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/bsearch.h>
  19 #include <linux/module.h>
  20 #include <linux/ctype.h>
  21 #include <linux/sort.h>
  22 #include <linux/slab.h>
  23 #include <linux/delay.h>
  24
  25 #include <trace/events/sched.h>
  26
  27 #include <asm/setup.h>
  28
  29 #include "trace_output.h"
  30
  31 #undef TRACE_SYSTEM
  32 #define TRACE_SYSTEM "TRACE_SYSTEM"
  33
  34 DEFINE_MUTEX(event_mutex);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_generic_fields);
  38 static LIST_HEAD(ftrace_common_fields);
  39
  40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  41
  42 static struct kmem_cache *field_cachep;
  43 static struct kmem_cache *file_cachep;
  44
  45 static inline int system_refcount(struct event_subsystem *system)
  46 {
  47         return system->ref_count;
  48 }
  49
  50 static int system_refcount_inc(struct event_subsystem *system)
  51 {
  52         return system->ref_count++;
  53 }
  54
  55 static int system_refcount_dec(struct event_subsystem *system)
  56 {
  57         return --system->ref_count;
  58 }
  59
  60 /* Double loops, do not use break, only goto's work */
  61 #define do_for_each_event_file(tr, file)                        \
  62         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  63                 list_for_each_entry(file, &tr->events, list)
  64
  65 #define do_for_each_event_file_safe(tr, file)                   \
  66         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  67                 struct trace_event_file *___n;                          \
  68                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  69
  70 #define while_for_each_event_file()             \
  71         }
  72
  73 static struct list_head *
  74 trace_get_fields(struct trace_event_call *event_call)
  75 {
  76         if (!event_call->class->get_fields)
  77                 return &event_call->class->fields;
  78         return event_call->class->get_fields(event_call);
  79 }
  80
  81 static struct ftrace_event_field *
  82 __find_event_field(struct list_head *head, char *name)
  83 {
  84         struct ftrace_event_field *field;
  85
  86         list_for_each_entry(field, head, link) {
  87                 if (!strcmp(field->name, name))
  88                         return field;
  89         }
  90
  91         return NULL;
  92 }
  93
  94 struct ftrace_event_field *
  95 trace_find_event_field(struct trace_event_call *call, char *name)
  96 {
  97         struct ftrace_event_field *field;
  98         struct list_head *head;
  99
 100         field = __find_event_field(&ftrace_generic_fields, name);
 101         if (field)
 102                 return field;
 103
 104         field = __find_event_field(&ftrace_common_fields, name);
 105         if (field)
 106                 return field;
 107
 108         head = trace_get_fields(call);
 109         return __find_event_field(head, name);
 110 }
 111
 112 static int __trace_define_field(struct list_head *head, const char *type,
 113                                 const char *name, int offset, int size,
 114                                 int is_signed, int filter_type)
 115 {
 116         struct ftrace_event_field *field;
 117
 118         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 119         if (!field)
 120                 return -ENOMEM;
 121
 122         field->name = name;
 123         field->type = type;
 124
 125         if (filter_type == FILTER_OTHER)
 126                 field->filter_type = filter_assign_type(type);
 127         else
 128                 field->filter_type = filter_type;
 129
 130         field->offset = offset;
 131         field->size = size;
 132         field->is_signed = is_signed;
 133
 134         list_add(&field->link, head);
 135
 136         return 0;
 137 }
 138
 139 int trace_define_field(struct trace_event_call *call, const char *type,
 140                        const char *name, int offset, int size, int is_signed,
 141                        int filter_type)
 142 {
 143         struct list_head *head;
 144
 145         if (WARN_ON(!call->class))
 146                 return 0;
 147
 148         head = trace_get_fields(call);
 149         return __trace_define_field(head, type, name, offset, size,
 150                                     is_signed, filter_type);
 151 }
 152 EXPORT_SYMBOL_GPL(trace_define_field);
 153
 154 #define __generic_field(type, item, filter_type)                        \
 155         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 156                                    #item, 0, 0, is_signed_type(type),   \
 157                                    filter_type);                        \
 158         if (ret)                                                        \
 159                 return ret;
 160
 161 #define __common_field(type, item)                                      \
 162         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 163                                    "common_" #item,                     \
 164                                    offsetof(typeof(ent), item),         \
 165                                    sizeof(ent.item),                    \
 166                                    is_signed_type(type), FILTER_OTHER); \
 167         if (ret)                                                        \
 168                 return ret;
 169
 170 static int trace_define_generic_fields(void)
 171 {
 172         int ret;
 173
 174         __generic_field(int, cpu, FILTER_OTHER);
 175         __generic_field(char *, comm, FILTER_PTR_STRING);
 176
 177         return ret;
 178 }
 179
 180 static int trace_define_common_fields(void)
 181 {
 182         int ret;
 183         struct trace_entry ent;
 184
 185         __common_field(unsigned short, type);
 186         __common_field(unsigned char, flags);
 187         __common_field(unsigned char, preempt_count);
 188         __common_field(int, pid);
 189
 190         return ret;
 191 }
 192
 193 static void trace_destroy_fields(struct trace_event_call *call)
 194 {
 195         struct ftrace_event_field *field, *next;
 196         struct list_head *head;
 197
 198         head = trace_get_fields(call);
 199         list_for_each_entry_safe(field, next, head, link) {
 200                 list_del(&field->link);
 201                 kmem_cache_free(field_cachep, field);
 202         }
 203 }
 204
 205 int trace_event_raw_init(struct trace_event_call *call)
 206 {
 207         int id;
 208
 209         id = register_trace_event(&call->event);
 210         if (!id)
 211                 return -ENODEV;
 212
 213         return 0;
 214 }
 215 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 216
 217 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 218 {
 219         struct trace_array *tr = trace_file->tr;
 220         struct trace_array_cpu *data;
 221         struct trace_pid_list *pid_list;
 222
 223         pid_list = rcu_dereference_sched(tr->filtered_pids);
 224         if (!pid_list)
 225                 return false;
 226
 227         data = this_cpu_ptr(tr->trace_buffer.data);
 228
 229         return data->ignore_pid;
 230 }
 231 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 232
 233 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 234                                  struct trace_event_file *trace_file,
 235                                  unsigned long len)
 236 {
 237         struct trace_event_call *event_call = trace_file->event_call;
 238
 239         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 240             trace_event_ignore_this_pid(trace_file))
 241                 return NULL;
 242
 243         local_save_flags(fbuffer->flags);
 244         fbuffer->pc = preempt_count();
 245         fbuffer->trace_file = trace_file;
 246
 247         fbuffer->event =
 248                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 249                                                 event_call->event.type, len,
 250                                                 fbuffer->flags, fbuffer->pc);
 251         if (!fbuffer->event)
 252                 return NULL;
 253
 254         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 255         return fbuffer->entry;
 256 }
 257 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 258
 259 static DEFINE_SPINLOCK(tracepoint_iter_lock);
 260
 261 static void output_printk(struct trace_event_buffer *fbuffer)
 262 {
 263         struct trace_event_call *event_call;
 264         struct trace_event *event;
 265         unsigned long flags;
 266         struct trace_iterator *iter = tracepoint_print_iter;
 267
 268         if (!iter)
 269                 return;
 270
 271         event_call = fbuffer->trace_file->event_call;
 272         if (!event_call || !event_call->event.funcs ||
 273             !event_call->event.funcs->trace)
 274                 return;
 275
 276         event = &fbuffer->trace_file->event_call->event;
 277
 278         spin_lock_irqsave(&tracepoint_iter_lock, flags);
 279         trace_seq_init(&iter->seq);
 280         iter->ent = fbuffer->entry;
 281         event_call->event.funcs->trace(iter, 0, event);
 282         trace_seq_putc(&iter->seq, 0);
 283         printk("%s", iter->seq.buffer);
 284
 285         spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
 286 }
 287
 288 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
 289 {
 290         if (tracepoint_printk)
 291                 output_printk(fbuffer);
 292
 293         event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
 294                                     fbuffer->event, fbuffer->entry,
 295                                     fbuffer->flags, fbuffer->pc);
 296 }
 297 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
 298
 299 int trace_event_reg(struct trace_event_call *call,
 300                     enum trace_reg type, void *data)
 301 {
 302         struct trace_event_file *file = data;
 303
 304         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 305         switch (type) {
 306         case TRACE_REG_REGISTER:
 307                 return tracepoint_probe_register(call->tp,
 308                                                  call->class->probe,
 309                                                  file);
 310         case TRACE_REG_UNREGISTER:
 311                 tracepoint_probe_unregister(call->tp,
 312                                             call->class->probe,
 313                                             file);
 314                 return 0;
 315
 316 #ifdef CONFIG_PERF_EVENTS
 317         case TRACE_REG_PERF_REGISTER:
 318                 return tracepoint_probe_register(call->tp,
 319                                                  call->class->perf_probe,
 320                                                  call);
 321         case TRACE_REG_PERF_UNREGISTER:
 322                 tracepoint_probe_unregister(call->tp,
 323                                             call->class->perf_probe,
 324                                             call);
 325                 return 0;
 326         case TRACE_REG_PERF_OPEN:
 327         case TRACE_REG_PERF_CLOSE:
 328         case TRACE_REG_PERF_ADD:
 329         case TRACE_REG_PERF_DEL:
 330                 return 0;
 331 #endif
 332         }
 333         return 0;
 334 }
 335 EXPORT_SYMBOL_GPL(trace_event_reg);
 336
 337 void trace_event_enable_cmd_record(bool enable)
 338 {
 339         struct trace_event_file *file;
 340         struct trace_array *tr;
 341
 342         mutex_lock(&event_mutex);
 343         do_for_each_event_file(tr, file) {
 344
 345                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 346                         continue;
 347
 348                 if (enable) {
 349                         tracing_start_cmdline_record();
 350                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 351                 } else {
 352                         tracing_stop_cmdline_record();
 353                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 354                 }
 355         } while_for_each_event_file();
 356         mutex_unlock(&event_mutex);
 357 }
 358
 359 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 360                                          int enable, int soft_disable)
 361 {
 362         struct trace_event_call *call = file->event_call;
 363         struct trace_array *tr = file->tr;
 364         int ret = 0;
 365         int disable;
 366
 367         switch (enable) {
 368         case 0:
 369                 /*
 370                  * When soft_disable is set and enable is cleared, the sm_ref
 371                  * reference counter is decremented. If it reaches 0, we want
 372                  * to clear the SOFT_DISABLED flag but leave the event in the
 373                  * state that it was. That is, if the event was enabled and
 374                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 375                  * is set we do not want the event to be enabled before we
 376                  * clear the bit.
 377                  *
 378                  * When soft_disable is not set but the SOFT_MODE flag is,
 379                  * we do nothing. Do not disable the tracepoint, otherwise
 380                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 381                  */
 382                 if (soft_disable) {
 383                         if (atomic_dec_return(&file->sm_ref) > 0)
 384                                 break;
 385                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 386                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 387                 } else
 388                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 389
 390                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 391                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 392                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 393                                 tracing_stop_cmdline_record();
 394                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 395                         }
 396                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 397                 }
 398                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 399                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 400                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 401                 else
 402                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 403                 break;
 404         case 1:
 405                 /*
 406                  * When soft_disable is set and enable is set, we want to
 407                  * register the tracepoint for the event, but leave the event
 408                  * as is. That means, if the event was already enabled, we do
 409                  * nothing (but set SOFT_MODE). If the event is disabled, we
 410                  * set SOFT_DISABLED before enabling the event tracepoint, so
 411                  * it still seems to be disabled.
