_parse_sysfs_value(const char * filename,unsigned long * val)143 eal_parse_sysfs_value(const char *filename, unsigned long *val)
144 {
145 FILE *f;
146 char buf[BUFSIZ];
147 char *end = NULL;
148
149 if ((f = fopen(filename, "r")) == NULL) {
150 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
151 __func__, filename);
152 return -1;
153 }
154
155 if (fgets(buf, sizeof(buf), f) == NULL) {
156 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
157 __func__, filename);
158 fclose(f);
159 return -1;
160 }
161 *val = strtoul(buf, &end, 0);
162 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
163 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
164 __func__, filename);
165 fclose(f);
166 return -1;
167 }
168 fclose(f);
169 return 0;
170 }
171
172
173 /* create memory configuration in shared/mmap memory. Take out
174 * a write lock on the memsegs, so we can auto-detect primary/secondary.
175 * This means we never close the file while running (auto-close on exit).
176 * We also don't lock the whole file, so that in future we can use read-locks
177 * on other parts, e.g. memzones, to detect if there are running secondary
178 * processes. */
179 static int
rte_eal_config_create(void)180 rte_eal_config_create(void)
181 {
182 struct rte_config *config = rte_eal_get_configuration();
183 const struct internal_config *internal_conf =
184 eal_get_internal_configuration();
185 size_t page_sz = sysconf(_SC_PAGE_SIZE);
186 size_t cfg_len = sizeof(struct rte_mem_config);
187 size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
188 void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
189 int retval;
190
191 const char *pathname = eal_runtime_config_path();
192
193 if (internal_conf->no_shconf)
194 return 0;
195
196 /* map the config before base address so that we don't waste a page */
197 if (internal_conf->base_virtaddr != 0)
198 rte_mem_cfg_addr = (void *)
199 RTE_ALIGN_FLOOR(internal_conf->base_virtaddr -
200 sizeof(struct rte_mem_config), page_sz);
201 else
202 rte_mem_cfg_addr = NULL;
203
204 if (mem_cfg_fd < 0){
205 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
206 if (mem_cfg_fd < 0) {
207 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
208 pathname);
209 return -1;
210 }
211 }
212
213 retval = ftruncate(mem_cfg_fd, cfg_len);
214 if (retval < 0){
215 close(mem_cfg_fd);
216 mem_cfg_fd = -1;
217 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
218 pathname);
219 return -1;
220 }
221
222 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
223 if (retval < 0){
224 close(mem_cfg_fd);
225 mem_cfg_fd = -1;
226 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
227 "process running?\n", pathname);
228 return -1;
229 }
230
231 /* reserve space for config */
232 rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
233 &cfg_len_aligned, page_sz, 0, 0);
234 if (rte_mem_cfg_addr == NULL) {
235 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
236 close(mem_cfg_fd);
237 mem_cfg_fd = -1;
238 return -1;
239 }
240
241 /* remap the actual file into the space we've just reserved */
242 mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
243 cfg_len_aligned, PROT_READ | PROT_WRITE,
244 MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
245 if (mapped_mem_cfg_addr == MAP_FAILED) {
246 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
247 munmap(rte_mem_cfg_addr, cfg_len);
248 close(mem_cfg_fd);
249 mem_cfg_fd = -1;
250 return -1;
251 }
252
253 memcpy(rte_mem_cfg_addr, config->mem_config, sizeof(struct rte_mem_config));
254 config->mem_config = rte_mem_cfg_addr;
255
256 /* store address of the config in the config itself so that secondary
257 * processes could later map the config into this exact location
258 */
259 config->mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
260 return 0;
261 }
262
263 /* attach to an existing shared memory config */
264 static int
rte_eal_config_attach(void)265 rte_eal_config_attach(void)
266 {
267 void *rte_mem_cfg_addr;
268 const char *pathname = eal_runtime_config_path();
269 struct rte_config *config = rte_eal_get_configuration();
270 const struct internal_config *internal_conf =
271 eal_get_internal_configuration();
272
273
274 if (internal_conf->no_shconf)
275 return 0;
276
277 if (mem_cfg_fd < 0){
278 mem_cfg_fd = open(pathname, O_RDWR);
279 if (mem_cfg_fd < 0) {
