]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/lib/librte_eal/linux/eal/eal.c
import 15.2.0 Octopus source
[ceph.git] / ceph / src / spdk / dpdk / lib / librte_eal / linux / 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 <dirent.h>
17 #include <fcntl.h>
18 #include <fnmatch.h>
19 #include <stddef.h>
20 #include <errno.h>
21 #include <limits.h>
22 #include <sys/mman.h>
23 #include <sys/queue.h>
24 #include <sys/stat.h>
25 #if defined(RTE_ARCH_X86)
26 #include <sys/io.h>
27 #endif
28
29 #include <rte_compat.h>
30 #include <rte_common.h>
31 #include <rte_debug.h>
32 #include <rte_memory.h>
33 #include <rte_launch.h>
34 #include <rte_eal.h>
35 #include <rte_eal_memconfig.h>
36 #include <rte_errno.h>
37 #include <rte_per_lcore.h>
38 #include <rte_lcore.h>
39 #include <rte_service_component.h>
40 #include <rte_log.h>
41 #include <rte_random.h>
42 #include <rte_cycles.h>
43 #include <rte_string_fns.h>
44 #include <rte_cpuflags.h>
45 #include <rte_interrupts.h>
46 #include <rte_bus.h>
47 #include <rte_dev.h>
48 #include <rte_devargs.h>
49 #include <rte_version.h>
50 #include <rte_atomic.h>
51 #include <malloc_heap.h>
52 #include <rte_vfio.h>
53 #include <rte_option.h>
54
55 #include "eal_private.h"
56 #include "eal_thread.h"
57 #include "eal_internal_cfg.h"
58 #include "eal_filesystem.h"
59 #include "eal_hugepages.h"
60 #include "eal_options.h"
61 #include "eal_vfio.h"
62
63 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
64
65 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
66
67 /* Allow the application to print its usage message too if set */
68 static rte_usage_hook_t rte_application_usage_hook = NULL;
69
70 /* early configuration structure, when memory config is not mmapped */
71 static struct rte_mem_config early_mem_config;
72
73 /* define fd variable here, because file needs to be kept open for the
74 * duration of the program, as we hold a write lock on it in the primary proc */
75 static int mem_cfg_fd = -1;
76
77 static struct flock wr_lock = {
78 .l_type = F_WRLCK,
79 .l_whence = SEEK_SET,
80 .l_start = offsetof(struct rte_mem_config, memsegs),
81 .l_len = sizeof(early_mem_config.memsegs),
82 };
83
84 /* Address of global and public configuration */
85 static struct rte_config rte_config = {
86 .mem_config = &early_mem_config,
87 };
88
89 /* internal configuration (per-core) */
90 struct lcore_config lcore_config[RTE_MAX_LCORE];
91
92 /* internal configuration */
93 struct internal_config internal_config;
94
95 /* used by rte_rdtsc() */
96 int rte_cycles_vmware_tsc_map;
97
98 /* platform-specific runtime dir */
99 static char runtime_dir[PATH_MAX];
100
101 static const char *default_runtime_dir = "/var/run";
102
103 int
104 eal_create_runtime_dir(void)
105 {
106 const char *directory = default_runtime_dir;
107 const char *xdg_runtime_dir = getenv("XDG_RUNTIME_DIR");
108 const char *fallback = "/tmp";
109 char tmp[PATH_MAX];
110 int ret;
111
112 if (getuid() != 0) {
113 /* try XDG path first, fall back to /tmp */
114 if (xdg_runtime_dir != NULL)
115 directory = xdg_runtime_dir;
116 else
117 directory = fallback;
118 }
119 /* create DPDK subdirectory under runtime dir */
120 ret = snprintf(tmp, sizeof(tmp), "%s/dpdk", directory);
121 if (ret < 0 || ret == sizeof(tmp)) {
122 RTE_LOG(ERR, EAL, "Error creating DPDK runtime path name\n");
123 return -1;
124 }
125
126 /* create prefix-specific subdirectory under DPDK runtime dir */
127 ret = snprintf(runtime_dir, sizeof(runtime_dir), "%s/%s",
128 tmp, eal_get_hugefile_prefix());
129 if (ret < 0 || ret == sizeof(runtime_dir)) {
130 RTE_LOG(ERR, EAL, "Error creating prefix-specific runtime path name\n");
131 return -1;
132 }
133
134 /* create the path if it doesn't exist. no "mkdir -p" here, so do it
135 * step by step.
