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         __acquires(RCU)
 942 {
 943         struct trace_pid_list *pid_list;
 944         struct trace_array *tr = m->private;
 945
 946         /*
 947          * Grab the mutex, to keep calls to p_next() having the same
 948          * tr->filtered_pids as p_start() has.
 949          * If we just passed the tr->filtered_pids around, then RCU would
 950          * have been enough, but doing that makes things more complex.
 951          */
 952         mutex_lock(&event_mutex);
 953         rcu_read_lock_sched();
 954
 955         pid_list = rcu_dereference_sched(tr->filtered_pids);
 956
 957         if (!pid_list || *pos >= pid_list->nr_pids)
 958                 return NULL;
 959
 960         return (void *)&pid_list->pids[*pos];
 961 }
 962
 963 static void p_stop(struct seq_file *m, void *p)
 964         __releases(RCU)
 965 {
 966         rcu_read_unlock_sched();
 967         mutex_unlock(&event_mutex);
 968 }
 969
 970 static void *
 971 p_next(struct seq_file *m, void *v, loff_t *pos)
 972 {
 973         struct trace_array *tr = m->private;
 974         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 975
 976         (*pos)++;
 977
 978         if (*pos >= pid_list->nr_pids)
 979                 return NULL;
 980
 981         return (void *)&pid_list->pids[*pos];
 982 }
 983
 984 static int p_show(struct seq_file *m, void *v)
 985 {
 986         pid_t *pid = v;
 987
 988         seq_printf(m, "%d\n", *pid);
 989         return 0;
 990 }
 991
 992 static ssize_t
 993 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 994                   loff_t *ppos)
 995 {
 996         struct trace_event_file *file;
 997         unsigned long flags;
 998         char buf[4] = "0";
 999
1000         mutex_lock(&event_mutex);
1001         file = event_file_data(filp);
1002         if (likely(file))
1003                 flags = file->flags;
1004         mutex_unlock(&event_mutex);
1005
1006         if (!file)
1007                 return -ENODEV;
1008
1009         if (flags & EVENT_FILE_FL_ENABLED &&
1010             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1011                 strcpy(buf, "1");
1012
1013         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1014             flags & EVENT_FILE_FL_SOFT_MODE)
1015                 strcat(buf, "*");
1016
1017         strcat(buf, "\n");
1018
1019         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1020 }
1021
1022 static ssize_t
1023 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1024                    loff_t *ppos)
1025 {
1026         struct trace_event_file *file;
1027         unsigned long val;
1028         int ret;
1029
1030         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1031         if (ret)
1032                 return ret;
1033
1034         ret = tracing_update_buffers();
1035         if (ret < 0)
1036                 return ret;
1037
1038         switch (val) {
1039         case 0:
1040         case 1:
1041                 ret = -ENODEV;
1042                 mutex_lock(&event_mutex);
1043                 file = event_file_data(filp);
1044                 if (likely(file))
1045                         ret = ftrace_event_enable_disable(file, val);
1046                 mutex_unlock(&event_mutex);
1047                 break;
1048
1049         default:
1050                 return -EINVAL;
1051         }
1052
1053         *ppos += cnt;
1054
1055         return ret ? ret : cnt;
1056 }
1057
1058 static ssize_t
1059 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1060                    loff_t *ppos)
1061 {
1062         const char set_to_char[4] = { '?', '0', '1', 'X' };
1063         struct trace_subsystem_dir *dir = filp->private_data;
1064         struct event_subsystem *system = dir->subsystem;
1065         struct trace_event_call *call;
1066         struct trace_event_file *file;
1067         struct trace_array *tr = dir->tr;
1068         char buf[2];
1069         int set = 0;
1070         int ret;
1071
1072         mutex_lock(&event_mutex);
1073         list_for_each_entry(file, &tr->events, list) {
1074                 call = file->event_call;
1075                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1076                         continue;
1077
1078                 if (system && strcmp(call->class->system, system->name) != 0)
1079                         continue;
1080
1081                 /*
1082                  * We need to find out if all the events are set
1083                  * or if all events or cleared, or if we have
1084                  * a mixture.
1085                  */
1086                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1087
1088                 /*
1089                  * If we have a mixture, no need to look further.
1090                  */
1091                 if (set == 3)
1092                         break;
1093         }
1094         mutex_unlock(&event_mutex);
1095
1096         buf[0] = set_to_char[set];
1097         buf[1] = '\n';
1098
1099         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1100
1101         return ret;
1102 }
1103
1104 static ssize_t
1105 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1106                     loff_t *ppos)
1107 {
1108         struct trace_subsystem_dir *dir = filp->private_data;
1109         struct event_subsystem *system = dir->subsystem;
1110         const char *name = NULL;
1111         unsigned long val;
1112         ssize_t ret;
1113
1114         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1115         if (ret)
1116                 return ret;
1117
1118         ret = tracing_update_buffers();
1119         if (ret < 0)
1120                 return ret;
1121
1122         if (val != 0 && val != 1)
1123                 return -EINVAL;
1124
1125         /*
1126          * Opening of "enable" adds a ref count to system,
1127          * so the name is safe to use.
1128          */
1129         if (system)
1130                 name = system->name;
1131
1132         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1133         if (ret)
1134                 goto out;
1135
1136         ret = cnt;
1137
1138 out:
1139         *ppos += cnt;
1140
1141         return ret;
1142 }
1143
1144 enum {
1145         FORMAT_HEADER           = 1,
1146         FORMAT_FIELD_SEPERATOR  = 2,
1147         FORMAT_PRINTFMT         = 3,
1148 };
1149
1150 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1151 {
1152         struct trace_event_call *call = event_file_data(m->private);
1153         struct list_head *common_head = &ftrace_common_fields;
1154         struct list_head *head = trace_get_fields(call);
1155         struct list_head *node = v;
1156
1157         (*pos)++;
1158
1159         switch ((unsigned long)v) {
1160         case FORMAT_HEADER:
1161                 node = common_head;
1162                 break;
1163
1164         case FORMAT_FIELD_SEPERATOR:
1165                 node = head;
1166                 break;
1167
1168         case FORMAT_PRINTFMT:
1169                 /* all done */
1170                 return NULL;
1171         }
1172
1173         node = node->prev;
1174         if (node == common_head)
1175                 return (void *)FORMAT_FIELD_SEPERATOR;
1176         else if (node == head)
1177                 return (void *)FORMAT_PRINTFMT;
1178         else
1179                 return node;
1180 }
1181
1182 static int f_show(struct seq_file *m, void *v)
1183 {
1184         struct trace_event_call *call = event_file_data(m->private);
1185         struct ftrace_event_field *field;
1186         const char *array_descriptor;
1187
1188         switch ((unsigned long)v) {
1189         case FORMAT_HEADER:
1190                 seq_printf(m, "name: %s\n", trace_event_name(call));
1191                 seq_printf(m, "ID: %d\n", call->event.type);
1192                 seq_puts(m, "format:\n");
1193                 return 0;
1194
1195         case FORMAT_FIELD_SEPERATOR:
1196                 seq_putc(m, '\n');
1197                 return 0;
1198
1199         case FORMAT_PRINTFMT:
1200                 seq_printf(m, "\nprint fmt: %s\n",
1201                            call->print_fmt);
1202                 return 0;
1203         }
1204
1205         field = list_entry(v, struct ftrace_event_field, link);
1206         /*
1207          * Smartly shows the array type(except dynamic array).