280 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
281 pathname);
282 return -1;
283 }
284 }
285
286 rte_mem_cfg_addr = mmap(NULL, sizeof(*config->mem_config),
287 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
288 /* don't close the fd here, it will be closed on reattach */
289 if (rte_mem_cfg_addr == MAP_FAILED) {
290 close(mem_cfg_fd);
291 mem_cfg_fd = -1;
292 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
293 errno, strerror(errno));
294 return -1;
295 }
296
297 config->mem_config = rte_mem_cfg_addr;
298
299 return 0;
300 }
301
302 /* reattach the shared config at exact memory location primary process has it */
303 static int
rte_eal_config_reattach(void)304 rte_eal_config_reattach(void)
305 {
306 struct rte_mem_config *mem_config;
307 void *rte_mem_cfg_addr;
308 struct rte_config *config = rte_eal_get_configuration();
309 const struct internal_config *internal_conf =
310 eal_get_internal_configuration();
311
312 if (internal_conf->no_shconf)
313 return 0;
314
315 /* save the address primary process has mapped shared config to */
316 rte_mem_cfg_addr =
317 (void *)(uintptr_t)config->mem_config->mem_cfg_addr;
318
319 /* unmap original config */
320 munmap(config->mem_config, sizeof(struct rte_mem_config));
321
322 /* remap the config at proper address */
323 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
324 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
325 mem_cfg_fd, 0);
326 close(mem_cfg_fd);
327 mem_cfg_fd = -1;
328
329 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
330 if (mem_config != MAP_FAILED) {
331 /* errno is stale, don't use */
332 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
333 " - please use '--" OPT_BASE_VIRTADDR
334 "' option\n",
335 rte_mem_cfg_addr, mem_config);
336 munmap(mem_config, sizeof(struct rte_mem_config));
337 return -1;
338 }
339 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
340 errno, strerror(errno));
341 return -1;
342 }
343
344 config->mem_config = mem_config;
345
346 return 0;
347 }
348
349 /* Detect if we are a primary or a secondary process */
350 enum rte_proc_type_t
eal_proc_type_detect(void)351 eal_proc_type_detect(void)
352 {
353 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
354 const char *pathname = eal_runtime_config_path();
355 const struct internal_config *internal_conf =
356 eal_get_internal_configuration();
357
358 /* if there no shared config, there can be no secondary processes */
359 if (!internal_conf->no_shconf) {
360 /* if we can open the file but not get a write-lock we are a
361 * secondary process. NOTE: if we get a file handle back, we
362 * keep that open and don't close it to prevent a race condition
363 * between multiple opens.
364 */
365 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
366 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
367 ptype = RTE_PROC_SECONDARY;
368 }
369
370 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
371 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
372
373 return ptype;
374 }
375
376 /* Sets up rte_config structure with the pointer to shared memory config.*/
377 static int
rte_config_init(void)378 rte_config_init(void)
379 {
380 struct rte_config *config = rte_eal_get_configuration();
381 const struct internal_config *internal_conf =
382 eal_get_internal_configuration();
383
384 config->process_type = internal_conf->process_type;
385
386 switch (config->process_type) {
387 case RTE_PROC_PRIMARY:
388 if (rte_eal_config_create() < 0)
389 return -1;
390 eal_mcfg_update_from_internal();
391 break;
392 case RTE_PROC_SECONDARY:
393 if (rte_eal_config_attach() < 0)
394 return -1;
395 eal_mcfg_wait_complete();
396 if (eal_mcfg_check_version() < 0) {
397 RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
398 return -1;
399 }
400 if (rte_eal_config_reattach() < 0)
401 return -1;
402 if (!__rte_mp_enable()) {
403 RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
404 return -1;
405 }
406 eal_mcfg_update_internal();
407 break;
408 case RTE_PROC_AUTO:
409 case RTE_PROC_INVALID:
410 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
411 config->process_type);
412 return -1;
413 }
414
415 return 0;
416 }
417
418 /* display usage */
419 static void
eal_usage(const char * prgname)420 eal_usage(const char *prgname)
421 {
422 rte_usage_hook_t hook = eal_get_application_usage_hook();
