]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/lib/librte_eal/linuxapp/eal/eal.c
update download target update for octopus release
[ceph.git] / ceph / src / spdk / dpdk / lib / librte_eal / linuxapp / eal / eal.c
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 <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 <fcntl.h>
17 #include <stddef.h>
18 #include <errno.h>
19 #include <limits.h>
20 #include <sys/mman.h>
21 #include <sys/queue.h>
22 #include <sys/stat.h>
23 #if defined(RTE_ARCH_X86)
24 #include <sys/io.h>
25 #endif
26
27 #include <rte_compat.h>
28 #include <rte_common.h>
29 #include <rte_debug.h>
30 #include <rte_memory.h>
31 #include <rte_launch.h>
32 #include <rte_eal.h>
33 #include <rte_eal_memconfig.h>
34 #include <rte_errno.h>
35 #include <rte_per_lcore.h>
36 #include <rte_lcore.h>
37 #include <rte_service_component.h>
38 #include <rte_log.h>
39 #include <rte_random.h>
40 #include <rte_cycles.h>
41 #include <rte_string_fns.h>
42 #include <rte_cpuflags.h>
43 #include <rte_interrupts.h>
44 #include <rte_bus.h>
45 #include <rte_dev.h>
46 #include <rte_devargs.h>
47 #include <rte_version.h>
48 #include <rte_atomic.h>
49 #include <malloc_heap.h>
50 #include <rte_vfio.h>
51
52 #include "eal_private.h"
53 #include "eal_thread.h"
54 #include "eal_internal_cfg.h"
55 #include "eal_filesystem.h"
56 #include "eal_hugepages.h"
57 #include "eal_options.h"
58 #include "eal_vfio.h"
59
60 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
61
62 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
63
64 /* Allow the application to print its usage message too if set */
65 static rte_usage_hook_t rte_application_usage_hook = NULL;
66
67 /* early configuration structure, when memory config is not mmapped */
68 static struct rte_mem_config early_mem_config;
69
70 /* define fd variable here, because file needs to be kept open for the
71 * duration of the program, as we hold a write lock on it in the primary proc */
72 static int mem_cfg_fd = -1;
73
74 static struct flock wr_lock = {
75 .l_type = F_WRLCK,
76 .l_whence = SEEK_SET,
77 .l_start = offsetof(struct rte_mem_config, memsegs),
78 .l_len = sizeof(early_mem_config.memsegs),
79 };
80
81 /* Address of global and public configuration */
82 static struct rte_config rte_config = {
83 .mem_config = &early_mem_config,
84 };
85
86 /* internal configuration (per-core) */
87 struct lcore_config lcore_config[RTE_MAX_LCORE];
88
89 /* internal configuration */
90 struct internal_config internal_config;
91
92 /* used by rte_rdtsc() */
93 int rte_cycles_vmware_tsc_map;
94
95 /* platform-specific runtime dir */
96 static char runtime_dir[PATH_MAX];
97
98 static const char *default_runtime_dir = "/var/run";
99
100 int
101 eal_create_runtime_dir(void)
102 {
103 const char *directory = default_runtime_dir;
104 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
105 const char *fallback = "/tmp";
106 char tmp[PATH_MAX];
107 int ret;
108
109 if (getuid() != 0) {
110 /* try XDG path first, fall back to /tmp */
111 if (xdg_runtime_dir != NULL)
112 directory = xdg_runtime_dir;
113 else
114 directory = fallback;
115 }
116 /* create DPDK subdirectory under runtime dir */
117 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
118 if (ret < 0 || ret == sizeof(tmp)) {
119 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
120 return -1;
121 }
122
123 /* create prefix-specific subdirectory under DPDK runtime dir */
124 ret = snprintf(runtime_dir, sizeof(runtime_dir), "%s/%s",
125 tmp, internal_config.hugefile_prefix);
126 if (ret < 0 || ret == sizeof(runtime_dir)) {
127 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
128 return -1;
129 }
130
131 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
132 * step by step.
