kernel/trace/trace_events.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * event tracer
   4  *
   5  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   6  *
   7  *  - Added format output of fields of the trace point.
   8  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   9  *
  10  */
  11
  12 #define pr_fmt(fmt) fmt
  13
  14 #include <linux/workqueue.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/kthread.h>
  17 #include <linux/tracefs.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/module.h>
  20 #include <linux/ctype.h>
  21 #include <linux/sort.h>
  22 #include <linux/slab.h>
  23 #include <linux/delay.h>
  24
  25 #include <trace/events/sched.h>
  26
  27 #include <asm/setup.h>
  28
  29 #include "trace_output.h"
  30
  31 #undef TRACE_SYSTEM
  32 #define TRACE_SYSTEM "TRACE_SYSTEM"
  33
  34 DEFINE_MUTEX(event_mutex);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_generic_fields);
  38 static LIST_HEAD(ftrace_common_fields);
  39
  40 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  41
  42 static struct kmem_cache *field_cachep;
  43 static struct kmem_cache *file_cachep;
  44
  45 static inline int system_refcount(struct event_subsystem *system)
  46 {
  47         return system->ref_count;
  48 }
  49
  50 static int system_refcount_inc(struct event_subsystem *system)
  51 {
  52         return system->ref_count++;
  53 }
  54
  55 static int system_refcount_dec(struct event_subsystem *system)
  56 {
  57         return --system->ref_count;
  58 }
  59
  60 /* Double loops, do not use break, only goto's work */
  61 #define do_for_each_event_file(tr, file)                        \
  62         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  63                 list_for_each_entry(file, &tr->events, list)
  64
  65 #define do_for_each_event_file_safe(tr, file)                   \
  66         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  67                 struct trace_event_file *___n;                          \
  68                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  69
  70 #define while_for_each_event_file()             \
  71         }
  72
  73 static struct list_head *
  74 trace_get_fields(struct trace_event_call *event_call)
  75 {
  76         if (!event_call->class->get_fields)
  77                 return &event_call->class->fields;
  78         return event_call->class->get_fields(event_call);
  79 }
  80
  81 static struct ftrace_event_field *
  82 __find_event_field(struct list_head *head, char *name)
  83 {
  84         struct ftrace_event_field *field;
  85
  86         list_for_each_entry(field, head, link) {
  87                 if (!strcmp(field->name, name))
  88                         return field;
  89         }
  90
  91         return NULL;
  92 }
  93
  94 struct ftrace_event_field *
  95 trace_find_event_field(struct trace_event_call *call, char *name)
  96 {
  97         struct ftrace_event_field *field;
  98         struct list_head *head;
  99
 100         head = trace_get_fields(call);
 101         field = __find_event_field(head, name);
 102         if (field)
 103                 return field;
 104
 105         field = __find_event_field(&ftrace_generic_fields, name);
 106         if (field)
 107                 return field;
 108
 109         return __find_event_field(&ftrace_common_fields, name);
 110 }
 111
 112 static int __trace_define_field(struct list_head *head, const char *type,
 113                                 const char *name, int offset, int size,
 114                                 int is_signed, int filter_type)
 115 {
 116         struct ftrace_event_field *field;
 117
 118         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 119         if (!field)
 120                 return -ENOMEM;
 121
 122         field->name = name;
 123         field->type = type;
 124
 125         if (filter_type == FILTER_OTHER)
 126                 field->filter_type = filter_assign_type(type);
 127         else
 128                 field->filter_type = filter_type;
 129
 130         field->offset = offset;
 131         field->size = size;
 132         field->is_signed = is_signed;
 133
 134         list_add(&field->link, head);
 135
 136         return 0;
 137 }
 138
 139 int trace_define_field(struct trace_event_call *call, const char *type,
 140                        const char *name, int offset, int size, int is_signed,
 141                        int filter_type)
 142 {
 143         struct list_head *head;
 144
 145         if (WARN_ON(!call->class))
 146                 return 0;
 147
 148         head = trace_get_fields(call);
 149         return __trace_define_field(head, type, name, offset, size,
 150                                     is_signed, filter_type);
 151 }
 152 EXPORT_SYMBOL_GPL(trace_define_field);
 153
 154 #define __generic_field(type, item, filter_type)                        \
 155         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 156                                    #item, 0, 0, is_signed_type(type),   \
 157                                    filter_type);                        \
 158         if (ret)                                                        \
 159                 return ret;
 160
 161 #define __common_field(type, item)                                      \
 162         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 163                                    "common_" #item,                     \
 164                                    offsetof(typeof(ent), item),         \
 165                                    sizeof(ent.item),                    \
 166                                    is_signed_type(type), FILTER_OTHER); \
 167         if (ret)                                                        \
 168                 return ret;
 169
 170 static int trace_define_generic_fields(void)
 171 {
 172         int ret;
 173
 174         __generic_field(int, CPU, FILTER_CPU);
 175         __generic_field(int, cpu, FILTER_CPU);
 176         __generic_field(char *, COMM, FILTER_COMM);
 177         __generic_field(char *, comm, FILTER_COMM);
 178
 179         return ret;
 180 }
 181
 182 static int trace_define_common_fields(void)
 183 {
 184         int ret;
 185         struct trace_entry ent;
 186
 187         __common_field(unsigned short, type);
 188         __common_field(unsigned char, flags);
 189         __common_field(unsigned char, preempt_count);
 190         __common_field(int, pid);
 191
 192         return ret;
 193 }
 194
 195 static void trace_destroy_fields(struct trace_event_call *call)
 196 {
 197         struct ftrace_event_field *field, *next;
 198         struct list_head *head;
 199
 200         head = trace_get_fields(call);
 201         list_for_each_entry_safe(field, next, head, link) {
 202                 list_del(&field->link);
 203                 kmem_cache_free(field_cachep, field);
 204         }
 205 }
 206
 207 /*
 208  * run-time version of trace_event_get_offsets_<call>() that returns the last
 209  * accessible offset of trace fields excluding __dynamic_array bytes
 210  */
 211 int trace_event_get_offsets(struct trace_event_call *call)
 212 {
 213         struct ftrace_event_field *tail;
 214         struct list_head *head;
 215
 216         head = trace_get_fields(call);
 217         /*
 218          * head->next points to the last field with the largest offset,
 219          * since it was added last by trace_define_field()
 220          */
 221         tail = list_first_entry(head, struct ftrace_event_field, link);
 222         return tail->offset + tail->size;
 223 }
 224
 225 int trace_event_raw_init(struct trace_event_call *call)
 226 {
 227         int id;
 228
 229         id = register_trace_event(&call->event);
 230         if (!id)
 231                 return -ENODEV;
 232
 233         return 0;
 234 }
 235 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 236
 237 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 238 {
 239         struct trace_array *tr = trace_file->tr;
 240         struct trace_array_cpu *data;
 241         struct trace_pid_list *pid_list;
 242
 243         pid_list = rcu_dereference_raw(tr->filtered_pids);
 244         if (!pid_list)
 245                 return false;
 246
 247         data = this_cpu_ptr(tr->trace_buffer.data);
 248
 249         return data->ignore_pid;
 250 }
 251 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 252
 253 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 254                                  struct trace_event_file *trace_file,
 255                                  unsigned long len)
 256 {
 257         struct trace_event_call *event_call = trace_file->event_call;
 258
 259         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 260             trace_event_ignore_this_pid(trace_file))
 261                 return NULL;
 262
 263         local_save_flags(fbuffer->flags);
 264         fbuffer->pc = preempt_count();
 265         /*
 266          * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
 267          * preemption (adding one to the preempt_count). Since we are
 268          * interested in the preempt_count at the time the tracepoint was
 269          * hit, we need to subtract one to offset the increment.
 270          */
 271         if (IS_ENABLED(CONFIG_PREEMPT))
 272                 fbuffer->pc--;
 273         fbuffer->trace_file = trace_file;
 274
 275         fbuffer->event =
 276                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 277                                                 event_call->event.type, len,
 278                                                 fbuffer->flags, fbuffer->pc);
 279         if (!fbuffer->event)
 280                 return NULL;
 281
 282         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 283         return fbuffer->entry;
 284 }
 285 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 286
 287 int trace_event_reg(struct trace_event_call *call,
 288                     enum trace_reg type, void *data)
 289 {
 290         struct trace_event_file *file = data;
 291
 292         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 293         switch (type) {
 294         case TRACE_REG_REGISTER:
 295                 return tracepoint_probe_register(call->tp,
 296                                                  call->class->probe,
 297                                                  file);
 298         case TRACE_REG_UNREGISTER:
 299                 tracepoint_probe_unregister(call->tp,
 300                                             call->class->probe,
 301                                             file);
 302                 return 0;
 303
 304 #ifdef CONFIG_PERF_EVENTS
 305         case TRACE_REG_PERF_REGISTER:
 306                 return tracepoint_probe_register(call->tp,
 307                                                  call->class->perf_probe,
 308                                                  call);
 309         case TRACE_REG_PERF_UNREGISTER:
 310                 tracepoint_probe_unregister(call->tp,
 311                                             call->class->perf_probe,
 312                                             call);
 313                 return 0;
 314         case TRACE_REG_PERF_OPEN:
 315         case TRACE_REG_PERF_CLOSE:
 316         case TRACE_REG_PERF_ADD:
 317         case TRACE_REG_PERF_DEL:
 318                 return 0;
 319 #endif
 320         }
 321         return 0;
 322 }
 323 EXPORT_SYMBOL_GPL(trace_event_reg);
 324
 325 void trace_event_enable_cmd_record(bool enable)
 326 {
 327         struct trace_event_file *file;
 328         struct trace_array *tr;
 329
 330         mutex_lock(&event_mutex);
 331         do_for_each_event_file(tr, file) {
 332
 333                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 334                         continue;
 335
 336                 if (enable) {
 337                         tracing_start_cmdline_record();
 338                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 339                 } else {
 340                         tracing_stop_cmdline_record();
 341                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 342                 }
 343         } while_for_each_event_file();
 344         mutex_unlock(&event_mutex);
 345 }
 346
 347 void trace_event_enable_tgid_record(bool enable)
 348 {
 349         struct trace_event_file *file;
 350         struct trace_array *tr;
 351
 352         mutex_lock(&event_mutex);
 353         do_for_each_event_file(tr, file) {
 354                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 355                         continue;
 356
 357                 if (enable) {
 358                         tracing_start_tgid_record();
 359                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 360                 } else {
 361                         tracing_stop_tgid_record();
 362                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 363                                   &file->flags);
 364                 }
 365         } while_for_each_event_file();
 366         mutex_unlock(&event_mutex);
 367 }
 368
 369 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 370                                          int enable, int soft_disable)
 371 {
 372         struct trace_event_call *call = file->event_call;
 373         struct trace_array *tr = file->tr;
 374         unsigned long file_flags = file->flags;
 375         int ret = 0;
 376         int disable;
 377
 378         switch (enable) {
 379         case 0:
 380                 /*
 381                  * When soft_disable is set and enable is cleared, the sm_ref
 382                  * reference counter is decremented. If it reaches 0, we want
 383                  * to clear the SOFT_DISABLED flag but leave the event in the
 384                  * state that it was. That is, if the event was enabled and
 385                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 386                  * is set we do not want the event to be enabled before we
 387                  * clear the bit.
 388                  *
 389                  * When soft_disable is not set but the SOFT_MODE flag is,
 390                  * we do nothing. Do not disable the tracepoint, otherwise
 391                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 392                  */
 393                 if (soft_disable) {
 394                         if (atomic_dec_return(&file->sm_ref) > 0)
 395                                 break;
 396                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 397                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 398                 } else
 399                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 400
 401                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 402                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 403                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 404                                 tracing_stop_cmdline_record();
 405                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 406                         }
 407
 408                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 409                                 tracing_stop_tgid_record();
 410                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 411                         }
 412
 413                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 414                 }
 415                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 416                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 417                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 418                 else
 419                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 420                 break;
 421         case 1:
 422                 /*
 423                  * When soft_disable is set and enable is set, we want to
 424                  * register the tracepoint for the event, but leave the event
 425                  * as is. That means, if the event was already enabled, we do
 426                  * nothing (but set SOFT_MODE). If the event is disabled, we
 427                  * set SOFT_DISABLED before enabling the event tracepoint, so
 428                  * it still seems to be disabled.
