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 LIST_HEAD(ftrace_events);
  31 static LIST_HEAD(ftrace_common_fields);
  32
  33 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  34
  35 static struct kmem_cache *field_cachep;
  36 static struct kmem_cache *file_cachep;
  37
  38 #define SYSTEM_FL_FREE_NAME             (1 << 31)
  39
  40 static inline int system_refcount(struct event_subsystem *system)
  41 {
  42         return system->ref_count & ~SYSTEM_FL_FREE_NAME;
  43 }
  44
  45 static int system_refcount_inc(struct event_subsystem *system)
  46 {
  47         return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
  48 }
  49
  50 static int system_refcount_dec(struct event_subsystem *system)
  51 {
  52         return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
  53 }
  54
  55 /* Double loops, do not use break, only goto's work */
  56 #define do_for_each_event_file(tr, file)                        \
  57         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  58                 list_for_each_entry(file, &tr->events, list)
  59
  60 #define do_for_each_event_file_safe(tr, file)                   \
  61         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  62                 struct ftrace_event_file *___n;                         \
  63                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  64
  65 #define while_for_each_event_file()             \
  66         }
  67
  68 static struct list_head *
  69 trace_get_fields(struct ftrace_event_call *event_call)
  70 {
  71         if (!event_call->class->get_fields)
  72                 return &event_call->class->fields;
  73         return event_call->class->get_fields(event_call);
  74 }
  75
  76 static struct ftrace_event_field *
  77 __find_event_field(struct list_head *head, char *name)
  78 {
  79         struct ftrace_event_field *field;
  80
  81         list_for_each_entry(field, head, link) {
  82                 if (!strcmp(field->name, name))
  83                         return field;
  84         }
  85
  86         return NULL;
  87 }
  88
  89 struct ftrace_event_field *
  90 trace_find_event_field(struct ftrace_event_call *call, char *name)
  91 {
  92         struct ftrace_event_field *field;
  93         struct list_head *head;
  94
  95         field = __find_event_field(&ftrace_common_fields, name);
  96         if (field)
  97                 return field;
  98
  99         head = trace_get_fields(call);
 100         return __find_event_field(head, name);
 101 }
 102
 103 static int __trace_define_field(struct list_head *head, const char *type,
 104                                 const char *name, int offset, int size,
 105                                 int is_signed, int filter_type)
 106 {
 107         struct ftrace_event_field *field;
 108
 109         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 110         if (!field)
 111                 return -ENOMEM;
 112
 113         field->name = name;
 114         field->type = type;
 115
 116         if (filter_type == FILTER_OTHER)
 117                 field->filter_type = filter_assign_type(type);
 118         else
 119                 field->filter_type = filter_type;
 120
 121         field->offset = offset;
 122         field->size = size;
 123         field->is_signed = is_signed;
 124
 125         list_add(&field->link, head);
 126
 127         return 0;
 128 }
 129
 130 int trace_define_field(struct ftrace_event_call *call, const char *type,
 131                        const char *name, int offset, int size, int is_signed,
 132                        int filter_type)
 133 {
 134         struct list_head *head;
 135
 136         if (WARN_ON(!call->class))
 137                 return 0;
 138
 139         head = trace_get_fields(call);
 140         return __trace_define_field(head, type, name, offset, size,
 141                                     is_signed, filter_type);
 142 }
 143 EXPORT_SYMBOL_GPL(trace_define_field);
 144
 145 #define __common_field(type, item)                                      \
 146         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 147                                    "common_" #item,                     \
 148                                    offsetof(typeof(ent), item),         \
 149                                    sizeof(ent.item),                    \
 150                                    is_signed_type(type), FILTER_OTHER); \
 151         if (ret)                                                        \
 152                 return ret;
 153
 154 static int trace_define_common_fields(void)
 155 {
 156         int ret;
 157         struct trace_entry ent;
 158
 159         __common_field(unsigned short, type);
 160         __common_field(unsigned char, flags);
 161         __common_field(unsigned char, preempt_count);
 162         __common_field(int, pid);
 163
 164         return ret;
 165 }
 166
 167 static void trace_destroy_fields(struct ftrace_event_call *call)
 168 {
 169         struct ftrace_event_field *field, *next;
 170         struct list_head *head;
 171
 172         head = trace_get_fields(call);
 173         list_for_each_entry_safe(field, next, head, link) {
 174                 list_del(&field->link);
 175                 kmem_cache_free(field_cachep, field);
 176         }
 177 }
 178
 179 int trace_event_raw_init(struct ftrace_event_call *call)
 180 {
 181         int id;
 182
 183         id = register_ftrace_event(&call->event);
 184         if (!id)
 185                 return -ENODEV;
 186
 187         return 0;
 188 }
 189 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 190
 191 void *ftrace_event_buffer_reserve(struct ftrace_event_buffer *fbuffer,
 192                                   struct ftrace_event_file *ftrace_file,
 193                                   unsigned long len)
 194 {
 195         struct ftrace_event_call *event_call = ftrace_file->event_call;
 196
 197         local_save_flags(fbuffer->flags);
 198         fbuffer->pc = preempt_count();
 199         fbuffer->ftrace_file = ftrace_file;
 200
 201         fbuffer->event =
 202                 trace_event_buffer_lock_reserve(&fbuffer->buffer, ftrace_file,
 203                                                 event_call->event.type, len,
 204                                                 fbuffer->flags, fbuffer->pc);
 205         if (!fbuffer->event)
 206                 return NULL;
 207
 208         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 209         return fbuffer->entry;
 210 }
 211 EXPORT_SYMBOL_GPL(ftrace_event_buffer_reserve);
 212
 213 void ftrace_event_buffer_commit(struct ftrace_event_buffer *fbuffer)
 214 {
 215         event_trigger_unlock_commit(fbuffer->ftrace_file, fbuffer->buffer,
 216                                     fbuffer->event, fbuffer->entry,
 217                                     fbuffer->flags, fbuffer->pc);
 218 }
 219 EXPORT_SYMBOL_GPL(ftrace_event_buffer_commit);
 220
 221 int ftrace_event_reg(struct ftrace_event_call *call,
 222                      enum trace_reg type, void *data)
 223 {
 224         struct ftrace_event_file *file = data;
 225
 226         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 227         switch (type) {
 228         case TRACE_REG_REGISTER:
 229                 return tracepoint_probe_register(call->tp,
 230                                                  call->class->probe,
 231                                                  file);
 232         case TRACE_REG_UNREGISTER:
 233                 tracepoint_probe_unregister(call->tp,
 234                                             call->class->probe,
 235                                             file);
 236                 return 0;
 237
 238 #ifdef CONFIG_PERF_EVENTS
 239         case TRACE_REG_PERF_REGISTER:
 240                 return tracepoint_probe_register(call->tp,
 241                                                  call->class->perf_probe,
 242                                                  call);
 243         case TRACE_REG_PERF_UNREGISTER:
 244                 tracepoint_probe_unregister(call->tp,
 245                                             call->class->perf_probe,
 246                                             call);
 247                 return 0;
 248         case TRACE_REG_PERF_OPEN:
 249         case TRACE_REG_PERF_CLOSE:
 250         case TRACE_REG_PERF_ADD:
 251         case TRACE_REG_PERF_DEL:
 252                 return 0;
 253 #endif
 254         }
 255         return 0;
 256 }
 257 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 258
 259 void trace_event_enable_cmd_record(bool enable)
 260 {
 261         struct ftrace_event_file *file;
 262         struct trace_array *tr;
 263
 264         mutex_lock(&event_mutex);
 265         do_for_each_event_file(tr, file) {
 266
 267                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 268                         continue;
 269
 270                 if (enable) {
 271                         tracing_start_cmdline_record();
 272                         set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 273                 } else {
 274                         tracing_stop_cmdline_record();
 275                         clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 276                 }
 277         } while_for_each_event_file();
 278         mutex_unlock(&event_mutex);
 279 }
 280
 281 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
 282                                          int enable, int soft_disable)
 283 {
 284         struct ftrace_event_call *call = file->event_call;
 285         int ret = 0;
 286         int disable;
 287
 288         switch (enable) {
 289         case 0:
 290                 /*
 291                  * When soft_disable is set and enable is cleared, the sm_ref
 292                  * reference counter is decremented. If it reaches 0, we want
 293                  * to clear the SOFT_DISABLED flag but leave the event in the
 294                  * state that it was. That is, if the event was enabled and
 295                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 296                  * is set we do not want the event to be enabled before we
 297                  * clear the bit.
 298                  *
 299                  * When soft_disable is not set but the SOFT_MODE flag is,
 300                  * we do nothing. Do not disable the tracepoint, otherwise
 301                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 302                  */
 303                 if (soft_disable) {
 304                         if (atomic_dec_return(&file->sm_ref) > 0)
 305                                 break;
 306                         disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
 307                         clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 308                 } else
 309                         disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
 310
 311                 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
 312                         clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 313                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 314                                 tracing_stop_cmdline_record();
 315                                 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 316                         }
 317                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 318                 }
 319                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 320                 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 321                         set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 322                 else
 323                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 324                 break;
 325         case 1:
 326                 /*
 327                  * When soft_disable is set and enable is set, we want to
 328                  * register the tracepoint for the event, but leave the event
 329                  * as is. That means, if the event was already enabled, we do
 330                  * nothing (but set SOFT_MODE). If the event is disabled, we
 331                  * set SOFT_DISABLED before enabling the event tracepoint, so
 332                  * it still seems to be disabled.