133 */
134 ret = mkdir(tmp, 0700);
135 if (ret < 0 && errno != EEXIST) {
136 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
137 tmp, strerror(errno));
138 return -1;
139 }
140
141 ret = mkdir(runtime_dir, 0700);
142 if (ret < 0 && errno != EEXIST) {
143 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
144 runtime_dir, strerror(errno));
145 return -1;
146 }
147
148 return 0;
149 }
150
151 const char *
152 eal_get_runtime_dir(void)
153 {
154 return runtime_dir;
155 }
156
157 /* Return user provided mbuf pool ops name */
158 const char *
159 rte_eal_mbuf_user_pool_ops(void)
160 {
161 return internal_config.user_mbuf_pool_ops_name;
162 }
163
164 /* Return a pointer to the configuration structure */
165 struct rte_config *
166 rte_eal_get_configuration(void)
167 {
168 return &rte_config;
169 }
170
171 enum rte_iova_mode
172 rte_eal_iova_mode(void)
173 {
174 return rte_eal_get_configuration()->iova_mode;
175 }
176
177 /* parse a sysfs (or other) file containing one integer value */
178 int
179 eal_parse_sysfs_value(const char *filename, unsigned long *val)
180 {
181 FILE *f;
182 char buf[BUFSIZ];
183 char *end = NULL;
184
185 if ((f = fopen(filename, "r")) == NULL) {
186 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
187 __func__, filename);
188 return -1;
189 }
190
191 if (fgets(buf, sizeof(buf), f) == NULL) {
192 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
193 __func__, filename);
194 fclose(f);
195 return -1;
196 }
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",
200 __func__, filename);
201 fclose(f);
202 return -1;
203 }
204 fclose(f);
205 return 0;
206 }
207
208
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
214 * processes. */
215 static void
216 rte_eal_config_create(void)
217 {
218 void *rte_mem_cfg_addr;
219 int retval;
220
221 const char *pathname = eal_runtime_config_path();
222
223 if (internal_config.no_shconf)
224 return;
225
226 /* map the config before hugepage address so that we don't waste a page */
227 if (internal_config.base_virtaddr != 0)
228 rte_mem_cfg_addr = (void *)
229 RTE_ALIGN_FLOOR(internal_config.base_virtaddr -
230 sizeof(struct rte_mem_config), sysconf(_SC_PAGE_SIZE));
231 else
232 rte_mem_cfg_addr = NULL;
233
234 if (mem_cfg_fd < 0){
235 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
236 if (mem_cfg_fd < 0)
237 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
238 }
239
240 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
241 if (retval < 0){
242 close(mem_cfg_fd);
243 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
244 }
245
246 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
247 if (retval < 0){
248 close(mem_cfg_fd);
249 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
250 "process running?\n", pathname);
251 }
252
253 rte_mem_cfg_addr = mmap(rte_mem_cfg_addr, sizeof(*rte_config.mem_config),
254 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
255
256 if (rte_mem_cfg_addr == MAP_FAILED){
257 rte_panic("Cannot mmap memory for rte_config\n");
258 }
259 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
260 rte_config.mem_config = rte_mem_cfg_addr;
261
262 /* store address of the config in the config itself so that secondary
263 * processes could later map the config into this exact location */
264 rte_config.mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
265
266 }
267
268 /* attach to an existing shared memory config */
269 static void
270 rte_eal_config_attach(void)
271 {
272 struct rte_mem_config *mem_config;
273
274 const char *pathname = eal_runtime_config_path();
275
276 if (internal_config.no_shconf)
277 return;
278
279 if (mem_cfg_fd < 0){
280 mem_cfg_fd = open(pathname, O_RDWR);
281 if (mem_cfg_fd < 0)
282 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
283 }
284
285 /* map it as read-only first */
286 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
287 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
288 if (mem_config == MAP_FAILED)
289 rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
290 errno, strerror(errno));
291
292 rte_config.mem_config = mem_config;
293 }
294
295 /* reattach the shared config at exact memory location primary process has it */
296 static void
297 rte_eal_config_reattach(void)
298 {
299 struct rte_mem_config *mem_config;
300 void *rte_mem_cfg_addr;
301
302 if (internal_config.no_shconf)
303 return;
304
305 /* save the address primary process has mapped shared config to */
306 rte_mem_cfg_addr = (void *) (uintptr_t) rte_config.mem_config->mem_cfg_addr;
307
308 /* unmap original config */
309 munmap(rte_config.mem_config, sizeof(struct rte_mem_config));
310
311 /* remap the config at proper address */
312 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
313 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
314 mem_cfg_fd, 0);
315 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
316 if (mem_config != MAP_FAILED)
317 /* errno is stale, don't use */
318 rte_panic("Cannot mmap memory for rte_config at [%p], got [%p]"
319 " - please use '--base-virtaddr' option\n",
320 rte_mem_cfg_addr, mem_config);
321 else
322 rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
323 errno, strerror(errno));
324 }
325 close(mem_cfg_fd);
326
327 rte_config.mem_config = mem_config;
328 }
329
330 /* Detect if we are a primary or a secondary process */
331 enum rte_proc_type_t
332 eal_proc_type_detect(void)
333 {
334 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
335 const char *pathname = eal_runtime_config_path();
336
337 /* if there no shared config, there can be no secondary processes */
338 if (!internal_config.no_shconf) {
339 /* if we can open the file but not get a write-lock we are a
340 * secondary process. NOTE: if we get a file handle back, we
341 * keep that open and don't close it to prevent a race condition
342 * between multiple opens.
