kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #define pr_fmt(fmt) fmt
  12
  13 #include <linux/workqueue.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/kthread.h>
  16 #include <linux/tracefs.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/module.h>
  19 #include <linux/ctype.h>
  20 #include <linux/sort.h>
  21 #include <linux/slab.h>
  22 #include <linux/delay.h>
  23
  24 #include <trace/events/sched.h>
  25
  26 #include <asm/setup.h>
  27
  28 #include "trace_output.h"
  29
  30 #undef TRACE_SYSTEM
  31 #define TRACE_SYSTEM "TRACE_SYSTEM"
  32
  33 DEFINE_MUTEX(event_mutex);
  34
  35 LIST_HEAD(ftrace_events);
  36 static LIST_HEAD(ftrace_generic_fields);
  37 static LIST_HEAD(ftrace_common_fields);
  38
  39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41 static struct kmem_cache *field_cachep;
  42 static struct kmem_cache *file_cachep;
  43
  44 static inline int system_refcount(struct event_subsystem *system)
  45 {
  46         return system->ref_count;
  47 }
  48
  49 static int system_refcount_inc(struct event_subsystem *system)
  50 {
  51         return system->ref_count++;
  52 }
  53
  54 static int system_refcount_dec(struct event_subsystem *system)
  55 {
  56         return --system->ref_count;
  57 }
  58
  59 /* Double loops, do not use break, only goto's work */
  60 #define do_for_each_event_file(tr, file)                        \
  61         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  62                 list_for_each_entry(file, &tr->events, list)
  63
  64 #define do_for_each_event_file_safe(tr, file)                   \
  65         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  66                 struct trace_event_file *___n;                          \
  67                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  68
  69 #define while_for_each_event_file()             \
  70         }
  71
  72 static struct list_head *
  73 trace_get_fields(struct trace_event_call *event_call)
  74 {
  75         if (!event_call->class->get_fields)
  76                 return &event_call->class->fields;
  77         return event_call->class->get_fields(event_call);
  78 }
  79
  80 static struct ftrace_event_field *
  81 __find_event_field(struct list_head *head, char *name)
  82 {
  83         struct ftrace_event_field *field;
  84
  85         list_for_each_entry(field, head, link) {
  86                 if (!strcmp(field->name, name))
  87                         return field;
  88         }
  89
  90         return NULL;
  91 }
  92
  93 struct ftrace_event_field *
  94 trace_find_event_field(struct trace_event_call *call, char *name)
  95 {
  96         struct ftrace_event_field *field;
  97         struct list_head *head;
  98
  99         head = trace_get_fields(call);
 100         field = __find_event_field(head, name);
 101         if (field)
 102                 return field;
 103
 104         field = __find_event_field(&ftrace_generic_fields, name);
 105         if (field)
 106                 return field;
 107
 108         return __find_event_field(&ftrace_common_fields, name);
 109 }
 110
 111 static int __trace_define_field(struct list_head *head, const char *type,
 112                                 const char *name, int offset, int size,
 113                                 int is_signed, int filter_type)
 114 {
 115         struct ftrace_event_field *field;
 116
 117         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 118         if (!field)
 119                 return -ENOMEM;
 120
 121         field->name = name;
 122         field->type = type;
 123
 124         if (filter_type == FILTER_OTHER)
 125                 field->filter_type = filter_assign_type(type);
 126         else
 127                 field->filter_type = filter_type;
 128
 129         field->offset = offset;
 130         field->size = size;
 131         field->is_signed = is_signed;
 132
 133         list_add(&field->link, head);
 134
 135         return 0;
 136 }
 137
 138 int trace_define_field(struct trace_event_call *call, const char *type,
 139                        const char *name, int offset, int size, int is_signed,
 140                        int filter_type)
 141 {
 142         struct list_head *head;
 143
 144         if (WARN_ON(!call->class))
 145                 return 0;
 146
 147         head = trace_get_fields(call);
 148         return __trace_define_field(head, type, name, offset, size,
 149                                     is_signed, filter_type);
 150 }
 151 EXPORT_SYMBOL_GPL(trace_define_field);
 152
 153 #define __generic_field(type, item, filter_type)                        \
 154         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 155                                    #item, 0, 0, is_signed_type(type),   \
 156                                    filter_type);                        \
 157         if (ret)                                                        \
 158                 return ret;
 159
 160 #define __common_field(type, item)                                      \
 161         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 162                                    "common_" #item,                     \
 163                                    offsetof(typeof(ent), item),         \
 164                                    sizeof(ent.item),                    \
 165                                    is_signed_type(type), FILTER_OTHER); \
 166         if (ret)                                                        \
 167                 return ret;
 168
 169 static int trace_define_generic_fields(void)
 170 {
 171         int ret;
 172
 173         __generic_field(int, CPU, FILTER_CPU);
 174         __generic_field(int, cpu, FILTER_CPU);
 175         __generic_field(char *, COMM, FILTER_COMM);
 176         __generic_field(char *, comm, FILTER_COMM);
 177
 178         return ret;
 179 }
 180
 181 static int trace_define_common_fields(void)
 182 {
 183         int ret;
 184         struct trace_entry ent;
 185
 186         __common_field(unsigned short, type);
 187         __common_field(unsigned char, flags);
 188         __common_field(unsigned char, preempt_count);
 189         __common_field(int, pid);
 190
 191         return ret;
 192 }
 193
 194 static void trace_destroy_fields(struct trace_event_call *call)
 195 {
 196         struct ftrace_event_field *field, *next;
 197         struct list_head *head;
 198
 199         head = trace_get_fields(call);
 200         list_for_each_entry_safe(field, next, head, link) {
 201                 list_del(&field->link);
 202                 kmem_cache_free(field_cachep, field);
 203         }
 204 }
 205
 206 /*
 207  * run-time version of trace_event_get_offsets_<call>() that returns the last
 208  * accessible offset of trace fields excluding __dynamic_array bytes
 209  */
 210 int trace_event_get_offsets(struct trace_event_call *call)
 211 {
 212         struct ftrace_event_field *tail;
 213         struct list_head *head;
 214
 215         head = trace_get_fields(call);
 216         /*
 217          * head->next points to the last field with the largest offset,
 218          * since it was added last by trace_define_field()
 219          */
 220         tail = list_first_entry(head, struct ftrace_event_field, link);
 221         return tail->offset + tail->size;
 222 }
 223
 224 int trace_event_raw_init(struct trace_event_call *call)
 225 {
 226         int id;
 227
 228         id = register_trace_event(&call->event);
 229         if (!id)
 230                 return -ENODEV;
 231
 232         return 0;
 233 }
 234 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 235
 236 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 237 {
 238         struct trace_array *tr = trace_file->tr;
 239         struct trace_array_cpu *data;
 240         struct trace_pid_list *pid_list;
 241
 242         pid_list = rcu_dereference_sched(tr->filtered_pids);
 243         if (!pid_list)
 244                 return false;
 245
 246         data = this_cpu_ptr(tr->trace_buffer.data);
 247
 248         return data->ignore_pid;
 249 }
 250 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 251
 252 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 253                                  struct trace_event_file *trace_file,
 254                                  unsigned long len)
 255 {
 256         struct trace_event_call *event_call = trace_file->event_call;
 257
 258         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 259             trace_event_ignore_this_pid(trace_file))
 260                 return NULL;
 261
 262         local_save_flags(fbuffer->flags);
 263         fbuffer->pc = preempt_count();
 264         /*
 265          * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
 266          * preemption (adding one to the preempt_count). Since we are
 267          * interested in the preempt_count at the time the tracepoint was
 268          * hit, we need to subtract one to offset the increment.
 269          */
 270         if (IS_ENABLED(CONFIG_PREEMPT))
 271                 fbuffer->pc--;
 272         fbuffer->trace_file = trace_file;
 273
 274         fbuffer->event =
 275                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 276                                                 event_call->event.type, len,
 277                                                 fbuffer->flags, fbuffer->pc);
 278         if (!fbuffer->event)
 279                 return NULL;
 280
 281         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 282         return fbuffer->entry;
 283 }
 284 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 285
 286 int trace_event_reg(struct trace_event_call *call,
 287                     enum trace_reg type, void *data)
 288 {
 289         struct trace_event_file *file = data;
 290
 291         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 292         switch (type) {
 293         case TRACE_REG_REGISTER:
 294                 return tracepoint_probe_register(call->tp,
 295                                                  call->class->probe,
 296                                                  file);
 297         case TRACE_REG_UNREGISTER:
 298                 tracepoint_probe_unregister(call->tp,
 299                                             call->class->probe,
 300                                             file);
 301                 return 0;
 302
 303 #ifdef CONFIG_PERF_EVENTS
 304         case TRACE_REG_PERF_REGISTER:
 305                 return tracepoint_probe_register(call->tp,
 306                                                  call->class->perf_probe,
 307                                                  call);
 308         case TRACE_REG_PERF_UNREGISTER:
 309                 tracepoint_probe_unregister(call->tp,
 310                                             call->class->perf_probe,
 311                                             call);
 312                 return 0;
 313         case TRACE_REG_PERF_OPEN:
 314         case TRACE_REG_PERF_CLOSE:
 315         case TRACE_REG_PERF_ADD:
 316         case TRACE_REG_PERF_DEL:
 317                 return 0;
 318 #endif
 319         }
 320         return 0;
 321 }
 322 EXPORT_SYMBOL_GPL(trace_event_reg);
 323
 324 void trace_event_enable_cmd_record(bool enable)
 325 {
 326         struct trace_event_file *file;
 327         struct trace_array *tr;
 328
 329         mutex_lock(&event_mutex);
 330         do_for_each_event_file(tr, file) {
 331
 332                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 333                         continue;
 334
 335                 if (enable) {
 336                         tracing_start_cmdline_record();
 337                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 338                 } else {
 339                         tracing_stop_cmdline_record();
 340                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 341                 }
 342         } while_for_each_event_file();
 343         mutex_unlock(&event_mutex);
 344 }
 345
 346 void trace_event_enable_tgid_record(bool enable)
 347 {
 348         struct trace_event_file *file;
 349         struct trace_array *tr;
 350
 351         mutex_lock(&event_mutex);
 352         do_for_each_event_file(tr, file) {
 353                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 354                         continue;
 355
 356                 if (enable) {
 357                         tracing_start_tgid_record();
 358                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 359                 } else {
 360                         tracing_stop_tgid_record();
 361                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 362                                   &file->flags);
 363                 }
 364         } while_for_each_event_file();
 365         mutex_unlock(&event_mutex);
 366 }
 367
 368 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 369                                          int enable, int soft_disable)
 370 {
 371         struct trace_event_call *call = file->event_call;
 372         struct trace_array *tr = file->tr;
 373         unsigned long file_flags = file->flags;
 374         int ret = 0;
 375         int disable;
 376
 377         switch (enable) {
 378         case 0:
 379                 /*
 380                  * When soft_disable is set and enable is cleared, the sm_ref
 381                  * reference counter is decremented. If it reaches 0, we want
 382                  * to clear the SOFT_DISABLED flag but leave the event in the
 383                  * state that it was. That is, if the event was enabled and
 384                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 385                  * is set we do not want the event to be enabled before we
 386                  * clear the bit.
 387                  *
 388                  * When soft_disable is not set but the SOFT_MODE flag is,
 389                  * we do nothing. Do not disable the tracepoint, otherwise
 390                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 391                  */
 392                 if (soft_disable) {
 393                         if (atomic_dec_return(&file->sm_ref) > 0)
 394                                 break;
 395                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 396                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 397                 } else
 398                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 399
 400                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 401                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 402                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 403                                 tracing_stop_cmdline_record();
 404                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 405                         }
 406
 407                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 408                                 tracing_stop_tgid_record();
 409                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 410                         }
 411
 412                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 413                 }
 414                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 415                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 416                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 417                 else
 418                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 419                 break;
 420         case 1:
 421                 /*
 422                  * When soft_disable is set and enable is set, we want to
 423                  * register the tracepoint for the event, but leave the event
 424                  * as is. That means, if the event was already enabled, we do
 425                  * nothing (but set SOFT_MODE). If the event is disabled, we
 426                  * set SOFT_DISABLED before enabling the event tracepoint, so
 427                  * it still seems to be disabled.
