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/security.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/kthread.h>
  18 #include <linux/tracefs.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/module.h>
  21 #include <linux/ctype.h>
  22 #include <linux/sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/delay.h>
  25
  26 #include <trace/events/sched.h>
  27 #include <trace/syscall.h>
  28
  29 #include <asm/setup.h>
  30
  31 #include "trace_output.h"
  32
  33 #undef TRACE_SYSTEM
  34 #define TRACE_SYSTEM "TRACE_SYSTEM"
  35
  36 DEFINE_MUTEX(event_mutex);
  37
  38 LIST_HEAD(ftrace_events);
  39 static LIST_HEAD(ftrace_generic_fields);
  40 static LIST_HEAD(ftrace_common_fields);
  41 static bool eventdir_initialized;
  42
  43 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  44
  45 static struct kmem_cache *field_cachep;
  46 static struct kmem_cache *file_cachep;
  47
  48 static inline int system_refcount(struct event_subsystem *system)
  49 {
  50         return system->ref_count;
  51 }
  52
  53 static int system_refcount_inc(struct event_subsystem *system)
  54 {
  55         return system->ref_count++;
  56 }
  57
  58 static int system_refcount_dec(struct event_subsystem *system)
  59 {
  60         return --system->ref_count;
  61 }
  62
  63 /* Double loops, do not use break, only goto's work */
  64 #define do_for_each_event_file(tr, file)                        \
  65         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  66                 list_for_each_entry(file, &tr->events, list)
  67
  68 #define do_for_each_event_file_safe(tr, file)                   \
  69         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  70                 struct trace_event_file *___n;                          \
  71                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  72
  73 #define while_for_each_event_file()             \
  74         }
  75
  76 static struct ftrace_event_field *
  77 __find_event_field(struct list_head *head, char *name)
  78 {
  79         struct ftrace_event_field *field;
  80
  81         list_for_each_entry(field, head, link) {
  82                 if (!strcmp(field->name, name))
  83                         return field;
  84         }
  85
  86         return NULL;
  87 }
  88
  89 struct ftrace_event_field *
  90 trace_find_event_field(struct trace_event_call *call, char *name)
  91 {
  92         struct ftrace_event_field *field;
  93         struct list_head *head;
  94
  95         head = trace_get_fields(call);
  96         field = __find_event_field(head, name);
  97         if (field)
  98                 return field;
  99
 100         field = __find_event_field(&ftrace_generic_fields, name);
 101         if (field)
 102                 return field;
 103
 104         return __find_event_field(&ftrace_common_fields, name);
 105 }
 106
 107 static int __trace_define_field(struct list_head *head, const char *type,
 108                                 const char *name, int offset, int size,
 109                                 int is_signed, int filter_type)
 110 {
 111         struct ftrace_event_field *field;
 112
 113         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 114         if (!field)
 115                 return -ENOMEM;
 116
 117         field->name = name;
 118         field->type = type;
 119
 120         if (filter_type == FILTER_OTHER)
 121                 field->filter_type = filter_assign_type(type);
 122         else
 123                 field->filter_type = filter_type;
 124
 125         field->offset = offset;
 126         field->size = size;
 127         field->is_signed = is_signed;
 128
 129         list_add(&field->link, head);
 130
 131         return 0;
 132 }
 133
 134 int trace_define_field(struct trace_event_call *call, const char *type,
 135                        const char *name, int offset, int size, int is_signed,
 136                        int filter_type)
 137 {
 138         struct list_head *head;
 139
 140         if (WARN_ON(!call->class))
 141                 return 0;
 142
 143         head = trace_get_fields(call);
 144         return __trace_define_field(head, type, name, offset, size,
 145                                     is_signed, filter_type);
 146 }
 147 EXPORT_SYMBOL_GPL(trace_define_field);
 148
 149 #define __generic_field(type, item, filter_type)                        \
 150         ret = __trace_define_field(&ftrace_generic_fields, #type,       \
 151                                    #item, 0, 0, is_signed_type(type),   \
 152                                    filter_type);                        \
 153         if (ret)                                                        \
 154                 return ret;
 155
 156 #define __common_field(type, item)                                      \
 157         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 158                                    "common_" #item,                     \
 159                                    offsetof(typeof(ent), item),         \
 160                                    sizeof(ent.item),                    \
 161                                    is_signed_type(type), FILTER_OTHER); \
 162         if (ret)                                                        \
 163                 return ret;
 164
 165 static int trace_define_generic_fields(void)
 166 {
 167         int ret;
 168
 169         __generic_field(int, CPU, FILTER_CPU);
 170         __generic_field(int, cpu, FILTER_CPU);
 171         __generic_field(char *, COMM, FILTER_COMM);
 172         __generic_field(char *, comm, FILTER_COMM);
 173
 174         return ret;
 175 }
 176
 177 static int trace_define_common_fields(void)
 178 {
 179         int ret;
 180         struct trace_entry ent;
 181
 182         __common_field(unsigned short, type);
 183         __common_field(unsigned char, flags);
 184         __common_field(unsigned char, preempt_count);
 185         __common_field(int, pid);
 186
 187         return ret;
 188 }
 189
 190 static void trace_destroy_fields(struct trace_event_call *call)
 191 {
 192         struct ftrace_event_field *field, *next;
 193         struct list_head *head;
 194
 195         head = trace_get_fields(call);
 196         list_for_each_entry_safe(field, next, head, link) {
 197                 list_del(&field->link);