343 */
344 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
345 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
346 ptype = RTE_PROC_SECONDARY;
347 }
348
349 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
350 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
351
352 return ptype;
353 }
354
355 /* Sets up rte_config structure with the pointer to shared memory config.*/
356 static void
357 rte_config_init(void)
358 {
359 rte_config.process_type = internal_config.process_type;
360
361 switch (rte_config.process_type){
362 case RTE_PROC_PRIMARY:
363 rte_eal_config_create();
364 break;
365 case RTE_PROC_SECONDARY:
366 rte_eal_config_attach();
367 rte_eal_mcfg_wait_complete(rte_config.mem_config);
368 rte_eal_config_reattach();
369 break;
370 case RTE_PROC_AUTO:
371 case RTE_PROC_INVALID:
372 rte_panic("Invalid process type\n");
373 }
374 }
375
376 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
377 static void
378 eal_hugedirs_unlock(void)
379 {
380 int i;
381
382 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
383 {
384 /* skip uninitialized */
385 if (internal_config.hugepage_info[i].lock_descriptor < 0)
386 continue;
387 /* unlock hugepage file */
388 flock(internal_config.hugepage_info[i].lock_descriptor, LOCK_UN);
389 close(internal_config.hugepage_info[i].lock_descriptor);
390 /* reset the field */
391 internal_config.hugepage_info[i].lock_descriptor = -1;
392 }
393 }
394
395 /* display usage */
396 static void
397 eal_usage(const char *prgname)
398 {
399 printf("\nUsage: %s ", prgname);
400 eal_common_usage();
401 printf("EAL Linux options:\n"
402 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
403 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
404 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
405 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
406 " --"OPT_BASE_VIRTADDR" Base virtual address\n"
407 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
408 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
409 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
410 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
411 "\n");
412 /* Allow the application to print its usage message too if hook is set */
413 if ( rte_application_usage_hook ) {
414 printf("===== Application Usage =====\n\n");
415 rte_application_usage_hook(prgname);
416 }
417 }
418
419 /* Set a per-application usage message */
420 rte_usage_hook_t
421 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
422 {
423 rte_usage_hook_t old_func;
424
425 /* Will be NULL on the first call to denote the last usage routine. */
426 old_func = rte_application_usage_hook;
427 rte_application_usage_hook = usage_func;
428
429 return old_func;
430 }
431
432 static int
433 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
434 {
435 char * arg[RTE_MAX_NUMA_NODES];
436 char *end;
437 int arg_num, i, len;
438 uint64_t total_mem = 0;
439
440 len = strnlen(strval, SOCKET_MEM_STRLEN);
441 if (len == SOCKET_MEM_STRLEN) {
442 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
443 return -1;
444 }
445
446 /* all other error cases will be caught later */
447 if (!isdigit(strval[len-1]))
448 return -1;
449
450 /* split the optarg into separate socket values */
451 arg_num = rte_strsplit(strval, len,
452 arg, RTE_MAX_NUMA_NODES, ',');
453
454 /* if split failed, or 0 arguments */
455 if (arg_num <= 0)
456 return -1;
457
458 /* parse each defined socket option */
459 errno = 0;
460 for (i = 0; i < arg_num; i++) {
461 uint64_t val;
462 end = NULL;