 333                  */
 334                 if (!soft_disable)
 335                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 336                 else {
 337                         if (atomic_inc_return(&file->sm_ref) > 1)
 338                                 break;
 339                         set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 340                 }
 341
 342                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 343
 344                         /* Keep the event disabled, when going to SOFT_MODE. */
 345                         if (soft_disable)
 346                                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 347
 348                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 349                                 tracing_start_cmdline_record();
 350                                 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 351                         }
 352                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 353                         if (ret) {
 354                                 tracing_stop_cmdline_record();
 355                                 pr_info("event trace: Could not enable event "
 356                                         "%s\n", ftrace_event_name(call));
 357                                 break;
 358                         }
 359                         set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 360
 361                         /* WAS_ENABLED gets set but never cleared. */
 362                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 363                 }
 364                 break;
 365         }
 366
 367         return ret;
 368 }
 369
 370 int trace_event_enable_disable(struct ftrace_event_file *file,
 371                                int enable, int soft_disable)
 372 {
 373         return __ftrace_event_enable_disable(file, enable, soft_disable);
 374 }
 375
 376 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 377                                        int enable)
 378 {
 379         return __ftrace_event_enable_disable(file, enable, 0);
 380 }
 381
 382 static void ftrace_clear_events(struct trace_array *tr)
 383 {
 384         struct ftrace_event_file *file;
 385
 386         mutex_lock(&event_mutex);
 387         list_for_each_entry(file, &tr->events, list) {
 388                 ftrace_event_enable_disable(file, 0);
 389         }
 390         mutex_unlock(&event_mutex);
 391 }
 392
 393 static void __put_system(struct event_subsystem *system)
 394 {
 395         struct event_filter *filter = system->filter;
 396
 397         WARN_ON_ONCE(system_refcount(system) == 0);
 398         if (system_refcount_dec(system))
 399                 return;
 400
 401         list_del(&system->list);
 402
 403         if (filter) {
 404                 kfree(filter->filter_string);
 405                 kfree(filter);
 406         }
 407         if (system->ref_count & SYSTEM_FL_FREE_NAME)
 408                 kfree(system->name);
 409         kfree(system);
 410 }
 411
 412 static void __get_system(struct event_subsystem *system)
 413 {
 414         WARN_ON_ONCE(system_refcount(system) == 0);
 415         system_refcount_inc(system);
 416 }
 417
 418 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 419 {
 420         WARN_ON_ONCE(dir->ref_count == 0);
 421         dir->ref_count++;
 422         __get_system(dir->subsystem);
 423 }
 424
 425 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 426 {
 427         WARN_ON_ONCE(dir->ref_count == 0);
 428         /* If the subsystem is about to be freed, the dir must be too */
 429         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 430
 431         __put_system(dir->subsystem);
 432         if (!--dir->ref_count)
 433                 kfree(dir);
 434 }
 435
 436 static void put_system(struct ftrace_subsystem_dir *dir)
 437 {
 438         mutex_lock(&event_mutex);
 439         __put_system_dir(dir);
 440         mutex_unlock(&event_mutex);
 441 }
 442
 443 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
 444 {
 445         if (!dir)
 446                 return;
 447
 448         if (!--dir->nr_events) {
 449                 debugfs_remove_recursive(dir->entry);
 450                 list_del(&dir->list);
 451                 __put_system_dir(dir);
 452         }
 453 }
 454
 455 static void remove_event_file_dir(struct ftrace_event_file *file)
 456 {
 457         struct dentry *dir = file->dir;
 458         struct dentry *child;
 459
 460         if (dir) {
 461                 spin_lock(&dir->d_lock);        /* probably unneeded */
 462                 list_for_each_entry(child, &dir->d_subdirs, d_u.d_child) {
 463                         if (child->d_inode)     /* probably unneeded */
 464                                 child->d_inode->i_private = NULL;
 465                 }
 466                 spin_unlock(&dir->d_lock);
 467
 468                 debugfs_remove_recursive(dir);
 469         }
 470
 471         list_del(&file->list);
 472         remove_subsystem(file->system);
 473         kmem_cache_free(file_cachep, file);
 474 }
 475
 476 /*
 477  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 478  */
 479 static int
 480 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 481                               const char *sub, const char *event, int set)
 482 {
 483         struct ftrace_event_file *file;
 484         struct ftrace_event_call *call;
 485         const char *name;
 486         int ret = -EINVAL;
 487
 488         list_for_each_entry(file, &tr->events, list) {
 489
 490                 call = file->event_call;
 491                 name = ftrace_event_name(call);
 492
 493                 if (!name || !call->class || !call->class->reg)
 494                         continue;
 495
 496                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 497                         continue;
 498
 499                 if (match &&
 500                     strcmp(match, name) != 0 &&
 501                     strcmp(match, call->class->system) != 0)
 502                         continue;
 503
 504                 if (sub && strcmp(sub, call->class->system) != 0)
 505                         continue;
 506
 507                 if (event && strcmp(event, name) != 0)
 508                         continue;
 509
 510                 ftrace_event_enable_disable(file, set);
 511
 512                 ret = 0;
 513         }
 514
 515         return ret;
 516 }
 517
 518 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 519                                   const char *sub, const char *event, int set)
 520 {
 521         int ret;
 522
 523         mutex_lock(&event_mutex);
 524         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 525         mutex_unlock(&event_mutex);
 526
 527         return ret;
 528 }
 529
 530 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 531 {
 532         char *event = NULL, *sub = NULL, *match;
 533
 534         /*
 535          * The buf format can be <subsystem>:<event-name>
 536          *  *:<event-name> means any event by that name.
 537          *  :<event-name> is the same.
 538          *
 539          *  <subsystem>:* means all events in that subsystem
 540          *  <subsystem>: means the same.
 541          *
 542          *  <name> (no ':') means all events in a subsystem with
 543          *  the name <name> or any event that matches <name>
 544          */
 545
 546         match = strsep(&buf, ":");
 547         if (buf) {
 548                 sub = match;
 549                 event = buf;
 550                 match = NULL;
 551
 552                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 553                         sub = NULL;
 554                 if (!strlen(event) || strcmp(event, "*") == 0)
 555                         event = NULL;
 556         }
 557
 558         return __ftrace_set_clr_event(tr, match, sub, event, set);
 559 }
 560
 561 /**
 562  * trace_set_clr_event - enable or disable an event
 563  * @system: system name to match (NULL for any system)
 564  * @event: event name to match (NULL for all events, within system)
 565  * @set: 1 to enable, 0 to disable
 566  *
 567  * This is a way for other parts of the kernel to enable or disable
 568  * event recording.
 569  *
 570  * Returns 0 on success, -EINVAL if the parameters do not match any
 571  * registered events.
 572  */
 573 int trace_set_clr_event(const char *system, const char *event, int set)
 574 {
 575         struct trace_array *tr = top_trace_array();
 576
 577         if (!tr)
 578                 return -ENODEV;
 579
 580         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 581 }
 582 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 583
 584 /* 128 should be much more than enough */
 585 #define EVENT_BUF_SIZE          127
 586
 587 static ssize_t
 588 ftrace_event_write(struct file *file, const char __user *ubuf,
 589                    size_t cnt, loff_t *ppos)
 590 {
 591         struct trace_parser parser;
 592         struct seq_file *m = file->private_data;
 593         struct trace_array *tr = m->private;
 594         ssize_t read, ret;
 595
 596         if (!cnt)
 597                 return 0;
 598
 599         ret = tracing_update_buffers();
 600         if (ret < 0)
 601                 return ret;
 602
 603         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 604                 return -ENOMEM;
 605
 606         read = trace_get_user(&parser, ubuf, cnt, ppos);
 607
 608         if (read >= 0 && trace_parser_loaded((&parser))) {
 609                 int set = 1;
 610
 611                 if (*parser.buffer == '!')
 612                         set = 0;
 613
 614                 parser.buffer[parser.idx] = 0;
 615
 616                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 617                 if (ret)
 618                         goto out_put;
 619         }
 620
 621         ret = read;
 622
 623  out_put:
 624         trace_parser_put(&parser);
 625
 626         return ret;
 627 }
 628
 629 static void *
 630 t_next(struct seq_file *m, void *v, loff_t *pos)
 631 {
 632         struct ftrace_event_file *file = v;
 633         struct ftrace_event_call *call;
 634         struct trace_array *tr = m->private;
 635
 636         (*pos)++;
 637
 638         list_for_each_entry_continue(file, &tr->events, list) {
 639                 call = file->event_call;
 640                 /*
 641                  * The ftrace subsystem is for showing formats only.
 642                  * They can not be enabled or disabled via the event files.