1208          * Normal:
1209          *      field:TYPE VAR
1210          * If TYPE := TYPE[LEN], it is shown:
1211          *      field:TYPE VAR[LEN]
1212          */
1213         array_descriptor = strchr(field->type, '[');
1214
1215         if (!strncmp(field->type, "__data_loc", 10))
1216                 array_descriptor = NULL;
1217
1218         if (!array_descriptor)
1219                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1220                            field->type, field->name, field->offset,
1221                            field->size, !!field->is_signed);
1222         else
1223                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1224                            (int)(array_descriptor - field->type),
1225                            field->type, field->name,
1226                            array_descriptor, field->offset,
1227                            field->size, !!field->is_signed);
1228
1229         return 0;
1230 }
1231
1232 static void *f_start(struct seq_file *m, loff_t *pos)
1233 {
1234         void *p = (void *)FORMAT_HEADER;
1235         loff_t l = 0;
1236
1237         /* ->stop() is called even if ->start() fails */
1238         mutex_lock(&event_mutex);
1239         if (!event_file_data(m->private))
1240                 return ERR_PTR(-ENODEV);
1241
1242         while (l < *pos && p)
1243                 p = f_next(m, p, &l);
1244
1245         return p;
1246 }
1247
1248 static void f_stop(struct seq_file *m, void *p)
1249 {
1250         mutex_unlock(&event_mutex);
1251 }
1252
1253 static const struct seq_operations trace_format_seq_ops = {
1254         .start          = f_start,
1255         .next           = f_next,
1256         .stop           = f_stop,
1257         .show           = f_show,
1258 };
1259
1260 static int trace_format_open(struct inode *inode, struct file *file)
1261 {
1262         struct seq_file *m;
1263         int ret;
1264
1265         ret = seq_open(file, &trace_format_seq_ops);
1266         if (ret < 0)
1267                 return ret;
1268
1269         m = file->private_data;
1270         m->private = file;
1271
1272         return 0;
1273 }
1274
1275 static ssize_t
1276 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1277 {
1278         int id = (long)event_file_data(filp);
1279         char buf[32];
1280         int len;
1281
1282         if (*ppos)
1283                 return 0;
1284
1285         if (unlikely(!id))
1286                 return -ENODEV;
1287
1288         len = sprintf(buf, "%d\n", id);
1289
1290         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1291 }
1292
1293 static ssize_t
1294 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1295                   loff_t *ppos)
1296 {
1297         struct trace_event_file *file;
1298         struct trace_seq *s;
1299         int r = -ENODEV;
1300
1301         if (*ppos)
1302                 return 0;
1303
1304         s = kmalloc(sizeof(*s), GFP_KERNEL);
1305
1306         if (!s)
1307                 return -ENOMEM;
1308
1309         trace_seq_init(s);
1310
1311         mutex_lock(&event_mutex);
1312         file = event_file_data(filp);
1313         if (file)
1314                 print_event_filter(file, s);
1315         mutex_unlock(&event_mutex);
1316
1317         if (file)
1318                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1319                                             s->buffer, trace_seq_used(s));
1320
1321         kfree(s);
1322
1323         return r;
1324 }
1325
1326 static ssize_t
1327 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1328                    loff_t *ppos)
1329 {
1330         struct trace_event_file *file;
1331         char *buf;
1332         int err = -ENODEV;
1333
1334         if (cnt >= PAGE_SIZE)
1335                 return -EINVAL;
1336
1337         buf = (char *)__get_free_page(GFP_TEMPORARY);
1338         if (!buf)
1339                 return -ENOMEM;
1340
1341         if (copy_from_user(buf, ubuf, cnt)) {
1342                 free_page((unsigned long) buf);
1343                 return -EFAULT;
1344         }
1345         buf[cnt] = '\0';
1346
1347         mutex_lock(&event_mutex);
1348         file = event_file_data(filp);
1349         if (file)
1350                 err = apply_event_filter(file, buf);
1351         mutex_unlock(&event_mutex);
1352
1353         free_page((unsigned long) buf);
1354         if (err < 0)
1355                 return err;
1356
1357         *ppos += cnt;
1358
1359         return cnt;
1360 }
1361
1362 static LIST_HEAD(event_subsystems);
1363
1364 static int subsystem_open(struct inode *inode, struct file *filp)
1365 {
1366         struct event_subsystem *system = NULL;
1367         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1368         struct trace_array *tr;
1369         int ret;
1370
1371         if (tracing_is_disabled())
1372                 return -ENODEV;
1373
1374         /* Make sure the system still exists */
1375         mutex_lock(&trace_types_lock);
1376         mutex_lock(&event_mutex);
1377         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1378                 list_for_each_entry(dir, &tr->systems, list) {
1379                         if (dir == inode->i_private) {
1380                                 /* Don't open systems with no events */
1381                                 if (dir->nr_events) {
1382                                         __get_system_dir(dir);
1383                                         system = dir->subsystem;
1384                                 }
1385                                 goto exit_loop;
1386                         }
1387                 }
1388         }
1389  exit_loop:
1390         mutex_unlock(&event_mutex);
1391         mutex_unlock(&trace_types_lock);
1392
1393         if (!system)
1394                 return -ENODEV;
1395
1396         /* Some versions of gcc think dir can be uninitialized here */
1397         WARN_ON(!dir);
1398
1399         /* Still need to increment the ref count of the system */
1400         if (trace_array_get(tr) < 0) {
1401                 put_system(dir);
1402                 return -ENODEV;
1403         }
1404
1405         ret = tracing_open_generic(inode, filp);
1406         if (ret < 0) {
1407                 trace_array_put(tr);
1408                 put_system(dir);
1409         }
1410
1411         return ret;
1412 }
1413
1414 static int system_tr_open(struct inode *inode, struct file *filp)
1415 {
1416         struct trace_subsystem_dir *dir;
1417         struct trace_array *tr = inode->i_private;
1418         int ret;
1419
1420         if (tracing_is_disabled())
1421                 return -ENODEV;
1422
1423         if (trace_array_get(tr) < 0)
1424                 return -ENODEV;
1425
1426         /* Make a temporary dir that has no system but points to tr */
1427         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1428         if (!dir) {
1429                 trace_array_put(tr);
1430                 return -ENOMEM;
1431         }
1432
1433         dir->tr = tr;
1434
1435         ret = tracing_open_generic(inode, filp);
1436         if (ret < 0) {
1437                 trace_array_put(tr);
1438                 kfree(dir);
1439                 return ret;
1440         }
1441
1442         filp->private_data = dir;
1443
1444         return 0;
1445 }
1446
1447 static int subsystem_release(struct inode *inode, struct file *file)
1448 {
1449         struct trace_subsystem_dir *dir = file->private_data;
1450
1451         trace_array_put(dir->tr);
1452
1453         /*
1454          * If dir->subsystem is NULL, then this is a temporary
1455          * descriptor that was made for a trace_array to enable
1456          * all subsystems.
1457          */
1458         if (dir->subsystem)
1459                 put_system(dir);
1460         else
1461                 kfree(dir);
1462
1463         return 0;
1464 }
1465
1466 static ssize_t
1467 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1468                       loff_t *ppos)
1469 {
1470         struct trace_subsystem_dir *dir = filp->private_data;
1471         struct event_subsystem *system = dir->subsystem;
1472         struct trace_seq *s;
1473         int r;
1474
1475         if (*ppos)
1476                 return 0;
1477
1478         s = kmalloc(sizeof(*s), GFP_KERNEL);
1479         if (!s)
1480                 return -ENOMEM;
1481
1482         trace_seq_init(s);
1483
1484         print_subsystem_event_filter(system, s);
1485         r = simple_read_from_buffer(ubuf, cnt, ppos,
1486                                     s->buffer, trace_seq_used(s));
1487
1488         kfree(s);
1489
1490         return r;
1491 }
1492
1493 static ssize_t
1494 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1495                        loff_t *ppos)
1496 {
1497         struct trace_subsystem_dir *dir = filp->private_data;
1498         char *buf;
1499         int err;
1500
1501         if (cnt >= PAGE_SIZE)
1502                 return -EINVAL;
1503
1504         buf = (char *)__get_free_page(GFP_TEMPORARY);
1505         if (!buf)
1506                 return -ENOMEM;
1507
1508         if (copy_from_user(buf, ubuf, cnt)) {
1509                 free_page((unsigned long) buf);
1510                 return -EFAULT;
1511         }
1512         buf[cnt] = '\0';
1513
1514         err = apply_subsystem_event_filter(dir, buf);
1515         free_page((unsigned long) buf);
1516         if (err < 0)
1517                 return err;
1518
1519         *ppos += cnt;
1520
1521         return cnt;
1522 }
1523
1524 static ssize_t
1525 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1526 {
1527         int (*func)(struct trace_seq *s) = filp->private_data;
1528         struct trace_seq *s;
1529         int r;
1530
1531         if (*ppos)
1532                 return 0;
1533
1534         s = kmalloc(sizeof(*s), GFP_KERNEL);
1535         if (!s)
1536                 return -ENOMEM;
1537
1538         trace_seq_init(s);
1539
1540         func(s);
1541         r = simple_read_from_buffer(ubuf, cnt, ppos,
1542                                     s->buffer, trace_seq_used(s));
1543
1544         kfree(s);
1545
1546         return r;
1547 }
1548
1549 static int max_pids(struct trace_pid_list *pid_list)
1550 {
1551         return (PAGE_SIZE << pid_list->order) / sizeof(pid_t);
1552 }
1553
1554 static void ignore_task_cpu(void *data)
1555 {
1556         struct trace_array *tr = data;
1557         struct trace_pid_list *pid_list;
1558
1559         /*
1560          * This function is called by on_each_cpu() while the
1561          * event_mutex is held.
1562          */
1563         pid_list = rcu_dereference_protected(tr->filtered_pids,
1564                                              mutex_is_locked(&event_mutex));
1565
1566         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1567                        check_ignore_pid(pid_list, current));
1568 }
1569
1570 static ssize_t
1571 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1572                        size_t cnt, loff_t *ppos)
1573 {
1574         struct seq_file *m = filp->private_data;
1575         struct trace_array *tr = m->private;
1576         struct trace_pid_list *filtered_pids = NULL;
1577         struct trace_pid_list *pid_list = NULL;
1578         struct trace_event_file *file;
1579         struct trace_parser parser;
1580         unsigned long val;
1581         loff_t this_pos;
1582         ssize_t read = 0;
1583         ssize_t ret = 0;
1584         pid_t pid;
1585         int i;
1586
1587         if (!cnt)
1588                 return 0;
1589
1590         ret = tracing_update_buffers();
1591         if (ret < 0)
1592                 return ret;
1593
1594         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
1595                 return -ENOMEM;
1596
1597         mutex_lock(&event_mutex);
1598         /*
1599          * Load as many pids into the array before doing a
1600          * swap from the tr->filtered_pids to the new list.
