ceph/src/dpdk/lib/librte_eal/bsdapp/eal/eal.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   Copyright(c) 2014 6WIND S.A.
   6  *   All rights reserved.
   7  *
   8  *   Redistribution and use in source and binary forms, with or without
   9  *   modification, are permitted provided that the following conditions
  10  *   are met:
  11  *
  12  *     * Redistributions of source code must retain the above copyright
  13  *       notice, this list of conditions and the following disclaimer.
  14  *     * Redistributions in binary form must reproduce the above copyright
  15  *       notice, this list of conditions and the following disclaimer in
  16  *       the documentation and/or other materials provided with the
  17  *       distribution.
  18  *     * Neither the name of Intel Corporation nor the names of its
  19  *       contributors may be used to endorse or promote products derived
  20  *       from this software without specific prior written permission.
  21  *
  22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <string.h>
  39 #include <stdarg.h>
  40 #include <unistd.h>
  41 #include <pthread.h>
  42 #include <syslog.h>
  43 #include <getopt.h>
  44 #include <sys/file.h>
  45 #include <stddef.h>
  46 #include <errno.h>
  47 #include <limits.h>
  48 #include <errno.h>
  49 #include <sys/mman.h>
  50 #include <sys/queue.h>
  51
  52 #include <rte_common.h>
  53 #include <rte_debug.h>
  54 #include <rte_memory.h>
  55 #include <rte_memzone.h>
  56 #include <rte_launch.h>
  57 #include <rte_eal.h>
  58 #include <rte_eal_memconfig.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_lcore.h>
  61 #include <rte_log.h>
  62 #include <rte_random.h>
  63 #include <rte_cycles.h>
  64 #include <rte_string_fns.h>
  65 #include <rte_cpuflags.h>
  66 #include <rte_interrupts.h>
  67 #include <rte_pci.h>
  68 #include <rte_dev.h>
  69 #include <rte_devargs.h>
  70 #include <rte_common.h>
  71 #include <rte_version.h>
  72 #include <rte_atomic.h>
  73 #include <malloc_heap.h>
  74
  75 #include "eal_private.h"
  76 #include "eal_thread.h"
  77 #include "eal_internal_cfg.h"
  78 #include "eal_filesystem.h"
  79 #include "eal_hugepages.h"
  80 #include "eal_options.h"
  81
  82 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
  83
  84 /* Allow the application to print its usage message too if set */
  85 static rte_usage_hook_t rte_application_usage_hook = NULL;
  86 /* early configuration structure, when memory config is not mmapped */
  87 static struct rte_mem_config early_mem_config;
  88
  89 /* define fd variable here, because file needs to be kept open for the
  90  * duration of the program, as we hold a write lock on it in the primary proc */
  91 static int mem_cfg_fd = -1;
  92
  93 static struct flock wr_lock = {
  94                 .l_type = F_WRLCK,
  95                 .l_whence = SEEK_SET,
  96                 .l_start = offsetof(struct rte_mem_config, memseg),
  97                 .l_len = sizeof(early_mem_config.memseg),
  98 };
  99
 100 /* Address of global and public configuration */
 101 static struct rte_config rte_config = {
 102                 .mem_config = &early_mem_config,
 103 };
 104
 105 /* internal configuration (per-core) */
 106 struct lcore_config lcore_config[RTE_MAX_LCORE];
 107
 108 /* internal configuration */
 109 struct internal_config internal_config;
 110
 111 /* used by rte_rdtsc() */
 112 int rte_cycles_vmware_tsc_map;
 113
 114 /* Return a pointer to the configuration structure */
 115 struct rte_config *
 116 rte_eal_get_configuration(void)
 117 {
 118         return &rte_config;
 119 }
 120
 121 /* parse a sysfs (or other) file containing one integer value */
 122 int
 123 eal_parse_sysfs_value(const char *filename, unsigned long *val)
 124 {
 125         FILE *f;
 126         char buf[BUFSIZ];
 127         char *end = NULL;
 128
 129         if ((f = fopen(filename, "r")) == NULL) {
 130                 RTE_LOG(ERR, EAL, "%s(): cannot open sysfs value %s\n",
 131                         __func__, filename);
 132                 return -1;
 133         }
 134
 135         if (fgets(buf, sizeof(buf), f) == NULL) {
 136                 RTE_LOG(ERR, EAL, "%s(): cannot read sysfs value %s\n",
 137                         __func__, filename);
 138                 fclose(f);
 139                 return -1;
 140         }
 141         *val = strtoul(buf, &end, 0);
 142         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 143                 RTE_LOG(ERR, EAL, "%s(): cannot parse sysfs value %s\n",
 144                                 __func__, filename);
 145                 fclose(f);
 146                 return -1;
 147         }
 148         fclose(f);
 149         return 0;
 150 }
 151
 152
 153 /* create memory configuration in shared/mmap memory. Take out
 154  * a write lock on the memsegs, so we can auto-detect primary/secondary.