 429                  */
 430                 if (!soft_disable)
 431                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 432                 else {
 433                         if (atomic_inc_return(&file->sm_ref) > 1)
 434                                 break;
 435                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 436                 }
 437
 438                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 439                         bool cmd = false, tgid = false;
 440
 441                         /* Keep the event disabled, when going to SOFT_MODE. */
 442                         if (soft_disable)
 443                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 444
 445                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 446                                 cmd = true;
 447                                 tracing_start_cmdline_record();
 448                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 449                         }
 450
 451                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 452                                 tgid = true;
 453                                 tracing_start_tgid_record();
 454                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 455                         }
 456
 457                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 458                         if (ret) {
 459                                 if (cmd)
 460                                         tracing_stop_cmdline_record();
 461                                 if (tgid)
 462                                         tracing_stop_tgid_record();
 463                                 pr_info("event trace: Could not enable event "
 464                                         "%s\n", trace_event_name(call));
 465                                 break;
 466                         }
 467                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 468
 469                         /* WAS_ENABLED gets set but never cleared. */
 470                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 471                 }
 472                 break;
 473         }
 474
 475         /* Enable or disable use of trace_buffered_event */
 476         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 477             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 478                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 479                         trace_buffered_event_enable();
 480                 else
 481                         trace_buffered_event_disable();
 482         }
 483
 484         return ret;
 485 }
 486
 487 int trace_event_enable_disable(struct trace_event_file *file,
 488                                int enable, int soft_disable)
 489 {
 490         return __ftrace_event_enable_disable(file, enable, soft_disable);
 491 }
 492
 493 static int ftrace_event_enable_disable(struct trace_event_file *file,
 494                                        int enable)
 495 {
 496         return __ftrace_event_enable_disable(file, enable, 0);
 497 }
 498
 499 static void ftrace_clear_events(struct trace_array *tr)
 500 {
 501         struct trace_event_file *file;
 502
 503         mutex_lock(&event_mutex);
 504         list_for_each_entry(file, &tr->events, list) {
 505                 ftrace_event_enable_disable(file, 0);
 506         }
 507         mutex_unlock(&event_mutex);
 508 }
 509
 510 static void
 511 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 512 {
 513         struct trace_pid_list *pid_list;
 514         struct trace_array *tr = data;
 515
 516         pid_list = rcu_dereference_raw(tr->filtered_pids);
 517         trace_filter_add_remove_task(pid_list, NULL, task);
 518 }
 519
 520 static void
 521 event_filter_pid_sched_process_fork(void *data,
 522                                     struct task_struct *self,
 523                                     struct task_struct *task)
 524 {
 525         struct trace_pid_list *pid_list;
 526         struct trace_array *tr = data;
 527
 528         pid_list = rcu_dereference_sched(tr->filtered_pids);
 529         trace_filter_add_remove_task(pid_list, self, task);
 530 }
 531
 532 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 533 {
 534         if (enable) {
 535                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 536                                                        tr, INT_MIN);
 537                 register_trace_prio_sched_process_exit(event_filter_pid_sched_process_exit,
 538                                                        tr, INT_MAX);
 539         } else {
 540                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 541                                                     tr);
 542                 unregister_trace_sched_process_exit(event_filter_pid_sched_process_exit,
 543                                                     tr);
 544         }
 545 }
 546
 547 static void
 548 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 549                     struct task_struct *prev, struct task_struct *next)
 550 {
 551         struct trace_array *tr = data;
 552         struct trace_pid_list *pid_list;
 553
 554         pid_list = rcu_dereference_sched(tr->filtered_pids);
 555
 556         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 557                        trace_ignore_this_task(pid_list, prev) &&
 558                        trace_ignore_this_task(pid_list, next));
 559 }
 560
 561 static void
 562 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 563                     struct task_struct *prev, struct task_struct *next)
 564 {
 565         struct trace_array *tr = data;
 566         struct trace_pid_list *pid_list;
 567
 568         pid_list = rcu_dereference_sched(tr->filtered_pids);
 569
 570         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 571                        trace_ignore_this_task(pid_list, next));
 572 }
 573
 574 static void
 575 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 576 {
 577         struct trace_array *tr = data;
 578         struct trace_pid_list *pid_list;
 579
 580         /* Nothing to do if we are already tracing */
 581         if (!this_cpu_read(tr->trace_buffer.data->ignore_pid))
 582                 return;
 583
 584         pid_list = rcu_dereference_sched(tr->filtered_pids);
 585
 586         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 587                        trace_ignore_this_task(pid_list, task));
 588 }
 589
 590 static void
 591 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 592 {
 593         struct trace_array *tr = data;
 594         struct trace_pid_list *pid_list;
 595
 596         /* Nothing to do if we are not tracing */
 597         if (this_cpu_read(tr->trace_buffer.data->ignore_pid))
 598                 return;
 599
 600         pid_list = rcu_dereference_sched(tr->filtered_pids);
 601
 602         /* Set tracing if current is enabled */
 603         this_cpu_write(tr->trace_buffer.data->ignore_pid,
 604                        trace_ignore_this_task(pid_list, current));
 605 }
 606
 607 static void __ftrace_clear_event_pids(struct trace_array *tr)
 608 {
 609         struct trace_pid_list *pid_list;
 610         struct trace_event_file *file;
 611         int cpu;
 612
 613         pid_list = rcu_dereference_protected(tr->filtered_pids,
 614                                              lockdep_is_held(&event_mutex));
 615         if (!pid_list)
 616                 return;
 617
 618         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 619         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 620
 621         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 622         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 623
 624         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 625         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 626
 627         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 628         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 629
 630         list_for_each_entry(file, &tr->events, list) {
 631                 clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 632         }
 633
 634         for_each_possible_cpu(cpu)
 635                 per_cpu_ptr(tr->trace_buffer.data, cpu)->ignore_pid = false;
 636
 637         rcu_assign_pointer(tr->filtered_pids, NULL);
 638
 639         /* Wait till all users are no longer using pid filtering */
 640         tracepoint_synchronize_unregister();
 641
 642         trace_free_pid_list(pid_list);
 643 }
 644
 645 static void ftrace_clear_event_pids(struct trace_array *tr)
 646 {
 647         mutex_lock(&event_mutex);
 648         __ftrace_clear_event_pids(tr);
 649         mutex_unlock(&event_mutex);
 650 }
 651
 652 static void __put_system(struct event_subsystem *system)
 653 {
 654         struct event_filter *filter = system->filter;
 655
 656         WARN_ON_ONCE(system_refcount(system) == 0);
 657         if (system_refcount_dec(system))
 658                 return;
 659
 660         list_del(&system->list);
 661
 662         if (filter) {
 663                 kfree(filter->filter_string);
 664                 kfree(filter);
 665         }
 666         kfree_const(system->name);
 667         kfree(system);
 668 }
 669
 670 static void __get_system(struct event_subsystem *system)
 671 {
 672         WARN_ON_ONCE(system_refcount(system) == 0);
 673         system_refcount_inc(system);
 674 }
 675
 676 static void __get_system_dir(struct trace_subsystem_dir *dir)
 677 {
 678         WARN_ON_ONCE(dir->ref_count == 0);
 679         dir->ref_count++;
 680         __get_system(dir->subsystem);
 681 }
 682
 683 static void __put_system_dir(struct trace_subsystem_dir *dir)
 684 {
 685         WARN_ON_ONCE(dir->ref_count == 0);
 686         /* If the subsystem is about to be freed, the dir must be too */
 687         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 688
 689         __put_system(dir->subsystem);
 690         if (!--dir->ref_count)
 691                 kfree(dir);
 692 }
 693
 694 static void put_system(struct trace_subsystem_dir *dir)
 695 {
 696         mutex_lock(&event_mutex);
 697         __put_system_dir(dir);
 698         mutex_unlock(&event_mutex);
 699 }
 700
 701 static void remove_subsystem(struct trace_subsystem_dir *dir)
 702 {
 703         if (!dir)
 704                 return;
 705
 706         if (!--dir->nr_events) {
 707                 tracefs_remove_recursive(dir->entry);
 708                 list_del(&dir->list);
 709                 __put_system_dir(dir);
 710         }
 711 }
 712
 713 static void remove_event_file_dir(struct trace_event_file *file)
 714 {
 715         struct dentry *dir = file->dir;
 716         struct dentry *child;
 717
 718         if (dir) {
 719                 spin_lock(&dir->d_lock);        /* probably unneeded */
 720                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 721                         if (d_really_is_positive(child))        /* probably unneeded */
 722                                 d_inode(child)->i_private = NULL;
 723                 }
 724                 spin_unlock(&dir->d_lock);
 725
 726                 tracefs_remove_recursive(dir);
 727         }
 728
 729         list_del(&file->list);
 730         remove_subsystem(file->system);
 731         free_event_filter(file->filter);
 732         kmem_cache_free(file_cachep, file);
 733 }
 734
 735 /*
 736  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 737  */
 738 static int
 739 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 740                               const char *sub, const char *event, int set)
 741 {
 742         struct trace_event_file *file;
 743         struct trace_event_call *call;
 744         const char *name;
 745         int ret = -EINVAL;
 746         int eret = 0;
 747
 748         list_for_each_entry(file, &tr->events, list) {
 749
 750                 call = file->event_call;
 751                 name = trace_event_name(call);
 752
 753                 if (!name || !call->class || !call->class->reg)
 754                         continue;
 755
 756                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 757                         continue;
 758
 759                 if (match &&
 760                     strcmp(match, name) != 0 &&
 761                     strcmp(match, call->class->system) != 0)
 762                         continue;
 763
 764                 if (sub && strcmp(sub, call->class->system) != 0)
 765                         continue;
 766
 767                 if (event && strcmp(event, name) != 0)
 768                         continue;
 769
 770                 ret = ftrace_event_enable_disable(file, set);
 771
 772                 /*
 773                  * Save the first error and return that. Some events
 774                  * may still have been enabled, but let the user
 775                  * know that something went wrong.
 776                  */
 777                 if (ret && !eret)
 778                         eret = ret;
 779
 780                 ret = eret;
 781         }
 782
 783         return ret;
 784 }
 785
 786 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 787                                   const char *sub, const char *event, int set)
 788 {
 789         int ret;
 790
 791         mutex_lock(&event_mutex);
 792         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 793         mutex_unlock(&event_mutex);
 794
 795         return ret;
 796 }
 797
 798 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 799 {
 800         char *event = NULL, *sub = NULL, *match;
 801         int ret;
 802
 803         /*
 804          * The buf format can be <subsystem>:<event-name>
 805          *  *:<event-name> means any event by that name.
 806          *  :<event-name> is the same.
 807          *
 808          *  <subsystem>:* means all events in that subsystem
 809          *  <subsystem>: means the same.
 810          *
 811          *  <name> (no ':') means all events in a subsystem with
 812          *  the name <name> or any event that matches <name>
 813          */
 814
 815         match = strsep(&buf, ":");
 816         if (buf) {
 817                 sub = match;
 818                 event = buf;
 819                 match = NULL;
 820
 821                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 822                         sub = NULL;
 823                 if (!strlen(event) || strcmp(event, "*") == 0)
 824                         event = NULL;
 825         }
 826
 827         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 828
 829         /* Put back the colon to allow this to be called again */
 830         if (buf)
 831                 *(buf - 1) = ':';
 832
 833         return ret;
 834 }
 835
 836 /**
 837  * trace_set_clr_event - enable or disable an event
 838  * @system: system name to match (NULL for any system)
 839  * @event: event name to match (NULL for all events, within system)
 840  * @set: 1 to enable, 0 to disable
 841  *
 842  * This is a way for other parts of the kernel to enable or disable
 843  * event recording.
 844  *
 845  * Returns 0 on success, -EINVAL if the parameters do not match any
 846  * registered events.
