kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/slab.h>
  19 #include <linux/delay.h>
  20
  21 #include <asm/setup.h>
  22
  23 #include "trace_output.h"
  24
  25 #undef TRACE_SYSTEM
  26 #define TRACE_SYSTEM "TRACE_SYSTEM"
  27
  28 DEFINE_MUTEX(event_mutex);
  29
  30 DEFINE_MUTEX(event_storage_mutex);
  31 EXPORT_SYMBOL_GPL(event_storage_mutex);
  32
  33 char event_storage[EVENT_STORAGE_SIZE];
  34 EXPORT_SYMBOL_GPL(event_storage);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_common_fields);
  38
  39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41 static struct kmem_cache *field_cachep;
  42 static struct kmem_cache *file_cachep;
  43
  44 #define SYSTEM_FL_FREE_NAME             (1 << 31)
  45
  46 static inline int system_refcount(struct event_subsystem *system)
  47 {
  48         return system->ref_count & ~SYSTEM_FL_FREE_NAME;
  49 }
  50
  51 static int system_refcount_inc(struct event_subsystem *system)
  52 {
  53         return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
  54 }
  55
  56 static int system_refcount_dec(struct event_subsystem *system)
  57 {
  58         return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
  59 }
  60
  61 /* Double loops, do not use break, only goto's work */
  62 #define do_for_each_event_file(tr, file)                        \
  63         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  64                 list_for_each_entry(file, &tr->events, list)
  65
  66 #define do_for_each_event_file_safe(tr, file)                   \
  67         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  68                 struct ftrace_event_file *___n;                         \
  69                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  70
  71 #define while_for_each_event_file()             \
  72         }
  73
  74 static struct list_head *
  75 trace_get_fields(struct ftrace_event_call *event_call)
  76 {
  77         if (!event_call->class->get_fields)
  78                 return &event_call->class->fields;
  79         return event_call->class->get_fields(event_call);
  80 }
  81
  82 static struct ftrace_event_field *
  83 __find_event_field(struct list_head *head, char *name)
  84 {
  85         struct ftrace_event_field *field;
  86
  87         list_for_each_entry(field, head, link) {
  88                 if (!strcmp(field->name, name))
  89                         return field;
  90         }
  91
  92         return NULL;
  93 }
  94
  95 struct ftrace_event_field *
  96 trace_find_event_field(struct ftrace_event_call *call, char *name)
  97 {
  98         struct ftrace_event_field *field;
  99         struct list_head *head;
 100
 101         field = __find_event_field(&ftrace_common_fields, name);
 102         if (field)
 103                 return field;
 104
 105         head = trace_get_fields(call);
 106         return __find_event_field(head, name);
 107 }
 108
 109 static int __trace_define_field(struct list_head *head, const char *type,
 110                                 const char *name, int offset, int size,
 111                                 int is_signed, int filter_type)
 112 {
 113         struct ftrace_event_field *field;
 114
 115         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 116         if (!field)
 117                 return -ENOMEM;
 118
 119         field->name = name;
 120         field->type = type;
 121
 122         if (filter_type == FILTER_OTHER)
 123                 field->filter_type = filter_assign_type(type);
 124         else
 125                 field->filter_type = filter_type;
 126
 127         field->offset = offset;
 128         field->size = size;
 129         field->is_signed = is_signed;
 130
 131         list_add(&field->link, head);
 132
 133         return 0;
 134 }
 135
 136 int trace_define_field(struct ftrace_event_call *call, const char *type,
 137                        const char *name, int offset, int size, int is_signed,
 138                        int filter_type)
 139 {
 140         struct list_head *head;
 141
 142         if (WARN_ON(!call->class))
 143                 return 0;
 144
 145         head = trace_get_fields(call);
 146         return __trace_define_field(head, type, name, offset, size,
 147                                     is_signed, filter_type);
 148 }
 149 EXPORT_SYMBOL_GPL(trace_define_field);
 150
 151 #define __common_field(type, item)                                      \
 152         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 153                                    "common_" #item,                     \
 154                                    offsetof(typeof(ent), item),         \
 155                                    sizeof(ent.item),                    \
 156                                    is_signed_type(type), FILTER_OTHER); \
 157         if (ret)                                                        \
 158                 return ret;
 159
 160 static int trace_define_common_fields(void)
 161 {
 162         int ret;
 163         struct trace_entry ent;
 164
 165         __common_field(unsigned short, type);
 166         __common_field(unsigned char, flags);
 167         __common_field(unsigned char, preempt_count);
 168         __common_field(int, pid);
 169
 170         return ret;
 171 }
 172
 173 static void trace_destroy_fields(struct ftrace_event_call *call)
 174 {
 175         struct ftrace_event_field *field, *next;
 176         struct list_head *head;
 177
 178         head = trace_get_fields(call);
 179         list_for_each_entry_safe(field, next, head, link) {
 180                 list_del(&field->link);
 181                 kmem_cache_free(field_cachep, field);
 182         }
 183 }
 184
 185 int trace_event_raw_init(struct ftrace_event_call *call)
 186 {
 187         int id;
 188
 189         id = register_ftrace_event(&call->event);
 190         if (!id)
 191                 return -ENODEV;
 192
 193         return 0;
 194 }
 195 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 196
 197 void *ftrace_event_buffer_reserve(struct ftrace_event_buffer *fbuffer,
 198                                   struct ftrace_event_file *ftrace_file,
 199                                   unsigned long len)
 200 {
 201         struct ftrace_event_call *event_call = ftrace_file->event_call;
 202
 203         local_save_flags(fbuffer->flags);
 204         fbuffer->pc = preempt_count();
 205         fbuffer->ftrace_file = ftrace_file;
 206
 207         fbuffer->event =
 208                 trace_event_buffer_lock_reserve(&fbuffer->buffer, ftrace_file,
 209                                                 event_call->event.type, len,
 210                                                 fbuffer->flags, fbuffer->pc);
 211         if (!fbuffer->event)
 212                 return NULL;
 213
 214         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 215         return fbuffer->entry;
 216 }
 217 EXPORT_SYMBOL_GPL(ftrace_event_buffer_reserve);
 218
 219 void ftrace_event_buffer_commit(struct ftrace_event_buffer *fbuffer)
 220 {
 221         event_trigger_unlock_commit(fbuffer->ftrace_file, fbuffer->buffer,
 222                                     fbuffer->event, fbuffer->entry,
 223                                     fbuffer->flags, fbuffer->pc);
 224 }
 225 EXPORT_SYMBOL_GPL(ftrace_event_buffer_commit);
 226
 227 int ftrace_event_reg(struct ftrace_event_call *call,
 228                      enum trace_reg type, void *data)
 229 {
 230         struct ftrace_event_file *file = data;
 231
 232         switch (type) {
 233         case TRACE_REG_REGISTER:
 234                 return tracepoint_probe_register(call->name,
 235                                                  call->class->probe,
 236                                                  file);
 237         case TRACE_REG_UNREGISTER:
 238                 tracepoint_probe_unregister(call->name,
 239                                             call->class->probe,
 240                                             file);
 241                 return 0;
 242
 243 #ifdef CONFIG_PERF_EVENTS
