kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <[email protected]>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <[email protected]>.
   8  *
   9  */
  10
  11 #define pr_fmt(fmt) fmt
  12
  13 #include <linux/workqueue.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kthread.h>
  16 #include <linux/tracefs.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/module.h>
  19 #include <linux/ctype.h>
  20 #include <linux/sort.h>
  21 #include <linux/slab.h>
  22 #include <linux/delay.h>
  23
  24 #include <trace/events/sched.h>
  25
  26 #include <asm/setup.h>
  27
  28 #include "trace_output.h"
  29
  30 #undef TRACE_SYSTEM
  31 #define TRACE_SYSTEM "TRACE_SYSTEM"
  32
  33 DEFINE_MUTEX(event_mutex);
  34
  35 LIST_HEAD(ftrace_events);
  36 static LIST_HEAD(ftrace_generic_fields);
  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 static inline int system_refcount(struct event_subsystem *system)
  45 {
  46         return system->ref_count;
  47 }
  48
  49 static int system_refcount_inc(struct event_subsystem *system)
  50 {
  51         return system->ref_count++;
  52 }
  53
  54 static int system_refcount_dec(struct event_subsystem *system)
  55 {
  56         return --system->ref_count;
  57 }
  58
  59 /* Double loops, do not use break, only goto's work */
  60 #define do_for_each_event_file(tr, file)                        \
  61         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  62                 list_for_each_entry(file, &tr->events, list)
  63
  64 #define do_for_each_event_file_safe(tr, file)                   \
  65         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  66                 struct trace_event_file *___n;                          \
  67                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  68
  69 #define while_for_each_event_file()             \
  70         }
  71
  72 static struct list_head *
  73 trace_get_fields(struct trace_event_call *event_call)
  74 {
  75         if (!event_call->class->get_fields)
  76                 return &event_call->class->fields;
  77         return event_call->class->get_fields(event_call);
  78 }
  79
  80 static struct ftrace_event_field *
  81 __find_event_field(struct list_head *head, char *name)
  82 {
  83         struct ftrace_event_field *field;
  84
  85         list_for_each_entry(field, head, link) {
  86                 if (!strcmp(field->name, name))
  87                         return field;
  88         }
  89
  90         return NULL;
  91 }
  92
  93 struct ftrace_event_field *
  94 trace_find_event_field(struct trace_event_call *call, char *name)
  95 {
  96         struct ftrace_event_field *field;
  97         struct list_head *head;
  98
  99         head = trace_get_fields(call);
 100         field = __find_event_field(head, name);
 101         if (field)
 102                 return field;
 103
 104         field = __find_event_field(&ftrace_generic_fields, name);
 105         if (field)
 106                 return field;
 107
 108         return __find_event_field(&ftrace_common_fields, name);
 109 }
 110
 111 static int __trace_define_field(struct list_head *head, const char *type,
 112                                 const char *name, int offset, int size,
 113                                 int is_signed, int filter_type)
 114 {
 115         struct ftrace_event_field *field;
 116
 117         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 118         if (!field)
 119                 return -ENOMEM;
 120
 121         field->name = name;
 122         field->type = type;
 123
 124         if (filter_type == FILTER_OTHER)
 125                 field->filter_type = filter_assign_type(type);
 126         else
 127                 field->filter_type = filter_type;
 128
 129         field->offset = offset;
 130         field->size = size;
 131         field->is_signed = is_signed;
 132
 133         list_add(&field->link, head);
 134
 135         return 0;
 136 }
 137
 138 int trace_define_field(struct trace_event_call *call, const char *type,
 139                        const char *name, int offset, int size, int is_signed,
 140                        int filter_type)
 141 {
 142         struct list_head *head;
 143
 144         if (WARN_ON(!call->class))
 145                 return 0;
 146
 147         head = trace_get_fields(call);
 148         return __trace_define_field(head, type, name, offset, size,
 149                                     is_signed, filter_type);
 150 }
 151 EXPORT_SYMBOL_GPL(trace_define_field);
 152
 153 #define __generic_field(type, item, filter_type)                        \
 154         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 155                                    #item, 0, 0, is_signed_type(type),   \
 156                                    filter_type);                        \
 157         if (ret)                                                        \
 158                 return ret;
 159
 160 #define __common_field(type, item)                                      \
 161         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 162                                    "common_" #item,                     \
 163                                    offsetof(typeof(ent), item),         \
 164                                    sizeof(ent.item),                    \
 165                                    is_signed_type(type), FILTER_OTHER); \
 166         if (ret)                                                        \
 167                 return ret;
 168
 169 static int trace_define_generic_fields(void)
 170 {
 171         int ret;
 172
 173         __generic_field(int, CPU, FILTER_CPU);
 174         __generic_field(int, cpu, FILTER_CPU);
 175         __generic_field(char *, COMM, FILTER_COMM);
 176         __generic_field(char *, comm, FILTER_COMM);
 177
 178         return ret;
 179 }
 180
 181 static int trace_define_common_fields(void)
 182 {
 183         int ret;
 184         struct trace_entry ent;
 185
 186         __common_field(unsigned short, type);
 187         __common_field(unsigned char, flags);
 188         __common_field(unsigned char, preempt_count);
 189         __common_field(int, pid);
 190
 191         return ret;
 192 }
 193
 194 static void trace_destroy_fields(struct trace_event_call *call)
 195 {
 196         struct ftrace_event_field *field, *next;
 197         struct list_head *head;
 198
 199         head = trace_get_fields(call);
 200         list_for_each_entry_safe(field, next, head, link) {
 201                 list_del(&field->link);
 202                 kmem_cache_free(field_cachep, field);
 203         }
 204 }
 205
 206 /*
 207  * run-time version of trace_event_get_offsets_<call>() that returns the last
 208  * accessible offset of trace fields excluding __dynamic_array bytes
 209  */
 210 int trace_event_get_offsets(struct trace_event_call *call)
 211 {
 212         struct ftrace_event_field *tail;
 213         struct list_head *head;
 214
 215         head = trace_get_fields(call);
 216         /*
 217          * head->next points to the last field with the largest offset,
 218          * since it was added last by trace_define_field()
 219          */
 220         tail = list_first_entry(head, struct ftrace_event_field, link);
 221         return tail->offset + tail->size;
 222 }
 223
 224 int trace_event_raw_init(struct trace_event_call *call)
 225 {
 226         int id;
 227
 228         id = register_trace_event(&call->event);
 229         if (!id)
 230                 return -ENODEV;
 231
 232         return 0;
 233 }
 234 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 235
 236 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 237 {
 238         struct trace_array *tr = trace_file->tr;
 239         struct trace_array_cpu *data;
 240         struct trace_pid_list *pid_list;
 241
 242         pid_list = rcu_dereference_sched(tr->filtered_pids);
 243         if (!pid_list)
 244                 return false;
 245
 246         data = this_cpu_ptr(tr->trace_buffer.data);
 247
 248         return data->ignore_pid;
 249 }
 250 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 251
 252 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 253                                  struct trace_event_file *trace_file,
 254                                  unsigned long len)
 255 {
 256         struct trace_event_call *event_call = trace_file->event_call;
 257
 258         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 259             trace_event_ignore_this_pid(trace_file))
 260                 return NULL;
 261
 262         local_save_flags(fbuffer->flags);
 263         fbuffer->pc = preempt_count();
 264         /*
 265          * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
 266          * preemption (adding one to the preempt_count). Since we are
 267          * interested in the preempt_count at the time the tracepoint was
 268          * hit, we need to subtract one to offset the increment.
 269          */
 270         if (IS_ENABLED(CONFIG_PREEMPT))
 271                 fbuffer->pc--;
 272         fbuffer->trace_file = trace_file;
 273
 274         fbuffer->event =
 275                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 276                                                 event_call->event.type, len,
 277                                                 fbuffer->flags, fbuffer->pc);
 278         if (!fbuffer->event)
 279                 return NULL;
 280
 281         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 282         return fbuffer->entry;
 283 }
 284 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 285
 286 int trace_event_reg(struct trace_event_call *call,
 287                     enum trace_reg type, void *data)
 288 {
 289         struct trace_event_file *file = data;
 290
 291         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 292         switch (type) {
 293         case TRACE_REG_REGISTER:
 294                 return tracepoint_probe_register(call->tp,
 295                                                  call->class->probe,
 296                                                  file);
 297         case TRACE_REG_UNREGISTER:
 298                 tracepoint_probe_unregister(call->tp,
 299                                             call->class->probe,
 300                                             file);
 301                 return 0;
 302
 303 #ifdef CONFIG_PERF_EVENTS
 304         case TRACE_REG_PERF_REGISTER:
 305                 return tracepoint_probe_register(call->tp,
 306                                                  call->class->perf_probe,
 307                                                  call);
 308         case TRACE_REG_PERF_UNREGISTER:
 309                 tracepoint_probe_unregister(call->tp,
 310                                             call->class->perf_probe,
 311                                             call);
 312                 return 0;
 313         case TRACE_REG_PERF_OPEN:
 314         case TRACE_REG_PERF_CLOSE:
 315         case TRACE_REG_PERF_ADD:
 316         case TRACE_REG_PERF_DEL:
 317                 return 0;
 318 #endif
 319         }
 320         return 0;
 321 }
 322 EXPORT_SYMBOL_GPL(trace_event_reg);
 323
 324 void trace_event_enable_cmd_record(bool enable)
 325 {
 326         struct trace_event_file *file;
 327         struct trace_array *tr;
 328
 329         mutex_lock(&event_mutex);
 330         do_for_each_event_file(tr, file) {
 331
 332                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 333                         continue;
 334
 335                 if (enable) {
 336                         tracing_start_cmdline_record();
 337                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 338                 } else {
 339                         tracing_stop_cmdline_record();
 340                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 341                 }
 342         } while_for_each_event_file();
 343         mutex_unlock(&event_mutex);
 344 }
 345
 346 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 347                                          int enable, int soft_disable)
 348 {
 349         struct trace_event_call *call = file->event_call;
 350         struct trace_array *tr = file->tr;
 351         unsigned long file_flags = file->flags;
 352         int ret = 0;
 353         int disable;
 354
 355         switch (enable) {
 356         case 0:
 357                 /*
 358                  * When soft_disable is set and enable is cleared, the sm_ref
 359                  * reference counter is decremented. If it reaches 0, we want
 360                  * to clear the SOFT_DISABLED flag but leave the event in the
 361                  * state that it was. That is, if the event was enabled and
 362                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 363                  * is set we do not want the event to be enabled before we
 364                  * clear the bit.
 365                  *
 366                  * When soft_disable is not set but the SOFT_MODE flag is,
 367                  * we do nothing. Do not disable the tracepoint, otherwise
 368                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 369                  */
 370                 if (soft_disable) {
 371                         if (atomic_dec_return(&file->sm_ref) > 0)
 372                                 break;
 373                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 374                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 375                 } else
 376                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 377
 378                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 379                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 380                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 381                                 tracing_stop_cmdline_record();
 382                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 383                         }
 384                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 385                 }
 386                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 387                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 388                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 389                 else
 390                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 391                 break;
 392         case 1:
 393                 /*
 394                  * When soft_disable is set and enable is set, we want to
 395                  * register the tracepoint for the event, but leave the event
 396                  * as is. That means, if the event was already enabled, we do
 397                  * nothing (but set SOFT_MODE). If the event is disabled, we
 398                  * set SOFT_DISABLED before enabling the event tracepoint, so
 399                  * it still seems to be disabled.