463 val = strtoull(arg[i], &end, 10);
464
465 /* check for invalid input */
466 if ((errno != 0) ||
467 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
468 return -1;
469 val <<= 20;
470 total_mem += val;
471 socket_arg[i] = val;
472 }
473
474 /* check if we have a positive amount of total memory */
475 if (total_mem == 0)
476 return -1;
477
478 return 0;
479 }
480
481 static int
482 eal_parse_base_virtaddr(const char *arg)
483 {
484 char *end;
485 uint64_t addr;
486
487 errno = 0;
488 addr = strtoull(arg, &end, 16);
489
490 /* check for errors */
491 if ((errno != 0) || (arg[0] == '\0') || end == NULL || (*end != '\0'))
492 return -1;
493
494 /* make sure we don't exceed 32-bit boundary on 32-bit target */
495 #ifndef RTE_ARCH_64
496 if (addr >= UINTPTR_MAX)
497 return -1;
498 #endif
499
500 /* align the addr on 16M boundary, 16MB is the minimum huge page
501 * size on IBM Power architecture. If the addr is aligned to 16MB,
502 * it can align to 2MB for x86. So this alignment can also be used
503 * on x86 */
504 internal_config.base_virtaddr =
505 RTE_PTR_ALIGN_CEIL((uintptr_t)addr, (size_t)RTE_PGSIZE_16M);
506
507 return 0;
508 }
509
510 static int
511 eal_parse_vfio_intr(const char *mode)
512 {
513 unsigned i;
514 static struct {
515 const char *name;
516 enum rte_intr_mode value;
517 } map[] = {
518 { "legacy", RTE_INTR_MODE_LEGACY },
519 { "msi", RTE_INTR_MODE_MSI },
520 { "msix", RTE_INTR_MODE_MSIX },
521 };
522
523 for (i = 0; i < RTE_DIM(map); i++) {
524 if (!strcmp(mode, map[i].name)) {
525 internal_config.vfio_intr_mode = map[i].value;
526 return 0;
527 }
528 }
529 return -1;
530 }
531
532 /* Parse the arguments for --log-level only */
533 static void
534 eal_log_level_parse(int argc, char **argv)
535 {
536 int opt;
537 char **argvopt;
538 int option_index;
539 const int old_optind = optind;
540 const int old_optopt = optopt;
541 char * const old_optarg = optarg;
542
543 argvopt = argv;
544 optind = 1;
545
546 while ((opt = getopt_long(argc, argvopt, eal_short_options,
547 eal_long_options, &option_index)) != EOF) {
548
549 int ret;
550
551 /* getopt is not happy, stop right now */
552 if (opt == '?')
553 break;
554
555 ret = (opt == OPT_LOG_LEVEL_NUM) ?
556 eal_parse_common_option(opt, optarg, &internal_config) : 0;
557
558 /* common parser is not happy */
559 if (ret < 0)
560 break;
561 }
562
563 /* restore getopt lib */
564 optind = old_optind;
565 optopt = old_optopt;
566 optarg = old_optarg;
567 }
568
569 /* Parse the argument given in the command line of the application */
570 static int
571 eal_parse_args(int argc, char **argv)
572 {
573 int opt, ret;
574 char **argvopt;
575 int option_index;
576 char *prgname = argv[0];
577 const int old_optind = optind;
578 const int old_optopt = optopt;
579 char * const old_optarg = optarg;
580
581 argvopt = argv;
582 optind = 1;
583
584 while ((opt = getopt_long(argc, argvopt, eal_short_options,
585 eal_long_options, &option_index)) != EOF) {
586
587 /* getopt is not happy, stop right now */
588 if (opt == '?') {
589 eal_usage(prgname);
590 ret = -1;
591 goto out;
592 }
593
594 ret = eal_parse_common_option(opt, optarg, &internal_config);
595 /* common parser is not happy */
596 if (ret < 0) {
597 eal_usage(prgname);
598 ret = -1;
599 goto out;
600 }
601 /* common parser handled this option */
602 if (ret == 0)
603 continue;
604
605 switch (opt) {
606 case 'h':
607 eal_usage(prgname);
608 exit(EXIT_SUCCESS);
609
610 case OPT_HUGE_DIR_NUM:
611 internal_config.hugepage_dir = strdup(optarg);
612 break;
613