 412                  */
 413                 if (!soft_disable)
 414                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 415                 else {
 416                         if (atomic_inc_return(&file->sm_ref) > 1)
 417                                 break;
 418                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 419                 }
 420
 421                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 422
 423                         /* Keep the event disabled, when going to SOFT_MODE. */
 424                         if (soft_disable)
 425                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 426
 427                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 428                                 tracing_start_cmdline_record();
 429                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 430                         }
 431                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 432                         if (ret) {
 433                                 tracing_stop_cmdline_record();
 434                                 pr_info("event trace: Could not enable event "
 435                                         "%s\n", trace_event_name(call));
 436                                 break;
 437                         }
 438                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 439
 440                         /* WAS_ENABLED gets set but never cleared. */
 441                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 442                 }
 443                 break;
 444         }
 445
 446         return ret;
 447 }
 448
 449 int trace_event_enable_disable(struct trace_event_file *file,
 450                                int enable, int soft_disable)
 451 {
 452         return __ftrace_event_enable_disable(file, enable, soft_disable);
 453 }
 454
 455 static int ftrace_event_enable_disable(struct trace_event_file *file,
 456                                        int enable)
 457 {
 458         return __ftrace_event_enable_disable(file, enable, 0);
 459 }
 460
 461 static void ftrace_clear_events(struct trace_array *tr)
 462 {
 463         struct trace_event_file *file;
 464
 465         mutex_lock(&event_mutex);
 466         list_for_each_entry(file, &tr->events, list) {
 467                 ftrace_event_enable_disable(file, 0);
 468         }
 469         mutex_unlock(&event_mutex);
 470 }
 471
 472 static int cmp_pid(const void *key, const void *elt)
 473 {
 474         const pid_t *search_pid = key;
 475         const pid_t *pid = elt;
 476
 477         if (*search_pid == *pid)
 478                 return 0;
 479         if (*search_pid < *pid)
 480                 return -1;
 481         return 1;
 482 }
 483
 484 static bool
 485 check_ignore_pid(struct trace_pid_list *filtered_pids, struct task_struct *task)
 486 {
 487         pid_t search_pid;
 488         pid_t *pid;
 489
 490         /*
 491          * Return false, because if filtered_pids does not exist,
 492          * all pids are good to trace.
 493          */
 494         if (!filtered_pids)
 495                 return false;
 496
 497         search_pid = task->pid;
 498
 499         pid = bsearch(&search_pid, filtered_pids->pids,
 500                       filtered_pids->nr_pids, sizeof(pid_t),
 501                       cmp_pid);
 502         if (!pid)
 503                 return true;
 504
 505         return false;
 506 }
 507
 508 static void
 509 event_filter_pid_sched_switch_probe_pre(void *data,
 510                     struct task_struct *prev, struct task_struct *next)
 511 {
 512         struct trace_array *tr = data;
 513         struct trace_pid_list *pid_list;
 514
 515         pid_list = rcu_dereference_sched(tr->filtered_pids);
 516
 517         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 518                        check_ignore_pid(pid_list, prev) &&
 519                        check_ignore_pid(pid_list, next));
 520 }
 521
 522 static void
 523 event_filter_pid_sched_switch_probe_post(void *data,
 524                     struct task_struct *prev, struct task_struct *next)
 525 {
 526         struct trace_array *tr = data;
 527         struct trace_pid_list *pid_list;
 528
 529         pid_list = rcu_dereference_sched(tr->filtered_pids);
 530
 531         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 532                        check_ignore_pid(pid_list, next));
 533 }
 534
 535 static void
 536 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 537 {
 538         struct trace_array *tr = data;
 539         struct trace_pid_list *pid_list;
 540
 541         /* Nothing to do if we are already tracing */
 542         if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 543                 return;
 544
 545         pid_list = rcu_dereference_sched(tr->filtered_pids);
 546
 547         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 548                        check_ignore_pid(pid_list, task));
 549 }
 550
 551 static void
 552 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 553 {
 554         struct trace_array *tr = data;
 555         struct trace_pid_list *pid_list;
 556
 557         /* Nothing to do if we are not tracing */
 558         if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 559                 return;
 560
 561         pid_list = rcu_dereference_sched(tr->filtered_pids);
 562
 563         /* Set tracing if current is enabled */
 564         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 565                        check_ignore_pid(pid_list, current));
 566 }
 567
 568 static void __ftrace_clear_event_pids(struct trace_array *tr)
 569 {
 570         struct trace_pid_list *pid_list;
 571         struct trace_event_file *file;
 572         int cpu;
 573
 574         pid_list = rcu_dereference_protected(tr->filtered_pids,
 575                                              lockdep_is_held(&event_mutex));
 576         if (!pid_list)
 577                 return;
 578
 579         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 580         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 581
 582         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 583         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 584
 585         list_for_each_entry(file, &tr->events, list) {
 586                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 587         }
 588
 589         for_each_possible_cpu(cpu)
 590                 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 591
 592         rcu_assign_pointer(tr->filtered_pids, NULL);
 593
 594         /* Wait till all users are no longer using pid filtering */
 595         synchronize_sched();
 596
 597         free_pages((unsigned long)pid_list->pids, pid_list->order);
 598         kfree(pid_list);
 599 }
 600
 601 static void ftrace_clear_event_pids(struct trace_array *tr)
 602 {
 603         mutex_lock(&event_mutex);
 604         __ftrace_clear_event_pids(tr);
 605         mutex_unlock(&event_mutex);
 606 }
 607
 608 static void __put_system(struct event_subsystem *system)
 609 {
 610         struct event_filter *filter = system->filter;
 611
 612         WARN_ON_ONCE(system_refcount(system) == 0);
 613         if (system_refcount_dec(system))
 614                 return;
 615
 616         list_del(&system->list);
 617
 618         if (filter) {
 619                 kfree(filter->filter_string);
 620                 kfree(filter);
 621         }
 622         kfree_const(system->name);
 623         kfree(system);
 624 }
 625
 626 static void __get_system(struct event_subsystem *system)
 627 {
 628         WARN_ON_ONCE(system_refcount(system) == 0);
 629         system_refcount_inc(system);
 630 }
 631
 632 static void __get_system_dir(struct trace_subsystem_dir *dir)
 633 {
 634         WARN_ON_ONCE(dir->ref_count == 0);
 635         dir->ref_count++;
 636         __get_system(dir->subsystem);
 637 }
 638
 639 static void __put_system_dir(struct trace_subsystem_dir *dir)
 640 {
 641         WARN_ON_ONCE(dir->ref_count == 0);
 642         /* If the subsystem is about to be freed, the dir must be too */
 643         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 644
 645         __put_system(dir->subsystem);
 646         if (!--dir->ref_count)
 647                 kfree(dir);
 648 }
 649
 650 static void put_system(struct trace_subsystem_dir *dir)
 651 {
 652         mutex_lock(&event_mutex);
 653         __put_system_dir(dir);
 654         mutex_unlock(&event_mutex);
 655 }
 656
 657 static void remove_subsystem(struct trace_subsystem_dir *dir)
 658 {
 659         if (!dir)
 660                 return;
 661
 662         if (!--dir->nr_events) {
 663                 tracefs_remove_recursive(dir->entry);
 664                 list_del(&dir->list);
 665                 __put_system_dir(dir);
 666         }
 667 }
 668
 669 static void remove_event_file_dir(struct trace_event_file *file)
 670 {
 671         struct dentry *dir = file->dir;
 672         struct dentry *child;
 673
 674         if (dir) {
 675                 spin_lock(&dir->d_lock);        /* probably unneeded */
 676                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 677                         if (d_really_is_positive(child))        /* probably unneeded */
 678                                 d_inode(child)->i_private = NULL;
 679                 }
 680                 spin_unlock(&dir->d_lock);
 681
 682                 tracefs_remove_recursive(dir);
 683         }
 684
 685         list_del(&file->list);
 686         remove_subsystem(file->system);
 687         free_event_filter(file->filter);
 688         kmem_cache_free(file_cachep, file);
 689 }
 690
 691 /*
 692  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 693  */
 694 static int
 695 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 696                               const char *sub, const char *event, int set)
 697 {
 698         struct trace_event_file *file;
 699         struct trace_event_call *call;
 700         const char *name;
 701         int ret = -EINVAL;
 702
 703         list_for_each_entry(file, &tr->events, list) {
 704
 705                 call = file->event_call;
 706                 name = trace_event_name(call);
 707
 708                 if (!name || !call->class || !call->class->reg)
 709                         continue;
 710
 711                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 712                         continue;
 713
 714                 if (match &&
 715                     strcmp(match, name) != 0 &&
 716                     strcmp(match, call->class->system) != 0)
 717                         continue;
 718
 719                 if (sub && strcmp(sub, call->class->system) != 0)
 720                         continue;
 721
 722                 if (event && strcmp(event, name) != 0)
 723                         continue;
 724
 725                 ftrace_event_enable_disable(file, set);
 726
 727                 ret = 0;
 728         }
 729
 730         return ret;
 731 }
 732
 733 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 734                                   const char *sub, const char *event, int set)
 735 {
 736         int ret;
 737
 738         mutex_lock(&event_mutex);
 739         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 740         mutex_unlock(&event_mutex);
 741
 742         return ret;
 743 }
 744
 745 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 746 {
 747         char *event = NULL, *sub = NULL, *match;
 748         int ret;
 749
 750         /*
 751          * The buf format can be <subsystem>:<event-name>
 752          *  *:<event-name> means any event by that name.
 753          *  :<event-name> is the same.
 754          *
 755          *  <subsystem>:* means all events in that subsystem
 756          *  <subsystem>: means the same.
 757          *
 758          *  <name> (no ':') means all events in a subsystem with
 759          *  the name <name> or any event that matches <name>
 760          */
 761
 762         match = strsep(&buf, ":");
 763         if (buf) {
 764                 sub = match;
 765                 event = buf;
 766                 match = NULL;
 767
 768                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 769                         sub = NULL;
 770                 if (!strlen(event) || strcmp(event, "*") == 0)
 771                         event = NULL;
 772         }
 773
 774         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 775
 776         /* Put back the colon to allow this to be called again */
 777         if (buf)
 778                 *(buf - 1) = ':';
 779
 780         return ret;
 781 }
 782
 783 /**
 784  * trace_set_clr_event - enable or disable an event
 785  * @system: system name to match (NULL for any system)
 786  * @event: event name to match (NULL for all events, within system)
 787  * @set: 1 to enable, 0 to disable
 788  *
 789  * This is a way for other parts of the kernel to enable or disable
 790  * event recording.
 791  *
 792  * Returns 0 on success, -EINVAL if the parameters do not match any
 793  * registered events.
 794  */
 795 int trace_set_clr_event(const char *system, const char *event, int set)
 796 {
 797         struct trace_array *tr = top_trace_array();
 798
 799         if (!tr)
 800                 return -ENODEV;
 801
 802         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 803 }
 804 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 805
 806 /* 128 should be much more than enough */
 807 #define EVENT_BUF_SIZE          127
 808
 809 static ssize_t
 810 ftrace_event_write(struct file *file, const char __user *ubuf,
 811                    size_t cnt, loff_t *ppos)
 812 {
 813         struct trace_parser parser;
 814         struct seq_file *m = file->private_data;
 815         struct trace_array *tr = m->private;
 816         ssize_t read, ret;
 817
 818         if (!cnt)
 819                 return 0;
 820
 821         ret = tracing_update_buffers();
 822         if (ret < 0)
 823                 return ret;
 824
 825         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 826                 return -ENOMEM;
 827
 828         read = trace_get_user(&parser, ubuf, cnt, ppos);
 829
 830         if (read >= 0 && trace_parser_loaded((&parser))) {
 831                 int set = 1;
 832
 833                 if (*parser.buffer == '!')
 834                         set = 0;
 835
 836                 parser.buffer[parser.idx] = 0;
 837
 838                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 839                 if (ret)
 840                         goto out_put;
 841         }
 842
 843         ret = read;
 844
 845  out_put:
 846         trace_parser_put(&parser);
 847
 848         return ret;
 849 }
 850
 851 static void *
 852 t_next(struct seq_file *m, void *v, loff_t *pos)
 853 {
 854         struct trace_event_file *file = v;
 855         struct trace_event_call *call;
 856         struct trace_array *tr = m->private;
 857
 858         (*pos)++;
 859
 860         list_for_each_entry_continue(file, &tr->events, list) {
 861                 call = file->event_call;
 862                 /*
 863                  * The ftrace subsystem is for showing formats only.
 864                  * They can not be enabled or disabled via the event files.