 847  */
 848 int trace_set_clr_event(const char *system, const char *event, int set)
 849 {
 850         struct trace_array *tr = top_trace_array();
 851
 852         if (!tr)
 853                 return -ENODEV;
 854
 855         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 856 }
 857 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 858
 859 /* 128 should be much more than enough */
 860 #define EVENT_BUF_SIZE          127
 861
 862 static ssize_t
 863 ftrace_event_write(struct file *file, const char __user *ubuf,
 864                    size_t cnt, loff_t *ppos)
 865 {
 866         struct trace_parser parser;
 867         struct seq_file *m = file->private_data;
 868         struct trace_array *tr = m->private;
 869         ssize_t read, ret;
 870
 871         if (!cnt)
 872                 return 0;
 873
 874         ret = tracing_update_buffers();
 875         if (ret < 0)
 876                 return ret;
 877
 878         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 879                 return -ENOMEM;
 880
 881         read = trace_get_user(&parser, ubuf, cnt, ppos);
 882
 883         if (read >= 0 && trace_parser_loaded((&parser))) {
 884                 int set = 1;
 885
 886                 if (*parser.buffer == '!')
 887                         set = 0;
 888
 889                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 890                 if (ret)
 891                         goto out_put;
 892         }
 893
 894         ret = read;
 895
 896  out_put:
 897         trace_parser_put(&parser);
 898
 899         return ret;
 900 }
 901
 902 static void *
 903 t_next(struct seq_file *m, void *v, loff_t *pos)
 904 {
 905         struct trace_event_file *file = v;
 906         struct trace_event_call *call;
 907         struct trace_array *tr = m->private;
 908
 909         (*pos)++;
 910
 911         list_for_each_entry_continue(file, &tr->events, list) {
 912                 call = file->event_call;
 913                 /*
 914                  * The ftrace subsystem is for showing formats only.
 915                  * They can not be enabled or disabled via the event files.
 916                  */
 917                 if (call->class && call->class->reg &&
 918                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 919                         return file;
 920         }
 921
 922         return NULL;
 923 }
 924
 925 static void *t_start(struct seq_file *m, loff_t *pos)
 926 {
 927         struct trace_event_file *file;
 928         struct trace_array *tr = m->private;
 929         loff_t l;
 930
 931         mutex_lock(&event_mutex);
 932
 933         file = list_entry(&tr->events, struct trace_event_file, list);
 934         for (l = 0; l <= *pos; ) {
 935                 file = t_next(m, file, &l);
 936                 if (!file)
 937                         break;
 938         }
 939         return file;
 940 }
 941
 942 static void *
 943 s_next(struct seq_file *m, void *v, loff_t *pos)
 944 {
 945         struct trace_event_file *file = v;
 946         struct trace_array *tr = m->private;
 947
 948         (*pos)++;
 949
 950         list_for_each_entry_continue(file, &tr->events, list) {
 951                 if (file->flags & EVENT_FILE_FL_ENABLED)
 952                         return file;
 953         }
 954
 955         return NULL;
 956 }
 957
 958 static void *s_start(struct seq_file *m, loff_t *pos)
 959 {
 960         struct trace_event_file *file;
 961         struct trace_array *tr = m->private;
 962         loff_t l;
 963
 964         mutex_lock(&event_mutex);
 965
 966         file = list_entry(&tr->events, struct trace_event_file, list);
 967         for (l = 0; l <= *pos; ) {
 968                 file = s_next(m, file, &l);
 969                 if (!file)
 970                         break;
 971         }
 972         return file;
 973 }
 974
 975 static int t_show(struct seq_file *m, void *v)
 976 {
 977         struct trace_event_file *file = v;
 978         struct trace_event_call *call = file->event_call;
 979
 980         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 981                 seq_printf(m, "%s:", call->class->system);
 982         seq_printf(m, "%s\n", trace_event_name(call));
 983
 984         return 0;
 985 }
 986
 987 static void t_stop(struct seq_file *m, void *p)
 988 {
 989         mutex_unlock(&event_mutex);
 990 }
 991
 992 static void *
 993 p_next(struct seq_file *m, void *v, loff_t *pos)
 994 {
 995         struct trace_array *tr = m->private;
 996         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 997
 998         return trace_pid_next(pid_list, v, pos);
 999 }
1000
1001 static void *p_start(struct seq_file *m, loff_t *pos)
1002         __acquires(RCU)
1003 {
1004         struct trace_pid_list *pid_list;
1005         struct trace_array *tr = m->private;
1006
1007         /*
1008          * Grab the mutex, to keep calls to p_next() having the same
1009          * tr->filtered_pids as p_start() has.
1010          * If we just passed the tr->filtered_pids around, then RCU would
1011          * have been enough, but doing that makes things more complex.
1012          */
1013         mutex_lock(&event_mutex);
1014         rcu_read_lock_sched();
1015
1016         pid_list = rcu_dereference_sched(tr->filtered_pids);
1017
1018         if (!pid_list)
1019                 return NULL;
1020
1021         return trace_pid_start(pid_list, pos);
1022 }
1023
1024 static void p_stop(struct seq_file *m, void *p)
1025         __releases(RCU)
1026 {
1027         rcu_read_unlock_sched();
1028         mutex_unlock(&event_mutex);
1029 }
1030
1031 static ssize_t
1032 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1033                   loff_t *ppos)
1034 {
1035         struct trace_event_file *file;
1036         unsigned long flags;
1037         char buf[4] = "0";
1038
1039         mutex_lock(&event_mutex);
1040         file = event_file_data(filp);
1041         if (likely(file))
1042                 flags = file->flags;
1043         mutex_unlock(&event_mutex);
1044
1045         if (!file)
1046                 return -ENODEV;
1047
1048         if (flags & EVENT_FILE_FL_ENABLED &&
1049             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1050                 strcpy(buf, "1");
1051
1052         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1053             flags & EVENT_FILE_FL_SOFT_MODE)
1054                 strcat(buf, "*");
1055
1056         strcat(buf, "\n");
1057
1058         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1059 }
1060
1061 static ssize_t
1062 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1063                    loff_t *ppos)
1064 {
1065         struct trace_event_file *file;
1066         unsigned long val;
1067         int ret;
1068
1069         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1070         if (ret)
1071                 return ret;
1072
1073         ret = tracing_update_buffers();
1074         if (ret < 0)
1075                 return ret;
1076
1077         switch (val) {
1078         case 0:
1079         case 1:
1080                 ret = -ENODEV;
1081                 mutex_lock(&event_mutex);
1082                 file = event_file_data(filp);
1083                 if (likely(file))
1084                         ret = ftrace_event_enable_disable(file, val);
1085                 mutex_unlock(&event_mutex);
1086                 break;
1087
1088         default:
1089                 return -EINVAL;
1090         }
1091
1092         *ppos += cnt;
1093
1094         return ret ? ret : cnt;
1095 }
1096
1097 static ssize_t
1098 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1099                    loff_t *ppos)
1100 {
1101         const char set_to_char[4] = { '?', '0', '1', 'X' };
1102         struct trace_subsystem_dir *dir = filp->private_data;
1103         struct event_subsystem *system = dir->subsystem;
1104         struct trace_event_call *call;
1105         struct trace_event_file *file;
1106         struct trace_array *tr = dir->tr;
1107         char buf[2];
1108         int set = 0;
1109         int ret;
1110
1111         mutex_lock(&event_mutex);
1112         list_for_each_entry(file, &tr->events, list) {
1113                 call = file->event_call;
1114                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1115                         continue;
1116
1117                 if (system && strcmp(call->class->system, system->name) != 0)
1118                         continue;
1119
1120                 /*
1121                  * We need to find out if all the events are set
1122                  * or if all events or cleared, or if we have
1123                  * a mixture.
1124                  */
1125                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1126
1127                 /*
1128                  * If we have a mixture, no need to look further.
1129                  */
1130                 if (set == 3)
1131                         break;
1132         }
1133         mutex_unlock(&event_mutex);
1134
1135         buf[0] = set_to_char[set];
1136         buf[1] = '\n';
1137
1138         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1139
1140         return ret;
1141 }
1142
1143 static ssize_t
1144 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1145                     loff_t *ppos)
1146 {
1147         struct trace_subsystem_dir *dir = filp->private_data;
1148         struct event_subsystem *system = dir->subsystem;
1149         const char *name = NULL;
1150         unsigned long val;
1151         ssize_t ret;
1152
1153         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1154         if (ret)
1155                 return ret;
1156
1157         ret = tracing_update_buffers();
1158         if (ret < 0)
1159                 return ret;
1160
1161         if (val != 0 && val != 1)
1162                 return -EINVAL;
1163
1164         /*
1165          * Opening of "enable" adds a ref count to system,
1166          * so the name is safe to use.
1167          */
1168         if (system)
1169                 name = system->name;
1170
1171         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1172         if (ret)
1173                 goto out;
1174
1175         ret = cnt;
1176
1177 out:
1178         *ppos += cnt;
1179
1180         return ret;
1181 }
1182
1183 enum {
1184         FORMAT_HEADER           = 1,
1185         FORMAT_FIELD_SEPERATOR  = 2,
1186         FORMAT_PRINTFMT         = 3,
1187 };
1188
1189 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1190 {
1191         struct trace_event_call *call = event_file_data(m->private);
1192         struct list_head *common_head = &ftrace_common_fields;
1193         struct list_head *head = trace_get_fields(call);
1194         struct list_head *node = v;
1195
1196         (*pos)++;
1197
1198         switch ((unsigned long)v) {
1199         case FORMAT_HEADER:
1200                 node = common_head;
1201                 break;
1202
1203         case FORMAT_FIELD_SEPERATOR:
1204                 node = head;
1205                 break;
1206
1207         case FORMAT_PRINTFMT:
1208                 /* all done */
1209                 return NULL;
1210         }
1211
1212         node = node->prev;
1213         if (node == common_head)
1214                 return (void *)FORMAT_FIELD_SEPERATOR;
1215         else if (node == head)
1216                 return (void *)FORMAT_PRINTFMT;
1217         else
1218                 return node;
1219 }
1220
1221 static int f_show(struct seq_file *m, void *v)
1222 {
1223         struct trace_event_call *call = event_file_data(m->private);
1224         struct ftrace_event_field *field;
1225         const char *array_descriptor;
1226
1227         switch ((unsigned long)v) {
1228         case FORMAT_HEADER:
1229                 seq_printf(m, "name: %s\n", trace_event_name(call));
1230                 seq_printf(m, "ID: %d\n", call->event.type);
1231                 seq_puts(m, "format:\n");
1232                 return 0;
1233
1234         case FORMAT_FIELD_SEPERATOR:
1235                 seq_putc(m, '\n');
1236                 return 0;
1237
1238         case FORMAT_PRINTFMT:
1239                 seq_printf(m, "\nprint fmt: %s\n",
1240                            call->print_fmt);
1241                 return 0;
1242         }
1243
1244         field = list_entry(v, struct ftrace_event_field, link);
1245         /*
1246          * Smartly shows the array type(except dynamic array).