 643                  */
 644                 if (call->class && call->class->reg)
 645                         return file;
 646         }
 647
 648         return NULL;
 649 }
 650
 651 static void *t_start(struct seq_file *m, loff_t *pos)
 652 {
 653         struct ftrace_event_file *file;
 654         struct trace_array *tr = m->private;
 655         loff_t l;
 656
 657         mutex_lock(&event_mutex);
 658
 659         file = list_entry(&tr->events, struct ftrace_event_file, list);
 660         for (l = 0; l <= *pos; ) {
 661                 file = t_next(m, file, &l);
 662                 if (!file)
 663                         break;
 664         }
 665         return file;
 666 }
 667
 668 static void *
 669 s_next(struct seq_file *m, void *v, loff_t *pos)
 670 {
 671         struct ftrace_event_file *file = v;
 672         struct trace_array *tr = m->private;
 673
 674         (*pos)++;
 675
 676         list_for_each_entry_continue(file, &tr->events, list) {
 677                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
 678                         return file;
 679         }
 680
 681         return NULL;
 682 }
 683
 684 static void *s_start(struct seq_file *m, loff_t *pos)
 685 {
 686         struct ftrace_event_file *file;
 687         struct trace_array *tr = m->private;
 688         loff_t l;
 689
 690         mutex_lock(&event_mutex);
 691
 692         file = list_entry(&tr->events, struct ftrace_event_file, list);
 693         for (l = 0; l <= *pos; ) {
 694                 file = s_next(m, file, &l);
 695                 if (!file)
 696                         break;
 697         }
 698         return file;
 699 }
 700
 701 static int t_show(struct seq_file *m, void *v)
 702 {
 703         struct ftrace_event_file *file = v;
 704         struct ftrace_event_call *call = file->event_call;
 705
 706         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 707                 seq_printf(m, "%s:", call->class->system);
 708         seq_printf(m, "%s\n", ftrace_event_name(call));
 709
 710         return 0;
 711 }
 712
 713 static void t_stop(struct seq_file *m, void *p)
 714 {
 715         mutex_unlock(&event_mutex);
 716 }
 717
 718 static ssize_t
 719 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 720                   loff_t *ppos)
 721 {
 722         struct ftrace_event_file *file;
 723         unsigned long flags;
 724         char buf[4] = "0";
 725
 726         mutex_lock(&event_mutex);
 727         file = event_file_data(filp);
 728         if (likely(file))
 729                 flags = file->flags;
 730         mutex_unlock(&event_mutex);
 731
 732         if (!file)
 733                 return -ENODEV;
 734
 735         if (flags & FTRACE_EVENT_FL_ENABLED &&
 736             !(flags & FTRACE_EVENT_FL_SOFT_DISABLED))
 737                 strcpy(buf, "1");
 738
 739         if (flags & FTRACE_EVENT_FL_SOFT_DISABLED ||
 740             flags & FTRACE_EVENT_FL_SOFT_MODE)
 741                 strcat(buf, "*");
 742
 743         strcat(buf, "\n");
 744
 745         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
 746 }
 747
 748 static ssize_t
 749 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 750                    loff_t *ppos)
 751 {
 752         struct ftrace_event_file *file;
 753         unsigned long val;
 754         int ret;
 755
 756         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 757         if (ret)
 758                 return ret;
 759
 760         ret = tracing_update_buffers();
 761         if (ret < 0)
 762                 return ret;
 763
 764         switch (val) {
 765         case 0:
 766         case 1:
 767                 ret = -ENODEV;
 768                 mutex_lock(&event_mutex);
 769                 file = event_file_data(filp);
 770                 if (likely(file))
 771                         ret = ftrace_event_enable_disable(file, val);
 772                 mutex_unlock(&event_mutex);
 773                 break;
 774
 775         default:
 776                 return -EINVAL;
 777         }
 778
 779         *ppos += cnt;
 780
 781         return ret ? ret : cnt;
 782 }
 783
 784 static ssize_t
 785 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 786                    loff_t *ppos)
 787 {
 788         const char set_to_char[4] = { '?', '0', '1', 'X' };
 789         struct ftrace_subsystem_dir *dir = filp->private_data;
 790         struct event_subsystem *system = dir->subsystem;
 791         struct ftrace_event_call *call;
 792         struct ftrace_event_file *file;
 793         struct trace_array *tr = dir->tr;
 794         char buf[2];
 795         int set = 0;
 796         int ret;
 797
 798         mutex_lock(&event_mutex);
 799         list_for_each_entry(file, &tr->events, list) {
 800                 call = file->event_call;
 801                 if (!ftrace_event_name(call) || !call->class || !call->class->reg)
 802                         continue;
 803
 804                 if (system && strcmp(call->class->system, system->name) != 0)
 805                         continue;
 806
 807                 /*
 808                  * We need to find out if all the events are set
 809                  * or if all events or cleared, or if we have
 810                  * a mixture.
 811                  */
 812                 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
 813
 814                 /*
 815                  * If we have a mixture, no need to look further.
 816                  */
 817                 if (set == 3)
 818                         break;
 819         }
 820         mutex_unlock(&event_mutex);
 821
 822         buf[0] = set_to_char[set];
 823         buf[1] = '\n';
 824
 825         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 826
 827         return ret;
 828 }
 829
 830 static ssize_t
 831 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 832                     loff_t *ppos)
 833 {
 834         struct ftrace_subsystem_dir *dir = filp->private_data;
 835         struct event_subsystem *system = dir->subsystem;
 836         const char *name = NULL;
 837         unsigned long val;
 838         ssize_t ret;
 839
 840         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 841         if (ret)
 842                 return ret;
 843
 844         ret = tracing_update_buffers();
 845         if (ret < 0)
 846                 return ret;
 847
 848         if (val != 0 && val != 1)
 849                 return -EINVAL;
 850
 851         /*
 852          * Opening of "enable" adds a ref count to system,
 853          * so the name is safe to use.
 854          */
 855         if (system)
 856                 name = system->name;
 857
 858         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
 859         if (ret)
 860                 goto out;
 861
 862         ret = cnt;
 863
 864 out:
 865         *ppos += cnt;
 866
 867         return ret;
 868 }
 869
 870 enum {
 871         FORMAT_HEADER           = 1,
 872         FORMAT_FIELD_SEPERATOR  = 2,
 873         FORMAT_PRINTFMT         = 3,
 874 };
 875
 876 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 877 {
 878         struct ftrace_event_call *call = event_file_data(m->private);
 879         struct list_head *common_head = &ftrace_common_fields;
 880         struct list_head *head = trace_get_fields(call);
 881         struct list_head *node = v;
 882
 883         (*pos)++;
 884
 885         switch ((unsigned long)v) {
 886         case FORMAT_HEADER:
 887                 node = common_head;
 888                 break;
 889
 890         case FORMAT_FIELD_SEPERATOR:
 891                 node = head;
 892                 break;
 893
 894         case FORMAT_PRINTFMT:
 895                 /* all done */
 896                 return NULL;
 897         }
 898
 899         node = node->prev;
 900         if (node == common_head)
 901                 return (void *)FORMAT_FIELD_SEPERATOR;
 902         else if (node == head)
 903                 return (void *)FORMAT_PRINTFMT;
 904         else
 905                 return node;
 906 }
 907
 908 static int f_show(struct seq_file *m, void *v)
 909 {
 910         struct ftrace_event_call *call = event_file_data(m->private);
 911         struct ftrace_event_field *field;
 912         const char *array_descriptor;
 913
 914         switch ((unsigned long)v) {
 915         case FORMAT_HEADER:
 916                 seq_printf(m, "name: %s\n", ftrace_event_name(call));
 917                 seq_printf(m, "ID: %d\n", call->event.type);
 918                 seq_printf(m, "format:\n");
 919                 return 0;
 920
 921         case FORMAT_FIELD_SEPERATOR:
 922                 seq_putc(m, '\n');
 923                 return 0;
 924
 925         case FORMAT_PRINTFMT:
 926                 seq_printf(m, "\nprint fmt: %s\n",
 927                            call->print_fmt);
 928                 return 0;
 929         }
 930
 931         field = list_entry(v, struct ftrace_event_field, link);
 932         /*
 933          * Smartly shows the array type(except dynamic array).
 934          * Normal:
 935          *      field:TYPE VAR
 936          * If TYPE := TYPE[LEN], it is shown:
 937          *      field:TYPE VAR[LEN]
 938          */
 939         array_descriptor = strchr(field->type, '[');
 940
 941         if (!strncmp(field->type, "__data_loc", 10))
 942                 array_descriptor = NULL;
 943
 944         if (!array_descriptor)
 945                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 946                            field->type, field->name, field->offset,
 947                            field->size, !!field->is_signed);
 948         else
 949                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 950                            (int)(array_descriptor - field->type),
 951                            field->type, field->name,
 952                            array_descriptor, field->offset,
 953                            field->size, !!field->is_signed);
 954
 955         return 0;
 956 }
 957
 958 static void *f_start(struct seq_file *m, loff_t *pos)
 959 {
 960         void *p = (void *)FORMAT_HEADER;
 961         loff_t l = 0;
 962
 963         /* ->stop() is called even if ->start() fails */
 964         mutex_lock(&event_mutex);
 965         if (!event_file_data(m->private))
 966                 return ERR_PTR(-ENODEV);
 967
 968         while (l < *pos && p)
 969                 p = f_next(m, p, &l);
 970
 971         return p;
 972 }
 973
 974 static void f_stop(struct seq_file *m, void *p)
 975 {
 976         mutex_unlock(&event_mutex);
 977 }
 978
 979 static const struct seq_operations trace_format_seq_ops = {
 980         .start          = f_start,
 981         .next           = f_next,
 982         .stop           = f_stop,
 983         .show           = f_show,
 984 };
 985
 986 static int trace_format_open(struct inode *inode, struct file *file)
 987 {
 988         struct seq_file *m;
 989         int ret;
 990
 991         ret = seq_open(file, &trace_format_seq_ops);
 992         if (ret < 0)
 993                 return ret;
 994
 995         m = file->private_data;
 996         m->private = file;
 997
 998         return 0;
 999 }
1000
1001 static ssize_t
1002 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1003 {
1004         int id = (long)event_file_data(filp);