 244         case TRACE_REG_PERF_REGISTER:
 245                 return tracepoint_probe_register(call->name,
 246                                                  call->class->perf_probe,
 247                                                  call);
 248         case TRACE_REG_PERF_UNREGISTER:
 249                 tracepoint_probe_unregister(call->name,
 250                                             call->class->perf_probe,
 251                                             call);
 252                 return 0;
 253         case TRACE_REG_PERF_OPEN:
 254         case TRACE_REG_PERF_CLOSE:
 255         case TRACE_REG_PERF_ADD:
 256         case TRACE_REG_PERF_DEL:
 257                 return 0;
 258 #endif
 259         }
 260         return 0;
 261 }
 262 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 263
 264 void trace_event_enable_cmd_record(bool enable)
 265 {
 266         struct ftrace_event_file *file;
 267         struct trace_array *tr;
 268
 269         mutex_lock(&event_mutex);
 270         do_for_each_event_file(tr, file) {
 271
 272                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 273                         continue;
 274
 275                 if (enable) {
 276                         tracing_start_cmdline_record();
 277                         set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 278                 } else {
 279                         tracing_stop_cmdline_record();
 280                         clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 281                 }
 282         } while_for_each_event_file();
 283         mutex_unlock(&event_mutex);
 284 }
 285
 286 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
 287                                          int enable, int soft_disable)
 288 {
 289         struct ftrace_event_call *call = file->event_call;
 290         int ret = 0;
 291         int disable;
 292
 293         switch (enable) {
 294         case 0:
 295                 /*
 296                  * When soft_disable is set and enable is cleared, the sm_ref
 297                  * reference counter is decremented. If it reaches 0, we want
 298                  * to clear the SOFT_DISABLED flag but leave the event in the
 299                  * state that it was. That is, if the event was enabled and
 300                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 301                  * is set we do not want the event to be enabled before we
 302                  * clear the bit.
 303                  *
 304                  * When soft_disable is not set but the SOFT_MODE flag is,
 305                  * we do nothing. Do not disable the tracepoint, otherwise
 306                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 307                  */
 308                 if (soft_disable) {
 309                         if (atomic_dec_return(&file->sm_ref) > 0)
 310                                 break;
 311                         disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
 312                         clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 313                 } else
 314                         disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
 315
 316                 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
 317                         clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 318                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 319                                 tracing_stop_cmdline_record();
 320                                 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 321                         }
 322                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 323                 }
 324                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 325                 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 326                         set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 327                 else
 328                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 329                 break;
 330         case 1:
 331                 /*
 332                  * When soft_disable is set and enable is set, we want to
 333                  * register the tracepoint for the event, but leave the event
 334                  * as is. That means, if the event was already enabled, we do
 335                  * nothing (but set SOFT_MODE). If the event is disabled, we
 336                  * set SOFT_DISABLED before enabling the event tracepoint, so
 337                  * it still seems to be disabled.
 338                  */
 339                 if (!soft_disable)
 340                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 341                 else {
 342                         if (atomic_inc_return(&file->sm_ref) > 1)
 343                                 break;
 344                         set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 345                 }
 346
 347                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 348
 349                         /* Keep the event disabled, when going to SOFT_MODE. */
 350                         if (soft_disable)
 351                                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 352
 353                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 354                                 tracing_start_cmdline_record();
 355                                 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 356                         }
 357                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 358                         if (ret) {
 359                                 tracing_stop_cmdline_record();
 360                                 pr_info("event trace: Could not enable event "
 361                                         "%s\n", call->name);
 362                                 break;
 363                         }
 364                         set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 365
 366                         /* WAS_ENABLED gets set but never cleared. */
 367                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 368                 }
 369                 break;
 370         }
 371
 372         return ret;
 373 }
 374
 375 int trace_event_enable_disable(struct ftrace_event_file *file,
 376                                int enable, int soft_disable)
 377 {
 378         return __ftrace_event_enable_disable(file, enable, soft_disable);
 379 }
 380
 381 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 382                                        int enable)
 383 {
 384         return __ftrace_event_enable_disable(file, enable, 0);
 385 }
 386
 387 static void ftrace_clear_events(struct trace_array *tr)
 388 {
 389         struct ftrace_event_file *file;
 390
 391         mutex_lock(&event_mutex);
 392         list_for_each_entry(file, &tr->events, list) {
 393                 ftrace_event_enable_disable(file, 0);
 394         }
 395         mutex_unlock(&event_mutex);
 396 }
 397
 398 static void __put_system(struct event_subsystem *system)
 399 {
 400         struct event_filter *filter = system->filter;
 401
 402         WARN_ON_ONCE(system_refcount(system) == 0);
 403         if (system_refcount_dec(system))
 404                 return;
 405
 406         list_del(&system->list);
 407
 408         if (filter) {
 409                 kfree(filter->filter_string);
 410                 kfree(filter);
 411         }
 412         if (system->ref_count & SYSTEM_FL_FREE_NAME)
 413                 kfree(system->name);
 414         kfree(system);
 415 }
 416
 417 static void __get_system(struct event_subsystem *system)
 418 {
 419         WARN_ON_ONCE(system_refcount(system) == 0);
 420         system_refcount_inc(system);
 421 }
 422
 423 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 424 {
 425         WARN_ON_ONCE(dir->ref_count == 0);
 426         dir->ref_count++;
 427         __get_system(dir->subsystem);
 428 }
 429
 430 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 431 {