 155  * This means we never close the file while running (auto-close on exit).
 156  * We also don't lock the whole file, so that in future we can use read-locks
 157  * on other parts, e.g. memzones, to detect if there are running secondary
 158  * processes. */
 159 static void
 160 rte_eal_config_create(void)
 161 {
 162         void *rte_mem_cfg_addr;
 163         int retval;
 164
 165         const char *pathname = eal_runtime_config_path();
 166
 167         if (internal_config.no_shconf)
 168                 return;
 169
 170         if (mem_cfg_fd < 0){
 171                 mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0660);
 172                 if (mem_cfg_fd < 0)
 173                         rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
 174         }
 175
 176         retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
 177         if (retval < 0){
 178                 close(mem_cfg_fd);
 179                 rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
 180         }
 181
 182         retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
 183         if (retval < 0){
 184                 close(mem_cfg_fd);
 185                 rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
 186                                 "process running?\n", pathname);
 187         }
 188
 189         rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
 190                                 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
 191
 192         if (rte_mem_cfg_addr == MAP_FAILED){
 193                 rte_panic("Cannot mmap memory for rte_config\n");
 194         }
 195         memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
 196         rte_config.mem_config = (struct rte_mem_config *) rte_mem_cfg_addr;
 197 }
 198
 199 /* attach to an existing shared memory config */
 200 static void
 201 rte_eal_config_attach(void)
 202 {
 203         void *rte_mem_cfg_addr;
 204         const char *pathname = eal_runtime_config_path();
 205
 206         if (internal_config.no_shconf)
 207                 return;
 208
 209         if (mem_cfg_fd < 0){
 210                 mem_cfg_fd = open(pathname, O_RDWR);
 211                 if (mem_cfg_fd < 0)
 212                         rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
 213         }
 214
 215         rte_mem_cfg_addr = mmap(NULL, sizeof(*rte_config.mem_config),
 216                                 PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
 217         close(mem_cfg_fd);
 218         if (rte_mem_cfg_addr == MAP_FAILED)
 219                 rte_panic("Cannot mmap memory for rte_config\n");
 220
 221         rte_config.mem_config = (struct rte_mem_config *) rte_mem_cfg_addr;
 222 }
 223
 224 /* Detect if we are a primary or a secondary process */
 225 enum rte_proc_type_t
 226 eal_proc_type_detect(void)
 227 {
 228         enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
 229         const char *pathname = eal_runtime_config_path();
 230
 231         /* if we can open the file but not get a write-lock we are a secondary
 232          * process. NOTE: if we get a file handle back, we keep that open
 233          * and don't close it to prevent a race condition between multiple opens */
 234         if (((mem_cfg_fd = open(pathname, O_RDWR)) >= 0) &&
 235                         (fcntl(mem_cfg_fd, F_SETLK, &wr_lock) < 0))
 236                 ptype = RTE_PROC_SECONDARY;
 237
 238         RTE_LOG(INFO, EAL, "Auto-detected process type: %s\n",
 239                         ptype == RTE_PROC_PRIMARY ? "PRIMARY" : "SECONDARY");
 240
 241         return ptype;
 242 }
 243
 244 /* Sets up rte_config structure with the pointer to shared memory config.*/
 245 static void
 246 rte_config_init(void)
 247 {
 248         rte_config.process_type = internal_config.process_type;
 249
 250         switch (rte_config.process_type){
 251         case RTE_PROC_PRIMARY:
 252                 rte_eal_config_create();
 253                 break;
 254         case RTE_PROC_SECONDARY:
 255                 rte_eal_config_attach();
 256                 rte_eal_mcfg_wait_complete(rte_config.mem_config);
 257                 break;
 258         case RTE_PROC_AUTO:
 259         case RTE_PROC_INVALID:
 260                 rte_panic("Invalid process type\n");
 261         }
 262 }
 263
 264 /* display usage */
 265 static void