136 */
137 ret = mkdir(tmp, 0700);
138 if (ret < 0 && errno != EEXIST) {
139 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
140 tmp, strerror(errno));
141 return -1;
142 }
143
144 ret = mkdir(runtime_dir, 0700);
145 if (ret < 0 && errno != EEXIST) {
146 RTE_LOG(ERR, EAL, "Error creating '%s': %s\n",
147 runtime_dir, strerror(errno));
148 return -1;
149 }
150
151 return 0;
152 }
153
154 int
155 eal_clean_runtime_dir(void)
156 {
157 DIR *dir;
158 struct dirent *dirent;
159 int dir_fd, fd, lck_result;
160 static const char * const filters[] = {
161 "fbarray_*",
162 "mp_socket_*"
163 };
164
165 /* open directory */
166 dir = opendir(runtime_dir);
167 if (!dir) {
168 RTE_LOG(ERR, EAL, "Unable to open runtime directory %s\n",
169 runtime_dir);
170 goto error;
171 }
172 dir_fd = dirfd(dir);
173
174 /* lock the directory before doing anything, to avoid races */
175 if (flock(dir_fd, LOCK_EX) < 0) {
176 RTE_LOG(ERR, EAL, "Unable to lock runtime directory %s\n",
177 runtime_dir);
178 goto error;
179 }
180
181 dirent = readdir(dir);
182 if (!dirent) {
183 RTE_LOG(ERR, EAL, "Unable to read runtime directory %s\n",
184 runtime_dir);
185 goto error;
186 }
187
188 while (dirent != NULL) {
189 unsigned int f_idx;
190 bool skip = true;
191
192 /* skip files that don't match the patterns */
193 for (f_idx = 0; f_idx < RTE_DIM(filters); f_idx++) {
194 const char *filter = filters[f_idx];
195
196 if (fnmatch(filter, dirent->d_name, 0) == 0) {
197 skip = false;
198 break;
199 }
200 }
201 if (skip) {
202 dirent = readdir(dir);
203 continue;
204 }
205
206 /* try and lock the file */
207 fd = openat(dir_fd, dirent->d_name, O_RDONLY);
208
209 /* skip to next file */
210 if (fd == -1) {
211 dirent = readdir(dir);
212 continue;
213 }
214
215 /* non-blocking lock */
216 lck_result = flock(fd, LOCK_EX | LOCK_NB);
217
218 /* if lock succeeds, remove the file */
219 if (lck_result != -1)
220 unlinkat(dir_fd, dirent->d_name, 0);
221 close(fd);
222 dirent = readdir(dir);
223 }
224
225 /* closedir closes dir_fd and drops the lock */
226 closedir(dir);
227 return 0;
228
229 error:
230 if (dir)
231 closedir(dir);
232
233 RTE_LOG(ERR, EAL, "Error while clearing runtime dir: %s\n",
234 strerror(errno));
235
236 return -1;
237 }
238
239 const char *
240 rte_eal_get_runtime_dir(void)
241 {
242 return runtime_dir;
243 }
244
245 /* Return user provided mbuf pool ops name */
246 const char *
247 rte_eal_mbuf_user_pool_ops(void)
248 {
249 return internal_config.user_mbuf_pool_ops_name;
250 }
251
252 /* Return a pointer to the configuration structure */
253 struct rte_config *
254 rte_eal_get_configuration(void)
255 {
256 return &rte_config;
257 }
258
259 enum rte_iova_mode
260 rte_eal_iova_mode(void)
261 {
262 return rte_eal_get_configuration()->iova_mode;
263 }
264
265 /* parse a sysfs (or other) file containing one integer value */
266 int
267 eal_parse_sysfs_value(const char *filename, unsigned long *val)
268 {
269 FILE *f;
270 char buf[BUFSIZ];
271 char *end = NULL;
272
273 if ((f = fopen(filename, "r")) == NULL) {
274 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
275 __func__, filename);
276 return -1;
277 }
278
279 if (fgets(buf, sizeof(buf), f) == NULL) {
280 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
281 __func__, filename);
282 fclose(f);
283 return -1;
284 }
285 *val = strtoul(buf, &end, 0);
286 if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
287 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
288 __func__, filename);
289 fclose(f);
290 return -1;
291 }
292 fclose(f);
293 return 0;
294 }
295
296
297 /* create memory configuration in shared/mmap memory. Take out
298 * a write lock on the memsegs, so we can auto-detect primary/secondary.
299 * This means we never close the file while running (auto-close on exit).
300 * We also don't lock the whole file, so that in future we can use read-locks
301 * on other parts, e.g. memzones, to detect if there are running secondary
302 * processes. */
303 static void
304 rte_eal_config_create(void)
305 {
306 void *rte_mem_cfg_addr;
307 int retval;
308
309 const char *pathname = eal_runtime_config_path();
310
311 if (internal_config.no_shconf)
312 return;
313
314 /* map the config before hugepage address so that we don't waste a page */
315 if (internal_config.base_virtaddr != 0)
316 rte_mem_cfg_addr = (void *)
317 RTE_ALIGN_FLOOR(internal_config.base_virtaddr -
318 sizeof(struct rte_mem_config), sysconf(_SC_PAGE_SIZE));
319 else
320 rte_mem_cfg_addr = NULL;
321
322 if (mem_cfg_fd < 0){
323 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
324 if (mem_cfg_fd < 0)
325 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
326 }
327
328 retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
329 if (retval < 0){
330 close(mem_cfg_fd);
331 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
332 }
333
334 retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
