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