1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation.
3 * Copyright(c) 2012-2014 6WIND S.A.
4 */
5
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <stdint.h>
9 #include <string.h>
10 #include <unistd.h>
11 #include <pthread.h>
12 #include <getopt.h>
13 #include <sys/file.h>
14 #include <dirent.h>
15 #include <fcntl.h>
16 #include <fnmatch.h>
17 #include <stddef.h>
18 #include <errno.h>
19 #include <limits.h>
20 #include <sys/mman.h>
21 #include <sys/stat.h>
22 #if defined(RTE_ARCH_X86)
23 #include <sys/io.h>
24 #endif
25 #include <linux/version.h>
26
27 #include <rte_common.h>
28 #include <rte_debug.h>
29 #include <rte_memory.h>
30 #include <rte_launch.h>
31 #include <rte_eal.h>
32 #include <rte_errno.h>
33 #include <rte_lcore.h>
34 #include <rte_service_component.h>
35 #include <rte_log.h>
36 #include <rte_string_fns.h>
37 #include <rte_cpuflags.h>
38 #include <rte_bus.h>
39 #include <rte_version.h>
40 #include <malloc_heap.h>
41 #include <rte_vfio.h>
42
43 #include <telemetry_internal.h>
44 #include "eal_private.h"
45 #include "eal_thread.h"
46 #include "eal_internal_cfg.h"
47 #include "eal_filesystem.h"
48 #include "eal_hugepages.h"
49 #include "eal_memcfg.h"
50 #include "eal_trace.h"
51 #include "eal_log.h"
52 #include "eal_options.h"
53 #include "eal_vfio.h"
54 #include "hotplug_mp.h"
55
56 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
57
58 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
59
60 #define KERNEL_IOMMU_GROUPS_PATH "/sys/kernel/iommu_groups"
61
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;
65
66 static struct flock wr_lock = {
67 .l_type = F_WRLCK,
68 .l_whence = SEEK_SET,
69 .l_start = offsetof(struct rte_mem_config, memsegs),
70 .l_len = RTE_SIZEOF_FIELD(struct rte_mem_config, memsegs),
71 };
72
73 /* internal configuration (per-core) */
74 struct lcore_config lcore_config[RTE_MAX_LCORE];
75
76 /* used by rte_rdtsc() */
77 int rte_cycles_vmware_tsc_map;
78
79
80 int
eal_clean_runtime_dir(void)81 eal_clean_runtime_dir(void)
82 {
83 const char *runtime_dir = rte_eal_get_runtime_dir();
84 DIR *dir;
85 struct dirent *dirent;
86 int dir_fd, fd, lck_result;
87 static const char * const filters[] = {
88 "fbarray_*",
89 "mp_socket_*"
90 };
91
92 /* open directory */
93 dir = opendir(runtime_dir);
94 if (!dir) {
95 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
96 runtime_dir);
97 goto error;
98 }
99 dir_fd = dirfd(dir);
100
101 /* lock the directory before doing anything, to avoid races */
102 if (flock(dir_fd, LOCK_EX) < 0) {
103 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
104 runtime_dir);
105 goto error;
106 }
107
108 dirent = readdir(dir);
109 if (!dirent) {
110 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
111 runtime_dir);
112 goto error;
113 }
114
115 while (dirent != NULL) {
116 unsigned int f_idx;
117 bool skip = true;
118
119 /* skip files that don't match the patterns */
120 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
121 const char *filter = filters[f_idx];
122
123 if (fnmatch(filter, dirent->d_name, 0) == 0) {
124 skip = false;
125 break;
126 }
127 }
128 if (skip) {
129 dirent = readdir(dir);
130 continue;
131 }
132
133 /* try and lock the file */
134 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
135
136 /* skip to next file */
137 if (fd == -1) {
138 dirent = readdir(dir);
139 continue;
140 }
141
142 /* non-blocking lock */
143 lck_result = flock(fd, LOCK_EX | LOCK_NB);
144
145 /* if lock succeeds, remove the file */
146 if (lck_result != -1)
147 unlinkat(dir_fd, dirent->d_name, 0);
148 close(fd);
149 dirent = readdir(dir);
150 }
151
152 /* closedir closes dir_fd and drops the lock */
153 closedir(dir);
154 return 0;
155
156 error:
157 if (dir)
158 closedir(dir);
159
160 RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
161 strerror(errno));
162
163 return -1;
164 }
165
166
167 /* create memory configuration in shared/mmap memory. Take out
168 * a write lock on the memsegs, so we can auto-detect primary/secondary.
169 * This means we never close the file while running (auto-close on exit).