 400                  */
 401                 if (!soft_disable)
 402                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 403                 else {
 404                         if (atomic_inc_return(&file->sm_ref) > 1)
 405                                 break;
 406                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 407                 }
 408
 409                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 410
 411                         /* Keep the event disabled, when going to SOFT_MODE. */
 412                         if (soft_disable)
 413                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 414
 415                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 416                                 tracing_start_cmdline_record();
 417                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 418                         }
 419                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 420                         if (ret) {
 421                                 tracing_stop_cmdline_record();
 422                                 pr_info("event trace: Could not enable event "
 423                                         "%s\n", trace_event_name(call));
 424                                 break;
 425                         }
 426                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 427
 428                         /* WAS_ENABLED gets set but never cleared. */
 429                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 430                 }
 431                 break;
 432         }
 433
 434         /* Enable or disable use of trace_buffered_event */
 435         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 436             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 437                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 438                         trace_buffered_event_enable();
 439                 else
 440                         trace_buffered_event_disable();
 441         }
 442
 443         return ret;
 444 }
 445
 446 int trace_event_enable_disable(struct trace_event_file *file,
 447                                int enable, int soft_disable)
 448 {
 449         return __ftrace_event_enable_disable(file, enable, soft_disable);
 450 }
 451
 452 static int ftrace_event_enable_disable(struct trace_event_file *file,
 453                                        int enable)
 454 {
 455         return __ftrace_event_enable_disable(file, enable, 0);
 456 }
 457
 458 static void ftrace_clear_events(struct trace_array *tr)
 459 {
 460         struct trace_event_file *file;
 461
 462         mutex_lock(&event_mutex);
 463         list_for_each_entry(file, &tr->events, list) {
 464                 ftrace_event_enable_disable(file, 0);
 465         }
 466         mutex_unlock(&event_mutex);
 467 }
 468
 469 static void
 470 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 471 {
 472         struct trace_pid_list *pid_list;
 473         struct trace_array *tr = data;
 474
 475         pid_list = rcu_dereference_sched(tr->filtered_pids);
 476         trace_filter_add_remove_task(pid_list, NULL, task);
 477 }
 478
 479 static void
 480 event_filter_pid_sched_process_fork(void *data,
 481                                     struct task_struct *self,
 482                                     struct task_struct *task)
 483 {
 484         struct trace_pid_list *pid_list;
 485         struct trace_array *tr = data;
 486
 487         pid_list = rcu_dereference_sched(tr->filtered_pids);
 488         trace_filter_add_remove_task(pid_list, self, task);
 489 }
 490
 491 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 492 {
 493         if (enable) {
 494                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 495                                                        tr, INT_MIN);
 496                 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 497                                                        tr, INT_MAX);
 498         } else {
 499                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 500                                                     tr);
 501                 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 502                                                     tr);
 503         }
 504 }
 505
 506 static void
 507 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 508                     struct task_struct *prev, struct task_struct *next)
 509 {
 510         struct trace_array *tr = data;
 511         struct trace_pid_list *pid_list;
 512
 513         pid_list = rcu_dereference_sched(tr->filtered_pids);
 514
 515         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 516                        trace_ignore_this_task(pid_list, prev) &&
 517                        trace_ignore_this_task(pid_list, next));
 518 }
 519
 520 static void
 521 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 522                     struct task_struct *prev, struct task_struct *next)
 523 {
 524         struct trace_array *tr = data;
 525         struct trace_pid_list *pid_list;
 526
 527         pid_list = rcu_dereference_sched(tr->filtered_pids);
 528
 529         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 530                        trace_ignore_this_task(pid_list, next));
 531 }
 532
 533 static void
 534 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 535 {
 536         struct trace_array *tr = data;
 537         struct trace_pid_list *pid_list;
 538
 539         /* Nothing to do if we are already tracing */
 540         if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 541                 return;
 542
 543         pid_list = rcu_dereference_sched(tr->filtered_pids);
 544
 545         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 546                        trace_ignore_this_task(pid_list, task));
 547 }
 548
 549 static void
 550 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 551 {
 552         struct trace_array *tr = data;
 553         struct trace_pid_list *pid_list;
 554
 555         /* Nothing to do if we are not tracing */
 556         if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 557                 return;
 558
 559         pid_list = rcu_dereference_sched(tr->filtered_pids);
 560
 561         /* Set tracing if current is enabled */
 562         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 563                        trace_ignore_this_task(pid_list, current));
 564 }
 565
 566 static void __ftrace_clear_event_pids(struct trace_array *tr)
 567 {
 568         struct trace_pid_list *pid_list;
 569         struct trace_event_file *file;
 570         int cpu;
 571
 572         pid_list = rcu_dereference_protected(tr->filtered_pids,
 573                                              lockdep_is_held(&event_mutex));
 574         if (!pid_list)
 575                 return;
 576
 577         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 578         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 579
 580         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 581         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 582
 583         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 584         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 585
 586         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 587         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 588
 589         list_for_each_entry(file, &tr->events, list) {
 590                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 591         }
 592
 593         for_each_possible_cpu(cpu)
 594                 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 595
 596         rcu_assign_pointer(tr->filtered_pids, NULL);
 597
 598         /* Wait till all users are no longer using pid filtering */
 599         synchronize_sched();
 600
 601         trace_free_pid_list(pid_list);
 602 }
 603
 604 static void ftrace_clear_event_pids(struct trace_array *tr)
 605 {
 606         mutex_lock(&event_mutex);
 607         __ftrace_clear_event_pids(tr);
 608         mutex_unlock(&event_mutex);
 609 }
 610
 611 static void __put_system(struct event_subsystem *system)
 612 {
 613         struct event_filter *filter = system->filter;
 614
 615         WARN_ON_ONCE(system_refcount(system) == 0);
 616         if (system_refcount_dec(system))
 617                 return;
 618
 619         list_del(&system->list);
 620
 621         if (filter) {
 622                 kfree(filter->filter_string);
 623                 kfree(filter);
 624         }
 625         kfree_const(system->name);
 626         kfree(system);
 627 }
 628
 629 static void __get_system(struct event_subsystem *system)
 630 {
 631         WARN_ON_ONCE(system_refcount(system) == 0);
 632         system_refcount_inc(system);
 633 }
 634
 635 static void __get_system_dir(struct trace_subsystem_dir *dir)
 636 {
 637         WARN_ON_ONCE(dir->ref_count == 0);
 638         dir->ref_count++;
 639         __get_system(dir->subsystem);
 640 }
 641
 642 static void __put_system_dir(struct trace_subsystem_dir *dir)
 643 {
 644         WARN_ON_ONCE(dir->ref_count == 0);
 645         /* If the subsystem is about to be freed, the dir must be too */
 646         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 647
 648         __put_system(dir->subsystem);
 649         if (!--dir->ref_count)
 650                 kfree(dir);
 651 }
 652
 653 static void put_system(struct trace_subsystem_dir *dir)
 654 {
 655         mutex_lock(&event_mutex);
 656         __put_system_dir(dir);
 657         mutex_unlock(&event_mutex);
 658 }
 659
 660 static void remove_subsystem(struct trace_subsystem_dir *dir)
 661 {
 662         if (!dir)
 663                 return;
 664
 665         if (!--dir->nr_events) {
 666                 tracefs_remove_recursive(dir->entry);
 667                 list_del(&dir->list);
 668                 __put_system_dir(dir);
 669         }
 670 }
 671
 672 static void remove_event_file_dir(struct trace_event_file *file)
 673 {
 674         struct dentry *dir = file->dir;
 675         struct dentry *child;
 676
 677         if (dir) {
 678                 spin_lock(&dir->d_lock);        /* probably unneeded */
 679                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 680                         if (d_really_is_positive(child))        /* probably unneeded */
 681                                 d_inode(child)->i_private = NULL;
 682                 }
 683                 spin_unlock(&dir->d_lock);
 684
 685                 tracefs_remove_recursive(dir);
 686         }
 687
 688         list_del(&file->list);
 689         remove_subsystem(file->system);
 690         free_event_filter(file->filter);
 691         kmem_cache_free(file_cachep, file);
 692 }
 693
 694 /*
 695  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 696  */
 697 static int
 698 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 699                               const char *sub, const char *event, int set)
 700 {
 701         struct trace_event_file *file;
 702         struct trace_event_call *call;
 703         const char *name;
 704         int ret = -EINVAL;
 705         int eret = 0;
 706
 707         list_for_each_entry(file, &tr->events, list) {
 708
 709                 call = file->event_call;
 710                 name = trace_event_name(call);
 711
 712                 if (!name || !call->class || !call->class->reg)
 713                         continue;
 714
 715                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 716                         continue;
 717
 718                 if (match &&
 719                     strcmp(match, name) != 0 &&
 720                     strcmp(match, call->class->system) != 0)
 721                         continue;
 722
 723                 if (sub && strcmp(sub, call->class->system) != 0)
 724                         continue;
 725
 726                 if (event && strcmp(event, name) != 0)
 727                         continue;
 728
 729                 ret = ftrace_event_enable_disable(file, set);
 730
 731                 /*
 732                  * Save the first error and return that. Some events
 733                  * may still have been enabled, but let the user
 734                  * know that something went wrong.
 735                  */
 736                 if (ret && !eret)
 737                         eret = ret;
 738
 739                 ret = eret;
 740         }
 741
 742         return ret;
 743 }
 744
 745 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 746                                   const char *sub, const char *event, int set)
 747 {
 748         int ret;
 749
 750         mutex_lock(&event_mutex);
 751         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 752         mutex_unlock(&event_mutex);
 753
 754         return ret;
 755 }
 756
 757 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 758 {
 759         char *event = NULL, *sub = NULL, *match;
 760         int ret;
 761
 762         /*
 763          * The buf format can be <subsystem>:<event-name>
 764          *  *:<event-name> means any event by that name.
 765          *  :<event-name> is the same.
 766          *
 767          *  <subsystem>:* means all events in that subsystem
 768          *  <subsystem>: means the same.
 769          *
 770          *  <name> (no ':') means all events in a subsystem with
 771          *  the name <name> or any event that matches <name>
 772          */
 773
 774         match = strsep(&buf, ":");
 775         if (buf) {
 776                 sub = match;
 777                 event = buf;
 778                 match = NULL;
 779
 780                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 781                         sub = NULL;
 782                 if (!strlen(event) || strcmp(event, "*") == 0)
 783                         event = NULL;
 784         }
 785
 786         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 787
 788         /* Put back the colon to allow this to be called again */
 789         if (buf)
 790                 *(buf - 1) = ':';
 791
 792         return ret;
 793 }
 794
 795 /**
 796  * trace_set_clr_event - enable or disable an event
 797  * @system: system name to match (NULL for any system)
 798  * @event: event name to match (NULL for all events, within system)
 799  * @set: 1 to enable, 0 to disable
 800  *
 801  * This is a way for other parts of the kernel to enable or disable
 802  * event recording.
 803  *
 804  * Returns 0 on success, -EINVAL if the parameters do not match any
 805  * registered events.
 806  */
 807 int trace_set_clr_event(const char *system, const char *event, int set)
 808 {
 809         struct trace_array *tr = top_trace_array();
 810
 811         if (!tr)
 812                 return -ENODEV;
 813
 814         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 815 }
 816 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 817
 818 /* 128 should be much more than enough */
 819 #define EVENT_BUF_SIZE          127
 820
 821 static ssize_t
 822 ftrace_event_write(struct file *file, const char __user *ubuf,
 823                    size_t cnt, loff_t *ppos)
 824 {
 825         struct trace_parser parser;
 826         struct seq_file *m = file->private_data;
 827         struct trace_array *tr = m->private;
 828         ssize_t read, ret;
 829
 830         if (!cnt)
 831                 return 0;
 832
 833         ret = tracing_update_buffers();
 834         if (ret < 0)
 835                 return ret;
 836
 837         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 838                 return -ENOMEM;
 839
 840         read = trace_get_user(&parser, ubuf, cnt, ppos);
 841
 842         if (read >= 0 && trace_parser_loaded((&parser))) {
 843                 int set = 1;
 844
 845                 if (*parser.buffer == '!')
 846                         set = 0;
 847
 848                 parser.buffer[parser.idx] = 0;
 849
 850                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 851                 if (ret)
 852                         goto out_put;
 853         }
 854
 855         ret = read;
 856
 857  out_put:
 858         trace_parser_put(&parser);
 859
 860         return ret;
 861 }
 862
 863 static void *
 864 t_next(struct seq_file *m, void *v, loff_t *pos)
 865 {
 866         struct trace_event_file *file = v;
 867         struct trace_event_call *call;
 868         struct trace_array *tr = m->private;
 869
 870         (*pos)++;
 871
 872         list_for_each_entry_continue(file, &tr->events, list) {
 873                 call = file->event_call;
 874                 /*
 875                  * The ftrace subsystem is for showing formats only.