614 case OPT_FILE_PREFIX_NUM:
615 internal_config.hugefile_prefix = strdup(optarg);
616 break;
617
618 case OPT_SOCKET_MEM_NUM:
619 if (eal_parse_socket_arg(optarg,
620 internal_config.socket_mem) < 0) {
621 RTE_LOG(ERR, EAL, "invalid parameters for --"
622 OPT_SOCKET_MEM "\n");
623 eal_usage(prgname);
624 ret = -1;
625 goto out;
626 }
627 internal_config.force_sockets = 1;
628 break;
629
630 case OPT_SOCKET_LIMIT_NUM:
631 if (eal_parse_socket_arg(optarg,
632 internal_config.socket_limit) < 0) {
633 RTE_LOG(ERR, EAL, "invalid parameters for --"
634 OPT_SOCKET_LIMIT "\n");
635 eal_usage(prgname);
636 ret = -1;
637 goto out;
638 }
639 internal_config.force_socket_limits = 1;
640 break;
641
642 case OPT_BASE_VIRTADDR_NUM:
643 if (eal_parse_base_virtaddr(optarg) < 0) {
644 RTE_LOG(ERR, EAL, "invalid parameter for --"
645 OPT_BASE_VIRTADDR "\n");
646 eal_usage(prgname);
647 ret = -1;
648 goto out;
649 }
650 break;
651
652 case OPT_VFIO_INTR_NUM:
653 if (eal_parse_vfio_intr(optarg) < 0) {
654 RTE_LOG(ERR, EAL, "invalid parameters for --"
655 OPT_VFIO_INTR "\n");
656 eal_usage(prgname);
657 ret = -1;
658 goto out;
659 }
660 break;
661
662 case OPT_CREATE_UIO_DEV_NUM:
663 internal_config.create_uio_dev = 1;
664 break;
665
666 case OPT_MBUF_POOL_OPS_NAME_NUM:
667 internal_config.user_mbuf_pool_ops_name =
668 strdup(optarg);
669 break;
670
671 default:
672 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
673 RTE_LOG(ERR, EAL, "Option %c is not supported "
674 "on Linux\n", opt);
675 } else if (opt >= OPT_LONG_MIN_NUM &&
676 opt < OPT_LONG_MAX_NUM) {
677 RTE_LOG(ERR, EAL, "Option %s is not supported "
678 "on Linux\n",
679 eal_long_options[option_index].name);
680 } else {
681 RTE_LOG(ERR, EAL, "Option %d is not supported "
682 "on Linux\n", opt);
683 }
684 eal_usage(prgname);
685 ret = -1;
686 goto out;
687 }
688 }
689
690 /* create runtime data directory */
691 if (internal_config.no_shconf == 0 &&
692 eal_create_runtime_dir() < 0) {
693 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
694 ret = -1;
695 goto out;
696 }
697
698 if (eal_adjust_config(&internal_config) != 0) {
699 ret = -1;
700 goto out;
701 }
702
703 /* sanity checks */
704 if (eal_check_common_options(&internal_config) != 0) {
705 eal_usage(prgname);
706 ret = -1;
707 goto out;
708 }
709
710 if (optind >= 0)
711 argv[optind-1] = prgname;
712 ret = optind-1;
713
714 out:
715 /* restore getopt lib */
716 optind = old_optind;
717 optopt = old_optopt;
718 optarg = old_optarg;
719
720 return ret;
721 }
722
723 static int
724 check_socket(const struct rte_memseg_list *msl, void *arg)
725 {
726 int *socket_id = arg;
727
728 return *socket_id == msl->socket_id;
729 }
730
731 static void
732 eal_check_mem_on_local_socket(void)
733 {
734 int socket_id;
735
736 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
737
738 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
739 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
740 }
741
742 static int
743 sync_func(__attribute__((unused)) void *arg)
744 {
745 return 0;
746 }
747
748 inline static void
749 rte_eal_mcfg_complete(void)
750 {
751 /* ALL shared mem_config related INIT DONE */
752 if (rte_config.process_type == RTE_PROC_PRIMARY)
753 rte_config.mem_config->magic = RTE_MAGIC;
754
755 internal_config.init_complete = 1;
756 }
757
758 /*
759 * Request iopl privilege for all RPL, returns 0 on success
760 * iopl() call is mostly for the i386 architecture. For other architectures,
761 * return -1 to indicate IO privilege can't be changed in this way.