170 * We also don't lock the whole file, so that in future we can use read-locks
171 * on other parts, e.g. memzones, to detect if there are running secondary
172 * processes. */
173 static int
rte_eal_config_create(void)174 rte_eal_config_create(void)
175 {
176 struct rte_config *config = rte_eal_get_configuration();
177 size_t page_sz = sysconf(_SC_PAGE_SIZE);
178 size_t cfg_len = sizeof(*config->mem_config);
179 size_t cfg_len_aligned = RTE_ALIGN(cfg_len, page_sz);
180 void *rte_mem_cfg_addr, *mapped_mem_cfg_addr;
181 int retval;
182 const struct internal_config *internal_conf =
183 eal_get_internal_configuration();
184
185 const char *pathname = eal_runtime_config_path();
186
187 if (internal_conf->no_shconf)
188 return 0;
189
190 /* map the config before hugepage address so that we don't waste a page */
191 if (internal_conf->base_virtaddr != 0)
192 rte_mem_cfg_addr = (void *)
193 RTE_ALIGN_FLOOR(internal_conf->base_virtaddr -
194 sizeof(struct rte_mem_config), page_sz);
195 else
196 rte_mem_cfg_addr = NULL;
197
198 if (mem_cfg_fd < 0){
199 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
200 if (mem_cfg_fd < 0) {
201 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
202 pathname);
203 return -1;
204 }
205 }
206
207 retval = ftruncate(mem_cfg_fd, cfg_len);
208 if (retval < 0){
209 close(mem_cfg_fd);
210 mem_cfg_fd = -1;
211 RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
212 pathname);
213 return -1;
214 }
215
216 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
217 if (retval < 0){
218 close(mem_cfg_fd);
219 mem_cfg_fd = -1;
220 RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
221 "process running?\n", pathname);
222 return -1;
223 }
224
225 /* reserve space for config */
226 rte_mem_cfg_addr = eal_get_virtual_area(rte_mem_cfg_addr,
227 &cfg_len_aligned, page_sz, 0, 0);
228 if (rte_mem_cfg_addr == NULL) {
229 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
230 close(mem_cfg_fd);
231 mem_cfg_fd = -1;
232 return -1;
233 }
234
235 /* remap the actual file into the space we've just reserved */
236 mapped_mem_cfg_addr = mmap(rte_mem_cfg_addr,
237 cfg_len_aligned, PROT_READ | PROT_WRITE,
238 MAP_SHARED | MAP_FIXED, mem_cfg_fd, 0);
239 if (mapped_mem_cfg_addr == MAP_FAILED) {
240 munmap(rte_mem_cfg_addr, cfg_len);
241 close(mem_cfg_fd);
242 mem_cfg_fd = -1;
243 RTE_LOG(ERR, EAL, "Cannot remap memory for rte_config\n");
244 return -1;
245 }
246
247 memcpy(rte_mem_cfg_addr, config->mem_config, sizeof(struct rte_mem_config));
248 config->mem_config = rte_mem_cfg_addr;
249
250 /* store address of the config in the config itself so that secondary
251 * processes could later map the config into this exact location
252 */
253 config->mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
254 config->mem_config->dma_maskbits = 0;
255
256 return 0;
257 }
258
259 /* attach to an existing shared memory config */
260 static int
rte_eal_config_attach(void)261 rte_eal_config_attach(void)
262 {
263 struct rte_config *config = rte_eal_get_configuration();
264 struct rte_mem_config *mem_config;
265 const struct internal_config *internal_conf =
266 eal_get_internal_configuration();
267
268 const char *pathname = eal_runtime_config_path();
269
270 if (internal_conf->no_shconf)
271 return 0;
272
273 if (mem_cfg_fd < 0){
274 mem_cfg_fd = open(pathname, O_RDWR);
275 if (mem_cfg_fd < 0) {
276 RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
277 pathname);
278 return -1;
279 }
280 }
281
282 /* map it as read-only first */
283 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
284 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
285 if (mem_config == MAP_FAILED) {
286 close(mem_cfg_fd);
287 mem_cfg_fd = -1;
288 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
289 errno, strerror(errno));
290 return -1;
291 }
292
293 config->mem_config = mem_config;
294
295 return 0;
296 }
297
298 /* reattach the shared config at exact memory location primary process has it */
299 static int
rte_eal_config_reattach(void)300 rte_eal_config_reattach(void)
301 {
302 struct rte_config *config = rte_eal_get_configuration();
303 struct rte_mem_config *mem_config;
304 void *rte_mem_cfg_addr;
305 const struct internal_config *internal_conf =
306 eal_get_internal_configuration();
307
308 if (internal_conf->no_shconf)
309 return 0;
310
311 /* save the address primary process has mapped shared config to */
312 rte_mem_cfg_addr =
313 (void *) (uintptr_t) config->mem_config->mem_cfg_addr;
314
315 /* unmap original config */
316 munmap(config->mem_config, sizeof(struct rte_mem_config));
317
318 /* remap the config at proper address */
319 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
320 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
321 mem_cfg_fd, 0);
322
323 close(mem_cfg_fd);
324 mem_cfg_fd = -1;
325
326 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
327 if (mem_config != MAP_FAILED) {
328 /* errno is stale, don't use */
329 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
330 " - please use '--" OPT_BASE_VIRTADDR
331 "' option\n", rte_mem_cfg_addr, mem_config);
332 munmap(mem_config, sizeof(struct rte_mem_config));
333 return -1;
334 }
335 RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
336 errno, strerror(errno));
337 return -1;
338 }
339
340 config->mem_config = mem_config;
341
342 return 0;
343 }
344
345 /* Detect if we are a primary or a secondary process */
346 enum rte_proc_type_t
eal_proc_type_detect(void)347 eal_proc_type_detect(void)
348 {
349 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
350 const char *pathname = eal_runtime_config_path();
351 const struct internal_config *internal_conf =
352 eal_get_internal_configuration();
353
354 /* if there no shared config, there can be no secondary processes */
355 if (!internal_conf->no_shconf) {
356 /* if we can open the file but not get a write-lock we are a
357 * secondary process. NOTE: if we get a file handle back, we
358 * keep that open and don't close it to prevent a race condition
359 * between multiple opens.