 432         WARN_ON_ONCE(dir->ref_count == 0);
 433         /* If the subsystem is about to be freed, the dir must be too */
 434         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 435
 436         __put_system(dir->subsystem);
 437         if (!--dir->ref_count)
 438                 kfree(dir);
 439 }
 440
 441 static void put_system(struct ftrace_subsystem_dir *dir)
 442 {
 443         mutex_lock(&event_mutex);
 444         __put_system_dir(dir);
 445         mutex_unlock(&event_mutex);
 446 }
 447
 448 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
 449 {
 450         if (!dir)
 451                 return;
 452
 453         if (!--dir->nr_events) {
 454                 debugfs_remove_recursive(dir->entry);
 455                 list_del(&dir->list);
 456                 __put_system_dir(dir);
 457         }
 458 }
 459
 460 static void remove_event_file_dir(struct ftrace_event_file *file)
 461 {
 462         struct dentry *dir = file->dir;
 463         struct dentry *child;
 464
 465         if (dir) {
 466                 spin_lock(&dir->d_lock);        /* probably unneeded */
 467                 list_for_each_entry(child, &dir->d_subdirs, d_u.d_child) {
 468                         if (child->d_inode)     /* probably unneeded */
 469                                 child->d_inode->i_private = NULL;
 470                 }
 471                 spin_unlock(&dir->d_lock);
 472
 473                 debugfs_remove_recursive(dir);
 474         }
 475
 476         list_del(&file->list);
 477         remove_subsystem(file->system);
 478         kmem_cache_free(file_cachep, file);
 479 }
 480
 481 /*
 482  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 483  */
 484 static int
 485 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 486                               const char *sub, const char *event, int set)
 487 {
 488         struct ftrace_event_file *file;
 489         struct ftrace_event_call *call;
 490         int ret = -EINVAL;
 491
 492         list_for_each_entry(file, &tr->events, list) {
 493
 494                 call = file->event_call;
 495
 496                 if (!call->name || !call->class || !call->class->reg)
 497                         continue;
 498
 499                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 500                         continue;
 501
 502                 if (match &&
 503                     strcmp(match, call->name) != 0 &&
 504                     strcmp(match, call->class->system) != 0)
 505                         continue;
 506
 507                 if (sub && strcmp(sub, call->class->system) != 0)
 508                         continue;
 509
 510                 if (event && strcmp(event, call->name) != 0)
 511                         continue;
 512
 513                 ftrace_event_enable_disable(file, set);
 514
 515                 ret = 0;
 516         }
 517
 518         return ret;
 519 }
 520
 521 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 522                                   const char *sub, const char *event, int set)
 523 {
 524         int ret;
 525
 526         mutex_lock(&event_mutex);
 527         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 528         mutex_unlock(&event_mutex);
 529
 530         return ret;
 531 }
 532
 533 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 534 {
 535         char *event = NULL, *sub = NULL, *match;
 536
 537         /*
 538          * The buf format can be <subsystem>:<event-name>
 539          *  *:<event-name> means any event by that name.
 540          *  :<event-name> is the same.
 541          *
 542          *  <subsystem>:* means all events in that subsystem
 543          *  <subsystem>: means the same.
 544          *
 545          *  <name> (no ':') means all events in a subsystem with
 546          *  the name <name> or any event that matches <name>
 547          */
 548
 549         match = strsep(&buf, ":");
 550         if (buf) {
 551                 sub = match;
 552                 event = buf;
 553                 match = NULL;
 554
 555                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 556                         sub = NULL;
 557                 if (!strlen(event) || strcmp(event, "*") == 0)
 558                         event = NULL;
 559         }
 560
 561         return __ftrace_set_clr_event(tr, match, sub, event, set);
 562 }
 563
 564 /**
 565  * trace_set_clr_event - enable or disable an event
 566  * @system: system name to match (NULL for any system)
 567  * @event: event name to match (NULL for all events, within system)
 568  * @set: 1 to enable, 0 to disable
 569  *
 570  * This is a way for other parts of the kernel to enable or disable
 571  * event recording.
 572  *
 573  * Returns 0 on success, -EINVAL if the parameters do not match any
 574  * registered events.
 575  */
 576 int trace_set_clr_event(const char *system, const char *event, int set)
 577 {
 578         struct trace_array *tr = top_trace_array();
 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", call->name);
 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 (!call->name || !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", call->name);
 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         int ret;
1537
1538         /*
1539          * If the trace point header did not define TRACE_SYSTEM
1540          * then the system would be called "TRACE_SYSTEM".
1541          */
1542         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1543                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1544                 if (!d_events)
1545                         return -ENOMEM;
1546         } else
1547                 d_events = parent;
1548
1549         file->dir = debugfs_create_dir(call->name, d_events);
1550         if (!file->dir) {
1551                 pr_warning("Could not create debugfs '%s' directory\n",
1552                            call->name);
1553                 return -1;
1554         }
1555
1556         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1557                 trace_create_file("enable", 0644, file->dir, file,
1558                                   &ftrace_enable_fops);
1559
1560 #ifdef CONFIG_PERF_EVENTS
1561         if (call->event.type && call->class->reg)
1562                 trace_create_file("id", 0444, file->dir,
1563                                   (void *)(long)call->event.type,
1564                                   &ftrace_event_id_fops);
1565 #endif
1566
1567         /*
1568          * Other events may have the same class. Only update
1569          * the fields if they are not already defined.
1570          */
1571         head = trace_get_fields(call);
1572         if (list_empty(head)) {
1573                 ret = call->class->define_fields(call);
1574                 if (ret < 0) {
1575                         pr_warning("Could not initialize trace point"
1576                                    " events/%s\n", call->name);
1577                         return -1;
1578                 }
1579         }
1580         trace_create_file("filter", 0644, file->dir, file,
1581                           &ftrace_event_filter_fops);
1582
1583         trace_create_file("trigger", 0644, file->dir, file,
1584                           &event_trigger_fops);
1585
1586         trace_create_file("format", 0444, file->dir, call,
1587                           &ftrace_event_format_fops);
1588
1589         return 0;
1590 }
1591
1592 static void remove_event_from_tracers(struct ftrace_event_call *call)
1593 {
1594         struct ftrace_event_file *file;
1595         struct trace_array *tr;
1596
1597         do_for_each_event_file_safe(tr, file) {
1598                 if (file->event_call != call)
1599                         continue;
1600
1601                 remove_event_file_dir(file);
1602                 /*
1603                  * The do_for_each_event_file_safe() is
1604                  * a double loop. After finding the call for this
1605                  * trace_array, we use break to jump to the next
1606                  * trace_array.