 876                  * They can not be enabled or disabled via the event files.
 877                  */
 878                 if (call->class && call->class->reg &&
 879                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 880                         return file;
 881         }
 882
 883         return NULL;
 884 }
 885
 886 static void *t_start(struct seq_file *m, loff_t *pos)
 887 {
 888         struct trace_event_file *file;
 889         struct trace_array *tr = m->private;
 890         loff_t l;
 891
 892         mutex_lock(&event_mutex);
 893
 894         file = list_entry(&tr->events, struct trace_event_file, list);
 895         for (l = 0; l <= *pos; ) {
 896                 file = t_next(m, file, &l);
 897                 if (!file)
 898                         break;
 899         }
 900         return file;
 901 }
 902
 903 static void *
 904 s_next(struct seq_file *m, void *v, loff_t *pos)
 905 {
 906         struct trace_event_file *file = v;
 907         struct trace_array *tr = m->private;
 908
 909         (*pos)++;
 910
 911         list_for_each_entry_continue(file, &tr->events, list) {
 912                 if (file->flags & EVENT_FILE_FL_ENABLED)
 913                         return file;
 914         }
 915
 916         return NULL;
 917 }
 918
 919 static void *s_start(struct seq_file *m, loff_t *pos)
 920 {
 921         struct trace_event_file *file;
 922         struct trace_array *tr = m->private;
 923         loff_t l;
 924
 925         mutex_lock(&event_mutex);
 926
 927         file = list_entry(&tr->events, struct trace_event_file, list);
 928         for (l = 0; l <= *pos; ) {
 929                 file = s_next(m, file, &l);
 930                 if (!file)
 931                         break;
 932         }
 933         return file;
 934 }
 935
 936 static int t_show(struct seq_file *m, void *v)
 937 {
 938         struct trace_event_file *file = v;
 939         struct trace_event_call *call = file->event_call;
 940
 941         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 942                 seq_printf(m, "%s:", call->class->system);
 943         seq_printf(m, "%s\n", trace_event_name(call));
 944
 945         return 0;
 946 }
 947
 948 static void t_stop(struct seq_file *m, void *p)
 949 {
 950         mutex_unlock(&event_mutex);
 951 }
 952
 953 static void *
 954 p_next(struct seq_file *m, void *v, loff_t *pos)
 955 {
 956         struct trace_array *tr = m->private;
 957         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 958
 959         return trace_pid_next(pid_list, v, pos);
 960 }
 961
 962 static void *p_start(struct seq_file *m, loff_t *pos)
 963         __acquires(RCU)
 964 {
 965         struct trace_pid_list *pid_list;
 966         struct trace_array *tr = m->private;
 967
 968         /*
 969          * Grab the mutex, to keep calls to p_next() having the same
 970          * tr->filtered_pids as p_start() has.
 971          * If we just passed the tr->filtered_pids around, then RCU would
 972          * have been enough, but doing that makes things more complex.
 973          */
 974         mutex_lock(&event_mutex);
 975         rcu_read_lock_sched();
 976
 977         pid_list = rcu_dereference_sched(tr->filtered_pids);
 978
 979         if (!pid_list)
 980                 return NULL;
 981
 982         return trace_pid_start(pid_list, pos);
 983 }
 984
 985 static void p_stop(struct seq_file *m, void *p)
 986         __releases(RCU)
 987 {
 988         rcu_read_unlock_sched();
 989         mutex_unlock(&event_mutex);
 990 }
 991
 992 static ssize_t
 993 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 994                   loff_t *ppos)
 995 {
 996         struct trace_event_file *file;
 997         unsigned long flags;
 998         char buf[4] = "0";
 999
1000         mutex_lock(&event_mutex);
1001         file = event_file_data(filp);
1002         if (likely(file))
1003                 flags = file->flags;
1004         mutex_unlock(&event_mutex);
1005
1006         if (!file)
1007                 return -ENODEV;
1008
1009         if (flags & EVENT_FILE_FL_ENABLED &&
1010             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1011                 strcpy(buf, "1");
1012
1013         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1014             flags & EVENT_FILE_FL_SOFT_MODE)
1015                 strcat(buf, "*");
1016
1017         strcat(buf, "\n");
1018
1019         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1020 }
1021
1022 static ssize_t
1023 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1024                    loff_t *ppos)
1025 {
1026         struct trace_event_file *file;
1027         unsigned long val;
1028         int ret;
1029
1030         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1031         if (ret)
1032                 return ret;
1033
1034         ret = tracing_update_buffers();
1035         if (ret < 0)
1036                 return ret;
1037
1038         switch (val) {
1039         case 0:
1040         case 1:
1041                 ret = -ENODEV;
1042                 mutex_lock(&event_mutex);
1043                 file = event_file_data(filp);
1044                 if (likely(file))
1045                         ret = ftrace_event_enable_disable(file, val);
1046                 mutex_unlock(&event_mutex);
1047                 break;
1048
1049         default:
1050                 return -EINVAL;
1051         }
1052
1053         *ppos += cnt;
1054
1055         return ret ? ret : cnt;
1056 }
1057
1058 static ssize_t
1059 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1060                    loff_t *ppos)
1061 {
1062         const char set_to_char[4] = { '?', '0', '1', 'X' };
1063         struct trace_subsystem_dir *dir = filp->private_data;
1064         struct event_subsystem *system = dir->subsystem;
1065         struct trace_event_call *call;
1066         struct trace_event_file *file;
1067         struct trace_array *tr = dir->tr;
1068         char buf[2];
1069         int set = 0;
1070         int ret;
1071
1072         mutex_lock(&event_mutex);
1073         list_for_each_entry(file, &tr->events, list) {
1074                 call = file->event_call;
1075                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1076                         continue;
1077
1078                 if (system && strcmp(call->class->system, system->name) != 0)
1079                         continue;
1080
1081                 /*
1082                  * We need to find out if all the events are set
1083                  * or if all events or cleared, or if we have
1084                  * a mixture.
1085                  */
1086                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1087
1088                 /*
1089                  * If we have a mixture, no need to look further.
1090                  */
1091                 if (set == 3)
1092                         break;
1093         }
1094         mutex_unlock(&event_mutex);
1095
1096         buf[0] = set_to_char[set];
1097         buf[1] = '\n';
1098
1099         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1100
1101         return ret;
1102 }
1103
1104 static ssize_t
1105 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1106                     loff_t *ppos)
1107 {
1108         struct trace_subsystem_dir *dir = filp->private_data;
1109         struct event_subsystem *system = dir->subsystem;
1110         const char *name = NULL;
1111         unsigned long val;
1112         ssize_t ret;
1113
1114         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1115         if (ret)
1116                 return ret;
1117
1118         ret = tracing_update_buffers();
1119         if (ret < 0)
1120                 return ret;
1121
1122         if (val != 0 && val != 1)
1123                 return -EINVAL;
1124
1125         /*
1126          * Opening of "enable" adds a ref count to system,
1127          * so the name is safe to use.
1128          */
1129         if (system)
1130                 name = system->name;
1131
1132         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1133         if (ret)
1134                 goto out;
1135
1136         ret = cnt;
1137
1138 out:
1139         *ppos += cnt;
1140
1141         return ret;
1142 }
1143
1144 enum {
1145         FORMAT_HEADER           = 1,
1146         FORMAT_FIELD_SEPERATOR  = 2,
1147         FORMAT_PRINTFMT         = 3,
1148 };
1149
1150 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1151 {
1152         struct trace_event_call *call = event_file_data(m->private);
1153         struct list_head *common_head = &ftrace_common_fields;
1154         struct list_head *head = trace_get_fields(call);
1155         struct list_head *node = v;
1156
1157         (*pos)++;
1158
1159         switch ((unsigned long)v) {
1160         case FORMAT_HEADER:
1161                 node = common_head;
1162                 break;
1163
1164         case FORMAT_FIELD_SEPERATOR:
1165                 node = head;
1166                 break;
1167
1168         case FORMAT_PRINTFMT:
1169                 /* all done */
1170                 return NULL;
1171         }
1172
1173         node = node->prev;
1174         if (node == common_head)
1175                 return (void *)FORMAT_FIELD_SEPERATOR;
1176         else if (node == head)
1177                 return (void *)FORMAT_PRINTFMT;
1178         else
1179                 return node;
1180 }
1181
1182 static int f_show(struct seq_file *m, void *v)
1183 {
1184         struct trace_event_call *call = event_file_data(m->private);
1185         struct ftrace_event_field *field;
1186         const char *array_descriptor;
1187
1188         switch ((unsigned long)v) {
1189         case FORMAT_HEADER:
1190                 seq_printf(m, "name: %s\n", trace_event_name(call));
1191                 seq_printf(m, "ID: %d\n", call->event.type);
1192                 seq_puts(m, "format:\n");
1193                 return 0;
1194
1195         case FORMAT_FIELD_SEPERATOR:
1196                 seq_putc(m, '\n');
1197                 return 0;
1198
1199         case FORMAT_PRINTFMT:
1200                 seq_printf(m, "\nprint fmt: %s\n",
1201                            call->print_fmt);
1202                 return 0;
1203         }
1204
1205         field = list_entry(v, struct ftrace_event_field, link);
1206         /*
1207          * Smartly shows the array type(except dynamic array).