335 if (retval < 0){
336 close(mem_cfg_fd);
337 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
338 "process running?\n", pathname);
339 }
340
341 rte_mem_cfg_addr = mmap(rte_mem_cfg_addr, sizeof(*rte_config.mem_config),
342 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
343
344 if (rte_mem_cfg_addr == MAP_FAILED){
345 rte_panic("Cannot mmap memory for rte_config\n");
346 }
347 memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
348 rte_config.mem_config = rte_mem_cfg_addr;
349
350 /* store address of the config in the config itself so that secondary
351 * processes could later map the config into this exact location */
352 rte_config.mem_config->mem_cfg_addr = (uintptr_t) rte_mem_cfg_addr;
353
354 rte_config.mem_config->dma_maskbits = 0;
355
356 }
357
358 /* attach to an existing shared memory config */
359 static void
360 rte_eal_config_attach(void)
361 {
362 struct rte_mem_config *mem_config;
363
364 const char *pathname = eal_runtime_config_path();
365
366 if (internal_config.no_shconf)
367 return;
368
369 if (mem_cfg_fd < 0){
370 mem_cfg_fd = open(pathname, O_RDWR);
371 if (mem_cfg_fd < 0)
372 rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
373 }
374
375 /* map it as read-only first */
376 mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
377 PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
378 if (mem_config == MAP_FAILED)
379 rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
380 errno, strerror(errno));
381
382 rte_config.mem_config = mem_config;
383 }
384
385 /* reattach the shared config at exact memory location primary process has it */
386 static void
387 rte_eal_config_reattach(void)
388 {
389 struct rte_mem_config *mem_config;
390 void *rte_mem_cfg_addr;
391
392 if (internal_config.no_shconf)
393 return;
394
395 /* save the address primary process has mapped shared config to */
396 rte_mem_cfg_addr = (void *) (uintptr_t) rte_config.mem_config->mem_cfg_addr;
397
398 /* unmap original config */
399 munmap(rte_config.mem_config, sizeof(struct rte_mem_config));
400
401 /* remap the config at proper address */
402 mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
403 sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
404 mem_cfg_fd, 0);
405 if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
406 if (mem_config != MAP_FAILED)
407 /* errno is stale, don't use */
408 rte_panic("Cannot mmap memory for rte_config at [%p], got [%p]"
409 " - please use '--base-virtaddr' option\n",
410 rte_mem_cfg_addr, mem_config);
411 else
412 rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
413 errno, strerror(errno));
414 }
415 close(mem_cfg_fd);
416
417 rte_config.mem_config = mem_config;
418 }
419
420 /* Detect if we are a primary or a secondary process */
421 enum rte_proc_type_t
422 eal_proc_type_detect(void)
423 {
424 enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
425 const char *pathname = eal_runtime_config_path();
426
427 /* if there no shared config, there can be no secondary processes */
428 if (!internal_config.no_shconf) {
429 /* if we can open the file but not get a write-lock we are a
430 * secondary process. NOTE: if we get a file handle back, we
431 * keep that open and don't close it to prevent a race condition
432 * between multiple opens.
433 */
434 if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
435 (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
436 ptype = RTE_PROC_SECONDARY;
437 }
438
439 RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
440 ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
441
442 return ptype;
443 }
444
445 /* copies data from internal config to shared config */
446 static void
447 eal_update_mem_config(void)
448 {
449 struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
450 mcfg->legacy_mem = internal_config.legacy_mem;
451 mcfg->single_file_segments = internal_config.single_file_segments;
452 }
453
454 /* copies data from shared config to internal config */
455 static void
456 eal_update_internal_config(void)
457 {
458 struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
459 internal_config.legacy_mem = mcfg->legacy_mem;
460 internal_config.single_file_segments = mcfg->single_file_segments;
461 }
462
463 /* Sets up rte_config structure with the pointer to shared memory config.*/
464 static void
465 rte_config_init(void)
466 {
467 rte_config.process_type = internal_config.process_type;
468
469 switch (rte_config.process_type){
470 case RTE_PROC_PRIMARY:
471 rte_eal_config_create();
472 eal_update_mem_config();
473 break;
474 case RTE_PROC_SECONDARY:
475 rte_eal_config_attach();
476 rte_eal_mcfg_wait_complete(rte_config.mem_config);
477 rte_eal_config_reattach();
478 eal_update_internal_config();
479 break;
480 case RTE_PROC_AUTO:
481 case RTE_PROC_INVALID:
482 rte_panic("Invalid process type\n");
483 }
484 }
485
486 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
487 static void