 865                  */
 866                 if (call->class && call->class->reg)
 867                         return file;
 868         }
 869
 870         return NULL;
 871 }
 872
 873 static void *t_start(struct seq_file *m, loff_t *pos)
 874 {
 875         struct trace_event_file *file;
 876         struct trace_array *tr = m->private;
 877         loff_t l;
 878
 879         mutex_lock(&event_mutex);
 880
 881         file = list_entry(&tr->events, struct trace_event_file, list);
 882         for (l = 0; l <= *pos; ) {
 883                 file = t_next(m, file, &l);
 884                 if (!file)
 885                         break;
 886         }
 887         return file;
 888 }
 889
 890 static void *
 891 s_next(struct seq_file *m, void *v, loff_t *pos)
 892 {
 893         struct trace_event_file *file = v;
 894         struct trace_array *tr = m->private;
 895
 896         (*pos)++;
 897
 898         list_for_each_entry_continue(file, &tr->events, list) {
 899                 if (file->flags & EVENT_FILE_FL_ENABLED)
 900                         return file;
 901         }
 902
 903         return NULL;
 904 }
 905
 906 static void *s_start(struct seq_file *m, loff_t *pos)
 907 {
 908         struct trace_event_file *file;
 909         struct trace_array *tr = m->private;
 910         loff_t l;
 911
 912         mutex_lock(&event_mutex);
 913
 914         file = list_entry(&tr->events, struct trace_event_file, list);
 915         for (l = 0; l <= *pos; ) {
 916                 file = s_next(m, file, &l);
 917                 if (!file)
 918                         break;
 919         }
 920         return file;
 921 }
 922
 923 static int t_show(struct seq_file *m, void *v)
 924 {
 925         struct trace_event_file *file = v;
 926         struct trace_event_call *call = file->event_call;
 927
 928         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 929                 seq_printf(m, "%s:", call->class->system);
 930         seq_printf(m, "%s\n", trace_event_name(call));
 931
 932         return 0;
 933 }
 934
 935 static void t_stop(struct seq_file *m, void *p)
 936 {
 937         mutex_unlock(&event_mutex);
 938 }
 939
 940 static void *p_start(struct seq_file *m, loff_t *pos)
 941 {
 942         struct trace_pid_list *pid_list;
 943         struct trace_array *tr = m->private;
 944
 945         /*
 946          * Grab the mutex, to keep calls to p_next() having the same
 947          * tr->filtered_pids as p_start() has.
 948          * If we just passed the tr->filtered_pids around, then RCU would
 949          * have been enough, but doing that makes things more complex.
 950          */
 951         mutex_lock(&event_mutex);
 952         rcu_read_lock_sched();
 953
 954         pid_list = rcu_dereference_sched(tr->filtered_pids);
 955
 956         if (!pid_list || *pos >= pid_list->nr_pids)
 957                 return NULL;
 958
 959         return (void *)&pid_list->pids[*pos];
 960 }
 961
 962 static void p_stop(struct seq_file *m, void *p)
 963 {
 964         rcu_read_unlock_sched();
 965         mutex_unlock(&event_mutex);
 966 }
 967
 968 static void *
 969 p_next(struct seq_file *m, void *v, loff_t *pos)
 970 {
 971         struct trace_array *tr = m->private;
 972         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 973
 974         (*pos)++;
 975
 976         if (*pos >= pid_list->nr_pids)
 977                 return NULL;
 978
 979         return (void *)&pid_list->pids[*pos];
 980 }
 981
 982 static int p_show(struct seq_file *m, void *v)
 983 {
 984         pid_t *pid = v;
 985
 986         seq_printf(m, "%d\n", *pid);
 987         return 0;
 988 }
 989
 990 static ssize_t
 991 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 992                   loff_t *ppos)
 993 {
 994         struct trace_event_file *file;
 995         unsigned long flags;
 996         char buf[4] = "0";
 997
 998         mutex_lock(&event_mutex);
 999         file = event_file_data(filp);
1000         if (likely(file))
1001                 flags = file->flags;
1002         mutex_unlock(&event_mutex);
1003
1004         if (!file)
1005                 return -ENODEV;
1006
1007         if (flags & EVENT_FILE_FL_ENABLED &&
1008             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1009                 strcpy(buf, "1");
1010
1011         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1012             flags & EVENT_FILE_FL_SOFT_MODE)
1013                 strcat(buf, "*");
1014
1015         strcat(buf, "\n");
1016
1017         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1018 }
1019
1020 static ssize_t
1021 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1022                    loff_t *ppos)
1023 {
1024         struct trace_event_file *file;
1025         unsigned long val;
1026         int ret;
1027
1028         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1029         if (ret)
1030                 return ret;
1031
1032         ret = tracing_update_buffers();
1033         if (ret < 0)
1034                 return ret;
1035
1036         switch (val) {
1037         case 0:
1038         case 1:
1039                 ret = -ENODEV;
1040                 mutex_lock(&event_mutex);
1041                 file = event_file_data(filp);
1042                 if (likely(file))
1043                         ret = ftrace_event_enable_disable(file, val);
1044                 mutex_unlock(&event_mutex);
1045                 break;
1046
1047         default:
1048                 return -EINVAL;
1049         }
1050
1051         *ppos += cnt;
1052
1053         return ret ? ret : cnt;
1054 }
1055
1056 static ssize_t
1057 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1058                    loff_t *ppos)
1059 {
1060         const char set_to_char[4] = { '?', '0', '1', 'X' };
1061         struct trace_subsystem_dir *dir = filp->private_data;
1062         struct event_subsystem *system = dir->subsystem;
1063         struct trace_event_call *call;
1064         struct trace_event_file *file;
1065         struct trace_array *tr = dir->tr;
1066         char buf[2];
1067         int set = 0;
1068         int ret;
1069
1070         mutex_lock(&event_mutex);
1071         list_for_each_entry(file, &tr->events, list) {
1072                 call = file->event_call;
1073                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1074                         continue;
1075
1076                 if (system && strcmp(call->class->system, system->name) != 0)
1077                         continue;
1078
1079                 /*
1080                  * We need to find out if all the events are set
1081                  * or if all events or cleared, or if we have
1082                  * a mixture.
1083                  */
1084                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1085
1086                 /*
1087                  * If we have a mixture, no need to look further.
1088                  */
1089                 if (set == 3)
1090                         break;
1091         }
1092         mutex_unlock(&event_mutex);
1093
1094         buf[0] = set_to_char[set];
1095         buf[1] = '\n';
1096
1097         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1098
1099         return ret;
1100 }
1101
1102 static ssize_t
1103 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1104                     loff_t *ppos)
1105 {
1106         struct trace_subsystem_dir *dir = filp->private_data;
1107         struct event_subsystem *system = dir->subsystem;
1108         const char *name = NULL;
1109         unsigned long val;
1110         ssize_t ret;
1111
1112         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1113         if (ret)
1114                 return ret;
1115
1116         ret = tracing_update_buffers();
1117         if (ret < 0)
1118                 return ret;
1119
1120         if (val != 0 && val != 1)
1121                 return -EINVAL;
1122
1123         /*
1124          * Opening of "enable" adds a ref count to system,
1125          * so the name is safe to use.
1126          */
1127         if (system)
1128                 name = system->name;
1129
1130         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1131         if (ret)
1132                 goto out;
1133
1134         ret = cnt;
1135
1136 out:
1137         *ppos += cnt;
1138
1139         return ret;
1140 }
1141
1142 enum {
1143         FORMAT_HEADER           = 1,
1144         FORMAT_FIELD_SEPERATOR  = 2,
1145         FORMAT_PRINTFMT         = 3,
1146 };
1147
1148 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1149 {
1150         struct trace_event_call *call = event_file_data(m->private);
1151         struct list_head *common_head = &ftrace_common_fields;
1152         struct list_head *head = trace_get_fields(call);
1153         struct list_head *node = v;
1154
1155         (*pos)++;
1156
1157         switch ((unsigned long)v) {
1158         case FORMAT_HEADER:
1159                 node = common_head;
1160                 break;
1161
1162         case FORMAT_FIELD_SEPERATOR:
1163                 node = head;
1164                 break;
1165
1166         case FORMAT_PRINTFMT:
1167                 /* all done */
1168                 return NULL;
1169         }
1170
1171         node = node->prev;
1172         if (node == common_head)
1173                 return (void *)FORMAT_FIELD_SEPERATOR;
1174         else if (node == head)
1175                 return (void *)FORMAT_PRINTFMT;
1176         else
1177                 return node;
1178 }
1179
1180 static int f_show(struct seq_file *m, void *v)
1181 {
1182         struct trace_event_call *call = event_file_data(m->private);
1183         struct ftrace_event_field *field;
1184         const char *array_descriptor;
1185
1186         switch ((unsigned long)v) {
1187         case FORMAT_HEADER:
1188                 seq_printf(m, "name: %s\n", trace_event_name(call));
1189                 seq_printf(m, "ID: %d\n", call->event.type);
1190                 seq_puts(m, "format:\n");
1191                 return 0;
1192
1193         case FORMAT_FIELD_SEPERATOR:
1194                 seq_putc(m, '\n');
1195                 return 0;
1196
1197         case FORMAT_PRINTFMT:
1198                 seq_printf(m, "\nprint fmt: %s\n",
1199                            call->print_fmt);
1200                 return 0;
1201         }
1202
1203         field = list_entry(v, struct ftrace_event_field, link);
1204         /*
1205          * Smartly shows the array type(except dynamic array).