1208          * Normal:
1209          *      field:TYPE VAR
1210          * If TYPE := TYPE[LEN], it is shown:
1211          *      field:TYPE VAR[LEN]
1212          */
1213         array_descriptor = strchr(field->type, '[');
1214
1215         if (!strncmp(field->type, "__data_loc", 10))
1216                 array_descriptor = NULL;
1217
1218         if (!array_descriptor)
1219                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1220                            field->type, field->name, field->offset,
1221                            field->size, !!field->is_signed);
1222         else
1223                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1224                            (int)(array_descriptor - field->type),
1225                            field->type, field->name,
1226                            array_descriptor, field->offset,
1227                            field->size, !!field->is_signed);
1228
1229         return 0;
1230 }
1231
1232 static void *f_start(struct seq_file *m, loff_t *pos)
1233 {
1234         void *p = (void *)FORMAT_HEADER;
1235         loff_t l = 0;
1236
1237         /* ->stop() is called even if ->start() fails */
1238         mutex_lock(&event_mutex);
1239         if (!event_file_data(m->private))
1240                 return ERR_PTR(-ENODEV);
1241
1242         while (l < *pos && p)
1243                 p = f_next(m, p, &l);
1244
1245         return p;
1246 }
1247
1248 static void f_stop(struct seq_file *m, void *p)
1249 {
1250         mutex_unlock(&event_mutex);
1251 }
1252
1253 static const struct seq_operations trace_format_seq_ops = {
1254         .start          = f_start,
1255         .next           = f_next,
1256         .stop           = f_stop,
1257         .show           = f_show,
1258 };
1259
1260 static int trace_format_open(struct inode *inode, struct file *file)
1261 {
1262         struct seq_file *m;
1263         int ret;
1264
1265         ret = seq_open(file, &trace_format_seq_ops);
1266         if (ret < 0)
1267                 return ret;
1268
1269         m = file->private_data;
1270         m->private = file;
1271
1272         return 0;
1273 }
1274
1275 static ssize_t
1276 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1277 {
1278         int id = (long)event_file_data(filp);
1279         char buf[32];
1280         int len;
1281
1282         if (*ppos)
1283                 return 0;
1284
1285         if (unlikely(!id))
1286                 return -ENODEV;
1287
1288         len = sprintf(buf, "%d\n", id);
1289
1290         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1291 }
1292
1293 static ssize_t
1294 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1295                   loff_t *ppos)
1296 {
1297         struct trace_event_file *file;
1298         struct trace_seq *s;
1299         int r = -ENODEV;
1300
1301         if (*ppos)
1302                 return 0;
1303
1304         s = kmalloc(sizeof(*s), GFP_KERNEL);
1305
1306         if (!s)
1307                 return -ENOMEM;
1308
1309         trace_seq_init(s);
1310
1311         mutex_lock(&event_mutex);
1312         file = event_file_data(filp);
1313         if (file)
1314                 print_event_filter(file, s);
1315         mutex_unlock(&event_mutex);
1316
1317         if (file)
1318                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1319                                             s->buffer, trace_seq_used(s));
1320
1321         kfree(s);
1322
1323         return r;
1324 }
1325
1326 static ssize_t
1327 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1328                    loff_t *ppos)
1329 {
1330         struct trace_event_file *file;
1331         char *buf;
1332         int err = -ENODEV;
1333
1334         if (cnt >= PAGE_SIZE)
1335                 return -EINVAL;
1336
1337         buf = memdup_user_nul(ubuf, cnt);
1338         if (IS_ERR(buf))
1339                 return PTR_ERR(buf);
1340
1341         mutex_lock(&event_mutex);
1342         file = event_file_data(filp);
1343         if (file)
1344                 err = apply_event_filter(file, buf);
1345         mutex_unlock(&event_mutex);
1346
1347         kfree(buf);
1348         if (err < 0)
1349                 return err;
1350
1351         *ppos += cnt;
1352
1353         return cnt;
1354 }
1355
1356 static LIST_HEAD(event_subsystems);
1357
1358 static int subsystem_open(struct inode *inode, struct file *filp)
1359 {
1360         struct event_subsystem *system = NULL;
1361         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1362         struct trace_array *tr;
1363         int ret;
1364
1365         if (tracing_is_disabled())
1366                 return -ENODEV;
1367
1368         /* Make sure the system still exists */
1369         mutex_lock(&trace_types_lock);
1370         mutex_lock(&event_mutex);
1371         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1372                 list_for_each_entry(dir, &tr->systems, list) {
1373                         if (dir == inode->i_private) {
1374                                 /* Don't open systems with no events */
1375                                 if (dir->nr_events) {
1376                                         __get_system_dir(dir);
1377                                         system = dir->subsystem;
1378                                 }
1379                                 goto exit_loop;
1380                         }
1381                 }
1382         }
1383  exit_loop:
1384         mutex_unlock(&event_mutex);
1385         mutex_unlock(&trace_types_lock);
1386
1387         if (!system)
1388                 return -ENODEV;
1389
1390         /* Some versions of gcc think dir can be uninitialized here */
1391         WARN_ON(!dir);
1392
1393         /* Still need to increment the ref count of the system */
1394         if (trace_array_get(tr) < 0) {
1395                 put_system(dir);
1396                 return -ENODEV;
1397         }
1398
1399         ret = tracing_open_generic(inode, filp);
1400         if (ret < 0) {
1401                 trace_array_put(tr);
1402                 put_system(dir);
1403         }
1404
1405         return ret;
1406 }
1407
1408 static int system_tr_open(struct inode *inode, struct file *filp)
1409 {
1410         struct trace_subsystem_dir *dir;
1411         struct trace_array *tr = inode->i_private;
1412         int ret;
1413
1414         if (tracing_is_disabled())
1415                 return -ENODEV;
1416
1417         if (trace_array_get(tr) < 0)
1418                 return -ENODEV;
1419
1420         /* Make a temporary dir that has no system but points to tr */
1421         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1422         if (!dir) {
1423                 trace_array_put(tr);
1424                 return -ENOMEM;
1425         }
1426
1427         dir->tr = tr;
1428
1429         ret = tracing_open_generic(inode, filp);
1430         if (ret < 0) {
1431                 trace_array_put(tr);
1432                 kfree(dir);
1433                 return ret;
1434         }
1435
1436         filp->private_data = dir;
1437
1438         return 0;
1439 }
1440
1441 static int subsystem_release(struct inode *inode, struct file *file)
1442 {
1443         struct trace_subsystem_dir *dir = file->private_data;
1444
1445         trace_array_put(dir->tr);
1446
1447         /*
1448          * If dir->subsystem is NULL, then this is a temporary
1449          * descriptor that was made for a trace_array to enable
1450          * all subsystems.
1451          */
1452         if (dir->subsystem)
1453                 put_system(dir);
1454         else
1455                 kfree(dir);
1456
1457         return 0;
1458 }
1459
1460 static ssize_t
1461 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1462                       loff_t *ppos)
1463 {
1464         struct trace_subsystem_dir *dir = filp->private_data;
1465         struct event_subsystem *system = dir->subsystem;
1466         struct trace_seq *s;
1467         int r;
1468
1469         if (*ppos)
1470                 return 0;
1471
1472         s = kmalloc(sizeof(*s), GFP_KERNEL);
1473         if (!s)
1474                 return -ENOMEM;
1475
1476         trace_seq_init(s);
1477
1478         print_subsystem_event_filter(system, s);
1479         r = simple_read_from_buffer(ubuf, cnt, ppos,
1480                                     s->buffer, trace_seq_used(s));
1481
1482         kfree(s);
1483
1484         return r;
1485 }
1486
1487 static ssize_t
1488 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1489                        loff_t *ppos)
1490 {
1491         struct trace_subsystem_dir *dir = filp->private_data;
1492         char *buf;
1493         int err;
1494
1495         if (cnt >= PAGE_SIZE)
1496                 return -EINVAL;
1497
1498         buf = memdup_user_nul(ubuf, cnt);
1499         if (IS_ERR(buf))
1500                 return PTR_ERR(buf);
1501
1502         err = apply_subsystem_event_filter(dir, buf);
1503         kfree(buf);
1504         if (err < 0)
1505                 return err;
1506
1507         *ppos += cnt;
1508
1509         return cnt;
1510 }
1511
1512 static ssize_t
1513 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1514 {
1515         int (*func)(struct trace_seq *s) = filp->private_data;
1516         struct trace_seq *s;
1517         int r;
1518
1519         if (*ppos)
1520                 return 0;
1521
1522         s = kmalloc(sizeof(*s), GFP_KERNEL);
1523         if (!s)
1524                 return -ENOMEM;
1525
1526         trace_seq_init(s);
1527
1528         func(s);
1529         r = simple_read_from_buffer(ubuf, cnt, ppos,
1530                                     s->buffer, trace_seq_used(s));
1531
1532         kfree(s);
1533
1534         return r;
1535 }
1536
1537 static void ignore_task_cpu(void *data)
1538 {
1539         struct trace_array *tr = data;
1540         struct trace_pid_list *pid_list;
1541
1542         /*
1543          * This function is called by on_each_cpu() while the
1544          * event_mutex is held.
1545          */
1546         pid_list = rcu_dereference_protected(tr->filtered_pids,
1547                                              mutex_is_locked(&event_mutex));
1548
1549         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1550                        trace_ignore_this_task(pid_list, current));
1551 }
1552
1553 static ssize_t
1554 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1555                        size_t cnt, loff_t *ppos)
1556 {
1557         struct seq_file *m = filp->private_data;
1558         struct trace_array *tr = m->private;
1559         struct trace_pid_list *filtered_pids = NULL;
1560         struct trace_pid_list *pid_list;
1561         struct trace_event_file *file;
1562         ssize_t ret;
1563
1564         if (!cnt)
1565                 return 0;
1566
1567         ret = tracing_update_buffers();
1568         if (ret < 0)
1569                 return ret;
1570
1571         mutex_lock(&event_mutex);
1572
1573         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1574                                              lockdep_is_held(&event_mutex));
1575
1576         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1577         if (ret < 0)
1578                 goto out;
1579
1580         rcu_assign_pointer(tr->filtered_pids, pid_list);
1581
1582         list_for_each_entry(file, &tr->events, list) {
1583                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1584         }
1585
1586         if (filtered_pids) {
1587                 synchronize_sched();
1588                 trace_free_pid_list(filtered_pids);
1589         } else if (pid_list) {
1590                 /*
1591                  * Register a probe that is called before all other probes
1592                  * to set ignore_pid if next or prev do not match.
1593                  * Register a probe this is called after all other probes
1594                  * to only keep ignore_pid set if next pid matches.
1595                  */
1596                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1597                                                  tr, INT_MAX);
1598                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1599                                                  tr, 0);
1600
1601                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1602                                                  tr, INT_MAX);
1603                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1604                                                  tr, 0);
1605
1606                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1607                                                      tr, INT_MAX);
1608                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1609                                                      tr, 0);
1610
1611                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1612                                                  tr, INT_MAX);
1613                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1614                                                  tr, 0);
1615         }
1616
1617         /*
1618          * Ignoring of pids is done at task switch. But we have to
1619          * check for those tasks that are currently running.
1620          * Always do this in case a pid was appended or removed.