1601          */
1602         while (cnt > 0) {
1603
1604                 this_pos = 0;
1605
1606                 ret = trace_get_user(&parser, ubuf, cnt, &this_pos);
1607                 if (ret < 0 || !trace_parser_loaded(&parser))
1608                         break;
1609
1610                 read += ret;
1611                 ubuf += ret;
1612                 cnt -= ret;
1613
1614                 parser.buffer[parser.idx] = 0;
1615
1616                 ret = -EINVAL;
1617                 if (kstrtoul(parser.buffer, 0, &val))
1618                         break;
1619                 if (val > INT_MAX)
1620                         break;
1621
1622                 pid = (pid_t)val;
1623
1624                 ret = -ENOMEM;
1625                 if (!pid_list) {
1626                         pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
1627                         if (!pid_list)
1628                                 break;
1629
1630                         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1631                                                         lockdep_is_held(&event_mutex));
1632                         if (filtered_pids)
1633                                 pid_list->order = filtered_pids->order;
1634                         else
1635                                 pid_list->order = 0;
1636
1637                         pid_list->pids = (void *)__get_free_pages(GFP_KERNEL,
1638                                                                   pid_list->order);
1639                         if (!pid_list->pids)
1640                                 break;
1641
1642                         if (filtered_pids) {
1643                                 pid_list->nr_pids = filtered_pids->nr_pids;
1644                                 memcpy(pid_list->pids, filtered_pids->pids,
1645                                        pid_list->nr_pids * sizeof(pid_t));
1646                         } else
1647                                 pid_list->nr_pids = 0;
1648                 }
1649
1650                 if (pid_list->nr_pids >= max_pids(pid_list)) {
1651                         pid_t *pid_page;
1652
1653                         pid_page = (void *)__get_free_pages(GFP_KERNEL,
1654                                                             pid_list->order + 1);
1655                         if (!pid_page)
1656                                 break;
1657                         memcpy(pid_page, pid_list->pids,
1658                                pid_list->nr_pids * sizeof(pid_t));
1659                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1660
1661                         pid_list->order++;
1662                         pid_list->pids = pid_page;
1663                 }
1664
1665                 pid_list->pids[pid_list->nr_pids++] = pid;
1666                 trace_parser_clear(&parser);
1667                 ret = 0;
1668         }
1669         trace_parser_put(&parser);
1670
1671         if (ret < 0) {
1672                 if (pid_list)
1673                         free_pages((unsigned long)pid_list->pids, pid_list->order);
1674                 kfree(pid_list);
1675                 mutex_unlock(&event_mutex);
1676                 return ret;
1677         }
1678
1679         if (!pid_list) {
1680                 mutex_unlock(&event_mutex);
1681                 return ret;
1682         }
1683
1684         sort(pid_list->pids, pid_list->nr_pids, sizeof(pid_t), cmp_pid, NULL);
1685
1686         /* Remove duplicates */
1687         for (i = 1; i < pid_list->nr_pids; i++) {
1688                 int start = i;
1689
1690                 while (i < pid_list->nr_pids &&
1691                        pid_list->pids[i - 1] == pid_list->pids[i])
1692                         i++;
1693
1694                 if (start != i) {
1695                         if (i < pid_list->nr_pids) {
1696                                 memmove(&pid_list->pids[start], &pid_list->pids[i],
1697                                         (pid_list->nr_pids - i) * sizeof(pid_t));
1698                                 pid_list->nr_pids -= i - start;
1699                                 i = start;
1700                         } else
1701                                 pid_list->nr_pids = start;
1702                 }
1703         }
1704
1705         rcu_assign_pointer(tr->filtered_pids, pid_list);
1706
1707         list_for_each_entry(file, &tr->events, list) {
1708                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1709         }
1710
1711         if (filtered_pids) {
1712                 synchronize_sched();
1713
1714                 free_pages((unsigned long)filtered_pids->pids, filtered_pids->order);
1715                 kfree(filtered_pids);
1716         } else {
1717                 /*
1718                  * Register a probe that is called before all other probes
1719                  * to set ignore_pid if next or prev do not match.
1720                  * Register a probe this is called after all other probes
1721                  * to only keep ignore_pid set if next pid matches.
1722                  */
1723                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1724                                                  tr, INT_MAX);
1725                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1726                                                  tr, 0);
1727
1728                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1729                                                  tr, INT_MAX);
1730                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1731                                                  tr, 0);
1732
1733                 /*
1734                  * Ignoring of pids is done at task switch. But we have to
1735                  * check for those tasks that are currently running.
1736                  */
1737                 on_each_cpu(ignore_task_cpu, tr, 1);
1738         }
1739
1740         mutex_unlock(&event_mutex);
1741
1742         ret = read;
1743         *ppos += read;
1744
1745         return ret;
1746 }
1747
1748 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1749 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1750 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1751 static int ftrace_event_release(struct inode *inode, struct file *file);
1752
1753 static const struct seq_operations show_event_seq_ops = {
1754         .start = t_start,
1755         .next = t_next,
1756         .show = t_show,
1757         .stop = t_stop,
1758 };
1759
1760 static const struct seq_operations show_set_event_seq_ops = {
1761         .start = s_start,
1762         .next = s_next,
1763         .show = t_show,
1764         .stop = t_stop,
1765 };
1766
1767 static const struct seq_operations show_set_pid_seq_ops = {
1768         .start = p_start,
1769         .next = p_next,
1770         .show = p_show,
1771         .stop = p_stop,
1772 };
1773
1774 static const struct file_operations ftrace_avail_fops = {
1775         .open = ftrace_event_avail_open,
1776         .read = seq_read,
1777         .llseek = seq_lseek,
1778         .release = seq_release,
1779 };
1780
1781 static const struct file_operations ftrace_set_event_fops = {
1782         .open = ftrace_event_set_open,
1783         .read = seq_read,
1784         .write = ftrace_event_write,
1785         .llseek = seq_lseek,
1786         .release = ftrace_event_release,
1787 };
1788
1789 static const struct file_operations ftrace_set_event_pid_fops = {
1790         .open = ftrace_event_set_pid_open,
1791         .read = seq_read,
1792         .write = ftrace_event_pid_write,
1793         .llseek = seq_lseek,
1794         .release = ftrace_event_release,
1795 };
1796
1797 static const struct file_operations ftrace_enable_fops = {
1798         .open = tracing_open_generic,
1799         .read = event_enable_read,
1800         .write = event_enable_write,
1801         .llseek = default_llseek,
1802 };
1803
1804 static const struct file_operations ftrace_event_format_fops = {
1805         .open = trace_format_open,
1806         .read = seq_read,
1807         .llseek = seq_lseek,
1808         .release = seq_release,
1809 };
1810
1811 static const struct file_operations ftrace_event_id_fops = {
1812         .read = event_id_read,
1813         .llseek = default_llseek,
1814 };
1815
1816 static const struct file_operations ftrace_event_filter_fops = {
1817         .open = tracing_open_generic,
1818         .read = event_filter_read,
1819         .write = event_filter_write,
1820         .llseek = default_llseek,
1821 };
1822
1823 static const struct file_operations ftrace_subsystem_filter_fops = {
1824         .open = subsystem_open,
1825         .read = subsystem_filter_read,
1826         .write = subsystem_filter_write,
1827         .llseek = default_llseek,
1828         .release = subsystem_release,
1829 };
1830
1831 static const struct file_operations ftrace_system_enable_fops = {
1832         .open = subsystem_open,
1833         .read = system_enable_read,
1834         .write = system_enable_write,
1835         .llseek = default_llseek,
1836         .release = subsystem_release,
1837 };
1838
1839 static const struct file_operations ftrace_tr_enable_fops = {
1840         .open = system_tr_open,
1841         .read = system_enable_read,
1842         .write = system_enable_write,
1843         .llseek = default_llseek,
1844         .release = subsystem_release,
1845 };
1846
1847 static const struct file_operations ftrace_show_header_fops = {
1848         .open = tracing_open_generic,
1849         .read = show_header,
1850         .llseek = default_llseek,
1851 };
1852
1853 static int
1854 ftrace_event_open(struct inode *inode, struct file *file,
1855                   const struct seq_operations *seq_ops)
1856 {
1857         struct seq_file *m;
1858         int ret;
1859
1860         ret = seq_open(file, seq_ops);
1861         if (ret < 0)
1862                 return ret;
1863         m = file->private_data;
1864         /* copy tr over to seq ops */
1865         m->private = inode->i_private;
1866
1867         return ret;
1868 }
1869
1870 static int ftrace_event_release(struct inode *inode, struct file *file)
1871 {
1872         struct trace_array *tr = inode->i_private;
1873
1874         trace_array_put(tr);
1875
1876         return seq_release(inode, file);
1877 }
1878
1879 static int
1880 ftrace_event_avail_open(struct inode *inode, struct file *file)
1881 {
1882         const struct seq_operations *seq_ops = &show_event_seq_ops;
1883
1884         return ftrace_event_open(inode, file, seq_ops);
1885 }
1886
1887 static int
1888 ftrace_event_set_open(struct inode *inode, struct file *file)
1889 {
1890         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1891         struct trace_array *tr = inode->i_private;
1892         int ret;
1893
1894         if (trace_array_get(tr) < 0)
1895                 return -ENODEV;
1896
1897         if ((file->f_mode & FMODE_WRITE) &&
1898             (file->f_flags & O_TRUNC))
1899                 ftrace_clear_events(tr);
1900
1901         ret = ftrace_event_open(inode, file, seq_ops);
1902         if (ret < 0)
1903                 trace_array_put(tr);
1904         return ret;
1905 }
1906
1907 static int
1908 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1909 {