 198                 kmem_cache_free(field_cachep, field);
 199         }
 200 }
 201
 202 /*
 203  * run-time version of trace_event_get_offsets_<call>() that returns the last
 204  * accessible offset of trace fields excluding __dynamic_array bytes
 205  */
 206 int trace_event_get_offsets(struct trace_event_call *call)
 207 {
 208         struct ftrace_event_field *tail;
 209         struct list_head *head;
 210
 211         head = trace_get_fields(call);
 212         /*
 213          * head->next points to the last field with the largest offset,
 214          * since it was added last by trace_define_field()
 215          */
 216         tail = list_first_entry(head, struct ftrace_event_field, link);
 217         return tail->offset + tail->size;
 218 }
 219
 220 int trace_event_raw_init(struct trace_event_call *call)
 221 {
 222         int id;
 223
 224         id = register_trace_event(&call->event);
 225         if (!id)
 226                 return -ENODEV;
 227
 228         return 0;
 229 }
 230 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 231
 232 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 233 {
 234         struct trace_array *tr = trace_file->tr;
 235         struct trace_array_cpu *data;
 236         struct trace_pid_list *no_pid_list;
 237         struct trace_pid_list *pid_list;
 238
 239         pid_list = rcu_dereference_raw(tr->filtered_pids);
 240         no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 241
 242         if (!pid_list && !no_pid_list)
 243                 return false;
 244
 245         data = this_cpu_ptr(tr->array_buffer.data);
 246
 247         return data->ignore_pid;
 248 }
 249 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 250
 251 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 252                                  struct trace_event_file *trace_file,
 253                                  unsigned long len)
 254 {
 255         struct trace_event_call *event_call = trace_file->event_call;
 256
 257         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 258             trace_event_ignore_this_pid(trace_file))
 259                 return NULL;
 260
 261         local_save_flags(fbuffer->flags);
 262         fbuffer->pc = preempt_count();
 263         /*
 264          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 265          * preemption (adding one to the preempt_count). Since we are
 266          * interested in the preempt_count at the time the tracepoint was
 267          * hit, we need to subtract one to offset the increment.
 268          */
 269         if (IS_ENABLED(CONFIG_PREEMPTION))
 270                 fbuffer->pc--;
 271         fbuffer->trace_file = trace_file;
 272
 273         fbuffer->event =
 274                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 275                                                 event_call->event.type, len,
 276                                                 fbuffer->flags, fbuffer->pc);
 277         if (!fbuffer->event)
 278                 return NULL;
 279
 280         fbuffer->regs = NULL;
 281         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 282         return fbuffer->entry;
 283 }
 284 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 285
 286 int trace_event_reg(struct trace_event_call *call,
 287                     enum trace_reg type, void *data)
 288 {
 289         struct trace_event_file *file = data;
 290
 291         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 292         switch (type) {
 293         case TRACE_REG_REGISTER:
 294                 return tracepoint_probe_register(call->tp,
 295                                                  call->class->probe,
 296                                                  file);
 297         case TRACE_REG_UNREGISTER:
 298                 tracepoint_probe_unregister(call->tp,
 299                                             call->class->probe,
 300                                             file);
 301                 return 0;
 302
 303 #ifdef CONFIG_PERF_EVENTS
 304         case TRACE_REG_PERF_REGISTER:
 305                 return tracepoint_probe_register(call->tp,
 306                                                  call->class->perf_probe,
 307                                                  call);
 308         case TRACE_REG_PERF_UNREGISTER:
 309                 tracepoint_probe_unregister(call->tp,
 310                                             call->class->perf_probe,
 311                                             call);
 312                 return 0;
 313         case TRACE_REG_PERF_OPEN:
 314         case TRACE_REG_PERF_CLOSE:
 315         case TRACE_REG_PERF_ADD:
 316         case TRACE_REG_PERF_DEL:
 317                 return 0;
 318 #endif
 319         }
 320         return 0;
 321 }
 322 EXPORT_SYMBOL_GPL(trace_event_reg);
 323
 324 void trace_event_enable_cmd_record(bool enable)
 325 {
 326         struct trace_event_file *file;
 327         struct trace_array *tr;
 328
 329         lockdep_assert_held(&event_mutex);
 330
 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 }
 345
 346 void trace_event_enable_tgid_record(bool enable)
 347 {
 348         struct trace_event_file *file;
 349         struct trace_array *tr;
 350
 351         lockdep_assert_held(&event_mutex);
 352
 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 }
 367
 368 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 369                                          int enable, int soft_disable)
 370 {
 371         struct trace_event_call *call = file->event_call;
 372         struct trace_array *tr = file->tr;
 373         unsigned long file_flags = file->flags;
 374         int ret = 0;
 375         int disable;
 376
 377         switch (enable) {
 378         case 0:
 379                 /*
 380                  * When soft_disable is set and enable is cleared, the sm_ref
 381                  * reference counter is decremented. If it reaches 0, we want
 382                  * to clear the SOFT_DISABLED flag but leave the event in the
 383                  * state that it was. That is, if the event was enabled and
 384                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 385                  * is set we do not want the event to be enabled before we
 386                  * clear the bit.