1206          * Normal:
1207          *      field:TYPE VAR
1208          * If TYPE := TYPE[LEN], it is shown:
1209          *      field:TYPE VAR[LEN]
1210          */
1211         array_descriptor = strchr(field->type, '[');
1212
1213         if (!strncmp(field->type, "__data_loc", 10))
1214                 array_descriptor = NULL;
1215
1216         if (!array_descriptor)
1217                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1218                            field->type, field->name, field->offset,
1219                            field->size, !!field->is_signed);
1220         else
1221                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1222                            (int)(array_descriptor - field->type),
1223                            field->type, field->name,
1224                            array_descriptor, field->offset,
1225                            field->size, !!field->is_signed);
1226
1227         return 0;
1228 }
1229
1230 static void *f_start(struct seq_file *m, loff_t *pos)
1231 {
1232         void *p = (void *)FORMAT_HEADER;
1233         loff_t l = 0;
1234
1235         /* ->stop() is called even if ->start() fails */
1236         mutex_lock(&event_mutex);
1237         if (!event_file_data(m->private))
1238                 return ERR_PTR(-ENODEV);
1239
1240         while (l < *pos && p)
1241                 p = f_next(m, p, &l);
1242
1243         return p;
1244 }
1245
1246 static void f_stop(struct seq_file *m, void *p)
1247 {
1248         mutex_unlock(&event_mutex);
1249 }
1250
1251 static const struct seq_operations trace_format_seq_ops = {
1252         .start          = f_start,
1253         .next           = f_next,
1254         .stop           = f_stop,
1255         .show           = f_show,
1256 };
1257
1258 static int trace_format_open(struct inode *inode, struct file *file)
1259 {
1260         struct seq_file *m;
1261         int ret;
1262
1263         ret = seq_open(file, &trace_format_seq_ops);
1264         if (ret < 0)
1265                 return ret;
1266
1267         m = file->private_data;
1268         m->private = file;
1269
1270         return 0;
1271 }
1272
1273 static ssize_t
1274 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1275 {
1276         int id = (long)event_file_data(filp);
1277         char buf[32];
1278         int len;
1279
1280         if (*ppos)
1281                 return 0;
1282
1283         if (unlikely(!id))
1284                 return -ENODEV;
1285
1286         len = sprintf(buf, "%d\n", id);
1287
1288         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1289 }
1290
1291 static ssize_t
1292 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1293                   loff_t *ppos)
1294 {
1295         struct trace_event_file *file;
1296         struct trace_seq *s;
1297         int r = -ENODEV;
1298
1299         if (*ppos)
1300                 return 0;
1301
1302         s = kmalloc(sizeof(*s), GFP_KERNEL);
1303
1304         if (!s)
1305                 return -ENOMEM;
1306
1307         trace_seq_init(s);
1308
1309         mutex_lock(&event_mutex);
1310         file = event_file_data(filp);
1311         if (file)
1312                 print_event_filter(file, s);
1313         mutex_unlock(&event_mutex);
1314
1315         if (file)
1316                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1317                                             s->buffer, trace_seq_used(s));
1318
1319         kfree(s);
1320
1321         return r;
1322 }
1323
1324 static ssize_t
1325 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1326                    loff_t *ppos)
1327 {
1328         struct trace_event_file *file;
1329         char *buf;
1330         int err = -ENODEV;
1331
1332         if (cnt >= PAGE_SIZE)
1333                 return -EINVAL;
1334
1335         buf = (char *)__get_free_page(GFP_TEMPORARY);
1336         if (!buf)
1337                 return -ENOMEM;
1338
1339         if (copy_from_user(buf, ubuf, cnt)) {
1340                 free_page((unsigned long) buf);
1341                 return -EFAULT;
1342         }
1343         buf[cnt] = '\0';
1344
1345         mutex_lock(&event_mutex);
1346         file = event_file_data(filp);
1347         if (file)
1348                 err = apply_event_filter(file, buf);
1349         mutex_unlock(&event_mutex);
1350
1351         free_page((unsigned long) buf);
1352         if (err < 0)
1353                 return err;
1354
1355         *ppos += cnt;
1356
1357         return cnt;
1358 }
1359
1360 static LIST_HEAD(event_subsystems);
1361
1362 static int subsystem_open(struct inode *inode, struct file *filp)
1363 {
1364         struct event_subsystem *system = NULL;
1365         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1366         struct trace_array *tr;
1367         int ret;
1368
1369         if (tracing_is_disabled())
1370                 return -ENODEV;
1371
1372         /* Make sure the system still exists */
1373         mutex_lock(&trace_types_lock);
1374         mutex_lock(&event_mutex);
1375         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1376                 list_for_each_entry(dir, &tr->systems, list) {
1377                         if (dir == inode->i_private) {
1378                                 /* Don't open systems with no events */
1379                                 if (dir->nr_events) {
1380                                         __get_system_dir(dir);
1381                                         system = dir->subsystem;
1382                                 }
1383                                 goto exit_loop;
1384                         }
1385                 }
1386         }
1387  exit_loop:
1388         mutex_unlock(&event_mutex);
1389         mutex_unlock(&trace_types_lock);
1390
1391         if (!system)
1392                 return -ENODEV;
1393
1394         /* Some versions of gcc think dir can be uninitialized here */
1395         WARN_ON(!dir);
1396
1397         /* Still need to increment the ref count of the system */
1398         if (trace_array_get(tr) < 0) {
1399                 put_system(dir);
1400                 return -ENODEV;
1401         }
1402
1403         ret = tracing_open_generic(inode, filp);
1404         if (ret < 0) {
1405                 trace_array_put(tr);
1406                 put_system(dir);
1407         }
1408
1409         return ret;
1410 }
1411
1412 static int system_tr_open(struct inode *inode, struct file *filp)
1413 {
1414         struct trace_subsystem_dir *dir;
1415         struct trace_array *tr = inode->i_private;
1416         int ret;
1417
1418         if (tracing_is_disabled())
1419                 return -ENODEV;
1420
1421         if (trace_array_get(tr) < 0)
1422                 return -ENODEV;
1423
1424         /* Make a temporary dir that has no system but points to tr */
1425         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1426         if (!dir) {
1427                 trace_array_put(tr);
1428                 return -ENOMEM;
1429         }
1430
1431         dir->tr = tr;
1432
1433         ret = tracing_open_generic(inode, filp);
1434         if (ret < 0) {
1435                 trace_array_put(tr);
1436                 kfree(dir);
1437                 return ret;
1438         }
1439
1440         filp->private_data = dir;
1441
1442         return 0;
1443 }
1444
1445 static int subsystem_release(struct inode *inode, struct file *file)
1446 {
1447         struct trace_subsystem_dir *dir = file->private_data;
1448
1449         trace_array_put(dir->tr);
1450
1451         /*
1452          * If dir->subsystem is NULL, then this is a temporary
1453          * descriptor that was made for a trace_array to enable
1454          * all subsystems.
1455          */
1456         if (dir->subsystem)
1457                 put_system(dir);
1458         else
1459                 kfree(dir);
1460
1461         return 0;
1462 }
1463
1464 static ssize_t
1465 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1466                       loff_t *ppos)
1467 {
1468         struct trace_subsystem_dir *dir = filp->private_data;
1469         struct event_subsystem *system = dir->subsystem;
1470         struct trace_seq *s;
1471         int r;
1472
1473         if (*ppos)
1474                 return 0;
1475
1476         s = kmalloc(sizeof(*s), GFP_KERNEL);
1477         if (!s)
1478                 return -ENOMEM;
1479
1480         trace_seq_init(s);
1481
1482         print_subsystem_event_filter(system, s);
1483         r = simple_read_from_buffer(ubuf, cnt, ppos,
1484                                     s->buffer, trace_seq_used(s));
1485
1486         kfree(s);
1487
1488         return r;
1489 }
1490
1491 static ssize_t
1492 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1493                        loff_t *ppos)
1494 {
1495         struct trace_subsystem_dir *dir = filp->private_data;
1496         char *buf;
1497         int err;
1498
1499         if (cnt >= PAGE_SIZE)
1500                 return -EINVAL;
1501
1502         buf = (char *)__get_free_page(GFP_TEMPORARY);
1503         if (!buf)
1504                 return -ENOMEM;
1505
1506         if (copy_from_user(buf, ubuf, cnt)) {
1507                 free_page((unsigned long) buf);
1508                 return -EFAULT;
1509         }
1510         buf[cnt] = '\0';
1511
1512         err = apply_subsystem_event_filter(dir, buf);
1513         free_page((unsigned long) buf);
1514         if (err < 0)
1515                 return err;
1516
1517         *ppos += cnt;
1518
1519         return cnt;
1520 }
1521
1522 static ssize_t
1523 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1524 {
1525         int (*func)(struct trace_seq *s) = filp->private_data;
1526         struct trace_seq *s;
1527         int r;
1528
1529         if (*ppos)
1530                 return 0;
1531
1532         s = kmalloc(sizeof(*s), GFP_KERNEL);
1533         if (!s)
1534                 return -ENOMEM;
1535
1536         trace_seq_init(s);
1537
1538         func(s);
1539         r = simple_read_from_buffer(ubuf, cnt, ppos,
1540                                     s->buffer, trace_seq_used(s));
1541
1542         kfree(s);
1543
1544         return r;
1545 }
1546
1547 static int max_pids(struct trace_pid_list *pid_list)
1548 {
1549         return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1550 }
1551
1552 static ssize_t
1553 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1554                        size_t cnt, loff_t *ppos)
1555 {
1556         struct seq_file *m = filp->private_data;
1557         struct trace_array *tr = m->private;
1558         struct trace_pid_list *filtered_pids = NULL;
1559         struct trace_pid_list *pid_list = NULL;
1560         struct trace_event_file *file;
1561         struct trace_parser parser;
1562         unsigned long val;
1563         loff_t this_pos;
1564         ssize_t read = 0;
1565         ssize_t ret = 0;
1566         pid_t pid;
1567         int i;
1568
1569         if (!cnt)
1570                 return 0;
1571
1572         ret = tracing_update_buffers();
1573         if (ret < 0)
1574                 return ret;
1575
1576         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1577                 return -ENOMEM;
1578
1579         mutex_lock(&event_mutex);
1580         /*
1581          * Load as many pids into the array before doing a
1582          * swap from the tr->filtered_pids to the new list.
1583          */
1584         while (cnt > 0) {
1585
1586                 this_pos = 0;
1587
1588                 ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1589                 if (ret < 0 || !trace_parser_loaded(&parser))
1590                         break;
1591
1592                 read += ret;
1593                 ubuf += ret;
1594                 cnt -= ret;
1595
1596                 parser.buffer[parser.idx] = 0;
1597
1598                 ret = -EINVAL;
1599                 if (kstrtoul(parser.buffer, 0, &val))
1600                         break;
1601                 if (val > INT_MAX)
1602                         break;
1603
1604                 pid = (pid_t)val;
1605
1606                 ret = -ENOMEM;
1607                 if (!pid_list) {
1608                         pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1609                         if (!pid_list)
1610                                 break;
1611
1612                         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1613                                                         lockdep_is_held(&event_mutex));
1614                         if (filtered_pids)
1615                                 pid_list->order = filtered_pids->order;
1616                         else
1617                                 pid_list->order = 0;
1618
1619                         pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1620                                                                   pid_list->order);
1621                         if (!pid_list->pids)
1622                                 break;
1623
1624                         if (filtered_pids) {
1625                                 pid_list->nr_pids = filtered_pids->nr_pids;
1626                                 memcpy(pid_list->pids, filtered_pids->pids,
1627                                        pid_list->nr_pids * sizeof(pid_t));
1628                         } else
1629                                 pid_list->nr_pids = 0;
1630                 }
1631
1632                 if (pid_list->nr_pids >= max_pids(pid_list)) {
1633                         pid_t *pid_page;
1634
1635                         pid_page = (void *)__get_free_pages(GFP_KERNEL,
1636                                                             pid_list->order + 1);
1637                         if (!pid_page)
1638                                 break;
1639                         memcpy(pid_page, pid_list->pids,
1640                                pid_list->nr_pids * sizeof(pid_t));
1641                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1642
1643                         pid_list->order++;
1644                         pid_list->pids = pid_page;
1645                 }
1646
1647                 pid_list->pids[pid_list->nr_pids++] = pid;
1648                 trace_parser_clear(&parser);
1649                 ret = 0;
1650         }
1651         trace_parser_put(&parser);
1652
1653         if (ret < 0) {
1654                 if (pid_list)
1655                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1656                 kfree(pid_list);
1657                 mutex_unlock(&event_mutex);
1658                 return ret;
1659         }
1660
1661         if (!pid_list) {
1662                 mutex_unlock(&event_mutex);
1663                 return ret;
1664         }
1665
1666         sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1667
1668         /* Remove duplicates */
1669         for (i = 1; i < pid_list->nr_pids; i++) {
1670                 int start = i;
1671
1672                 while (i < pid_list->nr_pids &&
1673                        pid_list->pids[i - 1] == pid_list->pids[i])
1674                         i++;
1675
1676                 if (start != i) {
1677                         if (i < pid_list->nr_pids) {
1678                                 memmove(&pid_list->pids[start], &pid_list->pids[i],
1679                                         (pid_list->nr_pids - i) * sizeof(pid_t));
1680                                 pid_list->nr_pids -= i - start;
1681                                 i = start;
1682                         } else
1683                                 pid_list->nr_pids = start;
1684                 }
1685         }
1686
1687         rcu_assign_pointer(tr->filtered_pids, pid_list);
1688
1689         list_for_each_entry(file, &tr->events, list) {
1690                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1691         }
1692
1693         if (filtered_pids) {
1694                 synchronize_sched();
1695
1696                 free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1697                 kfree(filtered_pids);
1698         } else {
1699                 /*
1700                  * Register a probe that is called before all other probes
1701                  * to set ignore_pid if next or prev do not match.
1702                  * Register a probe this is called after all other probes
1703                  * to only keep ignore_pid set if next pid matches.