 428                  */
 429                 if (!soft_disable)
 430                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 431                 else {
 432                         if (atomic_inc_return(&file->sm_ref) > 1)
 433                                 break;
 434                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 435                 }
 436
 437                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 438                         bool cmd = false, tgid = false;
 439
 440                         /* Keep the event disabled, when going to SOFT_MODE. */
 441                         if (soft_disable)
 442                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 443
 444                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 445                                 cmd = true;
 446                                 tracing_start_cmdline_record();
 447                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 448                         }
 449
 450                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 451                                 tgid = true;
 452                                 tracing_start_tgid_record();
 453                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 454                         }
 455
 456                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 457                         if (ret) {
 458                                 if (cmd)
 459                                         tracing_stop_cmdline_record();
 460                                 if (tgid)
 461                                         tracing_stop_tgid_record();
 462                                 pr_info("event trace: Could not enable event "
 463                                         "%s\n", trace_event_name(call));
 464                                 break;
 465                         }
 466                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 467
 468                         /* WAS_ENABLED gets set but never cleared. */
 469                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 470                 }
 471                 break;
 472         }
 473
 474         /* Enable or disable use of trace_buffered_event */
 475         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 476             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 477                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 478                         trace_buffered_event_enable();
 479                 else
 480                         trace_buffered_event_disable();
 481         }
 482
 483         return ret;
 484 }
 485
 486 int trace_event_enable_disable(struct trace_event_file *file,
 487                                int enable, int soft_disable)
 488 {
 489         return __ftrace_event_enable_disable(file, enable, soft_disable);
 490 }
 491
 492 static int ftrace_event_enable_disable(struct trace_event_file *file,
 493                                        int enable)
 494 {
 495         return __ftrace_event_enable_disable(file, enable, 0);
 496 }
 497
 498 static void ftrace_clear_events(struct trace_array *tr)
 499 {
 500         struct trace_event_file *file;
 501
 502         mutex_lock(&event_mutex);
 503         list_for_each_entry(file, &tr->events, list) {
 504                 ftrace_event_enable_disable(file, 0);
 505         }
 506         mutex_unlock(&event_mutex);
 507 }
 508
 509 static void
 510 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 511 {
 512         struct trace_pid_list *pid_list;
 513         struct trace_array *tr = data;
 514
 515         pid_list = rcu_dereference_sched(tr->filtered_pids);
 516         trace_filter_add_remove_task(pid_list, NULL, task);
 517 }
 518
 519 static void
 520 event_filter_pid_sched_process_fork(void *data,
 521                                     struct task_struct *self,
 522                                     struct task_struct *task)
 523 {
 524         struct trace_pid_list *pid_list;
 525         struct trace_array *tr = data;
 526
 527         pid_list = rcu_dereference_sched(tr->filtered_pids);
 528         trace_filter_add_remove_task(pid_list, self, task);
 529 }
 530
 531 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 532 {
 533         if (enable) {
 534                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 535                                                        tr, INT_MIN);
 536                 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 537                                                        tr, INT_MAX);
 538         } else {
 539                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 540                                                     tr);
 541                 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 542                                                     tr);
 543         }
 544 }
 545
 546 static void
 547 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 548                     struct task_struct *prev, struct task_struct *next)
 549 {
 550         struct trace_array *tr = data;
 551         struct trace_pid_list *pid_list;
 552
 553         pid_list = rcu_dereference_sched(tr->filtered_pids);
 554
 555         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 556                        trace_ignore_this_task(pid_list, prev) &&
 557                        trace_ignore_this_task(pid_list, next));
 558 }
 559
 560 static void
 561 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 562                     struct task_struct *prev, struct task_struct *next)
 563 {
 564         struct trace_array *tr = data;
 565         struct trace_pid_list *pid_list;
 566
 567         pid_list = rcu_dereference_sched(tr->filtered_pids);
 568
 569         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 570                        trace_ignore_this_task(pid_list, next));
 571 }
 572
 573 static void
 574 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 575 {
 576         struct trace_array *tr = data;
 577         struct trace_pid_list *pid_list;
 578
 579         /* Nothing to do if we are already tracing */
 580         if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 581                 return;
 582
 583         pid_list = rcu_dereference_sched(tr->filtered_pids);
 584
 585         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 586                        trace_ignore_this_task(pid_list, task));
 587 }
 588
 589 static void
 590 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 591 {
 592         struct trace_array *tr = data;
 593         struct trace_pid_list *pid_list;
 594
 595         /* Nothing to do if we are not tracing */
 596         if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 597                 return;
 598
 599         pid_list = rcu_dereference_sched(tr->filtered_pids);
 600
 601         /* Set tracing if current is enabled */
 602         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 603                        trace_ignore_this_task(pid_list, current));
 604 }
 605
 606 static void __ftrace_clear_event_pids(struct trace_array *tr)
 607 {
 608         struct trace_pid_list *pid_list;
 609         struct trace_event_file *file;
 610         int cpu;
 611
 612         pid_list = rcu_dereference_protected(tr->filtered_pids,
 613                                              lockdep_is_held(&event_mutex));
 614         if (!pid_list)
 615                 return;
 616
 617         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 618         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 619
 620         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 621         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 622
 623         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 624         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 625
 626         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 627         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 628
 629         list_for_each_entry(file, &tr->events, list) {
 630                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 631         }
 632
 633         for_each_possible_cpu(cpu)
 634                 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 635
 636         rcu_assign_pointer(tr->filtered_pids, NULL);
 637
 638         /* Wait till all users are no longer using pid filtering */
 639         synchronize_sched();
 640
 641         trace_free_pid_list(pid_list);
 642 }
 643
 644 static void ftrace_clear_event_pids(struct trace_array *tr)
 645 {
 646         mutex_lock(&event_mutex);
 647         __ftrace_clear_event_pids(tr);
 648         mutex_unlock(&event_mutex);
 649 }
 650
 651 static void __put_system(struct event_subsystem *system)
 652 {
 653         struct event_filter *filter = system->filter;
 654
 655         WARN_ON_ONCE(system_refcount(system) == 0);
 656         if (system_refcount_dec(system))
 657                 return;
 658
 659         list_del(&system->list);
 660
 661         if (filter) {
 662                 kfree(filter->filter_string);
 663                 kfree(filter);
 664         }
 665         kfree_const(system->name);
 666         kfree(system);
 667 }
 668
 669 static void __get_system(struct event_subsystem *system)
 670 {
 671         WARN_ON_ONCE(system_refcount(system) == 0);
 672         system_refcount_inc(system);
 673 }
 674
 675 static void __get_system_dir(struct trace_subsystem_dir *dir)
 676 {
 677         WARN_ON_ONCE(dir->ref_count == 0);
 678         dir->ref_count++;
 679         __get_system(dir->subsystem);
 680 }
 681
 682 static void __put_system_dir(struct trace_subsystem_dir *dir)
 683 {
 684         WARN_ON_ONCE(dir->ref_count == 0);
 685         /* If the subsystem is about to be freed, the dir must be too */
 686         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 687
 688         __put_system(dir->subsystem);
 689         if (!--dir->ref_count)
 690                 kfree(dir);
 691 }
 692
 693 static void put_system(struct trace_subsystem_dir *dir)
 694 {
 695         mutex_lock(&event_mutex);
 696         __put_system_dir(dir);
 697         mutex_unlock(&event_mutex);
 698 }
 699
 700 static void remove_subsystem(struct trace_subsystem_dir *dir)
 701 {
 702         if (!dir)
 703                 return;
 704
 705         if (!--dir->nr_events) {
 706                 tracefs_remove_recursive(dir->entry);
 707                 list_del(&dir->list);
 708                 __put_system_dir(dir);
 709         }
 710 }
 711
 712 static void remove_event_file_dir(struct trace_event_file *file)
 713 {
 714         struct dentry *dir = file->dir;
 715         struct dentry *child;
 716
 717         if (dir) {
 718                 spin_lock(&dir->d_lock);        /* probably unneeded */
 719                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 720                         if (d_really_is_positive(child))        /* probably unneeded */
 721                                 d_inode(child)->i_private = NULL;
 722                 }
 723                 spin_unlock(&dir->d_lock);
 724
 725                 tracefs_remove_recursive(dir);
 726         }
 727
 728         list_del(&file->list);
 729         remove_subsystem(file->system);
 730         free_event_filter(file->filter);
 731         kmem_cache_free(file_cachep, file);
 732 }
 733
 734 /*
 735  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 736  */
 737 static int
 738 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 739                               const char *sub, const char *event, int set)
 740 {
 741         struct trace_event_file *file;
 742         struct trace_event_call *call;
 743         const char *name;
 744         int ret = -EINVAL;
 745         int eret = 0;
 746
 747         list_for_each_entry(file, &tr->events, list) {
 748
 749                 call = file->event_call;
 750                 name = trace_event_name(call);
 751
 752                 if (!name || !call->class || !call->class->reg)
 753                         continue;
 754
 755                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 756                         continue;
 757
 758                 if (match &&
 759                     strcmp(match, name) != 0 &&
 760                     strcmp(match, call->class->system) != 0)
 761                         continue;
 762
 763                 if (sub && strcmp(sub, call->class->system) != 0)
 764                         continue;
 765
 766                 if (event && strcmp(event, name) != 0)
 767                         continue;
 768
 769                 ret = ftrace_event_enable_disable(file, set);
 770
 771                 /*
 772                  * Save the first error and return that. Some events
 773                  * may still have been enabled, but let the user
 774                  * know that something went wrong.
 775                  */
 776                 if (ret && !eret)
 777                         eret = ret;
 778
 779                 ret = eret;
 780         }
 781
 782         return ret;
 783 }
 784
 785 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 786                                   const char *sub, const char *event, int set)
 787 {
 788         int ret;
 789
 790         mutex_lock(&event_mutex);
 791         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 792         mutex_unlock(&event_mutex);
 793
 794         return ret;
 795 }
 796
 797 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 798 {
 799         char *event = NULL, *sub = NULL, *match;
 800         int ret;
 801
 802         /*
 803          * The buf format can be <subsystem>:<event-name>
 804          *  *:<event-name> means any event by that name.
 805          *  :<event-name> is the same.
 806          *
 807          *  <subsystem>:* means all events in that subsystem
 808          *  <subsystem>: means the same.
 809          *
 810          *  <name> (no ':') means all events in a subsystem with
 811          *  the name <name> or any event that matches <name>
 812          */
 813
 814         match = strsep(&buf, ":");
 815         if (buf) {
 816                 sub = match;
 817                 event = buf;
 818                 match = NULL;
 819
 820                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 821                         sub = NULL;
 822                 if (!strlen(event) || strcmp(event, "*") == 0)
 823                         event = NULL;
 824         }
 825
 826         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 827
 828         /* Put back the colon to allow this to be called again */
 829         if (buf)
 830                 *(buf - 1) = ':';
 831
 832         return ret;
 833 }
 834
 835 /**
 836  * trace_set_clr_event - enable or disable an event
 837  * @system: system name to match (NULL for any system)
 838  * @event: event name to match (NULL for all events, within system)
 839  * @set: 1 to enable, 0 to disable
 840  *
 841  * This is a way for other parts of the kernel to enable or disable
 842  * event recording.
 843  *
 844  * Returns 0 on success, -EINVAL if the parameters do not match any
 845  * registered events.
 846  */
 847 int trace_set_clr_event(const char *system, const char *event, int set)
 848 {
 849         struct trace_array *tr = top_trace_array();
 850
 851         if (!tr)
 852                 return -ENODEV;
 853
 854         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 855 }
 856 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 857
 858 /* 128 should be much more than enough */
 859 #define EVENT_BUF_SIZE          127
 860
 861 static ssize_t
 862 ftrace_event_write(struct file *file, const char __user *ubuf,
 863                    size_t cnt, loff_t *ppos)
 864 {
 865         struct trace_parser parser;
 866         struct seq_file *m = file->private_data;
 867         struct trace_array *tr = m->private;
 868         ssize_t read, ret;
 869
 870         if (!cnt)
 871                 return 0;
 872
 873         ret = tracing_update_buffers();
 874         if (ret < 0)
 875                 return ret;
 876
 877         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 878                 return -ENOMEM;
 879
 880         read = trace_get_user(&parser, ubuf, cnt, ppos);
 881
 882         if (read >= 0 && trace_parser_loaded((&parser))) {
 883                 int set = 1;
 884
 885                 if (*parser.buffer == '!')