 266 eal_usage(const char *prgname)
 267 {
 268         printf("\nUsage: %s ", prgname);
 269         eal_common_usage();
 270         /* Allow the application to print its usage message too if hook is set */
 271         if ( rte_application_usage_hook ) {
 272                 printf("===== Application Usage =====\n\n");
 273                 rte_application_usage_hook(prgname);
 274         }
 275 }
 276
 277 /* Set a per-application usage message */
 278 rte_usage_hook_t
 279 rte_set_application_usage_hook( rte_usage_hook_t usage_func )
 280 {
 281         rte_usage_hook_t        old_func;
 282
 283         /* Will be NULL on the first call to denote the last usage routine. */
 284         old_func                                        = rte_application_usage_hook;
 285         rte_application_usage_hook      = usage_func;
 286
 287         return old_func;
 288 }
 289
 290 static inline size_t
 291 eal_get_hugepage_mem_size(void)
 292 {
 293         uint64_t size = 0;
 294         unsigned i, j;
 295
 296         for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
 297                 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
 298                 if (hpi->hugedir != NULL) {
 299                         for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
 300                                 size += hpi->hugepage_sz * hpi->num_pages[j];
 301                         }
 302                 }
 303         }
 304
 305         return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
 306 }
 307
 308 /* Parse the arguments for --log-level only */
 309 static void
 310 eal_log_level_parse(int argc, char **argv)
 311 {
 312         int opt;
 313         char **argvopt;
 314         int option_index;
 315         const int old_optind = optind;
 316         const int old_optopt = optopt;
 317         const int old_optreset = optreset;
 318         char * const old_optarg = optarg;
 319
 320         argvopt = argv;
 321         optind = 1;
 322         optreset = 1;
 323
 324         eal_reset_internal_config(&internal_config);
 325
 326         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 327                                   eal_long_options, &option_index)) != EOF) {
 328
 329                 int ret;
 330
 331                 /* getopt is not happy, stop right now */
 332                 if (opt == '?')
 333                         break;
 334
 335                 ret = (opt == OPT_LOG_LEVEL_NUM) ?
 336                         eal_parse_common_option(opt, optarg, &internal_config) : 0;
 337
 338                 /* common parser is not happy */
 339                 if (ret < 0)
 340                         break;
 341         }
 342
 343         /* restore getopt lib */
 344         optind = old_optind;
 345         optopt = old_optopt;
 346         optreset = old_optreset;
 347         optarg = old_optarg;
 348 }
 349
 350 /* Parse the argument given in the command line of the application */
 351 static int
 352 eal_parse_args(int argc, char **argv)
 353 {
 354         int opt, ret;
 355         char **argvopt;
 356         int option_index;
 357         char *prgname = argv[0];
 358         const int old_optind = optind;
 359         const int old_optopt = optopt;
 360         const int old_optreset = optreset;
 361         char * const old_optarg = optarg;
 362
 363         argvopt = argv;
 364         optind = 1;
 365         optreset = 1;
 366
 367         while ((opt = getopt_long(argc, argvopt, eal_short_options,
 368                                   eal_long_options, &option_index)) != EOF) {
 369
 370                 /* getopt is not happy, stop right now */
 371                 if (opt == '?') {
 372                         eal_usage(prgname);
 373                         ret = -1;
 374                         goto out;
 375                 }
 376
 377                 ret = eal_parse_common_option(opt, optarg, &internal_config);
 378                 /* common parser is not happy */
 379                 if (ret < 0) {
 380                         eal_usage(prgname);
 381                         ret = -1;
 382                         goto out;
 383                 }
 384                 /* common parser handled this option */
 385                 if (ret == 0)
 386                         continue;
 387
 388                 switch (opt) {
 389                 case 'h':
 390                         eal_usage(prgname);