1621          */
1622         on_each_cpu(ignore_task_cpu, tr, 1);
1623
1624  out:
1625         mutex_unlock(&event_mutex);
1626
1627         if (ret > 0)
1628                 *ppos += ret;
1629
1630         return ret;
1631 }
1632
1633 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1634 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1635 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1636 static int ftrace_event_release(struct inode *inode, struct file *file);
1637
1638 static const struct seq_operations show_event_seq_ops = {
1639         .start = t_start,
1640         .next = t_next,
1641         .show = t_show,
1642         .stop = t_stop,
1643 };
1644
1645 static const struct seq_operations show_set_event_seq_ops = {
1646         .start = s_start,
1647         .next = s_next,
1648         .show = t_show,
1649         .stop = t_stop,
1650 };
1651
1652 static const struct seq_operations show_set_pid_seq_ops = {
1653         .start = p_start,
1654         .next = p_next,
1655         .show = trace_pid_show,
1656         .stop = p_stop,
1657 };
1658
1659 static const struct file_operations ftrace_avail_fops = {
1660         .open = ftrace_event_avail_open,
1661         .read = seq_read,
1662         .llseek = seq_lseek,
1663         .release = seq_release,
1664 };
1665
1666 static const struct file_operations ftrace_set_event_fops = {
1667         .open = ftrace_event_set_open,
1668         .read = seq_read,
1669         .write = ftrace_event_write,
1670         .llseek = seq_lseek,
1671         .release = ftrace_event_release,
1672 };
1673
1674 static const struct file_operations ftrace_set_event_pid_fops = {
1675         .open = ftrace_event_set_pid_open,
1676         .read = seq_read,
1677         .write = ftrace_event_pid_write,
1678         .llseek = seq_lseek,
1679         .release = ftrace_event_release,
1680 };
1681
1682 static const struct file_operations ftrace_enable_fops = {
1683         .open = tracing_open_generic,
1684         .read = event_enable_read,
1685         .write = event_enable_write,
1686         .llseek = default_llseek,
1687 };
1688
1689 static const struct file_operations ftrace_event_format_fops = {
1690         .open = trace_format_open,
1691         .read = seq_read,
1692         .llseek = seq_lseek,
1693         .release = seq_release,
1694 };
1695
1696 static const struct file_operations ftrace_event_id_fops = {
1697         .read = event_id_read,
1698         .llseek = default_llseek,
1699 };
1700
1701 static const struct file_operations ftrace_event_filter_fops = {
1702         .open = tracing_open_generic,
1703         .read = event_filter_read,
1704         .write = event_filter_write,
1705         .llseek = default_llseek,
1706 };
1707
1708 static const struct file_operations ftrace_subsystem_filter_fops = {
1709         .open = subsystem_open,
1710         .read = subsystem_filter_read,
1711         .write = subsystem_filter_write,
1712         .llseek = default_llseek,
1713         .release = subsystem_release,
1714 };
1715
1716 static const struct file_operations ftrace_system_enable_fops = {
1717         .open = subsystem_open,
1718         .read = system_enable_read,
1719         .write = system_enable_write,
1720         .llseek = default_llseek,
1721         .release = subsystem_release,
1722 };
1723
1724 static const struct file_operations ftrace_tr_enable_fops = {
1725         .open = system_tr_open,
1726         .read = system_enable_read,
1727         .write = system_enable_write,
1728         .llseek = default_llseek,
1729         .release = subsystem_release,
1730 };
1731
1732 static const struct file_operations ftrace_show_header_fops = {
1733         .open = tracing_open_generic,
1734         .read = show_header,
1735         .llseek = default_llseek,
1736 };
1737
1738 static int
1739 ftrace_event_open(struct inode *inode, struct file *file,
1740                   const struct seq_operations *seq_ops)
1741 {
1742         struct seq_file *m;
1743         int ret;
1744
1745         ret = seq_open(file, seq_ops);
1746         if (ret < 0)
1747                 return ret;
1748         m = file->private_data;
1749         /* copy tr over to seq ops */
1750         m->private = inode->i_private;
1751
1752         return ret;
1753 }
1754
1755 static int ftrace_event_release(struct inode *inode, struct file *file)
1756 {
1757         struct trace_array *tr = inode->i_private;
1758
1759         trace_array_put(tr);
1760
1761         return seq_release(inode, file);
1762 }
1763
1764 static int
1765 ftrace_event_avail_open(struct inode *inode, struct file *file)
1766 {
1767         const struct seq_operations *seq_ops = &show_event_seq_ops;
1768
1769         return ftrace_event_open(inode, file, seq_ops);
1770 }
1771
1772 static int
1773 ftrace_event_set_open(struct inode *inode, struct file *file)
1774 {
1775         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1776         struct trace_array *tr = inode->i_private;
1777         int ret;
1778
1779         if (trace_array_get(tr) < 0)
1780                 return -ENODEV;
1781
1782         if ((file->f_mode & FMODE_WRITE) &&
1783             (file->f_flags & O_TRUNC))
1784                 ftrace_clear_events(tr);
1785
1786         ret = ftrace_event_open(inode, file, seq_ops);
1787         if (ret < 0)
1788                 trace_array_put(tr);
1789         return ret;
1790 }
1791
1792 static int
1793 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1794 {
1795         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1796         struct trace_array *tr = inode->i_private;
1797         int ret;
1798
1799         if (trace_array_get(tr) < 0)
1800                 return -ENODEV;
1801
1802         if ((file->f_mode & FMODE_WRITE) &&
1803             (file->f_flags & O_TRUNC))
1804                 ftrace_clear_event_pids(tr);
1805
1806         ret = ftrace_event_open(inode, file, seq_ops);
1807         if (ret < 0)
1808                 trace_array_put(tr);
1809         return ret;
1810 }
1811
1812 static struct event_subsystem *
1813 create_new_subsystem(const char *name)
1814 {
1815         struct event_subsystem *system;
1816
1817         /* need to create new entry */
1818         system = kmalloc(sizeof(*system), GFP_KERNEL);
1819         if (!system)
1820                 return NULL;
1821
1822         system->ref_count = 1;
1823
1824         /* Only allocate if dynamic (kprobes and modules) */
1825         system->name = kstrdup_const(name, GFP_KERNEL);
1826         if (!system->name)
1827                 goto out_free;
1828
1829         system->filter = NULL;
1830
1831         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1832         if (!system->filter)
1833                 goto out_free;
1834
1835         list_add(&system->list, &event_subsystems);
1836
1837         return system;
1838
1839  out_free:
1840         kfree_const(system->name);
1841         kfree(system);
1842         return NULL;
1843 }
1844
1845 static struct dentry *
1846 event_subsystem_dir(struct trace_array *tr, const char *name,
1847                     struct trace_event_file *file, struct dentry *parent)
1848 {
1849         struct trace_subsystem_dir *dir;
1850         struct event_subsystem *system;
1851         struct dentry *entry;
1852
1853         /* First see if we did not already create this dir */
1854         list_for_each_entry(dir, &tr->systems, list) {
1855                 system = dir->subsystem;
1856                 if (strcmp(system->name, name) == 0) {
1857                         dir->nr_events++;
1858                         file->system = dir;
1859                         return dir->entry;
1860                 }
1861         }
1862
1863         /* Now see if the system itself exists. */
1864         list_for_each_entry(system, &event_subsystems, list) {
1865                 if (strcmp(system->name, name) == 0)
1866                         break;
1867         }
1868         /* Reset system variable when not found */
1869         if (&system->list == &event_subsystems)
1870                 system = NULL;
1871
1872         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1873         if (!dir)
1874                 goto out_fail;
1875
1876         if (!system) {
1877                 system = create_new_subsystem(name);
1878                 if (!system)
1879                         goto out_free;
1880         } else
1881                 __get_system(system);
1882
1883         dir->entry = tracefs_create_dir(name, parent);
1884         if (!dir->entry) {
1885                 pr_warn("Failed to create system directory %s\n", name);
1886                 __put_system(system);
1887                 goto out_free;
1888         }
1889
1890         dir->tr = tr;
1891         dir->ref_count = 1;
1892         dir->nr_events = 1;
1893         dir->subsystem = system;
1894         file->system = dir;
1895
1896         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1897                                     &ftrace_subsystem_filter_fops);
1898         if (!entry) {
1899                 kfree(system->filter);
1900                 system->filter = NULL;
1901                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1902         }
1903
1904         trace_create_file("enable", 0644, dir->entry, dir,
1905                           &ftrace_system_enable_fops);
1906
1907         list_add(&dir->list, &tr->systems);
1908
1909         return dir->entry;
1910
1911  out_free:
1912         kfree(dir);
1913  out_fail:
1914         /* Only print this message if failed on memory allocation */
1915         if (!dir || !system)
1916                 pr_warn("No memory to create event subsystem %s\n", name);
1917         return NULL;
1918 }
1919
1920 static int
1921 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1922 {
1923         struct trace_event_call *call = file->event_call;
1924         struct trace_array *tr = file->tr;
1925         struct list_head *head;
1926         struct dentry *d_events;
1927         const char *name;
1928         int ret;
1929
1930         /*
1931          * If the trace point header did not define TRACE_SYSTEM
1932          * then the system would be called "TRACE_SYSTEM".
1933          */
1934         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1935                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1936                 if (!d_events)
1937                         return -ENOMEM;
1938         } else
1939                 d_events = parent;
1940
1941         name = trace_event_name(call);
1942         file->dir = tracefs_create_dir(name, d_events);
1943         if (!file->dir) {
1944                 pr_warn("Could not create tracefs '%s' directory\n", name);
1945                 return -1;
1946         }
1947
1948         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1949                 trace_create_file("enable", 0644, file->dir, file,
1950                                   &ftrace_enable_fops);
1951
1952 #ifdef CONFIG_PERF_EVENTS
1953         if (call->event.type && call->class->reg)
1954                 trace_create_file("id", 0444, file->dir,
1955                                   (void *)(long)call->event.type,
1956                                   &ftrace_event_id_fops);
1957 #endif
1958
1959         /*
1960          * Other events may have the same class. Only update
1961          * the fields if they are not already defined.
1962          */
1963         head = trace_get_fields(call);
1964         if (list_empty(head)) {
1965                 ret = call->class->define_fields(call);
1966                 if (ret < 0) {
1967                         pr_warn("Could not initialize trace point events/%s\n",
1968                                 name);
1969                         return -1;
1970                 }
1971         }
1972         trace_create_file("filter", 0644, file->dir, file,
1973                           &ftrace_event_filter_fops);
1974
1975         /*
1976          * Only event directories that can be enabled should have
1977          * triggers.
1978          */
1979         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1980                 trace_create_file("trigger", 0644, file->dir, file,
1981                                   &event_trigger_fops);
1982
1983 #ifdef CONFIG_HIST_TRIGGERS
1984         trace_create_file("hist", 0444, file->dir, file,
1985                           &event_hist_fops);
1986 #endif
1987         trace_create_file("format", 0444, file->dir, call,
1988                           &ftrace_event_format_fops);
1989
1990         return 0;
1991 }
1992
1993 static void remove_event_from_tracers(struct trace_event_call *call)
1994 {
1995         struct trace_event_file *file;
1996         struct trace_array *tr;
1997
1998         do_for_each_event_file_safe(tr, file) {
1999                 if (file->event_call != call)
2000                         continue;
2001
2002                 remove_event_file_dir(file);
2003                 /*
2004                  * The do_for_each_event_file_safe() is
2005                  * a double loop. After finding the call for this
2006                  * trace_array, we use break to jump to the next
2007                  * trace_array.
2008                  */
2009                 break;
2010         } while_for_each_event_file();
2011 }
2012
2013 static void event_remove(struct trace_event_call *call)
2014 {
2015         struct trace_array *tr;
2016         struct trace_event_file *file;
2017
2018         do_for_each_event_file(tr, file) {
2019                 if (file->event_call != call)
2020                         continue;
2021                 ftrace_event_enable_disable(file, 0);
2022                 /*
2023                  * The do_for_each_event_file() is
2024                  * a double loop. After finding the call for this
2025                  * trace_array, we use break to jump to the next
2026                  * trace_array.
2027                  */
2028                 break;
2029         } while_for_each_event_file();
2030
2031         if (call->event.funcs)
2032                 __unregister_trace_event(&call->event);
2033         remove_event_from_tracers(call);
2034         list_del(&call->list);
2035 }
2036
2037 static int event_init(struct trace_event_call *call)
2038 {
2039         int ret = 0;
2040         const char *name;
2041
2042         name = trace_event_name(call);
2043         if (WARN_ON(!name))
2044                 return -EINVAL;
2045
2046         if (call->class->raw_init) {
2047                 ret = call->class->raw_init(call);
2048                 if (ret < 0 && ret != -ENOSYS)
2049                         pr_warn("Could not initialize trace events/%s\n", name);
2050         }
2051
2052         return ret;
2053 }
2054
2055 static int
2056 __register_event(struct trace_event_call *call, struct module *mod)
2057 {
2058         int ret;
2059
2060         ret = event_init(call);
2061         if (ret < 0)
2062                 return ret;
2063
2064         list_add(&call->list, &ftrace_events);
2065         call->mod = mod;
2066
2067         return 0;
2068 }
2069
2070 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2071 {
2072         int rlen;
2073         int elen;
2074
2075         /* Find the length of the enum value as a string */
2076         elen = snprintf(ptr, 0, "%ld", map->enum_value);
2077         /* Make sure there's enough room to replace the string with the value */
2078         if (len < elen)
2079                 return NULL;
2080
2081         snprintf(ptr, elen + 1, "%ld", map->enum_value);
2082
2083         /* Get the rest of the string of ptr */
2084         rlen = strlen(ptr + len);
2085         memmove(ptr + elen, ptr + len, rlen);
2086         /* Make sure we end the new string */
2087         ptr[elen + rlen] = 0;
2088
2089         return ptr + elen;
2090 }
2091
2092 static void update_event_printk(struct trace_event_call *call,
2093                                 struct trace_enum_map *map)
2094 {
2095         char *ptr;
2096         int quote = 0;
2097         int len = strlen(map->enum_string);
2098
2099         for (ptr = call->print_fmt; *ptr; ptr++) {
2100                 if (*ptr == '\\') {
2101                         ptr++;
2102                         /* paranoid */
2103                         if (!*ptr)
2104                                 break;
2105                         continue;
2106                 }
2107                 if (*ptr == '"') {
2108                         quote ^= 1;
2109                         continue;
2110                 }
2111                 if (quote)
2112                         continue;
2113                 if (isdigit(*ptr)) {
2114                         /* skip numbers */
2115                         do {
2116                                 ptr++;
2117                                 /* Check for alpha chars like ULL */
2118                         } while (isalnum(*ptr));
2119                         if (!*ptr)
2120                                 break;
2121                         /*
2122                          * A number must have some kind of delimiter after
2123                          * it, and we can ignore that too.