762 */
763 int
764 rte_eal_iopl_init(void)
765 {
766 #if defined(RTE_ARCH_X86)
767 if (iopl(3) != 0)
768 return -1;
769 #endif
770 return 0;
771 }
772
773 #ifdef VFIO_PRESENT
774 static int rte_eal_vfio_setup(void)
775 {
776 if (rte_vfio_enable("vfio"))
777 return -1;
778
779 return 0;
780 }
781 #endif
782
783 static void rte_eal_init_alert(const char *msg)
784 {
785 fprintf(stderr, "EAL: FATAL: %s\n", msg);
786 RTE_LOG(ERR, EAL, "%s\n", msg);
787 }
788
789 /* Launch threads, called at application init(). */
790 int
791 rte_eal_init(int argc, char **argv)
792 {
793 int i, fctret, ret;
794 pthread_t thread_id;
795 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
796 char *logid_storage;
797 const char *logid;
798 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
799 char thread_name[RTE_MAX_THREAD_NAME_LEN];
800
801 /* checks if the machine is adequate */
802 if (!rte_cpu_is_supported()) {
803 rte_eal_init_alert("unsupported cpu type.");
804 rte_errno = ENOTSUP;
805 return -1;
806 }
807
808 if (!rte_atomic32_test_and_set(&run_once)) {
809 rte_eal_init_alert("already called initialization.");
810 rte_errno = EALREADY;
811 return -1;
812 }
813
814 logid_storage = strrchr(argv[0], '/');
815 logid_storage = strdup(logid_storage ? logid_storage + 1 : argv[0]);
816 logid = logid_storage;
817
818 thread_id = pthread_self();
819
820 eal_reset_internal_config(&internal_config);
821
822 /* set log level as early as possible */
823 eal_log_level_parse(argc, argv);
824
825 if (rte_eal_cpu_init() < 0) {
826 rte_eal_init_alert("Cannot detect lcores.");
827 rte_errno = ENOTSUP;
828 fctret = -1;
829 goto finished;
830 }
831
832 fctret = eal_parse_args(argc, argv);
833 if (fctret < 0) {
834 rte_eal_init_alert("Invalid 'command line' arguments.");
835 rte_errno = EINVAL;
836 rte_atomic32_clear(&run_once);
837 fctret = -1;
838 goto finished;
839 }
840
841 if (eal_plugins_init() < 0) {
842 rte_eal_init_alert("Cannot init plugins\n");
843 rte_errno = EINVAL;
844 rte_atomic32_clear(&run_once);
845 fctret = -1;
846 goto finished;
847 }
848
849 if (eal_option_device_parse()) {
850 rte_errno = ENODEV;
851 rte_atomic32_clear(&run_once);
852 fctret = -1;
853 goto finished;
854 }
855
856 rte_config_init();
857
858 if (rte_eal_intr_init() < 0) {
859 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
860 fctret = -1;
861 goto finished;
862 }
863
864 /* Put mp channel init before bus scan so that we can init the vdev
865 * bus through mp channel in the secondary process before the bus scan.
866 */
867 if (rte_mp_channel_init() < 0) {
868 rte_eal_init_alert("failed to init mp channel\n");
869 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
870 rte_errno = EFAULT;
871 fctret = -1;
872 goto finished;
873 }
874 }
875
876 if (rte_bus_scan()) {
877 rte_eal_init_alert("Cannot scan the buses for devices\n");
878 rte_errno = ENODEV;
879 rte_atomic32_clear(&run_once);
880 fctret = -1;
881 goto finished;
882 }
883
884 /* autodetect the iova mapping mode (default is iova_pa) */
885 rte_eal_get_configuration()->iova_mode = rte_bus_get_iommu_class();
886
887 /* Workaround for KNI which requires physical address to work */
888 if (rte_eal_get_configuration()->iova_mode == RTE_IOVA_VA &&
889 rte_eal_check_module("rte_kni") == 1) {
890 rte_eal_get_configuration()->iova_mode = RTE_IOVA_PA;
891 RTE_LOG(WARNING, EAL,
892 "Some devices want IOVA as VA but PA will be used because.. "
893 "KNI module inserted\n");
894 }
895
896 if (internal_config.no_hugetlbfs == 0) {
897 /* rte_config isn't initialized yet */
898 ret = internal_config.process_type == RTE_PROC_PRIMARY ?
