Pileus Git - ~andy/linux/blob - kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/slab.h>
  19 #include <linux/delay.h>
  20
  21 #include <asm/setup.h>
  22
  23 #include "trace_output.h"
  24
  25 #undef TRACE_SYSTEM
  26 #define TRACE_SYSTEM "TRACE_SYSTEM"
  27
  28 DEFINE_MUTEX(event_mutex);
  29
  30 DEFINE_MUTEX(event_storage_mutex);
  31 EXPORT_SYMBOL_GPL(event_storage_mutex);
  32
  33 char event_storage[EVENT_STORAGE_SIZE];
  34 EXPORT_SYMBOL_GPL(event_storage);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_common_fields);
  38
  39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41 static struct kmem_cache *field_cachep;
  42 static struct kmem_cache *file_cachep;
  43
  44 #define SYSTEM_FL_FREE_NAME             (1 << 31)
  45
  46 static inline int system_refcount(struct event_subsystem *system)
  47 {
  48         return system->ref_count & ~SYSTEM_FL_FREE_NAME;
  49 }
  50
  51 static int system_refcount_inc(struct event_subsystem *system)
  52 {
  53         return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
  54 }
  55
  56 static int system_refcount_dec(struct event_subsystem *system)
  57 {
  58         return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
  59 }
  60
  61 /* Double loops, do not use break, only goto's work */
  62 #define do_for_each_event_file(tr, file)                        \
  63         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  64                 list_for_each_entry(file, &tr->events, list)
  65
  66 #define do_for_each_event_file_safe(tr, file)                   \
  67         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  68                 struct ftrace_event_file *___n;                         \
  69                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  70
  71 #define while_for_each_event_file()             \
  72         }
  73
  74 static struct list_head *
  75 trace_get_fields(struct ftrace_event_call *event_call)
  76 {
  77         if (!event_call->class->get_fields)
  78                 return &event_call->class->fields;
  79         return event_call->class->get_fields(event_call);
  80 }
  81
  82 static struct ftrace_event_field *
  83 __find_event_field(struct list_head *head, char *name)
  84 {
  85         struct ftrace_event_field *field;
  86
  87         list_for_each_entry(field, head, link) {
  88                 if (!strcmp(field->name, name))
  89                         return field;
  90         }
  91
  92         return NULL;
  93 }
  94
  95 struct ftrace_event_field *
  96 trace_find_event_field(struct ftrace_event_call *call, char *name)
  97 {
  98         struct ftrace_event_field *field;
  99         struct list_head *head;
 100
 101         field = __find_event_field(&ftrace_common_fields, name);
 102         if (field)
 103                 return field;
 104
 105         head = trace_get_fields(call);
 106         return __find_event_field(head, name);
 107 }
 108
 109 static int __trace_define_field(struct list_head *head, const char *type,
 110                                 const char *name, int offset, int size,
 111                                 int is_signed, int filter_type)
 112 {
 113         struct ftrace_event_field *field;
 114
 115         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 116         if (!field)
 117                 return -ENOMEM;
 118
 119         field->name = name;
 120         field->type = type;
 121
 122         if (filter_type == FILTER_OTHER)
 123                 field->filter_type = filter_assign_type(type);
 124         else
 125                 field->filter_type = filter_type;
 126
 127         field->offset = offset;
 128         field->size = size;
 129         field->is_signed = is_signed;
 130
 131         list_add(&field->link, head);
 132
 133         return 0;
 134 }
 135
 136 int trace_define_field(struct ftrace_event_call *call, const char *type,
 137                        const char *name, int offset, int size, int is_signed,
 138                        int filter_type)
 139 {
 140         struct list_head *head;
 141
 142         if (WARN_ON(!call->class))
 143                 return 0;
 144
 145         head = trace_get_fields(call);
 146         return __trace_define_field(head, type, name, offset, size,
 147                                     is_signed, filter_type);
 148 }
 149 EXPORT_SYMBOL_GPL(trace_define_field);
 150
 151 #define __common_field(type, item)                                      \
 152         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 153                                    "common_" #item,                     \
 154                                    offsetof(typeof(ent), item),         \
 155                                    sizeof(ent.item),                    \
 156                                    is_signed_type(type), FILTER_OTHER); \
 157         if (ret)                                                        \
 158                 return ret;
 159
 160 static int trace_define_common_fields(void)
 161 {
 162         int ret;
 163         struct trace_entry ent;
 164
 165         __common_field(unsigned short, type);
 166         __common_field(unsigned char, flags);
 167         __common_field(unsigned char, preempt_count);
 168         __common_field(int, pid);
 169
 170         return ret;
 171 }
 172
 173 static void trace_destroy_fields(struct ftrace_event_call *call)
 174 {
 175         struct ftrace_event_field *field, *next;
 176         struct list_head *head;
 177
 178         head = trace_get_fields(call);
 179         list_for_each_entry_safe(field, next, head, link) {
 180                 list_del(&field->link);
 181                 kmem_cache_free(field_cachep, field);
 182         }
 183 }
 184
 185 int trace_event_raw_init(struct ftrace_event_call *call)
 186 {
 187         int id;
 188
 189         id = register_ftrace_event(&call->event);
 190         if (!id)
 191                 return -ENODEV;
 192
 193         return 0;
 194 }
 195 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 196
 197 int ftrace_event_reg(struct ftrace_event_call *call,
 198                      enum trace_reg type, void *data)
 199 {
 200         struct ftrace_event_file *file = data;
 201
 202         switch (type) {
 203         case TRACE_REG_REGISTER:
 204                 return tracepoint_probe_register(call->name,
 205                                                  call->class->probe,
 206                                                  file);
 207         case TRACE_REG_UNREGISTER:
 208                 tracepoint_probe_unregister(call->name,
 209                                             call->class->probe,
 210                                             file);
 211                 return 0;
 212
 213 #ifdef CONFIG_PERF_EVENTS
 214         case TRACE_REG_PERF_REGISTER:
 215                 return tracepoint_probe_register(call->name,
 216                                                  call->class->perf_probe,
 217                                                  call);
 218         case TRACE_REG_PERF_UNREGISTER:
 219                 tracepoint_probe_unregister(call->name,
 220                                             call->class->perf_probe,
 221                                             call);
 222                 return 0;
 223         case TRACE_REG_PERF_OPEN:
 224         case TRACE_REG_PERF_CLOSE:
 225         case TRACE_REG_PERF_ADD:
 226         case TRACE_REG_PERF_DEL:
 227                 return 0;
 228 #endif
 229         }
 230         return 0;
 231 }
 232 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 233
 234 void trace_event_enable_cmd_record(bool enable)
 235 {
 236         struct ftrace_event_file *file;
 237         struct trace_array *tr;
 238
 239         mutex_lock(&event_mutex);
 240         do_for_each_event_file(tr, file) {
 241
 242                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 243                         continue;
 244
 245                 if (enable) {
 246                         tracing_start_cmdline_record();
 247                         set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 248                 } else {
 249                         tracing_stop_cmdline_record();
 250                         clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 251                 }
 252         } while_for_each_event_file();
 253         mutex_unlock(&event_mutex);
 254 }
 255
 256 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
 257                                          int enable, int soft_disable)
 258 {
 259         struct ftrace_event_call *call = file->event_call;
 260         int ret = 0;
 261         int disable;
 262
 263         switch (enable) {
 264         case 0:
 265                 /*
 266                  * When soft_disable is set and enable is cleared, the sm_ref
 267                  * reference counter is decremented. If it reaches 0, we want
 268                  * to clear the SOFT_DISABLED flag but leave the event in the
 269                  * state that it was. That is, if the event was enabled and
 270                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 271                  * is set we do not want the event to be enabled before we
 272                  * clear the bit.
 273                  *
 274                  * When soft_disable is not set but the SOFT_MODE flag is,
 275                  * we do nothing. Do not disable the tracepoint, otherwise
 276                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 277                  */
 278                 if (soft_disable) {
 279                         if (atomic_dec_return(&file->sm_ref) > 0)
 280                                 break;
 281                         disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
 282                         clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 283                 } else
 284                         disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
 285
 286                 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
 287                         clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 288                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 289                                 tracing_stop_cmdline_record();
 290                                 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 291                         }
 292                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 293                 }
 294                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 295                 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 296                         set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 297                 else
 298                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 299                 break;
 300         case 1:
 301                 /*
 302                  * When soft_disable is set and enable is set, we want to
 303                  * register the tracepoint for the event, but leave the event
 304                  * as is. That means, if the event was already enabled, we do
 305                  * nothing (but set SOFT_MODE). If the event is disabled, we
 306                  * set SOFT_DISABLED before enabling the event tracepoint, so
 307                  * it still seems to be disabled.