 391                         exit(EXIT_SUCCESS);
 392                 default:
 393                         if (opt < OPT_LONG_MIN_NUM && isprint(opt)) {
 394                                 RTE_LOG(ERR, EAL, "Option %c is not supported "
 395                                         "on FreeBSD\n", opt);
 396                         } else if (opt >= OPT_LONG_MIN_NUM &&
 397                                    opt < OPT_LONG_MAX_NUM) {
 398                                 RTE_LOG(ERR, EAL, "Option %s is not supported "
 399                                         "on FreeBSD\n",
 400                                         eal_long_options[option_index].name);
 401                         } else {
 402                                 RTE_LOG(ERR, EAL, "Option %d is not supported "
 403                                         "on FreeBSD\n", opt);
 404                         }
 405                         eal_usage(prgname);
 406                         ret = -1;
 407                         goto out;
 408                 }
 409         }
 410
 411         if (eal_adjust_config(&internal_config) != 0) {
 412                 ret = -1;
 413                 goto out;
 414         }
 415
 416         /* sanity checks */
 417         if (eal_check_common_options(&internal_config) != 0) {
 418                 eal_usage(prgname);
 419                 ret = -1;
 420                 goto out;
 421         }
 422
 423         if (optind >= 0)
 424                 argv[optind-1] = prgname;
 425         ret = optind-1;
 426
 427 out:
 428         /* restore getopt lib */
 429         optind = old_optind;
 430         optopt = old_optopt;
 431         optreset = old_optreset;
 432         optarg = old_optarg;
 433
 434         return ret;
 435 }
 436
 437 static void
 438 eal_check_mem_on_local_socket(void)
 439 {
 440         const struct rte_memseg *ms;
 441         int i, socket_id;
 442
 443         socket_id = rte_lcore_to_socket_id(rte_config.master_lcore);
 444
 445         ms = rte_eal_get_physmem_layout();
 446
 447         for (i = 0; i < RTE_MAX_MEMSEG; i++)
 448                 if (ms[i].socket_id == socket_id &&
 449                                 ms[i].len > 0)
 450                         return;
 451
 452         RTE_LOG(WARNING, EAL, "WARNING: Master core has no "
 453                         "memory on local socket!\n");
 454 }
 455
 456 static int
 457 sync_func(__attribute__((unused)) void *arg)
 458 {
 459         return 0;
 460 }
 461
 462 inline static void
 463 rte_eal_mcfg_complete(void)
 464 {
 465         /* ALL shared mem_config related INIT DONE */
 466         if (rte_config.process_type == RTE_PROC_PRIMARY)
 467                 rte_config.mem_config->magic = RTE_MAGIC;
 468 }
 469
 470 /* return non-zero if hugepages are enabled. */
 471 int rte_eal_has_hugepages(void)
 472 {
 473         return !internal_config.no_hugetlbfs;
 474 }
 475
 476 /* Abstraction for port I/0 privilege */
 477 int
 478 rte_eal_iopl_init(void)
 479 {
 480         static int fd;
 481
 482         fd = open("/dev/io", O_RDWR);
 483         if (fd < 0)
 484                 return -1;
 485         /* keep fd open for iopl */
 486         return 0;
 487 }
 488
 489 /* Launch threads, called at application init(). */
 490 int
 491 rte_eal_init(int argc, char **argv)
 492 {
 493         int i, fctret, ret;
 494         pthread_t thread_id;
 495         static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
 496         char cpuset[RTE_CPU_AFFINITY_STR_LEN];
 497         char thread_name[RTE_MAX_THREAD_NAME_LEN];
 498
 499         /* checks if the machine is adequate */
 500         rte_cpu_check_supported();
 501
 502         if (!rte_atomic32_test_and_set(&run_once))
 503                 return -1;
 504
 505         thread_id = pthread_self();
 506
 507         eal_log_level_parse(argc, argv);
 508
 509         /* set log level as early as possible */
 510         rte_set_log_level(internal_config.log_level);
 511
 512         if (rte_eal_cpu_init() < 0)
 513                 rte_panic("Cannot detect lcores\n");
 514
 515         fctret = eal_parse_args(argc, argv);
 516         if (fctret < 0)
 517                 exit(1);
 518
 519         if (internal_config.no_hugetlbfs == 0 &&
 520                         internal_config.process_type != RTE_PROC_SECONDARY &&
 521                         eal_hugepage_info_init() < 0)