423
424 printf("\nUsage: %s ", prgname);
425 eal_common_usage();
426 /* Allow the application to print its usage message too if hook is set */
427 if (hook) {
428 printf("===== Application Usage =====\n\n");
429 (hook)(prgname);
430 }
431 }
432
433 static inline size_t
eal_get_hugepage_mem_size(void)434 eal_get_hugepage_mem_size(void)
435 {
436 uint64_t size = 0;
437 unsigned i, j;
438 struct internal_config *internal_conf =
439 eal_get_internal_configuration();
440
441 for (i = 0; i < internal_conf->num_hugepage_sizes; i++) {
442 struct hugepage_info *hpi = &internal_conf->hugepage_info[i];
443 if (strnlen(hpi->hugedir, sizeof(hpi->hugedir)) != 0) {
444 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
445 size += hpi->hugepage_sz * hpi->num_pages[j];
446 }
447 }
448 }
449
450 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
451 }
452
453 /* Parse the arguments for --log-level only */
454 static void
eal_log_level_parse(int argc,char ** argv)455 eal_log_level_parse(int argc, char **argv)
456 {
457 int opt;
458 char **argvopt;
459 int option_index;
460 const int old_optind = optind;
461 const int old_optopt = optopt;
462 const int old_optreset = optreset;
463 char * const old_optarg = optarg;
464 struct internal_config *internal_conf =
465 eal_get_internal_configuration();
466
467 argvopt = argv;
468 optind = 1;
469 optreset = 1;
470
471 while ((opt = getopt_long(argc, argvopt, eal_short_options,
472 eal_long_options, &option_index)) != EOF) {
473
474 int ret;
475
476 /* getopt is not happy, stop right now */
477 if (opt == '?')
478 break;
479
480 ret = (opt == OPT_LOG_LEVEL_NUM) ?
481 eal_parse_common_option(opt, optarg, internal_conf) : 0;
482
483 /* common parser is not happy */
484 if (ret < 0)
485 break;
486 }
487
488 /* restore getopt lib */
489 optind = old_optind;
490 optopt = old_optopt;
491 optreset = old_optreset;
492 optarg = old_optarg;
493 }
494
495 /* Parse the argument given in the command line of the application */
496 static int
eal_parse_args(int argc,char ** argv)497 eal_parse_args(int argc, char **argv)
498 {
499 int opt, ret;
500 char **argvopt;
501 int option_index;
502 char *prgname = argv[0];
503 const int old_optind = optind;
504 const int old_optopt = optopt;
505 const int old_optreset = optreset;
506 char * const old_optarg = optarg;
507 struct internal_config *internal_conf =
508 eal_get_internal_configuration();
509
510 argvopt = argv;
511 optind = 1;
512 optreset = 1;
513
514 while ((opt = getopt_long(argc, argvopt, eal_short_options,
515 eal_long_options, &option_index)) != EOF) {
516
517 /* getopt didn't recognise the option */
518 if (opt == '?') {
519 eal_usage(prgname);
520 ret = -1;
521 goto out;
522 }
523
524 ret = eal_parse_common_option(opt, optarg, internal_conf);
525 /* common parser is not happy */
526 if (ret < 0) {
527 eal_usage(prgname);
528 ret = -1;
529 goto out;
530 }
531 /* common parser handled this option */
532 if (ret == 0)
533 continue;
534
535 switch (opt) {
536 case OPT_MBUF_POOL_OPS_NAME_NUM:
537 {
538 char *ops_name = strdup(optarg);
539 if (ops_name == NULL)
540 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
541 else {
542 /* free old ops name */
543 if (internal_conf->user_mbuf_pool_ops_name !=
544 NULL)
545 free(internal_conf->user_mbuf_pool_ops_name);
546
547 internal_conf->user_mbuf_pool_ops_name =
548 ops_name;
549 }
550 break;
551 }
552 case 'h':
553 eal_usage(prgname);
554 exit(EXIT_SUCCESS);
555 default:
556 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
557 RTE_LOG(ERR, EAL, "Option %c is not supported "
558 "on FreeBSD\n", opt);
559 } else if (opt >= OPT_LONG_MIN_NUM &&
560 opt < OPT_LONG_MAX_NUM) {
561 RTE_LOG(ERR, EAL, "Option %s is not supported "
562 "on FreeBSD\n",
563 eal_long_options[option_index].name);
564 } else {
565 RTE_LOG(ERR, EAL, "Option %d is not supported "
566 "on FreeBSD\n", opt);
567 }
568 eal_usage(prgname);
569 ret = -1;
570 goto out;
571 }
572 }
573
574 /* create runtime data directory */
575 if (internal_conf->no_shconf == 0 &&
576 eal_create_runtime_dir() < 0) {
577 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
578 ret = -1;
579 goto out;
580 }
581
582 if (eal_adjust_config(internal_conf) != 0) {
583 ret = -1;
584 goto out;
585 }
586
587 /* sanity checks */