 308                  */
 309                 if (!soft_disable)
 310                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 311                 else {
 312                         if (atomic_inc_return(&file->sm_ref) > 1)
 313                                 break;
 314                         set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 315                 }
 316
 317                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 318
 319                         /* Keep the event disabled, when going to SOFT_MODE. */
 320                         if (soft_disable)
 321                                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 322
 323                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 324                                 tracing_start_cmdline_record();
 325                                 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 326                         }
 327                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 328                         if (ret) {
 329                                 tracing_stop_cmdline_record();
 330                                 pr_info("event trace: Could not enable event "
 331                                         "%s\n", call->name);
 332                                 break;
 333                         }
 334                         set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 335
 336                         /* WAS_ENABLED gets set but never cleared. */
 337                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 338                 }
 339                 break;
 340         }
 341
 342         return ret;
 343 }
 344
 345 int trace_event_enable_disable(struct ftrace_event_file *file,
 346                                int enable, int soft_disable)
 347 {
 348         return __ftrace_event_enable_disable(file, enable, soft_disable);
 349 }
 350
 351 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 352                                        int enable)
 353 {
 354         return __ftrace_event_enable_disable(file, enable, 0);
 355 }
 356
 357 static void ftrace_clear_events(struct trace_array *tr)
 358 {
 359         struct ftrace_event_file *file;
 360
 361         mutex_lock(&event_mutex);
 362         list_for_each_entry(file, &tr->events, list) {
 363                 ftrace_event_enable_disable(file, 0);
 364         }
 365         mutex_unlock(&event_mutex);
 366 }
 367
 368 static void __put_system(struct event_subsystem *system)
 369 {
 370         struct event_filter *filter = system->filter;
 371
 372         WARN_ON_ONCE(system_refcount(system) == 0);
 373         if (system_refcount_dec(system))
 374                 return;
 375
 376         list_del(&system->list);
 377
 378         if (filter) {
 379                 kfree(filter->filter_string);
 380                 kfree(filter);
 381         }
 382         if (system->ref_count & SYSTEM_FL_FREE_NAME)
 383                 kfree(system->name);
 384         kfree(system);
 385 }
 386
 387 static void __get_system(struct event_subsystem *system)
 388 {
 389         WARN_ON_ONCE(system_refcount(system) == 0);
 390         system_refcount_inc(system);
 391 }
 392
 393 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 394 {
 395         WARN_ON_ONCE(dir->ref_count == 0);
 396         dir->ref_count++;
 397         __get_system(dir->subsystem);
 398 }
 399
 400 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 401 {
 402         WARN_ON_ONCE(dir->ref_count == 0);
 403         /* If the subsystem is about to be freed, the dir must be too */
 404         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 405
 406         __put_system(dir->subsystem);
 407         if (!--dir->ref_count)
 408                 kfree(dir);
 409 }
 410
 411 static void put_system(struct ftrace_subsystem_dir *dir)
 412 {
 413         mutex_lock(&event_mutex);
 414         __put_system_dir(dir);
 415         mutex_unlock(&event_mutex);
 416 }
 417
 418 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
 419 {
 420         if (!dir)
 421                 return;
 422
 423         if (!--dir->nr_events) {
 424                 debugfs_remove_recursive(dir->entry);
 425                 list_del(&dir->list);
 426                 __put_system_dir(dir);
 427         }
 428 }
 429
 430 static void remove_event_file_dir(struct ftrace_event_file *file)
 431 {
 432         struct dentry *dir = file->dir;
 433         struct dentry *child;
 434
 435         if (dir) {
 436                 spin_lock(&dir->d_lock);        /* probably unneeded */
 437                 list_for_each_entry(child, &dir->d_subdirs, d_u.d_child) {
 438                         if (child->d_inode)     /* probably unneeded */
 439                                 child->d_inode->i_private = NULL;
 440                 }
 441                 spin_unlock(&dir->d_lock);
 442
 443                 debugfs_remove_recursive(dir);
 444         }
 445
 446         list_del(&file->list);
 447         remove_subsystem(file->system);
 448         kmem_cache_free(file_cachep, file);
 449 }
 450
 451 /*
 452  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 453  */
 454 static int
 455 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 456                               const char *sub, const char *event, int set)
 457 {
 458         struct ftrace_event_file *file;
 459         struct ftrace_event_call *call;
 460         int ret = -EINVAL;
 461
 462         list_for_each_entry(file, &tr->events, list) {
 463
 464                 call = file->event_call;
 465
 466                 if (!call->name || !call->class || !call->class->reg)
 467                         continue;
 468
 469                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 470                         continue;
 471
 472                 if (match &&
 473                     strcmp(match, call->name) != 0 &&
 474                     strcmp(match, call->class->system) != 0)
 475                         continue;
 476
 477                 if (sub && strcmp(sub, call->class->system) != 0)
 478                         continue;
 479
 480                 if (event && strcmp(event, call->name) != 0)
 481                         continue;
 482
 483                 ftrace_event_enable_disable(file, set);
 484
 485                 ret = 0;
 486         }
 487
 488         return ret;
 489 }
 490
 491 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 492                                   const char *sub, const char *event, int set)
 493 {
 494         int ret;
 495
 496         mutex_lock(&event_mutex);
 497         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 498         mutex_unlock(&event_mutex);
 499
 500         return ret;
 501 }
 502
 503 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 504 {
 505         char *event = NULL, *sub = NULL, *match;
 506
 507         /*
 508          * The buf format can be <subsystem>:<event-name>
 509          *  *:<event-name> means any event by that name.
 510          *  :<event-name> is the same.
 511          *
 512          *  <subsystem>:* means all events in that subsystem
 513          *  <subsystem>: means the same.
 514          *
 515          *  <name> (no ':') means all events in a subsystem with
 516          *  the name <name> or any event that matches <name>
 517          */
 518
 519         match = strsep(&buf, ":");
 520         if (buf) {
 521                 sub = match;
 522                 event = buf;
 523                 match = NULL;
 524
 525                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 526                         sub = NULL;
 527                 if (!strlen(event) || strcmp(event, "*") == 0)
 528                         event = NULL;
 529         }
 530
 531         return __ftrace_set_clr_event(tr, match, sub, event, set);
 532 }
 533
 534 /**
 535  * trace_set_clr_event - enable or disable an event
 536  * @system: system name to match (NULL for any system)
 537  * @event: event name to match (NULL for all events, within system)
 538  * @set: 1 to enable, 0 to disable
 539  *
 540  * This is a way for other parts of the kernel to enable or disable
 541  * event recording.
 542  *
 543  * Returns 0 on success, -EINVAL if the parameters do not match any
 544  * registered events.
 545  */
 546 int trace_set_clr_event(const char *system, const char *event, int set)
 547 {
 548         struct trace_array *tr = top_trace_array();
 549
 550         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 551 }
 552 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 553
 554 /* 128 should be much more than enough */
 555 #define EVENT_BUF_SIZE          127
 556
 557 static ssize_t
 558 ftrace_event_write(struct file *file, const char __user *ubuf,
 559                    size_t cnt, loff_t *ppos)
 560 {
 561         struct trace_parser parser;
 562         struct seq_file *m = file->private_data;
 563         struct trace_array *tr = m->private;
 564         ssize_t read, ret;
 565
 566         if (!cnt)
 567                 return 0;
 568
 569         ret = tracing_update_buffers();
 570         if (ret < 0)
 571                 return ret;
 572
 573         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 574                 return -ENOMEM;
 575
 576         read = trace_get_user(&parser, ubuf, cnt, ppos);
 577
 578         if (read >= 0 && trace_parser_loaded((&parser))) {
 579                 int set = 1;
 580
 581                 if (*parser.buffer == '!')
 582                         set = 0;
 583
 584                 parser.buffer[parser.idx] = 0;
 585
 586                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 587                 if (ret)
 588                         goto out_put;
 589         }
 590
 591         ret = read;
 592
 593  out_put:
 594         trace_parser_put(&parser);
 595
 596         return ret;
 597 }
 598
 599 static void *
 600 t_next(struct seq_file *m, void *v, loff_t *pos)
 601 {
 602         struct ftrace_event_file *file = v;
 603         struct ftrace_event_call *call;
 604         struct trace_array *tr = m->private;
 605
 606         (*pos)++;
 607
 608         list_for_each_entry_continue(file, &tr->events, list) {
 609                 call = file->event_call;
 610                 /*
 611                  * The ftrace subsystem is for showing formats only.
 612                  * They can not be enabled or disabled via the event files.