1005         char buf[32];
1006         int len;
1007
1008         if (*ppos)
1009                 return 0;
1010
1011         if (unlikely(!id))
1012                 return -ENODEV;
1013
1014         len = sprintf(buf, "%d\n", id);
1015
1016         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1017 }
1018
1019 static ssize_t
1020 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1021                   loff_t *ppos)
1022 {
1023         struct ftrace_event_file *file;
1024         struct trace_seq *s;
1025         int r = -ENODEV;
1026
1027         if (*ppos)
1028                 return 0;
1029
1030         s = kmalloc(sizeof(*s), GFP_KERNEL);
1031
1032         if (!s)
1033                 return -ENOMEM;
1034
1035         trace_seq_init(s);
1036
1037         mutex_lock(&event_mutex);
1038         file = event_file_data(filp);
1039         if (file)
1040                 print_event_filter(file, s);
1041         mutex_unlock(&event_mutex);
1042
1043         if (file)
1044                 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1045
1046         kfree(s);
1047
1048         return r;
1049 }
1050
1051 static ssize_t
1052 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1053                    loff_t *ppos)
1054 {
1055         struct ftrace_event_file *file;
1056         char *buf;
1057         int err = -ENODEV;
1058
1059         if (cnt >= PAGE_SIZE)
1060                 return -EINVAL;
1061
1062         buf = (char *)__get_free_page(GFP_TEMPORARY);
1063         if (!buf)
1064                 return -ENOMEM;
1065
1066         if (copy_from_user(buf, ubuf, cnt)) {
1067                 free_page((unsigned long) buf);
1068                 return -EFAULT;
1069         }
1070         buf[cnt] = '\0';
1071
1072         mutex_lock(&event_mutex);
1073         file = event_file_data(filp);
1074         if (file)
1075                 err = apply_event_filter(file, buf);
1076         mutex_unlock(&event_mutex);
1077
1078         free_page((unsigned long) buf);
1079         if (err < 0)
1080                 return err;
1081
1082         *ppos += cnt;
1083
1084         return cnt;
1085 }
1086
1087 static LIST_HEAD(event_subsystems);
1088
1089 static int subsystem_open(struct inode *inode, struct file *filp)
1090 {
1091         struct event_subsystem *system = NULL;
1092         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1093         struct trace_array *tr;
1094         int ret;
1095
1096         if (tracing_is_disabled())
1097                 return -ENODEV;
1098
1099         /* Make sure the system still exists */
1100         mutex_lock(&trace_types_lock);
1101         mutex_lock(&event_mutex);
1102         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1103                 list_for_each_entry(dir, &tr->systems, list) {
1104                         if (dir == inode->i_private) {
1105                                 /* Don't open systems with no events */
1106                                 if (dir->nr_events) {
1107                                         __get_system_dir(dir);
1108                                         system = dir->subsystem;
1109                                 }
1110                                 goto exit_loop;
1111                         }
1112                 }
1113         }
1114  exit_loop:
1115         mutex_unlock(&event_mutex);
1116         mutex_unlock(&trace_types_lock);
1117
1118         if (!system)
1119                 return -ENODEV;
1120
1121         /* Some versions of gcc think dir can be uninitialized here */
1122         WARN_ON(!dir);
1123
1124         /* Still need to increment the ref count of the system */
1125         if (trace_array_get(tr) < 0) {
1126                 put_system(dir);
1127                 return -ENODEV;
1128         }
1129
1130         ret = tracing_open_generic(inode, filp);
1131         if (ret < 0) {
1132                 trace_array_put(tr);
1133                 put_system(dir);
1134         }
1135
1136         return ret;
1137 }
1138
1139 static int system_tr_open(struct inode *inode, struct file *filp)
1140 {
1141         struct ftrace_subsystem_dir *dir;
1142         struct trace_array *tr = inode->i_private;
1143         int ret;
1144
1145         if (tracing_is_disabled())
1146                 return -ENODEV;
1147
1148         if (trace_array_get(tr) < 0)
1149                 return -ENODEV;
1150
1151         /* Make a temporary dir that has no system but points to tr */
1152         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1153         if (!dir) {
1154                 trace_array_put(tr);
1155                 return -ENOMEM;
1156         }
1157
1158         dir->tr = tr;
1159
1160         ret = tracing_open_generic(inode, filp);
1161         if (ret < 0) {
1162                 trace_array_put(tr);
1163                 kfree(dir);
1164                 return ret;
1165         }
1166
1167         filp->private_data = dir;
1168
1169         return 0;
1170 }
1171
1172 static int subsystem_release(struct inode *inode, struct file *file)
1173 {
1174         struct ftrace_subsystem_dir *dir = file->private_data;
1175
1176         trace_array_put(dir->tr);
1177
1178         /*
1179          * If dir->subsystem is NULL, then this is a temporary
1180          * descriptor that was made for a trace_array to enable
1181          * all subsystems.
1182          */
1183         if (dir->subsystem)
1184                 put_system(dir);
1185         else
1186                 kfree(dir);
1187
1188         return 0;
1189 }
1190
1191 static ssize_t
1192 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1193                       loff_t *ppos)
1194 {
1195         struct ftrace_subsystem_dir *dir = filp->private_data;
1196         struct event_subsystem *system = dir->subsystem;
1197         struct trace_seq *s;
1198         int r;
1199
1200         if (*ppos)
1201                 return 0;
1202
1203         s = kmalloc(sizeof(*s), GFP_KERNEL);
1204         if (!s)
1205                 return -ENOMEM;
1206
1207         trace_seq_init(s);
1208
1209         print_subsystem_event_filter(system, s);
1210         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1211
1212         kfree(s);
1213
1214         return r;
1215 }
1216
1217 static ssize_t
1218 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1219                        loff_t *ppos)
1220 {
1221         struct ftrace_subsystem_dir *dir = filp->private_data;
1222         char *buf;
1223         int err;
1224
1225         if (cnt >= PAGE_SIZE)
1226                 return -EINVAL;
1227
1228         buf = (char *)__get_free_page(GFP_TEMPORARY);
1229         if (!buf)
1230                 return -ENOMEM;
1231
1232         if (copy_from_user(buf, ubuf, cnt)) {
1233                 free_page((unsigned long) buf);
1234                 return -EFAULT;
1235         }
1236         buf[cnt] = '\0';
1237
1238         err = apply_subsystem_event_filter(dir, buf);
1239         free_page((unsigned long) buf);
1240         if (err < 0)
1241                 return err;
1242
1243         *ppos += cnt;
1244
1245         return cnt;
1246 }
1247
1248 static ssize_t
1249 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1250 {
1251         int (*func)(struct trace_seq *s) = filp->private_data;
1252         struct trace_seq *s;
1253         int r;
1254
1255         if (*ppos)
1256                 return 0;
1257
1258         s = kmalloc(sizeof(*s), GFP_KERNEL);
1259         if (!s)
1260                 return -ENOMEM;
1261
1262         trace_seq_init(s);
1263
1264         func(s);
1265         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1266
1267         kfree(s);
1268
1269         return r;
1270 }
1271
1272 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1273 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1274 static int ftrace_event_release(struct inode *inode, struct file *file);
1275
1276 static const struct seq_operations show_event_seq_ops = {
1277         .start = t_start,
1278         .next = t_next,
1279         .show = t_show,
1280         .stop = t_stop,
1281 };
1282
1283 static const struct seq_operations show_set_event_seq_ops = {
1284         .start = s_start,
1285         .next = s_next,
1286         .show = t_show,
1287         .stop = t_stop,
1288 };
1289
1290 static const struct file_operations ftrace_avail_fops = {
1291         .open = ftrace_event_avail_open,
1292         .read = seq_read,
1293         .llseek = seq_lseek,
1294         .release = seq_release,
1295 };
1296
1297 static const struct file_operations ftrace_set_event_fops = {
1298         .open = ftrace_event_set_open,
1299         .read = seq_read,
1300         .write = ftrace_event_write,
1301         .llseek = seq_lseek,
1302         .release = ftrace_event_release,
1303 };
1304
1305 static const struct file_operations ftrace_enable_fops = {
1306         .open = tracing_open_generic,
1307         .read = event_enable_read,
1308         .write = event_enable_write,
1309         .llseek = default_llseek,
1310 };
1311
1312 static const struct file_operations ftrace_event_format_fops = {
1313         .open = trace_format_open,
1314         .read = seq_read,
1315         .llseek = seq_lseek,
1316         .release = seq_release,
1317 };
1318
1319 static const struct file_operations ftrace_event_id_fops = {
1320         .read = event_id_read,
1321         .llseek = default_llseek,
1322 };
1323
1324 static const struct file_operations ftrace_event_filter_fops = {
1325         .open = tracing_open_generic,
1326         .read = event_filter_read,
1327         .write = event_filter_write,
1328         .llseek = default_llseek,
1329 };
1330
1331 static const struct file_operations ftrace_subsystem_filter_fops = {
1332         .open = subsystem_open,
1333         .read = subsystem_filter_read,
1334         .write = subsystem_filter_write,
1335         .llseek = default_llseek,
1336         .release = subsystem_release,
1337 };
1338
1339 static const struct file_operations ftrace_system_enable_fops = {
1340         .open = subsystem_open,
1341         .read = system_enable_read,
1342         .write = system_enable_write,
1343         .llseek = default_llseek,
1344         .release = subsystem_release,
1345 };
1346
1347 static const struct file_operations ftrace_tr_enable_fops = {
1348         .open = system_tr_open,
1349         .read = system_enable_read,
1350         .write = system_enable_write,
1351         .llseek = default_llseek,
1352         .release = subsystem_release,
1353 };
1354
1355 static const struct file_operations ftrace_show_header_fops = {
1356         .open = tracing_open_generic,
1357         .read = show_header,
1358         .llseek = default_llseek,
1359 };
1360
1361 static int
1362 ftrace_event_open(struct inode *inode, struct file *file,
1363                   const struct seq_operations *seq_ops)
1364 {
1365         struct seq_file *m;
1366         int ret;
1367
1368         ret = seq_open(file, seq_ops);
1369         if (ret < 0)
1370                 return ret;
1371         m = file->private_data;
1372         /* copy tr over to seq ops */
1373         m->private = inode->i_private;
1374
1375         return ret;
1376 }
1377
1378 static int ftrace_event_release(struct inode *inode, struct file *file)
1379 {
1380         struct trace_array *tr = inode->i_private;
1381
