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

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