360 */
361 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
362 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
363 ptype = RTE_PROC_SECONDARY;
364 }
365
366 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
367 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
368
369 return ptype;
370 }
371
372 /* Sets up rte_config structure with the pointer to shared memory config.*/
373 static int
rte_config_init(void)374 rte_config_init(void)
375 {
376 struct rte_config *config = rte_eal_get_configuration();
377 const struct internal_config *internal_conf =
378 eal_get_internal_configuration();
379
380 config->process_type = internal_conf->process_type;
381
382 switch (config->process_type) {
383 case RTE_PROC_PRIMARY:
384 if (rte_eal_config_create() < 0)
385 return -1;
386 eal_mcfg_update_from_internal();
387 break;
388 case RTE_PROC_SECONDARY:
389 if (rte_eal_config_attach() < 0)
390 return -1;
391 eal_mcfg_wait_complete();
392 if (eal_mcfg_check_version() < 0) {
393 RTE_LOG(ERR, EAL, "Primary and secondary process DPDK version mismatch\n");
394 return -1;
395 }
396 if (rte_eal_config_reattach() < 0)
397 return -1;
398 if (!__rte_mp_enable()) {
399 RTE_LOG(ERR, EAL, "Primary process refused secondary attachment\n");
400 return -1;
401 }
402 eal_mcfg_update_internal();
403 break;
404 case RTE_PROC_AUTO:
405 case RTE_PROC_INVALID:
406 RTE_LOG(ERR, EAL, "Invalid process type %d\n",
407 config->process_type);
408 return -1;
409 }
410
411 return 0;
412 }
413
414 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
415 static void
eal_hugedirs_unlock(void)416 eal_hugedirs_unlock(void)
417 {
418 int i;
419 struct internal_config *internal_conf =
420 eal_get_internal_configuration();
421
422 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
423 {
424 /* skip uninitialized */
425 if (internal_conf->hugepage_info[i].lock_descriptor < 0)
426 continue;
427 /* unlock hugepage file */
428 flock(internal_conf->hugepage_info[i].lock_descriptor, LOCK_UN);
429 close(internal_conf->hugepage_info[i].lock_descriptor);
430 /* reset the field */
431 internal_conf->hugepage_info[i].lock_descriptor = -1;
432 }
433 }
434
435 /* display usage */
436 static void
eal_usage(const char * prgname)437 eal_usage(const char *prgname)
438 {
439 rte_usage_hook_t hook = eal_get_application_usage_hook();
440
441 printf("\nUsage: %s ", prgname);
442 eal_common_usage();
443 printf("EAL Linux options:\n"
444 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
445 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
446 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
447 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
448 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
449 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
450 " --"OPT_VFIO_VF_TOKEN" VF token (UUID) shared between SR-IOV PF and VFs\n"
451 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
452 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
453 " --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
454 "\n");
455 /* Allow the application to print its usage message too if hook is set */
456 if (hook) {
457 printf("===== Application Usage =====\n\n");
458 (hook)(prgname);
459 }
460 }
461
462 static int
eal_parse_socket_arg(char * strval,volatile uint64_t * socket_arg)463 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
464 {
465 char * arg[RTE_MAX_NUMA_NODES];
466 char *end;
467 int arg_num, i, len;
468
469 len = strnlen(strval, SOCKET_MEM_STRLEN);
470 if (len == SOCKET_MEM_STRLEN) {
471 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
472 return -1;
473 }
474
475 /* all other error cases will be caught later */
476 if (!isdigit(strval[len-1]))
477 return -1;
478
479 /* split the optarg into separate socket values */
480 arg_num = rte_strsplit(strval, len,
481 arg, RTE_MAX_NUMA_NODES, ',');
482
483 /* if split failed, or 0 arguments */
484 if (arg_num <= 0)
485 return -1;
486
487 /* parse each defined socket option */
488 errno = 0;
489 for (i = 0; i < arg_num; i++) {
490 uint64_t val;
491 end = NULL;
492 val = strtoull(arg[i], &end, 10);
493
494 /* check for invalid input */
495 if ((errno != 0) ||
496 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
497 return -1;
498 val <<= 20;
499 socket_arg[i] = val;
500 }
501
502 return 0;
503 }
504
505 static int
eal_parse_vfio_intr(const char * mode)506 eal_parse_vfio_intr(const char *mode)
507 {
508 struct internal_config *internal_conf =
509 eal_get_internal_configuration();
510 unsigned i;
511 static struct {
512 const char *name;
513 enum rte_intr_mode value;
514 } map[] = {
515 { "legacy", RTE_INTR_MODE_LEGACY },
516 { "msi", RTE_INTR_MODE_MSI },
517 { "msix", RTE_INTR_MODE_MSIX },
518 };
519
520 for (i = 0; i < RTE_DIM(map); i++) {
521 if (!strcmp(mode, map[i].name)) {
522 internal_conf->vfio_intr_mode = map[i].value;
523 return 0;
524 }
525 }
526 return -1;
527 }
528
529 static int
eal_parse_vfio_vf_token(const char * vf_token)530 eal_parse_vfio_vf_token(const char *vf_token)
531 {
532 struct internal_config *cfg = eal_get_internal_configuration();
533 rte_uuid_t uuid;
534
535 if (!rte_uuid_parse(vf_token, uuid)) {
536 rte_uuid_copy(cfg->vfio_vf_token, uuid);
537 return 0;
538 }
539
540 return -1;
541 }
542
543 /* Parse the arguments for --log-level only */
544 static void
eal_log_level_parse(int argc,char ** argv)545 eal_log_level_parse(int argc, char **argv)
546 {
547 int opt;
548 char **argvopt;
549 int option_index;
550 const int old_optind = optind;
551 const int old_optopt = optopt;
552 char * const old_optarg = optarg;
553 struct internal_config *internal_conf =
554 eal_get_internal_configuration();
555
556 argvopt = argv;
557 optind = 1;
558
559 while ((opt = getopt_long(argc, argvopt, eal_short_options,
560 eal_long_options, &option_index)) != EOF) {
561
562 int ret;
563
564 /* getopt is not happy, stop right now */
565 if (opt == '?')