1704                  */
1705                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1706                                                  tr, INT_MAX);
1707                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1708                                                  tr, 0);
1709
1710                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1711                                                  tr, INT_MAX);
1712                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1713                                                  tr, 0);
1714         }
1715
1716         mutex_unlock(&event_mutex);
1717
1718         ret = read;
1719         *ppos += read;
1720
1721         return ret;
1722 }
1723
1724 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1725 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1726 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1727 static int ftrace_event_release(struct inode *inode, struct file *file);
1728
1729 static const struct seq_operations show_event_seq_ops = {
1730         .start = t_start,
1731         .next = t_next,
1732         .show = t_show,
1733         .stop = t_stop,
1734 };
1735
1736 static const struct seq_operations show_set_event_seq_ops = {
1737         .start = s_start,
1738         .next = s_next,
1739         .show = t_show,
1740         .stop = t_stop,
1741 };
1742
1743 static const struct seq_operations show_set_pid_seq_ops = {
1744         .start = p_start,
1745         .next = p_next,
1746         .show = p_show,
1747         .stop = p_stop,
1748 };
1749
1750 static const struct file_operations ftrace_avail_fops = {
1751         .open = ftrace_event_avail_open,
1752         .read = seq_read,
1753         .llseek = seq_lseek,
1754         .release = seq_release,
1755 };
1756
1757 static const struct file_operations ftrace_set_event_fops = {
1758         .open = ftrace_event_set_open,
1759         .read = seq_read,
1760         .write = ftrace_event_write,
1761         .llseek = seq_lseek,
1762         .release = ftrace_event_release,
1763 };
1764
1765 static const struct file_operations ftrace_set_event_pid_fops = {
1766         .open = ftrace_event_set_pid_open,
1767         .read = seq_read,
1768         .write = ftrace_event_pid_write,
1769         .llseek = seq_lseek,
1770         .release = ftrace_event_release,
1771 };
1772
1773 static const struct file_operations ftrace_enable_fops = {
1774         .open = tracing_open_generic,
1775         .read = event_enable_read,
1776         .write = event_enable_write,
1777         .llseek = default_llseek,
1778 };
1779
1780 static const struct file_operations ftrace_event_format_fops = {
1781         .open = trace_format_open,
1782         .read = seq_read,
1783         .llseek = seq_lseek,
1784         .release = seq_release,
1785 };
1786
1787 static const struct file_operations ftrace_event_id_fops = {
1788         .read = event_id_read,
1789         .llseek = default_llseek,
1790 };
1791
1792 static const struct file_operations ftrace_event_filter_fops = {
1793         .open = tracing_open_generic,
1794         .read = event_filter_read,
1795         .write = event_filter_write,
1796         .llseek = default_llseek,
1797 };
1798
1799 static const struct file_operations ftrace_subsystem_filter_fops = {
1800         .open = subsystem_open,
1801         .read = subsystem_filter_read,
1802         .write = subsystem_filter_write,
1803         .llseek = default_llseek,
1804         .release = subsystem_release,
1805 };
1806
1807 static const struct file_operations ftrace_system_enable_fops = {
1808         .open = subsystem_open,
1809         .read = system_enable_read,
1810         .write = system_enable_write,
1811         .llseek = default_llseek,
1812         .release = subsystem_release,
1813 };
1814
1815 static const struct file_operations ftrace_tr_enable_fops = {
1816         .open = system_tr_open,
1817         .read = system_enable_read,
1818         .write = system_enable_write,
1819         .llseek = default_llseek,
1820         .release = subsystem_release,
1821 };
1822
1823 static const struct file_operations ftrace_show_header_fops = {
1824         .open = tracing_open_generic,
1825         .read = show_header,
1826         .llseek = default_llseek,
1827 };
1828
1829 static int
1830 ftrace_event_open(struct inode *inode, struct file *file,
1831                   const struct seq_operations *seq_ops)
1832 {
1833         struct seq_file *m;
1834         int ret;
1835
1836         ret = seq_open(file, seq_ops);
1837         if (ret < 0)
1838                 return ret;
1839         m = file->private_data;
1840         /* copy tr over to seq ops */
1841         m->private = inode->i_private;
1842
1843         return ret;
1844 }
1845
1846 static int ftrace_event_release(struct inode *inode, struct file *file)
1847 {
1848         struct trace_array *tr = inode->i_private;
1849
1850         trace_array_put(tr);
1851
1852         return seq_release(inode, file);
1853 }
1854
1855 static int
1856 ftrace_event_avail_open(struct inode *inode, struct file *file)
1857 {
1858         const struct seq_operations *seq_ops = &show_event_seq_ops;
1859
1860         return ftrace_event_open(inode, file, seq_ops);
1861 }
1862
1863 static int
1864 ftrace_event_set_open(struct inode *inode, struct file *file)
1865 {
1866         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1867         struct trace_array *tr = inode->i_private;
1868         int ret;
1869
1870         if (trace_array_get(tr) < 0)
1871                 return -ENODEV;
1872
1873         if ((file->f_mode & FMODE_WRITE) &&
1874             (file->f_flags & O_TRUNC))
1875                 ftrace_clear_events(tr);
1876
1877         ret = ftrace_event_open(inode, file, seq_ops);
1878         if (ret < 0)
1879                 trace_array_put(tr);
1880         return ret;
1881 }
1882
1883 static int
1884 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1885 {
1886         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1887         struct trace_array *tr = inode->i_private;
1888         int ret;
1889
1890         if (trace_array_get(tr) < 0)
1891                 return -ENODEV;
1892
1893         if ((file->f_mode & FMODE_WRITE) &&
1894             (file->f_flags & O_TRUNC))
1895                 ftrace_clear_event_pids(tr);
1896
1897         ret = ftrace_event_open(inode, file, seq_ops);
1898         if (ret < 0)
1899                 trace_array_put(tr);
1900         return ret;
1901 }
1902
1903 static struct event_subsystem *
1904 create_new_subsystem(const char *name)
1905 {
1906         struct event_subsystem *system;
1907
1908         /* need to create new entry */
1909         system = kmalloc(sizeof(*system), GFP_KERNEL);
1910         if (!system)
1911                 return NULL;
1912
1913         system->ref_count = 1;
1914
1915         /* Only allocate if dynamic (kprobes and modules) */
1916         system->name = kstrdup_const(name, GFP_KERNEL);
1917         if (!system->name)
1918                 goto out_free;
1919
1920         system->filter = NULL;
1921
1922         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1923         if (!system->filter)
1924                 goto out_free;
1925
1926         list_add(&system->list, &event_subsystems);
1927
1928         return system;
1929
1930  out_free:
1931         kfree_const(system->name);
1932         kfree(system);
1933         return NULL;
1934 }
1935
1936 static struct dentry *
1937 event_subsystem_dir(struct trace_array *tr, const char *name,
1938                     struct trace_event_file *file, struct dentry *parent)
1939 {
1940         struct trace_subsystem_dir *dir;
1941         struct event_subsystem *system;
1942         struct dentry *entry;
1943
1944         /* First see if we did not already create this dir */
1945         list_for_each_entry(dir, &tr->systems, list) {
1946                 system = dir->subsystem;
1947                 if (strcmp(system->name, name) == 0) {
1948                         dir->nr_events++;
1949                         file->system = dir;
1950                         return dir->entry;
1951                 }
1952         }
1953
1954         /* Now see if the system itself exists. */
1955         list_for_each_entry(system, &event_subsystems, list) {
1956                 if (strcmp(system->name, name) == 0)
1957                         break;
1958         }
1959         /* Reset system variable when not found */
1960         if (&system->list == &event_subsystems)
1961                 system = NULL;
1962
1963         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1964         if (!dir)
1965                 goto out_fail;
1966
1967         if (!system) {
1968                 system = create_new_subsystem(name);
1969                 if (!system)
1970                         goto out_free;
1971         } else
1972                 __get_system(system);
1973
1974         dir->entry = tracefs_create_dir(name, parent);
1975         if (!dir->entry) {
1976                 pr_warn("Failed to create system directory %s\n", name);
1977                 __put_system(system);
1978                 goto out_free;
1979         }
1980
1981         dir->tr = tr;
1982         dir->ref_count = 1;
1983         dir->nr_events = 1;
1984         dir->subsystem = system;
1985         file->system = dir;
1986
1987         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1988                                     &ftrace_subsystem_filter_fops);
1989         if (!entry) {
1990                 kfree(system->filter);
1991                 system->filter = NULL;
1992                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1993         }
1994
1995         trace_create_file("enable", 0644, dir->entry, dir,
1996                           &ftrace_system_enable_fops);
1997
1998         list_add(&dir->list, &tr->systems);
1999
2000         return dir->entry;
2001
2002  out_free:
2003         kfree(dir);
2004  out_fail:
2005         /* Only print this message if failed on memory allocation */
2006         if (!dir || !system)
2007                 pr_warn("No memory to create event subsystem %s\n", name);
2008         return NULL;
2009 }
2010
2011 static int
2012 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2013 {
2014         struct trace_event_call *call = file->event_call;
2015         struct trace_array *tr = file->tr;
2016         struct list_head *head;
2017         struct dentry *d_events;
2018         const char *name;
2019         int ret;
2020
2021         /*
2022          * If the trace point header did not define TRACE_SYSTEM
2023          * then the system would be called "TRACE_SYSTEM".
2024          */
2025         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2026                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2027                 if (!d_events)
2028                         return -ENOMEM;
2029         } else
2030                 d_events = parent;
2031
2032         name = trace_event_name(call);
2033         file->dir = tracefs_create_dir(name, d_events);
2034         if (!file->dir) {
2035                 pr_warn("Could not create tracefs '%s' directory\n", name);
2036                 return -1;
2037         }
2038
2039         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2040                 trace_create_file("enable", 0644, file->dir, file,
2041                                   &ftrace_enable_fops);
2042
2043 #ifdef CONFIG_PERF_EVENTS
2044         if (call->event.type && call->class->reg)
2045                 trace_create_file("id", 0444, file->dir,
2046                                   (void *)(long)call->event.type,
2047                                   &ftrace_event_id_fops);
2048 #endif
2049
2050         /*
2051          * Other events may have the same class. Only update
2052          * the fields if they are not already defined.
2053          */
2054         head = trace_get_fields(call);
2055         if (list_empty(head)) {
2056                 ret = call->class->define_fields(call);
2057                 if (ret < 0) {
2058                         pr_warn("Could not initialize trace point events/%s\n",
2059                                 name);
2060                         return -1;
2061                 }
2062         }
2063         trace_create_file("filter", 0644, file->dir, file,
2064                           &ftrace_event_filter_fops);
2065
2066         trace_create_file("trigger", 0644, file->dir, file,
2067                           &event_trigger_fops);
2068
2069         trace_create_file("format", 0444, file->dir, call,
2070                           &ftrace_event_format_fops);
2071
2072         return 0;
2073 }
2074
2075 static void remove_event_from_tracers(struct trace_event_call *call)
2076 {
2077         struct trace_event_file *file;
2078         struct trace_array *tr;
2079
2080         do_for_each_event_file_safe(tr, file) {
2081                 if (file->event_call != call)
2082                         continue;
2083
2084                 remove_event_file_dir(file);
2085                 /*
2086                  * The do_for_each_event_file_safe() is
2087                  * a double loop. After finding the call for this
2088                  * trace_array, we use break to jump to the next
2089                  * trace_array.
2090                  */
2091                 break;
2092         } while_for_each_event_file();
2093 }
2094
2095 static void event_remove(struct trace_event_call *call)
2096 {
2097         struct trace_array *tr;
2098         struct trace_event_file *file;
2099
2100         do_for_each_event_file(tr, file) {
2101                 if (file->event_call != call)
2102                         continue;
2103                 ftrace_event_enable_disable(file, 0);
2104                 /*
2105                  * The do_for_each_event_file() is
2106                  * a double loop. After finding the call for this
2107                  * trace_array, we use break to jump to the next
2108                  * trace_array.