2124                          */
2125                         continue;
2126                 }
2127                 if (isalpha(*ptr) || *ptr == '_') {
2128                         if (strncmp(map->enum_string, ptr, len) == 0 &&
2129                             !isalnum(ptr[len]) && ptr[len] != '_') {
2130                                 ptr = enum_replace(ptr, map, len);
2131                                 /* Hmm, enum string smaller than value */
2132                                 if (WARN_ON_ONCE(!ptr))
2133                                         return;
2134                                 /*
2135                                  * No need to decrement here, as enum_replace()
2136                                  * returns the pointer to the character passed
2137                                  * the enum, and two enums can not be placed
2138                                  * back to back without something in between.
2139                                  * We can skip that something in between.
2140                                  */
2141                                 continue;
2142                         }
2143                 skip_more:
2144                         do {
2145                                 ptr++;
2146                         } while (isalnum(*ptr) || *ptr == '_');
2147                         if (!*ptr)
2148                                 break;
2149                         /*
2150                          * If what comes after this variable is a '.' or
2151                          * '->' then we can continue to ignore that string.
2152                          */
2153                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2154                                 ptr += *ptr == '.' ? 1 : 2;
2155                                 if (!*ptr)
2156                                         break;
2157                                 goto skip_more;
2158                         }
2159                         /*
2160                          * Once again, we can skip the delimiter that came
2161                          * after the string.
2162                          */
2163                         continue;
2164                 }
2165         }
2166 }
2167
2168 void trace_event_enum_update(struct trace_enum_map **map, int len)
2169 {
2170         struct trace_event_call *call, *p;
2171         const char *last_system = NULL;
2172         int last_i;
2173         int i;
2174
2175         down_write(&trace_event_sem);
2176         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2177                 /* events are usually grouped together with systems */
2178                 if (!last_system || call->class->system != last_system) {
2179                         last_i = 0;
2180                         last_system = call->class->system;
2181                 }
2182
2183                 for (i = last_i; i < len; i++) {
2184                         if (call->class->system == map[i]->system) {
2185                                 /* Save the first system if need be */
2186                                 if (!last_i)
2187                                         last_i = i;
2188                                 update_event_printk(call, map[i]);
2189                         }
2190                 }
2191         }
2192         up_write(&trace_event_sem);
2193 }
2194
2195 static struct trace_event_file *
2196 trace_create_new_event(struct trace_event_call *call,
2197                        struct trace_array *tr)
2198 {
2199         struct trace_event_file *file;
2200
2201         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2202         if (!file)
2203                 return NULL;
2204
2205         file->event_call = call;
2206         file->tr = tr;
2207         atomic_set(&file->sm_ref, 0);
2208         atomic_set(&file->tm_ref, 0);
2209         INIT_LIST_HEAD(&file->triggers);
2210         list_add(&file->list, &tr->events);
2211
2212         return file;
2213 }
2214
2215 /* Add an event to a trace directory */
2216 static int
2217 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2218 {
2219         struct trace_event_file *file;
2220
2221         file = trace_create_new_event(call, tr);
2222         if (!file)
2223                 return -ENOMEM;
2224
2225         return event_create_dir(tr->event_dir, file);
2226 }
2227
2228 /*
2229  * Just create a decriptor for early init. A descriptor is required
2230  * for enabling events at boot. We want to enable events before
2231  * the filesystem is initialized.
2232  */
2233 static __init int
2234 __trace_early_add_new_event(struct trace_event_call *call,
2235                             struct trace_array *tr)
2236 {
2237         struct trace_event_file *file;
2238
2239         file = trace_create_new_event(call, tr);
2240         if (!file)
2241                 return -ENOMEM;
2242
2243         return 0;
2244 }
2245
2246 struct ftrace_module_file_ops;
2247 static void __add_event_to_tracers(struct trace_event_call *call);
2248
2249 /* Add an additional event_call dynamically */
2250 int trace_add_event_call(struct trace_event_call *call)
2251 {
2252         int ret;
2253         mutex_lock(&trace_types_lock);
2254         mutex_lock(&event_mutex);
2255
2256         ret = __register_event(call, NULL);
2257         if (ret >= 0)
2258                 __add_event_to_tracers(call);
2259
2260         mutex_unlock(&event_mutex);
2261         mutex_unlock(&trace_types_lock);
2262         return ret;
2263 }
2264
2265 /*
2266  * Must be called under locking of trace_types_lock, event_mutex and
2267  * trace_event_sem.
2268  */
2269 static void __trace_remove_event_call(struct trace_event_call *call)
2270 {
2271         event_remove(call);
2272         trace_destroy_fields(call);
2273         free_event_filter(call->filter);
2274         call->filter = NULL;
2275 }
2276
2277 static int probe_remove_event_call(struct trace_event_call *call)
2278 {
2279         struct trace_array *tr;
2280         struct trace_event_file *file;
2281
2282 #ifdef CONFIG_PERF_EVENTS
2283         if (call->perf_refcount)
2284                 return -EBUSY;
2285 #endif
2286         do_for_each_event_file(tr, file) {
2287                 if (file->event_call != call)
2288                         continue;
2289                 /*
2290                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2291                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2292                  * TRACE_REG_UNREGISTER.
2293                  */
2294                 if (file->flags & EVENT_FILE_FL_ENABLED)
2295                         return -EBUSY;
2296                 /*
2297                  * The do_for_each_event_file_safe() is
2298                  * a double loop. After finding the call for this
2299                  * trace_array, we use break to jump to the next
2300                  * trace_array.
2301                  */
2302                 break;
2303         } while_for_each_event_file();
2304
2305         __trace_remove_event_call(call);
2306
2307         return 0;
2308 }
2309
2310 /* Remove an event_call */
2311 int trace_remove_event_call(struct trace_event_call *call)
2312 {
2313         int ret;
2314
2315         mutex_lock(&trace_types_lock);
2316         mutex_lock(&event_mutex);
2317         down_write(&trace_event_sem);
2318         ret = probe_remove_event_call(call);
2319         up_write(&trace_event_sem);
2320         mutex_unlock(&event_mutex);
2321         mutex_unlock(&trace_types_lock);
2322
2323         return ret;
2324 }
2325
2326 #define for_each_event(event, start, end)                       \
2327         for (event = start;                                     \
2328              (unsigned long)event < (unsigned long)end;         \
2329              event++)
2330
2331 #ifdef CONFIG_MODULES
2332
2333 static void trace_module_add_events(struct module *mod)
2334 {
2335         struct trace_event_call **call, **start, **end;
2336
2337         if (!mod->num_trace_events)
2338                 return;
2339
2340         /* Don't add infrastructure for mods without tracepoints */
2341         if (trace_module_has_bad_taint(mod)) {
2342                 pr_err("%s: module has bad taint, not creating trace events\n",
2343                        mod->name);
2344                 return;
2345         }
2346
2347         start = mod->trace_events;
2348         end = mod->trace_events + mod->num_trace_events;
2349
2350         for_each_event(call, start, end) {
2351                 __register_event(*call, mod);
2352                 __add_event_to_tracers(*call);
2353         }
2354 }
2355
2356 static void trace_module_remove_events(struct module *mod)
2357 {
2358         struct trace_event_call *call, *p;
2359         bool clear_trace = false;
2360
2361         down_write(&trace_event_sem);
2362         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2363                 if (call->mod == mod) {
2364                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2365                                 clear_trace = true;
2366                         __trace_remove_event_call(call);
2367                 }
2368         }
2369         up_write(&trace_event_sem);
2370
2371         /*
2372          * It is safest to reset the ring buffer if the module being unloaded
2373          * registered any events that were used. The only worry is if
2374          * a new module gets loaded, and takes on the same id as the events
2375          * of this module. When printing out the buffer, traced events left
2376          * over from this module may be passed to the new module events and
2377          * unexpected results may occur.
2378          */
2379         if (clear_trace)
2380                 tracing_reset_all_online_cpus();
2381 }
2382
2383 static int trace_module_notify(struct notifier_block *self,
2384                                unsigned long val, void *data)
2385 {
2386         struct module *mod = data;
2387
2388         mutex_lock(&trace_types_lock);
2389         mutex_lock(&event_mutex);
2390         switch (val) {
2391         case MODULE_STATE_COMING:
2392                 trace_module_add_events(mod);
2393                 break;
2394         case MODULE_STATE_GOING:
2395                 trace_module_remove_events(mod);
2396                 break;
2397         }
2398         mutex_unlock(&event_mutex);
2399         mutex_unlock(&trace_types_lock);
2400
2401         return 0;
2402 }
2403
2404 static struct notifier_block trace_module_nb = {
2405         .notifier_call = trace_module_notify,
2406         .priority = 1, /* higher than trace.c module notify */
2407 };
2408 #endif /* CONFIG_MODULES */
2409
2410 /* Create a new event directory structure for a trace directory. */
2411 static void
2412 __trace_add_event_dirs(struct trace_array *tr)
2413 {
2414         struct trace_event_call *call;
2415         int ret;
2416
2417         list_for_each_entry(call, &ftrace_events, list) {
2418                 ret = __trace_add_new_event(call, tr);
2419                 if (ret < 0)
2420                         pr_warn("Could not create directory for event %s\n",
2421                                 trace_event_name(call));
2422         }
2423 }
2424
2425 struct trace_event_file *
2426 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2427 {
2428         struct trace_event_file *file;
2429         struct trace_event_call *call;
2430         const char *name;
2431
2432         list_for_each_entry(file, &tr->events, list) {
2433
2434                 call = file->event_call;
2435                 name = trace_event_name(call);
2436
2437                 if (!name || !call->class || !call->class->reg)
2438                         continue;
2439
2440                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2441                         continue;
2442
2443                 if (strcmp(event, name) == 0 &&
2444                     strcmp(system, call->class->system) == 0)
2445                         return file;
2446         }
2447         return NULL;
2448 }
2449
2450 #ifdef CONFIG_DYNAMIC_FTRACE
2451
2452 /* Avoid typos */
2453 #define ENABLE_EVENT_STR        "enable_event"
2454 #define DISABLE_EVENT_STR       "disable_event"
2455
2456 struct event_probe_data {
2457         struct trace_event_file *file;
2458         unsigned long                   count;
2459         int                             ref;
2460         bool                            enable;
2461 };
2462
2463 static void update_event_probe(struct event_probe_data *data)
2464 {
2465         if (data->enable)
2466                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2467         else
2468                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2469 }
2470
2471 static void
2472 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2473                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2474                    void *data)
2475 {
2476         struct ftrace_func_mapper *mapper = data;
2477         struct event_probe_data *edata;
2478         void **pdata;
2479
2480         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2481         if (!pdata || !*pdata)
2482                 return;
2483
2484         edata = *pdata;
2485         update_event_probe(edata);
2486 }
2487
2488 static void
2489 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2490                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2491                          void *data)
2492 {
2493         struct ftrace_func_mapper *mapper = data;
2494         struct event_probe_data *edata;
2495         void **pdata;
2496
2497         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2498         if (!pdata || !*pdata)
2499                 return;
2500
2501         edata = *pdata;
2502
2503         if (!edata->count)
2504                 return;
2505
2506         /* Skip if the event is in a state we want to switch to */
2507         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2508                 return;
2509
2510         if (edata->count != -1)
2511                 (edata->count)--;
2512
2513         update_event_probe(edata);
2514 }
2515
2516 static int
2517 event_enable_print(struct seq_file *m, unsigned long ip,
2518                    struct ftrace_probe_ops *ops, void *data)