 387                  *
 388                  * When soft_disable is not set but the SOFT_MODE flag is,
 389                  * we do nothing. Do not disable the tracepoint, otherwise
 390                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 391                  */
 392                 if (soft_disable) {
 393                         if (atomic_dec_return(&file->sm_ref) > 0)
 394                                 break;
 395                         disable = file->flags & EVENT_FILE_FL_SOFT_DISABLED;
 396                         clear_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 397                 } else
 398                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 399
 400                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 401                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 402                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 403                                 tracing_stop_cmdline_record();
 404                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 405                         }
 406
 407                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 408                                 tracing_stop_tgid_record();
 409                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 410                         }
 411
 412                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 413                 }
 414                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 415                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 416                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 417                 else
 418                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 419                 break;
 420         case 1:
 421                 /*
 422                  * When soft_disable is set and enable is set, we want to
 423                  * register the tracepoint for the event, but leave the event
 424                  * as is. That means, if the event was already enabled, we do
 425                  * nothing (but set SOFT_MODE). If the event is disabled, we
 426                  * set SOFT_DISABLED before enabling the event tracepoint, so
 427                  * it still seems to be disabled.
 428                  */
 429                 if (!soft_disable)
 430                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 431                 else {
 432                         if (atomic_inc_return(&file->sm_ref) > 1)
 433                                 break;
 434                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 435                 }
 436
 437                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 438                         bool cmd = false, tgid = false;
 439
 440                         /* Keep the event disabled, when going to SOFT_MODE. */
 441                         if (soft_disable)
 442                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 443
 444                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 445                                 cmd = true;
 446                                 tracing_start_cmdline_record();
 447                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 448                         }
 449
 450                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 451                                 tgid = true;
 452                                 tracing_start_tgid_record();
 453                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 454                         }
 455
 456                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 457                         if (ret) {
 458                                 if (cmd)
 459                                         tracing_stop_cmdline_record();
 460                                 if (tgid)
 461                                         tracing_stop_tgid_record();
 462                                 pr_info("event trace: Could not enable event "
 463                                         "%s\n", trace_event_name(call));
 464                                 break;
 465                         }
 466                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 467
 468                         /* WAS_ENABLED gets set but never cleared. */
 469                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 470                 }
 471                 break;
 472         }
 473
 474         /* Enable or disable use of trace_buffered_event */
 475         if ((file_flags & EVENT_FILE_FL_SOFT_DISABLED) !=
 476             (file->flags & EVENT_FILE_FL_SOFT_DISABLED)) {
 477                 if (file->flags & EVENT_FILE_FL_SOFT_DISABLED)
 478                         trace_buffered_event_enable();
 479                 else
 480                         trace_buffered_event_disable();
 481         }
 482
 483         return ret;
 484 }
 485
 486 int trace_event_enable_disable(struct trace_event_file *file,
 487                                int enable, int soft_disable)
 488 {
 489         return __ftrace_event_enable_disable(file, enable, soft_disable);
 490 }
 491
 492 static int ftrace_event_enable_disable(struct trace_event_file *file,
 493                                        int enable)
 494 {
 495         return __ftrace_event_enable_disable(file, enable, 0);
 496 }
 497
 498 static void ftrace_clear_events(struct trace_array *tr)
 499 {
 500         struct trace_event_file *file;
 501
 502         mutex_lock(&event_mutex);
 503         list_for_each_entry(file, &tr->events, list) {
 504                 ftrace_event_enable_disable(file, 0);
 505         }
 506         mutex_unlock(&event_mutex);
 507 }
 508
 509 static void
 510 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 511 {
 512         struct trace_pid_list *pid_list;
 513         struct trace_array *tr = data;
 514
 515         pid_list = rcu_dereference_raw(tr->filtered_pids);
 516         trace_filter_add_remove_task(pid_list, NULL, task);
 517
 518         pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 519         trace_filter_add_remove_task(pid_list, NULL, task);
 520 }
 521
 522 static void
 523 event_filter_pid_sched_process_fork(void *data,
 524                                     struct task_struct *self,
 525                                     struct task_struct *task)
 526 {
 527         struct trace_pid_list *pid_list;
 528         struct trace_array *tr = data;
 529
 530         pid_list = rcu_dereference_sched(tr->filtered_pids);
 531         trace_filter_add_remove_task(pid_list, self, task);
 532
 533         pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 534         trace_filter_add_remove_task(pid_list, self, task);
 535 }
 536
 537 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 538 {
 539         if (enable) {
 540                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 541                                                        tr, INT_MIN);
 542                 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 543                                                        tr, INT_MAX);
 544         } else {
 545                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 546                                                     tr);
 547                 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 548                                                     tr);
 549         }
 550 }
 551
 552 static void
 553 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 554                     struct task_struct *prev, struct task_struct *next)
 555 {
 556         struct trace_array *tr = data;
 557         struct trace_pid_list *no_pid_list;
 558         struct trace_pid_list *pid_list;
 559         bool ret;
 560
 561         pid_list = rcu_dereference_sched(tr->filtered_pids);
 562         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 563
 564         /*
 565          * Sched switch is funny, as we only want to ignore it
 566          * in the notrace case if both prev and next should be ignored.