488 eal_hugedirs_unlock(void)
489 {
490 int i;
491
492 for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
493 {
494 /* skip uninitialized */
495 if (internal_config.hugepage_info[i].lock_descriptor < 0)
496 continue;
497 /* unlock hugepage file */
498 flock(internal_config.hugepage_info[i].lock_descriptor, LOCK_UN);
499 close(internal_config.hugepage_info[i].lock_descriptor);
500 /* reset the field */
501 internal_config.hugepage_info[i].lock_descriptor = -1;
502 }
503 }
504
505 /* display usage */
506 static void
507 eal_usage(const char *prgname)
508 {
509 printf("\nUsage: %s ", prgname);
510 eal_common_usage();
511 printf("EAL Linux options:\n"
512 " --"OPT_SOCKET_MEM" Memory to allocate on sockets (comma separated values)\n"
513 " --"OPT_SOCKET_LIMIT" Limit memory allocation on sockets (comma separated values)\n"
514 " --"OPT_HUGE_DIR" Directory where hugetlbfs is mounted\n"
515 " --"OPT_FILE_PREFIX" Prefix for hugepage filenames\n"
516 " --"OPT_BASE_VIRTADDR" Base virtual address\n"
517 " --"OPT_CREATE_UIO_DEV" Create /dev/uioX (usually done by hotplug)\n"
518 " --"OPT_VFIO_INTR" Interrupt mode for VFIO (legacy|msi|msix)\n"
519 " --"OPT_LEGACY_MEM" Legacy memory mode (no dynamic allocation, contiguous segments)\n"
520 " --"OPT_SINGLE_FILE_SEGMENTS" Put all hugepage memory in single files\n"
521 " --"OPT_MATCH_ALLOCATIONS" Free hugepages exactly as allocated\n"
522 "\n");
523 /* Allow the application to print its usage message too if hook is set */
524 if ( rte_application_usage_hook ) {
525 printf("===== Application Usage =====\n\n");
526 rte_application_usage_hook(prgname);
527 }
528 }
529
530 /* Set a per-application usage message */
531 rte_usage_hook_t
532 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
533 {
534 rte_usage_hook_t old_func;
535
536 /* Will be NULL on the first call to denote the last usage routine. */
537 old_func = rte_application_usage_hook;
538 rte_application_usage_hook = usage_func;
539
540 return old_func;
541 }
542
543 static int
544 eal_parse_socket_arg(char *strval, volatile uint64_t *socket_arg)
545 {
546 char * arg[RTE_MAX_NUMA_NODES];
547 char *end;
548 int arg_num, i, len;
549 uint64_t total_mem = 0;
550
551 len = strnlen(strval, SOCKET_MEM_STRLEN);
552 if (len == SOCKET_MEM_STRLEN) {
553 RTE_LOG(ERR, EAL, "--socket-mem is too long\n");
554 return -1;
555 }
556
557 /* all other error cases will be caught later */
558 if (!isdigit(strval[len-1]))
559 return -1;
560
561 /* split the optarg into separate socket values */
562 arg_num = rte_strsplit(strval, len,
563 arg, RTE_MAX_NUMA_NODES, ',');
564
565 /* if split failed, or 0 arguments */
566 if (arg_num <= 0)
567 return -1;
568
569 /* parse each defined socket option */
570 errno = 0;
571 for (i = 0; i < arg_num; i++) {
572 uint64_t val;
573 end = NULL;
574 val = strtoull(arg[i], &end, 10);
575
576 /* check for invalid input */
577 if ((errno != 0) ||
578 (arg[i][0] == '\0') || (end == NULL) || (*end != '\0'))
579 return -1;
580 val <<= 20;
581 total_mem += val;
582 socket_arg[i] = val;
583 }
584
585 return 0;
586 }
587
588 static int
589 eal_parse_base_virtaddr(const char *arg)
590 {
591 char *end;
592 uint64_t addr;
593
594 errno = 0;
595 addr = strtoull(arg, &end, 16);
596
597 /* check for errors */
598 if ((errno != 0) || (arg[0] == '\0') || end == NULL || (*end != '\0'))
599 return -1;
600
601 /* make sure we don't exceed 32-bit boundary on 32-bit target */
602 #ifndef RTE_ARCH_64
603 if (addr >= UINTPTR_MAX)
604 return -1;
605 #endif
606
607 /* align the addr on 16M boundary, 16MB is the minimum huge page
608 * size on IBM Power architecture. If the addr is aligned to 16MB,
609 * it can align to 2MB for x86. So this alignment can also be used
610 * on x86 */
611 internal_config.base_virtaddr =
612 RTE_PTR_ALIGN_CEIL((uintptr_t)addr, (size_t)RTE_PGSIZE_16M);
613
614 return 0;
615 }
616
617 static int
618 eal_parse_vfio_intr(const char *mode)
619 {
620 unsigned i;
621 static struct {
622 const char *name;
623 enum rte_intr_mode value;
624 } map[] = {
625 { "legacy", RTE_INTR_MODE_LEGACY },
626 { "msi", RTE_INTR_MODE_MSI },
627 { "msix", RTE_INTR_MODE_MSIX },
628 };
629
630 for (i = 0; i < RTE_DIM(map); i++) {
631 if (!strcmp(mode, map[i].name)) {
632 internal_config.vfio_intr_mode = map[i].value;
633 return 0;
634 }
635 }
636 return -1;
637 }
638
639 /* Parse the arguments for --log-level only */
640 static void
641 eal_log_level_parse(int argc, char **argv)
642 {
643 int opt;
644 char **argvopt;
645 int option_index;
646 const int old_optind = optind;
647 const int old_optopt = optopt;
648 char * const old_optarg = optarg;
649
650 argvopt = argv;
651 optind = 1;
652
653 while ((opt = getopt_long(argc, argvopt, eal_short_options,
654 eal_long_options, &option_index)) != EOF) {
655
656 int ret;
657
658 /* getopt is not happy, stop right now */
659 if (opt == '?')
660 break;
661
662 ret = (opt == OPT_LOG_LEVEL_NUM) ?
