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 EXPORT_SYMBOL_GPL(ftrace_set_clr_event);
 836
 837 /**
 838  * trace_set_clr_event - enable or disable an event
 839  * @system: system name to match (NULL for any system)
 840  * @event: event name to match (NULL for all events, within system)
 841  * @set: 1 to enable, 0 to disable
 842  *
 843  * This is a way for other parts of the kernel to enable or disable
 844  * event recording.
 845  *
 846  * Returns 0 on success, -EINVAL if the parameters do not match any
 847  * registered events.
 848  */
 849 int trace_set_clr_event(const char *system, const char *event, int set)
 850 {
 851         struct trace_array *tr = top_trace_array();
 852
 853         if (!tr)
 854                 return -ENODEV;
 855
 856         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 857 }
 858 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 859
 860 /* 128 should be much more than enough */
 861 #define EVENT_BUF_SIZE          127
 862
 863 static ssize_t
 864 ftrace_event_write(struct file *file, const char __user *ubuf,
 865                    size_t cnt, loff_t *ppos)
 866 {
 867         struct trace_parser parser;
 868         struct seq_file *m = file->private_data;
 869         struct trace_array *tr = m->private;
 870         ssize_t read, ret;
 871
 872         if (!cnt)
 873                 return 0;
 874
 875         ret = tracing_update_buffers();
 876         if (ret < 0)
 877                 return ret;
 878
 879         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 880                 return -ENOMEM;
 881
 882         read = trace_get_user(&parser, ubuf, cnt, ppos);
 883
 884         if (read >= 0 && trace_parser_loaded((&parser))) {
 885                 int set = 1;
 886
 887                 if (*parser.buffer == '!')
 888                         set = 0;
 889
 890                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 891                 if (ret)
 892                         goto out_put;
 893         }
 894
 895         ret = read;
 896
 897  out_put:
 898         trace_parser_put(&parser);
 899
 900         return ret;
 901 }
 902
 903 static void *
 904 t_next(struct seq_file *m, void *v, loff_t *pos)
 905 {
 906         struct trace_event_file *file = v;
 907         struct trace_event_call *call;
 908         struct trace_array *tr = m->private;
 909
 910         (*pos)++;
 911
 912         list_for_each_entry_continue(file, &tr->events, list) {
 913                 call = file->event_call;
 914                 /*
 915                  * The ftrace subsystem is for showing formats only.
 916                  * They can not be enabled or disabled via the event files.
 917                  */
 918                 if (call->class && call->class->reg &&
 919                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 920                         return file;
 921         }
 922
 923         return NULL;
 924 }
 925
 926 static void *t_start(struct seq_file *m, loff_t *pos)
 927 {
 928         struct trace_event_file *file;
 929         struct trace_array *tr = m->private;
 930         loff_t l;
 931
 932         mutex_lock(&event_mutex);
 933
 934         file = list_entry(&tr->events, struct trace_event_file, list);
 935         for (l = 0; l <= *pos; ) {
 936                 file = t_next(m, file, &l);
 937                 if (!file)
 938                         break;
 939         }
 940         return file;
 941 }
 942
 943 static void *
 944 s_next(struct seq_file *m, void *v, loff_t *pos)
 945 {
 946         struct trace_event_file *file = v;
 947         struct trace_array *tr = m->private;
 948
 949         (*pos)++;
 950
 951         list_for_each_entry_continue(file, &tr->events, list) {
 952                 if (file->flags & EVENT_FILE_FL_ENABLED)
 953                         return file;
 954         }
 955
 956         return NULL;
 957 }
 958
 959 static void *s_start(struct seq_file *m, loff_t *pos)
 960 {
 961         struct trace_event_file *file;
 962         struct trace_array *tr = m->private;
 963         loff_t l;
 964
 965         mutex_lock(&event_mutex);
 966
 967         file = list_entry(&tr->events, struct trace_event_file, list);
 968         for (l = 0; l <= *pos; ) {
 969                 file = s_next(m, file, &l);
 970                 if (!file)
 971                         break;
 972         }
 973         return file;
 974 }
 975
 976 static int t_show(struct seq_file *m, void *v)
 977 {
 978         struct trace_event_file *file = v;
 979         struct trace_event_call *call = file->event_call;
 980
 981         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 982                 seq_printf(m, "%s:", call->class->system);
 983         seq_printf(m, "%s\n", trace_event_name(call));
 984
 985         return 0;
 986 }
 987
 988 static void t_stop(struct seq_file *m, void *p)
 989 {
 990         mutex_unlock(&event_mutex);
 991 }
 992
 993 static void *
 994 p_next(struct seq_file *m, void *v, loff_t *pos)
 995 {
 996         struct trace_array *tr = m->private;
 997         struct trace_pid_list *pid_list = rcu_dereference_sched(tr->filtered_pids);
 998
 999         return trace_pid_next(pid_list, v, pos);
1000 }
1001
1002 static void *p_start(struct seq_file *m, loff_t *pos)
1003         __acquires(RCU)
1004 {
1005         struct trace_pid_list *pid_list;
1006         struct trace_array *tr = m->private;
1007
1008         /*
1009          * Grab the mutex, to keep calls to p_next() having the same
1010          * tr->filtered_pids as p_start() has.
1011          * If we just passed the tr->filtered_pids around, then RCU would
1012          * have been enough, but doing that makes things more complex.
1013          */
1014         mutex_lock(&event_mutex);
1015         rcu_read_lock_sched();
1016
1017         pid_list = rcu_dereference_sched(tr->filtered_pids);
1018
1019         if (!pid_list)
1020                 return NULL;
1021
1022         return trace_pid_start(pid_list, pos);
1023 }
1024
1025 static void p_stop(struct seq_file *m, void *p)
1026         __releases(RCU)
1027 {
1028         rcu_read_unlock_sched();
1029         mutex_unlock(&event_mutex);
1030 }
1031
1032 static ssize_t
1033 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1034                   loff_t *ppos)
1035 {
1036         struct trace_event_file *file;
1037         unsigned long flags;
1038         char buf[4] = "0";
1039
1040         mutex_lock(&event_mutex);
1041         file = event_file_data(filp);
1042         if (likely(file))
1043                 flags = file->flags;
1044         mutex_unlock(&event_mutex);
1045
1046         if (!file)
1047                 return -ENODEV;
1048
1049         if (flags & EVENT_FILE_FL_ENABLED &&
1050             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1051                 strcpy(buf, "1");
1052
1053         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1054             flags & EVENT_FILE_FL_SOFT_MODE)
1055                 strcat(buf, "*");
1056
1057         strcat(buf, "\n");
1058
1059         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1060 }
1061
1062 static ssize_t
1063 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1064                    loff_t *ppos)
1065 {
1066         struct trace_event_file *file;
1067         unsigned long val;
1068         int ret;
1069
1070         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1071         if (ret)
1072                 return ret;
1073
1074         ret = tracing_update_buffers();
1075         if (ret < 0)
1076                 return ret;
1077
1078         switch (val) {
1079         case 0:
1080         case 1:
1081                 ret = -ENODEV;
1082                 mutex_lock(&event_mutex);
1083                 file = event_file_data(filp);
1084                 if (likely(file))
1085                         ret = ftrace_event_enable_disable(file, val);
1086                 mutex_unlock(&event_mutex);
1087                 break;
1088
1089         default:
1090                 return -EINVAL;
1091         }
1092
1093         *ppos += cnt;
1094
1095         return ret ? ret : cnt;
1096 }
1097
1098 static ssize_t
1099 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1100                    loff_t *ppos)
1101 {
1102         const char set_to_char[4] = { '?', '0', '1', 'X' };
1103         struct trace_subsystem_dir *dir = filp->private_data;
1104         struct event_subsystem *system = dir->subsystem;
1105         struct trace_event_call *call;
1106         struct trace_event_file *file;
1107         struct trace_array *tr = dir->tr;
1108         char buf[2];
1109         int set = 0;
1110         int ret;
1111
1112         mutex_lock(&event_mutex);
1113         list_for_each_entry(file, &tr->events, list) {
1114                 call = file->event_call;
1115                 if (!trace_event_name(call) || !call->class || !call->class->reg)
1116                         continue;
1117
1118                 if (system && strcmp(call->class->system, system->name) != 0)
1119                         continue;
1120
1121                 /*
1122                  * We need to find out if all the events are set
1123                  * or if all events or cleared, or if we have
1124                  * a mixture.
1125                  */
1126                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1127
1128                 /*
1129                  * If we have a mixture, no need to look further.
1130                  */
1131                 if (set == 3)
1132                         break;
1133         }
1134         mutex_unlock(&event_mutex);
1135
1136         buf[0] = set_to_char[set];
1137         buf[1] = '\n';
1138
1139         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1140
1141         return ret;
1142 }
1143
1144 static ssize_t
1145 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1146                     loff_t *ppos)
1147 {
1148         struct trace_subsystem_dir *dir = filp->private_data;
1149         struct event_subsystem *system = dir->subsystem;
1150         const char *name = NULL;
1151         unsigned long val;
1152         ssize_t ret;
1153
1154         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1155         if (ret)
1156                 return ret;
1157
1158         ret = tracing_update_buffers();
1159         if (ret < 0)
1160                 return ret;
1161
1162         if (val != 0 && val != 1)
1163                 return -EINVAL;
1164
1165         /*
1166          * Opening of "enable" adds a ref count to system,
1167          * so the name is safe to use.