 567          */
 568         ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 569                 trace_ignore_this_task(NULL, no_pid_list, next);
 570
 571         this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 572                        (trace_ignore_this_task(pid_list, NULL, prev) &&
 573                         trace_ignore_this_task(pid_list, NULL, next)));
 574 }
 575
 576 static void
 577 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 578                     struct task_struct *prev, struct task_struct *next)
 579 {
 580         struct trace_array *tr = data;
 581         struct trace_pid_list *no_pid_list;
 582         struct trace_pid_list *pid_list;
 583
 584         pid_list = rcu_dereference_sched(tr->filtered_pids);
 585         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 586
 587         this_cpu_write(tr->array_buffer.data->ignore_pid,
 588                        trace_ignore_this_task(pid_list, no_pid_list, next));
 589 }
 590
 591 static void
 592 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 593 {
 594         struct trace_array *tr = data;
 595         struct trace_pid_list *no_pid_list;
 596         struct trace_pid_list *pid_list;
 597
 598         /* Nothing to do if we are already tracing */
 599         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 600                 return;
 601
 602         pid_list = rcu_dereference_sched(tr->filtered_pids);
 603         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 604
 605         this_cpu_write(tr->array_buffer.data->ignore_pid,
 606                        trace_ignore_this_task(pid_list, no_pid_list, task));
 607 }
 608
 609 static void
 610 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 611 {
 612         struct trace_array *tr = data;
 613         struct trace_pid_list *no_pid_list;
 614         struct trace_pid_list *pid_list;
 615
 616         /* Nothing to do if we are not tracing */
 617         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 618                 return;
 619
 620         pid_list = rcu_dereference_sched(tr->filtered_pids);
 621         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 622
 623         /* Set tracing if current is enabled */
 624         this_cpu_write(tr->array_buffer.data->ignore_pid,
 625                        trace_ignore_this_task(pid_list, no_pid_list, current));
 626 }
 627
 628 static void unregister_pid_events(struct trace_array *tr)
 629 {
 630         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 631         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 632
 633         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 634         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 635
 636         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 637         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 638
 639         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 640         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 641 }
 642
 643 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 644 {
 645         struct trace_pid_list *pid_list;
 646         struct trace_pid_list *no_pid_list;
 647         struct trace_event_file *file;
 648         int cpu;
 649
 650         pid_list = rcu_dereference_protected(tr->filtered_pids,
 651                                              lockdep_is_held(&event_mutex));
 652         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 653                                              lockdep_is_held(&event_mutex));
 654
 655         /* Make sure there's something to do */
 656         if (!pid_type_enabled(type, pid_list, no_pid_list))
 657                 return;
 658
 659         if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 660                 unregister_pid_events(tr);
 661
 662                 list_for_each_entry(file, &tr->events, list) {
 663                         clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 664                 }
 665
 666                 for_each_possible_cpu(cpu)
 667                         per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 668         }
 669
 670         if (type & TRACE_PIDS)
 671                 rcu_assign_pointer(tr->filtered_pids, NULL);
 672
 673         if (type & TRACE_NO_PIDS)
 674                 rcu_assign_pointer(tr->filtered_no_pids, NULL);
 675
 676         /* Wait till all users are no longer using pid filtering */
 677         tracepoint_synchronize_unregister();
 678
 679         if ((type & TRACE_PIDS) && pid_list)
 680                 trace_free_pid_list(pid_list);
 681
 682         if ((type & TRACE_NO_PIDS) && no_pid_list)
 683                 trace_free_pid_list(no_pid_list);
 684 }
 685
 686 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 687 {
 688         mutex_lock(&event_mutex);
 689         __ftrace_clear_event_pids(tr, type);
 690         mutex_unlock(&event_mutex);
 691 }
 692
 693 static void __put_system(struct event_subsystem *system)
 694 {
 695         struct event_filter *filter = system->filter;
 696
 697         WARN_ON_ONCE(system_refcount(system) == 0);
 698         if (system_refcount_dec(system))
 699                 return;
 700
 701         list_del(&system->list);
 702
 703         if (filter) {
 704                 kfree(filter->filter_string);
 705                 kfree(filter);
 706         }
 707         kfree_const(system->name);
 708         kfree(system);
 709 }
 710
 711 static void __get_system(struct event_subsystem *system)
 712 {
 713         WARN_ON_ONCE(system_refcount(system) == 0);
 714         system_refcount_inc(system);
 715 }
 716
 717 static void __get_system_dir(struct trace_subsystem_dir *dir)
 718 {
 719         WARN_ON_ONCE(dir->ref_count == 0);
 720         dir->ref_count++;
 721         __get_system(dir->subsystem);
 722 }
 723
 724 static void __put_system_dir(struct trace_subsystem_dir *dir)
 725 {
 726         WARN_ON_ONCE(dir->ref_count == 0);
 727         /* If the subsystem is about to be freed, the dir must be too */
 728         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 729
 730         __put_system(dir->subsystem);
 731         if (!--dir->ref_count)