 613                  */
 614                 if (call->class && call->class->reg)
 615                         return file;
 616         }
 617
 618         return NULL;
 619 }
 620
 621 static void *t_start(struct seq_file *m, loff_t *pos)
 622 {
 623         struct ftrace_event_file *file;
 624         struct trace_array *tr = m->private;
 625         loff_t l;
 626
 627         mutex_lock(&event_mutex);
 628
 629         file = list_entry(&tr->events, struct ftrace_event_file, list);
 630         for (l = 0; l <= *pos; ) {
 631                 file = t_next(m, file, &l);
 632                 if (!file)
 633                         break;
 634         }
 635         return file;
 636 }
 637
 638 static void *
 639 s_next(struct seq_file *m, void *v, loff_t *pos)
 640 {
 641         struct ftrace_event_file *file = v;
 642         struct trace_array *tr = m->private;
 643
 644         (*pos)++;
 645
 646         list_for_each_entry_continue(file, &tr->events, list) {
 647                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
 648                         return file;
 649         }
 650
 651         return NULL;
 652 }
 653
 654 static void *s_start(struct seq_file *m, loff_t *pos)
 655 {
 656         struct ftrace_event_file *file;
 657         struct trace_array *tr = m->private;
 658         loff_t l;
 659
 660         mutex_lock(&event_mutex);
 661
 662         file = list_entry(&tr->events, struct ftrace_event_file, list);
 663         for (l = 0; l <= *pos; ) {
 664                 file = s_next(m, file, &l);
 665                 if (!file)
 666                         break;
 667         }
 668         return file;
 669 }
 670
 671 static int t_show(struct seq_file *m, void *v)
 672 {
 673         struct ftrace_event_file *file = v;
 674         struct ftrace_event_call *call = file->event_call;
 675
 676         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 677                 seq_printf(m, "%s:", call->class->system);
 678         seq_printf(m, "%s\n", call->name);
 679
 680         return 0;
 681 }
 682
 683 static void t_stop(struct seq_file *m, void *p)
 684 {
 685         mutex_unlock(&event_mutex);
 686 }
 687
 688 static ssize_t
 689 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 690                   loff_t *ppos)
 691 {
 692         struct ftrace_event_file *file;
 693         unsigned long flags;
 694         char buf[4] = "0";
 695
 696         mutex_lock(&event_mutex);
 697         file = event_file_data(filp);
 698         if (likely(file))
 699                 flags = file->flags;
 700         mutex_unlock(&event_mutex);
 701
 702         if (!file)
 703                 return -ENODEV;
 704
 705         if (flags & FTRACE_EVENT_FL_ENABLED &&
 706             !(flags & FTRACE_EVENT_FL_SOFT_DISABLED))
 707                 strcpy(buf, "1");
 708
 709         if (flags & FTRACE_EVENT_FL_SOFT_DISABLED ||
 710             flags & FTRACE_EVENT_FL_SOFT_MODE)
 711                 strcat(buf, "*");
 712
 713         strcat(buf, "\n");
 714
 715         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
 716 }
 717
 718 static ssize_t
 719 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 720                    loff_t *ppos)
 721 {
 722         struct ftrace_event_file *file;
 723         unsigned long val;
 724         int ret;
 725
 726         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 727         if (ret)
 728                 return ret;
 729
 730         ret = tracing_update_buffers();
 731         if (ret < 0)
 732                 return ret;
 733
 734         switch (val) {
 735         case 0:
 736         case 1:
 737                 ret = -ENODEV;
 738                 mutex_lock(&event_mutex);
 739                 file = event_file_data(filp);
 740                 if (likely(file))
 741                         ret = ftrace_event_enable_disable(file, val);
 742                 mutex_unlock(&event_mutex);
 743                 break;
 744
 745         default:
 746                 return -EINVAL;
 747         }
 748
 749         *ppos += cnt;
 750
 751         return ret ? ret : cnt;
 752 }
 753
 754 static ssize_t
 755 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 756                    loff_t *ppos)
 757 {
 758         const char set_to_char[4] = { '?', '0', '1', 'X' };
 759         struct ftrace_subsystem_dir *dir = filp->private_data;
 760         struct event_subsystem *system = dir->subsystem;
 761         struct ftrace_event_call *call;
 762         struct ftrace_event_file *file;
 763         struct trace_array *tr = dir->tr;
 764         char buf[2];
 765         int set = 0;
 766         int ret;
 767
 768         mutex_lock(&event_mutex);
 769         list_for_each_entry(file, &tr->events, list) {
 770                 call = file->event_call;
 771                 if (!call->name || !call->class || !call->class->reg)
 772                         continue;
 773
 774                 if (system && strcmp(call->class->system, system->name) != 0)
 775                         continue;
 776
 777                 /*
 778                  * We need to find out if all the events are set
 779                  * or if all events or cleared, or if we have
 780                  * a mixture.
 781                  */
 782                 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
 783
 784                 /*
 785                  * If we have a mixture, no need to look further.
 786                  */
 787                 if (set == 3)
 788                         break;
 789         }
 790         mutex_unlock(&event_mutex);
 791
 792         buf[0] = set_to_char[set];
 793         buf[1] = '\n';
 794
 795         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 796
 797         return ret;
 798 }
 799
 800 static ssize_t
 801 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 802                     loff_t *ppos)
 803 {
 804         struct ftrace_subsystem_dir *dir = filp->private_data;
 805         struct event_subsystem *system = dir->subsystem;
 806         const char *name = NULL;
 807         unsigned long val;
 808         ssize_t ret;
 809
 810         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 811         if (ret)
 812                 return ret;
 813
 814         ret = tracing_update_buffers();
 815         if (ret < 0)
 816                 return ret;
 817
 818         if (val != 0 && val != 1)
 819                 return -EINVAL;
 820
 821         /*
 822          * Opening of "enable" adds a ref count to system,
 823          * so the name is safe to use.
 824          */
 825         if (system)
 826                 name = system->name;
 827
 828         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
 829         if (ret)
 830                 goto out;
 831
 832         ret = cnt;
 833
 834 out:
 835         *ppos += cnt;
 836
 837         return ret;
 838 }
 839
 840 enum {
 841         FORMAT_HEADER           = 1,
 842         FORMAT_FIELD_SEPERATOR  = 2,
 843         FORMAT_PRINTFMT         = 3,
 844 };
 845
 846 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 847 {
 848         struct ftrace_event_call *call = event_file_data(m->private);
 849         struct list_head *common_head = &ftrace_common_fields;
 850         struct list_head *head = trace_get_fields(call);
 851         struct list_head *node = v;
 852
 853         (*pos)++;
 854
 855         switch ((unsigned long)v) {
 856         case FORMAT_HEADER:
 857                 node = common_head;
 858                 break;
 859
 860         case FORMAT_FIELD_SEPERATOR:
 861                 node = head;
 862                 break;
 863
 864         case FORMAT_PRINTFMT:
 865                 /* all done */
 866                 return NULL;
 867         }
 868
 869         node = node->prev;
 870         if (node == common_head)
 871                 return (void *)FORMAT_FIELD_SEPERATOR;
 872         else if (node == head)
 873                 return (void *)FORMAT_PRINTFMT;
 874         else
 875                 return node;
 876 }
 877
 878 static int f_show(struct seq_file *m, void *v)
 879 {
 880         struct ftrace_event_call *call = event_file_data(m->private);
 881         struct ftrace_event_field *field;
 882         const char *array_descriptor;
 883
 884         switch ((unsigned long)v) {
 885         case FORMAT_HEADER:
 886                 seq_printf(m, "name: %s\n", call->name);
 887                 seq_printf(m, "ID: %d\n", call->event.type);
 888                 seq_printf(m, "format:\n");
 889                 return 0;
 890
 891         case FORMAT_FIELD_SEPERATOR:
 892                 seq_putc(m, '\n');
 893                 return 0;
 894
 895         case FORMAT_PRINTFMT:
 896                 seq_printf(m, "\nprint fmt: %s\n",
 897                            call->print_fmt);
 898                 return 0;
 899         }
 900
 901         field = list_entry(v, struct ftrace_event_field, link);
 902         /*
 903          * Smartly shows the array type(except dynamic array).
 904          * Normal:
 905          *      field:TYPE VAR
 906          * If TYPE := TYPE[LEN], it is shown:
 907          *      field:TYPE VAR[LEN]
 908          */
 909         array_descriptor = strchr(field->type, '[');
 910
 911         if (!strncmp(field->type, "__data_loc", 10))
 912                 array_descriptor = NULL;
 913
 914         if (!array_descriptor)
 915                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 916                            field->type, field->name, field->offset,
 917                            field->size, !!field->is_signed);
 918         else
 919                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 920                            (int)(array_descriptor - field->type),
 921                            field->type, field->name,
 922                            array_descriptor, field->offset,
 923                            field->size, !!field->is_signed);
 924
 925         return 0;
 926 }
 927
 928 static void *f_start(struct seq_file *m, loff_t *pos)
 929 {
 930         void *p = (void *)FORMAT_HEADER;
 931         loff_t l = 0;
 932
 933         /* ->stop() is called even if ->start() fails */
 934         mutex_lock(&event_mutex);
 935         if (!event_file_data(m->private))
 936                 return ERR_PTR(-ENODEV);
 937
 938         while (l < *pos && p)
 939                 p = f_next(m, p, &l);
 940
 941         return p;
 942 }
 943
 944 static void f_stop(struct seq_file *m, void *p)
 945 {
 946         mutex_unlock(&event_mutex);
 947 }
 948
 949 static const struct seq_operations trace_format_seq_ops = {
 950         .start          = f_start,
 951         .next           = f_next,
 952         .stop           = f_stop,
 953         .show           = f_show,
 954 };
 955
 956 static int trace_format_open(struct inode *inode, struct file *file)
 957 {
 958         struct seq_file *m;
 959         int ret;
 960
 961         ret = seq_open(file, &trace_format_seq_ops);
 962         if (ret < 0)
 963                 return ret;
 964
 965         m = file->private_data;
 966         m->private = file;
 967
 968         return 0;
 969 }
 970
 971 static ssize_t
 972 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 973 {