588 if (eal_check_common_options(internal_conf) != 0) {
589 eal_usage(prgname);
590 ret = -1;
591 goto out;
592 }
593
594 if (optind >= 0)
595 argv[optind-1] = prgname;
596 ret = optind-1;
597
598 out:
599 /* restore getopt lib */
600 optind = old_optind;
601 optopt = old_optopt;
602 optreset = old_optreset;
603 optarg = old_optarg;
604
605 return ret;
606 }
607
608 static int
check_socket(const struct rte_memseg_list * msl,void * arg)609 check_socket(const struct rte_memseg_list *msl, void *arg)
610 {
611 int *socket_id = arg;
612
613 if (msl->external)
614 return 0;
615
616 if (msl->socket_id == *socket_id && msl->memseg_arr.count != 0)
617 return 1;
618
619 return 0;
620 }
621
622 static void
eal_check_mem_on_local_socket(void)623 eal_check_mem_on_local_socket(void)
624 {
625 int socket_id;
626 const struct rte_config *config = rte_eal_get_configuration();
627
628 socket_id = rte_lcore_to_socket_id(config->main_lcore);
629
630 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
631 RTE_LOG(WARNING, EAL, "WARNING: Main core has no memory on local socket!\n");
632 }
633
634
635 static int
sync_func(__rte_unused void * arg)636 sync_func(__rte_unused void *arg)
637 {
638 return 0;
639 }
640 /* Abstraction for port I/0 privilege */
641 int
rte_eal_iopl_init(void)642 rte_eal_iopl_init(void)
643 {
644 static int fd = -1;
645
646 if (fd < 0)
647 fd = open("/dev/io", O_RDWR);
648
649 if (fd < 0)
650 return -1;
651 /* keep fd open for iopl */
652 return 0;
653 }
654
rte_eal_init_alert(const char * msg)655 static void rte_eal_init_alert(const char *msg)
656 {
657 fprintf(stderr, "EAL: FATAL: %s\n", msg);
658 RTE_LOG(ERR, EAL, "%s\n", msg);
659 }
660
661 /* Launch threads, called at application init(). */
662 int
rte_eal_init(int argc,char ** argv)663 rte_eal_init(int argc, char **argv)
664 {
665 int i, fctret, ret;
666 pthread_t thread_id;
667 static uint32_t run_once;
668 uint32_t has_run = 0;
669 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
670 char thread_name[RTE_MAX_THREAD_NAME_LEN];
671 const struct rte_config *config = rte_eal_get_configuration();
672 struct internal_config *internal_conf =
673 eal_get_internal_configuration();
674
675 /* checks if the machine is adequate */
676 if (!rte_cpu_is_supported()) {
677 rte_eal_init_alert("unsupported cpu type.");
678 rte_errno = ENOTSUP;
679 return -1;
680 }
681
682 if (!__atomic_compare_exchange_n(&run_once, &has_run, 1, 0,
683 __ATOMIC_RELAXED, __ATOMIC_RELAXED)) {
684 rte_eal_init_alert("already called initialization.");
685 rte_errno = EALREADY;
686 return -1;
687 }
688
689 thread_id = pthread_self();
690
691 eal_reset_internal_config(internal_conf);
692
693 /* clone argv to report out later in telemetry */
694 eal_save_args(argc, argv);
695
696 /* set log level as early as possible */
697 eal_log_level_parse(argc, argv);
698
699 if (rte_eal_cpu_init() < 0) {
700 rte_eal_init_alert("Cannot detect lcores.");
701 rte_errno = ENOTSUP;
702 return -1;
703 }
704
705 fctret = eal_parse_args(argc, argv);
706 if (fctret < 0) {
707 rte_eal_init_alert("Invalid 'command line' arguments.");
708 rte_errno = EINVAL;
709 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
710 return -1;
711 }
712
713 /* FreeBSD always uses legacy memory model */
714 internal_conf->legacy_mem = true;
715
716 if (eal_plugins_init() < 0) {
717 rte_eal_init_alert("Cannot init plugins");
718 rte_errno = EINVAL;
719 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
720 return -1;
721 }
722
723 if (eal_trace_init() < 0) {
724 rte_eal_init_alert("Cannot init trace");
725 rte_errno = EFAULT;
726 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
727 return -1;
728 }
729
730 if (eal_option_device_parse()) {
731 rte_errno = ENODEV;
732 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
733 return -1;
734 }
735
736 if (rte_config_init() < 0) {
737 rte_eal_init_alert("Cannot init config");
738 return -1;
739 }
740
741 if (rte_eal_intr_init() < 0) {
742 rte_eal_init_alert("Cannot init interrupt-handling thread");
743 return -1;
744 }
745
746 if (rte_eal_alarm_init() < 0) {
747 rte_eal_init_alert("Cannot init alarm");
748 /* rte_eal_alarm_init sets rte_errno on failure. */
749 return -1;
750 }
751
752 /* Put mp channel init before bus scan so that we can init the vdev
753 * bus through mp channel in the secondary process before the bus scan.