 732                 kfree(dir);
 733 }
 734
 735 static void put_system(struct trace_subsystem_dir *dir)
 736 {
 737         mutex_lock(&event_mutex);
 738         __put_system_dir(dir);
 739         mutex_unlock(&event_mutex);
 740 }
 741
 742 static void remove_subsystem(struct trace_subsystem_dir *dir)
 743 {
 744         if (!dir)
 745                 return;
 746
 747         if (!--dir->nr_events) {
 748                 tracefs_remove(dir->entry);
 749                 list_del(&dir->list);
 750                 __put_system_dir(dir);
 751         }
 752 }
 753
 754 static void remove_event_file_dir(struct trace_event_file *file)
 755 {
 756         struct dentry *dir = file->dir;
 757         struct dentry *child;
 758
 759         if (dir) {
 760                 spin_lock(&dir->d_lock);        /* probably unneeded */
 761                 list_for_each_entry(child, &dir->d_subdirs, d_child) {
 762                         if (d_really_is_positive(child))        /* probably unneeded */
 763                                 d_inode(child)->i_private = NULL;
 764                 }
 765                 spin_unlock(&dir->d_lock);
 766
 767                 tracefs_remove(dir);
 768         }
 769
 770         list_del(&file->list);
 771         remove_subsystem(file->system);
 772         free_event_filter(file->filter);
 773         kmem_cache_free(file_cachep, file);
 774 }
 775
 776 /*
 777  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 778  */
 779 static int
 780 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 781                               const char *sub, const char *event, int set)
 782 {
 783         struct trace_event_file *file;
 784         struct trace_event_call *call;
 785         const char *name;
 786         int ret = -EINVAL;
 787         int eret = 0;
 788
 789         list_for_each_entry(file, &tr->events, list) {
 790
 791                 call = file->event_call;
 792                 name = trace_event_name(call);
 793
 794                 if (!name || !call->class || !call->class->reg)
 795                         continue;
 796
 797                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 798                         continue;
 799
 800                 if (match &&
 801                     strcmp(match, name) != 0 &&
 802                     strcmp(match, call->class->system) != 0)
 803                         continue;
 804
 805                 if (sub && strcmp(sub, call->class->system) != 0)
 806                         continue;
 807
 808                 if (event && strcmp(event, name) != 0)
 809                         continue;
 810
 811                 ret = ftrace_event_enable_disable(file, set);
 812
 813                 /*
 814                  * Save the first error and return that. Some events
 815                  * may still have been enabled, but let the user
 816                  * know that something went wrong.
 817                  */
 818                 if (ret && !eret)
 819                         eret = ret;
 820
 821                 ret = eret;
 822         }
 823
 824         return ret;
 825 }
 826
 827 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 828                                   const char *sub, const char *event, int set)
 829 {
 830         int ret;
 831
 832         mutex_lock(&event_mutex);
 833         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 834         mutex_unlock(&event_mutex);
 835
 836         return ret;
 837 }
 838
 839 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 840 {
 841         char *event = NULL, *sub = NULL, *match;
 842         int ret;
 843
 844         if (!tr)
 845                 return -ENOENT;
 846         /*
 847          * The buf format can be <subsystem>:<event-name>
 848          *  *:<event-name> means any event by that name.
 849          *  :<event-name> is the same.
 850          *
 851          *  <subsystem>:* means all events in that subsystem
 852          *  <subsystem>: means the same.
 853          *
 854          *  <name> (no ':') means all events in a subsystem with
 855          *  the name <name> or any event that matches <name>
 856          */
 857
 858         match = strsep(&buf, ":");
 859         if (buf) {
 860                 sub = match;
 861                 event = buf;
 862                 match = NULL;
 863
 864                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 865                         sub = NULL;
 866                 if (!strlen(event) || strcmp(event, "*") == 0)
 867                         event = NULL;
 868         }
 869
 870         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 871
 872         /* Put back the colon to allow this to be called again */
 873         if (buf)
 874                 *(buf - 1) = ':';
 875
 876         return ret;
 877 }
 878
 879 /**
 880  * trace_set_clr_event - enable or disable an event
 881  * @system: system name to match (NULL for any system)
 882  * @event: event name to match (NULL for all events, within system)
 883  * @set: 1 to enable, 0 to disable
 884  *
 885  * This is a way for other parts of the kernel to enable or disable
 886  * event recording.
 887  *
 888  * Returns 0 on success, -EINVAL if the parameters do not match any
 889  * registered events.
 890  */
 891 int trace_set_clr_event(const char *system, const char *event, int set)
 892 {
 893         struct trace_array *tr = top_trace_array();
 894
 895         if (!tr)
 896                 return -ENODEV;
 897
 898         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 899 }
 900 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 901
 902 /**
 903  * trace_array_set_clr_event - enable or disable an event for a trace array.
 904  * @tr: concerned trace array.
 905  * @system: system name to match (NULL for any system)
 906  * @event: event name to match (NULL for all events, within system)
 907  * @enable: true to enable, false to disable
 908  *
 909  * This is a way for other parts of the kernel to enable or disable
 910  * event recording.
 911  *
 912  * Returns 0 on success, -EINVAL if the parameters do not match any
 913  * registered events.