 974         int id = (long)event_file_data(filp);
 975         char buf[32];
 976         int len;
 977
 978         if (*ppos)
 979                 return 0;
 980
 981         if (unlikely(!id))
 982                 return -ENODEV;
 983
 984         len = sprintf(buf, "%d\n", id);
 985
 986         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
 987 }
 988
 989 static ssize_t
 990 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 991                   loff_t *ppos)
 992 {
 993         struct ftrace_event_file *file;
 994         struct trace_seq *s;
 995         int r = -ENODEV;
 996
 997         if (*ppos)
 998                 return 0;
 999
1000         s = kmalloc(sizeof(*s), GFP_KERNEL);
1001
1002         if (!s)
1003                 return -ENOMEM;
1004
1005         trace_seq_init(s);
1006
1007         mutex_lock(&event_mutex);
1008         file = event_file_data(filp);
1009         if (file)
1010                 print_event_filter(file, s);
1011         mutex_unlock(&event_mutex);
1012
1013         if (file)
1014                 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1015
1016         kfree(s);
1017
1018         return r;
1019 }
1020
1021 static ssize_t
1022 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1023                    loff_t *ppos)
1024 {
1025         struct ftrace_event_file *file;
1026         char *buf;
1027         int err = -ENODEV;
1028
1029         if (cnt >= PAGE_SIZE)
1030                 return -EINVAL;
1031
1032         buf = (char *)__get_free_page(GFP_TEMPORARY);
1033         if (!buf)
1034                 return -ENOMEM;
1035
1036         if (copy_from_user(buf, ubuf, cnt)) {
1037                 free_page((unsigned long) buf);
1038                 return -EFAULT;
1039         }
1040         buf[cnt] = '\0';
1041
1042         mutex_lock(&event_mutex);
1043         file = event_file_data(filp);
1044         if (file)
1045                 err = apply_event_filter(file, buf);
1046         mutex_unlock(&event_mutex);
1047
1048         free_page((unsigned long) buf);
1049         if (err < 0)
1050                 return err;
1051
1052         *ppos += cnt;
1053
1054         return cnt;
1055 }
1056
1057 static LIST_HEAD(event_subsystems);
1058
1059 static int subsystem_open(struct inode *inode, struct file *filp)
1060 {
1061         struct event_subsystem *system = NULL;
1062         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1063         struct trace_array *tr;
1064         int ret;
1065
1066         if (tracing_is_disabled())
1067                 return -ENODEV;
1068
1069         /* Make sure the system still exists */
1070         mutex_lock(&trace_types_lock);
1071         mutex_lock(&event_mutex);
1072         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1073                 list_for_each_entry(dir, &tr->systems, list) {
1074                         if (dir == inode->i_private) {
1075                                 /* Don't open systems with no events */
1076                                 if (dir->nr_events) {
1077                                         __get_system_dir(dir);
1078                                         system = dir->subsystem;
1079                                 }
1080                                 goto exit_loop;
1081                         }
1082                 }
1083         }
1084  exit_loop:
1085         mutex_unlock(&event_mutex);
1086         mutex_unlock(&trace_types_lock);
1087
1088         if (!system)
1089                 return -ENODEV;
1090
1091         /* Some versions of gcc think dir can be uninitialized here */
1092         WARN_ON(!dir);
1093
1094         /* Still need to increment the ref count of the system */
1095         if (trace_array_get(tr) < 0) {
1096                 put_system(dir);
1097                 return -ENODEV;
1098         }
1099
1100         ret = tracing_open_generic(inode, filp);
1101         if (ret < 0) {
1102                 trace_array_put(tr);
1103                 put_system(dir);
1104         }
1105
1106         return ret;
1107 }
1108
1109 static int system_tr_open(struct inode *inode, struct file *filp)
1110 {
1111         struct ftrace_subsystem_dir *dir;
1112         struct trace_array *tr = inode->i_private;
1113         int ret;
1114
1115         if (tracing_is_disabled())
1116                 return -ENODEV;
1117
1118         if (trace_array_get(tr) < 0)
1119                 return -ENODEV;
1120
1121         /* Make a temporary dir that has no system but points to tr */
1122         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1123         if (!dir) {
1124                 trace_array_put(tr);
1125                 return -ENOMEM;
1126         }
1127
1128         dir->tr = tr;
1129
1130         ret = tracing_open_generic(inode, filp);
1131         if (ret < 0) {
1132                 trace_array_put(tr);
1133                 kfree(dir);
1134                 return ret;
1135         }
1136
1137         filp->private_data = dir;
1138
1139         return 0;
1140 }
1141
1142 static int subsystem_release(struct inode *inode, struct file *file)
1143 {
1144         struct ftrace_subsystem_dir *dir = file->private_data;
1145
1146         trace_array_put(dir->tr);
1147
1148         /*
1149          * If dir->subsystem is NULL, then this is a temporary
1150          * descriptor that was made for a trace_array to enable
1151          * all subsystems.
1152          */
1153         if (dir->subsystem)
1154                 put_system(dir);
1155         else
1156                 kfree(dir);
1157
1158         return 0;
1159 }
1160
1161 static ssize_t
1162 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1163                       loff_t *ppos)
1164 {
1165         struct ftrace_subsystem_dir *dir = filp->private_data;
1166         struct event_subsystem *system = dir->subsystem;
1167         struct trace_seq *s;
1168         int r;
1169
1170         if (*ppos)
1171                 return 0;
1172
1173         s = kmalloc(sizeof(*s), GFP_KERNEL);
1174         if (!s)
1175                 return -ENOMEM;
1176
1177         trace_seq_init(s);
1178
1179         print_subsystem_event_filter(system, s);
1180         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1181
1182         kfree(s);
1183
1184         return r;
1185 }
1186
1187 static ssize_t
1188 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1189                        loff_t *ppos)
1190 {
1191         struct ftrace_subsystem_dir *dir = filp->private_data;
1192         char *buf;
1193         int err;
1194
1195         if (cnt >= PAGE_SIZE)
1196                 return -EINVAL;
1197
1198         buf = (char *)__get_free_page(GFP_TEMPORARY);
1199         if (!buf)
1200                 return -ENOMEM;
1201
1202         if (copy_from_user(buf, ubuf, cnt)) {
1203                 free_page((unsigned long) buf);
1204                 return -EFAULT;
1205         }
1206         buf[cnt] = '\0';
1207
1208         err = apply_subsystem_event_filter(dir, buf);
1209         free_page((unsigned long) buf);
1210         if (err < 0)
1211                 return err;
1212
1213         *ppos += cnt;
1214
1215         return cnt;
1216 }
1217
1218 static ssize_t
1219 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1220 {
1221         int (*func)(struct trace_seq *s) = filp->private_data;
1222         struct trace_seq *s;
1223         int r;
1224
1225         if (*ppos)
1226                 return 0;
1227
1228         s = kmalloc(sizeof(*s), GFP_KERNEL);
1229         if (!s)
1230                 return -ENOMEM;
1231
1232         trace_seq_init(s);
1233
1234         func(s);
1235         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1236
1237         kfree(s);
1238
1239         return r;
1240 }
1241
1242 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1243 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1244 static int ftrace_event_release(struct inode *inode, struct file *file);
1245
1246 static const struct seq_operations show_event_seq_ops = {
1247         .start = t_start,
1248         .next = t_next,
1249         .show = t_show,
1250         .stop = t_stop,
1251 };
1252
1253 static const struct seq_operations show_set_event_seq_ops = {
1254         .start = s_start,
1255         .next = s_next,
1256         .show = t_show,
1257         .stop = t_stop,
1258 };
1259
1260 static const struct file_operations ftrace_avail_fops = {
1261         .open = ftrace_event_avail_open,
1262         .read = seq_read,
1263         .llseek = seq_lseek,
1264         .release = seq_release,
1265 };
1266
1267 static const struct file_operations ftrace_set_event_fops = {
1268         .open = ftrace_event_set_open,
1269         .read = seq_read,
1270         .write = ftrace_event_write,
1271         .llseek = seq_lseek,
1272         .release = ftrace_event_release,
1273 };
1274
1275 static const struct file_operations ftrace_enable_fops = {
1276         .open = tracing_open_generic,
1277         .read = event_enable_read,
1278         .write = event_enable_write,
1279         .llseek = default_llseek,
1280 };
1281
1282 static const struct file_operations ftrace_event_format_fops = {
1283         .open = trace_format_open,
1284         .read = seq_read,
1285         .llseek = seq_lseek,
1286         .release = seq_release,
1287 };
1288
1289 static const struct file_operations ftrace_event_id_fops = {
1290         .read = event_id_read,
1291         .llseek = default_llseek,
1292 };
1293
1294 static const struct file_operations ftrace_event_filter_fops = {
1295         .open = tracing_open_generic,
1296         .read = event_filter_read,
1297         .write = event_filter_write,
1298         .llseek = default_llseek,
1299 };
1300
1301 static const struct file_operations ftrace_subsystem_filter_fops = {
1302         .open = subsystem_open,
1303         .read = subsystem_filter_read,
1304         .write = subsystem_filter_write,
1305         .llseek = default_llseek,
1306         .release = subsystem_release,
1307 };
1308
1309 static const struct file_operations ftrace_system_enable_fops = {
1310         .open = subsystem_open,
1311         .read = system_enable_read,
1312         .write = system_enable_write,
1313         .llseek = default_llseek,
1314         .release = subsystem_release,
1315 };
1316
1317 static const struct file_operations ftrace_tr_enable_fops = {
1318         .open = system_tr_open,
1319         .read = system_enable_read,
1320         .write = system_enable_write,
1321         .llseek = default_llseek,
1322         .release = subsystem_release,
1323 };
1324
1325 static const struct file_operations ftrace_show_header_fops = {
1326         .open = tracing_open_generic,
1327         .read = show_header,
1328         .llseek = default_llseek,
1329 };
1330
1331 static int
1332 ftrace_event_open(struct inode *inode, struct file *file,
1333                   const struct seq_operations *seq_ops)
1334 {
1335         struct seq_file *m;
1336         int ret;
1337
1338         ret = seq_open(file, seq_ops);
1339         if (ret < 0)
1340                 return ret;
1341         m = file->private_data;
1342         /* copy tr over to seq ops */
1343         m->private = inode->i_private;
1344
1345         return ret;
1346 }
1347
1348 static int ftrace_event_release(struct inode *inode, struct file *file)
1349 {