1168          */
1169         if (system)
1170                 name = system->name;
1171
1172         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1173         if (ret)
1174                 goto out;
1175
1176         ret = cnt;
1177
1178 out:
1179         *ppos += cnt;
1180
1181         return ret;
1182 }
1183
1184 enum {
1185         FORMAT_HEADER           = 1,
1186         FORMAT_FIELD_SEPERATOR  = 2,
1187         FORMAT_PRINTFMT         = 3,
1188 };
1189
1190 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1191 {
1192         struct trace_event_call *call = event_file_data(m->private);
1193         struct list_head *common_head = &ftrace_common_fields;
1194         struct list_head *head = trace_get_fields(call);
1195         struct list_head *node = v;
1196
1197         (*pos)++;
1198
1199         switch ((unsigned long)v) {
1200         case FORMAT_HEADER:
1201                 node = common_head;
1202                 break;
1203
1204         case FORMAT_FIELD_SEPERATOR:
1205                 node = head;
1206                 break;
1207
1208         case FORMAT_PRINTFMT:
1209                 /* all done */
1210                 return NULL;
1211         }
1212
1213         node = node->prev;
1214         if (node == common_head)
1215                 return (void *)FORMAT_FIELD_SEPERATOR;
1216         else if (node == head)
1217                 return (void *)FORMAT_PRINTFMT;
1218         else
1219                 return node;
1220 }
1221
1222 static int f_show(struct seq_file *m, void *v)
1223 {
1224         struct trace_event_call *call = event_file_data(m->private);
1225         struct ftrace_event_field *field;
1226         const char *array_descriptor;
1227
1228         switch ((unsigned long)v) {
1229         case FORMAT_HEADER:
1230                 seq_printf(m, "name: %s\n", trace_event_name(call));
1231                 seq_printf(m, "ID: %d\n", call->event.type);
1232                 seq_puts(m, "format:\n");
1233                 return 0;
1234
1235         case FORMAT_FIELD_SEPERATOR:
1236                 seq_putc(m, '\n');
1237                 return 0;
1238
1239         case FORMAT_PRINTFMT:
1240                 seq_printf(m, "\nprint fmt: %s\n",
1241                            call->print_fmt);
1242                 return 0;
1243         }
1244
1245         field = list_entry(v, struct ftrace_event_field, link);
1246         /*
1247          * Smartly shows the array type(except dynamic array).
1248          * Normal:
1249          *      field:TYPE VAR
1250          * If TYPE := TYPE[LEN], it is shown:
1251          *      field:TYPE VAR[LEN]
1252          */
1253         array_descriptor = strchr(field->type, '[');
1254
1255         if (str_has_prefix(field->type, "__data_loc"))
1256                 array_descriptor = NULL;
1257
1258         if (!array_descriptor)
1259                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1260                            field->type, field->name, field->offset,
1261                            field->size, !!field->is_signed);
1262         else
1263                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1264                            (int)(array_descriptor - field->type),
1265                            field->type, field->name,
1266                            array_descriptor, field->offset,
1267                            field->size, !!field->is_signed);
1268
1269         return 0;
1270 }
1271
1272 static void *f_start(struct seq_file *m, loff_t *pos)
1273 {
1274         void *p = (void *)FORMAT_HEADER;
1275         loff_t l = 0;
1276
1277         /* ->stop() is called even if ->start() fails */
1278         mutex_lock(&event_mutex);
1279         if (!event_file_data(m->private))
1280                 return ERR_PTR(-ENODEV);
1281
1282         while (l < *pos && p)
1283                 p = f_next(m, p, &l);
1284
1285         return p;
1286 }
1287
1288 static void f_stop(struct seq_file *m, void *p)
1289 {
1290         mutex_unlock(&event_mutex);
1291 }
1292
1293 static const struct seq_operations trace_format_seq_ops = {
1294         .start          = f_start,
1295         .next           = f_next,
1296         .stop           = f_stop,
1297         .show           = f_show,
1298 };
1299
1300 static int trace_format_open(struct inode *inode, struct file *file)
1301 {
1302         struct seq_file *m;
1303         int ret;
1304
1305         ret = seq_open(file, &trace_format_seq_ops);
1306         if (ret < 0)
1307                 return ret;
1308
1309         m = file->private_data;
1310         m->private = file;
1311
1312         return 0;
1313 }
1314
1315 static ssize_t
1316 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1317 {
1318         int id = (long)event_file_data(filp);
1319         char buf[32];
1320         int len;
1321
1322         if (unlikely(!id))
1323                 return -ENODEV;
1324
1325         len = sprintf(buf, "%d\n", id);
1326
1327         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1328 }
1329
1330 static ssize_t
1331 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1332                   loff_t *ppos)
1333 {
1334         struct trace_event_file *file;
1335         struct trace_seq *s;
1336         int r = -ENODEV;
1337
1338         if (*ppos)
1339                 return 0;
1340
1341         s = kmalloc(sizeof(*s), GFP_KERNEL);
1342
1343         if (!s)
1344                 return -ENOMEM;
1345
1346         trace_seq_init(s);
1347
1348         mutex_lock(&event_mutex);
1349         file = event_file_data(filp);
1350         if (file)
1351                 print_event_filter(file, s);
1352         mutex_unlock(&event_mutex);
1353
1354         if (file)
1355                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1356                                             s->buffer, trace_seq_used(s));
1357
1358         kfree(s);
1359
1360         return r;
1361 }
1362
1363 static ssize_t
1364 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1365                    loff_t *ppos)
1366 {
1367         struct trace_event_file *file;
1368         char *buf;
1369         int err = -ENODEV;
1370
1371         if (cnt >= PAGE_SIZE)
1372                 return -EINVAL;
1373
1374         buf = memdup_user_nul(ubuf, cnt);
1375         if (IS_ERR(buf))
1376                 return PTR_ERR(buf);
1377
1378         mutex_lock(&event_mutex);
1379         file = event_file_data(filp);
1380         if (file)
1381                 err = apply_event_filter(file, buf);
1382         mutex_unlock(&event_mutex);
1383
1384         kfree(buf);
1385         if (err < 0)
1386                 return err;
1387
1388         *ppos += cnt;
1389
1390         return cnt;
1391 }
1392
1393 static LIST_HEAD(event_subsystems);
1394
1395 static int subsystem_open(struct inode *inode, struct file *filp)
1396 {
1397         struct event_subsystem *system = NULL;
1398         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1399         struct trace_array *tr;
1400         int ret;
1401
1402         if (tracing_is_disabled())
1403                 return -ENODEV;
1404
1405         /* Make sure the system still exists */
1406         mutex_lock(&event_mutex);
1407         mutex_lock(&trace_types_lock);
1408         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1409                 list_for_each_entry(dir, &tr->systems, list) {
1410                         if (dir == inode->i_private) {
1411                                 /* Don't open systems with no events */
1412                                 if (dir->nr_events) {
1413                                         __get_system_dir(dir);
1414                                         system = dir->subsystem;
1415                                 }
1416                                 goto exit_loop;
1417                         }
1418                 }
1419         }
1420  exit_loop:
1421         mutex_unlock(&trace_types_lock);
1422         mutex_unlock(&event_mutex);
1423
1424         if (!system)
1425                 return -ENODEV;
1426
1427         /* Some versions of gcc think dir can be uninitialized here */
1428         WARN_ON(!dir);
1429
1430         /* Still need to increment the ref count of the system */
1431         if (trace_array_get(tr) < 0) {
1432                 put_system(dir);
1433                 return -ENODEV;
1434         }
1435
1436         ret = tracing_open_generic(inode, filp);
1437         if (ret < 0) {
1438                 trace_array_put(tr);
1439                 put_system(dir);
1440         }
1441
1442         return ret;
1443 }
1444
1445 static int system_tr_open(struct inode *inode, struct file *filp)
1446 {
1447         struct trace_subsystem_dir *dir;
1448         struct trace_array *tr = inode->i_private;
1449         int ret;
1450
1451         if (tracing_is_disabled())
1452                 return -ENODEV;
1453
1454         if (trace_array_get(tr) < 0)
1455                 return -ENODEV;
1456
1457         /* Make a temporary dir that has no system but points to tr */
1458         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1459         if (!dir) {
1460                 trace_array_put(tr);
1461                 return -ENOMEM;
1462         }
1463
1464         dir->tr = tr;
1465
1466         ret = tracing_open_generic(inode, filp);
1467         if (ret < 0) {
1468                 trace_array_put(tr);
1469                 kfree(dir);
1470                 return ret;
1471         }
1472
1473         filp->private_data = dir;
1474
1475         return 0;
1476 }
1477
1478 static int subsystem_release(struct inode *inode, struct file *file)
1479 {
1480         struct trace_subsystem_dir *dir = file->private_data;
1481
1482         trace_array_put(dir->tr);
1483
1484         /*
1485          * If dir->subsystem is NULL, then this is a temporary
1486          * descriptor that was made for a trace_array to enable
1487          * all subsystems.