1607                  */
1608                 break;
1609         } while_for_each_event_file();
1610 }
1611
1612 static void event_remove(struct ftrace_event_call *call)
1613 {
1614         struct trace_array *tr;
1615         struct ftrace_event_file *file;
1616
1617         do_for_each_event_file(tr, file) {
1618                 if (file->event_call != call)
1619                         continue;
1620                 ftrace_event_enable_disable(file, 0);
1621                 destroy_preds(file);
1622                 /*
1623                  * The do_for_each_event_file() is
1624                  * a double loop. After finding the call for this
1625                  * trace_array, we use break to jump to the next
1626                  * trace_array.
1627                  */
1628                 break;
1629         } while_for_each_event_file();
1630
1631         if (call->event.funcs)
1632                 __unregister_ftrace_event(&call->event);
1633         remove_event_from_tracers(call);
1634         list_del(&call->list);
1635 }
1636
1637 static int event_init(struct ftrace_event_call *call)
1638 {
1639         int ret = 0;
1640
1641         if (WARN_ON(!call->name))
1642                 return -EINVAL;
1643
1644         if (call->class->raw_init) {
1645                 ret = call->class->raw_init(call);
1646                 if (ret < 0 && ret != -ENOSYS)
1647                         pr_warn("Could not initialize trace events/%s\n",
1648                                 call->name);
1649         }
1650
1651         return ret;
1652 }
1653
1654 static int
1655 __register_event(struct ftrace_event_call *call, struct module *mod)
1656 {
1657         int ret;
1658
1659         ret = event_init(call);
1660         if (ret < 0)
1661                 return ret;
1662
1663         list_add(&call->list, &ftrace_events);
1664         call->mod = mod;
1665
1666         return 0;
1667 }
1668
1669 static struct ftrace_event_file *
1670 trace_create_new_event(struct ftrace_event_call *call,
1671                        struct trace_array *tr)
1672 {
1673         struct ftrace_event_file *file;
1674
1675         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1676         if (!file)
1677                 return NULL;
1678
1679         file->event_call = call;
1680         file->tr = tr;
1681         atomic_set(&file->sm_ref, 0);
1682         atomic_set(&file->tm_ref, 0);
1683         INIT_LIST_HEAD(&file->triggers);
1684         list_add(&file->list, &tr->events);
1685
1686         return file;
1687 }
1688
1689 /* Add an event to a trace directory */
1690 static int
1691 __trace_add_new_event(struct ftrace_event_call *call, struct trace_array *tr)
1692 {
1693         struct ftrace_event_file *file;
1694
1695         file = trace_create_new_event(call, tr);
1696         if (!file)
1697                 return -ENOMEM;
1698
1699         return event_create_dir(tr->event_dir, file);
1700 }
1701
1702 /*
1703  * Just create a decriptor for early init. A descriptor is required
1704  * for enabling events at boot. We want to enable events before
1705  * the filesystem is initialized.
1706  */
1707 static __init int
1708 __trace_early_add_new_event(struct ftrace_event_call *call,
1709                             struct trace_array *tr)
1710 {
1711         struct ftrace_event_file *file;
1712
1713         file = trace_create_new_event(call, tr);
1714         if (!file)
1715                 return -ENOMEM;
1716
1717         return 0;
1718 }
1719
1720 struct ftrace_module_file_ops;
1721 static void __add_event_to_tracers(struct ftrace_event_call *call);
1722
1723 /* Add an additional event_call dynamically */
1724 int trace_add_event_call(struct ftrace_event_call *call)
1725 {
1726         int ret;
1727         mutex_lock(&trace_types_lock);
1728         mutex_lock(&event_mutex);
1729
1730         ret = __register_event(call, NULL);
1731         if (ret >= 0)
1732                 __add_event_to_tracers(call);
1733
1734         mutex_unlock(&event_mutex);
1735         mutex_unlock(&trace_types_lock);
1736         return ret;
1737 }
1738
1739 /*
1740  * Must be called under locking of trace_types_lock, event_mutex and
1741  * trace_event_sem.
1742  */
1743 static void __trace_remove_event_call(struct ftrace_event_call *call)
1744 {
1745         event_remove(call);
1746         trace_destroy_fields(call);
1747         destroy_call_preds(call);
1748 }
1749
1750 static int probe_remove_event_call(struct ftrace_event_call *call)
1751 {
1752         struct trace_array *tr;
1753         struct ftrace_event_file *file;
1754
1755 #ifdef CONFIG_PERF_EVENTS
1756         if (call->perf_refcount)
1757                 return -EBUSY;
1758 #endif
1759         do_for_each_event_file(tr, file) {
1760                 if (file->event_call != call)
1761                         continue;
1762                 /*
1763                  * We can't rely on ftrace_event_enable_disable(enable => 0)
1764                  * we are going to do, FTRACE_EVENT_FL_SOFT_MODE can suppress
1765                  * TRACE_REG_UNREGISTER.
1766                  */
1767                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
1768                         return -EBUSY;
1769                 /*
1770                  * The do_for_each_event_file_safe() is
1771                  * a double loop. After finding the call for this
1772                  * trace_array, we use break to jump to the next
1773                  * trace_array.
1774                  */
1775                 break;
1776         } while_for_each_event_file();
1777
1778         __trace_remove_event_call(call);
1779
1780         return 0;
1781 }
1782
1783 /* Remove an event_call */
1784 int trace_remove_event_call(struct ftrace_event_call *call)
1785 {
1786         int ret;
1787
1788         mutex_lock(&trace_types_lock);
1789         mutex_lock(&event_mutex);
1790         down_write(&trace_event_sem);
1791         ret = probe_remove_event_call(call);
1792         up_write(&trace_event_sem);
1793         mutex_unlock(&event_mutex);
1794         mutex_unlock(&trace_types_lock);
1795
1796         return ret;
1797 }
1798
1799 #define for_each_event(event, start, end)                       \
1800         for (event = start;                                     \
1801              (unsigned long)event < (unsigned long)end;         \
1802              event++)
1803
1804 #ifdef CONFIG_MODULES
1805
1806 static void trace_module_add_events(struct module *mod)
1807 {
1808         struct ftrace_event_call **call, **start, **end;
1809
1810         start = mod->trace_events;
1811         end = mod->trace_events + mod->num_trace_events;
1812
1813         for_each_event(call, start, end) {
1814                 __register_event(*call, mod);
1815                 __add_event_to_tracers(*call);
1816         }
1817 }
1818
1819 static void trace_module_remove_events(struct module *mod)
1820 {
1821         struct ftrace_event_call *call, *p;
1822         bool clear_trace = false;
1823
1824         down_write(&trace_event_sem);
1825         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1826                 if (call->mod == mod) {
1827                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1828                                 clear_trace = true;
1829                         __trace_remove_event_call(call);
1830                 }
1831         }
1832         up_write(&trace_event_sem);
1833
1834         /*
1835          * It is safest to reset the ring buffer if the module being unloaded
1836          * registered any events that were used. The only worry is if
1837          * a new module gets loaded, and takes on the same id as the events
1838          * of this module. When printing out the buffer, traced events left
1839          * over from this module may be passed to the new module events and
1840          * unexpected results may occur.