663 eal_parse_common_option(opt, optarg, &internal_config) : 0;
664
665 /* common parser is not happy */
666 if (ret < 0)
667 break;
668 }
669
670 /* restore getopt lib */
671 optind = old_optind;
672 optopt = old_optopt;
673 optarg = old_optarg;
674 }
675
676 /* Parse the argument given in the command line of the application */
677 static int
678 eal_parse_args(int argc, char **argv)
679 {
680 int opt, ret;
681 char **argvopt;
682 int option_index;
683 char *prgname = argv[0];
684 const int old_optind = optind;
685 const int old_optopt = optopt;
686 char * const old_optarg = optarg;
687
688 argvopt = argv;
689 optind = 1;
690 opterr = 0;
691
692 while ((opt = getopt_long(argc, argvopt, eal_short_options,
693 eal_long_options, &option_index)) != EOF) {
694
695 /*
696 * getopt didn't recognise the option, lets parse the
697 * registered options to see if the flag is valid
698 */
699 if (opt == '?') {
700 ret = rte_option_parse(argv[optind-1]);
701 if (ret == 0)
702 continue;
703
704 eal_usage(prgname);
705 ret = -1;
706 goto out;
707 }
708
709 ret = eal_parse_common_option(opt, optarg, &internal_config);
710 /* common parser is not happy */
711 if (ret < 0) {
712 eal_usage(prgname);
713 ret = -1;
714 goto out;
715 }
716 /* common parser handled this option */
717 if (ret == 0)
718 continue;
719
720 switch (opt) {
721 case 'h':
722 eal_usage(prgname);
723 exit(EXIT_SUCCESS);
724
725 case OPT_HUGE_DIR_NUM:
726 {
727 char *hdir = strdup(optarg);
728 if (hdir == NULL)
729 RTE_LOG(ERR, EAL, "Could not store hugepage directory\n");
730 else {
731 /* free old hugepage dir */
732 if (internal_config.hugepage_dir != NULL)
733 free(internal_config.hugepage_dir);
734 internal_config.hugepage_dir = hdir;
735 }
736 break;
737 }
738 case OPT_FILE_PREFIX_NUM:
739 {
740 char *prefix = strdup(optarg);
741 if (prefix == NULL)
742 RTE_LOG(ERR, EAL, "Could not store file prefix\n");
743 else {
744 /* free old prefix */
745 if (internal_config.hugefile_prefix != NULL)
746 free(internal_config.hugefile_prefix);
747 internal_config.hugefile_prefix = prefix;
748 }
749 break;
750 }
751 case OPT_SOCKET_MEM_NUM:
752 if (eal_parse_socket_arg(optarg,
753 internal_config.socket_mem) < 0) {
754 RTE_LOG(ERR, EAL, "invalid parameters for --"
755 OPT_SOCKET_MEM "\n");
756 eal_usage(prgname);
757 ret = -1;
758 goto out;
759 }
760 internal_config.force_sockets = 1;
761 break;
762
763 case OPT_SOCKET_LIMIT_NUM:
764 if (eal_parse_socket_arg(optarg,
765 internal_config.socket_limit) < 0) {
766 RTE_LOG(ERR, EAL, "invalid parameters for --"
767 OPT_SOCKET_LIMIT "\n");
768 eal_usage(prgname);
769 ret = -1;
770 goto out;
771 }
772 internal_config.force_socket_limits = 1;
773 break;
774
775 case OPT_BASE_VIRTADDR_NUM:
776 if (eal_parse_base_virtaddr(optarg) < 0) {
777 RTE_LOG(ERR, EAL, "invalid parameter for --"
778 OPT_BASE_VIRTADDR "\n");
779 eal_usage(prgname);
780 ret = -1;
781 goto out;
782 }
783 break;
784
785 case OPT_VFIO_INTR_NUM:
786 if (eal_parse_vfio_intr(optarg) < 0) {
787 RTE_LOG(ERR, EAL, "invalid parameters for --"
788 OPT_VFIO_INTR "\n");
789 eal_usage(prgname);
790 ret = -1;
791 goto out;
792 }
793 break;
794
795 case OPT_CREATE_UIO_DEV_NUM:
796 internal_config.create_uio_dev = 1;
797 break;
798
799 case OPT_MBUF_POOL_OPS_NAME_NUM:
800 {
801 char *ops_name = strdup(optarg);
802 if (ops_name == NULL)
803 RTE_LOG(ERR, EAL, "Could not store mbuf pool ops name\n");
804 else {
805 /* free old ops name */
806 if (internal_config.user_mbuf_pool_ops_name !=
807 NULL)
808 free(internal_config.user_mbuf_pool_ops_name);
809
810 internal_config.user_mbuf_pool_ops_name =
811 ops_name;
812 }
813 break;
814 }
815 case OPT_MATCH_ALLOCATIONS_NUM:
816 internal_config.match_allocations = 1;
817 break;
818
819 default:
820 if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
821 RTE_LOG(ERR, EAL, "Option %c is not supported "
822 "on Linux\n", opt);
823 } else if (opt >= OPT_LONG_MIN_NUM &&
824 opt < OPT_LONG_MAX_NUM) {
825 RTE_LOG(ERR, EAL, "Option %s is not supported "
826 "on Linux\n",
827 eal_long_options[option_index].name);
828 } else {
829 RTE_LOG(ERR, EAL, "Option %d is not supported "
830 "on Linux\n", opt);
831 }
832 eal_usage(prgname);
833 ret = -1;
834 goto out;
835 }
836 }
837
838 /* create runtime data directory */