2109                  */
2110                 break;
2111         } while_for_each_event_file();
2112
2113         if (call->event.funcs)
2114                 __unregister_trace_event(&call->event);
2115         remove_event_from_tracers(call);
2116         list_del(&call->list);
2117 }
2118
2119 static int event_init(struct trace_event_call *call)
2120 {
2121         int ret = 0;
2122         const char *name;
2123
2124         name = trace_event_name(call);
2125         if (WARN_ON(!name))
2126                 return -EINVAL;
2127
2128         if (call->class->raw_init) {
2129                 ret = call->class->raw_init(call);
2130                 if (ret < 0 && ret != -ENOSYS)
2131                         pr_warn("Could not initialize trace events/%s\n", name);
2132         }
2133
2134         return ret;
2135 }
2136
2137 static int
2138 __register_event(struct trace_event_call *call, struct module *mod)
2139 {
2140         int ret;
2141
2142         ret = event_init(call);
2143         if (ret < 0)
2144                 return ret;
2145
2146         list_add(&call->list, &ftrace_events);
2147         call->mod = mod;
2148
2149         return 0;
2150 }
2151
2152 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2153 {
2154         int rlen;
2155         int elen;
2156
2157         /* Find the length of the enum value as a string */
2158         elen = snprintf(ptr, 0, "%ld", map->enum_value);
2159         /* Make sure there's enough room to replace the string with the value */
2160         if (len < elen)
2161                 return NULL;
2162
2163         snprintf(ptr, elen + 1, "%ld", map->enum_value);
2164
2165         /* Get the rest of the string of ptr */
2166         rlen = strlen(ptr + len);
2167         memmove(ptr + elen, ptr + len, rlen);
2168         /* Make sure we end the new string */
2169         ptr[elen + rlen] = 0;
2170
2171         return ptr + elen;
2172 }
2173
2174 static void update_event_printk(struct trace_event_call *call,
2175                                 struct trace_enum_map *map)
2176 {
2177         char *ptr;
2178         int quote = 0;
2179         int len = strlen(map->enum_string);
2180
2181         for (ptr = call->print_fmt; *ptr; ptr++) {
2182                 if (*ptr == '\\') {
2183                         ptr++;
2184                         /* paranoid */
2185                         if (!*ptr)
2186                                 break;
2187                         continue;
2188                 }
2189                 if (*ptr == '"') {
2190                         quote ^= 1;
2191                         continue;
2192                 }
2193                 if (quote)
2194                         continue;
2195                 if (isdigit(*ptr)) {
2196                         /* skip numbers */
2197                         do {
2198                                 ptr++;
2199                                 /* Check for alpha chars like ULL */
2200                         } while (isalnum(*ptr));
2201                         if (!*ptr)
2202                                 break;
2203                         /*
2204                          * A number must have some kind of delimiter after
2205                          * it, and we can ignore that too.
2206                          */
2207                         continue;
2208                 }
2209                 if (isalpha(*ptr) || *ptr == '_') {
2210                         if (strncmp(map->enum_string, ptr, len) == 0 &&
2211                             !isalnum(ptr[len]) && ptr[len] != '_') {
2212                                 ptr = enum_replace(ptr, map, len);
2213                                 /* Hmm, enum string smaller than value */
2214                                 if (WARN_ON_ONCE(!ptr))
2215                                         return;
2216                                 /*
2217                                  * No need to decrement here, as enum_replace()
2218                                  * returns the pointer to the character passed
2219                                  * the enum, and two enums can not be placed
2220                                  * back to back without something in between.
2221                                  * We can skip that something in between.
2222                                  */
2223                                 continue;
2224                         }
2225                 skip_more:
2226                         do {
2227                                 ptr++;
2228                         } while (isalnum(*ptr) || *ptr == '_');
2229                         if (!*ptr)
2230                                 break;
2231                         /*
2232                          * If what comes after this variable is a '.' or
2233                          * '->' then we can continue to ignore that string.
2234                          */
2235                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2236                                 ptr += *ptr == '.' ? 1 : 2;
2237                                 if (!*ptr)
2238                                         break;
2239                                 goto skip_more;
2240                         }
2241                         /*
2242                          * Once again, we can skip the delimiter that came
2243                          * after the string.
2244                          */
2245                         continue;
2246                 }
2247         }
2248 }
2249
2250 void trace_event_enum_update(struct trace_enum_map **map, int len)
2251 {
2252         struct trace_event_call *call, *p;
2253         const char *last_system = NULL;
2254         int last_i;
2255         int i;
2256
2257         down_write(&trace_event_sem);
2258         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2259                 /* events are usually grouped together with systems */
2260                 if (!last_system || call->class->system != last_system) {
2261                         last_i = 0;
2262                         last_system = call->class->system;
2263                 }
2264
2265                 for (i = last_i; i < len; i++) {
2266                         if (call->class->system == map[i]->system) {
2267                                 /* Save the first system if need be */
2268                                 if (!last_i)
2269                                         last_i = i;
2270                                 update_event_printk(call, map[i]);
2271                         }
2272                 }
2273         }
2274         up_write(&trace_event_sem);
2275 }
2276
2277 static struct trace_event_file *
2278 trace_create_new_event(struct trace_event_call *call,
2279                        struct trace_array *tr)
2280 {
2281         struct trace_event_file *file;
2282
2283         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2284         if (!file)
2285                 return NULL;
2286
2287         file->event_call = call;
2288         file->tr = tr;
2289         atomic_set(&file->sm_ref, 0);
2290         atomic_set(&file->tm_ref, 0);
2291         INIT_LIST_HEAD(&file->triggers);
2292         list_add(&file->list, &tr->events);
2293
2294         return file;
2295 }
2296
2297 /* Add an event to a trace directory */
2298 static int
2299 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2300 {
2301         struct trace_event_file *file;
2302
2303         file = trace_create_new_event(call, tr);
2304         if (!file)
2305                 return -ENOMEM;
2306
2307         return event_create_dir(tr->event_dir, file);
2308 }
2309
2310 /*
2311  * Just create a decriptor for early init. A descriptor is required
2312  * for enabling events at boot. We want to enable events before
2313  * the filesystem is initialized.
2314  */
2315 static __init int
2316 __trace_early_add_new_event(struct trace_event_call *call,
2317                             struct trace_array *tr)
2318 {
2319         struct trace_event_file *file;
2320
2321         file = trace_create_new_event(call, tr);
2322         if (!file)
2323                 return -ENOMEM;
2324
2325         return 0;
2326 }
2327
2328 struct ftrace_module_file_ops;
2329 static void __add_event_to_tracers(struct trace_event_call *call);
2330
2331 /* Add an additional event_call dynamically */
2332 int trace_add_event_call(struct trace_event_call *call)
2333 {
2334         int ret;
2335         mutex_lock(&trace_types_lock);
2336         mutex_lock(&event_mutex);
2337
2338         ret = __register_event(call, NULL);
2339         if (ret >= 0)
2340                 __add_event_to_tracers(call);
2341
2342         mutex_unlock(&event_mutex);
2343         mutex_unlock(&trace_types_lock);
2344         return ret;
2345 }
2346
2347 /*
2348  * Must be called under locking of trace_types_lock, event_mutex and
2349  * trace_event_sem.
2350  */
2351 static void __trace_remove_event_call(struct trace_event_call *call)
2352 {
2353         event_remove(call);
2354         trace_destroy_fields(call);
2355         free_event_filter(call->filter);
2356         call->filter = NULL;
2357 }
2358
2359 static int probe_remove_event_call(struct trace_event_call *call)
2360 {
2361         struct trace_array *tr;
2362         struct trace_event_file *file;
2363
2364 #ifdef CONFIG_PERF_EVENTS
2365         if (call->perf_refcount)
2366                 return -EBUSY;
2367 #endif
2368         do_for_each_event_file(tr, file) {
2369                 if (file->event_call != call)
2370                         continue;
2371                 /*
2372                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2373                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2374                  * TRACE_REG_UNREGISTER.
2375                  */
2376                 if (file->flags & EVENT_FILE_FL_ENABLED)
2377                         return -EBUSY;
2378                 /*
2379                  * The do_for_each_event_file_safe() is
2380                  * a double loop. After finding the call for this
2381                  * trace_array, we use break to jump to the next
2382                  * trace_array.
2383                  */
2384                 break;
2385         } while_for_each_event_file();
2386
2387         __trace_remove_event_call(call);
2388
2389         return 0;
2390 }
2391
2392 /* Remove an event_call */
2393 int trace_remove_event_call(struct trace_event_call *call)
2394 {
2395         int ret;
2396
2397         mutex_lock(&trace_types_lock);
2398         mutex_lock(&event_mutex);
2399         down_write(&trace_event_sem);
2400         ret = probe_remove_event_call(call);
2401         up_write(&trace_event_sem);
2402         mutex_unlock(&event_mutex);
2403         mutex_unlock(&trace_types_lock);
2404
2405         return ret;
2406 }
2407
2408 #define for_each_event(event, start, end)                       \
2409         for (event = start;                                     \
2410              (unsigned long)event < (unsigned long)end;         \
2411              event++)
2412
2413 #ifdef CONFIG_MODULES
2414
2415 static void trace_module_add_events(struct module *mod)
2416 {
2417         struct trace_event_call **call, **start, **end;
2418
2419         if (!mod->num_trace_events)
2420                 return;
2421
2422         /* Don't add infrastructure for mods without tracepoints */
2423         if (trace_module_has_bad_taint(mod)) {
2424                 pr_err("%s: module has bad taint, not creating trace events\n",
2425                        mod->name);
2426                 return;
2427         }
2428
2429         start = mod->trace_events;
2430         end = mod->trace_events + mod->num_trace_events;
2431
2432         for_each_event(call, start, end) {
2433                 __register_event(*call, mod);
2434                 __add_event_to_tracers(*call);
2435         }
2436 }
2437
2438 static void trace_module_remove_events(struct module *mod)
2439 {
2440         struct trace_event_call *call, *p;
2441         bool clear_trace = false;
2442
2443         down_write(&trace_event_sem);
2444         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2445                 if (call->mod == mod) {
2446                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2447                                 clear_trace = true;
2448                         __trace_remove_event_call(call);
2449                 }
2450         }
2451         up_write(&trace_event_sem);
2452
2453         /*
2454          * It is safest to reset the ring buffer if the module being unloaded
2455          * registered any events that were used. The only worry is if
2456          * a new module gets loaded, and takes on the same id as the events
2457          * of this module. When printing out the buffer, traced events left
2458          * over from this module may be passed to the new module events and
2459          * unexpected results may occur.