754 */
755 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
756 rte_eal_init_alert("failed to init mp channel");
757 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
758 rte_errno = EFAULT;
759 return -1;
760 }
761 }
762
763 if (rte_bus_scan()) {
764 rte_eal_init_alert("Cannot scan the buses for devices");
765 rte_errno = ENODEV;
766 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
767 return -1;
768 }
769
770 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
771 if (internal_conf->iova_mode == RTE_IOVA_DC) {
772 /* autodetect the IOVA mapping mode (default is RTE_IOVA_PA) */
773 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
774
775 if (iova_mode == RTE_IOVA_DC)
776 iova_mode = RTE_IOVA_PA;
777 rte_eal_get_configuration()->iova_mode = iova_mode;
778 } else {
779 rte_eal_get_configuration()->iova_mode =
780 internal_conf->iova_mode;
781 }
782
783 RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
784 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
785
786 if (internal_conf->no_hugetlbfs == 0) {
787 /* rte_config isn't initialized yet */
788 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
789 eal_hugepage_info_init() :
790 eal_hugepage_info_read();
791 if (ret < 0) {
792 rte_eal_init_alert("Cannot get hugepage information.");
793 rte_errno = EACCES;
794 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
795 return -1;
796 }
797 }
798
799 if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
800 if (internal_conf->no_hugetlbfs)
801 internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
802 else
803 internal_conf->memory = eal_get_hugepage_mem_size();
804 }
805
806 if (internal_conf->vmware_tsc_map == 1) {
807 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
808 rte_cycles_vmware_tsc_map = 1;
809 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
810 "you must have monitor_control.pseudo_perfctr = TRUE\n");
811 #else
812 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
813 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
814 #endif
815 }
816
817 /* in secondary processes, memory init may allocate additional fbarrays
818 * not present in primary processes, so to avoid any potential issues,
819 * initialize memzones first.