839 if (internal_config.no_shconf == 0 &&
840 eal_create_runtime_dir() < 0) {
841 RTE_LOG(ERR, EAL, "Cannot create runtime directory\n");
842 ret = -1;
843 goto out;
844 }
845
846 if (eal_adjust_config(&internal_config) != 0) {
847 ret = -1;
848 goto out;
849 }
850
851 /* sanity checks */
852 if (eal_check_common_options(&internal_config) != 0) {
853 eal_usage(prgname);
854 ret = -1;
855 goto out;
856 }
857
858 if (optind >= 0)
859 argv[optind-1] = prgname;
860 ret = optind-1;
861
862 out:
863 /* restore getopt lib */
864 optind = old_optind;
865 optopt = old_optopt;
866 optarg = old_optarg;
867
868 return ret;
869 }
870
871 static int
872 check_socket(const struct rte_memseg_list *msl, void *arg)
873 {
874 int *socket_id = arg;
875
876 if (msl->external)
877 return 0;
878
879 return *socket_id == msl->socket_id;
880 }
881
882 static void
883 eal_check_mem_on_local_socket(void)
884 {
885 int socket_id;
886
887 socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
888
889 if (rte_memseg_list_walk(check_socket, &socket_id) == 0)
890 RTE_LOG(WARNING, EAL, "WARNING: Master core has no memory on local socket!\n");
891 }
892
893 static int
894 sync_func(__attribute__((unused)) void *arg)
895 {
896 return 0;
897 }
898
899 inline static void
900 rte_eal_mcfg_complete(void)
901 {
902 /* ALL shared mem_config related INIT DONE */
903 if (rte_config.process_type == RTE_PROC_PRIMARY)
904 rte_config.mem_config->magic = RTE_MAGIC;
905
906 internal_config.init_complete = 1;
907 }
908
909 /*
910 * Request iopl privilege for all RPL, returns 0 on success
911 * iopl() call is mostly for the i386 architecture. For other architectures,
912 * return -1 to indicate IO privilege can't be changed in this way.
913 */
914 int
915 rte_eal_iopl_init(void)
916 {
917 #if defined(RTE_ARCH_X86)
918 if (iopl(3) != 0)
919 return -1;
920 #endif
921 return 0;
922 }
923
924 #ifdef VFIO_PRESENT
925 static int rte_eal_vfio_setup(void)
926 {
927 if (rte_vfio_enable("vfio"))
928 return -1;
929
930 return 0;
931 }
932 #endif
933
934 static void rte_eal_init_alert(const char *msg)
935 {
936 fprintf(stderr, "EAL: FATAL: %s\n", msg);
937 RTE_LOG(ERR, EAL, "%s\n", msg);
938 }
939
940 /* Launch threads, called at application init(). */
941 int
942 rte_eal_init(int argc, char **argv)
943 {
944 int i, fctret, ret;
945 pthread_t thread_id;
946 static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
947 const char *p;
948 static char logid[PATH_MAX];
949 char cpuset[RTE_CPU_AFFINITY_STR_LEN];
950 char thread_name[RTE_MAX_THREAD_NAME_LEN];
951
952 /* checks if the machine is adequate */
953 if (!rte_cpu_is_supported()) {
954 rte_eal_init_alert("unsupported cpu type.");
955 rte_errno = ENOTSUP;
956 return -1;
957 }
958
959 if (!rte_atomic32_test_and_set(&run_once)) {
960 rte_eal_init_alert("already called initialization.");
961 rte_errno = EALREADY;
962 return -1;
963 }
964
965 p = strrchr(argv[0], '/');
966 strlcpy(logid, p ? p + 1 : argv[0], sizeof(logid));
967 thread_id = pthread_self();
968
969 eal_reset_internal_config(&internal_config);
970
971 /* set log level as early as possible */
972 eal_log_level_parse(argc, argv);
973
974 if (rte_eal_cpu_init() < 0) {
975 rte_eal_init_alert("Cannot detect lcores.");
976 rte_errno = ENOTSUP;
977 return -1;
978 }
979
980 fctret = eal_parse_args(argc, argv);
981 if (fctret < 0) {
982 rte_eal_init_alert("Invalid 'command line' arguments.");
983 rte_errno = EINVAL;
984 rte_atomic32_clear(&run_once);
985 return -1;
986 }
987
988 if (eal_plugins_init() < 0) {
989 rte_eal_init_alert("Cannot init plugins");
990 rte_errno = EINVAL;
991 rte_atomic32_clear(&run_once);
992 return -1;
993 }
994
995 if (eal_option_device_parse()) {
996 rte_errno = ENODEV;
997 rte_atomic32_clear(&run_once);
998 return -1;
999 }
1000
1001 rte_config_init();
1002
1003 if (rte_eal_intr_init() < 0) {
1004 rte_eal_init_alert("Cannot init interrupt-handling thread");
1005 return -1;
1006 }
1007
1008 if (rte_eal_alarm_init() < 0) {
1009 rte_eal_init_alert("Cannot init interrupt-handling thread");
1010 /* rte_eal_alarm_init sets rte_errno on failure. */
1011 return -1;
1012 }
1013
1014 /* Put mp channel init before bus scan so that we can init the vdev
1015 * bus through mp channel in the secondary process before the bus scan.