 914  */
 915 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
 916                 const char *event, bool enable)
 917 {
 918         int set;
 919
 920         if (!tr)
 921                 return -ENOENT;
 922
 923         set = (enable == true) ? 1 : 0;
 924         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 925 }
 926 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
 927
 928 /* 128 should be much more than enough */
 929 #define EVENT_BUF_SIZE          127
 930
 931 static ssize_t
 932 ftrace_event_write(struct file *file, const char __user *ubuf,
 933                    size_t cnt, loff_t *ppos)
 934 {
 935         struct trace_parser parser;
 936         struct seq_file *m = file->private_data;
 937         struct trace_array *tr = m->private;
 938         ssize_t read, ret;
 939
 940         if (!cnt)
 941                 return 0;
 942
 943         ret = tracing_update_buffers();
 944         if (ret < 0)
 945                 return ret;
 946
 947         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 948                 return -ENOMEM;
 949
 950         read = trace_get_user(&parser, ubuf, cnt, ppos);
 951
 952         if (read >= 0 && trace_parser_loaded((&parser))) {
 953                 int set = 1;
 954
 955                 if (*parser.buffer == '!')
 956                         set = 0;
 957
 958                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 959                 if (ret)
 960                         goto out_put;
 961         }
 962
 963         ret = read;
 964
 965  out_put:
 966         trace_parser_put(&parser);
 967
 968         return ret;
 969 }
 970
 971 static void *
 972 t_next(struct seq_file *m, void *v, loff_t *pos)
 973 {
 974         struct trace_event_file *file = v;
 975         struct trace_event_call *call;
 976         struct trace_array *tr = m->private;
 977
 978         (*pos)++;
 979
 980         list_for_each_entry_continue(file, &tr->events, list) {
 981                 call = file->event_call;
 982                 /*
 983                  * The ftrace subsystem is for showing formats only.
 984                  * They can not be enabled or disabled via the event files.
 985                  */
 986                 if (call->class && call->class->reg &&
 987                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 988                         return file;
 989         }
 990
 991         return NULL;
 992 }
 993
 994 static void *t_start(struct seq_file *m, loff_t *pos)
 995 {
 996         struct trace_event_file *file;
 997         struct trace_array *tr = m->private;
 998         loff_t l;
 999
1000         mutex_lock(&event_mutex);
1001
1002         file = list_entry(&tr->events, struct trace_event_file, list);
1003         for (l = 0; l <= *pos; ) {
1004                 file = t_next(m, file, &l);
1005                 if (!file)
1006                         break;
1007         }
1008         return file;
1009 }
1010
1011 static void *
1012 s_next(struct seq_file *m, void *v, loff_t *pos)
1013 {
1014         struct trace_event_file *file = v;
1015         struct trace_array *tr = m->private;
1016
1017         (*pos)++;
1018
1019         list_for_each_entry_continue(file, &tr->events, list) {
1020                 if (file->flags & EVENT_FILE_FL_ENABLED)
1021                         return file;
1022         }
1023
1024         return NULL;
1025 }
1026
1027 static void *s_start(struct seq_file *m, loff_t *pos)
1028 {
1029         struct trace_event_file *file;
1030         struct trace_array *tr = m->private;
1031         loff_t l;
1032
1033         mutex_lock(&event_mutex);
1034
1035         file = list_entry(&tr->events, struct trace_event_file, list);
1036         for (l = 0; l <= *pos; ) {
1037                 file = s_next(m, file, &l);
1038                 if (!file)
1039                         break;
1040         }
1041         return file;
1042 }
1043
1044 static int t_show(struct seq_file *m, void *v)
1045 {
1046         struct trace_event_file *file = v;
1047         struct trace_event_call *call = file->event_call;
1048
1049         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1050                 seq_printf(m, "%s:", call->class->system);
1051         seq_printf(m, "%s\n", trace_event_name(call));
1052
1053         return 0;
1054 }
1055
1056 static void t_stop(struct seq_file *m, void *p)
1057 {
1058         mutex_unlock(&event_mutex);
1059 }
1060
1061 static void *
1062 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1063 {
1064         struct trace_array *tr = m->private;
1065         struct trace_pid_list *pid_list;
1066
1067         if (type == TRACE_PIDS)
1068                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1069         else
1070                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1071
1072         return trace_pid_next(pid_list, v, pos);
1073 }
1074
1075 static void *
1076 p_next(struct seq_file *m, void *v, loff_t *pos)
1077 {
1078         return __next(m, v, pos, TRACE_PIDS);
1079 }
1080
1081 static void *
1082 np_next(struct seq_file *m, void *v, loff_t *pos)
1083 {
1084         return __next(m, v, pos, TRACE_NO_PIDS);
1085 }
1086
1087 static void *__start(struct seq_file *m, loff_t *pos, int type)
1088         __acquires(RCU)
1089 {
1090         struct trace_pid_list *pid_list;
1091         struct trace_array *tr = m->private;
1092
1093         /*
1094          * Grab the mutex, to keep calls to p_next() having the same
1095          * tr->filtered_pids as p_start() has.
1096          * If we just passed the tr->filtered_pids around, then RCU would
1097          * have been enough, but doing that makes things more complex.