1350         struct trace_array *tr = inode->i_private;
1351
1352         trace_array_put(tr);
1353
1354         return seq_release(inode, file);
1355 }
1356
1357 static int
1358 ftrace_event_avail_open(struct inode *inode, struct file *file)
1359 {
1360         const struct seq_operations *seq_ops = &show_event_seq_ops;
1361
1362         return ftrace_event_open(inode, file, seq_ops);
1363 }
1364
1365 static int
1366 ftrace_event_set_open(struct inode *inode, struct file *file)
1367 {
1368         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1369         struct trace_array *tr = inode->i_private;
1370         int ret;
1371
1372         if (trace_array_get(tr) < 0)
1373                 return -ENODEV;
1374
1375         if ((file->f_mode & FMODE_WRITE) &&
1376             (file->f_flags & O_TRUNC))
1377                 ftrace_clear_events(tr);
1378
1379         ret = ftrace_event_open(inode, file, seq_ops);
1380         if (ret < 0)
1381                 trace_array_put(tr);
1382         return ret;
1383 }
1384
1385 static struct event_subsystem *
1386 create_new_subsystem(const char *name)
1387 {
1388         struct event_subsystem *system;
1389
1390         /* need to create new entry */
1391         system = kmalloc(sizeof(*system), GFP_KERNEL);
1392         if (!system)
1393                 return NULL;
1394
1395         system->ref_count = 1;
1396
1397         /* Only allocate if dynamic (kprobes and modules) */
1398         if (!core_kernel_data((unsigned long)name)) {
1399                 system->ref_count |= SYSTEM_FL_FREE_NAME;
1400                 system->name = kstrdup(name, GFP_KERNEL);
1401                 if (!system->name)
1402                         goto out_free;
1403         } else
1404                 system->name = name;
1405
1406         system->filter = NULL;
1407
1408         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1409         if (!system->filter)
1410                 goto out_free;
1411
1412         list_add(&system->list, &event_subsystems);
1413
1414         return system;
1415
1416  out_free:
1417         if (system->ref_count & SYSTEM_FL_FREE_NAME)
1418                 kfree(system->name);
1419         kfree(system);
1420         return NULL;
1421 }
1422
1423 static struct dentry *
1424 event_subsystem_dir(struct trace_array *tr, const char *name,
1425                     struct ftrace_event_file *file, struct dentry *parent)
1426 {
1427         struct ftrace_subsystem_dir *dir;
1428         struct event_subsystem *system;
1429         struct dentry *entry;
1430
1431         /* First see if we did not already create this dir */
1432         list_for_each_entry(dir, &tr->systems, list) {
1433                 system = dir->subsystem;
1434                 if (strcmp(system->name, name) == 0) {
1435                         dir->nr_events++;
1436                         file->system = dir;
1437                         return dir->entry;
1438                 }
1439         }
1440
1441         /* Now see if the system itself exists. */
1442         list_for_each_entry(system, &event_subsystems, list) {
1443                 if (strcmp(system->name, name) == 0)
1444                         break;
1445         }
1446         /* Reset system variable when not found */
1447         if (&system->list == &event_subsystems)
1448                 system = NULL;
1449
1450         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1451         if (!dir)
1452                 goto out_fail;
1453
1454         if (!system) {
1455                 system = create_new_subsystem(name);
1456                 if (!system)
1457                         goto out_free;
1458         } else
1459                 __get_system(system);
1460
1461         dir->entry = debugfs_create_dir(name, parent);
1462         if (!dir->entry) {
1463                 pr_warning("Failed to create system directory %s\n", name);
1464                 __put_system(system);
1465                 goto out_free;
1466         }
1467
1468         dir->tr = tr;
1469         dir->ref_count = 1;
1470         dir->nr_events = 1;
1471         dir->subsystem = system;
1472         file->system = dir;
1473
1474         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1475                                     &ftrace_subsystem_filter_fops);
1476         if (!entry) {
1477                 kfree(system->filter);
1478                 system->filter = NULL;
1479                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1480         }
1481
1482         trace_create_file("enable", 0644, dir->entry, dir,
1483                           &ftrace_system_enable_fops);
1484
1485         list_add(&dir->list, &tr->systems);
1486
1487         return dir->entry;
1488
1489  out_free:
1490         kfree(dir);
1491  out_fail:
1492         /* Only print this message if failed on memory allocation */
1493         if (!dir || !system)
1494                 pr_warning("No memory to create event subsystem %s\n",
1495                            name);
1496         return NULL;
1497 }
1498
1499 static int
1500 event_create_dir(struct dentry *parent, struct ftrace_event_file *file)
1501 {
1502         struct ftrace_event_call *call = file->event_call;
1503         struct trace_array *tr = file->tr;
1504         struct list_head *head;
1505         struct dentry *d_events;
1506         int ret;
1507
1508         /*
1509          * If the trace point header did not define TRACE_SYSTEM
1510          * then the system would be called "TRACE_SYSTEM".
1511          */
1512         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1513                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1514                 if (!d_events)
1515                         return -ENOMEM;
1516         } else
1517                 d_events = parent;
1518
1519         file->dir = debugfs_create_dir(call->name, d_events);
1520         if (!file->dir) {
1521                 pr_warning("Could not create debugfs '%s' directory\n",
1522                            call->name);
1523                 return -1;
1524         }
1525
1526         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1527                 trace_create_file("enable", 0644, file->dir, file,
1528                                   &ftrace_enable_fops);
1529
1530 #ifdef CONFIG_PERF_EVENTS
1531         if (call->event.type && call->class->reg)
1532                 trace_create_file("id", 0444, file->dir,
1533                                   (void *)(long)call->event.type,
1534                                   &ftrace_event_id_fops);
1535 #endif
1536
1537         /*
1538          * Other events may have the same class. Only update
1539          * the fields if they are not already defined.
1540          */
1541         head = trace_get_fields(call);
1542         if (list_empty(head)) {
1543                 ret = call->class->define_fields(call);
1544                 if (ret < 0) {
1545                         pr_warning("Could not initialize trace point"
1546                                    " events/%s\n", call->name);
1547                         return -1;
1548                 }
1549         }
1550         trace_create_file("filter", 0644, file->dir, file,
1551                           &ftrace_event_filter_fops);
1552
1553         trace_create_file("trigger", 0644, file->dir, file,
1554                           &event_trigger_fops);
1555
1556         trace_create_file("format", 0444, file->dir, call,
1557                           &ftrace_event_format_fops);
1558
1559         return 0;
1560 }
1561
1562 static void remove_event_from_tracers(struct ftrace_event_call *call)
1563 {
1564         struct ftrace_event_file *file;
1565         struct trace_array *tr;
1566
1567         do_for_each_event_file_safe(tr, file) {
1568                 if (file->event_call != call)
1569                         continue;
1570
1571                 remove_event_file_dir(file);
1572                 /*
1573                  * The do_for_each_event_file_safe() is
1574                  * a double loop. After finding the call for this
1575                  * trace_array, we use break to jump to the next
1576                  * trace_array.
1577                  */
1578                 break;
1579         } while_for_each_event_file();
1580 }
1581
1582 static void event_remove(struct ftrace_event_call *call)
1583 {
1584         struct trace_array *tr;
1585         struct ftrace_event_file *file;
1586
1587         do_for_each_event_file(tr, file) {
1588                 if (file->event_call != call)
1589                         continue;
1590                 ftrace_event_enable_disable(file, 0);
1591                 destroy_preds(file);
1592                 /*
1593                  * The do_for_each_event_file() is
1594                  * a double loop. After finding the call for this
1595                  * trace_array, we use break to jump to the next
1596                  * trace_array.
1597                  */
1598                 break;
1599         } while_for_each_event_file();
1600
1601         if (call->event.funcs)
1602                 __unregister_ftrace_event(&call->event);
1603         remove_event_from_tracers(call);
1604         list_del(&call->list);
1605 }
1606
1607 static int event_init(struct ftrace_event_call *call)
1608 {
1609         int ret = 0;
1610
1611         if (WARN_ON(!call->name))
1612                 return -EINVAL;
1613
1614         if (call->class->raw_init) {
1615                 ret = call->class->raw_init(call);
1616                 if (ret < 0 && ret != -ENOSYS)
1617                         pr_warn("Could not initialize trace events/%s\n",
1618                                 call->name);
1619         }
1620
1621         return ret;
1622 }
1623
1624 static int
1625 __register_event(struct ftrace_event_call *call, struct module *mod)
1626 {
1627         int ret;
1628
1629         ret = event_init(call);
1630         if (ret < 0)
1631                 return ret;
1632
1633         list_add(&call->list, &ftrace_events);
1634         call->mod = mod;
1635
1636         return 0;
1637 }
1638
1639 static struct ftrace_event_file *
1640 trace_create_new_event(struct ftrace_event_call *call,
1641                        struct trace_array *tr)
1642 {
1643         struct ftrace_event_file *file;
1644
1645         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1646         if (!file)
1647                 return NULL;
1648
1649         file->event_call = call;
1650         file->tr = tr;
1651         atomic_set(&file->sm_ref, 0);
1652         atomic_set(&file->tm_ref, 0);
1653         INIT_LIST_HEAD(&file->triggers);
1654         list_add(&file->list, &tr->events);
1655
1656         return file;
1657 }
1658
1659 /* Add an event to a trace directory */
1660 static int
1661 __trace_add_new_event(struct ftrace_event_call *call, struct trace_array *tr)
1662 {
1663         struct ftrace_event_file *file;
1664
1665         file = trace_create_new_event(call, tr);
1666         if (!file)
1667                 return -ENOMEM;
1668
1669         return event_create_dir(tr->event_dir, file);
1670 }
1671
1672 /*
1673  * Just create a decriptor for early init. A descriptor is required
1674  * for enabling events at boot. We want to enable events before
1675  * the filesystem is initialized.