 886                         set = 0;
 887
 888                 parser.buffer[parser.idx] = 0;
 889
 890                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 891                 if (ret)
 892                         goto out_put;
 893         }
 894
 895         ret = read;
 896
 897  out_put:
 898         trace_parser_put(&parser);
 899
 900         return ret;
 901 }
 902
 903 static void *
 904 t_next(struct seq_file *m, void *v, loff_t *pos)
 905 {
 906         struct trace_event_file *file = v;
 907         struct trace_event_call *call;
 908         struct trace_array *tr = m->private;
 909
 910         (*pos)++;
 911
 912         list_for_each_entry_continue(file, &tr->events, list) {
 913                 call = file->event_call;
 914                 /*
 915                  * The ftrace subsystem is for showing formats only.
 916                  * They can not be enabled or disabled via the event files.
 917                  */
 918                 if (call->class && call->class->reg &&
 919                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 920                         return file;
 921         }
 922
 923         return NULL;
 924 }
 925
 926 static void *t_start(struct seq_file *m, loff_t *pos)
 927 {
 928         struct trace_event_file *file;
 929         struct trace_array *tr = m->private;
 930         loff_t l;
 931
 932         mutex_lock(&event_mutex);
 933
 934         file = list_entry(&tr->events, struct trace_event_file, list);
 935         for (l = 0; l <= *pos; ) {
 936                 file = t_next(m, file, &l);
 937                 if (!file)
 938                         break;
 939         }
 940         return file;
 941 }
 942
 943 static void *
 944 s_next(struct seq_file *m, void *v, loff_t *pos)
 945 {
 946         struct trace_event_file *file = v;
 947         struct trace_array *tr = m->private;
 948
 949         (*pos)++;
 950
 951         list_for_each_entry_continue(file, &tr->events, list) {
 952                 if (file->flags & EVENT_FILE_FL_ENABLED)
 953                         return file;
 954         }
 955
 956         return NULL;
 957 }
 958
 959 static void *s_start(struct seq_file *m, loff_t *pos)
 960 {
 961         struct trace_event_file *file;
 962         struct trace_array *tr = m->private;
 963         loff_t l;
 964
 965         mutex_lock(&event_mutex);
 966
 967         file = list_entry(&tr->events, struct trace_event_file, list);
 968         for (l = 0; l <= *pos; ) {
 969                 file = s_next(m, file, &l);
 970                 if (!file)
 971                         break;
 972         }
 973         return file;
 974 }
 975
 976 static int t_show(struct seq_file *m, void *v)
 977 {
 978         struct trace_event_file *file = v;
 979         struct trace_event_call *call = file->event_call;
 980
 981         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 982                 seq_printf(m, "%s:", call->class->system);
 983         seq_printf(m, "%s\n", trace_event_name(call));
 984
 985         return 0;
 986 }
 987
 988 static void t_stop(struct seq_file *m, void *p)
 989 {
 990         mutex_unlock(&event_mutex);
 991 }
 992
 993 static void *
 994 p_next(struct seq_file *m, void *v, loff_t *pos)
 995 {
 996         struct trace_array *tr = m->private;
 997         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 998
 999         return trace_pid_next(pid_list, v, pos);
1000 }
1001
1002 static void *p_start(struct seq_file *m, loff_t *pos)
1003         __acquires(RCU)
1004 {
1005         struct trace_pid_list *pid_list;
1006         struct trace_array *tr = m->private;
1007
1008         /*
1009          * Grab the mutex, to keep calls to p_next() having the same
1010          * tr->filtered_pids as p_start() has.
1011          * If we just passed the tr->filtered_pids around, then RCU would
1012          * have been enough, but doing that makes things more complex.
1013          */
1014         mutex_lock(&event_mutex);
1015         rcu_read_lock_sched();
1016
1017         pid_list = rcu_dereference_sched(tr->filtered_pids);
1018
1019         if (!pid_list)
1020                 return NULL;
1021
1022         return trace_pid_start(pid_list, pos);
1023 }
1024
1025 static void p_stop(struct seq_file *m, void *p)
1026         __releases(RCU)
1027 {
1028         rcu_read_unlock_sched();
1029         mutex_unlock(&event_mutex);
1030 }
1031
1032 static ssize_t
1033 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1034                   loff_t *ppos)
1035 {
1036         struct trace_event_file *file;
1037         unsigned long flags;
1038         char buf[4] = "0";
1039
1040         mutex_lock(&event_mutex);
1041         file = event_file_data(filp);
1042         if (likely(file))
1043                 flags = file->flags;
1044         mutex_unlock(&event_mutex);
1045
1046         if (!file)
1047                 return -ENODEV;
1048
1049         if (flags & EVENT_FILE_FL_ENABLED &&
1050             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1051                 strcpy(buf, "1");
1052
1053         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1054             flags & EVENT_FILE_FL_SOFT_MODE)
1055                 strcat(buf, "*");
1056
1057         strcat(buf, "\n");
1058
1059         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1060 }
1061
1062 static ssize_t
1063 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1064                    loff_t *ppos)
1065 {
1066         struct trace_event_file *file;
1067         unsigned long val;
1068         int ret;
1069
1070         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1071         if (ret)
1072                 return ret;
1073
1074         ret = tracing_update_buffers();
1075         if (ret < 0)
1076                 return ret;
1077
1078         switch (val) {
1079         case 0:
1080         case 1:
1081                 ret = -ENODEV;
1082                 mutex_lock(&event_mutex);
1083                 file = event_file_data(filp);
1084                 if (likely(file))
1085                         ret = ftrace_event_enable_disable(file, val);
1086                 mutex_unlock(&event_mutex);
1087                 break;
1088
1089         default:
1090                 return -EINVAL;
1091         }
1092
1093         *ppos += cnt;
1094
1095         return ret ? ret : cnt;
1096 }
1097
1098 static ssize_t
1099 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1100                    loff_t *ppos)
1101 {
1102         const char set_to_char[4] = { '?', '0', '1', 'X' };
1103         struct trace_subsystem_dir *dir = filp->private_data;
1104         struct event_subsystem *system = dir->subsystem;
1105         struct trace_event_call *call;
1106         struct trace_event_file *file;
1107         struct trace_array *tr = dir->tr;
1108         char buf[2];
1109         int set = 0;
1110         int ret;
1111
1112         mutex_lock(&event_mutex);
1113         list_for_each_entry(file, &tr->events, list) {
1114                 call = file->event_call;
1115                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1116                         continue;
1117
1118                 if (system && strcmp(call->class->system, system->name) != 0)
1119                         continue;
1120
1121                 /*
1122                  * We need to find out if all the events are set
1123                  * or if all events or cleared, or if we have
1124                  * a mixture.
1125                  */
1126                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1127
1128                 /*
1129                  * If we have a mixture, no need to look further.
1130                  */
1131                 if (set == 3)
1132                         break;
1133         }
1134         mutex_unlock(&event_mutex);
1135
1136         buf[0] = set_to_char[set];
1137         buf[1] = '\n';
1138
1139         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1140
1141         return ret;
1142 }
1143
1144 static ssize_t
1145 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1146                     loff_t *ppos)
1147 {
1148         struct trace_subsystem_dir *dir = filp->private_data;
1149         struct event_subsystem *system = dir->subsystem;
1150         const char *name = NULL;
1151         unsigned long val;
1152         ssize_t ret;
1153
1154         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1155         if (ret)
1156                 return ret;
1157
1158         ret = tracing_update_buffers();
1159         if (ret < 0)
1160                 return ret;
1161
1162         if (val != 0 && val != 1)
1163                 return -EINVAL;
1164
1165         /*
1166          * Opening of "enable" adds a ref count to system,
1167          * so the name is safe to use.
1168          */
1169         if (system)
1170                 name = system->name;
1171
1172         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1173         if (ret)
1174                 goto out;
1175
1176         ret = cnt;
1177
1178 out:
1179         *ppos += cnt;
1180
1181         return ret;
1182 }
1183
1184 enum {
1185         FORMAT_HEADER           = 1,
1186         FORMAT_FIELD_SEPERATOR  = 2,
1187         FORMAT_PRINTFMT         = 3,
1188 };
1189
1190 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1191 {
1192         struct trace_event_call *call = event_file_data(m->private);
1193         struct list_head *common_head = &ftrace_common_fields;
1194         struct list_head *head = trace_get_fields(call);
1195         struct list_head *node = v;
1196
1197         (*pos)++;
1198
1199         switch ((unsigned long)v) {
1200         case FORMAT_HEADER:
1201                 node = common_head;
1202                 break;
1203
1204         case FORMAT_FIELD_SEPERATOR:
1205                 node = head;
1206                 break;
1207
1208         case FORMAT_PRINTFMT:
1209                 /* all done */
1210                 return NULL;
1211         }
1212
1213         node = node->prev;
1214         if (node == common_head)
1215                 return (void *)FORMAT_FIELD_SEPERATOR;
1216         else if (node == head)
1217                 return (void *)FORMAT_PRINTFMT;
1218         else
1219                 return node;
1220 }
1221
1222 static int f_show(struct seq_file *m, void *v)
1223 {
1224         struct trace_event_call *call = event_file_data(m->private);
1225         struct ftrace_event_field *field;
1226         const char *array_descriptor;
1227
1228         switch ((unsigned long)v) {
1229         case FORMAT_HEADER:
1230                 seq_printf(m, "name: %s\n", trace_event_name(call));
1231                 seq_printf(m, "ID: %d\n", call->event.type);
1232                 seq_puts(m, "format:\n");
1233                 return 0;
1234
1235         case FORMAT_FIELD_SEPERATOR:
1236                 seq_putc(m, '\n');
1237                 return 0;
1238
1239         case FORMAT_PRINTFMT:
1240                 seq_printf(m, "\nprint fmt: %s\n",
1241                            call->print_fmt);
1242                 return 0;
1243         }
1244
1245         field = list_entry(v, struct ftrace_event_field, link);
1246         /*
1247          * Smartly shows the array type(except dynamic array).