1841          */
1842         if (clear_trace)
1843                 tracing_reset_all_online_cpus();
1844 }
1845
1846 static int trace_module_notify(struct notifier_block *self,
1847                                unsigned long val, void *data)
1848 {
1849         struct module *mod = data;
1850
1851         mutex_lock(&trace_types_lock);
1852         mutex_lock(&event_mutex);
1853         switch (val) {
1854         case MODULE_STATE_COMING:
1855                 trace_module_add_events(mod);
1856                 break;
1857         case MODULE_STATE_GOING:
1858                 trace_module_remove_events(mod);
1859                 break;
1860         }
1861         mutex_unlock(&event_mutex);
1862         mutex_unlock(&trace_types_lock);
1863
1864         return 0;
1865 }
1866
1867 static struct notifier_block trace_module_nb = {
1868         .notifier_call = trace_module_notify,
1869         .priority = 0,
1870 };
1871 #endif /* CONFIG_MODULES */
1872
1873 /* Create a new event directory structure for a trace directory. */
1874 static void
1875 __trace_add_event_dirs(struct trace_array *tr)
1876 {
1877         struct ftrace_event_call *call;
1878         int ret;
1879
1880         list_for_each_entry(call, &ftrace_events, list) {
1881                 ret = __trace_add_new_event(call, tr);
1882                 if (ret < 0)
1883                         pr_warning("Could not create directory for event %s\n",
1884                                    call->name);
1885         }
1886 }
1887
1888 struct ftrace_event_file *
1889 find_event_file(struct trace_array *tr, const char *system,  const char *event)
1890 {
1891         struct ftrace_event_file *file;
1892         struct ftrace_event_call *call;
1893
1894         list_for_each_entry(file, &tr->events, list) {
1895
1896                 call = file->event_call;
1897
1898                 if (!call->name || !call->class || !call->class->reg)
1899                         continue;
1900
1901                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1902                         continue;
1903
1904                 if (strcmp(event, call->name) == 0 &&
1905                     strcmp(system, call->class->system) == 0)
1906                         return file;
1907         }
1908         return NULL;
1909 }
1910
1911 #ifdef CONFIG_DYNAMIC_FTRACE
1912
1913 /* Avoid typos */
1914 #define ENABLE_EVENT_STR        "enable_event"
1915 #define DISABLE_EVENT_STR       "disable_event"
1916
1917 struct event_probe_data {
1918         struct ftrace_event_file        *file;
1919         unsigned long                   count;
1920         int                             ref;
1921         bool                            enable;
1922 };
1923
1924 static void
1925 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1926 {
1927         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1928         struct event_probe_data *data = *pdata;
1929
1930         if (!data)
1931                 return;
1932
1933         if (data->enable)
1934                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1935         else
1936                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1937 }
1938
1939 static void
1940 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1941 {
1942         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1943         struct event_probe_data *data = *pdata;
1944
1945         if (!data)
1946                 return;
1947
1948         if (!data->count)
1949                 return;
1950
1951         /* Skip if the event is in a state we want to switch to */
1952         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1953                 return;
1954
1955         if (data->count != -1)
1956                 (data->count)--;
1957
1958         event_enable_probe(ip, parent_ip, _data);
1959 }
1960
1961 static int
1962 event_enable_print(struct seq_file *m, unsigned long ip,
1963                       struct ftrace_probe_ops *ops, void *_data)
1964 {
1965         struct event_probe_data *data = _data;
1966
1967         seq_printf(m, "%ps:", (void *)ip);
1968
1969         seq_printf(m, "%s:%s:%s",
1970                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1971                    data->file->event_call->class->system,
1972                    data->file->event_call->name);
1973
1974         if (data->count == -1)
1975                 seq_printf(m, ":unlimited\n");
1976         else
1977                 seq_printf(m, ":count=%ld\n", data->count);
1978
1979         return 0;
1980 }
1981
1982 static int
1983 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
1984                   void **_data)
1985 {
1986         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1987         struct event_probe_data *data = *pdata;
1988
1989         data->ref++;
1990         return 0;
1991 }
1992
1993 static void
1994 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
1995                   void **_data)
1996 {
1997         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1998         struct event_probe_data *data = *pdata;
1999
2000         if (WARN_ON_ONCE(data->ref <= 0))
2001                 return;
2002
2003         data->ref--;
2004         if (!data->ref) {
2005                 /* Remove the SOFT_MODE flag */
2006                 __ftrace_event_enable_disable(data->file, 0, 1);
2007                 module_put(data->file->event_call->mod);
2008                 kfree(data);
2009         }
2010         *pdata = NULL;
2011 }
2012
2013 static struct ftrace_probe_ops event_enable_probe_ops = {
2014         .func                   = event_enable_probe,
2015         .print                  = event_enable_print,
2016         .init                   = event_enable_init,
2017         .free                   = event_enable_free,
2018 };
2019
2020 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2021         .func                   = event_enable_count_probe,
2022         .print                  = event_enable_print,
2023         .init                   = event_enable_init,
2024         .free                   = event_enable_free,
2025 };
2026
2027 static struct ftrace_probe_ops event_disable_probe_ops = {
2028         .func                   = event_enable_probe,
2029         .print                  = event_enable_print,
2030         .init                   = event_enable_init,
2031         .free                   = event_enable_free,
2032 };
2033
2034 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2035         .func                   = event_enable_count_probe,
2036         .print                  = event_enable_print,
2037         .init                   = event_enable_init,
2038         .free                   = event_enable_free,
2039 };
2040
2041 static int
2042 event_enable_func(struct ftrace_hash *hash,
2043                   char *glob, char *cmd, char *param, int enabled)
2044 {
2045         struct trace_array *tr = top_trace_array();
2046         struct ftrace_event_file *file;
2047         struct ftrace_probe_ops *ops;
2048         struct event_probe_data *data;
2049         const char *system;
2050         const char *event;
2051         char *number;
2052         bool enable;
2053         int ret;
2054
2055         /* hash funcs only work with set_ftrace_filter */
2056         if (!enabled || !param)
2057                 return -EINVAL;
2058
2059         system = strsep(&param, ":");
2060         if (!param)
2061                 return -EINVAL;
2062
2063         event = strsep(&param, ":");
2064
2065         mutex_lock(&event_mutex);
2066
2067         ret = -EINVAL;
2068         file = find_event_file(tr, system, event);
2069         if (!file)
2070                 goto out;
2071
2072         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2073
2074         if (enable)
2075                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2076         else
2077                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2078
2079         if (glob[0] == '!') {
2080                 unregister_ftrace_function_probe_func(glob+1, ops);
2081                 ret = 0;
2082                 goto out;
2083         }
2084
2085         ret = -ENOMEM;
2086         data = kzalloc(sizeof(*data), GFP_KERNEL);
2087         if (!data)
2088                 goto out;
2089
2090         data->enable = enable;
2091         data->count = -1;
2092         data->file = file;
2093
2094         if (!param)
2095                 goto out_reg;
2096
2097         number = strsep(&param, ":");
2098
2099         ret = -EINVAL;
2100         if (!strlen(number))
2101                 goto out_free;
2102
2103         /*
2104          * We use the callback data field (which is a pointer)
2105          * as our counter.