1382         trace_array_put(tr);
1383
1384         return seq_release(inode, file);
1385 }
1386
1387 static int
1388 ftrace_event_avail_open(struct inode *inode, struct file *file)
1389 {
1390         const struct seq_operations *seq_ops = &show_event_seq_ops;
1391
1392         return ftrace_event_open(inode, file, seq_ops);
1393 }
1394
1395 static int
1396 ftrace_event_set_open(struct inode *inode, struct file *file)
1397 {
1398         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1399         struct trace_array *tr = inode->i_private;
1400         int ret;
1401
1402         if (trace_array_get(tr) < 0)
1403                 return -ENODEV;
1404
1405         if ((file->f_mode & FMODE_WRITE) &&
1406             (file->f_flags & O_TRUNC))
1407                 ftrace_clear_events(tr);
1408
1409         ret = ftrace_event_open(inode, file, seq_ops);
1410         if (ret < 0)
1411                 trace_array_put(tr);
1412         return ret;
1413 }
1414
1415 static struct event_subsystem *
1416 create_new_subsystem(const char *name)
1417 {
1418         struct event_subsystem *system;
1419
1420         /* need to create new entry */
1421         system = kmalloc(sizeof(*system), GFP_KERNEL);
1422         if (!system)
1423                 return NULL;
1424
1425         system->ref_count = 1;
1426
1427         /* Only allocate if dynamic (kprobes and modules) */
1428         if (!core_kernel_data((unsigned long)name)) {
1429                 system->ref_count |= SYSTEM_FL_FREE_NAME;
1430                 system->name = kstrdup(name, GFP_KERNEL);
1431                 if (!system->name)
1432                         goto out_free;
1433         } else
1434                 system->name = name;
1435
1436         system->filter = NULL;
1437
1438         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1439         if (!system->filter)
1440                 goto out_free;
1441
1442         list_add(&system->list, &event_subsystems);
1443
1444         return system;
1445
1446  out_free:
1447         if (system->ref_count & SYSTEM_FL_FREE_NAME)
1448                 kfree(system->name);
1449         kfree(system);
1450         return NULL;
1451 }
1452
1453 static struct dentry *
1454 event_subsystem_dir(struct trace_array *tr, const char *name,
1455                     struct ftrace_event_file *file, struct dentry *parent)
1456 {
1457         struct ftrace_subsystem_dir *dir;
1458         struct event_subsystem *system;
1459         struct dentry *entry;
1460
1461         /* First see if we did not already create this dir */
1462         list_for_each_entry(dir, &tr->systems, list) {
1463                 system = dir->subsystem;
1464                 if (strcmp(system->name, name) == 0) {
1465                         dir->nr_events++;
1466                         file->system = dir;
1467                         return dir->entry;
1468                 }
1469         }
1470
1471         /* Now see if the system itself exists. */
1472         list_for_each_entry(system, &event_subsystems, list) {
1473                 if (strcmp(system->name, name) == 0)
1474                         break;
1475         }
1476         /* Reset system variable when not found */
1477         if (&system->list == &event_subsystems)
1478                 system = NULL;
1479
1480         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1481         if (!dir)
1482                 goto out_fail;
1483
1484         if (!system) {
1485                 system = create_new_subsystem(name);
1486                 if (!system)
1487                         goto out_free;
1488         } else
1489                 __get_system(system);
1490
1491         dir->entry = debugfs_create_dir(name, parent);
1492         if (!dir->entry) {
1493                 pr_warning("Failed to create system directory %s\n", name);
1494                 __put_system(system);
1495                 goto out_free;
1496         }
1497
1498         dir->tr = tr;
1499         dir->ref_count = 1;
1500         dir->nr_events = 1;
1501         dir->subsystem = system;
1502         file->system = dir;
1503
1504         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1505                                     &ftrace_subsystem_filter_fops);
1506         if (!entry) {
1507                 kfree(system->filter);
1508                 system->filter = NULL;
1509                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1510         }
1511
1512         trace_create_file("enable", 0644, dir->entry, dir,
1513                           &ftrace_system_enable_fops);
1514
1515         list_add(&dir->list, &tr->systems);
1516
1517         return dir->entry;
1518
1519  out_free:
1520         kfree(dir);
1521  out_fail:
1522         /* Only print this message if failed on memory allocation */
1523         if (!dir || !system)
1524                 pr_warning("No memory to create event subsystem %s\n",
1525                            name);
1526         return NULL;
1527 }
1528
1529 static int
1530 event_create_dir(struct dentry *parent, struct ftrace_event_file *file)
1531 {
1532         struct ftrace_event_call *call = file->event_call;
1533         struct trace_array *tr = file->tr;
1534         struct list_head *head;
1535         struct dentry *d_events;
1536         const char *name;
1537         int ret;
1538
1539         /*
1540          * If the trace point header did not define TRACE_SYSTEM
1541          * then the system would be called "TRACE_SYSTEM".
1542          */
1543         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1544                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1545                 if (!d_events)
1546                         return -ENOMEM;
1547         } else
1548                 d_events = parent;
1549
1550         name = ftrace_event_name(call);
1551         file->dir = debugfs_create_dir(name, d_events);
1552         if (!file->dir) {
1553                 pr_warning("Could not create debugfs '%s' directory\n",
1554                            name);
1555                 return -1;
1556         }
1557
1558         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1559                 trace_create_file("enable", 0644, file->dir, file,
1560                                   &ftrace_enable_fops);
1561
1562 #ifdef CONFIG_PERF_EVENTS
1563         if (call->event.type && call->class->reg)
1564                 trace_create_file("id", 0444, file->dir,
1565                                   (void *)(long)call->event.type,
1566                                   &ftrace_event_id_fops);
1567 #endif
1568
1569         /*
1570          * Other events may have the same class. Only update
1571          * the fields if they are not already defined.
1572          */
1573         head = trace_get_fields(call);
1574         if (list_empty(head)) {
1575                 ret = call->class->define_fields(call);
1576                 if (ret < 0) {
1577                         pr_warning("Could not initialize trace point"
1578                                    " events/%s\n", name);
1579                         return -1;
1580                 }
1581         }
1582         trace_create_file("filter", 0644, file->dir, file,
1583                           &ftrace_event_filter_fops);
1584
1585         trace_create_file("trigger", 0644, file->dir, file,
1586                           &event_trigger_fops);
1587
1588         trace_create_file("format", 0444, file->dir, call,
1589                           &ftrace_event_format_fops);
1590
1591         return 0;
1592 }
1593
1594 static void remove_event_from_tracers(struct ftrace_event_call *call)
1595 {
1596         struct ftrace_event_file *file;
1597         struct trace_array *tr;
1598
1599         do_for_each_event_file_safe(tr, file) {
1600                 if (file->event_call != call)
1601                         continue;
1602
1603                 remove_event_file_dir(file);
1604                 /*
1605                  * The do_for_each_event_file_safe() is
1606                  * a double loop. After finding the call for this
1607                  * trace_array, we use break to jump to the next
1608                  * trace_array.
1609                  */
1610                 break;
1611         } while_for_each_event_file();
1612 }
1613
1614 static void event_remove(struct ftrace_event_call *call)
1615 {
1616         struct trace_array *tr;
1617         struct ftrace_event_file *file;
1618
1619         do_for_each_event_file(tr, file) {
1620                 if (file->event_call != call)
1621                         continue;
1622                 ftrace_event_enable_disable(file, 0);
1623                 destroy_preds(file);
1624                 /*
1625                  * The do_for_each_event_file() is
1626                  * a double loop. After finding the call for this
1627                  * trace_array, we use break to jump to the next
1628                  * trace_array.
1629                  */
1630                 break;
1631         } while_for_each_event_file();
1632
1633         if (call->event.funcs)
1634                 __unregister_ftrace_event(&call->event);
1635         remove_event_from_tracers(call);
1636         list_del(&call->list);
1637 }
1638
1639 static int event_init(struct ftrace_event_call *call)
1640 {
1641         int ret = 0;
1642         const char *name;
1643
1644         name = ftrace_event_name(call);
1645         if (WARN_ON(!name))
1646                 return -EINVAL;
1647
1648         if (call->class->raw_init) {
1649                 ret = call->class->raw_init(call);
1650                 if (ret < 0 && ret != -ENOSYS)
1651                         pr_warn("Could not initialize trace events/%s\n",
1652                                 name);
1653         }
1654
1655         return ret;
1656 }
1657
1658 static int
1659 __register_event(struct ftrace_event_call *call, struct module *mod)
1660 {
1661         int ret;
1662
1663         ret = event_init(call);
1664         if (ret < 0)
1665                 return ret;
1666
1667         list_add(&call->list, &ftrace_events);
1668         call->mod = mod;
1669
1670         return 0;
1671 }
1672
1673 static struct ftrace_event_file *
1674 trace_create_new_event(struct ftrace_event_call *call,
1675                        struct trace_array *tr)
1676 {
1677         struct ftrace_event_file *file;
1678
1679         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1680         if (!file)
1681                 return NULL;
1682
1683         file->event_call = call;
1684         file->tr = tr;
1685         atomic_set(&file->sm_ref, 0);
1686         atomic_set(&file->tm_ref, 0);
1687         INIT_LIST_HEAD(&file->triggers);
1688         list_add(&file->list, &tr->events);
1689
1690         return file;
1691 }
1692
1693 /* Add an event to a trace directory */
1694 static int
1695 __trace_add_new_event(struct ftrace_event_call *call, struct trace_array *tr)
1696 {
1697         struct ftrace_event_file *file;
1698
1699         file = trace_create_new_event(call, tr);
1700         if (!file)
1701                 return -ENOMEM;
1702
1703         return event_create_dir(tr->event_dir, file);
1704 }
1705
1706 /*
1707  * Just create a decriptor for early init. A descriptor is required
1708  * for enabling events at boot. We want to enable events before
1709  * the filesystem is initialized.