1248          * Normal:
1249          *      field:TYPE VAR
1250          * If TYPE := TYPE[LEN], it is shown:
1251          *      field:TYPE VAR[LEN]
1252          */
1253         array_descriptor = strchr(field->type, '[');
1254
1255         if (!strncmp(field->type, "__data_loc", 10))
1256                 array_descriptor = NULL;
1257
1258         if (!array_descriptor)
1259                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1260                            field->type, field->name, field->offset,
1261                            field->size, !!field->is_signed);
1262         else
1263                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1264                            (int)(array_descriptor - field->type),
1265                            field->type, field->name,
1266                            array_descriptor, field->offset,
1267                            field->size, !!field->is_signed);
1268
1269         return 0;
1270 }
1271
1272 static void *f_start(struct seq_file *m, loff_t *pos)
1273 {
1274         void *p = (void *)FORMAT_HEADER;
1275         loff_t l = 0;
1276
1277         /* ->stop() is called even if ->start() fails */
1278         mutex_lock(&event_mutex);
1279         if (!event_file_data(m->private))
1280                 return ERR_PTR(-ENODEV);
1281
1282         while (l < *pos && p)
1283                 p = f_next(m, p, &l);
1284
1285         return p;
1286 }
1287
1288 static void f_stop(struct seq_file *m, void *p)
1289 {
1290         mutex_unlock(&event_mutex);
1291 }
1292
1293 static const struct seq_operations trace_format_seq_ops = {
1294         .start          = f_start,
1295         .next           = f_next,
1296         .stop           = f_stop,
1297         .show           = f_show,
1298 };
1299
1300 static int trace_format_open(struct inode *inode, struct file *file)
1301 {
1302         struct seq_file *m;
1303         int ret;
1304
1305         ret = seq_open(file, &trace_format_seq_ops);
1306         if (ret < 0)
1307                 return ret;
1308
1309         m = file->private_data;
1310         m->private = file;
1311
1312         return 0;
1313 }
1314
1315 static ssize_t
1316 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1317 {
1318         int id = (long)event_file_data(filp);
1319         char buf[32];
1320         int len;
1321
1322         if (*ppos)
1323                 return 0;
1324
1325         if (unlikely(!id))
1326                 return -ENODEV;
1327
1328         len = sprintf(buf, "%d\n", id);
1329
1330         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1331 }
1332
1333 static ssize_t
1334 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1335                   loff_t *ppos)
1336 {
1337         struct trace_event_file *file;
1338         struct trace_seq *s;
1339         int r = -ENODEV;
1340
1341         if (*ppos)
1342                 return 0;
1343
1344         s = kmalloc(sizeof(*s), GFP_KERNEL);
1345
1346         if (!s)
1347                 return -ENOMEM;
1348
1349         trace_seq_init(s);
1350
1351         mutex_lock(&event_mutex);
1352         file = event_file_data(filp);
1353         if (file)
1354                 print_event_filter(file, s);
1355         mutex_unlock(&event_mutex);
1356
1357         if (file)
1358                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1359                                             s->buffer, trace_seq_used(s));
1360
1361         kfree(s);
1362
1363         return r;
1364 }
1365
1366 static ssize_t
1367 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1368                    loff_t *ppos)
1369 {
1370         struct trace_event_file *file;
1371         char *buf;
1372         int err = -ENODEV;
1373
1374         if (cnt >= PAGE_SIZE)
1375                 return -EINVAL;
1376
1377         buf = memdup_user_nul(ubuf, cnt);
1378         if (IS_ERR(buf))
1379                 return PTR_ERR(buf);
1380
1381         mutex_lock(&event_mutex);
1382         file = event_file_data(filp);
1383         if (file)
1384                 err = apply_event_filter(file, buf);
1385         mutex_unlock(&event_mutex);
1386
1387         kfree(buf);
1388         if (err < 0)
1389                 return err;
1390
1391         *ppos += cnt;
1392
1393         return cnt;
1394 }
1395
1396 static LIST_HEAD(event_subsystems);
1397
1398 static int subsystem_open(struct inode *inode, struct file *filp)
1399 {
1400         struct event_subsystem *system = NULL;
1401         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1402         struct trace_array *tr;
1403         int ret;
1404
1405         if (tracing_is_disabled())
1406                 return -ENODEV;
1407
1408         /* Make sure the system still exists */
1409         mutex_lock(&trace_types_lock);
1410         mutex_lock(&event_mutex);
1411         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1412                 list_for_each_entry(dir, &tr->systems, list) {
1413                         if (dir == inode->i_private) {
1414                                 /* Don't open systems with no events */
1415                                 if (dir->nr_events) {
1416                                         __get_system_dir(dir);
1417                                         system = dir->subsystem;
1418                                 }
1419                                 goto exit_loop;
1420                         }
1421                 }
1422         }
1423  exit_loop:
1424         mutex_unlock(&event_mutex);
1425         mutex_unlock(&trace_types_lock);
1426
1427         if (!system)
1428                 return -ENODEV;
1429
1430         /* Some versions of gcc think dir can be uninitialized here */
1431         WARN_ON(!dir);
1432
1433         /* Still need to increment the ref count of the system */
1434         if (trace_array_get(tr) < 0) {
1435                 put_system(dir);
1436                 return -ENODEV;
1437         }
1438
1439         ret = tracing_open_generic(inode, filp);
1440         if (ret < 0) {
1441                 trace_array_put(tr);
1442                 put_system(dir);
1443         }
1444
1445         return ret;
1446 }
1447
1448 static int system_tr_open(struct inode *inode, struct file *filp)
1449 {
1450         struct trace_subsystem_dir *dir;
1451         struct trace_array *tr = inode->i_private;
1452         int ret;
1453
1454         if (tracing_is_disabled())
1455                 return -ENODEV;
1456
1457         if (trace_array_get(tr) < 0)
1458                 return -ENODEV;
1459
1460         /* Make a temporary dir that has no system but points to tr */
1461         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1462         if (!dir) {
1463                 trace_array_put(tr);
1464                 return -ENOMEM;
1465         }
1466
1467         dir->tr = tr;
1468
1469         ret = tracing_open_generic(inode, filp);
1470         if (ret < 0) {
1471                 trace_array_put(tr);
1472                 kfree(dir);
1473                 return ret;
1474         }
1475
1476         filp->private_data = dir;
1477
1478         return 0;
1479 }
1480
1481 static int subsystem_release(struct inode *inode, struct file *file)
1482 {
1483         struct trace_subsystem_dir *dir = file->private_data;
1484
1485         trace_array_put(dir->tr);
1486
1487         /*
1488          * If dir->subsystem is NULL, then this is a temporary
1489          * descriptor that was made for a trace_array to enable
1490          * all subsystems.
1491          */
1492         if (dir->subsystem)
1493                 put_system(dir);
1494         else
1495                 kfree(dir);
1496
1497         return 0;
1498 }
1499
1500 static ssize_t
1501 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1502                       loff_t *ppos)
1503 {
1504         struct trace_subsystem_dir *dir = filp->private_data;
1505         struct event_subsystem *system = dir->subsystem;
1506         struct trace_seq *s;
1507         int r;
1508
1509         if (*ppos)
1510                 return 0;
1511
1512         s = kmalloc(sizeof(*s), GFP_KERNEL);
1513         if (!s)
1514                 return -ENOMEM;
1515
1516         trace_seq_init(s);
1517
1518         print_subsystem_event_filter(system, s);
1519         r = simple_read_from_buffer(ubuf, cnt, ppos,
1520                                     s->buffer, trace_seq_used(s));
1521
1522         kfree(s);
1523
1524         return r;
1525 }
1526
1527 static ssize_t
1528 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1529                        loff_t *ppos)
1530 {
1531         struct trace_subsystem_dir *dir = filp->private_data;
1532         char *buf;
1533         int err;
1534
1535         if (cnt >= PAGE_SIZE)
1536                 return -EINVAL;
1537
1538         buf = memdup_user_nul(ubuf, cnt);
1539         if (IS_ERR(buf))
1540                 return PTR_ERR(buf);
1541
1542         err = apply_subsystem_event_filter(dir, buf);
1543         kfree(buf);
1544         if (err < 0)
1545                 return err;
1546
1547         *ppos += cnt;
1548
1549         return cnt;
1550 }
1551
1552 static ssize_t
1553 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1554 {
1555         int (*func)(struct trace_seq *s) = filp->private_data;
1556         struct trace_seq *s;
1557         int r;
1558
1559         if (*ppos)
1560                 return 0;
1561
1562         s = kmalloc(sizeof(*s), GFP_KERNEL);
1563         if (!s)
1564                 return -ENOMEM;
1565
1566         trace_seq_init(s);
1567
1568         func(s);
1569         r = simple_read_from_buffer(ubuf, cnt, ppos,
1570                                     s->buffer, trace_seq_used(s));
1571
1572         kfree(s);
1573
1574         return r;
1575 }
1576
1577 static void ignore_task_cpu(void *data)
1578 {
1579         struct trace_array *tr = data;
1580         struct trace_pid_list *pid_list;
1581
1582         /*
1583          * This function is called by on_each_cpu() while the
1584          * event_mutex is held.
1585          */
1586         pid_list = rcu_dereference_protected(tr->filtered_pids,
1587                                              mutex_is_locked(&event_mutex));
1588
1589         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1590                        trace_ignore_this_task(pid_list, current));
1591 }
1592
1593 static ssize_t
1594 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1595                        size_t cnt, loff_t *ppos)
1596 {
1597         struct seq_file *m = filp->private_data;
1598         struct trace_array *tr = m->private;
1599         struct trace_pid_list *filtered_pids = NULL;
1600         struct trace_pid_list *pid_list;
1601         struct trace_event_file *file;
1602         ssize_t ret;
1603
1604         if (!cnt)
1605                 return 0;
1606
1607         ret = tracing_update_buffers();
1608         if (ret < 0)
1609                 return ret;
1610
1611         mutex_lock(&event_mutex);
1612
1613         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1614                                              lockdep_is_held(&event_mutex));
1615
1616         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1617         if (ret < 0)
1618                 goto out;
1619
1620         rcu_assign_pointer(tr->filtered_pids, pid_list);
1621
1622         list_for_each_entry(file, &tr->events, list) {
1623                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1624         }
1625
1626         if (filtered_pids) {
1627                 synchronize_sched();
1628                 trace_free_pid_list(filtered_pids);
1629         } else if (pid_list) {
1630                 /*
1631                  * Register a probe that is called before all other probes
1632                  * to set ignore_pid if next or prev do not match.
1633                  * Register a probe this is called after all other probes
1634                  * to only keep ignore_pid set if next pid matches.
1635                  */
1636                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1637                                                  tr, INT_MAX);
1638                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1639                                                  tr, 0);
1640
1641                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1642                                                  tr, INT_MAX);
1643                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1644                                                  tr, 0);
1645
1646                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1647                                                      tr, INT_MAX);
1648                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1649                                                      tr, 0);
1650
1651                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1652                                                  tr, INT_MAX);
1653                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1654                                                  tr, 0);
1655         }
1656
1657         /*
1658          * Ignoring of pids is done at task switch. But we have to
1659          * check for those tasks that are currently running.
1660          * Always do this in case a pid was appended or removed.