1488          */
1489         if (dir->subsystem)
1490                 put_system(dir);
1491         else
1492                 kfree(dir);
1493
1494         return 0;
1495 }
1496
1497 static ssize_t
1498 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1499                       loff_t *ppos)
1500 {
1501         struct trace_subsystem_dir *dir = filp->private_data;
1502         struct event_subsystem *system = dir->subsystem;
1503         struct trace_seq *s;
1504         int r;
1505
1506         if (*ppos)
1507                 return 0;
1508
1509         s = kmalloc(sizeof(*s), GFP_KERNEL);
1510         if (!s)
1511                 return -ENOMEM;
1512
1513         trace_seq_init(s);
1514
1515         print_subsystem_event_filter(system, s);
1516         r = simple_read_from_buffer(ubuf, cnt, ppos,
1517                                     s->buffer, trace_seq_used(s));
1518
1519         kfree(s);
1520
1521         return r;
1522 }
1523
1524 static ssize_t
1525 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1526                        loff_t *ppos)
1527 {
1528         struct trace_subsystem_dir *dir = filp->private_data;
1529         char *buf;
1530         int err;
1531
1532         if (cnt >= PAGE_SIZE)
1533                 return -EINVAL;
1534
1535         buf = memdup_user_nul(ubuf, cnt);
1536         if (IS_ERR(buf))
1537                 return PTR_ERR(buf);
1538
1539         err = apply_subsystem_event_filter(dir, buf);
1540         kfree(buf);
1541         if (err < 0)
1542                 return err;
1543
1544         *ppos += cnt;
1545
1546         return cnt;
1547 }
1548
1549 static ssize_t
1550 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1551 {
1552         int (*func)(struct trace_seq *s) = filp->private_data;
1553         struct trace_seq *s;
1554         int r;
1555
1556         if (*ppos)
1557                 return 0;
1558
1559         s = kmalloc(sizeof(*s), GFP_KERNEL);
1560         if (!s)
1561                 return -ENOMEM;
1562
1563         trace_seq_init(s);
1564
1565         func(s);
1566         r = simple_read_from_buffer(ubuf, cnt, ppos,
1567                                     s->buffer, trace_seq_used(s));
1568
1569         kfree(s);
1570
1571         return r;
1572 }
1573
1574 static void ignore_task_cpu(void *data)
1575 {
1576         struct trace_array *tr = data;
1577         struct trace_pid_list *pid_list;
1578
1579         /*
1580          * This function is called by on_each_cpu() while the
1581          * event_mutex is held.
1582          */
1583         pid_list = rcu_dereference_protected(tr->filtered_pids,
1584                                              mutex_is_locked(&event_mutex));
1585
1586         this_cpu_write(tr->trace_buffer.data->ignore_pid,
1587                        trace_ignore_this_task(pid_list, current));
1588 }
1589
1590 static ssize_t
1591 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1592                        size_t cnt, loff_t *ppos)
1593 {
1594         struct seq_file *m = filp->private_data;
1595         struct trace_array *tr = m->private;
1596         struct trace_pid_list *filtered_pids = NULL;
1597         struct trace_pid_list *pid_list;
1598         struct trace_event_file *file;
1599         ssize_t ret;
1600
1601         if (!cnt)
1602                 return 0;
1603
1604         ret = tracing_update_buffers();
1605         if (ret < 0)
1606                 return ret;
1607
1608         mutex_lock(&event_mutex);
1609
1610         filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1611                                              lockdep_is_held(&event_mutex));
1612
1613         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1614         if (ret < 0)
1615                 goto out;
1616
1617         rcu_assign_pointer(tr->filtered_pids, pid_list);
1618
1619         list_for_each_entry(file, &tr->events, list) {
1620                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1621         }
1622
1623         if (filtered_pids) {
1624                 tracepoint_synchronize_unregister();
1625                 trace_free_pid_list(filtered_pids);
1626         } else if (pid_list) {
1627                 /*
1628                  * Register a probe that is called before all other probes
1629                  * to set ignore_pid if next or prev do not match.
1630                  * Register a probe this is called after all other probes
1631                  * to only keep ignore_pid set if next pid matches.
1632                  */
1633                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1634                                                  tr, INT_MAX);
1635                 register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1636                                                  tr, 0);
1637
1638                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1639                                                  tr, INT_MAX);
1640                 register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1641                                                  tr, 0);
1642
1643                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1644                                                      tr, INT_MAX);
1645                 register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1646                                                      tr, 0);
1647
1648                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1649                                                  tr, INT_MAX);
1650                 register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1651                                                  tr, 0);
1652         }
1653
1654         /*
1655          * Ignoring of pids is done at task switch. But we have to
1656          * check for those tasks that are currently running.
1657          * Always do this in case a pid was appended or removed.
1658          */
1659         on_each_cpu(ignore_task_cpu, tr, 1);
1660
1661  out:
1662         mutex_unlock(&event_mutex);
1663
1664         if (ret > 0)
1665                 *ppos += ret;
1666
1667         return ret;
1668 }
1669
1670 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1671 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1672 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1673 static int ftrace_event_release(struct inode *inode, struct file *file);
1674
1675 static const struct seq_operations show_event_seq_ops = {
1676         .start = t_start,
1677         .next = t_next,
1678         .show = t_show,
1679         .stop = t_stop,
1680 };
1681
1682 static const struct seq_operations show_set_event_seq_ops = {
1683         .start = s_start,
1684         .next = s_next,
1685         .show = t_show,
1686         .stop = t_stop,
1687 };
1688
1689 static const struct seq_operations show_set_pid_seq_ops = {
1690         .start = p_start,
1691         .next = p_next,
1692         .show = trace_pid_show,
1693         .stop = p_stop,
1694 };
1695
1696 static const struct file_operations ftrace_avail_fops = {
1697         .open = ftrace_event_avail_open,
1698         .read = seq_read,
1699         .llseek = seq_lseek,
1700         .release = seq_release,
1701 };
1702
1703 static const struct file_operations ftrace_set_event_fops = {
1704         .open = ftrace_event_set_open,
1705         .read = seq_read,
1706         .write = ftrace_event_write,
1707         .llseek = seq_lseek,
1708         .release = ftrace_event_release,
1709 };
1710
1711 static const struct file_operations ftrace_set_event_pid_fops = {
1712         .open = ftrace_event_set_pid_open,
1713         .read = seq_read,
1714         .write = ftrace_event_pid_write,
1715         .llseek = seq_lseek,
1716         .release = ftrace_event_release,
1717 };
1718
1719 static const struct file_operations ftrace_enable_fops = {
1720         .open = tracing_open_generic,
1721         .read = event_enable_read,
1722         .write = event_enable_write,
1723         .llseek = default_llseek,
1724 };
1725
1726 static const struct file_operations ftrace_event_format_fops = {
1727         .open = trace_format_open,
1728         .read = seq_read,
1729         .llseek = seq_lseek,
1730         .release = seq_release,
1731 };
1732
1733 static const struct file_operations ftrace_event_id_fops = {
1734         .read = event_id_read,
1735         .llseek = default_llseek,
1736 };
1737
1738 static const struct file_operations ftrace_event_filter_fops = {
1739         .open = tracing_open_generic,
1740         .read = event_filter_read,
1741         .write = event_filter_write,
1742         .llseek = default_llseek,
1743 };
1744
1745 static const struct file_operations ftrace_subsystem_filter_fops = {
1746         .open = subsystem_open,
1747         .read = subsystem_filter_read,
1748         .write = subsystem_filter_write,
1749         .llseek = default_llseek,
1750         .release = subsystem_release,
1751 };
1752
1753 static const struct file_operations ftrace_system_enable_fops = {
1754         .open = subsystem_open,
1755         .read = system_enable_read,
1756         .write = system_enable_write,
1757         .llseek = default_llseek,
1758         .release = subsystem_release,
1759 };
1760
1761 static const struct file_operations ftrace_tr_enable_fops = {
1762         .open = system_tr_open,
1763         .read = system_enable_read,
1764         .write = system_enable_write,
1765         .llseek = default_llseek,
1766         .release = subsystem_release,
1767 };
1768
1769 static const struct file_operations ftrace_show_header_fops = {
1770         .open = tracing_open_generic,
1771         .read = show_header,
1772         .llseek = default_llseek,
1773 };
1774
1775 static int
1776 ftrace_event_open(struct inode *inode, struct file *file,
1777                   const struct seq_operations *seq_ops)
1778 {
1779         struct seq_file *m;
1780         int ret;
1781
1782         ret = seq_open(file, seq_ops);
1783         if (ret < 0)
1784                 return ret;
1785         m = file->private_data;
1786         /* copy tr over to seq ops */
1787         m->private = inode->i_private;
1788
1789         return ret;
1790 }
1791
1792 static int ftrace_event_release(struct inode *inode, struct file *file)
1793 {
1794         struct trace_array *tr = inode->i_private;
1795
1796         trace_array_put(tr);
1797
1798         return seq_release(inode, file);
1799 }
1800
1801 static int
1802 ftrace_event_avail_open(struct inode *inode, struct file *file)
1803 {
1804         const struct seq_operations *seq_ops = &show_event_seq_ops;
1805
1806         return ftrace_event_open(inode, file, seq_ops);
1807 }
1808
1809 static int
1810 ftrace_event_set_open(struct inode *inode, struct file *file)
1811 {
1812         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1813         struct trace_array *tr = inode->i_private;
1814         int ret;
1815
1816         if (trace_array_get(tr) < 0)
1817                 return -ENODEV;
1818
1819         if ((file->f_mode & FMODE_WRITE) &&
1820             (file->f_flags & O_TRUNC))
1821                 ftrace_clear_events(tr);
1822
1823         ret = ftrace_event_open(inode, file, seq_ops);
1824         if (ret < 0)
1825                 trace_array_put(tr);
1826         return ret;
1827 }
1828
1829 static int
1830 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1831 {
1832         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1833         struct trace_array *tr = inode->i_private;
1834         int ret;
1835
1836         if (trace_array_get(tr) < 0)
1837                 return -ENODEV;
1838
1839         if ((file->f_mode & FMODE_WRITE) &&