1098          */
1099         mutex_lock(&event_mutex);
1100         rcu_read_lock_sched();
1101
1102         if (type == TRACE_PIDS)
1103                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1104         else
1105                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1106
1107         if (!pid_list)
1108                 return NULL;
1109
1110         return trace_pid_start(pid_list, pos);
1111 }
1112
1113 static void *p_start(struct seq_file *m, loff_t *pos)
1114         __acquires(RCU)
1115 {
1116         return __start(m, pos, TRACE_PIDS);
1117 }
1118
1119 static void *np_start(struct seq_file *m, loff_t *pos)
1120         __acquires(RCU)
1121 {
1122         return __start(m, pos, TRACE_NO_PIDS);
1123 }
1124
1125 static void p_stop(struct seq_file *m, void *p)
1126         __releases(RCU)
1127 {
1128         rcu_read_unlock_sched();
1129         mutex_unlock(&event_mutex);
1130 }
1131
1132 static ssize_t
1133 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1134                   loff_t *ppos)
1135 {
1136         struct trace_event_file *file;
1137         unsigned long flags;
1138         char buf[4] = "0";
1139
1140         mutex_lock(&event_mutex);
1141         file = event_file_data(filp);
1142         if (likely(file))
1143                 flags = file->flags;
1144         mutex_unlock(&event_mutex);
1145
1146         if (!file)
1147                 return -ENODEV;
1148
1149         if (flags & EVENT_FILE_FL_ENABLED &&
1150             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1151                 strcpy(buf, "1");
1152
1153         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1154             flags & EVENT_FILE_FL_SOFT_MODE)
1155                 strcat(buf, "*");
1156
1157         strcat(buf, "\n");
1158
1159         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1160 }
1161
1162 static ssize_t
1163 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1164                    loff_t *ppos)
1165 {
1166         struct trace_event_file *file;
1167         unsigned long val;
1168         int ret;
1169
1170         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1171         if (ret)
1172                 return ret;
1173
1174         ret = tracing_update_buffers();
1175         if (ret < 0)
1176                 return ret;
1177
1178         switch (val) {
1179         case 0:
1180         case 1:
1181                 ret = -ENODEV;
1182                 mutex_lock(&event_mutex);
1183                 file = event_file_data(filp);
1184                 if (likely(file))
1185                         ret = ftrace_event_enable_disable(file, val);
1186                 mutex_unlock(&event_mutex);
1187                 break;
1188
1189         default:
1190                 return -EINVAL;
1191         }
1192
1193         *ppos += cnt;
1194
1195         return ret ? ret : cnt;
1196 }
1197
1198 static ssize_t
1199 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1200                    loff_t *ppos)
1201 {
1202         const char set_to_char[4] = { '?', '0', '1', 'X' };
1203         struct trace_subsystem_dir *dir = filp->private_data;
1204         struct event_subsystem *system = dir->subsystem;
1205         struct trace_event_call *call;
1206         struct trace_event_file *file;
1207         struct trace_array *tr = dir->tr;
1208         char buf[2];
1209         int set = 0;
1210         int ret;
1211
1212         mutex_lock(&event_mutex);
1213         list_for_each_entry(file, &tr->events, list) {
1214                 call = file->event_call;
1215                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1216                     !trace_event_name(call) || !call->class || !call->class->reg)
1217                         continue;
1218
1219                 if (system && strcmp(call->class->system, system->name) != 0)
1220                         continue;
1221
1222                 /*
1223                  * We need to find out if all the events are set
1224                  * or if all events or cleared, or if we have
1225                  * a mixture.
1226                  */
1227                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1228
1229                 /*
1230                  * If we have a mixture, no need to look further.
1231                  */
1232                 if (set == 3)
1233                         break;
1234         }
1235         mutex_unlock(&event_mutex);
1236
1237         buf[0] = set_to_char[set];
1238         buf[1] = '\n';
1239
1240         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1241
1242         return ret;
1243 }
1244
1245 static ssize_t
1246 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1247                     loff_t *ppos)
1248 {
1249         struct trace_subsystem_dir *dir = filp->private_data;
1250         struct event_subsystem *system = dir->subsystem;
1251         const char *name = NULL;
1252         unsigned long val;
1253         ssize_t ret;
1254
1255         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1256         if (ret)
1257                 return ret;
1258
1259         ret = tracing_update_buffers();
1260         if (ret < 0)
1261                 return ret;
1262
1263         if (val != 0 && val != 1)
1264                 return -EINVAL;
1265
1266         /*
1267          * Opening of "enable" adds a ref count to system,
1268          * so the name is safe to use.