899 eal_hugepage_info_init() :
900 eal_hugepage_info_read();
901 if (ret < 0) {
902 rte_eal_init_alert("Cannot get hugepage information.");
903 rte_errno = EACCES;
904 rte_atomic32_clear(&run_once);
905 fctret = -1;
906 goto finished;
907 }
908 }
909
910 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
911 if (internal_config.no_hugetlbfs)
912 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
913 }
914
915 if (internal_config.vmware_tsc_map == 1) {
916 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
917 rte_cycles_vmware_tsc_map = 1;
918 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
919 "you must have monitor_control.pseudo_perfctr = TRUE\n");
920 #else
921 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
922 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
923 #endif
924 }
925
926 rte_srand(rte_rdtsc());
927
928 if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0) {
929 rte_eal_init_alert("Cannot init logging.");
930 rte_errno = ENOMEM;
931 rte_atomic32_clear(&run_once);
932 fctret = -1;
933 goto finished;
934 }
935
936 #ifdef VFIO_PRESENT
937 if (rte_eal_vfio_setup() < 0) {
938 rte_eal_init_alert("Cannot init VFIO\n");
939 rte_errno = EAGAIN;
940 rte_atomic32_clear(&run_once);
941 fctret = -1;
942 goto finished;
943 }
944 #endif
945 /* in secondary processes, memory init may allocate additional fbarrays
946 * not present in primary processes, so to avoid any potential issues,
947 * initialize memzones first.
948 */
949 if (rte_eal_memzone_init() < 0) {
950 rte_eal_init_alert("Cannot init memzone\n");
951 rte_errno = ENODEV;
952 fctret = -1;
953 goto finished;
954 }
955
956 if (rte_eal_memory_init() < 0) {
957 rte_eal_init_alert("Cannot init memory\n");
958 rte_errno = ENOMEM;
959 fctret = -1;
960 goto finished;
961 }
962
963 /* the directories are locked during eal_hugepage_info_init */
964 eal_hugedirs_unlock();
965
966 if (rte_eal_malloc_heap_init() < 0) {
967 rte_eal_init_alert("Cannot init malloc heap\n");
968 rte_errno = ENODEV;
969 fctret = -1;
970 goto finished;
971 }
972
973 if (rte_eal_tailqs_init() < 0) {
974 rte_eal_init_alert("Cannot init tail queues for objects\n");
975 rte_errno = EFAULT;
976 fctret = -1;
977 goto finished;
978 }
979
980 if (rte_eal_alarm_init() < 0) {
981 rte_eal_init_alert("Cannot init interrupt-handling thread\n");
982 /* rte_eal_alarm_init sets rte_errno on failure. */
983 fctret = -1;
984 goto finished;
985 }
986
987 if (rte_eal_timer_init() < 0) {
988 rte_eal_init_alert("Cannot init HPET or TSC timers\n");
989 rte_errno = ENOTSUP;
990 fctret = -1;
991 goto finished;
992 }
993
994 eal_check_mem_on_local_socket();
995
996 eal_thread_init_master(rte_config.master_lcore);
997
998 ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
999
1000 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%x;cpuset=[%s%s])\n",
1001 rte_config.master_lcore, (int)thread_id, cpuset,
1002 ret == 0 ? "" : "...");
1003
1004 RTE_LCORE_FOREACH_SLAVE(i) {
1005
1006 /*
1007 * create communication pipes between master thread
1008 * and children
1009 */
1010 if (pipe(lcore_config[i].pipe_master2slave) < 0)
1011 rte_panic("Cannot create pipe\n");
1012 if (pipe(lcore_config[i].pipe_slave2master) < 0)
1013 rte_panic("Cannot create pipe\n");
1014
1015 lcore_config[i].state = WAIT;
1016
1017 /* create a thread for each lcore */
1018 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1019 eal_thread_loop, NULL);
1020 if (ret != 0)
1021 rte_panic("Cannot create thread\n");
1022
1023 /* Set thread_name for aid in debugging. */
1024 snprintf(thread_name, sizeof(thread_name),
1025 "lcore-slave-%d", i);
1026 ret = rte_thread_setname(lcore_config[i].thread_id,
1027 thread_name);
1028 if (ret != 0)
1029 RTE_LOG(DEBUG, EAL,
1030 "Cannot set name for lcore thread\n");
1031 }
1032
1033 /*
1034 * Launch a dummy function on all slave lcores, so that master lcore
1035 * knows they are all ready when this function returns.