566 break;
567
568 ret = (opt == OPT_LOG_LEVEL_NUM) ?
569 eal_parse_common_option(opt, optarg, internal_conf) : 0;
570
571 /* common parser is not happy */
572 if (ret < 0)
573 break;
574 }
575
576 /* restore getopt lib */
577 optind = old_optind;
578 optopt = old_optopt;
579 optarg = old_optarg;
580 }
581
582 /* Parse the argument given in the command line of the application */
583 static int
eal_parse_args(int argc,char ** argv)584 eal_parse_args(int argc, char **argv)
585 {
586 int opt, ret;
587 char **argvopt;
588 int option_index;
589 char *prgname = argv[0];
590 const int old_optind = optind;
591 const int old_optopt = optopt;
592 char * const old_optarg = optarg;
593 struct internal_config *internal_conf =
594 eal_get_internal_configuration();
595
596 argvopt = argv;
597 optind = 1;
598
599 while ((opt = getopt_long(argc, argvopt, eal_short_options,
600 eal_long_options, &option_index)) != EOF) {
601
602 /* getopt didn't recognise the option */
603 if (opt == '?') {
604 eal_usage(prgname);
605 ret = -1;
606 goto out;
607 }
608
609 /* eal_log_level_parse() already handled this option */
610 if (opt == OPT_LOG_LEVEL_NUM)
611 continue;
612
613 ret = eal_parse_common_option(opt, optarg, internal_conf);
614 /* common parser is not happy */
615 if (ret < 0) {
616 eal_usage(prgname);
617 ret = -1;
618 goto out;
619 }
620 /* common parser handled this option */
621 if (ret == 0)
622 continue;
623
624 switch (opt) {
625 case 'h':
626 eal_usage(prgname);
627 exit(EXIT_SUCCESS);
628
629 case OPT_HUGE_DIR_NUM:
630 {
631 char *hdir = strdup(optarg);
632 if (hdir == NULL)
633 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
634 else {
635 /* free old hugepage dir */
636 free(internal_conf->hugepage_dir);
637 internal_conf->hugepage_dir = hdir;
638 }
639 break;
640 }
641 case OPT_FILE_PREFIX_NUM:
642 {
643 char *prefix = strdup(optarg);
644 if (prefix == NULL)
645 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
646 else {
647 /* free old prefix */
648 free(internal_conf->hugefile_prefix);
649 internal_conf->hugefile_prefix = prefix;
650 }
651 break;
652 }
653 case OPT_SOCKET_MEM_NUM:
654 if (eal_parse_socket_arg(optarg,
655 internal_conf->socket_mem) < 0) {
656 RTE_LOG(ERR, EAL, "invalid parameters for --"
657 OPT_SOCKET_MEM "\n");
658 eal_usage(prgname);
659 ret = -1;
660 goto out;
661 }
662 internal_conf->force_sockets = 1;
663 break;
664
665 case OPT_SOCKET_LIMIT_NUM:
666 if (eal_parse_socket_arg(optarg,
667 internal_conf->socket_limit) < 0) {
668 RTE_LOG(ERR, EAL, "invalid parameters for --"
669 OPT_SOCKET_LIMIT "\n");
670 eal_usage(prgname);
671 ret = -1;
672 goto out;
673 }
674 internal_conf->force_socket_limits = 1;
675 break;
676
677 case OPT_VFIO_INTR_NUM:
678 if (eal_parse_vfio_intr(optarg) < 0) {
679 RTE_LOG(ERR, EAL, "invalid parameters for --"
680 OPT_VFIO_INTR "\n");
681 eal_usage(prgname);
682 ret = -1;
683 goto out;
684 }
685 break;
686
687 case OPT_VFIO_VF_TOKEN_NUM:
688 if (eal_parse_vfio_vf_token(optarg) < 0) {
689 RTE_LOG(ERR, EAL, "invalid parameters for --"
690 OPT_VFIO_VF_TOKEN "\n");
691 eal_usage(prgname);
692 ret = -1;
693 goto out;
694 }
695 break;
696
697 case OPT_CREATE_UIO_DEV_NUM:
698 internal_conf->create_uio_dev = 1;
699 break;
700
701 case OPT_MBUF_POOL_OPS_NAME_NUM:
702 {
703 char *ops_name = strdup(optarg);
704 if (ops_name == NULL)
705 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
706 else {
