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