820 */
821 if (rte_eal_memzone_init() < 0) {
822 rte_eal_init_alert("Cannot init memzone");
823 rte_errno = ENODEV;
824 return -1;
825 }
826
827 if (rte_eal_memory_init() < 0) {
828 rte_eal_init_alert("Cannot init memory");
829 rte_errno = ENOMEM;
830 return -1;
831 }
832
833 if (rte_eal_malloc_heap_init() < 0) {
834 rte_eal_init_alert("Cannot init malloc heap");
835 rte_errno = ENODEV;
836 return -1;
837 }
838
839 if (rte_eal_tailqs_init() < 0) {
840 rte_eal_init_alert("Cannot init tail queues for objects");
841 rte_errno = EFAULT;
842 return -1;
843 }
844
845 if (rte_eal_timer_init() < 0) {
846 rte_eal_init_alert("Cannot init HPET or TSC timers");
847 rte_errno = ENOTSUP;
848 return -1;
849 }
850
851 eal_check_mem_on_local_socket();
852
853 if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
854 &lcore_config[config->main_lcore].cpuset) != 0) {
855 rte_eal_init_alert("Cannot set affinity");
856 rte_errno = EINVAL;
857 return -1;
858 }
859 __rte_thread_init(config->main_lcore,
860 &lcore_config[config->main_lcore].cpuset);
861
862 ret = eal_thread_dump_current_affinity(cpuset, sizeof(cpuset));
863
864 RTE_LOG(DEBUG, EAL, "Main lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
865 config->main_lcore, thread_id, cpuset,
866 ret == 0 ? "" : "...");
867
868 RTE_LCORE_FOREACH_WORKER(i) {
869
870 /*
871 * create communication pipes between main thread
872 * and children
873 */
874 if (pipe(lcore_config[i].pipe_main2worker) < 0)
875 rte_panic("Cannot create pipe\n");
876 if (pipe(lcore_config[i].pipe_worker2main) < 0)
877 rte_panic("Cannot create pipe\n");
878
879 lcore_config[i].state = WAIT;
880
881 /* create a thread for each lcore */
882 ret = pthread_create(&lcore_config[i].thread_id, NULL,
883 eal_thread_loop, NULL);
884 if (ret != 0)
885 rte_panic("Cannot create thread\n");
886
887 /* Set thread_name for aid in debugging. */
888 snprintf(thread_name, sizeof(thread_name),
889 "lcore-worker-%d", i);
890 rte_thread_setname(lcore_config[i].thread_id, thread_name);
891
892 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
893 sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
894 if (ret != 0)
895 rte_panic("Cannot set affinity\n");
896 }
897
898 /*
899 * Launch a dummy function on all worker lcores, so that main lcore
900 * knows they are all ready when this function returns.
901 */
902 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MAIN);
903 rte_eal_mp_wait_lcore();
904
905 /* initialize services so vdevs register service during bus_probe. */
906 ret = rte_service_init();
907 if (ret) {
908 rte_eal_init_alert("rte_service_init() failed");
909 rte_errno = ENOEXEC;
910 return -1;
911 }
912
913 /* Probe all the buses and devices/drivers on them */
914 if (rte_bus_probe()) {
915 rte_eal_init_alert("Cannot probe devices");
916 rte_errno = ENOTSUP;
917 return -1;
918 }
919
920 /* initialize default service/lcore mappings and start running. Ignore
921 * -ENOTSUP, as it indicates no service coremask passed to EAL.
922 */
923 ret = rte_service_start_with_defaults();
924 if (ret < 0 && ret != -ENOTSUP) {
925 rte_errno = ENOEXEC;
926 return -1;
927 }
928
929 /*
930 * Clean up unused files in runtime directory. We do this at the end of
931 * init and not at the beginning because we want to clean stuff up
932 * whether we are primary or secondary process, but we cannot remove
933 * primary process' files because secondary should be able to run even
934 * if primary process is dead.
935 *
936 * In no_shconf mode, no runtime directory is created in the first
937 * place, so no cleanup needed.
938 */
939 if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
940 rte_eal_init_alert("Cannot clear runtime directory");
941 return -1;
942 }
943 if (!internal_conf->no_telemetry) {
944 const char *error_str = NULL;
945 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
946 &internal_conf->ctrl_cpuset, &error_str)
947 != 0) {
948 rte_eal_init_alert(error_str);
949 return -1;
950 }
951 if (error_str != NULL)
952 RTE_LOG(NOTICE, EAL, "%s\n", error_str);
953 }
954
955 eal_mcfg_complete();
956
957 return fctret;
958 }