707 /* free old ops name */
708 free(internal_conf->user_mbuf_pool_ops_name);
709
710 internal_conf->user_mbuf_pool_ops_name =
711 ops_name;
712 }
713 break;
714 }
715 case OPT_MATCH_ALLOCATIONS_NUM:
716 internal_conf->match_allocations = 1;
717 break;
718
719 default:
720 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
721 RTE_LOG(ERR, EAL, "Option %c is not supported "
722 "on Linux\n", opt);
723 } else if (opt >= OPT_LONG_MIN_NUM &&
724 opt < OPT_LONG_MAX_NUM) {
725 RTE_LOG(ERR, EAL, "Option %s is not supported "
726 "on Linux\n",
727 eal_long_options[option_index].name);
728 } else {
729 RTE_LOG(ERR, EAL, "Option %d is not supported "
730 "on Linux\n", opt);
731 }
732 eal_usage(prgname);
733 ret = -1;
734 goto out;
735 }
736 }
737
738 /* create runtime data directory. In no_shconf mode, skip any errors */
739 if (eal_create_runtime_dir() < 0) {
740 if (internal_conf->no_shconf == 0) {
741 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
742 ret = -1;
743 goto out;
744 } else
745 RTE_LOG(WARNING, EAL, "No DPDK runtime directory created\n");
746 }
747
748 if (eal_adjust_config(internal_conf) != 0) {
749 ret = -1;
750 goto out;
751 }
752
753 /* sanity checks */
754 if (eal_check_common_options(internal_conf) != 0) {
755 eal_usage(prgname);
756 ret = -1;
757 goto out;
758 }
759
760 if (optind >= 0)
761 argv[optind-1] = prgname;
762 ret = optind-1;
763
764 out:
765 /* restore getopt lib */
766 optind = old_optind;
767 optopt = old_optopt;
768 optarg = old_optarg;
769
770 return ret;
771 }
772
773 static int
check_socket(const struct rte_memseg_list * msl,void * arg)774 check_socket(const struct rte_memseg_list *msl, void *arg)
775 {
776 int *socket_id = arg;
777
778 if (msl->external)
779 return 0;
780
781 return *socket_id == msl->socket_id;
782 }
783
784 static void
eal_check_mem_on_local_socket(void)785 eal_check_mem_on_local_socket(void)
786 {
787 int socket_id;
788 const struct rte_config *config = rte_eal_get_configuration();
789
790 socket_id = rte_lcore_to_socket_id(config->main_lcore);
791
792 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
793 RTE_LOG(WARNING, EAL, "WARNING: Main core has no memory on local socket!\n");
794 }
795
796 static int
sync_func(__rte_unused void * arg)797 sync_func(__rte_unused void *arg)
798 {
799 return 0;
800 }
801
802 /*
803 * Request iopl privilege for all RPL, returns 0 on success
804 * iopl() call is mostly for the i386 architecture. For other architectures,
805 * return -1 to indicate IO privilege can't be changed in this way.
806 */
807 int
rte_eal_iopl_init(void)808 rte_eal_iopl_init(void)
809 {
810 #if defined(RTE_ARCH_X86)
811 if (iopl(3) != 0)
812 return -1;
813 #endif
814 return 0;
815 }
816
817 #ifdef VFIO_PRESENT
rte_eal_vfio_setup(void)818 static int rte_eal_vfio_setup(void)
819 {
820 if (rte_vfio_enable("vfio"))
821 return -1;
822
823 return 0;
824 }
825 #endif
826
rte_eal_init_alert(const char * msg)827 static void rte_eal_init_alert(const char *msg)
828 {
829 fprintf(stderr, "EAL: FATAL: %s\n", msg);
830 RTE_LOG(ERR, EAL, "%s\n", msg);
831 }
832
833 /*
834 * On Linux 3.6+, even if VFIO is not loaded, whenever IOMMU is enabled in the
835 * BIOS and in the kernel, /sys/kernel/iommu_groups path will contain kernel
836 * IOMMU groups. If IOMMU is not enabled, that path would be empty.
837 * Therefore, checking if the path is empty will tell us if IOMMU is enabled.