1840             (file->f_flags & O_TRUNC))
1841                 ftrace_clear_event_pids(tr);
1842
1843         ret = ftrace_event_open(inode, file, seq_ops);
1844         if (ret < 0)
1845                 trace_array_put(tr);
1846         return ret;
1847 }
1848
1849 static struct event_subsystem *
1850 create_new_subsystem(const char *name)
1851 {
1852         struct event_subsystem *system;
1853
1854         /* need to create new entry */
1855         system = kmalloc(sizeof(*system), GFP_KERNEL);
1856         if (!system)
1857                 return NULL;
1858
1859         system->ref_count = 1;
1860
1861         /* Only allocate if dynamic (kprobes and modules) */
1862         system->name = kstrdup_const(name, GFP_KERNEL);
1863         if (!system->name)
1864                 goto out_free;
1865
1866         system->filter = NULL;
1867
1868         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1869         if (!system->filter)
1870                 goto out_free;
1871
1872         list_add(&system->list, &event_subsystems);
1873
1874         return system;
1875
1876  out_free:
1877         kfree_const(system->name);
1878         kfree(system);
1879         return NULL;
1880 }
1881
1882 static struct dentry *
1883 event_subsystem_dir(struct trace_array *tr, const char *name,
1884                     struct trace_event_file *file, struct dentry *parent)
1885 {
1886         struct trace_subsystem_dir *dir;
1887         struct event_subsystem *system;
1888         struct dentry *entry;
1889
1890         /* First see if we did not already create this dir */
1891         list_for_each_entry(dir, &tr->systems, list) {
1892                 system = dir->subsystem;
1893                 if (strcmp(system->name, name) == 0) {
1894                         dir->nr_events++;
1895                         file->system = dir;
1896                         return dir->entry;
1897                 }
1898         }
1899
1900         /* Now see if the system itself exists. */
1901         list_for_each_entry(system, &event_subsystems, list) {
1902                 if (strcmp(system->name, name) == 0)
1903                         break;
1904         }
1905         /* Reset system variable when not found */
1906         if (&system->list == &event_subsystems)
1907                 system = NULL;
1908
1909         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1910         if (!dir)
1911                 goto out_fail;
1912
1913         if (!system) {
1914                 system = create_new_subsystem(name);
1915                 if (!system)
1916                         goto out_free;
1917         } else
1918                 __get_system(system);
1919
1920         dir->entry = tracefs_create_dir(name, parent);
1921         if (!dir->entry) {
1922                 pr_warn("Failed to create system directory %s\n", name);
1923                 __put_system(system);
1924                 goto out_free;
1925         }
1926
1927         dir->tr = tr;
1928         dir->ref_count = 1;
1929         dir->nr_events = 1;
1930         dir->subsystem = system;
1931         file->system = dir;
1932
1933         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
1934                                     &ftrace_subsystem_filter_fops);
1935         if (!entry) {
1936                 kfree(system->filter);
1937                 system->filter = NULL;
1938                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
1939         }
1940
1941         trace_create_file("enable", 0644, dir->entry, dir,
1942                           &ftrace_system_enable_fops);
1943
1944         list_add(&dir->list, &tr->systems);
1945
1946         return dir->entry;
1947
1948  out_free:
1949         kfree(dir);
1950  out_fail:
1951         /* Only print this message if failed on memory allocation */
1952         if (!dir || !system)
1953                 pr_warn("No memory to create event subsystem %s\n", name);
1954         return NULL;
1955 }
1956
1957 static int
1958 event_create_dir(struct dentry *parent, struct trace_event_file *file)
1959 {
1960         struct trace_event_call *call = file->event_call;
1961         struct trace_array *tr = file->tr;
1962         struct list_head *head;
1963         struct dentry *d_events;
1964         const char *name;
1965         int ret;
1966
1967         /*
1968          * If the trace point header did not define TRACE_SYSTEM
1969          * then the system would be called "TRACE_SYSTEM".
1970          */
1971         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1972                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1973                 if (!d_events)
1974                         return -ENOMEM;
1975         } else
1976                 d_events = parent;
1977
1978         name = trace_event_name(call);
1979         file->dir = tracefs_create_dir(name, d_events);
1980         if (!file->dir) {
1981                 pr_warn("Could not create tracefs '%s' directory\n", name);
1982                 return -1;
1983         }
1984
1985         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1986                 trace_create_file("enable", 0644, file->dir, file,
1987                                   &ftrace_enable_fops);
1988
1989 #ifdef CONFIG_PERF_EVENTS
1990         if (call->event.type && call->class->reg)
1991                 trace_create_file("id", 0444, file->dir,
1992                                   (void *)(long)call->event.type,
1993                                   &ftrace_event_id_fops);
1994 #endif
1995
1996         /*
1997          * Other events may have the same class. Only update
1998          * the fields if they are not already defined.
1999          */
2000         head = trace_get_fields(call);
2001         if (list_empty(head)) {
2002                 ret = call->class->define_fields(call);
2003                 if (ret < 0) {
2004                         pr_warn("Could not initialize trace point events/%s\n",
2005                                 name);
2006                         return -1;
2007                 }
2008         }
2009
2010         /*
2011          * Only event directories that can be enabled should have
2012          * triggers or filters.
2013          */
2014         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2015                 trace_create_file("filter", 0644, file->dir, file,
2016                                   &ftrace_event_filter_fops);
2017
2018                 trace_create_file("trigger", 0644, file->dir, file,
2019                                   &event_trigger_fops);
2020         }
2021
2022 #ifdef CONFIG_HIST_TRIGGERS
2023         trace_create_file("hist", 0444, file->dir, file,
2024                           &event_hist_fops);
2025 #endif
2026         trace_create_file("format", 0444, file->dir, call,
2027                           &ftrace_event_format_fops);
2028
2029         return 0;
2030 }
2031
2032 static void remove_event_from_tracers(struct trace_event_call *call)
2033 {
2034         struct trace_event_file *file;
2035         struct trace_array *tr;
2036
2037         do_for_each_event_file_safe(tr, file) {
2038                 if (file->event_call != call)
2039                         continue;
2040
2041                 remove_event_file_dir(file);
2042                 /*
2043                  * The do_for_each_event_file_safe() is
2044                  * a double loop. After finding the call for this
2045                  * trace_array, we use break to jump to the next
2046                  * trace_array.
2047                  */
2048                 break;
2049         } while_for_each_event_file();
2050 }
2051
2052 static void event_remove(struct trace_event_call *call)
2053 {
2054         struct trace_array *tr;
2055         struct trace_event_file *file;
2056
2057         do_for_each_event_file(tr, file) {
2058                 if (file->event_call != call)
2059                         continue;
2060
2061                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2062                         tr->clear_trace = true;
2063
2064                 ftrace_event_enable_disable(file, 0);
2065                 /*
2066                  * The do_for_each_event_file() is
2067                  * a double loop. After finding the call for this
2068                  * trace_array, we use break to jump to the next
2069                  * trace_array.
2070                  */
2071                 break;
2072         } while_for_each_event_file();
2073
2074         if (call->event.funcs)
2075                 __unregister_trace_event(&call->event);
2076         remove_event_from_tracers(call);
2077         list_del(&call->list);
2078 }
2079
2080 static int event_init(struct trace_event_call *call)
2081 {
2082         int ret = 0;
2083         const char *name;
2084
2085         name = trace_event_name(call);
2086         if (WARN_ON(!name))
2087                 return -EINVAL;
2088
2089         if (call->class->raw_init) {
2090                 ret = call->class->raw_init(call);
2091                 if (ret < 0 && ret != -ENOSYS)
2092                         pr_warn("Could not initialize trace events/%s\n", name);
2093         }
2094
2095         return ret;
2096 }
2097
2098 static int
2099 __register_event(struct trace_event_call *call, struct module *mod)
2100 {
2101         int ret;
2102
2103         ret = event_init(call);
2104         if (ret < 0)
2105                 return ret;
2106
2107         list_add(&call->list, &ftrace_events);
2108         call->mod = mod;
2109
2110         return 0;
2111 }
2112
2113 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2114 {
2115         int rlen;
2116         int elen;
2117
2118         /* Find the length of the eval value as a string */
2119         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2120         /* Make sure there's enough room to replace the string with the value */
2121         if (len < elen)
2122                 return NULL;
2123
2124         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2125
2126         /* Get the rest of the string of ptr */
2127         rlen = strlen(ptr + len);
2128         memmove(ptr + elen, ptr + len, rlen);
2129         /* Make sure we end the new string */
2130         ptr[elen + rlen] = 0;
2131
2132         return ptr + elen;
2133 }
2134
2135 static void update_event_printk(struct trace_event_call *call,
2136                                 struct trace_eval_map *map)
2137 {
2138         char *ptr;
2139         int quote = 0;
2140         int len = strlen(map->eval_string);
2141
2142         for (ptr = call->print_fmt; *ptr; ptr++) {
2143                 if (*ptr == '\\') {
2144                         ptr++;
2145                         /* paranoid */
2146                         if (!*ptr)
2147                                 break;
2148                         continue;
2149                 }
2150                 if (*ptr == '"') {
2151                         quote ^= 1;
2152                         continue;
2153                 }
2154                 if (quote)
2155                         continue;
2156                 if (isdigit(*ptr)) {
2157                         /* skip numbers */
2158                         do {
2159                                 ptr++;
2160                                 /* Check for alpha chars like ULL */
2161                         } while (isalnum(*ptr));
2162                         if (!*ptr)
2163                                 break;
2164                         /*
2165                          * A number must have some kind of delimiter after
2166                          * it, and we can ignore that too.