1661          */
1662         on_each_cpu(ignore_task_cpu, tr, 1);
1663
1664  out:
1665         mutex_unlock(&event_mutex);
1666
1667         if (ret > 0)
1668                 *ppos += ret;
1669
1670         return ret;
1671 }
1672
1673 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1674 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1675 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1676 static int ftrace_event_release(struct inode *inode, struct file *file);
1677
1678 static const struct seq_operations show_event_seq_ops = {
1679         .start = t_start,
1680         .next = t_next,
1681         .show = t_show,
1682         .stop = t_stop,
1683 };
1684
1685 static const struct seq_operations show_set_event_seq_ops = {
1686         .start = s_start,
1687         .next = s_next,
1688         .show = t_show,
1689         .stop = t_stop,
1690 };
1691
1692 static const struct seq_operations show_set_pid_seq_ops = {
1693         .start = p_start,
1694         .next = p_next,
1695         .show = trace_pid_show,
1696         .stop = p_stop,
1697 };
1698
1699 static const struct file_operations ftrace_avail_fops = {
1700         .open = ftrace_event_avail_open,
1701         .read = seq_read,
1702         .llseek = seq_lseek,
1703         .release = seq_release,
1704 };
1705
1706 static const struct file_operations ftrace_set_event_fops = {
1707         .open = ftrace_event_set_open,
1708         .read = seq_read,
1709         .write = ftrace_event_write,
1710         .llseek = seq_lseek,
1711         .release = ftrace_event_release,
1712 };
1713
1714 static const struct file_operations ftrace_set_event_pid_fops = {
1715         .open = ftrace_event_set_pid_open,
1716         .read = seq_read,
1717         .write = ftrace_event_pid_write,
1718         .llseek = seq_lseek,
1719         .release = ftrace_event_release,
1720 };
1721
1722 static const struct file_operations ftrace_enable_fops = {
1723         .open = tracing_open_generic,
1724         .read = event_enable_read,
1725         .write = event_enable_write,
1726         .llseek = default_llseek,
1727 };
1728
1729 static const struct file_operations ftrace_event_format_fops = {
1730         .open = trace_format_open,
1731         .read = seq_read,
1732         .llseek = seq_lseek,
1733         .release = seq_release,
1734 };
1735
1736 static const struct file_operations ftrace_event_id_fops = {
1737         .read = event_id_read,
1738         .llseek = default_llseek,
1739 };
1740
1741 static const struct file_operations ftrace_event_filter_fops = {
1742         .open = tracing_open_generic,
1743         .read = event_filter_read,
1744         .write = event_filter_write,
1745         .llseek = default_llseek,
1746 };
1747
1748 static const struct file_operations ftrace_subsystem_filter_fops = {
1749         .open = subsystem_open,
1750         .read = subsystem_filter_read,
1751         .write = subsystem_filter_write,
1752         .llseek = default_llseek,
1753         .release = subsystem_release,
1754 };
1755
1756 static const struct file_operations ftrace_system_enable_fops = {
1757         .open = subsystem_open,
1758         .read = system_enable_read,
1759         .write = system_enable_write,
1760         .llseek = default_llseek,
1761         .release = subsystem_release,
1762 };
1763
1764 static const struct file_operations ftrace_tr_enable_fops = {
1765         .open = system_tr_open,
1766         .read = system_enable_read,
1767         .write = system_enable_write,
1768         .llseek = default_llseek,
1769         .release = subsystem_release,
1770 };
1771
1772 static const struct file_operations ftrace_show_header_fops = {
1773         .open = tracing_open_generic,
1774         .read = show_header,
1775         .llseek = default_llseek,
1776 };
1777
1778 static int
1779 ftrace_event_open(struct inode *inode, struct file *file,
1780                   const struct seq_operations *seq_ops)
1781 {
1782         struct seq_file *m;
1783         int ret;
1784
1785         ret = seq_open(file, seq_ops);
1786         if (ret < 0)
1787                 return ret;
1788         m = file->private_data;
1789         /* copy tr over to seq ops */
1790         m->private = inode->i_private;
1791
1792         return ret;
1793 }
1794
1795 static int ftrace_event_release(struct inode *inode, struct file *file)
1796 {
1797         struct trace_array *tr = inode->i_private;
1798
1799         trace_array_put(tr);
1800
1801         return seq_release(inode, file);
1802 }
1803
1804 static int
1805 ftrace_event_avail_open(struct inode *inode, struct file *file)
1806 {
1807         const struct seq_operations *seq_ops = &show_event_seq_ops;
1808
1809         return ftrace_event_open(inode, file, seq_ops);
1810 }
1811
1812 static int
1813 ftrace_event_set_open(struct inode *inode, struct file *file)
1814 {
1815         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1816         struct trace_array *tr = inode->i_private;
1817         int ret;
1818
1819         if (trace_array_get(tr) < 0)
1820                 return -ENODEV;
1821
1822         if ((file->f_mode & FMODE_WRITE) &&
1823             (file->f_flags & O_TRUNC))
1824                 ftrace_clear_events(tr);
1825
1826         ret = ftrace_event_open(inode, file, seq_ops);
1827         if (ret < 0)
1828                 trace_array_put(tr);
1829         return ret;
1830 }
1831
1832 static int
1833 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1834 {
1835         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1836         struct trace_array *tr = inode->i_private;
1837         int ret;
1838
1839         if (trace_array_get(tr) < 0)
1840                 return -ENODEV;
1841
1842         if ((file->f_mode & FMODE_WRITE) &&
1843             (file->f_flags & O_TRUNC))
1844                 ftrace_clear_event_pids(tr);
1845
1846         ret = ftrace_event_open(inode, file, seq_ops);
1847         if (ret < 0)
1848                 trace_array_put(tr);
1849         return ret;
1850 }
1851
1852 static struct event_subsystem *
1853 create_new_subsystem(const char *name)
1854 {
1855         struct event_subsystem *system;
1856
1857         /* need to create new entry */
1858         system = kmalloc(sizeof(*system), GFP_KERNEL);
1859         if (!system)
1860                 return NULL;
1861
1862         system->ref_count = 1;
1863
1864         /* Only allocate if dynamic (kprobes and modules) */
1865         system->name = kstrdup_const(name, GFP_KERNEL);
1866         if (!system->name)
1867                 goto out_free;
1868
1869         system->filter = NULL;
1870
1871         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1872         if (!system->filter)
1873                 goto out_free;
1874
1875         list_add(&system->list, &event_subsystems);
1876
1877         return system;
1878
1879  out_free:
1880         kfree_const(system->name);
1881         kfree(system);
1882         return NULL;
1883 }
1884
1885 static struct dentry *
1886 event_subsystem_dir(struct trace_array *tr, const char *name,
1887                     struct trace_event_file *file, struct dentry *parent)
1888 {
1889         struct trace_subsystem_dir *dir;
1890         struct event_subsystem *system;
1891         struct dentry *entry;
1892
1893         /* First see if we did not already create this dir */
1894         list_for_each_entry(dir, &tr->systems, list) {
1895                 system = dir->subsystem;
1896                 if (strcmp(system->name, name) == 0) {
1897                         dir->nr_events++;
1898                         file->system = dir;
1899                         return dir->entry;
1900                 }
1901         }
1902
1903         /* Now see if the system itself exists. */
1904         list_for_each_entry(system, &event_subsystems, list) {
1905                 if (strcmp(system->name, name) == 0)
1906                         break;
1907         }
1908         /* Reset system variable when not found */
1909         if (&system->list == &event_subsystems)
1910                 system = NULL;
1911
1912         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1913         if (!dir)
1914                 goto out_fail;
1915
1916         if (!system) {
1917                 system = create_new_subsystem(name);
1918                 if (!system)
1919                         goto out_free;
1920         } else
1921                 __get_system(system);
1922
1923         dir->entry = tracefs_create_dir(name, parent);
1924         if (!dir->entry) {
1925                 pr_warn("Failed to create system directory %s\n", name);
1926                 __put_system(system);
1927                 goto out_free;
1928         }
1929
1930         dir->tr = tr;
1931         dir->ref_count = 1;
1932         dir->nr_events = 1;
1933         dir->subsystem = system;
1934         file->system = dir;
1935
1936         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1937                                     &ftrace_subsystem_filter_fops);
1938         if (!entry) {
1939                 kfree(system->filter);
1940                 system->filter = NULL;
1941                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1942         }
1943
1944         trace_create_file("enable", 0644, dir->entry, dir,
1945                           &ftrace_system_enable_fops);
1946
1947         list_add(&dir->list, &tr->systems);
1948
1949         return dir->entry;
1950
1951  out_free:
1952         kfree(dir);
1953  out_fail:
1954         /* Only print this message if failed on memory allocation */
1955         if (!dir || !system)
1956                 pr_warn("No memory to create event subsystem %s\n", name);
1957         return NULL;
1958 }
1959
1960 static int
1961 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1962 {
1963         struct trace_event_call *call = file->event_call;
1964         struct trace_array *tr = file->tr;
1965         struct list_head *head;
1966         struct dentry *d_events;
1967         const char *name;
1968         int ret;
1969
1970         /*
1971          * If the trace point header did not define TRACE_SYSTEM
1972          * then the system would be called "TRACE_SYSTEM".
1973          */
1974         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1975                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1976                 if (!d_events)
1977                         return -ENOMEM;
1978         } else
1979                 d_events = parent;
1980
1981         name = trace_event_name(call);
1982         file->dir = tracefs_create_dir(name, d_events);
1983         if (!file->dir) {
1984                 pr_warn("Could not create tracefs '%s' directory\n", name);
1985                 return -1;
1986         }
1987
1988         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1989                 trace_create_file("enable", 0644, file->dir, file,
1990                                   &ftrace_enable_fops);
1991
1992 #ifdef CONFIG_PERF_EVENTS
1993         if (call->event.type && call->class->reg)
1994                 trace_create_file("id", 0444, file->dir,
1995                                   (void *)(long)call->event.type,
1996                                   &ftrace_event_id_fops);
1997 #endif
1998
1999         /*
2000          * Other events may have the same class. Only update
2001          * the fields if they are not already defined.
2002          */
2003         head = trace_get_fields(call);
2004         if (list_empty(head)) {
2005                 ret = call->class->define_fields(call);
2006                 if (ret < 0) {
2007                         pr_warn("Could not initialize trace point events/%s\n",
2008                                 name);
2009                         return -1;
2010                 }
2011         }
2012         trace_create_file("filter", 0644, file->dir, file,
2013                           &ftrace_event_filter_fops);
2014
2015         /*
2016          * Only event directories that can be enabled should have
2017          * triggers.
2018          */
2019         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2020                 trace_create_file("trigger", 0644, file->dir, file,
2021                                   &event_trigger_fops);
2022
2023 #ifdef CONFIG_HIST_TRIGGERS
2024         trace_create_file("hist", 0444, file->dir, file,
2025                           &event_hist_fops);
2026 #endif
2027         trace_create_file("format", 0444, file->dir, call,
2028                           &ftrace_event_format_fops);
2029
2030         return 0;
2031 }
2032
2033 static void remove_event_from_tracers(struct trace_event_call *call)
2034 {
2035         struct trace_event_file *file;
2036         struct trace_array *tr;
2037
2038         do_for_each_event_file_safe(tr, file) {
2039                 if (file->event_call != call)
2040                         continue;
2041
2042                 remove_event_file_dir(file);
2043                 /*
2044                  * The do_for_each_event_file_safe() is
2045                  * a double loop. After finding the call for this
2046                  * trace_array, we use break to jump to the next
2047                  * trace_array.
2048                  */
2049                 break;
2050         } while_for_each_event_file();
2051 }
2052
2053 static void event_remove(struct trace_event_call *call)
2054 {
2055         struct trace_array *tr;
2056         struct trace_event_file *file;
2057
2058         do_for_each_event_file(tr, file) {
2059                 if (file->event_call != call)
2060                         continue;
2061                 ftrace_event_enable_disable(file, 0);
2062                 /*
2063                  * The do_for_each_event_file() is
2064                  * a double loop. After finding the call for this
2065                  * trace_array, we use break to jump to the next
2066                  * trace_array.
2067                  */
2068                 break;
2069         } while_for_each_event_file();
2070
2071         if (call->event.funcs)
2072                 __unregister_trace_event(&call->event);
2073         remove_event_from_tracers(call);
2074         list_del(&call->list);
2075 }
2076
2077 static int event_init(struct trace_event_call *call)
2078 {
2079         int ret = 0;
2080         const char *name;
2081
2082         name = trace_event_name(call);
2083         if (WARN_ON(!name))
2084                 return -EINVAL;
2085
2086         if (call->class->raw_init) {
2087                 ret = call->class->raw_init(call);
2088                 if (ret < 0 && ret != -ENOSYS)
2089                         pr_warn("Could not initialize trace events/%s\n", name);
2090         }
2091
2092         return ret;
2093 }
2094
2095 static int
2096 __register_event(struct trace_event_call *call, struct module *mod)
2097 {
2098         int ret;
2099
2100         ret = event_init(call);
2101         if (ret < 0)
2102                 return ret;
2103
2104         list_add(&call->list, &ftrace_events);
2105         call->mod = mod;
2106
2107         return 0;
2108 }
2109
2110 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2111 {
2112         int rlen;
2113         int elen;
2114
2115         /* Find the length of the eval value as a string */
2116         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2117         /* Make sure there's enough room to replace the string with the value */
2118         if (len < elen)
2119                 return NULL;
2120
2121         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2122
2123         /* Get the rest of the string of ptr */
2124         rlen = strlen(ptr + len);
2125         memmove(ptr + elen, ptr + len, rlen);
2126         /* Make sure we end the new string */
2127         ptr[elen + rlen] = 0;
2128
2129         return ptr + elen;
2130 }
2131
2132 static void update_event_printk(struct trace_event_call *call,
2133                                 struct trace_eval_map *map)
2134 {
2135         char *ptr;
2136         int quote = 0;
2137         int len = strlen(map->eval_string);
2138
2139         for (ptr = call->print_fmt; *ptr; ptr++) {
2140                 if (*ptr == '\\') {
2141                         ptr++;
2142                         /* paranoid */
2143                         if (!*ptr)
2144                                 break;
2145                         continue;
2146                 }
2147                 if (*ptr == '"') {
2148                         quote ^= 1;
2149                         continue;
2150                 }
2151                 if (quote)
2152                         continue;
2153                 if (isdigit(*ptr)) {
2154                         /* skip numbers */
2155                         do {
2156                                 ptr++;
2157                                 /* Check for alpha chars like ULL */
2158                         } while (isalnum(*ptr));
2159                         if (!*ptr)
2160                                 break;
2161                         /*
2162                          * A number must have some kind of delimiter after
2163                          * it, and we can ignore that too.