1036 */
1037 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
1038 rte_eal_mp_wait_lcore();
1039
1040 /* initialize services so vdevs register service during bus_probe. */
1041 ret = rte_service_init();
1042 if (ret) {
1043 rte_eal_init_alert("rte_service_init() failed\n");
1044 rte_errno = ENOEXEC;
1045 fctret = -1;
1046 goto finished;
1047 }
1048
1049 /* Probe all the buses and devices/drivers on them */
1050 if (rte_bus_probe()) {
1051 rte_eal_init_alert("Cannot probe devices\n");
1052 rte_errno = ENOTSUP;
1053 fctret = -1;
1054 goto finished;
1055 }
1056
1057 #ifdef VFIO_PRESENT
1058 /* Register mp action after probe() so that we got enough info */
1059 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0) {
1060 fctret = -1;
1061 goto finished;
1062 }
1063 #endif
1064
1065 /* initialize default service/lcore mappings and start running. Ignore
1066 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1067 */
1068 ret = rte_service_start_with_defaults();
1069 if (ret < 0 && ret != -ENOTSUP) {
1070 rte_errno = ENOEXEC;
1071 fctret = -1;
1072 goto finished;
1073 }
1074
1075 rte_eal_mcfg_complete();
1076
1077 finished:
1078 free(logid_storage);
1079 return fctret;
1080 }
1081
1082 static int
1083 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1084 void *arg __rte_unused)
1085 {
1086 /* ms is const, so find this memseg */
1087 struct rte_memseg *found = rte_mem_virt2memseg(ms->addr, msl);
1088
1089 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1090
1091 return 0;
1092 }
1093
1094 int __rte_experimental
1095 rte_eal_cleanup(void)
1096 {
1097 /* if we're in a primary process, we need to mark hugepages as freeable
1098 * so that finalization can release them back to the system.
1099 */
1100 if (rte_eal_process_type() == RTE_PROC_PRIMARY)
1101 rte_memseg_walk(mark_freeable, NULL);
1102 rte_service_finalize();
1103 return 0;
1104 }
1105
1106 /* get core role */
1107 enum rte_lcore_role_t
1108 rte_eal_lcore_role(unsigned lcore_id)
1109 {
1110 return rte_config.lcore_role[lcore_id];
1111 }
1112
1113 enum rte_proc_type_t
1114 rte_eal_process_type(void)
1115 {
1116 return rte_config.process_type;
1117 }
1118
1119 int rte_eal_has_hugepages(void)
1120 {
1121 return ! internal_config.no_hugetlbfs;
1122 }
1123
1124 int rte_eal_has_pci(void)
1125 {
1126 return !internal_config.no_pci;
1127 }
1128
1129 int rte_eal_create_uio_dev(void)
1130 {
1131 return internal_config.create_uio_dev;
1132 }
1133
1134 enum rte_intr_mode
1135 rte_eal_vfio_intr_mode(void)
1136 {
1137 return internal_config.vfio_intr_mode;
1138 }
1139
1140 int
1141 rte_eal_check_module(const char *module_name)
1142 {
1143 char sysfs_mod_name[PATH_MAX];
1144 struct stat st;
1145 int n;
1146
1147 if (NULL == module_name)
1148 return -1;
1149
1150 /* Check if there is sysfs mounted */
1151 if (stat("/sys/module", &st) != 0) {
1152 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1153 errno, strerror(errno));
1154 return -1;
1155 }
1156
1157 /* A module might be built-in, therefore try sysfs */
1158 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1159 if (n < 0 || n > PATH_MAX) {
1160 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1161 return -1;
1162 }
1163
1164 if (stat(sysfs_mod_name, &st) != 0) {
1165 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1166 sysfs_mod_name, errno, strerror(errno));
1167 return 0;
1168 }
1169
1170 /* Module has been found */
1171 return 1;
1172 }