2167                          */
2168                         continue;
2169                 }
2170                 if (isalpha(*ptr) || *ptr == '_') {
2171                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2172                             !isalnum(ptr[len]) && ptr[len] != '_') {
2173                                 ptr = eval_replace(ptr, map, len);
2174                                 /* enum/sizeof string smaller than value */
2175                                 if (WARN_ON_ONCE(!ptr))
2176                                         return;
2177                                 /*
2178                                  * No need to decrement here, as eval_replace()
2179                                  * returns the pointer to the character passed
2180                                  * the eval, and two evals can not be placed
2181                                  * back to back without something in between.
2182                                  * We can skip that something in between.
2183                                  */
2184                                 continue;
2185                         }
2186                 skip_more:
2187                         do {
2188                                 ptr++;
2189                         } while (isalnum(*ptr) || *ptr == '_');
2190                         if (!*ptr)
2191                                 break;
2192                         /*
2193                          * If what comes after this variable is a '.' or
2194                          * '->' then we can continue to ignore that string.
2195                          */
2196                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2197                                 ptr += *ptr == '.' ? 1 : 2;
2198                                 if (!*ptr)
2199                                         break;
2200                                 goto skip_more;
2201                         }
2202                         /*
2203                          * Once again, we can skip the delimiter that came
2204                          * after the string.
2205                          */
2206                         continue;
2207                 }
2208         }
2209 }
2210
2211 void trace_event_eval_update(struct trace_eval_map **map, int len)
2212 {
2213         struct trace_event_call *call, *p;
2214         const char *last_system = NULL;
2215         bool first = false;
2216         int last_i;
2217         int i;
2218
2219         down_write(&trace_event_sem);
2220         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2221                 /* events are usually grouped together with systems */
2222                 if (!last_system || call->class->system != last_system) {
2223                         first = true;
2224                         last_i = 0;
2225                         last_system = call->class->system;
2226                 }
2227
2228                 /*
2229                  * Since calls are grouped by systems, the likelyhood that the
2230                  * next call in the iteration belongs to the same system as the
2231                  * previous call is high. As an optimization, we skip seaching
2232                  * for a map[] that matches the call's system if the last call
2233                  * was from the same system. That's what last_i is for. If the
2234                  * call has the same system as the previous call, then last_i
2235                  * will be the index of the first map[] that has a matching
2236                  * system.
2237                  */
2238                 for (i = last_i; i < len; i++) {
2239                         if (call->class->system == map[i]->system) {
2240                                 /* Save the first system if need be */
2241                                 if (first) {
2242                                         last_i = i;
2243                                         first = false;
2244                                 }
2245                                 update_event_printk(call, map[i]);
2246                         }
2247                 }
2248         }
2249         up_write(&trace_event_sem);
2250 }
2251
2252 static struct trace_event_file *
2253 trace_create_new_event(struct trace_event_call *call,
2254                        struct trace_array *tr)
2255 {
2256         struct trace_event_file *file;
2257
2258         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2259         if (!file)
2260                 return NULL;
2261
2262         file->event_call = call;
2263         file->tr = tr;
2264         atomic_set(&file->sm_ref, 0);
2265         atomic_set(&file->tm_ref, 0);
2266         INIT_LIST_HEAD(&file->triggers);
2267         list_add(&file->list, &tr->events);
2268
2269         return file;
2270 }
2271
2272 /* Add an event to a trace directory */
2273 static int
2274 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2275 {
2276         struct trace_event_file *file;
2277
2278         file = trace_create_new_event(call, tr);
2279         if (!file)
2280                 return -ENOMEM;
2281
2282         return event_create_dir(tr->event_dir, file);
2283 }
2284
2285 /*
2286  * Just create a decriptor for early init. A descriptor is required
2287  * for enabling events at boot. We want to enable events before
2288  * the filesystem is initialized.
2289  */
2290 static __init int
2291 __trace_early_add_new_event(struct trace_event_call *call,
2292                             struct trace_array *tr)
2293 {
2294         struct trace_event_file *file;
2295
2296         file = trace_create_new_event(call, tr);
2297         if (!file)
2298                 return -ENOMEM;
2299
2300         return 0;
2301 }
2302
2303 struct ftrace_module_file_ops;
2304 static void __add_event_to_tracers(struct trace_event_call *call);
2305
2306 /* Add an additional event_call dynamically */
2307 int trace_add_event_call(struct trace_event_call *call)
2308 {
2309         int ret;
2310         lockdep_assert_held(&event_mutex);
2311
2312         mutex_lock(&trace_types_lock);
2313
2314         ret = __register_event(call, NULL);
2315         if (ret >= 0)
2316                 __add_event_to_tracers(call);
2317
2318         mutex_unlock(&trace_types_lock);
2319         return ret;
2320 }
2321
2322 /*
2323  * Must be called under locking of trace_types_lock, event_mutex and
2324  * trace_event_sem.
2325  */
2326 static void __trace_remove_event_call(struct trace_event_call *call)
2327 {
2328         event_remove(call);
2329         trace_destroy_fields(call);
2330         free_event_filter(call->filter);
2331         call->filter = NULL;
2332 }
2333
2334 static int probe_remove_event_call(struct trace_event_call *call)
2335 {
2336         struct trace_array *tr;
2337         struct trace_event_file *file;
2338
2339 #ifdef CONFIG_PERF_EVENTS
2340         if (call->perf_refcount)
2341                 return -EBUSY;
2342 #endif
2343         do_for_each_event_file(tr, file) {
2344                 if (file->event_call != call)
2345                         continue;
2346                 /*
2347                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2348                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2349                  * TRACE_REG_UNREGISTER.
2350                  */
2351                 if (file->flags & EVENT_FILE_FL_ENABLED)
2352                         return -EBUSY;
2353                 /*
2354                  * The do_for_each_event_file_safe() is
2355                  * a double loop. After finding the call for this
2356                  * trace_array, we use break to jump to the next
2357                  * trace_array.
2358                  */
2359                 break;
2360         } while_for_each_event_file();
2361
2362         __trace_remove_event_call(call);
2363
2364         return 0;
2365 }
2366
2367 /* Remove an event_call */
2368 int trace_remove_event_call(struct trace_event_call *call)
2369 {
2370         int ret;
2371
2372         lockdep_assert_held(&event_mutex);
2373
2374         mutex_lock(&trace_types_lock);
2375         down_write(&trace_event_sem);
2376         ret = probe_remove_event_call(call);
2377         up_write(&trace_event_sem);
2378         mutex_unlock(&trace_types_lock);
2379
2380         return ret;
2381 }
2382
2383 #define for_each_event(event, start, end)                       \
2384         for (event = start;                                     \
2385              (unsigned long)event < (unsigned long)end;         \
2386              event++)
2387
2388 #ifdef CONFIG_MODULES
2389
2390 static void trace_module_add_events(struct module *mod)
2391 {
2392         struct trace_event_call **call, **start, **end;
2393
2394         if (!mod->num_trace_events)
2395                 return;
2396
2397         /* Don't add infrastructure for mods without tracepoints */
2398         if (trace_module_has_bad_taint(mod)) {
2399                 pr_err("%s: module has bad taint, not creating trace events\n",
2400                        mod->name);
2401                 return;
2402         }
2403
2404         start = mod->trace_events;
2405         end = mod->trace_events + mod->num_trace_events;
2406
2407         for_each_event(call, start, end) {
2408                 __register_event(*call, mod);
2409                 __add_event_to_tracers(*call);
2410         }
2411 }
2412
2413 static void trace_module_remove_events(struct module *mod)
2414 {
2415         struct trace_event_call *call, *p;
2416
2417         down_write(&trace_event_sem);
2418         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2419                 if (call->mod == mod)
2420                         __trace_remove_event_call(call);
2421         }
2422         up_write(&trace_event_sem);
2423
2424         /*
2425          * It is safest to reset the ring buffer if the module being unloaded
2426          * registered any events that were used. The only worry is if
2427          * a new module gets loaded, and takes on the same id as the events
2428          * of this module. When printing out the buffer, traced events left
2429          * over from this module may be passed to the new module events and
2430          * unexpected results may occur.