2164                          */
2165                         continue;
2166                 }
2167                 if (isalpha(*ptr) || *ptr == '_') {
2168                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2169                             !isalnum(ptr[len]) && ptr[len] != '_') {
2170                                 ptr = eval_replace(ptr, map, len);
2171                                 /* enum/sizeof string smaller than value */
2172                                 if (WARN_ON_ONCE(!ptr))
2173                                         return;
2174                                 /*
2175                                  * No need to decrement here, as eval_replace()
2176                                  * returns the pointer to the character passed
2177                                  * the eval, and two evals can not be placed
2178                                  * back to back without something in between.
2179                                  * We can skip that something in between.
2180                                  */
2181                                 continue;
2182                         }
2183                 skip_more:
2184                         do {
2185                                 ptr++;
2186                         } while (isalnum(*ptr) || *ptr == '_');
2187                         if (!*ptr)
2188                                 break;
2189                         /*
2190                          * If what comes after this variable is a '.' or
2191                          * '->' then we can continue to ignore that string.
2192                          */
2193                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2194                                 ptr += *ptr == '.' ? 1 : 2;
2195                                 if (!*ptr)
2196                                         break;
2197                                 goto skip_more;
2198                         }
2199                         /*
2200                          * Once again, we can skip the delimiter that came
2201                          * after the string.
2202                          */
2203                         continue;
2204                 }
2205         }
2206 }
2207
2208 void trace_event_eval_update(struct trace_eval_map **map, int len)
2209 {
2210         struct trace_event_call *call, *p;
2211         const char *last_system = NULL;
2212         int last_i;
2213         int i;
2214
2215         down_write(&trace_event_sem);
2216         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2217                 /* events are usually grouped together with systems */
2218                 if (!last_system || call->class->system != last_system) {
2219                         last_i = 0;
2220                         last_system = call->class->system;
2221                 }
2222
2223                 for (i = last_i; i < len; i++) {
2224                         if (call->class->system == map[i]->system) {
2225                                 /* Save the first system if need be */
2226                                 if (!last_i)
2227                                         last_i = i;
2228                                 update_event_printk(call, map[i]);
2229                         }
2230                 }
2231         }
2232         up_write(&trace_event_sem);
2233 }
2234
2235 static struct trace_event_file *
2236 trace_create_new_event(struct trace_event_call *call,
2237                        struct trace_array *tr)
2238 {
2239         struct trace_event_file *file;
2240
2241         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2242         if (!file)
2243                 return NULL;
2244
2245         file->event_call = call;
2246         file->tr = tr;
2247         atomic_set(&file->sm_ref, 0);
2248         atomic_set(&file->tm_ref, 0);
2249         INIT_LIST_HEAD(&file->triggers);
2250         list_add(&file->list, &tr->events);
2251
2252         return file;
2253 }
2254
2255 /* Add an event to a trace directory */
2256 static int
2257 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2258 {
2259         struct trace_event_file *file;
2260
2261         file = trace_create_new_event(call, tr);
2262         if (!file)
2263                 return -ENOMEM;
2264
2265         return event_create_dir(tr->event_dir, file);
2266 }
2267
2268 /*
2269  * Just create a decriptor for early init. A descriptor is required
2270  * for enabling events at boot. We want to enable events before
2271  * the filesystem is initialized.
2272  */
2273 static __init int
2274 __trace_early_add_new_event(struct trace_event_call *call,
2275                             struct trace_array *tr)
2276 {
2277         struct trace_event_file *file;
2278
2279         file = trace_create_new_event(call, tr);
2280         if (!file)
2281                 return -ENOMEM;
2282
2283         return 0;
2284 }
2285
2286 struct ftrace_module_file_ops;
2287 static void __add_event_to_tracers(struct trace_event_call *call);
2288
2289 /* Add an additional event_call dynamically */
2290 int trace_add_event_call(struct trace_event_call *call)
2291 {
2292         int ret;
2293         mutex_lock(&trace_types_lock);
2294         mutex_lock(&event_mutex);
2295
2296         ret = __register_event(call, NULL);
2297         if (ret >= 0)
2298                 __add_event_to_tracers(call);
2299
2300         mutex_unlock(&event_mutex);
2301         mutex_unlock(&trace_types_lock);
2302         return ret;
2303 }
2304
2305 /*
2306  * Must be called under locking of trace_types_lock, event_mutex and
2307  * trace_event_sem.
2308  */
2309 static void __trace_remove_event_call(struct trace_event_call *call)
2310 {
2311         event_remove(call);
2312         trace_destroy_fields(call);
2313         free_event_filter(call->filter);
2314         call->filter = NULL;
2315 }
2316
2317 static int probe_remove_event_call(struct trace_event_call *call)
2318 {
2319         struct trace_array *tr;
2320         struct trace_event_file *file;
2321
2322 #ifdef CONFIG_PERF_EVENTS
2323         if (call->perf_refcount)
2324                 return -EBUSY;
2325 #endif
2326         do_for_each_event_file(tr, file) {
2327                 if (file->event_call != call)
2328                         continue;
2329                 /*
2330                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2331                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2332                  * TRACE_REG_UNREGISTER.
2333                  */
2334                 if (file->flags & EVENT_FILE_FL_ENABLED)
2335                         return -EBUSY;
2336                 /*
2337                  * The do_for_each_event_file_safe() is
2338                  * a double loop. After finding the call for this
2339                  * trace_array, we use break to jump to the next
2340                  * trace_array.
2341                  */
2342                 break;
2343         } while_for_each_event_file();
2344
2345         __trace_remove_event_call(call);
2346
2347         return 0;
2348 }
2349
2350 /* Remove an event_call */
2351 int trace_remove_event_call(struct trace_event_call *call)
2352 {
2353         int ret;
2354
2355         mutex_lock(&trace_types_lock);
2356         mutex_lock(&event_mutex);
2357         down_write(&trace_event_sem);
2358         ret = probe_remove_event_call(call);
2359         up_write(&trace_event_sem);
2360         mutex_unlock(&event_mutex);
2361         mutex_unlock(&trace_types_lock);
2362
2363         return ret;
2364 }
2365
2366 #define for_each_event(event, start, end)                       \
2367         for (event = start;                                     \
2368              (unsigned long)event < (unsigned long)end;         \
2369              event++)
2370
2371 #ifdef CONFIG_MODULES
2372
2373 static void trace_module_add_events(struct module *mod)
2374 {
2375         struct trace_event_call **call, **start, **end;
2376
2377         if (!mod->num_trace_events)
2378                 return;
2379
2380         /* Don't add infrastructure for mods without tracepoints */
2381         if (trace_module_has_bad_taint(mod)) {
2382                 pr_err("%s: module has bad taint, not creating trace events\n",
2383                        mod->name);
2384                 return;
2385         }
2386
2387         start = mod->trace_events;
2388         end = mod->trace_events + mod->num_trace_events;
2389
2390         for_each_event(call, start, end) {
2391                 __register_event(*call, mod);
2392                 __add_event_to_tracers(*call);
2393         }
2394 }
2395
2396 static void trace_module_remove_events(struct module *mod)
2397 {
2398         struct trace_event_call *call, *p;
2399         bool clear_trace = false;
2400
2401         down_write(&trace_event_sem);
2402         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2403                 if (call->mod == mod) {
2404                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
2405                                 clear_trace = true;
2406                         __trace_remove_event_call(call);
2407                 }
2408         }
2409         up_write(&trace_event_sem);
2410
2411         /*
2412          * It is safest to reset the ring buffer if the module being unloaded
2413          * registered any events that were used. The only worry is if
2414          * a new module gets loaded, and takes on the same id as the events
2415          * of this module. When printing out the buffer, traced events left
2416          * over from this module may be passed to the new module events and
2417          * unexpected results may occur.
2418          */
2419         if (clear_trace)
2420                 tracing_reset_all_online_cpus();
2421 }
2422
2423 static int trace_module_notify(struct notifier_block *self,
2424                                unsigned long val, void *data)
2425 {
2426         struct module *mod = data;
2427
2428         mutex_lock(&trace_types_lock);
2429         mutex_lock(&event_mutex);
2430         switch (val) {
2431         case MODULE_STATE_COMING:
2432                 trace_module_add_events(mod);
2433                 break;
2434         case MODULE_STATE_GOING:
2435                 trace_module_remove_events(mod);
2436                 break;
2437         }
2438         mutex_unlock(&event_mutex);
2439         mutex_unlock(&trace_types_lock);
2440
2441         return 0;
2442 }
2443
2444 static struct notifier_block trace_module_nb = {
2445         .notifier_call = trace_module_notify,
2446         .priority = 1, /* higher than trace.c module notify */
2447 };
2448 #endif /* CONFIG_MODULES */
2449
2450 /* Create a new event directory structure for a trace directory. */
2451 static void
2452 __trace_add_event_dirs(struct trace_array *tr)
2453 {
2454         struct trace_event_call *call;
2455         int ret;
2456
2457         list_for_each_entry(call, &ftrace_events, list) {
2458                 ret = __trace_add_new_event(call, tr);
2459                 if (ret < 0)
2460                         pr_warn("Could not create directory for event %s\n",
2461                                 trace_event_name(call));
2462         }
2463 }
2464
2465 struct trace_event_file *
2466 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2467 {
2468         struct trace_event_file *file;
2469         struct trace_event_call *call;
2470         const char *name;
2471
2472         list_for_each_entry(file, &tr->events, list) {
2473
2474                 call = file->event_call;
2475                 name = trace_event_name(call);
2476
2477                 if (!name || !call->class || !call->class->reg)
2478                         continue;
2479
2480                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2481                         continue;
2482
2483                 if (strcmp(event, name) == 0 &&
2484                     strcmp(system, call->class->system) == 0)
2485                         return file;
2486         }
2487         return NULL;
2488 }
2489
2490 #ifdef CONFIG_DYNAMIC_FTRACE
2491
2492 /* Avoid typos */
2493 #define ENABLE_EVENT_STR        "enable_event"
2494 #define DISABLE_EVENT_STR       "disable_event"
2495
2496 struct event_probe_data {
2497         struct trace_event_file *file;
2498         unsigned long                   count;
2499         int                             ref;
2500         bool                            enable;
2501 };
2502
2503 static void update_event_probe(struct event_probe_data *data)
2504 {
2505         if (data->enable)
2506                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2507         else
2508                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2509 }
2510
2511 static void
2512 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2513                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2514                    void *data)
2515 {
2516         struct ftrace_func_mapper *mapper = data;
2517         struct event_probe_data *edata;
2518         void **pdata;
2519
2520         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2521         if (!pdata || !*pdata)
2522                 return;
2523
2524         edata = *pdata;
2525         update_event_probe(edata);
2526 }
2527
2528 static void
2529 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2530                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2531                          void *data)
2532 {
2533         struct ftrace_func_mapper *mapper = data;
2534         struct event_probe_data *edata;
2535         void **pdata;
2536
2537         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2538         if (!pdata || !*pdata)
2539                 return;
2540
2541         edata = *pdata;
2542
2543         if (!edata->count)
2544                 return;
2545
2546         /* Skip if the event is in a state we want to switch to */
2547         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2548                 return;
2549
2550         if (edata->count != -1)
2551                 (edata->count)--;
2552
2553         update_event_probe(edata);
2554 }
2555
2556 static int
2557 event_enable_print(struct seq_file *m, unsigned long ip,
2558                    struct ftrace_probe_ops *ops, void *data)