838 */
839 static bool
is_iommu_enabled(void)840 is_iommu_enabled(void)
841 {
842 DIR *dir = opendir(KERNEL_IOMMU_GROUPS_PATH);
843 struct dirent *d;
844 int n = 0;
845
846 /* if directory doesn't exist, assume IOMMU is not enabled */
847 if (dir == NULL)
848 return false;
849
850 while ((d = readdir(dir)) != NULL) {
851 /* skip dot and dot-dot */
852 if (++n > 2)
853 break;
854 }
855 closedir(dir);
856
857 return n > 2;
858 }
859
860 /* Launch threads, called at application init(). */
861 int
rte_eal_init(int argc,char ** argv)862 rte_eal_init(int argc, char **argv)
863 {
864 int i, fctret, ret;
865 static uint32_t run_once;
866 uint32_t has_run = 0;
867 const char *p;
868 static char logid[PATH_MAX];
869 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
870 char thread_name[RTE_MAX_THREAD_NAME_LEN];
871 bool phys_addrs;
872 const struct rte_config *config = rte_eal_get_configuration();
873 struct internal_config *internal_conf =
874 eal_get_internal_configuration();
875
876 /* checks if the machine is adequate */
877 if (!rte_cpu_is_supported()) {
878 rte_eal_init_alert("unsupported cpu type.");
879 rte_errno = ENOTSUP;
880 return -1;
881 }
882
883 if (!__atomic_compare_exchange_n(&run_once, &has_run, 1, 0,
884 __ATOMIC_RELAXED, __ATOMIC_RELAXED)) {
885 rte_eal_init_alert("already called initialization.");
886 rte_errno = EALREADY;
887 return -1;
888 }
889
890 p = strrchr(argv[0], '/');
891 strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
892
893 eal_reset_internal_config(internal_conf);
894
895 /* set log level as early as possible */
896 eal_log_level_parse(argc, argv);
897
898 /* clone argv to report out later in telemetry */
899 eal_save_args(argc, argv);
900
901 if (rte_eal_cpu_init() < 0) {
902 rte_eal_init_alert("Cannot detect lcores.");
903 rte_errno = ENOTSUP;
904 return -1;
905 }
906
907 fctret = eal_parse_args(argc, argv);
908 if (fctret < 0) {
909 rte_eal_init_alert("Invalid 'command line' arguments.");
910 rte_errno = EINVAL;
911 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
912 return -1;
913 }
914
915 if (eal_plugins_init() < 0) {
916 rte_eal_init_alert("Cannot init plugins");
917 rte_errno = EINVAL;
918 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
919 return -1;
920 }
921
922 if (eal_trace_init() < 0) {
923 rte_eal_init_alert("Cannot init trace");
924 rte_errno = EFAULT;
925 return -1;
926 }
927
928 if (eal_option_device_parse()) {
929 rte_errno = ENODEV;
930 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
931 return -1;
932 }
933
934 if (rte_config_init() < 0) {
935 rte_eal_init_alert("Cannot init config");
936 return -1;
937 }
938
939 if (rte_eal_intr_init() < 0) {
940 rte_eal_init_alert("Cannot init interrupt-handling thread");
941 return -1;
942 }
943
944 if (rte_eal_alarm_init() < 0) {
945 rte_eal_init_alert("Cannot init alarm");
946 /* rte_eal_alarm_init sets rte_errno on failure. */
947 return -1;
948 }
949
950 /* Put mp channel init before bus scan so that we can init the vdev
951 * bus through mp channel in the secondary process before the bus scan.
952 */
953 if (rte_mp_channel_init() < 0 && rte_errno != ENOTSUP) {
954 rte_eal_init_alert("failed to init mp channel");
955 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
956 rte_errno = EFAULT;
957 return -1;
958 }
959 }
960
961 /* register multi-process action callbacks for hotplug */
962 if (eal_mp_dev_hotplug_init() < 0) {
963 rte_eal_init_alert("failed to register mp callback for hotplug");
964 return -1;
965 }
966
967 if (rte_bus_scan()) {
968 rte_eal_init_alert("Cannot scan the buses for devices");
969 rte_errno = ENODEV;
970 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
971 return -1;
972 }
973
974 phys_addrs = rte_eal_using_phys_addrs() != 0;
975
976 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
977 if (internal_conf->iova_mode == RTE_IOVA_DC) {
978 /* autodetect the IOVA mapping mode */
979 enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
980
981 if (iova_mode == RTE_IOVA_DC) {
982 RTE_LOG(DEBUG, EAL, "Buses did not request a specific IOVA mode.\n");
983
984 if (!phys_addrs) {
985 /* if we have no access to physical addresses,
986 * pick IOVA as VA mode.
987 */
988 iova_mode = RTE_IOVA_VA;
989 RTE_LOG(DEBUG, EAL, "Physical addresses are unavailable, selecting IOVA as VA mode.\n");
990 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
991 } else if (rte_eal_check_module("rte_kni") == 1) {
992 iova_mode = RTE_IOVA_PA;
993 RTE_LOG(DEBUG, EAL, "KNI is loaded, selecting IOVA as PA mode for better KNI performance.\n");
994 #endif
995 } else if (is_iommu_enabled()) {
996 /* we have an IOMMU, pick IOVA as VA mode */
997 iova_mode = RTE_IOVA_VA;
998 RTE_LOG(DEBUG, EAL, "IOMMU is available, selecting IOVA as VA mode.\n");
999 } else {
1000 /* physical addresses available, and no IOMMU
1001 * found, so pick IOVA as PA.
1002 */
1003 iova_mode = RTE_IOVA_PA;
1004 RTE_LOG(DEBUG, EAL, "IOMMU is not available, selecting IOVA as PA mode.\n");
1005 }
1006 }
1007 #if defined(RTE_LIB_KNI) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 10, 0)
1008 /* Workaround for KNI which requires physical address to work
1009 * in kernels < 4.10
1010 */
1011 if (iova_mode == RTE_IOVA_VA &&
1012 rte_eal_check_module("rte_kni") == 1) {
1013 if (phys_addrs) {
1014 iova_mode = RTE_IOVA_PA;
1015 RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
1016 } else {
1017 RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
1018 }
1019 }
1020 #endif
1021 rte_eal_get_configuration()->iova_mode = iova_mode;
1022 } else {
1023 rte_eal_get_configuration()->iova_mode =
1024 internal_conf->iova_mode;
1025 }
1026
1027 if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
1028 rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
1029 rte_errno = EINVAL;
1030 return -1;
1031 }
1032
1033 RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
1034 rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
1035
1036 if (internal_conf->no_hugetlbfs == 0) {
1037 /* rte_config isn't initialized yet */
1038 ret = internal_conf->process_type == RTE_PROC_PRIMARY ?