1247          * Normal:
1248          *      field:TYPE VAR
1249          * If TYPE := TYPE[LEN], it is shown:
1250          *      field:TYPE VAR[LEN]
1251          */
1252         array_descriptor = strchr(field->type, '[');
1253
1254         if (str_has_prefix(field->type, "__data_loc"))
1255                 array_descriptor = NULL;
1256
1257         if (!array_descriptor)
1258                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1259                            field->type, field->name, field->offset,
1260                            field->size, !!field->is_signed);
1261         else
1262                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1263                            (int)(array_descriptor - field->type),
1264                            field->type, field->name,
1265                            array_descriptor, field->offset,
1266                            field->size, !!field->is_signed);
1267
1268         return 0;
1269 }
1270
1271 static void *f_start(struct seq_file *m, loff_t *pos)
1272 {
1273         void *p = (void *)FORMAT_HEADER;
1274         loff_t l = 0;
1275
1276         /* ->stop() is called even if ->start() fails */
1277         mutex_lock(&event_mutex);
1278         if (!event_file_data(m->private))
1279                 return ERR_PTR(-ENODEV);
1280
1281         while (l < *pos && p)
1282                 p = f_next(m, p, &l);
1283
1284         return p;
1285 }
1286
1287 static void f_stop(struct seq_file *m, void *p)
1288 {
1289         mutex_unlock(&event_mutex);
1290 }
1291
1292 static const struct seq_operations trace_format_seq_ops = {
1293         .start          = f_start,
1294         .next           = f_next,
1295         .stop           = f_stop,
1296         .show           = f_show,
1297 };
1298
1299 static int trace_format_open(struct inode *inode, struct file *file)
1300 {
1301         struct seq_file *m;
1302         int ret;
1303
1304         ret = seq_open(file, &trace_format_seq_ops);
1305         if (ret < 0)
1306                 return ret;
1307
1308         m = file->private_data;
1309         m->private = file;
1310
1311         return 0;
1312 }
1313
1314 static ssize_t
1315 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1316 {
1317         int id = (long)event_file_data(filp);
1318         char buf[32];
1319         int len;
1320
1321         if (*ppos)
1322                 return 0;
1323
1324         if (unlikely(!id))
1325                 return -ENODEV;
1326
1327         len = sprintf(buf, "%d\n", id);
1328
1329         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1330 }
1331
1332 static ssize_t
1333 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1334                   loff_t *ppos)
1335 {
1336         struct trace_event_file *file;
1337         struct trace_seq *s;
1338         int r = -ENODEV;
1339
1340         if (*ppos)
1341                 return 0;
1342
1343         s = kmalloc(sizeof(*s), GFP_KERNEL);
1344
1345         if (!s)
1346                 return -ENOMEM;
1347
1348         trace_seq_init(s);
1349
1350         mutex_lock(&event_mutex);
1351         file = event_file_data(filp);
1352         if (file)
1353                 print_event_filter(file, s);
1354         mutex_unlock(&event_mutex);
1355
1356         if (file)
1357                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1358                                             s->buffer, trace_seq_used(s));
1359
1360         kfree(s);
1361
1362         return r;
1363 }
1364
1365 static ssize_t
1366 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1367                    loff_t *ppos)
1368 {
1369         struct trace_event_file *file;
1370         char *buf;
1371         int err = -ENODEV;
1372
1373         if (cnt >= PAGE_SIZE)
1374                 return -EINVAL;
1375
1376         buf = memdup_user_nul(ubuf, cnt);
1377         if (IS_ERR(buf))
1378                 return PTR_ERR(buf);
1379
1380         mutex_lock(&event_mutex);
1381         file = event_file_data(filp);
1382         if (file)
1383                 err = apply_event_filter(file, buf);
1384         mutex_unlock(&event_mutex);
1385
1386         kfree(buf);
1387         if (err < 0)
1388                 return err;
1389
1390         *ppos += cnt;
1391
1392         return cnt;
1393 }
1394
1395 static LIST_HEAD(event_subsystems);
1396
1397 static int subsystem_open(struct inode *inode, struct file *filp)
1398 {
1399         struct event_subsystem *system = NULL;
1400         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1401         struct trace_array *tr;
1402         int ret;
1403
1404         if (tracing_is_disabled())
1405                 return -ENODEV;
1406
1407         /* Make sure the system still exists */
1408         mutex_lock(&event_mutex);
1409         mutex_lock(&trace_types_lock);
1410         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1411                 list_for_each_entry(dir, &tr->systems, list) {
1412                         if (dir == inode->i_private) {
1413                                 /* Don't open systems with no events */
1414                                 if (dir->nr_events) {
1415                                         __get_system_dir(dir);
1416                                         system = dir->subsystem;
1417                                 }
1418                                 goto exit_loop;
1419                         }
1420                 }
1421         }
1422  exit_loop:
1423         mutex_unlock(&trace_types_lock);
1424         mutex_unlock(&event_mutex);
1425
1426         if (!system)
1427                 return -ENODEV;
1428
1429         /* Some versions of gcc think dir can be uninitialized here */
1430         WARN_ON(!dir);
1431
1432         /* Still need to increment the ref count of the system */
1433         if (trace_array_get(tr) < 0) {
1434                 put_system(dir);
1435                 return -ENODEV;
1436         }
1437
1438         ret = tracing_open_generic(inode, filp);
1439         if (ret < 0) {
1440                 trace_array_put(tr);
1441                 put_system(dir);
1442         }
1443
1444         return ret;
1445 }
1446
1447 static int system_tr_open(struct inode *inode, struct file *filp)
1448 {
1449         struct trace_subsystem_dir *dir;
1450         struct trace_array *tr = inode->i_private;
1451         int ret;
1452
1453         if (tracing_is_disabled())
1454                 return -ENODEV;
1455
1456         if (trace_array_get(tr) < 0)
1457                 return -ENODEV;
1458
1459         /* Make a temporary dir that has no system but points to tr */
1460         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1461         if (!dir) {
1462                 trace_array_put(tr);
1463                 return -ENOMEM;
1464         }
1465
1466         dir->tr = tr;
1467
1468         ret = tracing_open_generic(inode, filp);
1469         if (ret < 0) {
1470                 trace_array_put(tr);
1471                 kfree(dir);
1472                 return ret;
1473         }
1474
1475         filp->private_data = dir;
1476
1477         return 0;
1478 }
1479
1480 static int subsystem_release(struct inode *inode, struct file *file)
1481 {
1482         struct trace_subsystem_dir *dir = file->private_data;
1483
1484         trace_array_put(dir->tr);
1485
1486         /*
1487          * If dir->subsystem is NULL, then this is a temporary
1488          * descriptor that was made for a trace_array to enable
1489          * all subsystems.
1490          */
1491         if (dir->subsystem)
1492                 put_system(dir);
1493         else
1494                 kfree(dir);
1495
1496         return 0;
1497 }
1498
1499 static ssize_t
1500 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1501                       loff_t *ppos)
1502 {
1503         struct trace_subsystem_dir *dir = filp->private_data;
1504         struct event_subsystem *system = dir->subsystem;
1505         struct trace_seq *s;
1506         int r;
1507
1508         if (*ppos)
1509                 return 0;
1510
1511         s = kmalloc(sizeof(*s), GFP_KERNEL);
1512         if (!s)
1513                 return -ENOMEM;
1514
1515         trace_seq_init(s);
1516
1517         print_subsystem_event_filter(system, s);
1518         r = simple_read_from_buffer(ubuf, cnt, ppos,
1519                                     s->buffer, trace_seq_used(s));
1520
1521         kfree(s);
1522
1523         return r;
1524 }
1525
1526 static ssize_t
1527 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1528                        loff_t *ppos)
1529 {
1530         struct trace_subsystem_dir *dir = filp->private_data;
1531         char *buf;
1532         int err;
1533
1534         if (cnt >= PAGE_SIZE)
1535                 return -EINVAL;
1536
1537         buf = memdup_user_nul(ubuf, cnt);
1538         if (IS_ERR(buf))
1539                 return PTR_ERR(buf);
1540
1541         err = apply_subsystem_event_filter(dir, buf);
1542         kfree(buf);
1543         if (err < 0)
1544                 return err;
1545
1546         *ppos += cnt;
1547
1548         return cnt;
1549 }
1550
1551 static ssize_t
1552 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1553 {
1554         int (*func)(struct trace_seq *s) = filp->private_data;
1555         struct trace_seq *s;
1556         int r;
1557
1558         if (*ppos)
1559                 return 0;
1560
1561         s = kmalloc(sizeof(*s), GFP_KERNEL);
1562         if (!s)
1563                 return -ENOMEM;
1564
1565         trace_seq_init(s);
1566
1567         func(s);
1568         r = simple_read_from_buffer(ubuf, cnt, ppos,
1569                                     s->buffer, trace_seq_used(s));
1570
1571         kfree(s);
1572
1573         return r;
1574 }
1575
1576 static void ignore_task_cpu(void *data)
1577 {
1578         struct trace_array *tr = data;
1579         struct trace_pid_list *pid_list;
1580
1581         /*
1582          * This function is called by on_each_cpu() while the
1583          * event_mutex is held.
1584          */
1585         pid_list = rcu_dereference_protected(tr->filtered_pids,
1586                                              mutex_is_locked(&event_mutex));
1587
1588         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1589                        trace_ignore_this_task(pid_list, current));
1590 }
1591
1592 static ssize_t
1593 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1594                        size_t cnt, loff_t *ppos)
1595 {
1596         struct seq_file *m = filp->private_data;
1597         struct trace_array *tr = m->private;
1598         struct trace_pid_list *filtered_pids = NULL;
1599         struct trace_pid_list *pid_list;
1600         struct trace_event_file *file;
1601         ssize_t ret;
1602
1603         if (!cnt)
1604                 return 0;
1605
1606         ret = tracing_update_buffers();
1607         if (ret < 0)
1608                 return ret;
1609
1610         mutex_lock(&event_mutex);
1611
1612         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1613                                              lockdep_is_held(&event_mutex));
1614
1615         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1616         if (ret < 0)
1617                 goto out;
1618
1619         rcu_assign_pointer(tr->filtered_pids, pid_list);
1620
1621         list_for_each_entry(file, &tr->events, list) {
1622                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1623         }
1624
1625         if (filtered_pids) {
1626                 tracepoint_synchronize_unregister();
1627                 trace_free_pid_list(filtered_pids);
1628         } else if (pid_list) {
1629                 /*
1630                  * Register a probe that is called before all other probes
1631                  * to set ignore_pid if next or prev do not match.
1632                  * Register a probe this is called after all other probes
1633                  * to only keep ignore_pid set if next pid matches.
1634                  */
1635                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1636                                                  tr, INT_MAX);
1637                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1638                                                  tr, 0);
1639
1640                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1641                                                  tr, INT_MAX);
1642                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1643                                                  tr, 0);
1644
1645                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1646                                                      tr, INT_MAX);
1647                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1648                                                      tr, 0);
1649
1650                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1651                                                  tr, INT_MAX);
1652                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1653                                                  tr, 0);
1654         }
1655
1656         /*
1657          * Ignoring of pids is done at task switch. But we have to
1658          * check for those tasks that are currently running.
1659          * Always do this in case a pid was appended or removed.