1016 */
1017 if (rte_mp_channel_init() < 0) {
1018 rte_eal_init_alert("failed to init mp channel");
1019 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1020 rte_errno = EFAULT;
1021 return -1;
1022 }
1023 }
1024
1025 /* register multi-process action callbacks for hotplug */
1026 if (rte_mp_dev_hotplug_init() < 0) {
1027 rte_eal_init_alert("failed to register mp callback for hotplug");
1028 return -1;
1029 }
1030
1031 if (rte_bus_scan()) {
1032 rte_eal_init_alert("Cannot scan the buses for devices");
1033 rte_errno = ENODEV;
1034 rte_atomic32_clear(&run_once);
1035 return -1;
1036 }
1037
1038 /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
1039 if (internal_config.iova_mode == RTE_IOVA_DC) {
1040 /* autodetect the IOVA mapping mode (default is RTE_IOVA_PA) */
1041 rte_eal_get_configuration()->iova_mode =
1042 rte_bus_get_iommu_class();
1043
1044 /* Workaround for KNI which requires physical address to work */
1045 if (rte_eal_get_configuration()->iova_mode == RTE_IOVA_VA &&
1046 rte_eal_check_module("rte_kni") == 1) {
1047 rte_eal_get_configuration()->iova_mode = RTE_IOVA_PA;
1048 RTE_LOG(WARNING, EAL,
1049 "Some devices want IOVA as VA but PA will be used because.. "
1050 "KNI module inserted\n");
1051 }
1052 } else {
1053 rte_eal_get_configuration()->iova_mode =
1054 internal_config.iova_mode;
1055 }
1056
1057 if (internal_config.no_hugetlbfs == 0) {
1058 /* rte_config isn't initialized yet */
1059 ret = internal_config.process_type == RTE_PROC_PRIMARY ?
1060 eal_hugepage_info_init() :
1061 eal_hugepage_info_read();
1062 if (ret < 0) {
1063 rte_eal_init_alert("Cannot get hugepage information.");
1064 rte_errno = EACCES;
1065 rte_atomic32_clear(&run_once);
1066 return -1;
1067 }
1068 }
1069
1070 if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
1071 if (internal_config.no_hugetlbfs)
1072 internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
1073 }
1074
1075 if (internal_config.vmware_tsc_map == 1) {
1076 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
1077 rte_cycles_vmware_tsc_map = 1;
1078 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
1079 "you must have monitor_control.pseudo_perfctr = TRUE\n");
1080 #else
1081 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
1082 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
1083 #endif
1084 }
1085
1086 rte_srand(rte_rdtsc());
1087
1088 if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0) {
1089 rte_eal_init_alert("Cannot init logging.");
1090 rte_errno = ENOMEM;
1091 rte_atomic32_clear(&run_once);
1092 return -1;
1093 }
1094
1095 #ifdef VFIO_PRESENT
1096 if (rte_eal_vfio_setup() < 0) {
1097 rte_eal_init_alert("Cannot init VFIO");
1098 rte_errno = EAGAIN;
1099 rte_atomic32_clear(&run_once);
1100 return -1;
1101 }
1102 #endif
1103 /* in secondary processes, memory init may allocate additional fbarrays
1104 * not present in primary processes, so to avoid any potential issues,
1105 * initialize memzones first.
1106 */
1107 if (rte_eal_memzone_init() < 0) {
1108 rte_eal_init_alert("Cannot init memzone");
1109 rte_errno = ENODEV;
1110 return -1;
1111 }
1112
1113 if (rte_eal_memory_init() < 0) {
1114 rte_eal_init_alert("Cannot init memory");
1115 rte_errno = ENOMEM;
1116 return -1;
1117 }
1118
1119 /* the directories are locked during eal_hugepage_info_init */
1120 eal_hugedirs_unlock();
1121
1122 if (rte_eal_malloc_heap_init() < 0) {
1123 rte_eal_init_alert("Cannot init malloc heap");
1124 rte_errno = ENODEV;
1125 return -1;
1126 }
1127
1128 if (rte_eal_tailqs_init() < 0) {
1129 rte_eal_init_alert("Cannot init tail queues for objects");
1130 rte_errno = EFAULT;
1131 return -1;
1132 }
1133
1134 if (rte_eal_timer_init() < 0) {
1135 rte_eal_init_alert("Cannot init HPET or TSC timers");
1136 rte_errno = ENOTSUP;
1137 return -1;
1138 }
1139
1140 eal_check_mem_on_local_socket();
1141
1142 eal_thread_init_master(rte_config.master_lcore);
1143
1144 ret = eal_thread_dump_affinity(cpuset, sizeof(cpuset));
1145
1146 RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%zx;cpuset=[%s%s])\n",
1147 rte_config.master_lcore, (uintptr_t)thread_id, cpuset,
1148 ret == 0 ? "" : "...");
1149
1150 RTE_LCORE_FOREACH_SLAVE(i) {
1151
1152 /*
1153 * create communication pipes between master thread
1154 * and children
1155 */
1156 if (pipe(lcore_config[i].pipe_master2slave) < 0)
1157 rte_panic("Cannot create pipe\n");
1158 if (pipe(lcore_config[i].pipe_slave2master) < 0)
1159 rte_panic("Cannot create pipe\n");
1160
1161 lcore_config[i].state = WAIT;
1162
1163 /* create a thread for each lcore */
1164 ret = pthread_create(&lcore_config[i].thread_id, NULL,
1165 eal_thread_loop, NULL);
1166 if (ret != 0)
1167 rte_panic("Cannot create thread\n");
1168
1169 /* Set thread_name for aid in debugging. */
1170 snprintf(thread_name, sizeof(thread_name),
1171 "lcore-slave-%d", i);
1172 ret = rte_thread_setname(lcore_config[i].thread_id,
1173 thread_name);
1174 if (ret != 0)
1175 RTE_LOG(DEBUG, EAL,
1176 "Cannot set name for lcore thread\n");
1177 }
1178
1179 /*
1180 * Launch a dummy function on all slave lcores, so that master lcore
1181 * knows they are all ready when this function returns.