1910         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1911         struct trace_array *tr = inode->i_private;
1912         int ret;
1913
1914         if (trace_array_get(tr) < 0)
1915                 return -ENODEV;
1916
1917         if ((file->f_mode & FMODE_WRITE) &&
1918             (file->f_flags & O_TRUNC))
1919                 ftrace_clear_event_pids(tr);
1920
1921         ret = ftrace_event_open(inode, file, seq_ops);
1922         if (ret < 0)
1923                 trace_array_put(tr);
1924         return ret;
1925 }
1926
1927 static struct event_subsystem *
1928 create_new_subsystem(const char *name)
1929 {
1930         struct event_subsystem *system;
1931
1932         /* need to create new entry */
1933         system = kmalloc(sizeof(*system), GFP_KERNEL);
1934         if (!system)
1935                 return NULL;
1936
1937         system->ref_count = 1;
1938
1939         /* Only allocate if dynamic (kprobes and modules) */
1940         system->name = kstrdup_const(name, GFP_KERNEL);
1941         if (!system->name)
1942                 goto out_free;
1943
1944         system->filter = NULL;
1945
1946         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1947         if (!system->filter)
1948                 goto out_free;
1949
1950         list_add(&system->list, &event_subsystems);
1951
1952         return system;
1953
1954  out_free:
1955         kfree_const(system->name);
1956         kfree(system);
1957         return NULL;
1958 }
1959
1960 static struct dentry *
1961 event_subsystem_dir(struct trace_array *tr, const char *name,
1962                     struct trace_event_file *file, struct dentry *parent)
1963 {
1964         struct trace_subsystem_dir *dir;
1965         struct event_subsystem *system;
1966         struct dentry *entry;
1967
1968         /* First see if we did not already create this dir */
1969         list_for_each_entry(dir, &tr->systems, list) {
1970                 system = dir->subsystem;
1971                 if (strcmp(system->name, name) == 0) {
1972                         dir->nr_events++;
1973                         file->system = dir;
1974                         return dir->entry;
1975                 }
1976         }
1977
1978         /* Now see if the system itself exists. */
1979         list_for_each_entry(system, &event_subsystems, list) {
1980                 if (strcmp(system->name, name) == 0)
1981                         break;
1982         }
1983         /* Reset system variable when not found */
1984         if (&system->list == &event_subsystems)
1985                 system = NULL;
1986
1987         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1988         if (!dir)
1989                 goto out_fail;
1990
1991         if (!system) {
1992                 system = create_new_subsystem(name);
1993                 if (!system)
1994                         goto out_free;
1995         } else
1996                 __get_system(system);
1997
1998         dir->entry = tracefs_create_dir(name, parent);
1999         if (!dir->entry) {
2000                 pr_warn("Failed to create system directory %s\n", name);
2001                 __put_system(system);
2002                 goto out_free;
2003         }
2004
2005         dir->tr = tr;
2006         dir->ref_count = 1;
2007         dir->nr_events = 1;
2008         dir->subsystem = system;
2009         file->system = dir;
2010
2011         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2012                                     &ftrace_subsystem_filter_fops);
2013         if (!entry) {
2014                 kfree(system->filter);
2015                 system->filter = NULL;
2016                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2017         }
2018
2019         trace_create_file("enable", 0644, dir->entry, dir,
2020                           &ftrace_system_enable_fops);
2021
2022         list_add(&dir->list, &tr->systems);
2023
2024         return dir->entry;
2025
2026  out_free:
2027         kfree(dir);
2028  out_fail:
2029         /* Only print this message if failed on memory allocation */
2030         if (!dir || !system)
2031                 pr_warn("No memory to create event subsystem %s\n", name);
2032         return NULL;
2033 }
2034
2035 static int
2036 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2037 {
2038         struct trace_event_call *call = file->event_call;
2039         struct trace_array *tr = file->tr;
2040         struct list_head *head;
2041         struct dentry *d_events;
2042         const char *name;
2043         int ret;
2044
2045         /*
2046          * If the trace point header did not define TRACE_SYSTEM
2047          * then the system would be called "TRACE_SYSTEM".
2048          */
2049         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2050                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2051                 if (!d_events)
2052                         return -ENOMEM;
2053         } else
2054                 d_events = parent;
2055
2056         name = trace_event_name(call);
2057         file->dir = tracefs_create_dir(name, d_events);
2058         if (!file->dir) {
2059                 pr_warn("Could not create tracefs '%s' directory\n", name);
2060                 return -1;
2061         }
2062
2063         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2064                 trace_create_file("enable", 0644, file->dir, file,
2065                                   &ftrace_enable_fops);
2066
2067 #ifdef CONFIG_PERF_EVENTS
2068         if (call->event.type && call->class->reg)
2069                 trace_create_file("id", 0444, file->dir,
2070                                   (void *)(long)call->event.type,
2071                                   &ftrace_event_id_fops);
2072 #endif
2073
2074         /*
2075          * Other events may have the same class. Only update
2076          * the fields if they are not already defined.
2077          */
2078         head = trace_get_fields(call);
2079         if (list_empty(head)) {
2080                 ret = call->class->define_fields(call);
2081                 if (ret < 0) {
2082                         pr_warn("Could not initialize trace point events/%s\n",
2083                                 name);
2084                         return -1;
2085                 }
2086         }
2087         trace_create_file("filter", 0644, file->dir, file,
2088                           &ftrace_event_filter_fops);
2089
2090         trace_create_file("trigger", 0644, file->dir, file,
2091                           &event_trigger_fops);
2092
2093         trace_create_file("format", 0444, file->dir, call,
2094                           &ftrace_event_format_fops);
2095
2096         return 0;
2097 }
2098
2099 static void remove_event_from_tracers(struct trace_event_call *call)
2100 {
2101         struct trace_event_file *file;
2102         struct trace_array *tr;
2103
2104         do_for_each_event_file_safe(tr, file) {
2105                 if (file->event_call != call)
2106                         continue;
2107
2108                 remove_event_file_dir(file);
2109                 /*
2110                  * The do_for_each_event_file_safe() is
2111                  * a double loop. After finding the call for this
2112                  * trace_array, we use break to jump to the next
2113                  * trace_array.
2114                  */
2115                 break;
2116         } while_for_each_event_file();
2117 }
2118
2119 static void event_remove(struct trace_event_call *call)
2120 {
2121         struct trace_array *tr;
2122         struct trace_event_file *file;
2123
2124         do_for_each_event_file(tr, file) {
2125                 if (file->event_call != call)
2126                         continue;
2127                 ftrace_event_enable_disable(file, 0);
2128                 /*
2129                  * The do_for_each_event_file() is
2130                  * a double loop. After finding the call for this
2131                  * trace_array, we use break to jump to the next
2132                  * trace_array.
2133                  */
2134                 break;
2135         } while_for_each_event_file();
2136
2137         if (call->event.funcs)
2138                 __unregister_trace_event(&call->event);
2139         remove_event_from_tracers(call);
2140         list_del(&call->list);
2141 }
2142
2143 static int event_init(struct trace_event_call *call)
2144 {
2145         int ret = 0;
2146         const char *name;
2147
2148         name = trace_event_name(call);
2149         if (WARN_ON(!name))
2150                 return -EINVAL;
2151
2152         if (call->class->raw_init) {
2153                 ret = call->class->raw_init(call);
2154                 if (ret < 0 && ret != -ENOSYS)
2155                         pr_warn("Could not initialize trace events/%s\n", name);
2156         }
2157
2158         return ret;
2159 }
2160
2161 static int
2162 __register_event(struct trace_event_call *call, struct module *mod)
2163 {
2164         int ret;
2165
2166         ret = event_init(call);
2167         if (ret < 0)
2168                 return ret;
2169
2170         list_add(&call->list, &ftrace_events);
2171         call->mod = mod;
2172
2173         return 0;
2174 }
2175
2176 static char *enum_replace(char *ptr, struct trace_enum_map *map, int len)
2177 {
2178         int rlen;
2179         int elen;
2180
2181         /* Find the length of the enum value as a string */
2182         elen = snprintf(ptr, 0, "%ld", map->enum_value);
2183         /* Make sure there's enough room to replace the string with the value */
2184         if (len < elen)
2185                 return NULL;
2186
2187         snprintf(ptr, elen + 1, "%ld", map->enum_value);
2188
2189         /* Get the rest of the string of ptr */
2190         rlen = strlen(ptr + len);
2191         memmove(ptr + elen, ptr + len, rlen);
2192         /* Make sure we end the new string */
2193         ptr[elen + rlen] = 0;
2194
2195         return ptr + elen;
2196 }
2197
2198 static void update_event_printk(struct trace_event_call *call,
2199                                 struct trace_enum_map *map)
2200 {
2201         char *ptr;
2202         int quote = 0;
2203         int len = strlen(map->enum_string);
2204
2205         for (ptr = call->print_fmt; *ptr; ptr++) {
2206                 if (*ptr == '\\') {
2207                         ptr++;
2208                         /* paranoid */
2209                         if (!*ptr)
2210                                 break;
2211                         continue;
2212                 }
2213                 if (*ptr == '"') {
2214                         quote ^= 1;
2215                         continue;
2216                 }
2217                 if (quote)
2218                         continue;
2219                 if (isdigit(*ptr)) {
2220                         /* skip numbers */
2221                         do {
2222                                 ptr++;
2223                                 /* Check for alpha chars like ULL */
2224                         } while (isalnum(*ptr));
2225                         if (!*ptr)
2226                                 break;
2227                         /*
2228                          * A number must have some kind of delimiter after
2229                          * it, and we can ignore that too.