 522                 rte_panic("Cannot get hugepage information\n");
 523
 524         if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
 525                 if (internal_config.no_hugetlbfs)
 526                         internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
 527                 else
 528                         internal_config.memory = eal_get_hugepage_mem_size();
 529         }
 530
 531         if (internal_config.vmware_tsc_map == 1) {
 532 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
 533                 rte_cycles_vmware_tsc_map = 1;
 534                 RTE_LOG (DEBUG, EAL, "Using VMWARE TSC MAP, "
 535                                 "you must have monitor_control.pseudo_perfctr = TRUE\n");
 536 #else
 537                 RTE_LOG (WARNING, EAL, "Ignoring --vmware-tsc-map because "
 538                                 "RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set\n");
 539 #endif
 540         }
 541
 542         rte_srand(rte_rdtsc());
 543
 544         rte_config_init();
 545
 546         if (rte_eal_memory_init() < 0)
 547                 rte_panic("Cannot init memory\n");
 548
 549         if (rte_eal_memzone_init() < 0)
 550                 rte_panic("Cannot init memzone\n");
 551
 552         if (rte_eal_tailqs_init() < 0)
 553                 rte_panic("Cannot init tail queues for objects\n");
 554
 555         if (rte_eal_alarm_init() < 0)
 556                 rte_panic("Cannot init interrupt-handling thread\n");
 557
 558         if (rte_eal_intr_init() < 0)
 559                 rte_panic("Cannot init interrupt-handling thread\n");
 560
 561         if (rte_eal_timer_init() < 0)
 562                 rte_panic("Cannot init HPET or TSC timers\n");
 563
 564         if (rte_eal_pci_init() < 0)
 565                 rte_panic("Cannot init PCI\n");
 566
 567         eal_check_mem_on_local_socket();
 568
 569         if (eal_plugins_init() < 0)
 570                 rte_panic("Cannot init plugins\n");
 571
 572         eal_thread_init_master(rte_config.master_lcore);
 573
 574         ret = eal_thread_dump_affinity(cpuset, RTE_CPU_AFFINITY_STR_LEN);
 575
 576         RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%p;cpuset=[%s%s])\n",
 577                 rte_config.master_lcore, thread_id, cpuset,
 578                 ret == 0 ? "" : "...");
 579
 580         if (rte_eal_dev_init() < 0)
 581                 rte_panic("Cannot init pmd devices\n");
 582
 583         RTE_LCORE_FOREACH_SLAVE(i) {
 584
 585                 /*
 586                  * create communication pipes between master thread
 587                  * and children
 588                  */
 589                 if (pipe(lcore_config[i].pipe_master2slave) < 0)
 590                         rte_panic("Cannot create pipe\n");
 591                 if (pipe(lcore_config[i].pipe_slave2master) < 0)
 592                         rte_panic("Cannot create pipe\n");
 593
 594                 lcore_config[i].state = WAIT;
 595
 596                 /* create a thread for each lcore */
 597                 ret = pthread_create(&lcore_config[i].thread_id, NULL,
 598                                      eal_thread_loop, NULL);
 599                 if (ret != 0)
 600                         rte_panic("Cannot create thread\n");
 601
 602                 /* Set thread_name for aid in debugging. */
 603                 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
 604                                 "lcore-slave-%d", i);
 605                 rte_thread_setname(lcore_config[i].thread_id, thread_name);
 606         }
 607
 608         /*
 609          * Launch a dummy function on all slave lcores, so that master lcore
 610          * knows they are all ready when this function returns.
 611          */
 612         rte_eal_mp_remote_launch(sync_func, NULL, SKIP_MASTER);
 613         rte_eal_mp_wait_lcore();
 614
 615         /* Probe & Initialize PCI devices */
 616         if (rte_eal_pci_probe())
 617                 rte_panic("Cannot probe PCI\n");
 618
 619         rte_eal_mcfg_complete();
 620
 621         return fctret;
 622 }
 623
 624 /* get core role */
 625 enum rte_lcore_role_t
 626 rte_eal_lcore_role(unsigned lcore_id)
 627 {
 628         return rte_config.lcore_role[lcore_id];
 629 }
 630
 631 enum rte_proc_type_t
 632 rte_eal_process_type(void)
 633 {
 634         return rte_config.process_type;
 635 }