2431          */
2432         tracing_reset_all_online_cpus();
2433 }
2434
2435 static int trace_module_notify(struct notifier_block *self,
2436                                unsigned long val, void *data)
2437 {
2438         struct module *mod = data;
2439
2440         mutex_lock(&event_mutex);
2441         mutex_lock(&trace_types_lock);
2442         switch (val) {
2443         case MODULE_STATE_COMING:
2444                 trace_module_add_events(mod);
2445                 break;
2446         case MODULE_STATE_GOING:
2447                 trace_module_remove_events(mod);
2448                 break;
2449         }
2450         mutex_unlock(&trace_types_lock);
2451         mutex_unlock(&event_mutex);
2452
2453         return 0;
2454 }
2455
2456 static struct notifier_block trace_module_nb = {
2457         .notifier_call = trace_module_notify,
2458         .priority = 1, /* higher than trace.c module notify */
2459 };
2460 #endif /* CONFIG_MODULES */
2461
2462 /* Create a new event directory structure for a trace directory. */
2463 static void
2464 __trace_add_event_dirs(struct trace_array *tr)
2465 {
2466         struct trace_event_call *call;
2467         int ret;
2468
2469         list_for_each_entry(call, &ftrace_events, list) {
2470                 ret = __trace_add_new_event(call, tr);
2471                 if (ret < 0)
2472                         pr_warn("Could not create directory for event %s\n",
2473                                 trace_event_name(call));
2474         }
2475 }
2476
2477 /* Returns any file that matches the system and event */
2478 struct trace_event_file *
2479 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2480 {
2481         struct trace_event_file *file;
2482         struct trace_event_call *call;
2483         const char *name;
2484
2485         list_for_each_entry(file, &tr->events, list) {
2486
2487                 call = file->event_call;
2488                 name = trace_event_name(call);
2489
2490                 if (!name || !call->class)
2491                         continue;
2492
2493                 if (strcmp(event, name) == 0 &&
2494                     strcmp(system, call->class->system) == 0)
2495                         return file;
2496         }
2497         return NULL;
2498 }
2499
2500 /* Returns valid trace event files that match system and event */
2501 struct trace_event_file *
2502 find_event_file(struct trace_array *tr, const char *system, const char *event)
2503 {
2504         struct trace_event_file *file;
2505
2506         file = __find_event_file(tr, system, event);
2507         if (!file || !file->event_call->class->reg ||
2508             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2509                 return NULL;
2510
2511         return file;
2512 }
2513
2514 #ifdef CONFIG_DYNAMIC_FTRACE
2515
2516 /* Avoid typos */
2517 #define ENABLE_EVENT_STR        "enable_event"
2518 #define DISABLE_EVENT_STR       "disable_event"
2519
2520 struct event_probe_data {
2521         struct trace_event_file *file;
2522         unsigned long                   count;
2523         int                             ref;
2524         bool                            enable;
2525 };
2526
2527 static void update_event_probe(struct event_probe_data *data)
2528 {
2529         if (data->enable)
2530                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2531         else
2532                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2533 }
2534
2535 static void
2536 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2537                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2538                    void *data)
2539 {
2540         struct ftrace_func_mapper *mapper = data;
2541         struct event_probe_data *edata;
2542         void **pdata;
2543
2544         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2545         if (!pdata || !*pdata)
2546                 return;
2547
2548         edata = *pdata;
2549         update_event_probe(edata);
2550 }
2551
2552 static void
2553 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2554                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2555                          void *data)
2556 {
2557         struct ftrace_func_mapper *mapper = data;
2558         struct event_probe_data *edata;
2559         void **pdata;
2560
2561         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2562         if (!pdata || !*pdata)
2563                 return;
2564
2565         edata = *pdata;
2566
2567         if (!edata->count)
2568                 return;
2569
2570         /* Skip if the event is in a state we want to switch to */
2571         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2572                 return;
2573
2574         if (edata->count != -1)
2575                 (edata->count)--;
2576
2577         update_event_probe(edata);
2578 }
2579
2580 static int
2581 event_enable_print(struct seq_file *m, unsigned long ip,
2582                    struct ftrace_probe_ops *ops, void *data)
2583 {
2584         struct ftrace_func_mapper *mapper = data;
2585         struct event_probe_data *edata;
2586         void **pdata;
2587
2588         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2589
2590         if (WARN_ON_ONCE(!pdata || !*pdata))
2591                 return 0;
2592
2593         edata = *pdata;
2594
2595         seq_printf(m, "%ps:", (void *)ip);
2596
2597         seq_printf(m, "%s:%s:%s",
2598                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2599                    edata->file->event_call->class->system,
2600                    trace_event_name(edata->file->event_call));
2601
2602         if (edata->count == -1)
2603                 seq_puts(m, ":unlimited\n");
2604         else
2605                 seq_printf(m, ":count=%ld\n", edata->count);
2606
2607         return 0;
2608 }
2609
2610 static int
2611 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2612                   unsigned long ip, void *init_data, void **data)
2613 {
2614         struct ftrace_func_mapper *mapper = *data;
2615         struct event_probe_data *edata = init_data;
2616         int ret;
2617
2618         if (!mapper) {
2619                 mapper = allocate_ftrace_func_mapper();
2620                 if (!mapper)
2621                         return -ENODEV;
2622                 *data = mapper;
2623         }
2624
2625         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2626         if (ret < 0)
2627                 return ret;
2628
2629         edata->ref++;
2630
2631         return 0;
2632 }
2633
2634 static int free_probe_data(void *data)
2635 {
2636         struct event_probe_data *edata = data;
2637
2638         edata->ref--;
2639         if (!edata->ref) {
2640                 /* Remove the SOFT_MODE flag */
2641                 __ftrace_event_enable_disable(edata->file, 0, 1);
2642                 module_put(edata->file->event_call->mod);
2643                 kfree(edata);
2644         }
2645         return 0;
2646 }
2647
2648 static void
2649 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2650                   unsigned long ip, void *data)
2651 {
2652         struct ftrace_func_mapper *mapper = data;
2653         struct event_probe_data *edata;
2654
2655         if (!ip) {
2656                 if (!mapper)
2657                         return;
2658                 free_ftrace_func_mapper(mapper, free_probe_data);
2659                 return;
2660         }
2661
2662         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2663
2664         if (WARN_ON_ONCE(!edata))
2665                 return;
2666
2667         if (WARN_ON_ONCE(edata->ref <= 0))
2668                 return;
2669
2670         free_probe_data(edata);
2671 }
2672
2673 static struct ftrace_probe_ops event_enable_probe_ops = {
2674         .func                   = event_enable_probe,
2675         .print                  = event_enable_print,
2676         .init                   = event_enable_init,
2677         .free                   = event_enable_free,
2678 };
2679
2680 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2681         .func                   = event_enable_count_probe,
2682         .print                  = event_enable_print,
2683         .init                   = event_enable_init,
2684         .free                   = event_enable_free,
2685 };
2686
2687 static struct ftrace_probe_ops event_disable_probe_ops = {
2688         .func                   = event_enable_probe,
2689         .print                  = event_enable_print,
2690         .init                   = event_enable_init,
2691         .free                   = event_enable_free,
2692 };
2693
2694 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2695         .func                   = event_enable_count_probe,
2696         .print                  = event_enable_print,
2697         .init                   = event_enable_init,
2698         .free                   = event_enable_free,
2699 };
2700
2701 static int
2702 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
2703                   char *glob, char *cmd, char *param, int enabled)
2704 {
2705         struct trace_event_file *file;
2706         struct ftrace_probe_ops *ops;
2707         struct event_probe_data *data;
2708         const char *system;
2709         const char *event;
2710         char *number;
2711         bool enable;
2712         int ret;
2713
2714         if (!tr)
2715                 return -ENODEV;
2716
2717         /* hash funcs only work with set_ftrace_filter */
2718         if (!enabled || !param)
2719                 return -EINVAL;
2720
2721         system = strsep(&param, ":");
2722         if (!param)
2723                 return -EINVAL;
2724
2725         event = strsep(&param, ":");
2726
2727         mutex_lock(&event_mutex);
2728
2729         ret = -EINVAL;
2730         file = find_event_file(tr, system, event);
2731         if (!file)
2732                 goto out;
2733
2734         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2735
2736         if (enable)
2737                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2738         else
2739                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2740
2741         if (glob[0] == '!') {
2742                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
2743                 goto out;
2744         }
2745
2746         ret = -ENOMEM;
2747
2748         data = kzalloc(sizeof(*data), GFP_KERNEL);
2749         if (!data)
2750                 goto out;
2751
2752         data->enable = enable;
2753         data->count = -1;
2754         data->file = file;
2755
2756         if (!param)
2757                 goto out_reg;
2758
2759         number = strsep(&param, ":");
2760
2761         ret = -EINVAL;
2762         if (!strlen(number))
2763                 goto out_free;
2764
2765         /*
2766          * We use the callback data field (which is a pointer)
2767          * as our counter.
2768          */
2769         ret = kstrtoul(number, 0, &data->count);
2770         if (ret)
2771                 goto out_free;
2772
2773  out_reg:
2774         /* Don't let event modules unload while probe registered */
2775         ret = try_module_get(file->event_call->mod);
2776         if (!ret) {
2777                 ret = -EBUSY;
2778                 goto out_free;
2779         }
2780
2781         ret = __ftrace_event_enable_disable(file, 1, 1);
2782         if (ret < 0)
2783                 goto out_put;
2784
2785         ret = register_ftrace_function_probe(glob, tr, ops, data);
2786         /*
2787          * The above returns on success the # of functions enabled,
2788          * but if it didn't find any functions it returns zero.
2789          * Consider no functions a failure too.