1660          */
1661         on_each_cpu(ignore_task_cpu, tr, 1);
1662
1663  out:
1664         mutex_unlock(&event_mutex);
1665
1666         if (ret > 0)
1667                 *ppos += ret;
1668
1669         return ret;
1670 }
1671
1672 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1673 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1674 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1675 static int ftrace_event_release(struct inode *inode, struct file *file);
1676
1677 static const struct seq_operations show_event_seq_ops = {
1678         .start = t_start,
1679         .next = t_next,
1680         .show = t_show,
1681         .stop = t_stop,
1682 };
1683
1684 static const struct seq_operations show_set_event_seq_ops = {
1685         .start = s_start,
1686         .next = s_next,
1687         .show = t_show,
1688         .stop = t_stop,
1689 };
1690
1691 static const struct seq_operations show_set_pid_seq_ops = {
1692         .start = p_start,
1693         .next = p_next,
1694         .show = trace_pid_show,
1695         .stop = p_stop,
1696 };
1697
1698 static const struct file_operations ftrace_avail_fops = {
1699         .open = ftrace_event_avail_open,
1700         .read = seq_read,
1701         .llseek = seq_lseek,
1702         .release = seq_release,
1703 };
1704
1705 static const struct file_operations ftrace_set_event_fops = {
1706         .open = ftrace_event_set_open,
1707         .read = seq_read,
1708         .write = ftrace_event_write,
1709         .llseek = seq_lseek,
1710         .release = ftrace_event_release,
1711 };
1712
1713 static const struct file_operations ftrace_set_event_pid_fops = {
1714         .open = ftrace_event_set_pid_open,
1715         .read = seq_read,
1716         .write = ftrace_event_pid_write,
1717         .llseek = seq_lseek,
1718         .release = ftrace_event_release,
1719 };
1720
1721 static const struct file_operations ftrace_enable_fops = {
1722         .open = tracing_open_generic,
1723         .read = event_enable_read,
1724         .write = event_enable_write,
1725         .llseek = default_llseek,
1726 };
1727
1728 static const struct file_operations ftrace_event_format_fops = {
1729         .open = trace_format_open,
1730         .read = seq_read,
1731         .llseek = seq_lseek,
1732         .release = seq_release,
1733 };
1734
1735 static const struct file_operations ftrace_event_id_fops = {
1736         .read = event_id_read,
1737         .llseek = default_llseek,
1738 };
1739
1740 static const struct file_operations ftrace_event_filter_fops = {
1741         .open = tracing_open_generic,
1742         .read = event_filter_read,
1743         .write = event_filter_write,
1744         .llseek = default_llseek,
1745 };
1746
1747 static const struct file_operations ftrace_subsystem_filter_fops = {
1748         .open = subsystem_open,
1749         .read = subsystem_filter_read,
1750         .write = subsystem_filter_write,
1751         .llseek = default_llseek,
1752         .release = subsystem_release,
1753 };
1754
1755 static const struct file_operations ftrace_system_enable_fops = {
1756         .open = subsystem_open,
1757         .read = system_enable_read,
1758         .write = system_enable_write,
1759         .llseek = default_llseek,
1760         .release = subsystem_release,
1761 };
1762
1763 static const struct file_operations ftrace_tr_enable_fops = {
1764         .open = system_tr_open,
1765         .read = system_enable_read,
1766         .write = system_enable_write,
1767         .llseek = default_llseek,
1768         .release = subsystem_release,
1769 };
1770
1771 static const struct file_operations ftrace_show_header_fops = {
1772         .open = tracing_open_generic,
1773         .read = show_header,
1774         .llseek = default_llseek,
1775 };
1776
1777 static int
1778 ftrace_event_open(struct inode *inode, struct file *file,
1779                   const struct seq_operations *seq_ops)
1780 {
1781         struct seq_file *m;
1782         int ret;
1783
1784         ret = seq_open(file, seq_ops);
1785         if (ret < 0)
1786                 return ret;
1787         m = file->private_data;
1788         /* copy tr over to seq ops */
1789         m->private = inode->i_private;
1790
1791         return ret;
1792 }
1793
1794 static int ftrace_event_release(struct inode *inode, struct file *file)
1795 {
1796         struct trace_array *tr = inode->i_private;
1797
1798         trace_array_put(tr);
1799
1800         return seq_release(inode, file);
1801 }
1802
1803 static int
1804 ftrace_event_avail_open(struct inode *inode, struct file *file)
1805 {
1806         const struct seq_operations *seq_ops = &show_event_seq_ops;
1807
1808         return ftrace_event_open(inode, file, seq_ops);
1809 }
1810
1811 static int
1812 ftrace_event_set_open(struct inode *inode, struct file *file)
1813 {
1814         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1815         struct trace_array *tr = inode->i_private;
1816         int ret;
1817
1818         if (trace_array_get(tr) < 0)
1819                 return -ENODEV;
1820
1821         if ((file->f_mode & FMODE_WRITE) &&
1822             (file->f_flags & O_TRUNC))
1823                 ftrace_clear_events(tr);
1824
1825         ret = ftrace_event_open(inode, file, seq_ops);
1826         if (ret < 0)
1827                 trace_array_put(tr);
1828         return ret;
1829 }
1830
1831 static int
1832 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1833 {
1834         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1835         struct trace_array *tr = inode->i_private;
1836         int ret;
1837
1838         if (trace_array_get(tr) < 0)
1839                 return -ENODEV;
1840
1841         if ((file->f_mode & FMODE_WRITE) &&
1842             (file->f_flags & O_TRUNC))
1843                 ftrace_clear_event_pids(tr);
1844
1845         ret = ftrace_event_open(inode, file, seq_ops);
1846         if (ret < 0)
1847                 trace_array_put(tr);
1848         return ret;
1849 }
1850
1851 static struct event_subsystem *
1852 create_new_subsystem(const char *name)
1853 {
1854         struct event_subsystem *system;
1855
1856         /* need to create new entry */
1857         system = kmalloc(sizeof(*system), GFP_KERNEL);
1858         if (!system)
1859                 return NULL;
1860
1861         system->ref_count = 1;
1862
1863         /* Only allocate if dynamic (kprobes and modules) */
1864         system->name = kstrdup_const(name, GFP_KERNEL);
1865         if (!system->name)
1866                 goto out_free;
1867
1868         system->filter = NULL;
1869
1870         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1871         if (!system->filter)
1872                 goto out_free;
1873
1874         list_add(&system->list, &event_subsystems);
1875
1876         return system;
1877
1878  out_free:
1879         kfree_const(system->name);
1880         kfree(system);
1881         return NULL;
1882 }
1883
1884 static struct dentry *
1885 event_subsystem_dir(struct trace_array *tr, const char *name,
1886                     struct trace_event_file *file, struct dentry *parent)
1887 {
1888         struct trace_subsystem_dir *dir;
1889         struct event_subsystem *system;
1890         struct dentry *entry;
1891
1892         /* First see if we did not already create this dir */
1893         list_for_each_entry(dir, &tr->systems, list) {
1894                 system = dir->subsystem;
1895                 if (strcmp(system->name, name) == 0) {
1896                         dir->nr_events++;
1897                         file->system = dir;
1898                         return dir->entry;
1899                 }
1900         }
1901
1902         /* Now see if the system itself exists. */
1903         list_for_each_entry(system, &event_subsystems, list) {
1904                 if (strcmp(system->name, name) == 0)
1905                         break;
1906         }
1907         /* Reset system variable when not found */
1908         if (&system->list == &event_subsystems)
1909                 system = NULL;
1910
1911         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1912         if (!dir)
1913                 goto out_fail;
1914
1915         if (!system) {
1916                 system = create_new_subsystem(name);
1917                 if (!system)
1918                         goto out_free;
1919         } else
1920                 __get_system(system);
1921
1922         dir->entry = tracefs_create_dir(name, parent);
1923         if (!dir->entry) {
1924                 pr_warn("Failed to create system directory %s\n", name);
1925                 __put_system(system);
1926                 goto out_free;
1927         }
1928
1929         dir->tr = tr;
1930         dir->ref_count = 1;
1931         dir->nr_events = 1;
1932         dir->subsystem = system;
1933         file->system = dir;
1934
1935         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1936                                     &ftrace_subsystem_filter_fops);
1937         if (!entry) {
1938                 kfree(system->filter);
1939                 system->filter = NULL;
1940                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1941         }
1942
1943         trace_create_file("enable", 0644, dir->entry, dir,
1944                           &ftrace_system_enable_fops);
1945
1946         list_add(&dir->list, &tr->systems);
1947
1948         return dir->entry;
1949
1950  out_free:
1951         kfree(dir);
1952  out_fail:
1953         /* Only print this message if failed on memory allocation */
1954         if (!dir || !system)
1955                 pr_warn("No memory to create event subsystem %s\n", name);
1956         return NULL;
1957 }
1958
1959 static int
1960 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1961 {
1962         struct trace_event_call *call = file->event_call;
1963         struct trace_array *tr = file->tr;
1964         struct list_head *head;
1965         struct dentry *d_events;
1966         const char *name;
1967         int ret;
1968
1969         /*
1970          * If the trace point header did not define TRACE_SYSTEM
1971          * then the system would be called "TRACE_SYSTEM".
1972          */
1973         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1974                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1975                 if (!d_events)
1976                         return -ENOMEM;
1977         } else
1978                 d_events = parent;
1979
1980         name = trace_event_name(call);
1981         file->dir = tracefs_create_dir(name, d_events);
1982         if (!file->dir) {
1983                 pr_warn("Could not create tracefs '%s' directory\n", name);
1984                 return -1;
1985         }
1986
1987         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1988                 trace_create_file("enable", 0644, file->dir, file,
1989                                   &ftrace_enable_fops);
1990
1991 #ifdef CONFIG_PERF_EVENTS
1992         if (call->event.type && call->class->reg)
1993                 trace_create_file("id", 0444, file->dir,
1994                                   (void *)(long)call->event.type,
1995                                   &ftrace_event_id_fops);
1996 #endif
1997
1998         /*
1999          * Other events may have the same class. Only update
2000          * the fields if they are not already defined.
2001          */
2002         head = trace_get_fields(call);
2003         if (list_empty(head)) {
2004                 ret = call->class->define_fields(call);
2005                 if (ret < 0) {
2006                         pr_warn("Could not initialize trace point events/%s\n",
2007                                 name);
2008                         return -1;
2009                 }
2010         }
2011
2012         /*
2013          * Only event directories that can be enabled should have
2014          * triggers or filters.
2015          */
2016         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2017                 trace_create_file("filter", 0644, file->dir, file,
2018                                   &ftrace_event_filter_fops);
2019
2020                 trace_create_file("trigger", 0644, file->dir, file,
2021                                   &event_trigger_fops);
2022         }
2023
2024 #ifdef CONFIG_HIST_TRIGGERS
2025         trace_create_file("hist", 0444, file->dir, file,
2026                           &event_hist_fops);
2027 #endif
2028         trace_create_file("format", 0444, file->dir, call,
2029                           &ftrace_event_format_fops);
2030
2031         return 0;
2032 }
2033
2034 static void remove_event_from_tracers(struct trace_event_call *call)
2035 {
2036         struct trace_event_file *file;
2037         struct trace_array *tr;
2038
2039         do_for_each_event_file_safe(tr, file) {
2040                 if (file->event_call != call)
2041                         continue;
2042
2043                 remove_event_file_dir(file);
2044                 /*
2045                  * The do_for_each_event_file_safe() is
2046                  * a double loop. After finding the call for this
2047                  * trace_array, we use break to jump to the next
2048                  * trace_array.
2049                  */
2050                 break;
2051         } while_for_each_event_file();
2052 }
2053
2054 static void event_remove(struct trace_event_call *call)
2055 {
2056         struct trace_array *tr;
2057         struct trace_event_file *file;
2058
2059         do_for_each_event_file(tr, file) {
2060                 if (file->event_call != call)
2061                         continue;
2062
2063                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2064                         tr->clear_trace = true;
2065
2066                 ftrace_event_enable_disable(file, 0);
2067                 /*
2068                  * The do_for_each_event_file() is
2069                  * a double loop. After finding the call for this
2070                  * trace_array, we use break to jump to the next
2071                  * trace_array.
2072                  */
2073                 break;
2074         } while_for_each_event_file();
2075
2076         if (call->event.funcs)
2077                 __unregister_trace_event(&call->event);
2078         remove_event_from_tracers(call);
2079         list_del(&call->list);
2080 }
2081
2082 static int event_init(struct trace_event_call *call)
2083 {
2084         int ret = 0;
2085         const char *name;
2086
2087         name = trace_event_name(call);
2088         if (WARN_ON(!name))
2089                 return -EINVAL;
2090
2091         if (call->class->raw_init) {
2092                 ret = call->class->raw_init(call);
2093                 if (ret < 0 && ret != -ENOSYS)
2094                         pr_warn("Could not initialize trace events/%s\n", name);
2095         }
2096
2097         return ret;
2098 }
2099
2100 static int
2101 __register_event(struct trace_event_call *call, struct module *mod)
2102 {
2103         int ret;
2104
2105         ret = event_init(call);
2106         if (ret < 0)
2107                 return ret;
2108
2109         list_add(&call->list, &ftrace_events);
2110         call->mod = mod;
2111
2112         return 0;
2113 }
2114
2115 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2116 {
2117         int rlen;
2118         int elen;
2119
2120         /* Find the length of the eval value as a string */
2121         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2122         /* Make sure there's enough room to replace the string with the value */
2123         if (len < elen)
2124                 return NULL;
2125
2126         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2127
2128         /* Get the rest of the string of ptr */
2129         rlen = strlen(ptr + len);
2130         memmove(ptr + elen, ptr + len, rlen);
2131         /* Make sure we end the new string */
2132         ptr[elen + rlen] = 0;
2133
2134         return ptr + elen;
2135 }
2136
2137 static void update_event_printk(struct trace_event_call *call,
2138                                 struct trace_eval_map *map)
2139 {
2140         char *ptr;
2141         int quote = 0;
2142         int len = strlen(map->eval_string);
2143
2144         for (ptr = call->print_fmt; *ptr; ptr++) {
2145                 if (*ptr == '\\') {
2146                         ptr++;
2147                         /* paranoid */
2148                         if (!*ptr)
2149                                 break;
2150                         continue;
2151                 }
2152                 if (*ptr == '"') {
2153                         quote ^= 1;
2154                         continue;
2155                 }
2156                 if (quote)
2157                         continue;
2158                 if (isdigit(*ptr)) {
2159                         /* skip numbers */
2160                         do {
2161                                 ptr++;
2162                                 /* Check for alpha chars like ULL */
2163                         } while (isalnum(*ptr));
2164                         if (!*ptr)
2165                                 break;
2166                         /*
2167                          * A number must have some kind of delimiter after
2168                          * it, and we can ignore that too.