1710  */
1711 static __init int
1712 __trace_early_add_new_event(struct ftrace_event_call *call,
1713                             struct trace_array *tr)
1714 {
1715         struct ftrace_event_file *file;
1716
1717         file = trace_create_new_event(call, tr);
1718         if (!file)
1719                 return -ENOMEM;
1720
1721         return 0;
1722 }
1723
1724 struct ftrace_module_file_ops;
1725 static void __add_event_to_tracers(struct ftrace_event_call *call);
1726
1727 /* Add an additional event_call dynamically */
1728 int trace_add_event_call(struct ftrace_event_call *call)
1729 {
1730         int ret;
1731         mutex_lock(&trace_types_lock);
1732         mutex_lock(&event_mutex);
1733
1734         ret = __register_event(call, NULL);
1735         if (ret >= 0)
1736                 __add_event_to_tracers(call);
1737
1738         mutex_unlock(&event_mutex);
1739         mutex_unlock(&trace_types_lock);
1740         return ret;
1741 }
1742
1743 /*
1744  * Must be called under locking of trace_types_lock, event_mutex and
1745  * trace_event_sem.
1746  */
1747 static void __trace_remove_event_call(struct ftrace_event_call *call)
1748 {
1749         event_remove(call);
1750         trace_destroy_fields(call);
1751         destroy_call_preds(call);
1752 }
1753
1754 static int probe_remove_event_call(struct ftrace_event_call *call)
1755 {
1756         struct trace_array *tr;
1757         struct ftrace_event_file *file;
1758
1759 #ifdef CONFIG_PERF_EVENTS
1760         if (call->perf_refcount)
1761                 return -EBUSY;
1762 #endif
1763         do_for_each_event_file(tr, file) {
1764                 if (file->event_call != call)
1765                         continue;
1766                 /*
1767                  * We can't rely on ftrace_event_enable_disable(enable => 0)
1768                  * we are going to do, FTRACE_EVENT_FL_SOFT_MODE can suppress
1769                  * TRACE_REG_UNREGISTER.
1770                  */
1771                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
1772                         return -EBUSY;
1773                 /*
1774                  * The do_for_each_event_file_safe() is
1775                  * a double loop. After finding the call for this
1776                  * trace_array, we use break to jump to the next
1777                  * trace_array.
1778                  */
1779                 break;
1780         } while_for_each_event_file();
1781
1782         __trace_remove_event_call(call);
1783
1784         return 0;
1785 }
1786
1787 /* Remove an event_call */
1788 int trace_remove_event_call(struct ftrace_event_call *call)
1789 {
1790         int ret;
1791
1792         mutex_lock(&trace_types_lock);
1793         mutex_lock(&event_mutex);
1794         down_write(&trace_event_sem);
1795         ret = probe_remove_event_call(call);
1796         up_write(&trace_event_sem);
1797         mutex_unlock(&event_mutex);
1798         mutex_unlock(&trace_types_lock);
1799
1800         return ret;
1801 }
1802
1803 #define for_each_event(event, start, end)                       \
1804         for (event = start;                                     \
1805              (unsigned long)event < (unsigned long)end;         \
1806              event++)
1807
1808 #ifdef CONFIG_MODULES
1809
1810 static void trace_module_add_events(struct module *mod)
1811 {
1812         struct ftrace_event_call **call, **start, **end;
1813
1814         if (!mod->num_trace_events)
1815                 return;
1816
1817         /* Don't add infrastructure for mods without tracepoints */
1818         if (trace_module_has_bad_taint(mod)) {
1819                 pr_err("%s: module has bad taint, not creating trace events\n",
1820                        mod->name);
1821                 return;
1822         }
1823
1824         start = mod->trace_events;
1825         end = mod->trace_events + mod->num_trace_events;
1826
1827         for_each_event(call, start, end) {
1828                 __register_event(*call, mod);
1829                 __add_event_to_tracers(*call);
1830         }
1831 }
1832
1833 static void trace_module_remove_events(struct module *mod)
1834 {
1835         struct ftrace_event_call *call, *p;
1836         bool clear_trace = false;
1837
1838         down_write(&trace_event_sem);
1839         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1840                 if (call->mod == mod) {
1841                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1842                                 clear_trace = true;
1843                         __trace_remove_event_call(call);
1844                 }
1845         }
1846         up_write(&trace_event_sem);
1847
1848         /*
1849          * It is safest to reset the ring buffer if the module being unloaded
1850          * registered any events that were used. The only worry is if
1851          * a new module gets loaded, and takes on the same id as the events
1852          * of this module. When printing out the buffer, traced events left
1853          * over from this module may be passed to the new module events and
1854          * unexpected results may occur.
1855          */
1856         if (clear_trace)
1857                 tracing_reset_all_online_cpus();
1858 }
1859
1860 static int trace_module_notify(struct notifier_block *self,
1861                                unsigned long val, void *data)
1862 {
1863         struct module *mod = data;
1864
1865         mutex_lock(&trace_types_lock);
1866         mutex_lock(&event_mutex);
1867         switch (val) {
1868         case MODULE_STATE_COMING:
1869                 trace_module_add_events(mod);
1870                 break;
1871         case MODULE_STATE_GOING:
1872                 trace_module_remove_events(mod);
1873                 break;
1874         }
1875         mutex_unlock(&event_mutex);
1876         mutex_unlock(&trace_types_lock);
1877
1878         return 0;
1879 }
1880
1881 static struct notifier_block trace_module_nb = {
1882         .notifier_call = trace_module_notify,
1883         .priority = 0,
1884 };
1885 #endif /* CONFIG_MODULES */
1886
1887 /* Create a new event directory structure for a trace directory. */
1888 static void
1889 __trace_add_event_dirs(struct trace_array *tr)
1890 {
1891         struct ftrace_event_call *call;
1892         int ret;
1893
1894         list_for_each_entry(call, &ftrace_events, list) {
1895                 ret = __trace_add_new_event(call, tr);
1896                 if (ret < 0)
1897                         pr_warning("Could not create directory for event %s\n",
1898                                    ftrace_event_name(call));
1899         }
1900 }
1901
1902 struct ftrace_event_file *
1903 find_event_file(struct trace_array *tr, const char *system,  const char *event)
1904 {
1905         struct ftrace_event_file *file;
1906         struct ftrace_event_call *call;
1907         const char *name;
1908
1909         list_for_each_entry(file, &tr->events, list) {
1910
1911                 call = file->event_call;
1912                 name = ftrace_event_name(call);
1913
1914                 if (!name || !call->class || !call->class->reg)
1915                         continue;
1916
1917                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1918                         continue;
1919
1920                 if (strcmp(event, name) == 0 &&
1921                     strcmp(system, call->class->system) == 0)
1922                         return file;
1923         }
1924         return NULL;
1925 }
1926
1927 #ifdef CONFIG_DYNAMIC_FTRACE
1928
1929 /* Avoid typos */
1930 #define ENABLE_EVENT_STR        "enable_event"
1931 #define DISABLE_EVENT_STR       "disable_event"
1932
1933 struct event_probe_data {
1934         struct ftrace_event_file        *file;
1935         unsigned long                   count;
1936         int                             ref;
1937         bool                            enable;
1938 };
1939
1940 static void
1941 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1942 {
1943         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1944         struct event_probe_data *data = *pdata;
1945
1946         if (!data)
1947                 return;
1948
1949         if (data->enable)
1950                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1951         else
1952                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1953 }
1954
1955 static void
1956 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1957 {
1958         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1959         struct event_probe_data *data = *pdata;
1960
1961         if (!data)
1962                 return;
1963
1964         if (!data->count)
1965                 return;
1966
1967         /* Skip if the event is in a state we want to switch to */
1968         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1969                 return;
1970
1971         if (data->count != -1)
1972                 (data->count)--;
1973
1974         event_enable_probe(ip, parent_ip, _data);
1975 }
1976
1977 static int
1978 event_enable_print(struct seq_file *m, unsigned long ip,
1979                       struct ftrace_probe_ops *ops, void *_data)
1980 {
1981         struct event_probe_data *data = _data;
1982
1983         seq_printf(m, "%ps:", (void *)ip);
1984
1985         seq_printf(m, "%s:%s:%s",
1986                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1987                    data->file->event_call->class->system,
1988                    ftrace_event_name(data->file->event_call));
1989
1990         if (data->count == -1)
1991                 seq_printf(m, ":unlimited\n");
1992         else
1993                 seq_printf(m, ":count=%ld\n", data->count);
1994
1995         return 0;
1996 }
1997
1998 static int
1999 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2000                   void **_data)
2001 {
2002         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2003         struct event_probe_data *data = *pdata;
2004
2005         data->ref++;
2006         return 0;
2007 }
2008
2009 static void
2010 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2011                   void **_data)
2012 {
2013         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2014         struct event_probe_data *data = *pdata;
2015
2016         if (WARN_ON_ONCE(data->ref <= 0))
2017                 return;
2018
2019         data->ref--;
2020         if (!data->ref) {
2021                 /* Remove the SOFT_MODE flag */
2022                 __ftrace_event_enable_disable(data->file, 0, 1);
2023                 module_put(data->file->event_call->mod);
2024                 kfree(data);
2025         }
2026         *pdata = NULL;
2027 }
2028
2029 static struct ftrace_probe_ops event_enable_probe_ops = {
2030         .func                   = event_enable_probe,
2031         .print                  = event_enable_print,
2032         .init                   = event_enable_init,
2033         .free                   = event_enable_free,
2034 };
2035
2036 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2037         .func                   = event_enable_count_probe,
2038         .print                  = event_enable_print,
2039         .init                   = event_enable_init,
2040         .free                   = event_enable_free,
2041 };
2042
2043 static struct ftrace_probe_ops event_disable_probe_ops = {
2044         .func                   = event_enable_probe,
2045         .print                  = event_enable_print,
2046         .init                   = event_enable_init,
2047         .free                   = event_enable_free,
2048 };
2049
2050 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2051         .func                   = event_enable_count_probe,
2052         .print                  = event_enable_print,
2053         .init                   = event_enable_init,
2054         .free                   = event_enable_free,
2055 };
2056
2057 static int
2058 event_enable_func(struct ftrace_hash *hash,
2059                   char *glob, char *cmd, char *param, int enabled)
2060 {
2061         struct trace_array *tr = top_trace_array();
2062         struct ftrace_event_file *file;
2063         struct ftrace_probe_ops *ops;
2064         struct event_probe_data *data;
2065         const char *system;
2066         const char *event;
2067         char *number;
2068         bool enable;
2069         int ret;
2070
2071         if (!tr)
2072                 return -ENODEV;
2073
2074         /* hash funcs only work with set_ftrace_filter */
2075         if (!enabled || !param)
2076                 return -EINVAL;
2077
2078         system = strsep(&param, ":");
2079         if (!param)
2080                 return -EINVAL;
2081
2082         event = strsep(&param, ":");
2083
2084         mutex_lock(&event_mutex);
2085
2086         ret = -EINVAL;
2087         file = find_event_file(tr, system, event);
2088         if (!file)
2089                 goto out;
2090
2091         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2092
2093         if (enable)
2094                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2095         else
2096                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2097
2098         if (glob[0] == '!') {
2099                 unregister_ftrace_function_probe_func(glob+1, ops);
2100                 ret = 0;
2101                 goto out;
2102         }
2103
2104         ret = -ENOMEM;
2105         data = kzalloc(sizeof(*data), GFP_KERNEL);
2106         if (!data)
2107                 goto out;
2108
2109         data->enable = enable;
2110         data->count = -1;
2111         data->file = file;
2112
2113         if (!param)
2114                 goto out_reg;
2115
2116         number = strsep(&param, ":");
2117
2118         ret = -EINVAL;
2119         if (!strlen(number))
2120                 goto out_free;
2121
2122         /*
2123          * We use the callback data field (which is a pointer)
2124          * as our counter.