2230                          */
2231                         continue;
2232                 }
2233                 if (isalpha(*ptr) || *ptr == '_') {
2234                         if (strncmp(map->enum_string, ptr, len) == 0 &&
2235                             !isalnum(ptr[len]) && ptr[len] != '_') {
2236                                 ptr = enum_replace(ptr, map, len);
2237                                 /* Hmm, enum string smaller than value */
2238                                 if (WARN_ON_ONCE(!ptr))
2239                                         return;
2240                                 /*
2241                                  * No need to decrement here, as enum_replace()
2242                                  * returns the pointer to the character passed
2243                                  * the enum, and two enums can not be placed
2244                                  * back to back without something in between.
2245                                  * We can skip that something in between.
2246                                  */
2247                                 continue;
2248                         }
2249                 skip_more:
2250                         do {
2251                                 ptr++;
2252                         } while (isalnum(*ptr) || *ptr == '_');
2253                         if (!*ptr)
2254                                 break;
2255                         /*
2256                          * If what comes after this variable is a '.' or
2257                          * '->' then we can continue to ignore that string.
2258                          */
2259                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2260                                 ptr += *ptr == '.' ? 1 : 2;
2261                                 if (!*ptr)
2262                                         break;
2263                                 goto skip_more;
2264                         }
2265                         /*
2266                          * Once again, we can skip the delimiter that came
2267                          * after the string.
2268                          */
2269                         continue;
2270                 }
2271         }
2272 }
2273
2274 void trace_event_enum_update(struct trace_enum_map **map, int len)
2275 {
2276         struct trace_event_call *call, *p;
2277         const char *last_system = NULL;
2278         int last_i;
2279         int i;
2280
2281         down_write(&trace_event_sem);
2282         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2283                 /* events are usually grouped together with systems */
2284                 if (!last_system || call->class->system != last_system) {
2285                         last_i = 0;
2286                         last_system = call->class->system;
2287                 }
2288
2289                 for (i = last_i; i < len; i++) {
2290                         if (call->class->system == map[i]->system) {
2291                                 /* Save the first system if need be */
2292                                 if (!last_i)
2293                                         last_i = i;
2294                                 update_event_printk(call, map[i]);
2295                         }
2296                 }
2297         }
2298         up_write(&trace_event_sem);
2299 }
2300
2301 static struct trace_event_file *
2302 trace_create_new_event(struct trace_event_call *call,
2303                        struct trace_array *tr)
2304 {
2305         struct trace_event_file *file;
2306
2307         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2308         if (!file)
2309                 return NULL;
2310
2311         file->event_call = call;
2312         file->tr = tr;
2313         atomic_set(&file->sm_ref, 0);
2314         atomic_set(&file->tm_ref, 0);
2315         INIT_LIST_HEAD(&file->triggers);
2316         list_add(&file->list, &tr->events);
2317
2318         return file;
2319 }
2320
2321 /* Add an event to a trace directory */
2322 static int
2323 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2324 {
2325         struct trace_event_file *file;
2326
2327         file = trace_create_new_event(call, tr);
2328         if (!file)
2329                 return -ENOMEM;
2330
2331         return event_create_dir(tr->event_dir, file);
2332 }
2333
2334 /*
2335  * Just create a decriptor for early init. A descriptor is required
2336  * for enabling events at boot. We want to enable events before
2337  * the filesystem is initialized.
2338  */
2339 static __init int
2340 __trace_early_add_new_event(struct trace_event_call *call,
2341                             struct trace_array *tr)
2342 {
2343         struct trace_event_file *file;
2344
2345         file = trace_create_new_event(call, tr);
2346         if (!file)
2347                 return -ENOMEM;
2348
2349         return 0;
2350 }
2351
2352 struct ftrace_module_file_ops;
2353 static void __add_event_to_tracers(struct trace_event_call *call);
2354
2355 /* Add an additional event_call dynamically */
2356 int trace_add_event_call(struct trace_event_call *call)
2357 {
2358         int ret;
2359         mutex_lock(&trace_types_lock);
2360         mutex_lock(&event_mutex);
2361
2362         ret = __register_event(call, NULL);
2363         if (ret >= 0)
2364                 __add_event_to_tracers(call);
2365
2366         mutex_unlock(&event_mutex);
2367         mutex_unlock(&trace_types_lock);
2368         return ret;
2369 }
2370
2371 /*
2372  * Must be called under locking of trace_types_lock, event_mutex and
2373  * trace_event_sem.
2374  */
2375 static void __trace_remove_event_call(struct trace_event_call *call)
2376 {
2377         event_remove(call);
2378         trace_destroy_fields(call);
2379         free_event_filter(call->filter);
2380         call->filter = NULL;
2381 }
2382
2383 static int probe_remove_event_call(struct trace_event_call *call)
2384 {
2385         struct trace_array *tr;
2386         struct trace_event_file *file;
2387
2388 #ifdef CONFIG_PERF_EVENTS
2389         if (call->perf_refcount)
2390                 return -EBUSY;
2391 #endif
2392         do_for_each_event_file(tr, file) {
2393                 if (file->event_call != call)
2394                         continue;
2395                 /*
2396                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2397                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2398                  * TRACE_REG_UNREGISTER.
2399                  */
2400                 if (file->flags & EVENT_FILE_FL_ENABLED)
2401                         return -EBUSY;
2402                 /*
2403                  * The do_for_each_event_file_safe() is
2404                  * a double loop. After finding the call for this
2405                  * trace_array, we use break to jump to the next
2406                  * trace_array.
2407                  */
2408                 break;
2409         } while_for_each_event_file();
2410
2411         __trace_remove_event_call(call);
2412
2413         return 0;
2414 }
2415
2416 /* Remove an event_call */
2417 int trace_remove_event_call(struct trace_event_call *call)
2418 {
2419         int ret;
2420
2421         mutex_lock(&trace_types_lock);
2422         mutex_lock(&event_mutex);
2423         down_write(&trace_event_sem);
2424         ret = probe_remove_event_call(call);
2425         up_write(&trace_event_sem);
2426         mutex_unlock(&event_mutex);
2427         mutex_unlock(&trace_types_lock);
2428
2429         return ret;
2430 }
2431
2432 #define for_each_event(event, start, end)                       \
2433         for (event = start;                                     \
2434              (unsigned long)event < (unsigned long)end;         \
2435              event++)
2436
2437 #ifdef CONFIG_MODULES
2438
2439 static void trace_module_add_events(struct module *mod)
2440 {
2441         struct trace_event_call **call, **start, **end;
2442
2443         if (!mod->num_trace_events)
2444                 return;
2445
2446         /* Don't add infrastructure for mods without tracepoints */
2447         if (trace_module_has_bad_taint(mod)) {
2448                 pr_err("%s: module has bad taint, not creating trace events\n",
2449                        mod->name);
2450                 return;
2451         }
2452
2453         start = mod->trace_events;
2454         end = mod->trace_events + mod->num_trace_events;
2455
2456         for_each_event(call, start, end) {
2457                 __register_event(*call, mod);
2458                 __add_event_to_tracers(*call);
2459         }
2460 }
2461
2462 static void trace_module_remove_events(struct module *mod)
2463 {
2464         struct trace_event_call *call, *p;
2465         bool clear_trace = false;
2466
2467         down_write(&trace_event_sem);
2468         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2469                 if (call->mod == mod) {
2470                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2471                                 clear_trace = true;
2472                         __trace_remove_event_call(call);
2473                 }
2474         }
2475         up_write(&trace_event_sem);
2476
2477         /*
2478          * It is safest to reset the ring buffer if the module being unloaded
2479          * registered any events that were used. The only worry is if
2480          * a new module gets loaded, and takes on the same id as the events
2481          * of this module. When printing out the buffer, traced events left
2482          * over from this module may be passed to the new module events and
2483          * unexpected results may occur.
