kernel/trace/trace_events.c

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