2169                          */
2170                         continue;
2171                 }
2172                 if (isalpha(*ptr) || *ptr == '_') {
2173                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2174                             !isalnum(ptr[len]) && ptr[len] != '_') {
2175                                 ptr = eval_replace(ptr, map, len);
2176                                 /* enum/sizeof string smaller than value */
2177                                 if (WARN_ON_ONCE(!ptr))
2178                                         return;
2179                                 /*
2180                                  * No need to decrement here, as eval_replace()
2181                                  * returns the pointer to the character passed
2182                                  * the eval, and two evals can not be placed
2183                                  * back to back without something in between.
2184                                  * We can skip that something in between.
2185                                  */
2186                                 continue;
2187                         }
2188                 skip_more:
2189                         do {
2190                                 ptr++;
2191                         } while (isalnum(*ptr) || *ptr == '_');
2192                         if (!*ptr)
2193                                 break;
2194                         /*
2195                          * If what comes after this variable is a '.' or
2196                          * '->' then we can continue to ignore that string.
2197                          */
2198                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2199                                 ptr += *ptr == '.' ? 1 : 2;
2200                                 if (!*ptr)
2201                                         break;
2202                                 goto skip_more;
2203                         }
2204                         /*
2205                          * Once again, we can skip the delimiter that came
2206                          * after the string.
2207                          */
2208                         continue;
2209                 }
2210         }
2211 }
2212
2213 void trace_event_eval_update(struct trace_eval_map **map, int len)
2214 {
2215         struct trace_event_call *call, *p;
2216         const char *last_system = NULL;
2217         bool first = false;
2218         int last_i;
2219         int i;
2220
2221         down_write(&trace_event_sem);
2222         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2223                 /* events are usually grouped together with systems */
2224                 if (!last_system || call->class->system != last_system) {
2225                         first = true;
2226                         last_i = 0;
2227                         last_system = call->class->system;
2228                 }
2229
2230                 /*
2231                  * Since calls are grouped by systems, the likelyhood that the
2232                  * next call in the iteration belongs to the same system as the
2233                  * previous call is high. As an optimization, we skip seaching
2234                  * for a map[] that matches the call's system if the last call
2235                  * was from the same system. That's what last_i is for. If the
2236                  * call has the same system as the previous call, then last_i
2237                  * will be the index of the first map[] that has a matching
2238                  * system.
2239                  */
2240                 for (i = last_i; i < len; i++) {
2241                         if (call->class->system == map[i]->system) {
2242                                 /* Save the first system if need be */
2243                                 if (first) {
2244                                         last_i = i;
2245                                         first = false;
2246                                 }
2247                                 update_event_printk(call, map[i]);
2248                         }
2249                 }
2250         }
2251         up_write(&trace_event_sem);
2252 }
2253
2254 static struct trace_event_file *
2255 trace_create_new_event(struct trace_event_call *call,
2256                        struct trace_array *tr)
2257 {
2258         struct trace_event_file *file;
2259
2260         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2261         if (!file)
2262                 return NULL;
2263
2264         file->event_call = call;
2265         file->tr = tr;
2266         atomic_set(&file->sm_ref, 0);
2267         atomic_set(&file->tm_ref, 0);
2268         INIT_LIST_HEAD(&file->triggers);
2269         list_add(&file->list, &tr->events);
2270
2271         return file;
2272 }
2273
2274 /* Add an event to a trace directory */
2275 static int
2276 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2277 {
2278         struct trace_event_file *file;
2279
2280         file = trace_create_new_event(call, tr);
2281         if (!file)
2282                 return -ENOMEM;
2283
2284         return event_create_dir(tr->event_dir, file);
2285 }
2286
2287 /*
2288  * Just create a decriptor for early init. A descriptor is required
2289  * for enabling events at boot. We want to enable events before
2290  * the filesystem is initialized.
2291  */
2292 static __init int
2293 __trace_early_add_new_event(struct trace_event_call *call,
2294                             struct trace_array *tr)
2295 {
2296         struct trace_event_file *file;
2297
2298         file = trace_create_new_event(call, tr);
2299         if (!file)
2300                 return -ENOMEM;
2301
2302         return 0;
2303 }
2304
2305 struct ftrace_module_file_ops;
2306 static void __add_event_to_tracers(struct trace_event_call *call);
2307
2308 /* Add an additional event_call dynamically */
2309 int trace_add_event_call(struct trace_event_call *call)
2310 {
2311         int ret;
2312         lockdep_assert_held(&event_mutex);
2313
2314         mutex_lock(&trace_types_lock);
2315
2316         ret = __register_event(call, NULL);
2317         if (ret >= 0)
2318                 __add_event_to_tracers(call);
2319
2320         mutex_unlock(&trace_types_lock);
2321         return ret;
2322 }
2323
2324 /*
2325  * Must be called under locking of trace_types_lock, event_mutex and
2326  * trace_event_sem.
2327  */
2328 static void __trace_remove_event_call(struct trace_event_call *call)
2329 {
2330         event_remove(call);
2331         trace_destroy_fields(call);
2332         free_event_filter(call->filter);
2333         call->filter = NULL;
2334 }
2335
2336 static int probe_remove_event_call(struct trace_event_call *call)
2337 {
2338         struct trace_array *tr;
2339         struct trace_event_file *file;
2340
2341 #ifdef CONFIG_PERF_EVENTS
2342         if (call->perf_refcount)
2343                 return -EBUSY;
2344 #endif
2345         do_for_each_event_file(tr, file) {
2346                 if (file->event_call != call)
2347                         continue;
2348                 /*
2349                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2350                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2351                  * TRACE_REG_UNREGISTER.
2352                  */
2353                 if (file->flags & EVENT_FILE_FL_ENABLED)
2354                         return -EBUSY;
2355                 /*
2356                  * The do_for_each_event_file_safe() is
2357                  * a double loop. After finding the call for this
2358                  * trace_array, we use break to jump to the next
2359                  * trace_array.
2360                  */
2361                 break;
2362         } while_for_each_event_file();
2363
2364         __trace_remove_event_call(call);
2365
2366         return 0;
2367 }
2368
2369 /* Remove an event_call */
2370 int trace_remove_event_call(struct trace_event_call *call)
2371 {
2372         int ret;
2373
2374         lockdep_assert_held(&event_mutex);
2375
2376         mutex_lock(&trace_types_lock);
2377         down_write(&trace_event_sem);
2378         ret = probe_remove_event_call(call);
2379         up_write(&trace_event_sem);
2380         mutex_unlock(&trace_types_lock);
2381
2382         return ret;
2383 }
2384
2385 #define for_each_event(event, start, end)                       \
2386         for (event = start;                                     \
2387              (unsigned long)event < (unsigned long)end;         \
2388              event++)
2389
2390 #ifdef CONFIG_MODULES
2391
2392 static void trace_module_add_events(struct module *mod)
2393 {
2394         struct trace_event_call **call, **start, **end;
2395
2396         if (!mod->num_trace_events)
2397                 return;
2398
2399         /* Don't add infrastructure for mods without tracepoints */
2400         if (trace_module_has_bad_taint(mod)) {
2401                 pr_err("%s: module has bad taint, not creating trace events\n",
2402                        mod->name);
2403                 return;
2404         }
2405
2406         start = mod->trace_events;
2407         end = mod->trace_events + mod->num_trace_events;
2408
2409         for_each_event(call, start, end) {
2410                 __register_event(*call, mod);
2411                 __add_event_to_tracers(*call);
2412         }
2413 }
2414
2415 static void trace_module_remove_events(struct module *mod)
2416 {
2417         struct trace_event_call *call, *p;
2418
2419         down_write(&trace_event_sem);
2420         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2421                 if (call->mod == mod)
2422                         __trace_remove_event_call(call);
2423         }
2424         up_write(&trace_event_sem);
2425
2426         /*
2427          * It is safest to reset the ring buffer if the module being unloaded
2428          * registered any events that were used. The only worry is if
2429          * a new module gets loaded, and takes on the same id as the events
2430          * of this module. When printing out the buffer, traced events left
2431          * over from this module may be passed to the new module events and
2432          * unexpected results may occur.
2433          */
2434         tracing_reset_all_online_cpus();
2435 }
2436
2437 static int trace_module_notify(struct notifier_block *self,
2438                                unsigned long val, void *data)
2439 {
2440         struct module *mod = data;
2441
2442         mutex_lock(&event_mutex);
2443         mutex_lock(&trace_types_lock);
2444         switch (val) {
2445         case MODULE_STATE_COMING:
2446                 trace_module_add_events(mod);
2447                 break;
2448         case MODULE_STATE_GOING:
2449                 trace_module_remove_events(mod);
2450                 break;
2451         }
2452         mutex_unlock(&trace_types_lock);
2453         mutex_unlock(&event_mutex);
2454
2455         return 0;
2456 }
2457
2458 static struct notifier_block trace_module_nb = {
2459         .notifier_call = trace_module_notify,
2460         .priority = 1, /* higher than trace.c module notify */
2461 };
2462 #endif /* CONFIG_MODULES */
2463
2464 /* Create a new event directory structure for a trace directory. */
2465 static void
2466 __trace_add_event_dirs(struct trace_array *tr)
2467 {
2468         struct trace_event_call *call;
2469         int ret;
2470
2471         list_for_each_entry(call, &ftrace_events, list) {
2472                 ret = __trace_add_new_event(call, tr);
2473                 if (ret < 0)
2474                         pr_warn("Could not create directory for event %s\n",
2475                                 trace_event_name(call));
2476         }
2477 }
2478
2479 /* Returns any file that matches the system and event */
2480 struct trace_event_file *
2481 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2482 {
2483         struct trace_event_file *file;
2484         struct trace_event_call *call;
2485         const char *name;
2486
2487         list_for_each_entry(file, &tr->events, list) {
2488
2489                 call = file->event_call;
2490                 name = trace_event_name(call);
2491
2492                 if (!name || !call->class)
2493                         continue;
2494
2495                 if (strcmp(event, name) == 0 &&
2496                     strcmp(system, call->class->system) == 0)
2497                         return file;
2498         }
2499         return NULL;
2500 }
2501
2502 /* Returns valid trace event files that match system and event */
2503 struct trace_event_file *
2504 find_event_file(struct trace_array *tr, const char *system, const char *event)
2505 {
2506         struct trace_event_file *file;
2507
2508         file = __find_event_file(tr, system, event);
2509         if (!file || !file->event_call->class->reg ||
2510             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2511                 return NULL;
2512
2513         return file;
2514 }
2515
2516 #ifdef CONFIG_DYNAMIC_FTRACE
2517
2518 /* Avoid typos */
2519 #define ENABLE_EVENT_STR        "enable_event"
2520 #define DISABLE_EVENT_STR       "disable_event"
2521
2522 struct event_probe_data {
2523         struct trace_event_file *file;
2524         unsigned long                   count;
2525         int                             ref;
2526         bool                            enable;
2527 };
2528
2529 static void update_event_probe(struct event_probe_data *data)
2530 {
2531         if (data->enable)
2532                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2533         else
2534                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2535 }
2536
2537 static void
2538 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2539                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2540                    void *data)
2541 {
2542         struct ftrace_func_mapper *mapper = data;
2543         struct event_probe_data *edata;
2544         void **pdata;
2545
2546         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2547         if (!pdata || !*pdata)
2548                 return;
2549
2550         edata = *pdata;
2551         update_event_probe(edata);
2552 }
2553
2554 static void
2555 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2556                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2557                          void *data)
2558 {
2559         struct ftrace_func_mapper *mapper = data;
2560         struct event_probe_data *edata;
2561         void **pdata;
2562
2563         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2564         if (!pdata || !*pdata)
2565                 return;
2566
2567         edata = *pdata;
2568