2106          */
2107         ret = kstrtoul(number, 0, &data->count);
2108         if (ret)
2109                 goto out_free;
2110
2111  out_reg:
2112         /* Don't let event modules unload while probe registered */
2113         ret = try_module_get(file->event_call->mod);
2114         if (!ret) {
2115                 ret = -EBUSY;
2116                 goto out_free;
2117         }
2118
2119         ret = __ftrace_event_enable_disable(file, 1, 1);
2120         if (ret < 0)
2121                 goto out_put;
2122         ret = register_ftrace_function_probe(glob, ops, data);
2123         /*
2124          * The above returns on success the # of functions enabled,
2125          * but if it didn't find any functions it returns zero.
2126          * Consider no functions a failure too.
2127          */
2128         if (!ret) {
2129                 ret = -ENOENT;
2130                 goto out_disable;
2131         } else if (ret < 0)
2132                 goto out_disable;
2133         /* Just return zero, not the number of enabled functions */
2134         ret = 0;
2135  out:
2136         mutex_unlock(&event_mutex);
2137         return ret;
2138
2139  out_disable:
2140         __ftrace_event_enable_disable(file, 0, 1);
2141  out_put:
2142         module_put(file->event_call->mod);
2143  out_free:
2144         kfree(data);
2145         goto out;
2146 }
2147
2148 static struct ftrace_func_command event_enable_cmd = {
2149         .name                   = ENABLE_EVENT_STR,
2150         .func                   = event_enable_func,
2151 };
2152
2153 static struct ftrace_func_command event_disable_cmd = {
2154         .name                   = DISABLE_EVENT_STR,
2155         .func                   = event_enable_func,
2156 };
2157
2158 static __init int register_event_cmds(void)
2159 {
2160         int ret;
2161
2162         ret = register_ftrace_command(&event_enable_cmd);
2163         if (WARN_ON(ret < 0))
2164                 return ret;
2165         ret = register_ftrace_command(&event_disable_cmd);
2166         if (WARN_ON(ret < 0))
2167                 unregister_ftrace_command(&event_enable_cmd);
2168         return ret;
2169 }
2170 #else
2171 static inline int register_event_cmds(void) { return 0; }
2172 #endif /* CONFIG_DYNAMIC_FTRACE */
2173
2174 /*
2175  * The top level array has already had its ftrace_event_file
2176  * descriptors created in order to allow for early events to
2177  * be recorded. This function is called after the debugfs has been
2178  * initialized, and we now have to create the files associated
2179  * to the events.
2180  */
2181 static __init void
2182 __trace_early_add_event_dirs(struct trace_array *tr)
2183 {
2184         struct ftrace_event_file *file;
2185         int ret;
2186
2187
2188         list_for_each_entry(file, &tr->events, list) {
2189                 ret = event_create_dir(tr->event_dir, file);
2190                 if (ret < 0)
2191                         pr_warning("Could not create directory for event %s\n",
2192                                    file->event_call->name);
2193         }
2194 }
2195
2196 /*
2197  * For early boot up, the top trace array requires to have
2198  * a list of events that can be enabled. This must be done before
2199  * the filesystem is set up in order to allow events to be traced
2200  * early.
2201  */
2202 static __init void
2203 __trace_early_add_events(struct trace_array *tr)
2204 {
2205         struct ftrace_event_call *call;
2206         int ret;
2207
2208         list_for_each_entry(call, &ftrace_events, list) {
2209                 /* Early boot up should not have any modules loaded */
2210                 if (WARN_ON_ONCE(call->mod))
2211                         continue;
2212
2213                 ret = __trace_early_add_new_event(call, tr);
2214                 if (ret < 0)
2215                         pr_warning("Could not create early event %s\n",
2216                                    call->name);
2217         }
2218 }
2219
2220 /* Remove the event directory structure for a trace directory. */
2221 static void
2222 __trace_remove_event_dirs(struct trace_array *tr)
2223 {
2224         struct ftrace_event_file *file, *next;
2225
2226         list_for_each_entry_safe(file, next, &tr->events, list)
2227                 remove_event_file_dir(file);
2228 }
2229
2230 static void __add_event_to_tracers(struct ftrace_event_call *call)
2231 {
2232         struct trace_array *tr;
2233
2234         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2235                 __trace_add_new_event(call, tr);
2236 }
2237
2238 extern struct ftrace_event_call *__start_ftrace_events[];
2239 extern struct ftrace_event_call *__stop_ftrace_events[];
2240
2241 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2242
2243 static __init int setup_trace_event(char *str)
2244 {
2245         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2246         ring_buffer_expanded = true;
2247         tracing_selftest_disabled = true;
2248
2249         return 1;
2250 }
2251 __setup("trace_event=", setup_trace_event);
2252
2253 /* Expects to have event_mutex held when called */
2254 static int
2255 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2256 {
2257         struct dentry *d_events;
2258         struct dentry *entry;
2259
2260         entry = debugfs_create_file("set_event", 0644, parent,
2261                                     tr, &ftrace_set_event_fops);
2262         if (!entry) {
2263                 pr_warning("Could not create debugfs 'set_event' entry\n");
2264                 return -ENOMEM;
2265         }
2266
2267         d_events = debugfs_create_dir("events", parent);
2268         if (!d_events) {
2269                 pr_warning("Could not create debugfs 'events' directory\n");
2270                 return -ENOMEM;
2271         }
2272
2273         /* ring buffer internal formats */
2274         trace_create_file("header_page", 0444, d_events,
2275                           ring_buffer_print_page_header,
2276                           &ftrace_show_header_fops);
2277
2278         trace_create_file("header_event", 0444, d_events,
2279                           ring_buffer_print_entry_header,
2280                           &ftrace_show_header_fops);
2281
2282         trace_create_file("enable", 0644, d_events,
2283                           tr, &ftrace_tr_enable_fops);
2284
2285         tr->event_dir = d_events;
2286
2287         return 0;
2288 }
2289
2290 /**
2291  * event_trace_add_tracer - add a instance of a trace_array to events
2292  * @parent: The parent dentry to place the files/directories for events in
2293  * @tr: The trace array associated with these events
2294  *
2295  * When a new instance is created, it needs to set up its events
2296  * directory, as well as other files associated with events. It also
2297  * creates the event hierachry in the @parent/events directory.