2484          */
2485         if (clear_trace)
2486                 tracing_reset_all_online_cpus();
2487 }
2488
2489 static int trace_module_notify(struct notifier_block *self,
2490                                unsigned long val, void *data)
2491 {
2492         struct module *mod = data;
2493
2494         mutex_lock(&trace_types_lock);
2495         mutex_lock(&event_mutex);
2496         switch (val) {
2497         case MODULE_STATE_COMING:
2498                 trace_module_add_events(mod);
2499                 break;
2500         case MODULE_STATE_GOING:
2501                 trace_module_remove_events(mod);
2502                 break;
2503         }
2504         mutex_unlock(&event_mutex);
2505         mutex_unlock(&trace_types_lock);
2506
2507         return 0;
2508 }
2509
2510 static struct notifier_block trace_module_nb = {
2511         .notifier_call = trace_module_notify,
2512         .priority = 1, /* higher than trace.c module notify */
2513 };
2514 #endif /* CONFIG_MODULES */
2515
2516 /* Create a new event directory structure for a trace directory. */
2517 static void
2518 __trace_add_event_dirs(struct trace_array *tr)
2519 {
2520         struct trace_event_call *call;
2521         int ret;
2522
2523         list_for_each_entry(call, &ftrace_events, list) {
2524                 ret = __trace_add_new_event(call, tr);
2525                 if (ret < 0)
2526                         pr_warn("Could not create directory for event %s\n",
2527                                 trace_event_name(call));
2528         }
2529 }
2530
2531 struct trace_event_file *
2532 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2533 {
2534         struct trace_event_file *file;
2535         struct trace_event_call *call;
2536         const char *name;
2537
2538         list_for_each_entry(file, &tr->events, list) {
2539
2540                 call = file->event_call;
2541                 name = trace_event_name(call);
2542
2543                 if (!name || !call->class || !call->class->reg)
2544                         continue;
2545
2546                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2547                         continue;
2548
2549                 if (strcmp(event, name) == 0 &&
2550                     strcmp(system, call->class->system) == 0)
2551                         return file;
2552         }
2553         return NULL;
2554 }
2555
2556 #ifdef CONFIG_DYNAMIC_FTRACE
2557
2558 /* Avoid typos */
2559 #define ENABLE_EVENT_STR        "enable_event"
2560 #define DISABLE_EVENT_STR       "disable_event"
2561
2562 struct event_probe_data {
2563         struct trace_event_file *file;
2564         unsigned long                   count;
2565         int                             ref;
2566         bool                            enable;
2567 };
2568
2569 static void
2570 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2571 {
2572         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2573         struct event_probe_data *data = *pdata;
2574
2575         if (!data)
2576                 return;
2577
2578         if (data->enable)
2579                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2580         else
2581                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2582 }
2583
2584 static void
2585 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2586 {
2587         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2588         struct event_probe_data *data = *pdata;
2589
2590         if (!data)
2591                 return;
2592
2593         if (!data->count)
2594                 return;
2595
2596         /* Skip if the event is in a state we want to switch to */
2597         if (data->enable == !(data->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2598                 return;
2599
2600         if (data->count != -1)
2601                 (data->count)--;
2602
2603         event_enable_probe(ip, parent_ip, _data);
2604 }
2605
2606 static int
2607 event_enable_print(struct seq_file *m, unsigned long ip,
2608                       struct ftrace_probe_ops *ops, void *_data)
2609 {
2610         struct event_probe_data *data = _data;
2611
2612         seq_printf(m, "%ps:", (void *)ip);
2613
2614         seq_printf(m, "%s:%s:%s",
2615                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2616                    data->file->event_call->class->system,
2617                    trace_event_name(data->file->event_call));
2618
2619         if (data->count == -1)
2620                 seq_puts(m, ":unlimited\n");
2621         else
2622                 seq_printf(m, ":count=%ld\n", data->count);
2623
2624         return 0;
2625 }
2626
2627 static int
2628 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2629                   void **_data)
2630 {
2631         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2632         struct event_probe_data *data = *pdata;
2633
2634         data->ref++;
2635         return 0;
2636 }
2637
2638 static void
2639 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2640                   void **_data)
2641 {
2642         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2643         struct event_probe_data *data = *pdata;
2644
2645         if (WARN_ON_ONCE(data->ref <= 0))
2646                 return;
2647
2648         data->ref--;
2649         if (!data->ref) {
2650                 /* Remove the SOFT_MODE flag */
2651                 __ftrace_event_enable_disable(data->file, 0, 1);
2652                 module_put(data->file->event_call->mod);
2653                 kfree(data);
2654         }
2655         *pdata = NULL;
2656 }
2657
2658 static struct ftrace_probe_ops event_enable_probe_ops = {
2659         .func                   = event_enable_probe,
2660         .print                  = event_enable_print,
2661         .init                   = event_enable_init,
2662         .free                   = event_enable_free,
2663 };
2664
2665 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2666         .func                   = event_enable_count_probe,
2667         .print                  = event_enable_print,
2668         .init                   = event_enable_init,
2669         .free                   = event_enable_free,
2670 };
2671
2672 static struct ftrace_probe_ops event_disable_probe_ops = {
2673         .func                   = event_enable_probe,
2674         .print                  = event_enable_print,
2675         .init                   = event_enable_init,
2676         .free                   = event_enable_free,
2677 };
2678
2679 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2680         .func                   = event_enable_count_probe,
2681         .print                  = event_enable_print,
2682         .init                   = event_enable_init,
2683         .free                   = event_enable_free,
2684 };
2685
2686 static int
2687 event_enable_func(struct ftrace_hash *hash,
2688                   char *glob, char *cmd, char *param, int enabled)
2689 {
2690         struct trace_array *tr = top_trace_array();
2691         struct trace_event_file *file;
2692         struct ftrace_probe_ops *ops;
2693         struct event_probe_data *data;
2694         const char *system;
2695         const char *event;
2696         char *number;
2697         bool enable;
2698         int ret;
2699
2700         if (!tr)
2701                 return -ENODEV;
2702
2703         /* hash funcs only work with set_ftrace_filter */
2704         if (!enabled || !param)
2705                 return -EINVAL;
2706
2707         system = strsep(&param, ":");
2708         if (!param)
2709                 return -EINVAL;
2710
2711         event = strsep(&param, ":");
2712
2713         mutex_lock(&event_mutex);
2714
2715         ret = -EINVAL;
2716         file = find_event_file(tr, system, event);
2717         if (!file)
2718                 goto out;
2719
2720         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2721
2722         if (enable)
2723                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2724         else
2725                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2726
2727         if (glob[0] == '!') {
2728                 unregister_ftrace_function_probe_func(glob+1, ops);
2729                 ret = 0;
2730                 goto out;
2731         }
2732
2733         ret = -ENOMEM;
2734         data = kzalloc(sizeof(*data), GFP_KERNEL);
2735         if (!data)
2736                 goto out;
2737
2738         data->enable = enable;
2739         data->count = -1;
2740         data->file = file;
2741
2742         if (!param)
2743                 goto out_reg;
2744
2745         number = strsep(&param, ":");
2746
2747         ret = -EINVAL;
2748         if (!strlen(number))
2749                 goto out_free;
2750
2751         /*
2752          * We use the callback data field (which is a pointer)
2753          * as our counter.
2754          */
2755         ret = kstrtoul(number, 0, &data->count);
2756         if (ret)
2757                 goto out_free;
2758
2759  out_reg:
2760         /* Don't let event modules unload while probe registered */
2761         ret = try_module_get(file->event_call->mod);
2762         if (!ret) {
2763                 ret = -EBUSY;
2764                 goto out_free;
2765         }
2766
2767         ret = __ftrace_event_enable_disable(file, 1, 1);
2768         if (ret < 0)
2769                 goto out_put;
2770         ret = register_ftrace_function_probe(glob, ops, data);
2771         /*
2772          * The above returns on success the # of functions enabled,
2773          * but if it didn't find any functions it returns zero.
2774          * Consider no functions a failure too.