2559 {
2560         struct ftrace_func_mapper *mapper = data;
2561         struct event_probe_data *edata;
2562         void **pdata;
2563
2564         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2565
2566         if (WARN_ON_ONCE(!pdata || !*pdata))
2567                 return 0;
2568
2569         edata = *pdata;
2570
2571         seq_printf(m, "%ps:", (void *)ip);
2572
2573         seq_printf(m, "%s:%s:%s",
2574                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2575                    edata->file->event_call->class->system,
2576                    trace_event_name(edata->file->event_call));
2577
2578         if (edata->count == -1)
2579                 seq_puts(m, ":unlimited\n");
2580         else
2581                 seq_printf(m, ":count=%ld\n", edata->count);
2582
2583         return 0;
2584 }
2585
2586 static int
2587 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2588                   unsigned long ip, void *init_data, void **data)
2589 {
2590         struct ftrace_func_mapper *mapper = *data;
2591         struct event_probe_data *edata = init_data;
2592         int ret;
2593
2594         if (!mapper) {
2595                 mapper = allocate_ftrace_func_mapper();
2596                 if (!mapper)
2597                         return -ENODEV;
2598                 *data = mapper;
2599         }
2600
2601         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2602         if (ret < 0)
2603                 return ret;
2604
2605         edata->ref++;
2606
2607         return 0;
2608 }
2609
2610 static int free_probe_data(void *data)
2611 {
2612         struct event_probe_data *edata = data;
2613
2614         edata->ref--;
2615         if (!edata->ref) {
2616                 /* Remove the SOFT_MODE flag */
2617                 __ftrace_event_enable_disable(edata->file, 0, 1);
2618                 module_put(edata->file->event_call->mod);
2619                 kfree(edata);
2620         }
2621         return 0;
2622 }
2623
2624 static void
2625 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2626                   unsigned long ip, void *data)
2627 {
2628         struct ftrace_func_mapper *mapper = data;
2629         struct event_probe_data *edata;
2630
2631         if (!ip) {
2632                 if (!mapper)
2633                         return;
2634                 free_ftrace_func_mapper(mapper, free_probe_data);
2635                 return;
2636         }
2637
2638         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2639
2640         if (WARN_ON_ONCE(!edata))
2641                 return;
2642
2643         if (WARN_ON_ONCE(edata->ref <= 0))
2644                 return;
2645
2646         free_probe_data(edata);
2647 }
2648
2649 static struct ftrace_probe_ops event_enable_probe_ops = {
2650         .func                   = event_enable_probe,
2651         .print                  = event_enable_print,
2652         .init                   = event_enable_init,
2653         .free                   = event_enable_free,
2654 };
2655
2656 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2657         .func                   = event_enable_count_probe,
2658         .print                  = event_enable_print,
2659         .init                   = event_enable_init,
2660         .free                   = event_enable_free,
2661 };
2662
2663 static struct ftrace_probe_ops event_disable_probe_ops = {
2664         .func                   = event_enable_probe,
2665         .print                  = event_enable_print,
2666         .init                   = event_enable_init,
2667         .free                   = event_enable_free,
2668 };
2669
2670 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2671         .func                   = event_enable_count_probe,
2672         .print                  = event_enable_print,
2673         .init                   = event_enable_init,
2674         .free                   = event_enable_free,
2675 };
2676
2677 static int
2678 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2679                   char *glob, char *cmd, char *param, int enabled)
2680 {
2681         struct trace_event_file *file;
2682         struct ftrace_probe_ops *ops;
2683         struct event_probe_data *data;
2684         const char *system;
2685         const char *event;
2686         char *number;
2687         bool enable;
2688         int ret;
2689
2690         if (!tr)
2691                 return -ENODEV;
2692
2693         /* hash funcs only work with set_ftrace_filter */
2694         if (!enabled || !param)
2695                 return -EINVAL;
2696
2697         system = strsep(&param, ":");
2698         if (!param)
2699                 return -EINVAL;
2700
2701         event = strsep(&param, ":");
2702
2703         mutex_lock(&event_mutex);
2704
2705         ret = -EINVAL;
2706         file = find_event_file(tr, system, event);
2707         if (!file)
2708                 goto out;
2709
2710         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2711
2712         if (enable)
2713                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2714         else
2715                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2716
2717         if (glob[0] == '!') {
2718                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2719                 goto out;
2720         }
2721
2722         ret = -ENOMEM;
2723
2724         data = kzalloc(sizeof(*data), GFP_KERNEL);
2725         if (!data)
2726                 goto out;
2727
2728         data->enable = enable;
2729         data->count = -1;
2730         data->file = file;
2731
2732         if (!param)
2733                 goto out_reg;
2734
2735         number = strsep(&param, ":");
2736
2737         ret = -EINVAL;
2738         if (!strlen(number))
2739                 goto out_free;
2740
2741         /*
2742          * We use the callback data field (which is a pointer)
2743          * as our counter.
2744          */
2745         ret = kstrtoul(number, 0, &data->count);
2746         if (ret)
2747                 goto out_free;
2748
2749  out_reg:
2750         /* Don't let event modules unload while probe registered */
2751         ret = try_module_get(file->event_call->mod);
2752         if (!ret) {
2753                 ret = -EBUSY;
2754                 goto out_free;
2755         }
2756
2757         ret = __ftrace_event_enable_disable(file, 1, 1);
2758         if (ret < 0)
2759                 goto out_put;
2760
2761         ret = register_ftrace_function_probe(glob, tr, ops, data);
2762         /*
2763          * The above returns on success the # of functions enabled,
2764          * but if it didn't find any functions it returns zero.
2765          * Consider no functions a failure too.
2766          */
2767         if (!ret) {
2768                 ret = -ENOENT;
2769                 goto out_disable;
2770         } else if (ret < 0)
2771                 goto out_disable;
2772         /* Just return zero, not the number of enabled functions */
2773         ret = 0;
2774  out:
2775         mutex_unlock(&event_mutex);
2776         return ret;
2777
2778  out_disable:
2779         __ftrace_event_enable_disable(file, 0, 1);
2780  out_put:
2781         module_put(file->event_call->mod);
2782  out_free:
2783         kfree(data);
2784         goto out;
2785 }
2786
2787 static struct ftrace_func_command event_enable_cmd = {
2788         .name                   = ENABLE_EVENT_STR,
2789         .func                   = event_enable_func,
2790 };
2791
2792 static struct ftrace_func_command event_disable_cmd = {
2793         .name                   = DISABLE_EVENT_STR,
2794         .func                   = event_enable_func,
2795 };
2796
2797 static __init int register_event_cmds(void)
2798 {
2799         int ret;
2800
2801         ret = register_ftrace_command(&event_enable_cmd);
2802         if (WARN_ON(ret < 0))
2803                 return ret;
2804         ret = register_ftrace_command(&event_disable_cmd);
2805         if (WARN_ON(ret < 0))
2806                 unregister_ftrace_command(&event_enable_cmd);
2807         return ret;
2808 }
2809 #else
2810 static inline int register_event_cmds(void) { return 0; }
2811 #endif /* CONFIG_DYNAMIC_FTRACE */
2812
2813 /*
2814  * The top level array has already had its trace_event_file
2815  * descriptors created in order to allow for early events to
2816  * be recorded. This function is called after the tracefs has been
2817  * initialized, and we now have to create the files associated
2818  * to the events.
2819  */
2820 static __init void
2821 __trace_early_add_event_dirs(struct trace_array *tr)
2822 {
2823         struct trace_event_file *file;
2824         int ret;
2825
2826
2827         list_for_each_entry(file, &tr->events, list) {
2828                 ret = event_create_dir(tr->event_dir, file);
2829                 if (ret < 0)
2830                         pr_warn("Could not create directory for event %s\n",
2831                                 trace_event_name(file->event_call));
2832         }
2833 }
2834
2835 /*
2836  * For early boot up, the top trace array requires to have
2837  * a list of events that can be enabled. This must be done before
2838  * the filesystem is set up in order to allow events to be traced
2839  * early.
2840  */
2841 static __init void
2842 __trace_early_add_events(struct trace_array *tr)
2843 {
2844         struct trace_event_call *call;
2845         int ret;
2846
2847         list_for_each_entry(call, &ftrace_events, list) {
2848                 /* Early boot up should not have any modules loaded */
2849                 if (WARN_ON_ONCE(call->mod))
2850                         continue;
2851
2852                 ret = __trace_early_add_new_event(call, tr);
2853                 if (ret < 0)
2854                         pr_warn("Could not create early event %s\n",
2855                                 trace_event_name(call));
2856         }
2857 }
2858
2859 /* Remove the event directory structure for a trace directory. */
2860 static void
2861 __trace_remove_event_dirs(struct trace_array *tr)
2862 {
2863         struct trace_event_file *file, *next;
2864
2865         list_for_each_entry_safe(file, next, &tr->events, list)
2866                 remove_event_file_dir(file);
2867 }
2868
2869 static void __add_event_to_tracers(struct trace_event_call *call)
2870 {
2871         struct trace_array *tr;
2872
2873         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2874                 __trace_add_new_event(call, tr);
2875 }
2876
2877 extern struct trace_event_call *__start_ftrace_events[];
2878 extern struct trace_event_call *__stop_ftrace_events[];
2879
2880 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2881
2882 static __init int setup_trace_event(char *str)
2883 {
2884         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2885         ring_buffer_expanded = true;
2886         tracing_selftest_disabled = true;
2887
2888         return 1;
2889 }
2890 __setup("trace_event=", setup_trace_event);
2891
2892 /* Expects to have event_mutex held when called */
2893 static int
2894 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2895 {
2896         struct dentry *d_events;
2897         struct dentry *entry;
2898
2899         entry = tracefs_create_file("set_event", 0644, parent,
2900                                     tr, &ftrace_set_event_fops);
2901         if (!entry) {
2902                 pr_warn("Could not create tracefs 'set_event' entry\n");
2903                 return -ENOMEM;
2904         }
2905
2906         d_events = tracefs_create_dir("events", parent);
2907         if (!d_events) {
2908                 pr_warn("Could not create tracefs 'events' directory\n");
2909                 return -ENOMEM;
2910         }
2911
2912         entry = trace_create_file("enable", 0644, d_events,
2913                                   tr, &ftrace_tr_enable_fops);
2914         if (!entry) {
2915                 pr_warn("Could not create tracefs 'enable' entry\n");
2916                 return -ENOMEM;
2917         }
2918
2919         /* There are not as crucial, just warn if they are not created */
2920
2921         entry = tracefs_create_file("set_event_pid", 0644, parent,
2922                                     tr, &ftrace_set_event_pid_fops);
2923         if (!entry)
2924                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2925
2926         /* ring buffer internal formats */
2927         entry = trace_create_file("header_page", 0444, d_events,
2928                                   ring_buffer_print_page_header,
2929                                   &ftrace_show_header_fops);
2930         if (!entry)
2931                 pr_warn("Could not create tracefs 'header_page' entry\n");
2932
2933         entry = trace_create_file("header_event", 0444, d_events,
2934                                   ring_buffer_print_entry_header,
2935                                   &ftrace_show_header_fops);
2936         if (!entry)
2937                 pr_warn("Could not create tracefs 'header_event' entry\n");
2938
2939         tr->event_dir = d_events;
2940
2941         return 0;
2942 }
2943
2944 /**
2945  * event_trace_add_tracer - add a instance of a trace_array to events
2946  * @parent: The parent dentry to place the files/directories for events in
2947  * @tr: The trace array associated with these events
2948  *
2949  * When a new instance is created, it needs to set up its events
2950  * directory, as well as other files associated with events. It also
2951  * creates the event hierachry in the @parent/events directory.
2952  *
2953  * Returns 0 on success.
2954  */
2955 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2956 {
2957         int ret;
2958
2959         mutex_lock(&event_mutex);
2960
2961         ret = create_event_toplevel_files(parent, tr);
2962         if (ret)
2963                 goto out_unlock;
2964
2965         down_write(&trace_event_sem);
2966         __trace_add_event_dirs(tr);
2967         up_write(&trace_event_sem);
2968
2969  out_unlock:
2970         mutex_unlock(&event_mutex);
2971
2972         return ret;
2973 }
2974
2975 /*
2976  * The top trace array already had its file descriptors created.
2977  * Now the files themselves need to be created.