2298  *
2299  * Returns 0 on success.
2300  */
2301 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2302 {
2303         int ret;
2304
2305         mutex_lock(&event_mutex);
2306
2307         ret = create_event_toplevel_files(parent, tr);
2308         if (ret)
2309                 goto out_unlock;
2310
2311         down_write(&trace_event_sem);
2312         __trace_add_event_dirs(tr);
2313         up_write(&trace_event_sem);
2314
2315  out_unlock:
2316         mutex_unlock(&event_mutex);
2317
2318         return ret;
2319 }
2320
2321 /*
2322  * The top trace array already had its file descriptors created.
2323  * Now the files themselves need to be created.
2324  */
2325 static __init int
2326 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2327 {
2328         int ret;
2329
2330         mutex_lock(&event_mutex);
2331
2332         ret = create_event_toplevel_files(parent, tr);
2333         if (ret)
2334                 goto out_unlock;
2335
2336         down_write(&trace_event_sem);
2337         __trace_early_add_event_dirs(tr);
2338         up_write(&trace_event_sem);
2339
2340  out_unlock:
2341         mutex_unlock(&event_mutex);
2342
2343         return ret;
2344 }
2345
2346 int event_trace_del_tracer(struct trace_array *tr)
2347 {
2348         mutex_lock(&event_mutex);
2349
2350         /* Disable any event triggers and associated soft-disabled events */
2351         clear_event_triggers(tr);
2352
2353         /* Disable any running events */
2354         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2355
2356         /* Access to events are within rcu_read_lock_sched() */
2357         synchronize_sched();
2358
2359         down_write(&trace_event_sem);
2360         __trace_remove_event_dirs(tr);
2361         debugfs_remove_recursive(tr->event_dir);
2362         up_write(&trace_event_sem);
2363
2364         tr->event_dir = NULL;
2365
2366         mutex_unlock(&event_mutex);
2367
2368         return 0;
2369 }
2370
2371 static __init int event_trace_memsetup(void)
2372 {
2373         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2374         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2375         return 0;
2376 }
2377
2378 static __init int event_trace_enable(void)
2379 {
2380         struct trace_array *tr = top_trace_array();
2381         struct ftrace_event_call **iter, *call;
2382         char *buf = bootup_event_buf;
2383         char *token;
2384         int ret;
2385
2386         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2387
2388                 call = *iter;
2389                 ret = event_init(call);
2390                 if (!ret)
2391                         list_add(&call->list, &ftrace_events);
2392         }
2393
2394         /*
2395          * We need the top trace array to have a working set of trace
2396          * points at early init, before the debug files and directories
2397          * are created. Create the file entries now, and attach them
2398          * to the actual file dentries later.
2399          */
2400         __trace_early_add_events(tr);
2401
2402         while (true) {
2403                 token = strsep(&buf, ",");
2404
2405                 if (!token)
2406                         break;
2407                 if (!*token)
2408                         continue;
2409
2410                 ret = ftrace_set_clr_event(tr, token, 1);
2411                 if (ret)
2412                         pr_warn("Failed to enable trace event: %s\n", token);
2413         }
2414
2415         trace_printk_start_comm();
2416
2417         register_event_cmds();
2418
2419         register_trigger_cmds();
2420
2421         return 0;
2422 }
2423
2424 static __init int event_trace_init(void)
2425 {
2426         struct trace_array *tr;
2427         struct dentry *d_tracer;
2428         struct dentry *entry;
2429         int ret;
2430
2431         tr = top_trace_array();
2432
2433         d_tracer = tracing_init_dentry();
2434         if (!d_tracer)
2435                 return 0;
2436
2437         entry = debugfs_create_file("available_events", 0444, d_tracer,
2438                                     tr, &ftrace_avail_fops);
2439         if (!entry)
2440                 pr_warning("Could not create debugfs "
2441                            "'available_events' entry\n");
2442
2443         if (trace_define_common_fields())
2444                 pr_warning("tracing: Failed to allocate common fields");
2445
2446         ret = early_event_add_tracer(d_tracer, tr);
2447         if (ret)
2448                 return ret;
2449
2450 #ifdef CONFIG_MODULES
2451         ret = register_module_notifier(&trace_module_nb);
2452         if (ret)
2453                 pr_warning("Failed to register trace events module notifier\n");
2454 #endif
2455         return 0;
2456 }
2457 early_initcall(event_trace_memsetup);
2458 core_initcall(event_trace_enable);
2459 fs_initcall(event_trace_init);
2460
2461 #ifdef CONFIG_FTRACE_STARTUP_TEST
2462
2463 static DEFINE_SPINLOCK(test_spinlock);
2464 static DEFINE_SPINLOCK(test_spinlock_irq);
2465 static DEFINE_MUTEX(test_mutex);
2466
2467 static __init void test_work(struct work_struct *dummy)
2468 {
2469         spin_lock(&test_spinlock);
2470         spin_lock_irq(&test_spinlock_irq);
2471         udelay(1);
2472         spin_unlock_irq(&test_spinlock_irq);
2473         spin_unlock(&test_spinlock);
2474
2475         mutex_lock(&test_mutex);
2476         msleep(1);
2477         mutex_unlock(&test_mutex);
2478 }
2479
2480 static __init int event_test_thread(void *unused)
2481 {
2482         void *test_malloc;
2483
2484         test_malloc = kmalloc(1234, GFP_KERNEL);
2485         if (!test_malloc)
2486                 pr_info("failed to kmalloc\n");
2487
2488         schedule_on_each_cpu(test_work);
2489
2490         kfree(test_malloc);
2491
2492         set_current_state(TASK_INTERRUPTIBLE);
2493         while (!kthread_should_stop())
2494                 schedule();
2495
2496         return 0;
2497 }
2498
2499 /*
2500  * Do various things that may trigger events.