2775          */
2776         if (!ret) {
2777                 ret = -ENOENT;
2778                 goto out_disable;
2779         } else if (ret < 0)
2780                 goto out_disable;
2781         /* Just return zero, not the number of enabled functions */
2782         ret = 0;
2783  out:
2784         mutex_unlock(&event_mutex);
2785         return ret;
2786
2787  out_disable:
2788         __ftrace_event_enable_disable(file, 0, 1);
2789  out_put:
2790         module_put(file->event_call->mod);
2791  out_free:
2792         kfree(data);
2793         goto out;
2794 }
2795
2796 static struct ftrace_func_command event_enable_cmd = {
2797         .name                   = ENABLE_EVENT_STR,
2798         .func                   = event_enable_func,
2799 };
2800
2801 static struct ftrace_func_command event_disable_cmd = {
2802         .name                   = DISABLE_EVENT_STR,
2803         .func                   = event_enable_func,
2804 };
2805
2806 static __init int register_event_cmds(void)
2807 {
2808         int ret;
2809
2810         ret = register_ftrace_command(&event_enable_cmd);
2811         if (WARN_ON(ret < 0))
2812                 return ret;
2813         ret = register_ftrace_command(&event_disable_cmd);
2814         if (WARN_ON(ret < 0))
2815                 unregister_ftrace_command(&event_enable_cmd);
2816         return ret;
2817 }
2818 #else
2819 static inline int register_event_cmds(void) { return 0; }
2820 #endif /* CONFIG_DYNAMIC_FTRACE */
2821
2822 /*
2823  * The top level array has already had its trace_event_file
2824  * descriptors created in order to allow for early events to
2825  * be recorded. This function is called after the tracefs has been
2826  * initialized, and we now have to create the files associated
2827  * to the events.
2828  */
2829 static __init void
2830 __trace_early_add_event_dirs(struct trace_array *tr)
2831 {
2832         struct trace_event_file *file;
2833         int ret;
2834
2835
2836         list_for_each_entry(file, &tr->events, list) {
2837                 ret = event_create_dir(tr->event_dir, file);
2838                 if (ret < 0)
2839                         pr_warn("Could not create directory for event %s\n",
2840                                 trace_event_name(file->event_call));
2841         }
2842 }
2843
2844 /*
2845  * For early boot up, the top trace array requires to have
2846  * a list of events that can be enabled. This must be done before
2847  * the filesystem is set up in order to allow events to be traced
2848  * early.
2849  */
2850 static __init void
2851 __trace_early_add_events(struct trace_array *tr)
2852 {
2853         struct trace_event_call *call;
2854         int ret;
2855
2856         list_for_each_entry(call, &ftrace_events, list) {
2857                 /* Early boot up should not have any modules loaded */
2858                 if (WARN_ON_ONCE(call->mod))
2859                         continue;
2860
2861                 ret = __trace_early_add_new_event(call, tr);
2862                 if (ret < 0)
2863                         pr_warn("Could not create early event %s\n",
2864                                 trace_event_name(call));
2865         }
2866 }
2867
2868 /* Remove the event directory structure for a trace directory. */
2869 static void
2870 __trace_remove_event_dirs(struct trace_array *tr)
2871 {
2872         struct trace_event_file *file, *next;
2873
2874         list_for_each_entry_safe(file, next, &tr->events, list)
2875                 remove_event_file_dir(file);
2876 }
2877
2878 static void __add_event_to_tracers(struct trace_event_call *call)
2879 {
2880         struct trace_array *tr;
2881
2882         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2883                 __trace_add_new_event(call, tr);
2884 }
2885
2886 extern struct trace_event_call *__start_ftrace_events[];
2887 extern struct trace_event_call *__stop_ftrace_events[];
2888
2889 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2890
2891 static __init int setup_trace_event(char *str)
2892 {
2893         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2894         ring_buffer_expanded = true;
2895         tracing_selftest_disabled = true;
2896
2897         return 1;
2898 }
2899 __setup("trace_event=", setup_trace_event);
2900
2901 /* Expects to have event_mutex held when called */
2902 static int
2903 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2904 {
2905         struct dentry *d_events;
2906         struct dentry *entry;
2907
2908         entry = tracefs_create_file("set_event", 0644, parent,
2909                                     tr, &ftrace_set_event_fops);
2910         if (!entry) {
2911                 pr_warn("Could not create tracefs 'set_event' entry\n");
2912                 return -ENOMEM;
2913         }
2914
2915         d_events = tracefs_create_dir("events", parent);
2916         if (!d_events) {
2917                 pr_warn("Could not create tracefs 'events' directory\n");
2918                 return -ENOMEM;
2919         }
2920
2921         entry = tracefs_create_file("set_event_pid", 0644, parent,
2922                                     tr, &ftrace_set_event_pid_fops);
2923
2924         /* ring buffer internal formats */
2925         trace_create_file("header_page", 0444, d_events,
2926                           ring_buffer_print_page_header,
2927                           &ftrace_show_header_fops);
2928
2929         trace_create_file("header_event", 0444, d_events,
2930                           ring_buffer_print_entry_header,
2931                           &ftrace_show_header_fops);
2932
2933         trace_create_file("enable", 0644, d_events,
2934                           tr, &ftrace_tr_enable_fops);
2935
2936         tr->event_dir = d_events;
2937
2938         return 0;
2939 }
2940
2941 /**
2942  * event_trace_add_tracer - add a instance of a trace_array to events
2943  * @parent: The parent dentry to place the files/directories for events in
2944  * @tr: The trace array associated with these events
2945  *
2946  * When a new instance is created, it needs to set up its events
2947  * directory, as well as other files associated with events. It also
2948  * creates the event hierachry in the @parent/events directory.
2949  *
2950  * Returns 0 on success.
2951  */
2952 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2953 {
2954         int ret;
2955
2956         mutex_lock(&event_mutex);
2957
2958         ret = create_event_toplevel_files(parent, tr);
2959         if (ret)
2960                 goto out_unlock;
2961
2962         down_write(&trace_event_sem);
2963         __trace_add_event_dirs(tr);
2964         up_write(&trace_event_sem);
2965
2966  out_unlock:
2967         mutex_unlock(&event_mutex);
2968
2969         return ret;
2970 }
2971
2972 /*
2973  * The top trace array already had its file descriptors created.
2974  * Now the files themselves need to be created.
2975  */
2976 static __init int
2977 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2978 {
2979         int ret;
2980
2981         mutex_lock(&event_mutex);
2982
2983         ret = create_event_toplevel_files(parent, tr);
2984         if (ret)
2985                 goto out_unlock;
2986
2987         down_write(&trace_event_sem);
2988         __trace_early_add_event_dirs(tr);
2989         up_write(&trace_event_sem);
2990
2991  out_unlock:
2992         mutex_unlock(&event_mutex);
2993
2994         return ret;
2995 }
2996
2997 int event_trace_del_tracer(struct trace_array *tr)
2998 {
2999         mutex_lock(&event_mutex);
3000
3001         /* Disable any event triggers and associated soft-disabled events */
3002         clear_event_triggers(tr);
3003
3004         /* Clear the pid list */
3005         __ftrace_clear_event_pids(tr);
3006
3007         /* Disable any running events */
3008         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3009
3010         /* Access to events are within rcu_read_lock_sched() */
3011         synchronize_sched();
3012
3013         down_write(&trace_event_sem);
3014         __trace_remove_event_dirs(tr);
3015         tracefs_remove_recursive(tr->event_dir);
3016         up_write(&trace_event_sem);
3017
3018         tr->event_dir = NULL;
3019
3020         mutex_unlock(&event_mutex);
3021
3022         return 0;
3023 }
3024
3025 static __init int event_trace_memsetup(void)
3026 {
3027         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3028         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3029         return 0;
3030 }
3031
3032 static __init void
3033 early_enable_events(struct trace_array *tr, bool disable_first)
3034 {
3035         char *buf = bootup_event_buf;
3036         char *token;
3037         int ret;
3038
3039         while (true) {
3040                 token = strsep(&buf, ",");
3041
3042                 if (!token)
3043                         break;
3044                 if (!*token)
3045                         continue;
3046
3047                 /* Restarting syscalls requires that we stop them first */
3048                 if (disable_first)
3049                         ftrace_set_clr_event(tr, token, 0);
3050
3051                 ret = ftrace_set_clr_event(tr, token, 1);
3052                 if (ret)
3053                         pr_warn("Failed to enable trace event: %s\n", token);
3054
3055                 /* Put back the comma to allow this to be called again */
3056                 if (buf)
3057                         *(buf - 1) = ',';
3058         }
3059 }
3060
3061 static __init int event_trace_enable(void)
3062 {
3063         struct trace_array *tr = top_trace_array();
3064         struct trace_event_call **iter, *call;
3065         int ret;
3066
3067         if (!tr)
3068                 return -ENODEV;
3069
3070         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3071
3072                 call = *iter;
3073                 ret = event_init(call);
3074                 if (!ret)
3075                         list_add(&call->list, &ftrace_events);
3076         }
3077
3078         /*
3079          * We need the top trace array to have a working set of trace
3080          * points at early init, before the debug files and directories
3081          * are created. Create the file entries now, and attach them
3082          * to the actual file dentries later.
3083          */
3084         __trace_early_add_events(tr);
3085
3086         early_enable_events(tr, false);
3087
3088         trace_printk_start_comm();
3089
3090         register_event_cmds();
3091
3092         register_trigger_cmds();
3093
3094         return 0;
3095 }
3096
3097 /*
3098  * event_trace_enable() is called from trace_event_init() first to
3099  * initialize events and perhaps start any events that are on the
3100  * command line. Unfortunately, there are some events that will not
3101  * start this early, like the system call tracepoints that need
3102  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3103  * is called before pid 1 starts, and this flag is never set, making
3104  * the syscall tracepoint never get reached, but the event is enabled
3105  * regardless (and not doing anything).