2569         if (!edata->count)
2570                 return;
2571
2572         /* Skip if the event is in a state we want to switch to */
2573         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2574                 return;
2575
2576         if (edata->count != -1)
2577                 (edata->count)--;
2578
2579         update_event_probe(edata);
2580 }
2581
2582 static int
2583 event_enable_print(struct seq_file *m, unsigned long ip,
2584                    struct ftrace_probe_ops *ops, void *data)
2585 {
2586         struct ftrace_func_mapper *mapper = data;
2587         struct event_probe_data *edata;
2588         void **pdata;
2589
2590         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2591
2592         if (WARN_ON_ONCE(!pdata || !*pdata))
2593                 return 0;
2594
2595         edata = *pdata;
2596
2597         seq_printf(m, "%ps:", (void *)ip);
2598
2599         seq_printf(m, "%s:%s:%s",
2600                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2601                    edata->file->event_call->class->system,
2602                    trace_event_name(edata->file->event_call));
2603
2604         if (edata->count == -1)
2605                 seq_puts(m, ":unlimited\n");
2606         else
2607                 seq_printf(m, ":count=%ld\n", edata->count);
2608
2609         return 0;
2610 }
2611
2612 static int
2613 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2614                   unsigned long ip, void *init_data, void **data)
2615 {
2616         struct ftrace_func_mapper *mapper = *data;
2617         struct event_probe_data *edata = init_data;
2618         int ret;
2619
2620         if (!mapper) {
2621                 mapper = allocate_ftrace_func_mapper();
2622                 if (!mapper)
2623                         return -ENODEV;
2624                 *data = mapper;
2625         }
2626
2627         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2628         if (ret < 0)
2629                 return ret;
2630
2631         edata->ref++;
2632
2633         return 0;
2634 }
2635
2636 static int free_probe_data(void *data)
2637 {
2638         struct event_probe_data *edata = data;
2639
2640         edata->ref--;
2641         if (!edata->ref) {
2642                 /* Remove the SOFT_MODE flag */
2643                 __ftrace_event_enable_disable(edata->file, 0, 1);
2644                 module_put(edata->file->event_call->mod);
2645                 kfree(edata);
2646         }
2647         return 0;
2648 }
2649
2650 static void
2651 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2652                   unsigned long ip, void *data)
2653 {
2654         struct ftrace_func_mapper *mapper = data;
2655         struct event_probe_data *edata;
2656
2657         if (!ip) {
2658                 if (!mapper)
2659                         return;
2660                 free_ftrace_func_mapper(mapper, free_probe_data);
2661                 return;
2662         }
2663
2664         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2665
2666         if (WARN_ON_ONCE(!edata))
2667                 return;
2668
2669         if (WARN_ON_ONCE(edata->ref <= 0))
2670                 return;
2671
2672         free_probe_data(edata);
2673 }
2674
2675 static struct ftrace_probe_ops event_enable_probe_ops = {
2676         .func                   = event_enable_probe,
2677         .print                  = event_enable_print,
2678         .init                   = event_enable_init,
2679         .free                   = event_enable_free,
2680 };
2681
2682 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2683         .func                   = event_enable_count_probe,
2684         .print                  = event_enable_print,
2685         .init                   = event_enable_init,
2686         .free                   = event_enable_free,
2687 };
2688
2689 static struct ftrace_probe_ops event_disable_probe_ops = {
2690         .func                   = event_enable_probe,
2691         .print                  = event_enable_print,
2692         .init                   = event_enable_init,
2693         .free                   = event_enable_free,
2694 };
2695
2696 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2697         .func                   = event_enable_count_probe,
2698         .print                  = event_enable_print,
2699         .init                   = event_enable_init,
2700         .free                   = event_enable_free,
2701 };
2702
2703 static int
2704 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2705                   char *glob, char *cmd, char *param, int enabled)
2706 {
2707         struct trace_event_file *file;
2708         struct ftrace_probe_ops *ops;
2709         struct event_probe_data *data;
2710         const char *system;
2711         const char *event;
2712         char *number;
2713         bool enable;
2714         int ret;
2715
2716         if (!tr)
2717                 return -ENODEV;
2718
2719         /* hash funcs only work with set_ftrace_filter */
2720         if (!enabled || !param)
2721                 return -EINVAL;
2722
2723         system = strsep(&param, ":");
2724         if (!param)
2725                 return -EINVAL;
2726
2727         event = strsep(&param, ":");
2728
2729         mutex_lock(&event_mutex);
2730
2731         ret = -EINVAL;
2732         file = find_event_file(tr, system, event);
2733         if (!file)
2734                 goto out;
2735
2736         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2737
2738         if (enable)
2739                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2740         else
2741                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2742
2743         if (glob[0] == '!') {
2744                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2745                 goto out;
2746         }
2747
2748         ret = -ENOMEM;
2749
2750         data = kzalloc(sizeof(*data), GFP_KERNEL);
2751         if (!data)
2752                 goto out;
2753
2754         data->enable = enable;
2755         data->count = -1;
2756         data->file = file;
2757
2758         if (!param)
2759                 goto out_reg;
2760
2761         number = strsep(&param, ":");
2762
2763         ret = -EINVAL;
2764         if (!strlen(number))
2765                 goto out_free;
2766
2767         /*
2768          * We use the callback data field (which is a pointer)
2769          * as our counter.
2770          */
2771         ret = kstrtoul(number, 0, &data->count);
2772         if (ret)
2773                 goto out_free;
2774
2775  out_reg:
2776         /* Don't let event modules unload while probe registered */
2777         ret = try_module_get(file->event_call->mod);
2778         if (!ret) {
2779                 ret = -EBUSY;
2780                 goto out_free;
2781         }
2782
2783         ret = __ftrace_event_enable_disable(file, 1, 1);
2784         if (ret < 0)
2785                 goto out_put;
2786
2787         ret = register_ftrace_function_probe(glob, tr, ops, data);
2788         /*
2789          * The above returns on success the # of functions enabled,
2790          * but if it didn't find any functions it returns zero.
2791          * Consider no functions a failure too.
2792          */
2793         if (!ret) {
2794                 ret = -ENOENT;
2795                 goto out_disable;
2796         } else if (ret < 0)
2797                 goto out_disable;
2798         /* Just return zero, not the number of enabled functions */
2799         ret = 0;
2800  out:
2801         mutex_unlock(&event_mutex);
2802         return ret;
2803
2804  out_disable:
2805         __ftrace_event_enable_disable(file, 0, 1);
2806  out_put:
2807         module_put(file->event_call->mod);
2808  out_free:
2809         kfree(data);
2810         goto out;
2811 }
2812
2813 static struct ftrace_func_command event_enable_cmd = {
2814         .name                   = ENABLE_EVENT_STR,
2815         .func                   = event_enable_func,
2816 };
2817
2818 static struct ftrace_func_command event_disable_cmd = {
2819         .name                   = DISABLE_EVENT_STR,
2820         .func                   = event_enable_func,
2821 };
2822
2823 static __init int register_event_cmds(void)
2824 {
2825         int ret;
2826
2827         ret = register_ftrace_command(&event_enable_cmd);
2828         if (WARN_ON(ret < 0))
2829                 return ret;
2830         ret = register_ftrace_command(&event_disable_cmd);
2831         if (WARN_ON(ret < 0))
2832                 unregister_ftrace_command(&event_enable_cmd);
2833         return ret;
2834 }
2835 #else
2836 static inline int register_event_cmds(void) { return 0; }
2837 #endif /* CONFIG_DYNAMIC_FTRACE */
2838
2839 /*
2840  * The top level array has already had its trace_event_file
2841  * descriptors created in order to allow for early events to
2842  * be recorded. This function is called after the tracefs has been
2843  * initialized, and we now have to create the files associated
2844  * to the events.
2845  */
2846 static __init void
2847 __trace_early_add_event_dirs(struct trace_array *tr)
2848 {
2849         struct trace_event_file *file;
2850         int ret;
2851
2852
2853         list_for_each_entry(file, &tr->events, list) {
2854                 ret = event_create_dir(tr->event_dir, file);
2855                 if (ret < 0)
2856                         pr_warn("Could not create directory for event %s\n",
2857                                 trace_event_name(file->event_call));
2858         }
2859 }
2860
2861 /*
2862  * For early boot up, the top trace array requires to have
2863  * a list of events that can be enabled. This must be done before
2864  * the filesystem is set up in order to allow events to be traced
2865  * early.
2866  */
2867 static __init void
2868 __trace_early_add_events(struct trace_array *tr)
2869 {
2870         struct trace_event_call *call;
2871         int ret;
2872
2873         list_for_each_entry(call, &ftrace_events, list) {
2874                 /* Early boot up should not have any modules loaded */
2875                 if (WARN_ON_ONCE(call->mod))
2876                         continue;
2877
2878                 ret = __trace_early_add_new_event(call, tr);
2879                 if (ret < 0)
2880                         pr_warn("Could not create early event %s\n",
2881                                 trace_event_name(call));
2882         }
2883 }
2884
2885 /* Remove the event directory structure for a trace directory. */
2886 static void
2887 __trace_remove_event_dirs(struct trace_array *tr)
2888 {
2889         struct trace_event_file *file, *next;
2890
2891         list_for_each_entry_safe(file, next, &tr->events, list)
2892                 remove_event_file_dir(file);
2893 }
2894
2895 static void __add_event_to_tracers(struct trace_event_call *call)
2896 {
2897         struct trace_array *tr;
2898
2899         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2900                 __trace_add_new_event(call, tr);
2901 }
2902
2903 extern struct trace_event_call *__start_ftrace_events[];
2904 extern struct trace_event_call *__stop_ftrace_events[];
2905
2906 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2907
2908 static __init int setup_trace_event(char *str)
2909 {
2910         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2911         ring_buffer_expanded = true;
2912         tracing_selftest_disabled = true;
2913
2914         return 1;
2915 }
2916 __setup("trace_event=", setup_trace_event);
2917
2918 /* Expects to have event_mutex held when called */
2919 static int
2920 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2921 {
2922         struct dentry *d_events;
2923         struct dentry *entry;
2924
2925         entry = tracefs_create_file("set_event", 0644, parent,
2926                                     tr, &ftrace_set_event_fops);
2927         if (!entry) {
2928                 pr_warn("Could not create tracefs 'set_event' entry\n");
2929                 return -ENOMEM;
2930         }
2931
2932         d_events = tracefs_create_dir("events", parent);
2933         if (!d_events) {
2934                 pr_warn("Could not create tracefs 'events' directory\n");
2935                 return -ENOMEM;
2936         }
2937
2938         entry = trace_create_file("enable", 0644, d_events,
2939                                   tr, &ftrace_tr_enable_fops);
2940         if (!entry) {
2941                 pr_warn("Could not create tracefs 'enable' entry\n");
2942                 return -ENOMEM;
2943         }
2944
2945         /* There are not as crucial, just warn if they are not created */
2946
2947         entry = tracefs_create_file("set_event_pid", 0644, parent,
2948                                     tr, &ftrace_set_event_pid_fops);
2949         if (!entry)
2950                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2951
2952         /* ring buffer internal formats */
2953         entry = trace_create_file("header_page", 0444, d_events,
2954                                   ring_buffer_print_page_header,
2955                                   &ftrace_show_header_fops);
2956         if (!entry)
2957                 pr_warn("Could not create tracefs 'header_page' entry\n");
2958
2959         entry = trace_create_file("header_event", 0444, d_events,
2960                                   ring_buffer_print_entry_header,
2961                                   &ftrace_show_header_fops);
2962         if (!entry)
2963                 pr_warn("Could not create tracefs 'header_event' entry\n");
2964
2965         tr->event_dir = d_events;
2966
2967         return 0;
2968 }
2969
2970 /**
2971  * event_trace_add_tracer - add a instance of a trace_array to events
2972  * @parent: The parent dentry to place the files/directories for events in
2973  * @tr: The trace array associated with these events
2974  *
2975  * When a new instance is created, it needs to set up its events
2976  * directory, as well as other files associated with events. It also
2977  * creates the event hierachry in the @parent/events directory.
2978  *
2979  * Returns 0 on success.
2980  *
2981  * Must be called with event_mutex held.
2982  */
2983 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2984 {
2985         int ret;
2986
2987         lockdep_assert_held(&event_mutex);
2988
2989         ret = create_event_toplevel_files(parent, tr);
2990         if (ret)
2991                 goto out;
2992
2993         down_write(&trace_event_sem);
2994         __trace_add_event_dirs(tr);
2995         up_write(&trace_event_sem);
2996
2997  out:
2998         return ret;
2999 }
3000
3001 /*
3002  * The top trace array already had its file descriptors created.
3003  * Now the files themselves need to be created.