1676  */
1677 static __init int
1678 __trace_early_add_new_event(struct ftrace_event_call *call,
1679                             struct trace_array *tr)
1680 {
1681         struct ftrace_event_file *file;
1682
1683         file = trace_create_new_event(call, tr);
1684         if (!file)
1685                 return -ENOMEM;
1686
1687         return 0;
1688 }
1689
1690 struct ftrace_module_file_ops;
1691 static void __add_event_to_tracers(struct ftrace_event_call *call);
1692
1693 /* Add an additional event_call dynamically */
1694 int trace_add_event_call(struct ftrace_event_call *call)
1695 {
1696         int ret;
1697         mutex_lock(&trace_types_lock);
1698         mutex_lock(&event_mutex);
1699
1700         ret = __register_event(call, NULL);
1701         if (ret >= 0)
1702                 __add_event_to_tracers(call);
1703
1704         mutex_unlock(&event_mutex);
1705         mutex_unlock(&trace_types_lock);
1706         return ret;
1707 }
1708
1709 /*
1710  * Must be called under locking of trace_types_lock, event_mutex and
1711  * trace_event_sem.
1712  */
1713 static void __trace_remove_event_call(struct ftrace_event_call *call)
1714 {
1715         event_remove(call);
1716         trace_destroy_fields(call);
1717         destroy_call_preds(call);
1718 }
1719
1720 static int probe_remove_event_call(struct ftrace_event_call *call)
1721 {
1722         struct trace_array *tr;
1723         struct ftrace_event_file *file;
1724
1725 #ifdef CONFIG_PERF_EVENTS
1726         if (call->perf_refcount)
1727                 return -EBUSY;
1728 #endif
1729         do_for_each_event_file(tr, file) {
1730                 if (file->event_call != call)
1731                         continue;
1732                 /*
1733                  * We can't rely on ftrace_event_enable_disable(enable => 0)
1734                  * we are going to do, FTRACE_EVENT_FL_SOFT_MODE can suppress
1735                  * TRACE_REG_UNREGISTER.
1736                  */
1737                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
1738                         return -EBUSY;
1739                 /*
1740                  * The do_for_each_event_file_safe() is
1741                  * a double loop. After finding the call for this
1742                  * trace_array, we use break to jump to the next
1743                  * trace_array.
1744                  */
1745                 break;
1746         } while_for_each_event_file();
1747
1748         __trace_remove_event_call(call);
1749
1750         return 0;
1751 }
1752
1753 /* Remove an event_call */
1754 int trace_remove_event_call(struct ftrace_event_call *call)
1755 {
1756         int ret;
1757
1758         mutex_lock(&trace_types_lock);
1759         mutex_lock(&event_mutex);
1760         down_write(&trace_event_sem);
1761         ret = probe_remove_event_call(call);
1762         up_write(&trace_event_sem);
1763         mutex_unlock(&event_mutex);
1764         mutex_unlock(&trace_types_lock);
1765
1766         return ret;
1767 }
1768
1769 #define for_each_event(event, start, end)                       \
1770         for (event = start;                                     \
1771              (unsigned long)event < (unsigned long)end;         \
1772              event++)
1773
1774 #ifdef CONFIG_MODULES
1775
1776 static void trace_module_add_events(struct module *mod)
1777 {
1778         struct ftrace_event_call **call, **start, **end;
1779
1780         if (!mod->num_trace_events)
1781                 return;
1782
1783         /* Don't add infrastructure for mods without tracepoints */
1784         if (trace_module_has_bad_taint(mod)) {
1785                 pr_err("%s: module has bad taint, not creating trace events\n",
1786                        mod->name);
1787                 return;
1788         }
1789
1790         start = mod->trace_events;
1791         end = mod->trace_events + mod->num_trace_events;
1792
1793         for_each_event(call, start, end) {
1794                 __register_event(*call, mod);
1795                 __add_event_to_tracers(*call);
1796         }
1797 }
1798
1799 static void trace_module_remove_events(struct module *mod)
1800 {
1801         struct ftrace_event_call *call, *p;
1802         bool clear_trace = false;
1803
1804         down_write(&trace_event_sem);
1805         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1806                 if (call->mod == mod) {
1807                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1808                                 clear_trace = true;
1809                         __trace_remove_event_call(call);
1810                 }
1811         }
1812         up_write(&trace_event_sem);
1813
1814         /*
1815          * It is safest to reset the ring buffer if the module being unloaded
1816          * registered any events that were used. The only worry is if
1817          * a new module gets loaded, and takes on the same id as the events
1818          * of this module. When printing out the buffer, traced events left
1819          * over from this module may be passed to the new module events and
1820          * unexpected results may occur.
1821          */
1822         if (clear_trace)
1823                 tracing_reset_all_online_cpus();
1824 }
1825
1826 static int trace_module_notify(struct notifier_block *self,
1827                                unsigned long val, void *data)
1828 {
1829         struct module *mod = data;
1830
1831         mutex_lock(&trace_types_lock);
1832         mutex_lock(&event_mutex);
1833         switch (val) {
1834         case MODULE_STATE_COMING:
1835                 trace_module_add_events(mod);
1836                 break;
1837         case MODULE_STATE_GOING:
1838                 trace_module_remove_events(mod);
1839                 break;
1840         }
1841         mutex_unlock(&event_mutex);
1842         mutex_unlock(&trace_types_lock);
1843
1844         return 0;
1845 }
1846
1847 static struct notifier_block trace_module_nb = {
1848         .notifier_call = trace_module_notify,
1849         .priority = 0,
1850 };
1851 #endif /* CONFIG_MODULES */
1852
1853 /* Create a new event directory structure for a trace directory. */
1854 static void
1855 __trace_add_event_dirs(struct trace_array *tr)
1856 {
1857         struct ftrace_event_call *call;
1858         int ret;
1859
1860         list_for_each_entry(call, &ftrace_events, list) {
1861                 ret = __trace_add_new_event(call, tr);
1862                 if (ret < 0)
1863                         pr_warning("Could not create directory for event %s\n",
1864                                    call->name);
1865         }
1866 }
1867
1868 struct ftrace_event_file *
1869 find_event_file(struct trace_array *tr, const char *system,  const char *event)
1870 {
1871         struct ftrace_event_file *file;
1872         struct ftrace_event_call *call;
1873
1874         list_for_each_entry(file, &tr->events, list) {
1875
1876                 call = file->event_call;
1877
1878                 if (!call->name || !call->class || !call->class->reg)
1879                         continue;
1880
1881                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1882                         continue;
1883
1884                 if (strcmp(event, call->name) == 0 &&
1885                     strcmp(system, call->class->system) == 0)
1886                         return file;
1887         }
1888         return NULL;
1889 }
1890
1891 #ifdef CONFIG_DYNAMIC_FTRACE
1892
1893 /* Avoid typos */
1894 #define ENABLE_EVENT_STR        "enable_event"
1895 #define DISABLE_EVENT_STR       "disable_event"
1896
1897 struct event_probe_data {
1898         struct ftrace_event_file        *file;
1899         unsigned long                   count;
1900         int                             ref;
1901         bool                            enable;
1902 };
1903
1904 static void
1905 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1906 {
1907         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1908         struct event_probe_data *data = *pdata;
1909
1910         if (!data)
1911                 return;
1912
1913         if (data->enable)
1914                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1915         else
1916                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1917 }
1918
1919 static void
1920 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1921 {
1922         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1923         struct event_probe_data *data = *pdata;
1924
1925         if (!data)
1926                 return;
1927
1928         if (!data->count)
1929                 return;
1930
1931         /* Skip if the event is in a state we want to switch to */
1932         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1933                 return;
1934
1935         if (data->count != -1)
1936                 (data->count)--;
1937
1938         event_enable_probe(ip, parent_ip, _data);
1939 }
1940
1941 static int
1942 event_enable_print(struct seq_file *m, unsigned long ip,
1943                       struct ftrace_probe_ops *ops, void *_data)
1944 {
1945         struct event_probe_data *data = _data;
1946
1947         seq_printf(m, "%ps:", (void *)ip);
1948
1949         seq_printf(m, "%s:%s:%s",
1950                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1951                    data->file->event_call->class->system,
1952                    data->file->event_call->name);
1953
1954         if (data->count == -1)
1955                 seq_printf(m, ":unlimited\n");
1956         else
1957                 seq_printf(m, ":count=%ld\n", data->count);
1958
1959         return 0;
1960 }
1961
1962 static int
1963 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
1964                   void **_data)
1965 {
1966         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1967         struct event_probe_data *data = *pdata;
1968
1969         data->ref++;
1970         return 0;
1971 }
1972
1973 static void
1974 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
1975                   void **_data)
1976 {
1977         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1978         struct event_probe_data *data = *pdata;
1979
1980         if (WARN_ON_ONCE(data->ref <= 0))
1981                 return;
1982
1983         data->ref--;
1984         if (!data->ref) {
1985                 /* Remove the SOFT_MODE flag */
1986                 __ftrace_event_enable_disable(data->file, 0, 1);
1987                 module_put(data->file->event_call->mod);
1988                 kfree(data);
1989         }
1990         *pdata = NULL;
1991 }
1992
1993 static struct ftrace_probe_ops event_enable_probe_ops = {
1994         .func                   = event_enable_probe,
1995         .print                  = event_enable_print,
1996         .init                   = event_enable_init,
1997         .free                   = event_enable_free,
1998 };
1999
2000 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2001         .func                   = event_enable_count_probe,
2002         .print                  = event_enable_print,
2003         .init                   = event_enable_init,
2004         .free                   = event_enable_free,
2005 };
2006
2007 static struct ftrace_probe_ops event_disable_probe_ops = {
2008         .func                   = event_enable_probe,
2009         .print                  = event_enable_print,
2010         .init                   = event_enable_init,
2011         .free                   = event_enable_free,
2012 };
2013
2014 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2015         .func                   = event_enable_count_probe,
2016         .print                  = event_enable_print,
2017         .init                   = event_enable_init,
2018         .free                   = event_enable_free,
2019 };
2020
2021 static int
2022 event_enable_func(struct ftrace_hash *hash,
2023                   char *glob, char *cmd, char *param, int enabled)
2024 {
2025         struct trace_array *tr = top_trace_array();
2026         struct ftrace_event_file *file;
2027         struct ftrace_probe_ops *ops;
2028         struct event_probe_data *data;
2029         const char *system;
2030         const char *event;
2031         char *number;
2032         bool enable;
2033         int ret;
2034
2035         /* hash funcs only work with set_ftrace_filter */
2036         if (!enabled || !param)
2037                 return -EINVAL;
2038
2039         system = strsep(&param, ":");
2040         if (!param)
2041                 return -EINVAL;
2042
2043         event = strsep(&param, ":");
2044
2045         mutex_lock(&event_mutex);
2046
2047         ret = -EINVAL;
2048         file = find_event_file(tr, system, event);
2049         if (!file)
2050                 goto out;
2051
2052         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2053
2054         if (enable)
2055                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2056         else
2057                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2058
2059         if (glob[0] == '!') {
2060                 unregister_ftrace_function_probe_func(glob+1, ops);
2061                 ret = 0;
2062                 goto out;
2063         }
2064
2065         ret = -ENOMEM;
2066         data = kzalloc(sizeof(*data), GFP_KERNEL);
2067         if (!data)
2068                 goto out;
2069
2070         data->enable = enable;
2071         data->count = -1;
2072         data->file = file;
2073
2074         if (!param)
2075                 goto out_reg;
2076
2077         number = strsep(&param, ":");
2078
2079         ret = -EINVAL;
2080         if (!strlen(number))
2081                 goto out_free;
2082
2083         /*
2084          * We use the callback data field (which is a pointer)
2085          * as our counter.