2519 {
2520         struct ftrace_func_mapper *mapper = data;
2521         struct event_probe_data *edata;
2522         void **pdata;
2523
2524         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2525
2526         if (WARN_ON_ONCE(!pdata || !*pdata))
2527                 return 0;
2528
2529         edata = *pdata;
2530
2531         seq_printf(m, "%ps:", (void *)ip);
2532
2533         seq_printf(m, "%s:%s:%s",
2534                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2535                    edata->file->event_call->class->system,
2536                    trace_event_name(edata->file->event_call));
2537
2538         if (edata->count == -1)
2539                 seq_puts(m, ":unlimited\n");
2540         else
2541                 seq_printf(m, ":count=%ld\n", edata->count);
2542
2543         return 0;
2544 }
2545
2546 static int
2547 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2548                   unsigned long ip, void *init_data, void **data)
2549 {
2550         struct ftrace_func_mapper *mapper = *data;
2551         struct event_probe_data *edata = init_data;
2552         int ret;
2553
2554         if (!mapper) {
2555                 mapper = allocate_ftrace_func_mapper();
2556                 if (!mapper)
2557                         return -ENODEV;
2558                 *data = mapper;
2559         }
2560
2561         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2562         if (ret < 0)
2563                 return ret;
2564
2565         edata->ref++;
2566
2567         return 0;
2568 }
2569
2570 static int free_probe_data(void *data)
2571 {
2572         struct event_probe_data *edata = data;
2573
2574         edata->ref--;
2575         if (!edata->ref) {
2576                 /* Remove the SOFT_MODE flag */
2577                 __ftrace_event_enable_disable(edata->file, 0, 1);
2578                 module_put(edata->file->event_call->mod);
2579                 kfree(edata);
2580         }
2581         return 0;
2582 }
2583
2584 static void
2585 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2586                   unsigned long ip, void *data)
2587 {
2588         struct ftrace_func_mapper *mapper = data;
2589         struct event_probe_data *edata;
2590
2591         if (!ip) {
2592                 if (!mapper)
2593                         return;
2594                 free_ftrace_func_mapper(mapper, free_probe_data);
2595                 return;
2596         }
2597
2598         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2599
2600         if (WARN_ON_ONCE(!edata))
2601                 return;
2602
2603         if (WARN_ON_ONCE(edata->ref <= 0))
2604                 return;
2605
2606         free_probe_data(edata);
2607 }
2608
2609 static struct ftrace_probe_ops event_enable_probe_ops = {
2610         .func                   = event_enable_probe,
2611         .print                  = event_enable_print,
2612         .init                   = event_enable_init,
2613         .free                   = event_enable_free,
2614 };
2615
2616 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2617         .func                   = event_enable_count_probe,
2618         .print                  = event_enable_print,
2619         .init                   = event_enable_init,
2620         .free                   = event_enable_free,
2621 };
2622
2623 static struct ftrace_probe_ops event_disable_probe_ops = {
2624         .func                   = event_enable_probe,
2625         .print                  = event_enable_print,
2626         .init                   = event_enable_init,
2627         .free                   = event_enable_free,
2628 };
2629
2630 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2631         .func                   = event_enable_count_probe,
2632         .print                  = event_enable_print,
2633         .init                   = event_enable_init,
2634         .free                   = event_enable_free,
2635 };
2636
2637 static int
2638 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2639                   char *glob, char *cmd, char *param, int enabled)
2640 {
2641         struct trace_event_file *file;
2642         struct ftrace_probe_ops *ops;
2643         struct event_probe_data *data;
2644         const char *system;
2645         const char *event;
2646         char *number;
2647         bool enable;
2648         int ret;
2649
2650         if (!tr)
2651                 return -ENODEV;
2652
2653         /* hash funcs only work with set_ftrace_filter */
2654         if (!enabled || !param)
2655                 return -EINVAL;
2656
2657         system = strsep(&param, ":");
2658         if (!param)
2659                 return -EINVAL;
2660
2661         event = strsep(&param, ":");
2662
2663         mutex_lock(&event_mutex);
2664
2665         ret = -EINVAL;
2666         file = find_event_file(tr, system, event);
2667         if (!file)
2668                 goto out;
2669
2670         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2671
2672         if (enable)
2673                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2674         else
2675                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2676
2677         if (glob[0] == '!') {
2678                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2679                 goto out;
2680         }
2681
2682         ret = -ENOMEM;
2683
2684         data = kzalloc(sizeof(*data), GFP_KERNEL);
2685         if (!data)
2686                 goto out;
2687
2688         data->enable = enable;
2689         data->count = -1;
2690         data->file = file;
2691
2692         if (!param)
2693                 goto out_reg;
2694
2695         number = strsep(&param, ":");
2696
2697         ret = -EINVAL;
2698         if (!strlen(number))
2699                 goto out_free;
2700
2701         /*
2702          * We use the callback data field (which is a pointer)
2703          * as our counter.
2704          */
2705         ret = kstrtoul(number, 0, &data->count);
2706         if (ret)
2707                 goto out_free;
2708
2709  out_reg:
2710         /* Don't let event modules unload while probe registered */
2711         ret = try_module_get(file->event_call->mod);
2712         if (!ret) {
2713                 ret = -EBUSY;
2714                 goto out_free;
2715         }
2716
2717         ret = __ftrace_event_enable_disable(file, 1, 1);
2718         if (ret < 0)
2719                 goto out_put;
2720
2721         ret = register_ftrace_function_probe(glob, tr, ops, data);
2722         /*
2723          * The above returns on success the # of functions enabled,
2724          * but if it didn't find any functions it returns zero.
2725          * Consider no functions a failure too.
2726          */
2727         if (!ret) {
2728                 ret = -ENOENT;
2729                 goto out_disable;
2730         } else if (ret < 0)
2731                 goto out_disable;
2732         /* Just return zero, not the number of enabled functions */
2733         ret = 0;
2734  out:
2735         mutex_unlock(&event_mutex);
2736         return ret;
2737
2738  out_disable:
2739         __ftrace_event_enable_disable(file, 0, 1);
2740  out_put:
2741         module_put(file->event_call->mod);
2742  out_free:
2743         kfree(data);
2744         goto out;
2745 }
2746
2747 static struct ftrace_func_command event_enable_cmd = {
2748         .name                   = ENABLE_EVENT_STR,
2749         .func                   = event_enable_func,
2750 };
2751
2752 static struct ftrace_func_command event_disable_cmd = {
2753         .name                   = DISABLE_EVENT_STR,
2754         .func                   = event_enable_func,
2755 };
2756
2757 static __init int register_event_cmds(void)
2758 {
2759         int ret;
2760
2761         ret = register_ftrace_command(&event_enable_cmd);
2762         if (WARN_ON(ret < 0))
2763                 return ret;
2764         ret = register_ftrace_command(&event_disable_cmd);
2765         if (WARN_ON(ret < 0))
2766                 unregister_ftrace_command(&event_enable_cmd);
2767         return ret;
2768 }
2769 #else
2770 static inline int register_event_cmds(void) { return 0; }
2771 #endif /* CONFIG_DYNAMIC_FTRACE */
2772
2773 /*
2774  * The top level array has already had its trace_event_file
2775  * descriptors created in order to allow for early events to
2776  * be recorded. This function is called after the tracefs has been
2777  * initialized, and we now have to create the files associated
2778  * to the events.
2779  */
2780 static __init void
2781 __trace_early_add_event_dirs(struct trace_array *tr)
2782 {
2783         struct trace_event_file *file;
2784         int ret;
2785
2786
2787         list_for_each_entry(file, &tr->events, list) {
2788                 ret = event_create_dir(tr->event_dir, file);
2789                 if (ret < 0)
2790                         pr_warn("Could not create directory for event %s\n",
2791                                 trace_event_name(file->event_call));
2792         }
2793 }
2794
2795 /*
2796  * For early boot up, the top trace array requires to have
2797  * a list of events that can be enabled. This must be done before
2798  * the filesystem is set up in order to allow events to be traced
2799  * early.
2800  */
2801 static __init void
2802 __trace_early_add_events(struct trace_array *tr)
2803 {
2804         struct trace_event_call *call;
2805         int ret;
2806
2807         list_for_each_entry(call, &ftrace_events, list) {
2808                 /* Early boot up should not have any modules loaded */
2809                 if (WARN_ON_ONCE(call->mod))
2810                         continue;
2811
2812                 ret = __trace_early_add_new_event(call, tr);
2813                 if (ret < 0)
2814                         pr_warn("Could not create early event %s\n",
2815                                 trace_event_name(call));
2816         }
2817 }
2818
2819 /* Remove the event directory structure for a trace directory. */
2820 static void
2821 __trace_remove_event_dirs(struct trace_array *tr)
2822 {
2823         struct trace_event_file *file, *next;
2824
2825         list_for_each_entry_safe(file, next, &tr->events, list)
2826                 remove_event_file_dir(file);
2827 }
2828
2829 static void __add_event_to_tracers(struct trace_event_call *call)
2830 {
2831         struct trace_array *tr;
2832
2833         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2834                 __trace_add_new_event(call, tr);
2835 }
2836
2837 extern struct trace_event_call *__start_ftrace_events[];
2838 extern struct trace_event_call *__stop_ftrace_events[];
2839
2840 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2841
2842 static __init int setup_trace_event(char *str)
2843 {
2844         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2845         ring_buffer_expanded = true;
2846         tracing_selftest_disabled = true;
2847
2848         return 1;
2849 }
2850 __setup("trace_event=", setup_trace_event);
2851
2852 /* Expects to have event_mutex held when called */
2853 static int
2854 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2855 {
2856         struct dentry *d_events;
2857         struct dentry *entry;
2858
2859         entry = tracefs_create_file("set_event", 0644, parent,
2860                                     tr, &ftrace_set_event_fops);
2861         if (!entry) {
2862                 pr_warn("Could not create tracefs 'set_event' entry\n");
2863                 return -ENOMEM;
2864         }
2865
2866         d_events = tracefs_create_dir("events", parent);
2867         if (!d_events) {
2868                 pr_warn("Could not create tracefs 'events' directory\n");
2869                 return -ENOMEM;
2870         }
2871
2872         entry = trace_create_file("enable", 0644, d_events,
2873                                   tr, &ftrace_tr_enable_fops);
2874         if (!entry) {
2875                 pr_warn("Could not create tracefs 'enable' entry\n");
2876                 return -ENOMEM;
2877         }
2878
2879         /* There are not as crucial, just warn if they are not created */
2880
2881         entry = tracefs_create_file("set_event_pid", 0644, parent,
2882                                     tr, &ftrace_set_event_pid_fops);
2883         if (!entry)
2884                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2885
2886         /* ring buffer internal formats */
2887         entry = trace_create_file("header_page", 0444, d_events,
2888                                   ring_buffer_print_page_header,
2889                                   &ftrace_show_header_fops);
2890         if (!entry)
2891                 pr_warn("Could not create tracefs 'header_page' entry\n");
2892
2893         entry = trace_create_file("header_event", 0444, d_events,
2894                                   ring_buffer_print_entry_header,
2895                                   &ftrace_show_header_fops);
2896         if (!entry)
2897                 pr_warn("Could not create tracefs 'header_event' entry\n");
2898
2899         tr->event_dir = d_events;
2900
2901         return 0;
2902 }
2903
2904 /**
2905  * event_trace_add_tracer - add a instance of a trace_array to events
2906  * @parent: The parent dentry to place the files/directories for events in
2907  * @tr: The trace array associated with these events
2908  *
2909  * When a new instance is created, it needs to set up its events
2910  * directory, as well as other files associated with events. It also
2911  * creates the event hierachry in the @parent/events directory.
2912  *
2913  * Returns 0 on success.
2914  */
2915 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2916 {
2917         int ret;
2918
2919         mutex_lock(&event_mutex);
2920
2921         ret = create_event_toplevel_files(parent, tr);
2922         if (ret)
2923                 goto out_unlock;
2924
2925         down_write(&trace_event_sem);
2926         __trace_add_event_dirs(tr);
2927         up_write(&trace_event_sem);
2928
2929  out_unlock:
2930         mutex_unlock(&event_mutex);
2931
2932         return ret;
2933 }
2934
2935 /*
2936  * The top trace array already had its file descriptors created.
2937  * Now the files themselves need to be created.