2460          */
2461         if (clear_trace)
2462                 tracing_reset_all_online_cpus();
2463 }
2464
2465 static int trace_module_notify(struct notifier_block *self,
2466                                unsigned long val, void *data)
2467 {
2468         struct module *mod = data;
2469
2470         mutex_lock(&trace_types_lock);
2471         mutex_lock(&event_mutex);
2472         switch (val) {
2473         case MODULE_STATE_COMING:
2474                 trace_module_add_events(mod);
2475                 break;
2476         case MODULE_STATE_GOING:
2477                 trace_module_remove_events(mod);
2478                 break;
2479         }
2480         mutex_unlock(&event_mutex);
2481         mutex_unlock(&trace_types_lock);
2482
2483         return 0;
2484 }
2485
2486 static struct notifier_block trace_module_nb = {
2487         .notifier_call = trace_module_notify,
2488         .priority = 1, /* higher than trace.c module notify */
2489 };
2490 #endif /* CONFIG_MODULES */
2491
2492 /* Create a new event directory structure for a trace directory. */
2493 static void
2494 __trace_add_event_dirs(struct trace_array *tr)
2495 {
2496         struct trace_event_call *call;
2497         int ret;
2498
2499         list_for_each_entry(call, &ftrace_events, list) {
2500                 ret = __trace_add_new_event(call, tr);
2501                 if (ret < 0)
2502                         pr_warn("Could not create directory for event %s\n",
2503                                 trace_event_name(call));
2504         }
2505 }
2506
2507 struct trace_event_file *
2508 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2509 {
2510         struct trace_event_file *file;
2511         struct trace_event_call *call;
2512         const char *name;
2513
2514         list_for_each_entry(file, &tr->events, list) {
2515
2516                 call = file->event_call;
2517                 name = trace_event_name(call);
2518
2519                 if (!name || !call->class || !call->class->reg)
2520                         continue;
2521
2522                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2523                         continue;
2524
2525                 if (strcmp(event, name) == 0 &&
2526                     strcmp(system, call->class->system) == 0)
2527                         return file;
2528         }
2529         return NULL;
2530 }
2531
2532 #ifdef CONFIG_DYNAMIC_FTRACE
2533
2534 /* Avoid typos */
2535 #define ENABLE_EVENT_STR        "enable_event"
2536 #define DISABLE_EVENT_STR       "disable_event"
2537
2538 struct event_probe_data {
2539         struct trace_event_file *file;
2540         unsigned long                   count;
2541         int                             ref;
2542         bool                            enable;
2543 };
2544
2545 static void
2546 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2547 {
2548         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2549         struct event_probe_data *data = *pdata;
2550
2551         if (!data)
2552                 return;
2553
2554         if (data->enable)
2555                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2556         else
2557                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2558 }
2559
2560 static void
2561 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2562 {
2563         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2564         struct event_probe_data *data = *pdata;
2565
2566         if (!data)
2567                 return;
2568
2569         if (!data->count)
2570                 return;
2571
2572         /* Skip if the event is in a state we want to switch to */
2573         if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2574                 return;
2575
2576         if (data->count != -1)
2577                 (data->count)--;
2578
2579         event_enable_probe(ip, parent_ip, _data);
2580 }
2581
2582 static int
2583 event_enable_print(struct seq_file *m, unsigned long ip,
2584                       struct ftrace_probe_ops *ops, void *_data)
2585 {
2586         struct event_probe_data *data = _data;
2587
2588         seq_printf(m, "%ps:", (void *)ip);
2589
2590         seq_printf(m, "%s:%s:%s",
2591                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2592                    data->file->event_call->class->system,
2593                    trace_event_name(data->file->event_call));
2594
2595         if (data->count == -1)
2596                 seq_puts(m, ":unlimited\n");
2597         else
2598                 seq_printf(m, ":count=%ld\n", data->count);
2599
2600         return 0;
2601 }
2602
2603 static int
2604 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2605                   void **_data)
2606 {
2607         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2608         struct event_probe_data *data = *pdata;
2609
2610         data->ref++;
2611         return 0;
2612 }
2613
2614 static void
2615 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2616                   void **_data)
2617 {
2618         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2619         struct event_probe_data *data = *pdata;
2620
2621         if (WARN_ON_ONCE(data->ref <= 0))
2622                 return;
2623
2624         data->ref--;
2625         if (!data->ref) {
2626                 /* Remove the SOFT_MODE flag */
2627                 __ftrace_event_enable_disable(data->file, 0, 1);
2628                 module_put(data->file->event_call->mod);
2629                 kfree(data);
2630         }
2631         *pdata = NULL;
2632 }
2633
2634 static struct ftrace_probe_ops event_enable_probe_ops = {
2635         .func                   = event_enable_probe,
2636         .print                  = event_enable_print,
2637         .init                   = event_enable_init,
2638         .free                   = event_enable_free,
2639 };
2640
2641 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2642         .func                   = event_enable_count_probe,
2643         .print                  = event_enable_print,
2644         .init                   = event_enable_init,
2645         .free                   = event_enable_free,
2646 };
2647
2648 static struct ftrace_probe_ops event_disable_probe_ops = {
2649         .func                   = event_enable_probe,
2650         .print                  = event_enable_print,
2651         .init                   = event_enable_init,
2652         .free                   = event_enable_free,
2653 };
2654
2655 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2656         .func                   = event_enable_count_probe,
2657         .print                  = event_enable_print,
2658         .init                   = event_enable_init,
2659         .free                   = event_enable_free,
2660 };
2661
2662 static int
2663 event_enable_func(struct ftrace_hash *hash,
2664                   char *glob, char *cmd, char *param, int enabled)
2665 {
2666         struct trace_array *tr = top_trace_array();
2667         struct trace_event_file *file;
2668         struct ftrace_probe_ops *ops;
2669         struct event_probe_data *data;
2670         const char *system;
2671         const char *event;
2672         char *number;
2673         bool enable;
2674         int ret;
2675
2676         if (!tr)
2677                 return -ENODEV;
2678
2679         /* hash funcs only work with set_ftrace_filter */
2680         if (!enabled || !param)
2681                 return -EINVAL;
2682
2683         system = strsep(&param, ":");
2684         if (!param)
2685                 return -EINVAL;
2686
2687         event = strsep(&param, ":");
2688
2689         mutex_lock(&event_mutex);
2690
2691         ret = -EINVAL;
2692         file = find_event_file(tr, system, event);
2693         if (!file)
2694                 goto out;
2695
2696         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2697
2698         if (enable)
2699                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2700         else
2701                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2702
2703         if (glob[0] == '!') {
2704                 unregister_ftrace_function_probe_func(glob+1, ops);
2705                 ret = 0;
2706                 goto out;
2707         }
2708
2709         ret = -ENOMEM;
2710         data = kzalloc(sizeof(*data), GFP_KERNEL);
2711         if (!data)
2712                 goto out;
2713
2714         data->enable = enable;
2715         data->count = -1;
2716         data->file = file;
2717
2718         if (!param)
2719                 goto out_reg;
2720
2721         number = strsep(&param, ":");
2722
2723         ret = -EINVAL;
2724         if (!strlen(number))
2725                 goto out_free;
2726
2727         /*
2728          * We use the callback data field (which is a pointer)
2729          * as our counter.
2730          */
2731         ret = kstrtoul(number, 0, &data->count);
2732         if (ret)
2733                 goto out_free;
2734
2735  out_reg:
2736         /* Don't let event modules unload while probe registered */
2737         ret = try_module_get(file->event_call->mod);
2738         if (!ret) {
2739                 ret = -EBUSY;
2740                 goto out_free;
2741         }
2742
2743         ret = __ftrace_event_enable_disable(file, 1, 1);
2744         if (ret < 0)
2745                 goto out_put;
2746         ret = register_ftrace_function_probe(glob, ops, data);
2747         /*
2748          * The above returns on success the # of functions enabled,
2749          * but if it didn't find any functions it returns zero.
2750          * Consider no functions a failure too.
2751          */
2752         if (!ret) {
2753                 ret = -ENOENT;
2754                 goto out_disable;
2755         } else if (ret < 0)
2756                 goto out_disable;
2757         /* Just return zero, not the number of enabled functions */
2758         ret = 0;
2759  out:
2760         mutex_unlock(&event_mutex);
2761         return ret;
2762
2763  out_disable:
2764         __ftrace_event_enable_disable(file, 0, 1);
2765  out_put:
2766         module_put(file->event_call->mod);
2767  out_free:
2768         kfree(data);
2769         goto out;
2770 }
2771
2772 static struct ftrace_func_command event_enable_cmd = {
2773         .name                   = ENABLE_EVENT_STR,
2774         .func                   = event_enable_func,
2775 };
2776
2777 static struct ftrace_func_command event_disable_cmd = {
2778         .name                   = DISABLE_EVENT_STR,
2779         .func                   = event_enable_func,
2780 };
2781
2782 static __init int register_event_cmds(void)
2783 {
2784         int ret;
2785
2786         ret = register_ftrace_command(&event_enable_cmd);
2787         if (WARN_ON(ret < 0))
2788                 return ret;
2789         ret = register_ftrace_command(&event_disable_cmd);
2790         if (WARN_ON(ret < 0))
2791                 unregister_ftrace_command(&event_enable_cmd);
2792         return ret;
2793 }
2794 #else
2795 static inline int register_event_cmds(void) { return 0; }
2796 #endif /* CONFIG_DYNAMIC_FTRACE */
2797
2798 /*
2799  * The top level array has already had its trace_event_file
2800  * descriptors created in order to allow for early events to
2801  * be recorded. This function is called after the tracefs has been
2802  * initialized, and we now have to create the files associated
2803  * to the events.
2804  */
2805 static __init void
2806 __trace_early_add_event_dirs(struct trace_array *tr)
2807 {
2808         struct trace_event_file *file;
2809         int ret;
2810
2811
2812         list_for_each_entry(file, &tr->events, list) {
2813                 ret = event_create_dir(tr->event_dir, file);
2814                 if (ret < 0)
2815                         pr_warn("Could not create directory for event %s\n",
2816                                 trace_event_name(file->event_call));
2817         }
2818 }
2819
2820 /*
2821  * For early boot up, the top trace array requires to have
2822  * a list of events that can be enabled. This must be done before
2823  * the filesystem is set up in order to allow events to be traced
2824  * early.
2825  */
2826 static __init void
2827 __trace_early_add_events(struct trace_array *tr)
2828 {
2829         struct trace_event_call *call;
2830         int ret;
2831
2832         list_for_each_entry(call, &ftrace_events, list) {
2833                 /* Early boot up should not have any modules loaded */
2834                 if (WARN_ON_ONCE(call->mod))
2835                         continue;
2836
2837                 ret = __trace_early_add_new_event(call, tr);
2838                 if (ret < 0)
2839                         pr_warn("Could not create early event %s\n",
2840                                 trace_event_name(call));
2841         }
2842 }
2843
2844 /* Remove the event directory structure for a trace directory. */
2845 static void
2846 __trace_remove_event_dirs(struct trace_array *tr)
2847 {
2848         struct trace_event_file *file, *next;
2849
2850         list_for_each_entry_safe(file, next, &tr->events, list)
2851                 remove_event_file_dir(file);
2852 }
2853
2854 static void __add_event_to_tracers(struct trace_event_call *call)
2855 {
2856         struct trace_array *tr;
2857
2858         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2859                 __trace_add_new_event(call, tr);
2860 }
2861
2862 extern struct trace_event_call *__start_ftrace_events[];
2863 extern struct trace_event_call *__stop_ftrace_events[];
2864
2865 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2866
2867 static __init int setup_trace_event(char *str)
2868 {
2869         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2870         ring_buffer_expanded = true;
2871         tracing_selftest_disabled = true;
2872
2873         return 1;
2874 }
2875 __setup("trace_event=", setup_trace_event);
2876
2877 /* Expects to have event_mutex held when called */
2878 static int
2879 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2880 {
2881         struct dentry *d_events;
2882         struct dentry *entry;
2883
2884         entry = tracefs_create_file("set_event", 0644, parent,
2885                                     tr, &ftrace_set_event_fops);
2886         if (!entry) {
2887                 pr_warn("Could not create tracefs 'set_event' entry\n");
2888                 return -ENOMEM;
2889         }
2890
2891         d_events = tracefs_create_dir("events", parent);
2892         if (!d_events) {
2893                 pr_warn("Could not create tracefs 'events' directory\n");
2894                 return -ENOMEM;
2895         }
2896
2897         entry = tracefs_create_file("set_event_pid", 0644, parent,
2898                                     tr, &ftrace_set_event_pid_fops);
2899
2900         /* ring buffer internal formats */
2901         trace_create_file("header_page", 0444, d_events,
2902                           ring_buffer_print_page_header,
2903                           &ftrace_show_header_fops);
2904
2905         trace_create_file("header_event", 0444, d_events,
2906                           ring_buffer_print_entry_header,
2907                           &ftrace_show_header_fops);
2908
2909         trace_create_file("enable", 0644, d_events,
2910                           tr, &ftrace_tr_enable_fops);
2911
2912         tr->event_dir = d_events;
2913
2914         return 0;
2915 }
2916
2917 /**
2918  * event_trace_add_tracer - add a instance of a trace_array to events
2919  * @parent: The parent dentry to place the files/directories for events in
2920  * @tr: The trace array associated with these events
2921  *
2922  * When a new instance is created, it needs to set up its events
2923  * directory, as well as other files associated with events. It also
2924  * creates the event hierachry in the @parent/events directory.
2925  *
2926  * Returns 0 on success.
2927  */
2928 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2929 {
2930         int ret;
2931
2932         mutex_lock(&event_mutex);
2933
2934         ret = create_event_toplevel_files(parent, tr);
2935         if (ret)
2936                 goto out_unlock;
2937
2938         down_write(&trace_event_sem);
2939         __trace_add_event_dirs(tr);
2940         up_write(&trace_event_sem);
2941
2942  out_unlock:
2943         mutex_unlock(&event_mutex);
2944
2945         return ret;
2946 }
2947
2948 /*
2949  * The top trace array already had its file descriptors created.
2950  * Now the files themselves need to be created.