2790          */
2791         if (!ret) {
2792                 ret = -ENOENT;
2793                 goto out_disable;
2794         } else if (ret < 0)
2795                 goto out_disable;
2796         /* Just return zero, not the number of enabled functions */
2797         ret = 0;
2798  out:
2799         mutex_unlock(&event_mutex);
2800         return ret;
2801
2802  out_disable:
2803         __ftrace_event_enable_disable(file, 0, 1);
2804  out_put:
2805         module_put(file->event_call->mod);
2806  out_free:
2807         kfree(data);
2808         goto out;
2809 }
2810
2811 static struct ftrace_func_command event_enable_cmd = {
2812         .name                   = ENABLE_EVENT_STR,
2813         .func                   = event_enable_func,
2814 };
2815
2816 static struct ftrace_func_command event_disable_cmd = {
2817         .name                   = DISABLE_EVENT_STR,
2818         .func                   = event_enable_func,
2819 };
2820
2821 static __init int register_event_cmds(void)
2822 {
2823         int ret;
2824
2825         ret = register_ftrace_command(&event_enable_cmd);
2826         if (WARN_ON(ret < 0))
2827                 return ret;
2828         ret = register_ftrace_command(&event_disable_cmd);
2829         if (WARN_ON(ret < 0))
2830                 unregister_ftrace_command(&event_enable_cmd);
2831         return ret;
2832 }
2833 #else
2834 static inline int register_event_cmds(void) { return 0; }
2835 #endif /* CONFIG_DYNAMIC_FTRACE */
2836
2837 /*
2838  * The top level array has already had its trace_event_file
2839  * descriptors created in order to allow for early events to
2840  * be recorded. This function is called after the tracefs has been
2841  * initialized, and we now have to create the files associated
2842  * to the events.
2843  */
2844 static __init void
2845 __trace_early_add_event_dirs(struct trace_array *tr)
2846 {
2847         struct trace_event_file *file;
2848         int ret;
2849
2850
2851         list_for_each_entry(file, &tr->events, list) {
2852                 ret = event_create_dir(tr->event_dir, file);
2853                 if (ret < 0)
2854                         pr_warn("Could not create directory for event %s\n",
2855                                 trace_event_name(file->event_call));
2856         }
2857 }
2858
2859 /*
2860  * For early boot up, the top trace array requires to have
2861  * a list of events that can be enabled. This must be done before
2862  * the filesystem is set up in order to allow events to be traced
2863  * early.
2864  */
2865 static __init void
2866 __trace_early_add_events(struct trace_array *tr)
2867 {
2868         struct trace_event_call *call;
2869         int ret;
2870
2871         list_for_each_entry(call, &ftrace_events, list) {
2872                 /* Early boot up should not have any modules loaded */
2873                 if (WARN_ON_ONCE(call->mod))
2874                         continue;
2875
2876                 ret = __trace_early_add_new_event(call, tr);
2877                 if (ret < 0)
2878                         pr_warn("Could not create early event %s\n",
2879                                 trace_event_name(call));
2880         }
2881 }
2882
2883 /* Remove the event directory structure for a trace directory. */
2884 static void
2885 __trace_remove_event_dirs(struct trace_array *tr)
2886 {
2887         struct trace_event_file *file, *next;
2888
2889         list_for_each_entry_safe(file, next, &tr->events, list)
2890                 remove_event_file_dir(file);
2891 }
2892
2893 static void __add_event_to_tracers(struct trace_event_call *call)
2894 {
2895         struct trace_array *tr;
2896
2897         list_for_each_entry(tr, &ftrace_trace_arrays, list)
2898                 __trace_add_new_event(call, tr);
2899 }
2900
2901 extern struct trace_event_call *__start_ftrace_events[];
2902 extern struct trace_event_call *__stop_ftrace_events[];
2903
2904 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2905
2906 static __init int setup_trace_event(char *str)
2907 {
2908         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2909         ring_buffer_expanded = true;
2910         tracing_selftest_disabled = true;
2911
2912         return 1;
2913 }
2914 __setup("trace_event=", setup_trace_event);
2915
2916 /* Expects to have event_mutex held when called */
2917 static int
2918 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2919 {
2920         struct dentry *d_events;
2921         struct dentry *entry;
2922
2923         entry = tracefs_create_file("set_event", 0644, parent,
2924                                     tr, &ftrace_set_event_fops);
2925         if (!entry) {
2926                 pr_warn("Could not create tracefs 'set_event' entry\n");
2927                 return -ENOMEM;
2928         }
2929
2930         d_events = tracefs_create_dir("events", parent);
2931         if (!d_events) {
2932                 pr_warn("Could not create tracefs 'events' directory\n");
2933                 return -ENOMEM;
2934         }
2935
2936         entry = trace_create_file("enable", 0644, d_events,
2937                                   tr, &ftrace_tr_enable_fops);
2938         if (!entry) {
2939                 pr_warn("Could not create tracefs 'enable' entry\n");
2940                 return -ENOMEM;
2941         }
2942
2943         /* There are not as crucial, just warn if they are not created */
2944
2945         entry = tracefs_create_file("set_event_pid", 0644, parent,
2946                                     tr, &ftrace_set_event_pid_fops);
2947         if (!entry)
2948                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
2949
2950         /* ring buffer internal formats */
2951         entry = trace_create_file("header_page", 0444, d_events,
2952                                   ring_buffer_print_page_header,
2953                                   &ftrace_show_header_fops);
2954         if (!entry)
2955                 pr_warn("Could not create tracefs 'header_page' entry\n");
2956
2957         entry = trace_create_file("header_event", 0444, d_events,
2958                                   ring_buffer_print_entry_header,
2959                                   &ftrace_show_header_fops);
2960         if (!entry)
2961                 pr_warn("Could not create tracefs 'header_event' entry\n");
2962
2963         tr->event_dir = d_events;
2964
2965         return 0;
2966 }
2967
2968 /**
2969  * event_trace_add_tracer - add a instance of a trace_array to events
2970  * @parent: The parent dentry to place the files/directories for events in
2971  * @tr: The trace array associated with these events
2972  *
2973  * When a new instance is created, it needs to set up its events
2974  * directory, as well as other files associated with events. It also
2975  * creates the event hierachry in the @parent/events directory.
2976  *
2977  * Returns 0 on success.
2978  *
2979  * Must be called with event_mutex held.
2980  */
2981 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2982 {
2983         int ret;
2984
2985         lockdep_assert_held(&event_mutex);
2986
2987         ret = create_event_toplevel_files(parent, tr);
2988         if (ret)
2989                 goto out;
2990
2991         down_write(&trace_event_sem);
2992         __trace_add_event_dirs(tr);
2993         up_write(&trace_event_sem);
2994
2995  out:
2996         return ret;
2997 }
2998
2999 /*
3000  * The top trace array already had its file descriptors created.
3001  * Now the files themselves need to be created.
3002  */
3003 static __init int
3004 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3005 {
3006         int ret;
3007
3008         mutex_lock(&event_mutex);
3009
3010         ret = create_event_toplevel_files(parent, tr);
3011         if (ret)
3012                 goto out_unlock;
3013
3014         down_write(&trace_event_sem);
3015         __trace_early_add_event_dirs(tr);
3016         up_write(&trace_event_sem);
3017
3018  out_unlock:
3019         mutex_unlock(&event_mutex);
3020
3021         return ret;
3022 }
3023
3024 /* Must be called with event_mutex held */
3025 int event_trace_del_tracer(struct trace_array *tr)
3026 {
3027         lockdep_assert_held(&event_mutex);
3028
3029         /* Disable any event triggers and associated soft-disabled events */
3030         clear_event_triggers(tr);
3031
3032         /* Clear the pid list */
3033         __ftrace_clear_event_pids(tr);
3034
3035         /* Disable any running events */
3036         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3037
3038         /* Make sure no more events are being executed */
3039         tracepoint_synchronize_unregister();
3040
3041         down_write(&trace_event_sem);
3042         __trace_remove_event_dirs(tr);
3043         tracefs_remove_recursive(tr->event_dir);
3044         up_write(&trace_event_sem);
3045
3046         tr->event_dir = NULL;
3047
3048         return 0;
3049 }
3050
3051 static __init int event_trace_memsetup(void)
3052 {
3053         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3054         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3055         return 0;
3056 }
3057
3058 static __init void
3059 early_enable_events(struct trace_array *tr, bool disable_first)
3060 {
3061         char *buf = bootup_event_buf;
3062         char *token;
3063         int ret;
3064
3065         while (true) {
3066                 token = strsep(&buf, ",");
3067
3068                 if (!token)
3069                         break;
3070
3071                 if (*token) {
3072                         /* Restarting syscalls requires that we stop them first */
3073                         if (disable_first)
3074                                 ftrace_set_clr_event(tr, token, 0);
3075
3076                         ret = ftrace_set_clr_event(tr, token, 1);
3077                         if (ret)
3078                                 pr_warn("Failed to enable trace event: %s\n", token);
3079                 }
3080
3081                 /* Put back the comma to allow this to be called again */
3082                 if (buf)
3083                         *(buf - 1) = ',';
3084         }
3085 }
3086
3087 static __init int event_trace_enable(void)
3088 {
3089         struct trace_array *tr = top_trace_array();
3090         struct trace_event_call **iter, *call;
3091         int ret;
3092
3093         if (!tr)
3094                 return -ENODEV;
3095
3096         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3097
3098                 call = *iter;
3099                 ret = event_init(call);
3100                 if (!ret)
3101                         list_add(&call->list, &ftrace_events);
3102         }
3103
3104         /*
3105          * We need the top trace array to have a working set of trace
3106          * points at early init, before the debug files and directories
3107          * are created. Create the file entries now, and attach them
3108          * to the actual file dentries later.