2125          */
2126         ret = kstrtoul(number, 0, &data->count);
2127         if (ret)
2128                 goto out_free;
2129
2130  out_reg:
2131         /* Don't let event modules unload while probe registered */
2132         ret = try_module_get(file->event_call->mod);
2133         if (!ret) {
2134                 ret = -EBUSY;
2135                 goto out_free;
2136         }
2137
2138         ret = __ftrace_event_enable_disable(file, 1, 1);
2139         if (ret < 0)
2140                 goto out_put;
2141         ret = register_ftrace_function_probe(glob, ops, data);
2142         /*
2143          * The above returns on success the # of functions enabled,
2144          * but if it didn't find any functions it returns zero.
2145          * Consider no functions a failure too.
2146          */
2147         if (!ret) {
2148                 ret = -ENOENT;
2149                 goto out_disable;
2150         } else if (ret < 0)
2151                 goto out_disable;
2152         /* Just return zero, not the number of enabled functions */
2153         ret = 0;
2154  out:
2155         mutex_unlock(&event_mutex);
2156         return ret;
2157
2158  out_disable:
2159         __ftrace_event_enable_disable(file, 0, 1);
2160  out_put:
2161         module_put(file->event_call->mod);
2162  out_free:
2163         kfree(data);
2164         goto out;
2165 }
2166
2167 static struct ftrace_func_command event_enable_cmd = {
2168         .name                   = ENABLE_EVENT_STR,
2169         .func                   = event_enable_func,
2170 };
2171
2172 static struct ftrace_func_command event_disable_cmd = {
2173         .name                   = DISABLE_EVENT_STR,
2174         .func                   = event_enable_func,
2175 };
2176
2177 static __init int register_event_cmds(void)
2178 {
2179         int ret;
2180
2181         ret = register_ftrace_command(&event_enable_cmd);
2182         if (WARN_ON(ret < 0))
2183                 return ret;
2184         ret = register_ftrace_command(&event_disable_cmd);
2185         if (WARN_ON(ret < 0))
2186                 unregister_ftrace_command(&event_enable_cmd);
2187         return ret;
2188 }
2189 #else
2190 static inline int register_event_cmds(void) { return 0; }
2191 #endif /* CONFIG_DYNAMIC_FTRACE */
2192
2193 /*
2194  * The top level array has already had its ftrace_event_file
2195  * descriptors created in order to allow for early events to
2196  * be recorded. This function is called after the debugfs has been
2197  * initialized, and we now have to create the files associated
2198  * to the events.
2199  */
2200 static __init void
2201 __trace_early_add_event_dirs(struct trace_array *tr)
2202 {
2203         struct ftrace_event_file *file;
2204         int ret;
2205
2206
2207         list_for_each_entry(file, &tr->events, list) {
2208                 ret = event_create_dir(tr->event_dir, file);
2209                 if (ret < 0)
2210                         pr_warning("Could not create directory for event %s\n",
2211                                    ftrace_event_name(file->event_call));
2212         }
2213 }
2214
2215 /*
2216  * For early boot up, the top trace array requires to have
2217  * a list of events that can be enabled. This must be done before
2218  * the filesystem is set up in order to allow events to be traced
2219  * early.
2220  */
2221 static __init void
2222 __trace_early_add_events(struct trace_array *tr)
2223 {
2224         struct ftrace_event_call *call;
2225         int ret;
2226
2227         list_for_each_entry(call, &ftrace_events, list) {
2228                 /* Early boot up should not have any modules loaded */
2229                 if (WARN_ON_ONCE(call->mod))
2230                         continue;
2231
2232                 ret = __trace_early_add_new_event(call, tr);
2233                 if (ret < 0)
2234                         pr_warning("Could not create early event %s\n",
2235                                    ftrace_event_name(call));
2236         }
2237 }
2238
2239 /* Remove the event directory structure for a trace directory. */
2240 static void
2241 __trace_remove_event_dirs(struct trace_array *tr)
2242 {
2243         struct ftrace_event_file *file, *next;
2244
2245         list_for_each_entry_safe(file, next, &tr->events, list)
2246                 remove_event_file_dir(file);
2247 }
2248
2249 static void __add_event_to_tracers(struct ftrace_event_call *call)
2250 {
2251         struct trace_array *tr;
2252
2253         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2254                 __trace_add_new_event(call, tr);
2255 }
2256
2257 extern struct ftrace_event_call *__start_ftrace_events[];
2258 extern struct ftrace_event_call *__stop_ftrace_events[];
2259
2260 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2261
2262 static __init int setup_trace_event(char *str)
2263 {
2264         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2265         ring_buffer_expanded = true;
2266         tracing_selftest_disabled = true;
2267
2268         return 1;
2269 }
2270 __setup("trace_event=", setup_trace_event);
2271
2272 /* Expects to have event_mutex held when called */
2273 static int
2274 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2275 {
2276         struct dentry *d_events;
2277         struct dentry *entry;
2278
2279         entry = debugfs_create_file("set_event", 0644, parent,
2280                                     tr, &ftrace_set_event_fops);
2281         if (!entry) {
2282                 pr_warning("Could not create debugfs 'set_event' entry\n");
2283                 return -ENOMEM;
2284         }
2285
2286         d_events = debugfs_create_dir("events", parent);
2287         if (!d_events) {
2288                 pr_warning("Could not create debugfs 'events' directory\n");
2289                 return -ENOMEM;
2290         }
2291
2292         /* ring buffer internal formats */
2293         trace_create_file("header_page", 0444, d_events,
2294                           ring_buffer_print_page_header,
2295                           &ftrace_show_header_fops);
2296
2297         trace_create_file("header_event", 0444, d_events,
2298                           ring_buffer_print_entry_header,
2299                           &ftrace_show_header_fops);
2300
2301         trace_create_file("enable", 0644, d_events,
2302                           tr, &ftrace_tr_enable_fops);
2303
2304         tr->event_dir = d_events;
2305
2306         return 0;
2307 }
2308
2309 /**
2310  * event_trace_add_tracer - add a instance of a trace_array to events
2311  * @parent: The parent dentry to place the files/directories for events in
2312  * @tr: The trace array associated with these events
2313  *
2314  * When a new instance is created, it needs to set up its events
2315  * directory, as well as other files associated with events. It also
2316  * creates the event hierachry in the @parent/events directory.
2317  *
2318  * Returns 0 on success.
2319  */
2320 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2321 {
2322         int ret;
2323
2324         mutex_lock(&event_mutex);
2325
2326         ret = create_event_toplevel_files(parent, tr);
2327         if (ret)
2328                 goto out_unlock;
2329
2330         down_write(&trace_event_sem);
2331         __trace_add_event_dirs(tr);
2332         up_write(&trace_event_sem);
2333
2334  out_unlock:
2335         mutex_unlock(&event_mutex);
2336
2337         return ret;
2338 }
2339
2340 /*
2341  * The top trace array already had its file descriptors created.
2342  * Now the files themselves need to be created.