2951  */
2952 static __init int
2953 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2954 {
2955         int ret;
2956
2957         mutex_lock(&event_mutex);
2958
2959         ret = create_event_toplevel_files(parent, tr);
2960         if (ret)
2961                 goto out_unlock;
2962
2963         down_write(&trace_event_sem);
2964         __trace_early_add_event_dirs(tr);
2965         up_write(&trace_event_sem);
2966
2967  out_unlock:
2968         mutex_unlock(&event_mutex);
2969
2970         return ret;
2971 }
2972
2973 int event_trace_del_tracer(struct trace_array *tr)
2974 {
2975         mutex_lock(&event_mutex);
2976
2977         /* Disable any event triggers and associated soft-disabled events */
2978         clear_event_triggers(tr);
2979
2980         /* Clear the pid list */
2981         __ftrace_clear_event_pids(tr);
2982
2983         /* Disable any running events */
2984         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2985
2986         /* Access to events are within rcu_read_lock_sched() */
2987         synchronize_sched();
2988
2989         down_write(&trace_event_sem);
2990         __trace_remove_event_dirs(tr);
2991         tracefs_remove_recursive(tr->event_dir);
2992         up_write(&trace_event_sem);
2993
2994         tr->event_dir = NULL;
2995
2996         mutex_unlock(&event_mutex);
2997
2998         return 0;
2999 }
3000
3001 static __init int event_trace_memsetup(void)
3002 {
3003         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3004         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3005         return 0;
3006 }
3007
3008 static __init void
3009 early_enable_events(struct trace_array *tr, bool disable_first)
3010 {
3011         char *buf = bootup_event_buf;
3012         char *token;
3013         int ret;
3014
3015         while (true) {
3016                 token = strsep(&buf, ",");
3017
3018                 if (!token)
3019                         break;
3020                 if (!*token)
3021                         continue;
3022
3023                 /* Restarting syscalls requires that we stop them first */
3024                 if (disable_first)
3025                         ftrace_set_clr_event(tr, token, 0);
3026
3027                 ret = ftrace_set_clr_event(tr, token, 1);
3028                 if (ret)
3029                         pr_warn("Failed to enable trace event: %s\n", token);
3030
3031                 /* Put back the comma to allow this to be called again */
3032                 if (buf)
3033                         *(buf - 1) = ',';
3034         }
3035 }
3036
3037 static __init int event_trace_enable(void)
3038 {
3039         struct trace_array *tr = top_trace_array();
3040         struct trace_event_call **iter, *call;
3041         int ret;
3042
3043         if (!tr)
3044                 return -ENODEV;
3045
3046         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3047
3048                 call = *iter;
3049                 ret = event_init(call);
3050                 if (!ret)
3051                         list_add(&call->list, &ftrace_events);
3052         }
3053
3054         /*
3055          * We need the top trace array to have a working set of trace
3056          * points at early init, before the debug files and directories
3057          * are created. Create the file entries now, and attach them
3058          * to the actual file dentries later.
3059          */
3060         __trace_early_add_events(tr);
3061
3062         early_enable_events(tr, false);
3063
3064         trace_printk_start_comm();
3065
3066         register_event_cmds();
3067
3068         register_trigger_cmds();
3069
3070         return 0;
3071 }
3072
3073 /*
3074  * event_trace_enable() is called from trace_event_init() first to
3075  * initialize events and perhaps start any events that are on the
3076  * command line. Unfortunately, there are some events that will not
3077  * start this early, like the system call tracepoints that need
3078  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3079  * is called before pid 1 starts, and this flag is never set, making
3080  * the syscall tracepoint never get reached, but the event is enabled
3081  * regardless (and not doing anything).
3082  */
3083 static __init int event_trace_enable_again(void)
3084 {
3085         struct trace_array *tr;
3086
3087         tr = top_trace_array();
3088         if (!tr)
3089                 return -ENODEV;
3090
3091         early_enable_events(tr, true);
3092
3093         return 0;
3094 }
3095
3096 early_initcall(event_trace_enable_again);
3097
3098 static __init int event_trace_init(void)
3099 {
3100         struct trace_array *tr;
3101         struct dentry *d_tracer;
3102         struct dentry *entry;
3103         int ret;
3104
3105         tr = top_trace_array();
3106         if (!tr)
3107                 return -ENODEV;
3108
3109         d_tracer = tracing_init_dentry();
3110         if (IS_ERR(d_tracer))
3111                 return 0;
3112
3113         entry = tracefs_create_file("available_events", 0444, d_tracer,
3114                                     tr, &ftrace_avail_fops);
3115         if (!entry)
3116                 pr_warn("Could not create tracefs 'available_events' entry\n");
3117
3118         if (trace_define_generic_fields())
3119                 pr_warn("tracing: Failed to allocated generic fields");
3120
3121         if (trace_define_common_fields())
3122                 pr_warn("tracing: Failed to allocate common fields");
3123
3124         ret = early_event_add_tracer(d_tracer, tr);
3125         if (ret)
3126                 return ret;
3127
3128 #ifdef CONFIG_MODULES
3129         ret = register_module_notifier(&trace_module_nb);
3130         if (ret)
3131                 pr_warn("Failed to register trace events module notifier\n");
3132 #endif
3133         return 0;
3134 }
3135
3136 void __init trace_event_init(void)
3137 {
3138         event_trace_memsetup();
3139         init_ftrace_syscalls();
3140         event_trace_enable();
3141 }
3142
3143 fs_initcall(event_trace_init);
3144
3145 #ifdef CONFIG_FTRACE_STARTUP_TEST
3146
3147 static DEFINE_SPINLOCK(test_spinlock);
3148 static DEFINE_SPINLOCK(test_spinlock_irq);
3149 static DEFINE_MUTEX(test_mutex);
3150
3151 static __init void test_work(struct work_struct *dummy)
3152 {
3153         spin_lock(&test_spinlock);
3154         spin_lock_irq(&test_spinlock_irq);
3155         udelay(1);
3156         spin_unlock_irq(&test_spinlock_irq);
3157         spin_unlock(&test_spinlock);
3158
3159         mutex_lock(&test_mutex);
3160         msleep(1);
3161         mutex_unlock(&test_mutex);
3162 }
3163
3164 static __init int event_test_thread(void *unused)
3165 {
3166         void *test_malloc;
3167
3168         test_malloc = kmalloc(1234, GFP_KERNEL);
3169         if (!test_malloc)
3170                 pr_info("failed to kmalloc\n");
3171
3172         schedule_on_each_cpu(test_work);
3173
3174         kfree(test_malloc);
3175
3176         set_current_state(TASK_INTERRUPTIBLE);
3177         while (!kthread_should_stop()) {
3178                 schedule();
3179                 set_current_state(TASK_INTERRUPTIBLE);
3180         }
3181         __set_current_state(TASK_RUNNING);
3182
3183         return 0;
3184 }
3185
3186 /*
3187  * Do various things that may trigger events.
3188  */
3189 static __init void event_test_stuff(void)
3190 {
3191         struct task_struct *test_thread;
3192
3193         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3194         msleep(1);
3195         kthread_stop(test_thread);
3196 }
3197
3198 /*
3199  * For every trace event defined, we will test each trace point separately,
3200  * and then by groups, and finally all trace points.
3201  */
3202 static __init void event_trace_self_tests(void)
3203 {
3204         struct trace_subsystem_dir *dir;
3205         struct trace_event_file *file;
3206         struct trace_event_call *call;
3207         struct event_subsystem *system;
3208         struct trace_array *tr;
3209         int ret;
3210
3211         tr = top_trace_array();
3212         if (!tr)
3213                 return;
3214
3215         pr_info("Running tests on trace events:\n");
3216
3217         list_for_each_entry(file, &tr->events, list) {
3218
3219                 call = file->event_call;
3220
3221                 /* Only test those that have a probe */
3222                 if (!call->class || !call->class->probe)
3223                         continue;
3224
3225 /*
3226  * Testing syscall events here is pretty useless, but
3227  * we still do it if configured. But this is time consuming.
3228  * What we really need is a user thread to perform the
3229  * syscalls as we test.
3230  */
3231 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3232                 if (call->class->system &&
3233                     strcmp(call->class->system, "syscalls") == 0)
3234                         continue;
3235 #endif
3236
3237                 pr_info("Testing event %s: ", trace_event_name(call));
3238
3239                 /*
3240                  * If an event is already enabled, someone is using
3241                  * it and the self test should not be on.
3242                  */
3243                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3244                         pr_warn("Enabled event during self test!\n");
3245                         WARN_ON_ONCE(1);
3246                         continue;
3247                 }
3248
3249                 ftrace_event_enable_disable(file, 1);
3250                 event_test_stuff();
3251                 ftrace_event_enable_disable(file, 0);
3252
3253                 pr_cont("OK\n");
3254         }
3255
3256         /* Now test at the sub system level */
3257
3258         pr_info("Running tests on trace event systems:\n");
3259
3260         list_for_each_entry(dir, &tr->systems, list) {
3261
3262                 system = dir->subsystem;
3263
3264                 /* the ftrace system is special, skip it */
3265                 if (strcmp(system->name, "ftrace") == 0)
3266                         continue;
3267
3268                 pr_info("Testing event system %s: ", system->name);
3269
3270                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3271                 if (WARN_ON_ONCE(ret)) {
3272                         pr_warn("error enabling system %s\n",
3273                                 system->name);
3274                         continue;
3275                 }
3276
3277                 event_test_stuff();
3278
3279                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3280                 if (WARN_ON_ONCE(ret)) {
3281                         pr_warn("error disabling system %s\n",
3282                                 system->name);
3283                         continue;
3284                 }
3285
3286                 pr_cont("OK\n");
3287         }
3288
3289         /* Test with all events enabled */
3290
3291         pr_info("Running tests on all trace events:\n");
3292         pr_info("Testing all events: ");
3293
3294         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3295         if (WARN_ON_ONCE(ret)) {
3296                 pr_warn("error enabling all events\n");
3297                 return;
3298         }
3299
3300         event_test_stuff();
3301
3302         /* reset sysname */
3303         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3304         if (WARN_ON_ONCE(ret)) {
3305                 pr_warn("error disabling all events\n");
3306                 return;
3307         }
3308
3309         pr_cont("OK\n");
3310 }
3311
3312 #ifdef CONFIG_FUNCTION_TRACER
3313
3314 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3315
3316 static struct trace_array *event_tr;
3317
3318 static void __init
3319 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3320                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3321 {
3322         struct ring_buffer_event *event;
3323         struct ring_buffer *buffer;
3324         struct ftrace_entry *entry;
3325         unsigned long flags;
3326         long disabled;
3327         int cpu;
3328         int pc;
3329
3330         pc = preempt_count();
3331         preempt_disable_notrace();
3332         cpu = raw_smp_processor_id();
3333         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3334
3335         if (disabled != 1)
3336                 goto out;
3337
3338         local_save_flags(flags);
3339
3340         event = trace_current_buffer_lock_reserve(&buffer,
3341                                                   TRACE_FN, sizeof(*entry),
3342                                                   flags, pc);
3343         if (!event)
3344                 goto out;
3345         entry   = ring_buffer_event_data(event);
3346         entry->ip                       = ip;
3347         entry->parent_ip                = parent_ip;
3348
3349         trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3350
3351  out:
3352         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3353         preempt_enable_notrace();
3354 }
3355
3356 static struct ftrace_ops trace_ops __initdata  =
3357 {
3358         .func = function_test_events_call,
3359         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3360 };
3361
3362 static __init void event_trace_self_test_with_function(void)
3363 {
3364         int ret;
3365         event_tr = top_trace_array();
3366         if (WARN_ON(!event_tr))
3367                 return;
3368         ret = register_ftrace_function(&trace_ops);
3369         if (WARN_ON(ret < 0)) {
3370                 pr_info("Failed to enable function tracer for event tests\n");
3371                 return;
3372         }
3373         pr_info("Running tests again, along with the function tracer\n");
3374         event_trace_self_tests();
3375         unregister_ftrace_function(&trace_ops);
3376 }
3377 #else
3378 static __init void event_trace_self_test_with_function(void)
3379 {
3380 }
3381 #endif
3382
3383 static __init int event_trace_self_tests_init(void)
3384 {
3385         if (!tracing_selftest_disabled) {
3386                 event_trace_self_tests();
3387                 event_trace_self_test_with_function();
3388         }
3389
3390         return 0;
3391 }
3392
3393 late_initcall(event_trace_self_tests_init);
3394
3395 #endif