3109          */
3110         __trace_early_add_events(tr);
3111
3112         early_enable_events(tr, false);
3113
3114         trace_printk_start_comm();
3115
3116         register_event_cmds();
3117
3118         register_trigger_cmds();
3119
3120         return 0;
3121 }
3122
3123 /*
3124  * event_trace_enable() is called from trace_event_init() first to
3125  * initialize events and perhaps start any events that are on the
3126  * command line. Unfortunately, there are some events that will not
3127  * start this early, like the system call tracepoints that need
3128  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3129  * is called before pid 1 starts, and this flag is never set, making
3130  * the syscall tracepoint never get reached, but the event is enabled
3131  * regardless (and not doing anything).
3132  */
3133 static __init int event_trace_enable_again(void)
3134 {
3135         struct trace_array *tr;
3136
3137         tr = top_trace_array();
3138         if (!tr)
3139                 return -ENODEV;
3140
3141         early_enable_events(tr, true);
3142
3143         return 0;
3144 }
3145
3146 early_initcall(event_trace_enable_again);
3147
3148 __init int event_trace_init(void)
3149 {
3150         struct trace_array *tr;
3151         struct dentry *d_tracer;
3152         struct dentry *entry;
3153         int ret;
3154
3155         tr = top_trace_array();
3156         if (!tr)
3157                 return -ENODEV;
3158
3159         d_tracer = tracing_init_dentry();
3160         if (IS_ERR(d_tracer))
3161                 return 0;
3162
3163         entry = tracefs_create_file("available_events", 0444, d_tracer,
3164                                     tr, &ftrace_avail_fops);
3165         if (!entry)
3166                 pr_warn("Could not create tracefs 'available_events' entry\n");
3167
3168         if (trace_define_generic_fields())
3169                 pr_warn("tracing: Failed to allocated generic fields");
3170
3171         if (trace_define_common_fields())
3172                 pr_warn("tracing: Failed to allocate common fields");
3173
3174         ret = early_event_add_tracer(d_tracer, tr);
3175         if (ret)
3176                 return ret;
3177
3178 #ifdef CONFIG_MODULES
3179         ret = register_module_notifier(&trace_module_nb);
3180         if (ret)
3181                 pr_warn("Failed to register trace events module notifier\n");
3182 #endif
3183         return 0;
3184 }
3185
3186 void __init trace_event_init(void)
3187 {
3188         event_trace_memsetup();
3189         init_ftrace_syscalls();
3190         event_trace_enable();
3191 }
3192
3193 #ifdef CONFIG_FTRACE_STARTUP_TEST
3194
3195 static DEFINE_SPINLOCK(test_spinlock);
3196 static DEFINE_SPINLOCK(test_spinlock_irq);
3197 static DEFINE_MUTEX(test_mutex);
3198
3199 static __init void test_work(struct work_struct *dummy)
3200 {
3201         spin_lock(&test_spinlock);
3202         spin_lock_irq(&test_spinlock_irq);
3203         udelay(1);
3204         spin_unlock_irq(&test_spinlock_irq);
3205         spin_unlock(&test_spinlock);
3206
3207         mutex_lock(&test_mutex);
3208         msleep(1);
3209         mutex_unlock(&test_mutex);
3210 }
3211
3212 static __init int event_test_thread(void *unused)
3213 {
3214         void *test_malloc;
3215
3216         test_malloc = kmalloc(1234, GFP_KERNEL);
3217         if (!test_malloc)
3218                 pr_info("failed to kmalloc\n");
3219
3220         schedule_on_each_cpu(test_work);
3221
3222         kfree(test_malloc);
3223
3224         set_current_state(TASK_INTERRUPTIBLE);
3225         while (!kthread_should_stop()) {
3226                 schedule();
3227                 set_current_state(TASK_INTERRUPTIBLE);
3228         }
3229         __set_current_state(TASK_RUNNING);
3230
3231         return 0;
3232 }
3233
3234 /*
3235  * Do various things that may trigger events.
3236  */
3237 static __init void event_test_stuff(void)
3238 {
3239         struct task_struct *test_thread;
3240
3241         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3242         msleep(1);
3243         kthread_stop(test_thread);
3244 }
3245
3246 /*
3247  * For every trace event defined, we will test each trace point separately,
3248  * and then by groups, and finally all trace points.
3249  */
3250 static __init void event_trace_self_tests(void)
3251 {
3252         struct trace_subsystem_dir *dir;
3253         struct trace_event_file *file;
3254         struct trace_event_call *call;
3255         struct event_subsystem *system;
3256         struct trace_array *tr;
3257         int ret;
3258
3259         tr = top_trace_array();
3260         if (!tr)
3261                 return;
3262
3263         pr_info("Running tests on trace events:\n");
3264
3265         list_for_each_entry(file, &tr->events, list) {
3266
3267                 call = file->event_call;
3268
3269                 /* Only test those that have a probe */
3270                 if (!call->class || !call->class->probe)
3271                         continue;
3272
3273 /*
3274  * Testing syscall events here is pretty useless, but
3275  * we still do it if configured. But this is time consuming.
3276  * What we really need is a user thread to perform the
3277  * syscalls as we test.
3278  */
3279 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3280                 if (call->class->system &&
3281                     strcmp(call->class->system, "syscalls") == 0)
3282                         continue;
3283 #endif
3284
3285                 pr_info("Testing event %s: ", trace_event_name(call));
3286
3287                 /*
3288                  * If an event is already enabled, someone is using
3289                  * it and the self test should not be on.
3290                  */
3291                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3292                         pr_warn("Enabled event during self test!\n");
3293                         WARN_ON_ONCE(1);
3294                         continue;
3295                 }
3296
3297                 ftrace_event_enable_disable(file, 1);
3298                 event_test_stuff();
3299                 ftrace_event_enable_disable(file, 0);
3300
3301                 pr_cont("OK\n");
3302         }
3303
3304         /* Now test at the sub system level */
3305
3306         pr_info("Running tests on trace event systems:\n");
3307
3308         list_for_each_entry(dir, &tr->systems, list) {
3309
3310                 system = dir->subsystem;
3311
3312                 /* the ftrace system is special, skip it */
3313                 if (strcmp(system->name, "ftrace") == 0)
3314                         continue;
3315
3316                 pr_info("Testing event system %s: ", system->name);
3317
3318                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3319                 if (WARN_ON_ONCE(ret)) {
3320                         pr_warn("error enabling system %s\n",
3321                                 system->name);
3322                         continue;
3323                 }
3324
3325                 event_test_stuff();
3326
3327                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3328                 if (WARN_ON_ONCE(ret)) {
3329                         pr_warn("error disabling system %s\n",
3330                                 system->name);
3331                         continue;
3332                 }
3333
3334                 pr_cont("OK\n");
3335         }
3336
3337         /* Test with all events enabled */
3338
3339         pr_info("Running tests on all trace events:\n");
3340         pr_info("Testing all events: ");
3341
3342         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3343         if (WARN_ON_ONCE(ret)) {
3344                 pr_warn("error enabling all events\n");
3345                 return;
3346         }
3347
3348         event_test_stuff();
3349
3350         /* reset sysname */
3351         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3352         if (WARN_ON_ONCE(ret)) {
3353                 pr_warn("error disabling all events\n");
3354                 return;
3355         }
3356
3357         pr_cont("OK\n");
3358 }
3359
3360 #ifdef CONFIG_FUNCTION_TRACER
3361
3362 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3363
3364 static struct trace_event_file event_trace_file __initdata;
3365
3366 static void __init
3367 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3368                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3369 {
3370         struct ring_buffer_event *event;
3371         struct ring_buffer *buffer;
3372         struct ftrace_entry *entry;
3373         unsigned long flags;
3374         long disabled;
3375         int cpu;
3376         int pc;
3377
3378         pc = preempt_count();
3379         preempt_disable_notrace();
3380         cpu = raw_smp_processor_id();
3381         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3382
3383         if (disabled != 1)
3384                 goto out;
3385
3386         local_save_flags(flags);
3387
3388         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3389                                                 TRACE_FN, sizeof(*entry),
3390                                                 flags, pc);
3391         if (!event)
3392                 goto out;
3393         entry   = ring_buffer_event_data(event);
3394         entry->ip                       = ip;
3395         entry->parent_ip                = parent_ip;
3396
3397         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3398                                     entry, flags, pc);
3399  out:
3400         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3401         preempt_enable_notrace();
3402 }
3403
3404 static struct ftrace_ops trace_ops __initdata  =
3405 {
3406         .func = function_test_events_call,
3407         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3408 };
3409
3410 static __init void event_trace_self_test_with_function(void)
3411 {
3412         int ret;
3413
3414         event_trace_file.tr = top_trace_array();
3415         if (WARN_ON(!event_trace_file.tr))
3416                 return;
3417
3418         ret = register_ftrace_function(&trace_ops);
3419         if (WARN_ON(ret < 0)) {
3420                 pr_info("Failed to enable function tracer for event tests\n");
3421                 return;
3422         }
3423         pr_info("Running tests again, along with the function tracer\n");
3424         event_trace_self_tests();
3425         unregister_ftrace_function(&trace_ops);
3426 }
3427 #else
3428 static __init void event_trace_self_test_with_function(void)
3429 {
3430 }
3431 #endif
3432
3433 static __init int event_trace_self_tests_init(void)
3434 {
3435         if (!tracing_selftest_disabled) {
3436                 event_trace_self_tests();
3437                 event_trace_self_test_with_function();
3438         }
3439
3440         return 0;
3441 }
3442
3443 late_initcall(event_trace_self_tests_init);
3444
3445 #endif