2501  */
2502 static __init void event_test_stuff(void)
2503 {
2504         struct task_struct *test_thread;
2505
2506         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2507         msleep(1);
2508         kthread_stop(test_thread);
2509 }
2510
2511 /*
2512  * For every trace event defined, we will test each trace point separately,
2513  * and then by groups, and finally all trace points.
2514  */
2515 static __init void event_trace_self_tests(void)
2516 {
2517         struct ftrace_subsystem_dir *dir;
2518         struct ftrace_event_file *file;
2519         struct ftrace_event_call *call;
2520         struct event_subsystem *system;
2521         struct trace_array *tr;
2522         int ret;
2523
2524         tr = top_trace_array();
2525
2526         pr_info("Running tests on trace events:\n");
2527
2528         list_for_each_entry(file, &tr->events, list) {
2529
2530                 call = file->event_call;
2531
2532                 /* Only test those that have a probe */
2533                 if (!call->class || !call->class->probe)
2534                         continue;
2535
2536 /*
2537  * Testing syscall events here is pretty useless, but
2538  * we still do it if configured. But this is time consuming.
2539  * What we really need is a user thread to perform the
2540  * syscalls as we test.
2541  */
2542 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2543                 if (call->class->system &&
2544                     strcmp(call->class->system, "syscalls") == 0)
2545                         continue;
2546 #endif
2547
2548                 pr_info("Testing event %s: ", call->name);
2549
2550                 /*
2551                  * If an event is already enabled, someone is using
2552                  * it and the self test should not be on.
2553                  */
2554                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2555                         pr_warning("Enabled event during self test!\n");
2556                         WARN_ON_ONCE(1);
2557                         continue;
2558                 }
2559
2560                 ftrace_event_enable_disable(file, 1);
2561                 event_test_stuff();
2562                 ftrace_event_enable_disable(file, 0);
2563
2564                 pr_cont("OK\n");
2565         }
2566
2567         /* Now test at the sub system level */
2568
2569         pr_info("Running tests on trace event systems:\n");
2570
2571         list_for_each_entry(dir, &tr->systems, list) {
2572
2573                 system = dir->subsystem;
2574
2575                 /* the ftrace system is special, skip it */
2576                 if (strcmp(system->name, "ftrace") == 0)
2577                         continue;
2578
2579                 pr_info("Testing event system %s: ", system->name);
2580
2581                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2582                 if (WARN_ON_ONCE(ret)) {
2583                         pr_warning("error enabling system %s\n",
2584                                    system->name);
2585                         continue;
2586                 }
2587
2588                 event_test_stuff();
2589
2590                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2591                 if (WARN_ON_ONCE(ret)) {
2592                         pr_warning("error disabling system %s\n",
2593                                    system->name);
2594                         continue;
2595                 }
2596
2597                 pr_cont("OK\n");
2598         }
2599
2600         /* Test with all events enabled */
2601
2602         pr_info("Running tests on all trace events:\n");
2603         pr_info("Testing all events: ");
2604
2605         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2606         if (WARN_ON_ONCE(ret)) {
2607                 pr_warning("error enabling all events\n");
2608                 return;
2609         }
2610
2611         event_test_stuff();
2612
2613         /* reset sysname */
2614         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2615         if (WARN_ON_ONCE(ret)) {
2616                 pr_warning("error disabling all events\n");
2617                 return;
2618         }
2619
2620         pr_cont("OK\n");
2621 }
2622
2623 #ifdef CONFIG_FUNCTION_TRACER
2624
2625 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2626
2627 static void
2628 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2629                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2630 {
2631         struct ring_buffer_event *event;
2632         struct ring_buffer *buffer;
2633         struct ftrace_entry *entry;
2634         unsigned long flags;
2635         long disabled;
2636         int cpu;
2637         int pc;
2638
2639         pc = preempt_count();
2640         preempt_disable_notrace();
2641         cpu = raw_smp_processor_id();
2642         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2643
2644         if (disabled != 1)
2645                 goto out;
2646
2647         local_save_flags(flags);
2648
2649         event = trace_current_buffer_lock_reserve(&buffer,
2650                                                   TRACE_FN, sizeof(*entry),
2651                                                   flags, pc);
2652         if (!event)
2653                 goto out;
2654         entry   = ring_buffer_event_data(event);
2655         entry->ip                       = ip;
2656         entry->parent_ip                = parent_ip;
2657
2658         trace_buffer_unlock_commit(buffer, event, flags, pc);
2659
2660  out:
2661         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2662         preempt_enable_notrace();
2663 }
2664
2665 static struct ftrace_ops trace_ops __initdata  =
2666 {
2667         .func = function_test_events_call,
2668         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2669 };
2670
2671 static __init void event_trace_self_test_with_function(void)
2672 {
2673         int ret;
2674         ret = register_ftrace_function(&trace_ops);
2675         if (WARN_ON(ret < 0)) {
2676                 pr_info("Failed to enable function tracer for event tests\n");
2677                 return;
2678         }
2679         pr_info("Running tests again, along with the function tracer\n");
2680         event_trace_self_tests();
2681         unregister_ftrace_function(&trace_ops);
2682 }
2683 #else
2684 static __init void event_trace_self_test_with_function(void)
2685 {
2686 }
2687 #endif
2688
2689 static __init int event_trace_self_tests_init(void)
2690 {
2691         if (!tracing_selftest_disabled) {
2692                 event_trace_self_tests();
2693                 event_trace_self_test_with_function();
2694         }
2695
2696         return 0;
2697 }
2698
2699 late_initcall(event_trace_self_tests_init);
2700
2701 #endif