2978  */
2979 static __init int
2980 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2981 {
2982         int ret;
2983
2984         mutex_lock(&event_mutex);
2985
2986         ret = create_event_toplevel_files(parent, tr);
2987         if (ret)
2988                 goto out_unlock;
2989
2990         down_write(&trace_event_sem);
2991         __trace_early_add_event_dirs(tr);
2992         up_write(&trace_event_sem);
2993
2994  out_unlock:
2995         mutex_unlock(&event_mutex);
2996
2997         return ret;
2998 }
2999
3000 int event_trace_del_tracer(struct trace_array *tr)
3001 {
3002         mutex_lock(&event_mutex);
3003
3004         /* Disable any event triggers and associated soft-disabled events */
3005         clear_event_triggers(tr);
3006
3007         /* Clear the pid list */
3008         __ftrace_clear_event_pids(tr);
3009
3010         /* Disable any running events */
3011         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3012
3013         /* Access to events are within rcu_read_lock_sched() */
3014         synchronize_sched();
3015
3016         down_write(&trace_event_sem);
3017         __trace_remove_event_dirs(tr);
3018         tracefs_remove_recursive(tr->event_dir);
3019         up_write(&trace_event_sem);
3020
3021         tr->event_dir = NULL;
3022
3023         mutex_unlock(&event_mutex);
3024
3025         return 0;
3026 }
3027
3028 static __init int event_trace_memsetup(void)
3029 {
3030         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3031         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3032         return 0;
3033 }
3034
3035 static __init void
3036 early_enable_events(struct trace_array *tr, bool disable_first)
3037 {
3038         char *buf = bootup_event_buf;
3039         char *token;
3040         int ret;
3041
3042         while (true) {
3043                 token = strsep(&buf, ",");
3044
3045                 if (!token)
3046                         break;
3047
3048                 if (*token) {
3049                         /* Restarting syscalls requires that we stop them first */
3050                         if (disable_first)
3051                                 ftrace_set_clr_event(tr, token, 0);
3052
3053                         ret = ftrace_set_clr_event(tr, token, 1);
3054                         if (ret)
3055                                 pr_warn("Failed to enable trace event: %s\n", token);
3056                 }
3057
3058                 /* Put back the comma to allow this to be called again */
3059                 if (buf)
3060                         *(buf - 1) = ',';
3061         }
3062 }
3063
3064 static __init int event_trace_enable(void)
3065 {
3066         struct trace_array *tr = top_trace_array();
3067         struct trace_event_call **iter, *call;
3068         int ret;
3069
3070         if (!tr)
3071                 return -ENODEV;
3072
3073         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3074
3075                 call = *iter;
3076                 ret = event_init(call);
3077                 if (!ret)
3078                         list_add(&call->list, &ftrace_events);
3079         }
3080
3081         /*
3082          * We need the top trace array to have a working set of trace
3083          * points at early init, before the debug files and directories
3084          * are created. Create the file entries now, and attach them
3085          * to the actual file dentries later.
3086          */
3087         __trace_early_add_events(tr);
3088
3089         early_enable_events(tr, false);
3090
3091         trace_printk_start_comm();
3092
3093         register_event_cmds();
3094
3095         register_trigger_cmds();
3096
3097         return 0;
3098 }
3099
3100 /*
3101  * event_trace_enable() is called from trace_event_init() first to
3102  * initialize events and perhaps start any events that are on the
3103  * command line. Unfortunately, there are some events that will not
3104  * start this early, like the system call tracepoints that need
3105  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3106  * is called before pid 1 starts, and this flag is never set, making
3107  * the syscall tracepoint never get reached, but the event is enabled
3108  * regardless (and not doing anything).
3109  */
3110 static __init int event_trace_enable_again(void)
3111 {
3112         struct trace_array *tr;
3113
3114         tr = top_trace_array();
3115         if (!tr)
3116                 return -ENODEV;
3117
3118         early_enable_events(tr, true);
3119
3120         return 0;
3121 }
3122
3123 early_initcall(event_trace_enable_again);
3124
3125 static __init int event_trace_init(void)
3126 {
3127         struct trace_array *tr;
3128         struct dentry *d_tracer;
3129         struct dentry *entry;
3130         int ret;
3131
3132         tr = top_trace_array();
3133         if (!tr)
3134                 return -ENODEV;
3135
3136         d_tracer = tracing_init_dentry();
3137         if (IS_ERR(d_tracer))
3138                 return 0;
3139
3140         entry = tracefs_create_file("available_events", 0444, d_tracer,
3141                                     tr, &ftrace_avail_fops);
3142         if (!entry)
3143                 pr_warn("Could not create tracefs 'available_events' entry\n");
3144
3145         if (trace_define_generic_fields())
3146                 pr_warn("tracing: Failed to allocated generic fields");
3147
3148         if (trace_define_common_fields())
3149                 pr_warn("tracing: Failed to allocate common fields");
3150
3151         ret = early_event_add_tracer(d_tracer, tr);
3152         if (ret)
3153                 return ret;
3154
3155 #ifdef CONFIG_MODULES
3156         ret = register_module_notifier(&trace_module_nb);
3157         if (ret)
3158                 pr_warn("Failed to register trace events module notifier\n");
3159 #endif
3160         return 0;
3161 }
3162
3163 void __init trace_event_init(void)
3164 {
3165         event_trace_memsetup();
3166         init_ftrace_syscalls();
3167         event_trace_enable();
3168 }
3169
3170 fs_initcall(event_trace_init);
3171
3172 #ifdef CONFIG_FTRACE_STARTUP_TEST
3173
3174 static DEFINE_SPINLOCK(test_spinlock);
3175 static DEFINE_SPINLOCK(test_spinlock_irq);
3176 static DEFINE_MUTEX(test_mutex);
3177
3178 static __init void test_work(struct work_struct *dummy)
3179 {
3180         spin_lock(&test_spinlock);
3181         spin_lock_irq(&test_spinlock_irq);
3182         udelay(1);
3183         spin_unlock_irq(&test_spinlock_irq);
3184         spin_unlock(&test_spinlock);
3185
3186         mutex_lock(&test_mutex);
3187         msleep(1);
3188         mutex_unlock(&test_mutex);
3189 }
3190
3191 static __init int event_test_thread(void *unused)
3192 {
3193         void *test_malloc;
3194
3195         test_malloc = kmalloc(1234, GFP_KERNEL);
3196         if (!test_malloc)
3197                 pr_info("failed to kmalloc\n");
3198
3199         schedule_on_each_cpu(test_work);
3200
3201         kfree(test_malloc);
3202
3203         set_current_state(TASK_INTERRUPTIBLE);
3204         while (!kthread_should_stop()) {
3205                 schedule();
3206                 set_current_state(TASK_INTERRUPTIBLE);
3207         }
3208         __set_current_state(TASK_RUNNING);
3209
3210         return 0;
3211 }
3212
3213 /*
3214  * Do various things that may trigger events.
3215  */
3216 static __init void event_test_stuff(void)
3217 {
3218         struct task_struct *test_thread;
3219
3220         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3221         msleep(1);
3222         kthread_stop(test_thread);
3223 }
3224
3225 /*
3226  * For every trace event defined, we will test each trace point separately,
3227  * and then by groups, and finally all trace points.
3228  */
3229 static __init void event_trace_self_tests(void)
3230 {
3231         struct trace_subsystem_dir *dir;
3232         struct trace_event_file *file;
3233         struct trace_event_call *call;
3234         struct event_subsystem *system;
3235         struct trace_array *tr;
3236         int ret;
3237
3238         tr = top_trace_array();
3239         if (!tr)
3240                 return;
3241
3242         pr_info("Running tests on trace events:\n");
3243
3244         list_for_each_entry(file, &tr->events, list) {
3245
3246                 call = file->event_call;
3247
3248                 /* Only test those that have a probe */
3249                 if (!call->class || !call->class->probe)
3250                         continue;
3251
3252 /*
3253  * Testing syscall events here is pretty useless, but
3254  * we still do it if configured. But this is time consuming.
3255  * What we really need is a user thread to perform the
3256  * syscalls as we test.
3257  */
3258 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3259                 if (call->class->system &&
3260                     strcmp(call->class->system, "syscalls") == 0)
3261                         continue;
3262 #endif
3263
3264                 pr_info("Testing event %s: ", trace_event_name(call));
3265
3266                 /*
3267                  * If an event is already enabled, someone is using
3268                  * it and the self test should not be on.
3269                  */
3270                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3271                         pr_warn("Enabled event during self test!\n");
3272                         WARN_ON_ONCE(1);
3273                         continue;
3274                 }
3275
3276                 ftrace_event_enable_disable(file, 1);
3277                 event_test_stuff();
3278                 ftrace_event_enable_disable(file, 0);
3279
3280                 pr_cont("OK\n");
3281         }
3282
3283         /* Now test at the sub system level */
3284
3285         pr_info("Running tests on trace event systems:\n");
3286
3287         list_for_each_entry(dir, &tr->systems, list) {
3288
3289                 system = dir->subsystem;
3290
3291                 /* the ftrace system is special, skip it */
3292                 if (strcmp(system->name, "ftrace") == 0)
3293                         continue;
3294
3295                 pr_info("Testing event system %s: ", system->name);
3296
3297                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3298                 if (WARN_ON_ONCE(ret)) {
3299                         pr_warn("error enabling system %s\n",
3300                                 system->name);
3301                         continue;
3302                 }
3303
3304                 event_test_stuff();
3305
3306                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3307                 if (WARN_ON_ONCE(ret)) {
3308                         pr_warn("error disabling system %s\n",
3309                                 system->name);
3310                         continue;
3311                 }
3312
3313                 pr_cont("OK\n");
3314         }
3315
3316         /* Test with all events enabled */
3317
3318         pr_info("Running tests on all trace events:\n");
3319         pr_info("Testing all events: ");
3320
3321         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3322         if (WARN_ON_ONCE(ret)) {
3323                 pr_warn("error enabling all events\n");
3324                 return;
3325         }
3326
3327         event_test_stuff();
3328
3329         /* reset sysname */
3330         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3331         if (WARN_ON_ONCE(ret)) {
3332                 pr_warn("error disabling all events\n");
3333                 return;
3334         }
3335
3336         pr_cont("OK\n");
3337 }
3338
3339 #ifdef CONFIG_FUNCTION_TRACER
3340
3341 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3342
3343 static struct trace_event_file event_trace_file __initdata;
3344
3345 static void __init
3346 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3347                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3348 {
3349         struct ring_buffer_event *event;
3350         struct ring_buffer *buffer;
3351         struct ftrace_entry *entry;
3352         unsigned long flags;
3353         long disabled;
3354         int cpu;
3355         int pc;
3356
3357         pc = preempt_count();
3358         preempt_disable_notrace();
3359         cpu = raw_smp_processor_id();
3360         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3361
3362         if (disabled != 1)
3363                 goto out;
3364
3365         local_save_flags(flags);
3366
3367         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3368                                                 TRACE_FN, sizeof(*entry),
3369                                                 flags, pc);
3370         if (!event)
3371                 goto out;
3372         entry   = ring_buffer_event_data(event);
3373         entry->ip                       = ip;
3374         entry->parent_ip                = parent_ip;
3375
3376         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3377                                     entry, flags, pc);
3378  out:
3379         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3380         preempt_enable_notrace();
3381 }
3382
3383 static struct ftrace_ops trace_ops __initdata  =
3384 {
3385         .func = function_test_events_call,
3386         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3387 };
3388
3389 static __init void event_trace_self_test_with_function(void)
3390 {
3391         int ret;
3392
3393         event_trace_file.tr = top_trace_array();
3394         if (WARN_ON(!event_trace_file.tr))
3395                 return;
3396
3397         ret = register_ftrace_function(&trace_ops);
3398         if (WARN_ON(ret < 0)) {
3399                 pr_info("Failed to enable function tracer for event tests\n");
3400                 return;
3401         }
3402         pr_info("Running tests again, along with the function tracer\n");
3403         event_trace_self_tests();
3404         unregister_ftrace_function(&trace_ops);
3405 }
3406 #else
3407 static __init void event_trace_self_test_with_function(void)
3408 {
3409 }
3410 #endif
3411
3412 static __init int event_trace_self_tests_init(void)
3413 {
3414         if (!tracing_selftest_disabled) {
3415                 event_trace_self_tests();
3416                 event_trace_self_test_with_function();
3417         }
3418
3419         return 0;
3420 }
3421
3422 late_initcall(event_trace_self_tests_init);
3423
3424 #endif