1039 eal_hugepage_info_init() :
1040 eal_hugepage_info_read();
1041 if (ret < 0) {
1042 rte_eal_init_alert("Cannot get hugepage information.");
1043 rte_errno = EACCES;
1044 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1045 return -1;
1046 }
1047 }
1048
1049 if (internal_conf->memory == 0 && internal_conf->force_sockets == 0) {
1050 if (internal_conf->no_hugetlbfs)
1051 internal_conf->memory = MEMSIZE_IF_NO_HUGE_PAGE;
1052 }
1053
1054 if (internal_conf->vmware_tsc_map == 1) {
1055 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1056 rte_cycles_vmware_tsc_map = 1;
1057 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1058 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1059 #else
1060 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1061 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1062 #endif
1063 }
1064
1065 if (eal_log_init(logid, internal_conf->syslog_facility) < 0) {
1066 rte_eal_init_alert("Cannot init logging.");
1067 rte_errno = ENOMEM;
1068 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1069 return -1;
1070 }
1071
1072 #ifdef VFIO_PRESENT
1073 if (rte_eal_vfio_setup() < 0) {
1074 rte_eal_init_alert("Cannot init VFIO");
1075 rte_errno = EAGAIN;
1076 __atomic_store_n(&run_once, 0, __ATOMIC_RELAXED);
1077 return -1;
1078 }
1079 #endif
1080 /* in secondary processes, memory init may allocate additional fbarrays
1081 * not present in primary processes, so to avoid any potential issues,
1082 * initialize memzones first.
1083 */
1084 if (rte_eal_memzone_init() < 0) {
1085 rte_eal_init_alert("Cannot init memzone");
1086 rte_errno = ENODEV;
1087 return -1;
1088 }
1089
1090 if (rte_eal_memory_init() < 0) {
1091 rte_eal_init_alert("Cannot init memory");
1092 rte_errno = ENOMEM;
1093 return -1;
1094 }
1095
1096 /* the directories are locked during eal_hugepage_info_init */
1097 eal_hugedirs_unlock();
1098
1099 if (rte_eal_malloc_heap_init() < 0) {
1100 rte_eal_init_alert("Cannot init malloc heap");
1101 rte_errno = ENODEV;
1102 return -1;
1103 }
1104
1105 if (rte_eal_tailqs_init() < 0) {
1106 rte_eal_init_alert("Cannot init tail queues for objects");
1107 rte_errno = EFAULT;
1108 return -1;
1109 }
1110
1111 if (rte_eal_timer_init() < 0) {
1112 rte_eal_init_alert("Cannot init HPET or TSC timers");
1113 rte_errno = ENOTSUP;
1114 return -1;
1115 }
1116
1117 eal_check_mem_on_local_socket();
1118
1119 if (pthread_setaffinity_np(pthread_self(), sizeof(rte_cpuset_t),
1120 &lcore_config[config->main_lcore].cpuset) != 0) {
1121 rte_eal_init_alert("Cannot set affinity");
1122 rte_errno = EINVAL;
1123 return -1;
1124 }
1125 __rte_thread_init(config->main_lcore,
1126 &lcore_config[config->main_lcore].cpuset);
1127
1128 ret = eal_thread_dump_current_affinity(cpuset, sizeof(cpuset));
1129 RTE_LOG(DEBUG, EAL, "Main lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1130 config->main_lcore, (uintptr_t)pthread_self(), cpuset,
1131 ret == 0 ? "" : "...");
1132
1133 RTE_LCORE_FOREACH_WORKER(i) {
1134
1135 /*
1136 * create communication pipes between main thread
1137 * and children
1138 */
1139 if (pipe(lcore_config[i].pipe_main2worker) < 0)
1140 rte_panic("Cannot create pipe\n");
1141 if (pipe(lcore_config[i].pipe_worker2main) < 0)
1142 rte_panic("Cannot create pipe\n");
1143
1144 lcore_config[i].state = WAIT;
1145
1146 /* create a thread for each lcore */
1147 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1148 eal_thread_loop, (void *)(uintptr_t)i);
1149 if (ret != 0)
1150 rte_panic("Cannot create thread\n");
1151
1152 /* Set thread_name for aid in debugging. */
1153 snprintf(thread_name, sizeof(thread_name),
1154 "lcore-worker-%d", i);
1155 ret = rte_thread_setname(lcore_config[i].thread_id,
1156 thread_name);
1157 if (ret != 0)
1158 RTE_LOG(DEBUG, EAL,
1159 "Cannot set name for lcore thread\n");
1160
1161 ret = pthread_setaffinity_np(lcore_config[i].thread_id,
1162 sizeof(rte_cpuset_t), &lcore_config[i].cpuset);
1163 if (ret != 0)
1164 rte_panic("Cannot set affinity\n");
1165 }
1166
1167 /*
1168 * Launch a dummy function on all worker lcores, so that main lcore
1169 * knows they are all ready when this function returns.