1182 */
1183 rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
1184 rte_eal_mp_wait_lcore();
1185
1186 /* initialize services so vdevs register service during bus_probe. */
1187 ret = rte_service_init();
1188 if (ret) {
1189 rte_eal_init_alert("rte_service_init() failed");
1190 rte_errno = ENOEXEC;
1191 return -1;
1192 }
1193
1194 /* Probe all the buses and devices/drivers on them */
1195 if (rte_bus_probe()) {
1196 rte_eal_init_alert("Cannot probe devices");
1197 rte_errno = ENOTSUP;
1198 return -1;
1199 }
1200
1201 #ifdef VFIO_PRESENT
1202 /* Register mp action after probe() so that we got enough info */
1203 if (rte_vfio_is_enabled("vfio") && vfio_mp_sync_setup() < 0)
1204 return -1;
1205 #endif
1206
1207 /* initialize default service/lcore mappings and start running. Ignore
1208 * -ENOTSUP, as it indicates no service coremask passed to EAL.
1209 */
1210 ret = rte_service_start_with_defaults();
1211 if (ret < 0 && ret != -ENOTSUP) {
1212 rte_errno = ENOEXEC;
1213 return -1;
1214 }
1215
1216 /*
1217 * Clean up unused files in runtime directory. We do this at the end of
1218 * init and not at the beginning because we want to clean stuff up
1219 * whether we are primary or secondary process, but we cannot remove
1220 * primary process' files because secondary should be able to run even
1221 * if primary process is dead.
1222 *
1223 * In no_shconf mode, no runtime directory is created in the first
1224 * place, so no cleanup needed.
1225 */
1226 if (!internal_config.no_shconf && eal_clean_runtime_dir() < 0) {
1227 rte_eal_init_alert("Cannot clear runtime directory\n");
1228 return -1;
1229 }
1230
1231 rte_eal_mcfg_complete();
1232
1233 /* Call each registered callback, if enabled */
1234 rte_option_init();
1235
1236 return fctret;
1237 }
1238
1239 static int
1240 mark_freeable(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
1241 void *arg __rte_unused)
1242 {
1243 /* ms is const, so find this memseg */
1244 struct rte_memseg *found;
1245
1246 if (msl->external)
1247 return 0;
1248
1249 found = rte_mem_virt2memseg(ms->addr, msl);
1250
1251 found->flags &= ~RTE_MEMSEG_FLAG_DO_NOT_FREE;
1252
1253 return 0;
1254 }
1255
1256 int
1257 rte_eal_cleanup(void)
1258 {
1259 /* if we're in a primary process, we need to mark hugepages as freeable
1260 * so that finalization can release them back to the system.
1261 */
1262 if (rte_eal_process_type() == RTE_PROC_PRIMARY)
1263 rte_memseg_walk(mark_freeable, NULL);
1264 rte_service_finalize();
1265 rte_mp_channel_cleanup();
1266 eal_cleanup_config(&internal_config);
1267 return 0;
1268 }
1269
1270 /* get core role */
1271 enum rte_lcore_role_t
1272 rte_eal_lcore_role(unsigned lcore_id)
1273 {
1274 return rte_config.lcore_role[lcore_id];
1275 }
1276
1277 enum rte_proc_type_t
1278 rte_eal_process_type(void)
1279 {
1280 return rte_config.process_type;
1281 }
1282
1283 int rte_eal_has_hugepages(void)
1284 {
1285 return ! internal_config.no_hugetlbfs;
1286 }
1287
1288 int rte_eal_has_pci(void)
1289 {
1290 return !internal_config.no_pci;
1291 }
1292
1293 int rte_eal_create_uio_dev(void)
1294 {
1295 return internal_config.create_uio_dev;
1296 }
1297
1298 enum rte_intr_mode
1299 rte_eal_vfio_intr_mode(void)
1300 {
1301 return internal_config.vfio_intr_mode;
1302 }
1303
1304 int
1305 rte_eal_check_module(const char *module_name)
1306 {
1307 char sysfs_mod_name[PATH_MAX];
1308 struct stat st;
1309 int n;
1310
1311 if (NULL == module_name)
1312 return -1;
1313
1314 /* Check if there is sysfs mounted */
1315 if (stat("/sys/module", &st) != 0) {
1316 RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
1317 errno, strerror(errno));
1318 return -1;
1319 }
1320
1321 /* A module might be built-in, therefore try sysfs */
1322 n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
1323 if (n < 0 || n > PATH_MAX) {
1324 RTE_LOG(DEBUG, EAL, "Could not format module path\n");
1325 return -1;
1326 }
1327
1328 if (stat(sysfs_mod_name, &st) != 0) {
1329 RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
1330 sysfs_mod_name, errno, strerror(errno));
1331 return 0;
1332 }
1333
1334 /* Module has been found */
1335 return 1;
1336 }