2343  */
2344 static __init int
2345 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2346 {
2347         int ret;
2348
2349         mutex_lock(&event_mutex);
2350
2351         ret = create_event_toplevel_files(parent, tr);
2352         if (ret)
2353                 goto out_unlock;
2354
2355         down_write(&trace_event_sem);
2356         __trace_early_add_event_dirs(tr);
2357         up_write(&trace_event_sem);
2358
2359  out_unlock:
2360         mutex_unlock(&event_mutex);
2361
2362         return ret;
2363 }
2364
2365 int event_trace_del_tracer(struct trace_array *tr)
2366 {
2367         mutex_lock(&event_mutex);
2368
2369         /* Disable any event triggers and associated soft-disabled events */
2370         clear_event_triggers(tr);
2371
2372         /* Disable any running events */
2373         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2374
2375         /* Access to events are within rcu_read_lock_sched() */
2376         synchronize_sched();
2377
2378         down_write(&trace_event_sem);
2379         __trace_remove_event_dirs(tr);
2380         debugfs_remove_recursive(tr->event_dir);
2381         up_write(&trace_event_sem);
2382
2383         tr->event_dir = NULL;
2384
2385         mutex_unlock(&event_mutex);
2386
2387         return 0;
2388 }
2389
2390 static __init int event_trace_memsetup(void)
2391 {
2392         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2393         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2394         return 0;
2395 }
2396
2397 static __init int event_trace_enable(void)
2398 {
2399         struct trace_array *tr = top_trace_array();
2400         struct ftrace_event_call **iter, *call;
2401         char *buf = bootup_event_buf;
2402         char *token;
2403         int ret;
2404
2405         if (!tr)
2406                 return -ENODEV;
2407
2408         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2409
2410                 call = *iter;
2411                 ret = event_init(call);
2412                 if (!ret)
2413                         list_add(&call->list, &ftrace_events);
2414         }
2415
2416         /*
2417          * We need the top trace array to have a working set of trace
2418          * points at early init, before the debug files and directories
2419          * are created. Create the file entries now, and attach them
2420          * to the actual file dentries later.
2421          */
2422         __trace_early_add_events(tr);
2423
2424         while (true) {
2425                 token = strsep(&buf, ",");
2426
2427                 if (!token)
2428                         break;
2429                 if (!*token)
2430                         continue;
2431
2432                 ret = ftrace_set_clr_event(tr, token, 1);
2433                 if (ret)
2434                         pr_warn("Failed to enable trace event: %s\n", token);
2435         }
2436
2437         trace_printk_start_comm();
2438
2439         register_event_cmds();
2440
2441         register_trigger_cmds();
2442
2443         return 0;
2444 }
2445
2446 static __init int event_trace_init(void)
2447 {
2448         struct trace_array *tr;
2449         struct dentry *d_tracer;
2450         struct dentry *entry;
2451         int ret;
2452
2453         tr = top_trace_array();
2454         if (!tr)
2455                 return -ENODEV;
2456
2457         d_tracer = tracing_init_dentry();
2458         if (!d_tracer)
2459                 return 0;
2460
2461         entry = debugfs_create_file("available_events", 0444, d_tracer,
2462                                     tr, &ftrace_avail_fops);
2463         if (!entry)
2464                 pr_warning("Could not create debugfs "
2465                            "'available_events' entry\n");
2466
2467         if (trace_define_common_fields())
2468                 pr_warning("tracing: Failed to allocate common fields");
2469
2470         ret = early_event_add_tracer(d_tracer, tr);
2471         if (ret)
2472                 return ret;
2473
2474 #ifdef CONFIG_MODULES
2475         ret = register_module_notifier(&trace_module_nb);
2476         if (ret)
2477                 pr_warning("Failed to register trace events module notifier\n");
2478 #endif
2479         return 0;
2480 }
2481 early_initcall(event_trace_memsetup);
2482 core_initcall(event_trace_enable);
2483 fs_initcall(event_trace_init);
2484
2485 #ifdef CONFIG_FTRACE_STARTUP_TEST
2486
2487 static DEFINE_SPINLOCK(test_spinlock);
2488 static DEFINE_SPINLOCK(test_spinlock_irq);
2489 static DEFINE_MUTEX(test_mutex);
2490
2491 static __init void test_work(struct work_struct *dummy)
2492 {
2493         spin_lock(&test_spinlock);
2494         spin_lock_irq(&test_spinlock_irq);
2495         udelay(1);
2496         spin_unlock_irq(&test_spinlock_irq);
2497         spin_unlock(&test_spinlock);
2498
2499         mutex_lock(&test_mutex);
2500         msleep(1);
2501         mutex_unlock(&test_mutex);
2502 }
2503
2504 static __init int event_test_thread(void *unused)
2505 {
2506         void *test_malloc;
2507
2508         test_malloc = kmalloc(1234, GFP_KERNEL);
2509         if (!test_malloc)
2510                 pr_info("failed to kmalloc\n");
2511
2512         schedule_on_each_cpu(test_work);
2513
2514         kfree(test_malloc);
2515
2516         set_current_state(TASK_INTERRUPTIBLE);
2517         while (!kthread_should_stop())
2518                 schedule();
2519
2520         return 0;
2521 }
2522
2523 /*
2524  * Do various things that may trigger events.
2525  */
2526 static __init void event_test_stuff(void)
2527 {
2528         struct task_struct *test_thread;
2529
2530         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2531         msleep(1);
2532         kthread_stop(test_thread);
2533 }
2534
2535 /*
2536  * For every trace event defined, we will test each trace point separately,
2537  * and then by groups, and finally all trace points.
2538  */
2539 static __init void event_trace_self_tests(void)
2540 {
2541         struct ftrace_subsystem_dir *dir;
2542         struct ftrace_event_file *file;
2543         struct ftrace_event_call *call;
2544         struct event_subsystem *system;
2545         struct trace_array *tr;
2546         int ret;
2547
2548         tr = top_trace_array();
2549         if (!tr)
2550                 return;
2551
2552         pr_info("Running tests on trace events:\n");
2553
2554         list_for_each_entry(file, &tr->events, list) {
2555
2556                 call = file->event_call;
2557
2558                 /* Only test those that have a probe */
2559                 if (!call->class || !call->class->probe)
2560                         continue;
2561
2562 /*
2563  * Testing syscall events here is pretty useless, but
2564  * we still do it if configured. But this is time consuming.
2565  * What we really need is a user thread to perform the
2566  * syscalls as we test.
2567  */
2568 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2569                 if (call->class->system &&
2570                     strcmp(call->class->system, "syscalls") == 0)
2571                         continue;
2572 #endif
2573
2574                 pr_info("Testing event %s: ", ftrace_event_name(call));
2575
2576                 /*
2577                  * If an event is already enabled, someone is using
2578                  * it and the self test should not be on.
2579                  */
2580                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2581                         pr_warning("Enabled event during self test!\n");
2582                         WARN_ON_ONCE(1);
2583                         continue;
2584                 }
2585
2586                 ftrace_event_enable_disable(file, 1);
2587                 event_test_stuff();
2588                 ftrace_event_enable_disable(file, 0);
2589
2590                 pr_cont("OK\n");
2591         }
2592
2593         /* Now test at the sub system level */
2594
2595         pr_info("Running tests on trace event systems:\n");
2596
2597         list_for_each_entry(dir, &tr->systems, list) {
2598
2599                 system = dir->subsystem;
2600
2601                 /* the ftrace system is special, skip it */
2602                 if (strcmp(system->name, "ftrace") == 0)
2603                         continue;
2604
2605                 pr_info("Testing event system %s: ", system->name);
2606
2607                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2608                 if (WARN_ON_ONCE(ret)) {
2609                         pr_warning("error enabling system %s\n",
2610                                    system->name);
2611                         continue;
2612                 }
2613
2614                 event_test_stuff();
2615
2616                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2617                 if (WARN_ON_ONCE(ret)) {
2618                         pr_warning("error disabling system %s\n",
2619                                    system->name);
2620                         continue;
2621                 }
2622
2623                 pr_cont("OK\n");
2624         }
2625
2626         /* Test with all events enabled */
2627
2628         pr_info("Running tests on all trace events:\n");
2629         pr_info("Testing all events: ");
2630
2631         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2632         if (WARN_ON_ONCE(ret)) {
2633                 pr_warning("error enabling all events\n");
2634                 return;
2635         }
2636
2637         event_test_stuff();
2638
2639         /* reset sysname */
2640         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2641         if (WARN_ON_ONCE(ret)) {
2642                 pr_warning("error disabling all events\n");
2643                 return;
2644         }
2645
2646         pr_cont("OK\n");
2647 }
2648
2649 #ifdef CONFIG_FUNCTION_TRACER
2650
2651 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2652
2653 static void
2654 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2655                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2656 {
2657         struct ring_buffer_event *event;
2658         struct ring_buffer *buffer;
2659         struct ftrace_entry *entry;
2660         unsigned long flags;
2661         long disabled;
2662         int cpu;
2663         int pc;
2664
2665         pc = preempt_count();
2666         preempt_disable_notrace();
2667         cpu = raw_smp_processor_id();
2668         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2669
2670         if (disabled != 1)
2671                 goto out;
2672
2673         local_save_flags(flags);
2674
2675         event = trace_current_buffer_lock_reserve(&buffer,
2676                                                   TRACE_FN, sizeof(*entry),
2677                                                   flags, pc);
2678         if (!event)
2679                 goto out;
2680         entry   = ring_buffer_event_data(event);
2681         entry->ip                       = ip;
2682         entry->parent_ip                = parent_ip;
2683
2684         trace_buffer_unlock_commit(buffer, event, flags, pc);
2685
2686  out:
2687         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2688         preempt_enable_notrace();
2689 }
2690
2691 static struct ftrace_ops trace_ops __initdata  =
2692 {
2693         .func = function_test_events_call,
2694         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2695 };
2696
2697 static __init void event_trace_self_test_with_function(void)
2698 {
2699         int ret;
2700         ret = register_ftrace_function(&trace_ops);
2701         if (WARN_ON(ret < 0)) {
2702                 pr_info("Failed to enable function tracer for event tests\n");
2703                 return;
2704         }
2705         pr_info("Running tests again, along with the function tracer\n");
2706         event_trace_self_tests();
2707         unregister_ftrace_function(&trace_ops);
2708 }
2709 #else
2710 static __init void event_trace_self_test_with_function(void)
2711 {
2712 }
2713 #endif
2714
2715 static __init int event_trace_self_tests_init(void)
2716 {
2717         if (!tracing_selftest_disabled) {
2718                 event_trace_self_tests();
2719                 event_trace_self_test_with_function();
2720         }
2721
2722         return 0;
2723 }
2724
2725 late_initcall(event_trace_self_tests_init);
2726
2727 #endif