1170 */
1171 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MAIN);
1172 rte_eal_mp_wait_lcore();
1173
1174 /* initialize services so vdevs register service during bus_probe. */
1175 ret = rte_service_init();
1176 if (ret) {
1177 rte_eal_init_alert("rte_service_init() failed");
1178 rte_errno = -ret;
1179 return -1;
1180 }
1181
1182 /* Probe all the buses and devices/drivers on them */
1183 if (rte_bus_probe()) {
1184 rte_eal_init_alert("Cannot probe devices");
1185 rte_errno = ENOTSUP;
1186 return -1;
1187 }
1188
1189 #ifdef VFIO_PRESENT
1190 /* Register mp action after probe() so that we got enough info */
1191 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1192 return -1;
1193 #endif
1194
1195 /* initialize default service/lcore mappings and start running. Ignore
1196 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1197 */
1198 ret = rte_service_start_with_defaults();
1199 if (ret < 0 && ret != -ENOTSUP) {
1200 rte_errno = -ret;
1201 return -1;
1202 }
1203
1204 /*
1205 * Clean up unused files in runtime directory. We do this at the end of
1206 * init and not at the beginning because we want to clean stuff up
1207 * whether we are primary or secondary process, but we cannot remove
1208 * primary process' files because secondary should be able to run even
1209 * if primary process is dead.
1210 *
1211 * In no_shconf mode, no runtime directory is created in the first
1212 * place, so no cleanup needed.
1213 */
1214 if (!internal_conf->no_shconf && eal_clean_runtime_dir() < 0) {
1215 rte_eal_init_alert("Cannot clear runtime directory");
1216 return -1;
1217 }
1218 if (rte_eal_process_type() == RTE_PROC_PRIMARY && !internal_conf->no_telemetry) {
1219 int tlog = rte_log_register_type_and_pick_level(
1220 "lib.telemetry", RTE_LOG_WARNING);
1221 if (tlog < 0)
1222 tlog = RTE_LOGTYPE_EAL;
1223 if (rte_telemetry_init(rte_eal_get_runtime_dir(),
1224 rte_version(),
1225 &internal_conf->ctrl_cpuset, rte_log, tlog) != 0)
1226 return -1;
1227 }
1228
1229 eal_mcfg_complete();
1230
1231 return fctret;
1232 }
1233
1234 static int
mark_freeable(const struct rte_memseg_list * msl,const struct rte_memseg * ms,void * arg __rte_unused)1235 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1236 void *arg __rte_unused)
1237 {
1238 /* ms is const, so find this memseg */
1239 struct rte_memseg *found;
1240
1241 if (msl->external)
1242 return 0;
1243
1244 found = rte_mem_virt2memseg(ms->addr, msl);
1245
1246 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1247
1248 return 0;
1249 }
1250
1251 int
rte_eal_cleanup(void)1252 rte_eal_cleanup(void)
1253 {
1254 /* if we're in a primary process, we need to mark hugepages as freeable
1255 * so that finalization can release them back to the system.
1256 */
1257 struct internal_config *internal_conf =
1258 eal_get_internal_configuration();
1259
1260 if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
1261 internal_conf->hugepage_file.unlink_existing)
1262 rte_memseg_walk(mark_freeable, NULL);
1263
1264 rte_service_finalize();
1265 #ifdef VFIO_PRESENT
1266 vfio_mp_sync_cleanup();
1267 #endif
1268 rte_mp_channel_cleanup();
1269 /* after this point, any DPDK pointers will become dangling */
1270 rte_eal_memory_detach();
1271 eal_mp_dev_hotplug_cleanup();
1272 rte_eal_malloc_heap_cleanup();
1273 rte_eal_alarm_cleanup();
1274 rte_trace_save();
1275 eal_trace_fini();
1276 eal_cleanup_config(internal_conf);
1277 rte_eal_log_cleanup();
1278 return 0;
1279 }
1280
rte_eal_create_uio_dev(void)1281 int rte_eal_create_uio_dev(void)
1282 {
1283 const struct internal_config *internal_conf =
1284 eal_get_internal_configuration();
1285
1286 return internal_conf->create_uio_dev;
1287 }
1288
1289 enum rte_intr_mode
rte_eal_vfio_intr_mode(void)1290 rte_eal_vfio_intr_mode(void)
1291 {
1292 const struct internal_config *internal_conf =
1293 eal_get_internal_configuration();
1294
1295 return internal_conf->vfio_intr_mode;
1296 }
1297
1298 void
rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)1299 rte_eal_vfio_get_vf_token(rte_uuid_t vf_token)
1300 {
1301 struct internal_config *cfg = eal_get_internal_configuration();
1302
1303 rte_uuid_copy(vf_token, cfg->vfio_vf_token);
1304 }
1305
1306 int
rte_eal_check_module(const char * module_name)1307 rte_eal_check_module(const char *module_name)
1308 {
1309 char sysfs_mod_name[PATH_MAX];
1310 struct stat st;
1311 int n;
1312
1313 if (NULL == module_name)
1314 return -1;
1315
1316 /* Check if there is sysfs mounted */
1317 if (stat("/sys/module", &st) != 0) {
1318 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1319 errno, strerror(errno));
1320 return -1;
1321 }
1322
1323 /* A module might be built-in, therefore try sysfs */
1324 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1325 if (n < 0 || n > PATH_MAX) {
1326 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1327 return -1;
1328 }
1329
1330 if (stat(sysfs_mod_name, &st) != 0) {
1331 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1332 sysfs_mod_name, errno, strerror(errno));
1333 return 0;
1334 }
1335
1336 /* Module has been found */
1337 return 1;
1338 }
1339