2086          */
2087         ret = kstrtoul(number, 0, &data->count);
2088         if (ret)
2089                 goto out_free;
2090
2091  out_reg:
2092         /* Don't let event modules unload while probe registered */
2093         ret = try_module_get(file->event_call->mod);
2094         if (!ret) {
2095                 ret = -EBUSY;
2096                 goto out_free;
2097         }
2098
2099         ret = __ftrace_event_enable_disable(file, 1, 1);
2100         if (ret < 0)
2101                 goto out_put;
2102         ret = register_ftrace_function_probe(glob, ops, data);
2103         /*
2104          * The above returns on success the # of functions enabled,
2105          * but if it didn't find any functions it returns zero.
2106          * Consider no functions a failure too.
2107          */
2108         if (!ret) {
2109                 ret = -ENOENT;
2110                 goto out_disable;
2111         } else if (ret < 0)
2112                 goto out_disable;
2113         /* Just return zero, not the number of enabled functions */
2114         ret = 0;
2115  out:
2116         mutex_unlock(&event_mutex);
2117         return ret;
2118
2119  out_disable:
2120         __ftrace_event_enable_disable(file, 0, 1);
2121  out_put:
2122         module_put(file->event_call->mod);
2123  out_free:
2124         kfree(data);
2125         goto out;
2126 }
2127
2128 static struct ftrace_func_command event_enable_cmd = {
2129         .name                   = ENABLE_EVENT_STR,
2130         .func                   = event_enable_func,
2131 };
2132
2133 static struct ftrace_func_command event_disable_cmd = {
2134         .name                   = DISABLE_EVENT_STR,
2135         .func                   = event_enable_func,
2136 };
2137
2138 static __init int register_event_cmds(void)
2139 {
2140         int ret;
2141
2142         ret = register_ftrace_command(&event_enable_cmd);
2143         if (WARN_ON(ret < 0))
2144                 return ret;
2145         ret = register_ftrace_command(&event_disable_cmd);
2146         if (WARN_ON(ret < 0))
2147                 unregister_ftrace_command(&event_enable_cmd);
2148         return ret;
2149 }
2150 #else
2151 static inline int register_event_cmds(void) { return 0; }
2152 #endif /* CONFIG_DYNAMIC_FTRACE */
2153
2154 /*
2155  * The top level array has already had its ftrace_event_file
2156  * descriptors created in order to allow for early events to
2157  * be recorded. This function is called after the debugfs has been
2158  * initialized, and we now have to create the files associated
2159  * to the events.
2160  */
2161 static __init void
2162 __trace_early_add_event_dirs(struct trace_array *tr)
2163 {
2164         struct ftrace_event_file *file;
2165         int ret;
2166
2167
2168         list_for_each_entry(file, &tr->events, list) {
2169                 ret = event_create_dir(tr->event_dir, file);
2170                 if (ret < 0)
2171                         pr_warning("Could not create directory for event %s\n",
2172                                    file->event_call->name);
2173         }
2174 }
2175
2176 /*
2177  * For early boot up, the top trace array requires to have
2178  * a list of events that can be enabled. This must be done before
2179  * the filesystem is set up in order to allow events to be traced
2180  * early.
2181  */
2182 static __init void
2183 __trace_early_add_events(struct trace_array *tr)
2184 {
2185         struct ftrace_event_call *call;
2186         int ret;
2187
2188         list_for_each_entry(call, &ftrace_events, list) {
2189                 /* Early boot up should not have any modules loaded */
2190                 if (WARN_ON_ONCE(call->mod))
2191                         continue;
2192
2193                 ret = __trace_early_add_new_event(call, tr);
2194                 if (ret < 0)
2195                         pr_warning("Could not create early event %s\n",
2196                                    call->name);
2197         }
2198 }
2199
2200 /* Remove the event directory structure for a trace directory. */
2201 static void
2202 __trace_remove_event_dirs(struct trace_array *tr)
2203 {
2204         struct ftrace_event_file *file, *next;
2205
2206         list_for_each_entry_safe(file, next, &tr->events, list)
2207                 remove_event_file_dir(file);
2208 }
2209
2210 static void __add_event_to_tracers(struct ftrace_event_call *call)
2211 {
2212         struct trace_array *tr;
2213
2214         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2215                 __trace_add_new_event(call, tr);
2216 }
2217
2218 extern struct ftrace_event_call *__start_ftrace_events[];
2219 extern struct ftrace_event_call *__stop_ftrace_events[];
2220
2221 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2222
2223 static __init int setup_trace_event(char *str)
2224 {
2225         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2226         ring_buffer_expanded = true;
2227         tracing_selftest_disabled = true;
2228
2229         return 1;
2230 }
2231 __setup("trace_event=", setup_trace_event);
2232
2233 /* Expects to have event_mutex held when called */
2234 static int
2235 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2236 {
2237         struct dentry *d_events;
2238         struct dentry *entry;
2239
2240         entry = debugfs_create_file("set_event", 0644, parent,
2241                                     tr, &ftrace_set_event_fops);
2242         if (!entry) {
2243                 pr_warning("Could not create debugfs 'set_event' entry\n");
2244                 return -ENOMEM;
2245         }
2246
2247         d_events = debugfs_create_dir("events", parent);
2248         if (!d_events) {
2249                 pr_warning("Could not create debugfs 'events' directory\n");
2250                 return -ENOMEM;
2251         }
2252
2253         /* ring buffer internal formats */
2254         trace_create_file("header_page", 0444, d_events,
2255                           ring_buffer_print_page_header,
2256                           &ftrace_show_header_fops);
2257
2258         trace_create_file("header_event", 0444, d_events,
2259                           ring_buffer_print_entry_header,
2260                           &ftrace_show_header_fops);
2261
2262         trace_create_file("enable", 0644, d_events,
2263                           tr, &ftrace_tr_enable_fops);
2264
2265         tr->event_dir = d_events;
2266
2267         return 0;
2268 }
2269
2270 /**
2271  * event_trace_add_tracer - add a instance of a trace_array to events
2272  * @parent: The parent dentry to place the files/directories for events in
2273  * @tr: The trace array associated with these events
2274  *
2275  * When a new instance is created, it needs to set up its events
2276  * directory, as well as other files associated with events. It also
2277  * creates the event hierachry in the @parent/events directory.
2278  *
2279  * Returns 0 on success.
2280  */
2281 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2282 {
2283         int ret;
2284
2285         mutex_lock(&event_mutex);
2286
2287         ret = create_event_toplevel_files(parent, tr);
2288         if (ret)
2289                 goto out_unlock;
2290
2291         down_write(&trace_event_sem);
2292         __trace_add_event_dirs(tr);
2293         up_write(&trace_event_sem);
2294
2295  out_unlock:
2296         mutex_unlock(&event_mutex);
2297
2298         return ret;
2299 }
2300
2301 /*
2302  * The top trace array already had its file descriptors created.
2303  * Now the files themselves need to be created.
2304  */
2305 static __init int
2306 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2307 {
2308         int ret;
2309
2310         mutex_lock(&event_mutex);
2311
2312         ret = create_event_toplevel_files(parent, tr);
2313         if (ret)
2314                 goto out_unlock;
2315
2316         down_write(&trace_event_sem);
2317         __trace_early_add_event_dirs(tr);
2318         up_write(&trace_event_sem);
2319
2320  out_unlock:
2321         mutex_unlock(&event_mutex);
2322
2323         return ret;
2324 }
2325
2326 int event_trace_del_tracer(struct trace_array *tr)
2327 {
2328         mutex_lock(&event_mutex);
2329
2330         /* Disable any event triggers and associated soft-disabled events */
2331         clear_event_triggers(tr);
2332
2333         /* Disable any running events */
2334         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2335
2336         /* Access to events are within rcu_read_lock_sched() */
2337         synchronize_sched();
2338
2339         down_write(&trace_event_sem);
2340         __trace_remove_event_dirs(tr);
2341         debugfs_remove_recursive(tr->event_dir);
2342         up_write(&trace_event_sem);
2343
2344         tr->event_dir = NULL;
2345
2346         mutex_unlock(&event_mutex);
2347
2348         return 0;
2349 }
2350
2351 static __init int event_trace_memsetup(void)
2352 {
2353         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2354         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2355         return 0;
2356 }
2357
2358 static __init int event_trace_enable(void)
2359 {
2360         struct trace_array *tr = top_trace_array();
2361         struct ftrace_event_call **iter, *call;
2362         char *buf = bootup_event_buf;
2363         char *token;
2364         int ret;
2365
2366         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2367
2368                 call = *iter;
2369                 ret = event_init(call);
2370                 if (!ret)
2371                         list_add(&call->list, &ftrace_events);
2372         }
2373
2374         /*
2375          * We need the top trace array to have a working set of trace
2376          * points at early init, before the debug files and directories
2377          * are created. Create the file entries now, and attach them
2378          * to the actual file dentries later.