3004  */
3005 static __init int
3006 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3007 {
3008         int ret;
3009
3010         mutex_lock(&event_mutex);
3011
3012         ret = create_event_toplevel_files(parent, tr);
3013         if (ret)
3014                 goto out_unlock;
3015
3016         down_write(&trace_event_sem);
3017         __trace_early_add_event_dirs(tr);
3018         up_write(&trace_event_sem);
3019
3020  out_unlock:
3021         mutex_unlock(&event_mutex);
3022
3023         return ret;
3024 }
3025
3026 /* Must be called with event_mutex held */
3027 int event_trace_del_tracer(struct trace_array *tr)
3028 {
3029         lockdep_assert_held(&event_mutex);
3030
3031         /* Disable any event triggers and associated soft-disabled events */
3032         clear_event_triggers(tr);
3033
3034         /* Clear the pid list */
3035         __ftrace_clear_event_pids(tr);
3036
3037         /* Disable any running events */
3038         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3039
3040         /* Make sure no more events are being executed */
3041         tracepoint_synchronize_unregister();
3042
3043         down_write(&trace_event_sem);
3044         __trace_remove_event_dirs(tr);
3045         tracefs_remove_recursive(tr->event_dir);
3046         up_write(&trace_event_sem);
3047
3048         tr->event_dir = NULL;
3049
3050         return 0;
3051 }
3052
3053 static __init int event_trace_memsetup(void)
3054 {
3055         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3056         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3057         return 0;
3058 }
3059
3060 static __init void
3061 early_enable_events(struct trace_array *tr, bool disable_first)
3062 {
3063         char *buf = bootup_event_buf;
3064         char *token;
3065         int ret;
3066
3067         while (true) {
3068                 token = strsep(&buf, ",");
3069
3070                 if (!token)
3071                         break;
3072
3073                 if (*token) {
3074                         /* Restarting syscalls requires that we stop them first */
3075                         if (disable_first)
3076                                 ftrace_set_clr_event(tr, token, 0);
3077
3078                         ret = ftrace_set_clr_event(tr, token, 1);
3079                         if (ret)
3080                                 pr_warn("Failed to enable trace event: %s\n", token);
3081                 }
3082
3083                 /* Put back the comma to allow this to be called again */
3084                 if (buf)
3085                         *(buf - 1) = ',';
3086         }
3087 }
3088
3089 static __init int event_trace_enable(void)
3090 {
3091         struct trace_array *tr = top_trace_array();
3092         struct trace_event_call **iter, *call;
3093         int ret;
3094
3095         if (!tr)
3096                 return -ENODEV;
3097
3098         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3099
3100                 call = *iter;
3101                 ret = event_init(call);
3102                 if (!ret)
3103                         list_add(&call->list, &ftrace_events);
3104         }
3105
3106         /*
3107          * We need the top trace array to have a working set of trace
3108          * points at early init, before the debug files and directories
3109          * are created. Create the file entries now, and attach them
3110          * to the actual file dentries later.
3111          */
3112         __trace_early_add_events(tr);
3113
3114         early_enable_events(tr, false);
3115
3116         trace_printk_start_comm();
3117
3118         register_event_cmds();
3119
3120         register_trigger_cmds();
3121
3122         return 0;
3123 }
3124
3125 /*
3126  * event_trace_enable() is called from trace_event_init() first to
3127  * initialize events and perhaps start any events that are on the
3128  * command line. Unfortunately, there are some events that will not
3129  * start this early, like the system call tracepoints that need
3130  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3131  * is called before pid 1 starts, and this flag is never set, making
3132  * the syscall tracepoint never get reached, but the event is enabled
3133  * regardless (and not doing anything).
3134  */
3135 static __init int event_trace_enable_again(void)
3136 {
3137         struct trace_array *tr;
3138
3139         tr = top_trace_array();
3140         if (!tr)
3141                 return -ENODEV;
3142
3143         early_enable_events(tr, true);
3144
3145         return 0;
3146 }
3147
3148 early_initcall(event_trace_enable_again);
3149
3150 __init int event_trace_init(void)
3151 {
3152         struct trace_array *tr;
3153         struct dentry *d_tracer;
3154         struct dentry *entry;
3155         int ret;
3156
3157         tr = top_trace_array();
3158         if (!tr)
3159                 return -ENODEV;
3160
3161         d_tracer = tracing_init_dentry();
3162         if (IS_ERR(d_tracer))
3163                 return 0;
3164
3165         entry = tracefs_create_file("available_events", 0444, d_tracer,
3166                                     tr, &ftrace_avail_fops);
3167         if (!entry)
3168                 pr_warn("Could not create tracefs 'available_events' entry\n");
3169
3170         if (trace_define_generic_fields())
3171                 pr_warn("tracing: Failed to allocated generic fields");
3172
3173         if (trace_define_common_fields())
3174                 pr_warn("tracing: Failed to allocate common fields");
3175
3176         ret = early_event_add_tracer(d_tracer, tr);
3177         if (ret)
3178                 return ret;
3179
3180 #ifdef CONFIG_MODULES
3181         ret = register_module_notifier(&trace_module_nb);
3182         if (ret)
3183                 pr_warn("Failed to register trace events module notifier\n");
3184 #endif
3185         return 0;
3186 }
3187
3188 void __init trace_event_init(void)
3189 {
3190         event_trace_memsetup();
3191         init_ftrace_syscalls();
3192         event_trace_enable();
3193 }
3194
3195 #ifdef CONFIG_FTRACE_STARTUP_TEST
3196
3197 static DEFINE_SPINLOCK(test_spinlock);
3198 static DEFINE_SPINLOCK(test_spinlock_irq);
3199 static DEFINE_MUTEX(test_mutex);
3200
3201 static __init void test_work(struct work_struct *dummy)
3202 {
3203         spin_lock(&test_spinlock);
3204         spin_lock_irq(&test_spinlock_irq);
3205         udelay(1);
3206         spin_unlock_irq(&test_spinlock_irq);
3207         spin_unlock(&test_spinlock);
3208
3209         mutex_lock(&test_mutex);
3210         msleep(1);
3211         mutex_unlock(&test_mutex);
3212 }
3213
3214 static __init int event_test_thread(void *unused)
3215 {
3216         void *test_malloc;
3217
3218         test_malloc = kmalloc(1234, GFP_KERNEL);
3219         if (!test_malloc)
3220                 pr_info("failed to kmalloc\n");
3221
3222         schedule_on_each_cpu(test_work);
3223
3224         kfree(test_malloc);
3225
3226         set_current_state(TASK_INTERRUPTIBLE);
3227         while (!kthread_should_stop()) {
3228                 schedule();
3229                 set_current_state(TASK_INTERRUPTIBLE);
3230         }
3231         __set_current_state(TASK_RUNNING);
3232
3233         return 0;
3234 }
3235
3236 /*
3237  * Do various things that may trigger events.
3238  */
3239 static __init void event_test_stuff(void)
3240 {
3241         struct task_struct *test_thread;
3242
3243         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3244         msleep(1);
3245         kthread_stop(test_thread);
3246 }
3247
3248 /*
3249  * For every trace event defined, we will test each trace point separately,
3250  * and then by groups, and finally all trace points.
3251  */
3252 static __init void event_trace_self_tests(void)
3253 {
3254         struct trace_subsystem_dir *dir;
3255         struct trace_event_file *file;
3256         struct trace_event_call *call;
3257         struct event_subsystem *system;
3258         struct trace_array *tr;
3259         int ret;
3260
3261         tr = top_trace_array();
3262         if (!tr)
3263                 return;
3264
3265         pr_info("Running tests on trace events:\n");
3266
3267         list_for_each_entry(file, &tr->events, list) {
3268
3269                 call = file->event_call;
3270
3271                 /* Only test those that have a probe */
3272                 if (!call->class || !call->class->probe)
3273                         continue;
3274
3275 /*
3276  * Testing syscall events here is pretty useless, but
3277  * we still do it if configured. But this is time consuming.
3278  * What we really need is a user thread to perform the
3279  * syscalls as we test.
3280  */
3281 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3282                 if (call->class->system &&
3283                     strcmp(call->class->system, "syscalls") == 0)
3284                         continue;
3285 #endif
3286
3287                 pr_info("Testing event %s: ", trace_event_name(call));
3288
3289                 /*
3290                  * If an event is already enabled, someone is using
3291                  * it and the self test should not be on.
3292                  */
3293                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3294                         pr_warn("Enabled event during self test!\n");
3295                         WARN_ON_ONCE(1);
3296                         continue;
3297                 }
3298
3299                 ftrace_event_enable_disable(file, 1);
3300                 event_test_stuff();
3301                 ftrace_event_enable_disable(file, 0);
3302
3303                 pr_cont("OK\n");
3304         }
3305
3306         /* Now test at the sub system level */
3307
3308         pr_info("Running tests on trace event systems:\n");
3309
3310         list_for_each_entry(dir, &tr->systems, list) {
3311
3312                 system = dir->subsystem;
3313
3314                 /* the ftrace system is special, skip it */
3315                 if (strcmp(system->name, "ftrace") == 0)
3316                         continue;
3317
3318                 pr_info("Testing event system %s: ", system->name);
3319
3320                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3321                 if (WARN_ON_ONCE(ret)) {
3322                         pr_warn("error enabling system %s\n",
3323                                 system->name);
3324                         continue;
3325                 }
3326
3327                 event_test_stuff();
3328
3329                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3330                 if (WARN_ON_ONCE(ret)) {
3331                         pr_warn("error disabling system %s\n",
3332                                 system->name);
3333                         continue;
3334                 }
3335
3336                 pr_cont("OK\n");
3337         }
3338
3339         /* Test with all events enabled */
3340
3341         pr_info("Running tests on all trace events:\n");
3342         pr_info("Testing all events: ");
3343
3344         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3345         if (WARN_ON_ONCE(ret)) {
3346                 pr_warn("error enabling all events\n");
3347                 return;
3348         }
3349
3350         event_test_stuff();
3351
3352         /* reset sysname */
3353         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3354         if (WARN_ON_ONCE(ret)) {
3355                 pr_warn("error disabling all events\n");
3356                 return;
3357         }
3358
3359         pr_cont("OK\n");
3360 }
3361
3362 #ifdef CONFIG_FUNCTION_TRACER
3363
3364 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3365
3366 static struct trace_event_file event_trace_file __initdata;
3367
3368 static void __init
3369 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3370                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3371 {
3372         struct ring_buffer_event *event;
3373         struct ring_buffer *buffer;
3374         struct ftrace_entry *entry;
3375         unsigned long flags;
3376         long disabled;
3377         int cpu;
3378         int pc;
3379
3380         pc = preempt_count();
3381         preempt_disable_notrace();
3382         cpu = raw_smp_processor_id();
3383         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3384
3385         if (disabled != 1)
3386                 goto out;
3387
3388         local_save_flags(flags);
3389
3390         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3391                                                 TRACE_FN, sizeof(*entry),
3392                                                 flags, pc);
3393         if (!event)
3394                 goto out;
3395         entry   = ring_buffer_event_data(event);
3396         entry->ip                       = ip;
3397         entry->parent_ip                = parent_ip;
3398
3399         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3400                                     entry, flags, pc);
3401  out:
3402         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3403         preempt_enable_notrace();
3404 }
3405
3406 static struct ftrace_ops trace_ops __initdata  =
3407 {
3408         .func = function_test_events_call,
3409         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3410 };
3411
3412 static __init void event_trace_self_test_with_function(void)
3413 {
3414         int ret;
3415
3416         event_trace_file.tr = top_trace_array();
3417         if (WARN_ON(!event_trace_file.tr))
3418                 return;
3419
3420         ret = register_ftrace_function(&trace_ops);
3421         if (WARN_ON(ret < 0)) {
3422                 pr_info("Failed to enable function tracer for event tests\n");
3423                 return;
3424         }
3425         pr_info("Running tests again, along with the function tracer\n");
3426         event_trace_self_tests();
3427         unregister_ftrace_function(&trace_ops);
3428 }
3429 #else
3430 static __init void event_trace_self_test_with_function(void)
3431 {
3432 }
3433 #endif
3434
3435 static __init int event_trace_self_tests_init(void)
3436 {
3437         if (!tracing_selftest_disabled) {
3438                 event_trace_self_tests();
3439                 event_trace_self_test_with_function();
3440         }
3441
3442         return 0;
3443 }
3444
3445 late_initcall(event_trace_self_tests_init);
3446
3447 #endif