959
960 int
rte_eal_cleanup(void)961 rte_eal_cleanup(void)
962 {
963 struct internal_config *internal_conf =
964 eal_get_internal_configuration();
965 rte_service_finalize();
966 rte_mp_channel_cleanup();
967 rte_trace_save();
968 eal_trace_fini();
969 eal_cleanup_config(internal_conf);
970 return 0;
971 }
972
rte_eal_create_uio_dev(void)973 int rte_eal_create_uio_dev(void)
974 {
975 const struct internal_config *internal_conf =
976 eal_get_internal_configuration();
977 return internal_conf->create_uio_dev;
978 }
979
980 enum rte_intr_mode
rte_eal_vfio_intr_mode(void)981 rte_eal_vfio_intr_mode(void)
982 {
983 return RTE_INTR_MODE_NONE;
984 }
985
986 void
rte_eal_vfio_get_vf_token(__rte_unused rte_uuid_t vf_token)987 rte_eal_vfio_get_vf_token(__rte_unused rte_uuid_t vf_token)
988 {
989 }
990
rte_vfio_setup_device(__rte_unused const char * sysfs_base,__rte_unused const char * dev_addr,__rte_unused int * vfio_dev_fd,__rte_unused struct vfio_device_info * device_info)991 int rte_vfio_setup_device(__rte_unused const char *sysfs_base,
992 __rte_unused const char *dev_addr,
993 __rte_unused int *vfio_dev_fd,
994 __rte_unused struct vfio_device_info *device_info)
995 {
996 return -1;
997 }
998
rte_vfio_release_device(__rte_unused const char * sysfs_base,__rte_unused const char * dev_addr,__rte_unused int fd)999 int rte_vfio_release_device(__rte_unused const char *sysfs_base,
1000 __rte_unused const char *dev_addr,
1001 __rte_unused int fd)
1002 {
1003 return -1;
1004 }
1005
rte_vfio_enable(__rte_unused const char * modname)1006 int rte_vfio_enable(__rte_unused const char *modname)
1007 {
1008 return -1;
1009 }
1010
rte_vfio_is_enabled(__rte_unused const char * modname)1011 int rte_vfio_is_enabled(__rte_unused const char *modname)
1012 {
1013 return 0;
1014 }
1015
rte_vfio_noiommu_is_enabled(void)1016 int rte_vfio_noiommu_is_enabled(void)
1017 {
1018 return 0;
1019 }
1020
rte_vfio_clear_group(__rte_unused int vfio_group_fd)1021 int rte_vfio_clear_group(__rte_unused int vfio_group_fd)
1022 {
1023 return 0;
1024 }
1025
1026 int
rte_vfio_get_group_num(__rte_unused const char * sysfs_base,__rte_unused const char * dev_addr,__rte_unused int * iommu_group_num)1027 rte_vfio_get_group_num(__rte_unused const char *sysfs_base,
1028 __rte_unused const char *dev_addr,
1029 __rte_unused int *iommu_group_num)
1030 {
1031 return -1;
1032 }
1033
1034 int
rte_vfio_get_container_fd(void)1035 rte_vfio_get_container_fd(void)
1036 {
1037 return -1;
1038 }
1039
1040 int
rte_vfio_get_group_fd(__rte_unused int iommu_group_num)1041 rte_vfio_get_group_fd(__rte_unused int iommu_group_num)
1042 {
1043 return -1;
1044 }
1045
1046 int
rte_vfio_container_create(void)1047 rte_vfio_container_create(void)
1048 {
1049 return -1;
1050 }
1051
1052 int
rte_vfio_container_destroy(__rte_unused int container_fd)1053 rte_vfio_container_destroy(__rte_unused int container_fd)
1054 {
1055 return -1;
1056 }
1057
1058 int
rte_vfio_container_group_bind(__rte_unused int container_fd,__rte_unused int iommu_group_num)1059 rte_vfio_container_group_bind(__rte_unused int container_fd,
1060 __rte_unused int iommu_group_num)
1061 {
1062 return -1;
1063 }
1064
1065 int
rte_vfio_container_group_unbind(__rte_unused int container_fd,__rte_unused int iommu_group_num)1066 rte_vfio_container_group_unbind(__rte_unused int container_fd,
1067 __rte_unused int iommu_group_num)
1068 {
1069 return -1;
1070 }
1071
1072 int
rte_vfio_container_dma_map(__rte_unused int container_fd,__rte_unused uint64_t vaddr,__rte_unused uint64_t iova,__rte_unused uint64_t len)1073 rte_vfio_container_dma_map(__rte_unused int container_fd,
1074 __rte_unused uint64_t vaddr,
1075 __rte_unused uint64_t iova,
1076 __rte_unused uint64_t len)
1077 {
1078 return -1;
1079 }
1080
1081 int
rte_vfio_container_dma_unmap(__rte_unused int container_fd,__rte_unused uint64_t vaddr,__rte_unused uint64_t iova,__rte_unused uint64_t len)1082 rte_vfio_container_dma_unmap(__rte_unused int container_fd,
1083 __rte_unused uint64_t vaddr,
1084 __rte_unused uint64_t iova,
1085 __rte_unused uint64_t len)
1086 {
1087 return -1;
1088 }
1089