2379          */
2380         __trace_early_add_events(tr);
2381
2382         while (true) {
2383                 token = strsep(&buf, ",");
2384
2385                 if (!token)
2386                         break;
2387                 if (!*token)
2388                         continue;
2389
2390                 ret = ftrace_set_clr_event(tr, token, 1);
2391                 if (ret)
2392                         pr_warn("Failed to enable trace event: %s\n", token);
2393         }
2394
2395         trace_printk_start_comm();
2396
2397         register_event_cmds();
2398
2399         register_trigger_cmds();
2400
2401         return 0;
2402 }
2403
2404 static __init int event_trace_init(void)
2405 {
2406         struct trace_array *tr;
2407         struct dentry *d_tracer;
2408         struct dentry *entry;
2409         int ret;
2410
2411         tr = top_trace_array();
2412
2413         d_tracer = tracing_init_dentry();
2414         if (!d_tracer)
2415                 return 0;
2416
2417         entry = debugfs_create_file("available_events", 0444, d_tracer,
2418                                     tr, &ftrace_avail_fops);
2419         if (!entry)
2420                 pr_warning("Could not create debugfs "
2421                            "'available_events' entry\n");
2422
2423         if (trace_define_common_fields())
2424                 pr_warning("tracing: Failed to allocate common fields");
2425
2426         ret = early_event_add_tracer(d_tracer, tr);
2427         if (ret)
2428                 return ret;
2429
2430 #ifdef CONFIG_MODULES
2431         ret = register_module_notifier(&trace_module_nb);
2432         if (ret)
2433                 pr_warning("Failed to register trace events module notifier\n");
2434 #endif
2435         return 0;
2436 }
2437 early_initcall(event_trace_memsetup);
2438 core_initcall(event_trace_enable);
2439 fs_initcall(event_trace_init);
2440
2441 #ifdef CONFIG_FTRACE_STARTUP_TEST
2442
2443 static DEFINE_SPINLOCK(test_spinlock);
2444 static DEFINE_SPINLOCK(test_spinlock_irq);
2445 static DEFINE_MUTEX(test_mutex);
2446
2447 static __init void test_work(struct work_struct *dummy)
2448 {
2449         spin_lock(&test_spinlock);
2450         spin_lock_irq(&test_spinlock_irq);
2451         udelay(1);
2452         spin_unlock_irq(&test_spinlock_irq);
2453         spin_unlock(&test_spinlock);
2454
2455         mutex_lock(&test_mutex);
2456         msleep(1);
2457         mutex_unlock(&test_mutex);
2458 }
2459
2460 static __init int event_test_thread(void *unused)
2461 {
2462         void *test_malloc;
2463
2464         test_malloc = kmalloc(1234, GFP_KERNEL);
2465         if (!test_malloc)
2466                 pr_info("failed to kmalloc\n");
2467
2468         schedule_on_each_cpu(test_work);
2469
2470         kfree(test_malloc);
2471
2472         set_current_state(TASK_INTERRUPTIBLE);
2473         while (!kthread_should_stop())
2474                 schedule();
2475
2476         return 0;
2477 }
2478
2479 /*
2480  * Do various things that may trigger events.
2481  */
2482 static __init void event_test_stuff(void)
2483 {
2484         struct task_struct *test_thread;
2485
2486         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2487         msleep(1);
2488         kthread_stop(test_thread);
2489 }
2490
2491 /*
2492  * For every trace event defined, we will test each trace point separately,
2493  * and then by groups, and finally all trace points.
2494  */
2495 static __init void event_trace_self_tests(void)
2496 {
2497         struct ftrace_subsystem_dir *dir;
2498         struct ftrace_event_file *file;
2499         struct ftrace_event_call *call;
2500         struct event_subsystem *system;
2501         struct trace_array *tr;
2502         int ret;
2503
2504         tr = top_trace_array();
2505
2506         pr_info("Running tests on trace events:\n");
2507
2508         list_for_each_entry(file, &tr->events, list) {
2509
2510                 call = file->event_call;
2511
2512                 /* Only test those that have a probe */
2513                 if (!call->class || !call->class->probe)
2514                         continue;
2515
2516 /*
2517  * Testing syscall events here is pretty useless, but
2518  * we still do it if configured. But this is time consuming.
2519  * What we really need is a user thread to perform the
2520  * syscalls as we test.
2521  */
2522 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2523                 if (call->class->system &&
2524                     strcmp(call->class->system, "syscalls") == 0)
2525                         continue;
2526 #endif
2527
2528                 pr_info("Testing event %s: ", call->name);
2529
2530                 /*
2531                  * If an event is already enabled, someone is using
2532                  * it and the self test should not be on.
2533                  */
2534                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2535                         pr_warning("Enabled event during self test!\n");
2536                         WARN_ON_ONCE(1);
2537                         continue;
2538                 }
2539
2540                 ftrace_event_enable_disable(file, 1);
2541                 event_test_stuff();
2542                 ftrace_event_enable_disable(file, 0);
2543
2544                 pr_cont("OK\n");
2545         }
2546
2547         /* Now test at the sub system level */
2548
2549         pr_info("Running tests on trace event systems:\n");
2550
2551         list_for_each_entry(dir, &tr->systems, list) {
2552
2553                 system = dir->subsystem;
2554
2555                 /* the ftrace system is special, skip it */
2556                 if (strcmp(system->name, "ftrace") == 0)
2557                         continue;
2558
2559                 pr_info("Testing event system %s: ", system->name);
2560
2561                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2562                 if (WARN_ON_ONCE(ret)) {
2563                         pr_warning("error enabling system %s\n",
2564                                    system->name);
2565                         continue;
2566                 }
2567
2568                 event_test_stuff();
2569
2570                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2571                 if (WARN_ON_ONCE(ret)) {
2572                         pr_warning("error disabling system %s\n",
2573                                    system->name);
2574                         continue;
2575                 }
2576
2577                 pr_cont("OK\n");
2578         }
2579
2580         /* Test with all events enabled */
2581
2582         pr_info("Running tests on all trace events:\n");
2583         pr_info("Testing all events: ");
2584
2585         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2586         if (WARN_ON_ONCE(ret)) {
2587                 pr_warning("error enabling all events\n");
2588                 return;
2589         }
2590
2591         event_test_stuff();
2592
2593         /* reset sysname */
2594         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2595         if (WARN_ON_ONCE(ret)) {
2596                 pr_warning("error disabling all events\n");
2597                 return;
2598         }
2599
2600         pr_cont("OK\n");
2601 }
2602
2603 #ifdef CONFIG_FUNCTION_TRACER
2604
2605 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2606
2607 static void
2608 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2609                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2610 {
2611         struct ring_buffer_event *event;
2612         struct ring_buffer *buffer;
2613         struct ftrace_entry *entry;
2614         unsigned long flags;
2615         long disabled;
2616         int cpu;
2617         int pc;
2618
2619         pc = preempt_count();
2620         preempt_disable_notrace();
2621         cpu = raw_smp_processor_id();
2622         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2623
2624         if (disabled != 1)
2625                 goto out;
2626
2627         local_save_flags(flags);
2628
2629         event = trace_current_buffer_lock_reserve(&buffer,
2630                                                   TRACE_FN, sizeof(*entry),
2631                                                   flags, pc);
2632         if (!event)
2633                 goto out;
2634         entry   = ring_buffer_event_data(event);
2635         entry->ip                       = ip;
2636         entry->parent_ip                = parent_ip;
2637
2638         trace_buffer_unlock_commit(buffer, event, flags, pc);
2639
2640  out:
2641         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2642         preempt_enable_notrace();
2643 }
2644
2645 static struct ftrace_ops trace_ops __initdata  =
2646 {
2647         .func = function_test_events_call,
2648         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2649 };
2650
2651 static __init void event_trace_self_test_with_function(void)
2652 {
2653         int ret;
2654         ret = register_ftrace_function(&trace_ops);
2655         if (WARN_ON(ret < 0)) {
2656                 pr_info("Failed to enable function tracer for event tests\n");
2657                 return;
2658         }
2659         pr_info("Running tests again, along with the function tracer\n");
2660         event_trace_self_tests();
2661         unregister_ftrace_function(&trace_ops);
2662 }
2663 #else
2664 static __init void event_trace_self_test_with_function(void)
2665 {
2666 }
2667 #endif
2668
2669 static __init int event_trace_self_tests_init(void)
2670 {
2671         if (!tracing_selftest_disabled) {
2672                 event_trace_self_tests();
2673                 event_trace_self_test_with_function();
2674         }
2675
2676         return 0;
2677 }
2678
2679 late_initcall(event_trace_self_tests_init);
2680
2681 #endif