3106  */
3107 static __init int event_trace_enable_again(void)
3108 {
3109         struct trace_array *tr;
3110
3111         tr = top_trace_array();
3112         if (!tr)
3113                 return -ENODEV;
3114
3115         early_enable_events(tr, true);
3116
3117         return 0;
3118 }
3119
3120 early_initcall(event_trace_enable_again);
3121
3122 static __init int event_trace_init(void)
3123 {
3124         struct trace_array *tr;
3125         struct dentry *d_tracer;
3126         struct dentry *entry;
3127         int ret;
3128
3129         tr = top_trace_array();
3130         if (!tr)
3131                 return -ENODEV;
3132
3133         d_tracer = tracing_init_dentry();
3134         if (IS_ERR(d_tracer))
3135                 return 0;
3136
3137         entry = tracefs_create_file("available_events", 0444, d_tracer,
3138                                     tr, &ftrace_avail_fops);
3139         if (!entry)
3140                 pr_warn("Could not create tracefs 'available_events' entry\n");
3141
3142         if (trace_define_generic_fields())
3143                 pr_warn("tracing: Failed to allocated generic fields");
3144
3145         if (trace_define_common_fields())
3146                 pr_warn("tracing: Failed to allocate common fields");
3147
3148         ret = early_event_add_tracer(d_tracer, tr);
3149         if (ret)
3150                 return ret;
3151
3152 #ifdef CONFIG_MODULES
3153         ret = register_module_notifier(&trace_module_nb);
3154         if (ret)
3155                 pr_warn("Failed to register trace events module notifier\n");
3156 #endif
3157         return 0;
3158 }
3159
3160 void __init trace_event_init(void)
3161 {
3162         event_trace_memsetup();
3163         init_ftrace_syscalls();
3164         event_trace_enable();
3165 }
3166
3167 fs_initcall(event_trace_init);
3168
3169 #ifdef CONFIG_FTRACE_STARTUP_TEST
3170
3171 static DEFINE_SPINLOCK(test_spinlock);
3172 static DEFINE_SPINLOCK(test_spinlock_irq);
3173 static DEFINE_MUTEX(test_mutex);
3174
3175 static __init void test_work(struct work_struct *dummy)
3176 {
3177         spin_lock(&test_spinlock);
3178         spin_lock_irq(&test_spinlock_irq);
3179         udelay(1);
3180         spin_unlock_irq(&test_spinlock_irq);
3181         spin_unlock(&test_spinlock);
3182
3183         mutex_lock(&test_mutex);
3184         msleep(1);
3185         mutex_unlock(&test_mutex);
3186 }
3187
3188 static __init int event_test_thread(void *unused)
3189 {
3190         void *test_malloc;
3191
3192         test_malloc = kmalloc(1234, GFP_KERNEL);
3193         if (!test_malloc)
3194                 pr_info("failed to kmalloc\n");
3195
3196         schedule_on_each_cpu(test_work);
3197
3198         kfree(test_malloc);
3199
3200         set_current_state(TASK_INTERRUPTIBLE);
3201         while (!kthread_should_stop()) {
3202                 schedule();
3203                 set_current_state(TASK_INTERRUPTIBLE);
3204         }
3205         __set_current_state(TASK_RUNNING);
3206
3207         return 0;
3208 }
3209
3210 /*
3211  * Do various things that may trigger events.
3212  */
3213 static __init void event_test_stuff(void)
3214 {
3215         struct task_struct *test_thread;
3216
3217         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3218         msleep(1);
3219         kthread_stop(test_thread);
3220 }
3221
3222 /*
3223  * For every trace event defined, we will test each trace point separately,
3224  * and then by groups, and finally all trace points.
3225  */
3226 static __init void event_trace_self_tests(void)
3227 {
3228         struct trace_subsystem_dir *dir;
3229         struct trace_event_file *file;
3230         struct trace_event_call *call;
3231         struct event_subsystem *system;
3232         struct trace_array *tr;
3233         int ret;
3234
3235         tr = top_trace_array();
3236         if (!tr)
3237                 return;
3238
3239         pr_info("Running tests on trace events:\n");
3240
3241         list_for_each_entry(file, &tr->events, list) {
3242
3243                 call = file->event_call;
3244
3245                 /* Only test those that have a probe */
3246                 if (!call->class || !call->class->probe)
3247                         continue;
3248
3249 /*
3250  * Testing syscall events here is pretty useless, but
3251  * we still do it if configured. But this is time consuming.
3252  * What we really need is a user thread to perform the
3253  * syscalls as we test.
3254  */
3255 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3256                 if (call->class->system &&
3257                     strcmp(call->class->system, "syscalls") == 0)
3258                         continue;
3259 #endif
3260
3261                 pr_info("Testing event %s: ", trace_event_name(call));
3262
3263                 /*
3264                  * If an event is already enabled, someone is using
3265                  * it and the self test should not be on.
3266                  */
3267                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3268                         pr_warn("Enabled event during self test!\n");
3269                         WARN_ON_ONCE(1);
3270                         continue;
3271                 }
3272
3273                 ftrace_event_enable_disable(file, 1);
3274                 event_test_stuff();
3275                 ftrace_event_enable_disable(file, 0);
3276
3277                 pr_cont("OK\n");
3278         }
3279
3280         /* Now test at the sub system level */
3281
3282         pr_info("Running tests on trace event systems:\n");
3283
3284         list_for_each_entry(dir, &tr->systems, list) {
3285
3286                 system = dir->subsystem;
3287
3288                 /* the ftrace system is special, skip it */
3289                 if (strcmp(system->name, "ftrace") == 0)
3290                         continue;
3291
3292                 pr_info("Testing event system %s: ", system->name);
3293
3294                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3295                 if (WARN_ON_ONCE(ret)) {
3296                         pr_warn("error enabling system %s\n",
3297                                 system->name);
3298                         continue;
3299                 }
3300
3301                 event_test_stuff();
3302
3303                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3304                 if (WARN_ON_ONCE(ret)) {
3305                         pr_warn("error disabling system %s\n",
3306                                 system->name);
3307                         continue;
3308                 }
3309
3310                 pr_cont("OK\n");
3311         }
3312
3313         /* Test with all events enabled */
3314
3315         pr_info("Running tests on all trace events:\n");
3316         pr_info("Testing all events: ");
3317
3318         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3319         if (WARN_ON_ONCE(ret)) {
3320                 pr_warn("error enabling all events\n");
3321                 return;
3322         }
3323
3324         event_test_stuff();
3325
3326         /* reset sysname */
3327         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3328         if (WARN_ON_ONCE(ret)) {
3329                 pr_warn("error disabling all events\n");
3330                 return;
3331         }
3332
3333         pr_cont("OK\n");
3334 }
3335
3336 #ifdef CONFIG_FUNCTION_TRACER
3337
3338 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3339
3340 static struct trace_array *event_tr;
3341
3342 static void __init
3343 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3344                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3345 {
3346         struct ring_buffer_event *event;
3347         struct ring_buffer *buffer;
3348         struct ftrace_entry *entry;
3349         unsigned long flags;
3350         long disabled;
3351         int cpu;
3352         int pc;
3353
3354         pc = preempt_count();
3355         preempt_disable_notrace();
3356         cpu = raw_smp_processor_id();
3357         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3358
3359         if (disabled != 1)
3360                 goto out;
3361
3362         local_save_flags(flags);
3363
3364         event = trace_current_buffer_lock_reserve(&buffer,
3365                                                   TRACE_FN, sizeof(*entry),
3366                                                   flags, pc);
3367         if (!event)
3368                 goto out;
3369         entry   = ring_buffer_event_data(event);
3370         entry->ip                       = ip;
3371         entry->parent_ip                = parent_ip;
3372
3373         trace_buffer_unlock_commit(event_tr, buffer, event, flags, pc);
3374
3375  out:
3376         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3377         preempt_enable_notrace();
3378 }
3379
3380 static struct ftrace_ops trace_ops __initdata  =
3381 {
3382         .func = function_test_events_call,
3383         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3384 };
3385
3386 static __init void event_trace_self_test_with_function(void)
3387 {
3388         int ret;
3389         event_tr = top_trace_array();
3390         if (WARN_ON(!event_tr))
3391                 return;
3392         ret = register_ftrace_function(&trace_ops);
3393         if (WARN_ON(ret < 0)) {
3394                 pr_info("Failed to enable function tracer for event tests\n");
3395                 return;
3396         }
3397         pr_info("Running tests again, along with the function tracer\n");
3398         event_trace_self_tests();
3399         unregister_ftrace_function(&trace_ops);
3400 }
3401 #else
3402 static __init void event_trace_self_test_with_function(void)
3403 {
3404 }
3405 #endif
3406
3407 static __init int event_trace_self_tests_init(void)
3408 {
3409         if (!tracing_selftest_disabled) {
3410                 event_trace_self_tests();
3411                 event_trace_self_test_with_function();
3412         }
3413
3414         return 0;
3415 }
3416
3417 late_initcall(event_trace_self_tests_init);
3418
3419 #endif