2938  */
2939 static __init int
2940 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2941 {
2942         int ret;
2943
2944         mutex_lock(&event_mutex);
2945
2946         ret = create_event_toplevel_files(parent, tr);
2947         if (ret)
2948                 goto out_unlock;
2949
2950         down_write(&trace_event_sem);
2951         __trace_early_add_event_dirs(tr);
2952         up_write(&trace_event_sem);
2953
2954  out_unlock:
2955         mutex_unlock(&event_mutex);
2956
2957         return ret;
2958 }
2959
2960 int event_trace_del_tracer(struct trace_array *tr)
2961 {
2962         mutex_lock(&event_mutex);
2963
2964         /* Disable any event triggers and associated soft-disabled events */
2965         clear_event_triggers(tr);
2966
2967         /* Clear the pid list */
2968         __ftrace_clear_event_pids(tr);
2969
2970         /* Disable any running events */
2971         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2972
2973         /* Access to events are within rcu_read_lock_sched() */
2974         synchronize_sched();
2975
2976         down_write(&trace_event_sem);
2977         __trace_remove_event_dirs(tr);
2978         tracefs_remove_recursive(tr->event_dir);
2979         up_write(&trace_event_sem);
2980
2981         tr->event_dir = NULL;
2982
2983         mutex_unlock(&event_mutex);
2984
2985         return 0;
2986 }
2987
2988 static __init int event_trace_memsetup(void)
2989 {
2990         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2991         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
2992         return 0;
2993 }
2994
2995 static __init void
2996 early_enable_events(struct trace_array *tr, bool disable_first)
2997 {
2998         char *buf = bootup_event_buf;
2999         char *token;
3000         int ret;
3001
3002         while (true) {
3003                 token = strsep(&buf, ",");
3004
3005                 if (!token)
3006                         break;
3007
3008                 if (*token) {
3009                         /* Restarting syscalls requires that we stop them first */
3010                         if (disable_first)
3011                                 ftrace_set_clr_event(tr, token, 0);
3012
3013                         ret = ftrace_set_clr_event(tr, token, 1);
3014                         if (ret)
3015                                 pr_warn("Failed to enable trace event: %s\n", token);
3016                 }
3017
3018                 /* Put back the comma to allow this to be called again */
3019                 if (buf)
3020                         *(buf - 1) = ',';
3021         }
3022 }
3023
3024 static __init int event_trace_enable(void)
3025 {
3026         struct trace_array *tr = top_trace_array();
3027         struct trace_event_call **iter, *call;
3028         int ret;
3029
3030         if (!tr)
3031                 return -ENODEV;
3032
3033         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3034
3035                 call = *iter;
3036                 ret = event_init(call);
3037                 if (!ret)
3038                         list_add(&call->list, &ftrace_events);
3039         }
3040
3041         /*
3042          * We need the top trace array to have a working set of trace
3043          * points at early init, before the debug files and directories
3044          * are created. Create the file entries now, and attach them
3045          * to the actual file dentries later.
3046          */
3047         __trace_early_add_events(tr);
3048
3049         early_enable_events(tr, false);
3050
3051         trace_printk_start_comm();
3052
3053         register_event_cmds();
3054
3055         register_trigger_cmds();
3056
3057         return 0;
3058 }
3059
3060 /*
3061  * event_trace_enable() is called from trace_event_init() first to
3062  * initialize events and perhaps start any events that are on the
3063  * command line. Unfortunately, there are some events that will not
3064  * start this early, like the system call tracepoints that need
3065  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3066  * is called before pid 1 starts, and this flag is never set, making
3067  * the syscall tracepoint never get reached, but the event is enabled
3068  * regardless (and not doing anything).
3069  */
3070 static __init int event_trace_enable_again(void)
3071 {
3072         struct trace_array *tr;
3073
3074         tr = top_trace_array();
3075         if (!tr)
3076                 return -ENODEV;
3077
3078         early_enable_events(tr, true);
3079
3080         return 0;
3081 }
3082
3083 early_initcall(event_trace_enable_again);
3084
3085 static __init int event_trace_init(void)
3086 {
3087         struct trace_array *tr;
3088         struct dentry *d_tracer;
3089         struct dentry *entry;
3090         int ret;
3091
3092         tr = top_trace_array();
3093         if (!tr)
3094                 return -ENODEV;
3095
3096         d_tracer = tracing_init_dentry();
3097         if (IS_ERR(d_tracer))
3098                 return 0;
3099
3100         entry = tracefs_create_file("available_events", 0444, d_tracer,
3101                                     tr, &ftrace_avail_fops);
3102         if (!entry)
3103                 pr_warn("Could not create tracefs 'available_events' entry\n");
3104
3105         if (trace_define_generic_fields())
3106                 pr_warn("tracing: Failed to allocated generic fields");
3107
3108         if (trace_define_common_fields())
3109                 pr_warn("tracing: Failed to allocate common fields");
3110
3111         ret = early_event_add_tracer(d_tracer, tr);
3112         if (ret)
3113                 return ret;
3114
3115 #ifdef CONFIG_MODULES
3116         ret = register_module_notifier(&trace_module_nb);
3117         if (ret)
3118                 pr_warn("Failed to register trace events module notifier\n");
3119 #endif
3120         return 0;
3121 }
3122
3123 void __init trace_event_init(void)
3124 {
3125         event_trace_memsetup();
3126         init_ftrace_syscalls();
3127         event_trace_enable();
3128 }
3129
3130 fs_initcall(event_trace_init);
3131
3132 #ifdef CONFIG_FTRACE_STARTUP_TEST
3133
3134 static DEFINE_SPINLOCK(test_spinlock);
3135 static DEFINE_SPINLOCK(test_spinlock_irq);
3136 static DEFINE_MUTEX(test_mutex);
3137
3138 static __init void test_work(struct work_struct *dummy)
3139 {
3140         spin_lock(&test_spinlock);
3141         spin_lock_irq(&test_spinlock_irq);
3142         udelay(1);
3143         spin_unlock_irq(&test_spinlock_irq);
3144         spin_unlock(&test_spinlock);
3145
3146         mutex_lock(&test_mutex);
3147         msleep(1);
3148         mutex_unlock(&test_mutex);
3149 }
3150
3151 static __init int event_test_thread(void *unused)
3152 {
3153         void *test_malloc;
3154
3155         test_malloc = kmalloc(1234, GFP_KERNEL);
3156         if (!test_malloc)
3157                 pr_info("failed to kmalloc\n");
3158
3159         schedule_on_each_cpu(test_work);
3160
3161         kfree(test_malloc);
3162
3163         set_current_state(TASK_INTERRUPTIBLE);
3164         while (!kthread_should_stop()) {
3165                 schedule();
3166                 set_current_state(TASK_INTERRUPTIBLE);
3167         }
3168         __set_current_state(TASK_RUNNING);
3169
3170         return 0;
3171 }
3172
3173 /*
3174  * Do various things that may trigger events.
3175  */
3176 static __init void event_test_stuff(void)
3177 {
3178         struct task_struct *test_thread;
3179
3180         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3181         msleep(1);
3182         kthread_stop(test_thread);
3183 }
3184
3185 /*
3186  * For every trace event defined, we will test each trace point separately,
3187  * and then by groups, and finally all trace points.
3188  */
3189 static __init void event_trace_self_tests(void)
3190 {
3191         struct trace_subsystem_dir *dir;
3192         struct trace_event_file *file;
3193         struct trace_event_call *call;
3194         struct event_subsystem *system;
3195         struct trace_array *tr;
3196         int ret;
3197
3198         tr = top_trace_array();
3199         if (!tr)
3200                 return;
3201
3202         pr_info("Running tests on trace events:\n");
3203
3204         list_for_each_entry(file, &tr->events, list) {
3205
3206                 call = file->event_call;
3207
3208                 /* Only test those that have a probe */
3209                 if (!call->class || !call->class->probe)
3210                         continue;
3211
3212 /*
3213  * Testing syscall events here is pretty useless, but
3214  * we still do it if configured. But this is time consuming.
3215  * What we really need is a user thread to perform the
3216  * syscalls as we test.
3217  */
3218 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3219                 if (call->class->system &&
3220                     strcmp(call->class->system, "syscalls") == 0)
3221                         continue;
3222 #endif
3223
3224                 pr_info("Testing event %s: ", trace_event_name(call));
3225
3226                 /*
3227                  * If an event is already enabled, someone is using
3228                  * it and the self test should not be on.
3229                  */
3230                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3231                         pr_warn("Enabled event during self test!\n");
3232                         WARN_ON_ONCE(1);
3233                         continue;
3234                 }
3235
3236                 ftrace_event_enable_disable(file, 1);
3237                 event_test_stuff();
3238                 ftrace_event_enable_disable(file, 0);
3239
3240                 pr_cont("OK\n");
3241         }
3242
3243         /* Now test at the sub system level */
3244
3245         pr_info("Running tests on trace event systems:\n");
3246
3247         list_for_each_entry(dir, &tr->systems, list) {
3248
3249                 system = dir->subsystem;
3250
3251                 /* the ftrace system is special, skip it */
3252                 if (strcmp(system->name, "ftrace") == 0)
3253                         continue;
3254
3255                 pr_info("Testing event system %s: ", system->name);
3256
3257                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3258                 if (WARN_ON_ONCE(ret)) {
3259                         pr_warn("error enabling system %s\n",
3260                                 system->name);
3261                         continue;
3262                 }
3263
3264                 event_test_stuff();
3265
3266                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3267                 if (WARN_ON_ONCE(ret)) {
3268                         pr_warn("error disabling system %s\n",
3269                                 system->name);
3270                         continue;
3271                 }
3272
3273                 pr_cont("OK\n");
3274         }
3275
3276         /* Test with all events enabled */
3277
3278         pr_info("Running tests on all trace events:\n");
3279         pr_info("Testing all events: ");
3280
3281         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3282         if (WARN_ON_ONCE(ret)) {
3283                 pr_warn("error enabling all events\n");
3284                 return;
3285         }
3286
3287         event_test_stuff();
3288
3289         /* reset sysname */
3290         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3291         if (WARN_ON_ONCE(ret)) {
3292                 pr_warn("error disabling all events\n");
3293                 return;
3294         }
3295
3296         pr_cont("OK\n");
3297 }
3298
3299 #ifdef CONFIG_FUNCTION_TRACER
3300
3301 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3302
3303 static struct trace_event_file event_trace_file __initdata;
3304
3305 static void __init
3306 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3307                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3308 {
3309         struct ring_buffer_event *event;
3310         struct ring_buffer *buffer;
3311         struct ftrace_entry *entry;
3312         unsigned long flags;
3313         long disabled;
3314         int cpu;
3315         int pc;
3316
3317         pc = preempt_count();
3318         preempt_disable_notrace();
3319         cpu = raw_smp_processor_id();
3320         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3321
3322         if (disabled != 1)
3323                 goto out;
3324
3325         local_save_flags(flags);
3326
3327         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3328                                                 TRACE_FN, sizeof(*entry),
3329                                                 flags, pc);
3330         if (!event)
3331                 goto out;
3332         entry   = ring_buffer_event_data(event);
3333         entry->ip                       = ip;
3334         entry->parent_ip                = parent_ip;
3335
3336         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3337                                     entry, flags, pc);
3338  out:
3339         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3340         preempt_enable_notrace();
3341 }
3342
3343 static struct ftrace_ops trace_ops __initdata  =
3344 {
3345         .func = function_test_events_call,
3346         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3347 };
3348
3349 static __init void event_trace_self_test_with_function(void)
3350 {
3351         int ret;
3352
3353         event_trace_file.tr = top_trace_array();
3354         if (WARN_ON(!event_trace_file.tr))
3355                 return;
3356
3357         ret = register_ftrace_function(&trace_ops);
3358         if (WARN_ON(ret < 0)) {
3359                 pr_info("Failed to enable function tracer for event tests\n");
3360                 return;
3361         }
3362         pr_info("Running tests again, along with the function tracer\n");
3363         event_trace_self_tests();
3364         unregister_ftrace_function(&trace_ops);
3365 }
3366 #else
3367 static __init void event_trace_self_test_with_function(void)
3368 {
3369 }
3370 #endif
3371
3372 static __init int event_trace_self_tests_init(void)
3373 {
3374         if (!tracing_selftest_disabled) {
3375                 event_trace_self_tests();
3376                 event_trace_self_test_with_function();
3377         }
3378
3379         return 0;
3380 }
3381
3382 late_initcall(event_trace_self_tests_init);
3383
3384 #endif