1269          */
1270         if (system)
1271                 name = system->name;
1272
1273         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1274         if (ret)
1275                 goto out;
1276
1277         ret = cnt;
1278
1279 out:
1280         *ppos += cnt;
1281
1282         return ret;
1283 }
1284
1285 enum {
1286         FORMAT_HEADER           = 1,
1287         FORMAT_FIELD_SEPERATOR  = 2,
1288         FORMAT_PRINTFMT         = 3,
1289 };
1290
1291 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1292 {
1293         struct trace_event_call *call = event_file_data(m->private);
1294         struct list_head *common_head = &ftrace_common_fields;
1295         struct list_head *head = trace_get_fields(call);
1296         struct list_head *node = v;
1297
1298         (*pos)++;
1299
1300         switch ((unsigned long)v) {
1301         case FORMAT_HEADER:
1302                 node = common_head;
1303                 break;
1304
1305         case FORMAT_FIELD_SEPERATOR:
1306                 node = head;
1307                 break;
1308
1309         case FORMAT_PRINTFMT:
1310                 /* all done */
1311                 return NULL;
1312         }
1313
1314         node = node->prev;
1315         if (node == common_head)
1316                 return (void *)FORMAT_FIELD_SEPERATOR;
1317         else if (node == head)
1318                 return (void *)FORMAT_PRINTFMT;
1319         else
1320                 return node;
1321 }
1322
1323 static int f_show(struct seq_file *m, void *v)
1324 {
1325         struct trace_event_call *call = event_file_data(m->private);
1326         struct ftrace_event_field *field;
1327         const char *array_descriptor;
1328
1329         switch ((unsigned long)v) {
1330         case FORMAT_HEADER:
1331                 seq_printf(m, "name: %s\n", trace_event_name(call));
1332                 seq_printf(m, "ID: %d\n", call->event.type);
1333                 seq_puts(m, "format:\n");
1334                 return 0;
1335
1336         case FORMAT_FIELD_SEPERATOR:
1337                 seq_putc(m, '\n');
1338                 return 0;
1339
1340         case FORMAT_PRINTFMT:
1341                 seq_printf(m, "\nprint fmt: %s\n",
1342                            call->print_fmt);
1343                 return 0;
1344         }
1345
1346         field = list_entry(v, struct ftrace_event_field, link);
1347         /*
1348          * Smartly shows the array type(except dynamic array).
1349          * Normal:
1350          *      field:TYPE VAR
1351          * If TYPE := TYPE[LEN], it is shown:
1352          *      field:TYPE VAR[LEN]
1353          */
1354         array_descriptor = strchr(field->type, '[');
1355
1356         if (str_has_prefix(field->type, "__data_loc"))
1357                 array_descriptor = NULL;
1358
1359         if (!array_descriptor)
1360                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1361                            field->type, field->name, field->offset,
1362                            field->size, !!field->is_signed);
1363         else
1364                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1365                            (int)(array_descriptor - field->type),
1366                            field->type, field->name,
1367                            array_descriptor, field->offset,
1368                            field->size, !!field->is_signed);
1369
1370         return 0;
1371 }
1372
1373 static void *f_start(struct seq_file *m, loff_t *pos)
1374 {
1375         void *p = (void *)FORMAT_HEADER;
1376         loff_t l = 0;
1377
1378         /* ->stop() is called even if ->start() fails */
1379         mutex_lock(&event_mutex);
1380         if (!event_file_data(m->private))
1381                 return ERR_PTR(-ENODEV);
1382
1383         while (l < *pos && p)
1384                 p = f_next(m, p, &l);
1385
1386         return p;
1387 }
1388
1389 static void f_stop(struct seq_file *m, void *p)
1390 {
1391         mutex_unlock(&event_mutex);
1392 }
1393
1394 static const struct seq_operations trace_format_seq_ops = {
1395         .start          = f_start,
1396         .next           = f_next,
1397         .stop           = f_stop,
1398         .show           = f_show,
1399 };
1400
1401 static int trace_format_open(struct inode *inode, struct file *file)
1402 {
1403         struct seq_file *m;
1404         int ret;
1405
1406         /* Do we want to hide event format files on tracefs lockdown? */
1407
1408         ret = seq_open(file, &trace_format_seq_ops);
1409         if (ret < 0)
1410                 return ret;
1411
1412         m = file->private_data;
1413         m->private = file;
1414
1415         return 0;
1416 }
1417
1418 static ssize_t
1419 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1420 {
1421         int id = (long)event_file_data(filp);
1422         char buf[32];
1423         int len;
1424
1425         if (unlikely(!id))
1426                 return -ENODEV;
1427
1428         len = sprintf(buf, "%d\n", id);
1429
1430         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1431 }
1432
1433 static ssize_t
1434 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1435                   loff_t *ppos)
1436 {
1437         struct trace_event_file *file;
1438         struct trace_seq *s;
1439         int r = -ENODEV;
1440
1441         if (*ppos)
1442                 return 0;
1443
1444         s = kmalloc(sizeof(*s), GFP_KERNEL);
1445
1446         if (!s)
1447                 return -ENOMEM;
1448
1449         trace_seq_init(s);
1450
1451         mutex_lock(&event_mutex);
1452         file = event_file_data(filp);
1453         if (file)
1454                 print_event_filter(file, s);
1455         mutex_unlock(&event_mutex);
1456
1457         if (file)
1458                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1459                                             s->buffer, trace_seq_used(s));
1460
1461         kfree(s);
1462
1463         return r;
1464 }
1465
1466 static ssize_t
1467 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1468                    loff_t *ppos)
1469 {
1470         struct trace_event_file *file;
1471         char *buf;
1472         int err = -ENODEV;
1473
1474         if (cnt >= PAGE_SIZE)
1475                 return -EINVAL;
1476
1477         buf = memdup_user_nul(ubuf, cnt);
1478         if (IS_ERR(buf))
1479                 return PTR_ERR(buf);
1480
1481         mutex_lock(&event_mutex);
1482         file = event_file_data(filp);
1483         if (file)
1484                 err = apply_event_filter(file, buf);
1485         mutex_unlock(&event_mutex);
1486
1487         kfree(buf);
1488         if (err < 0)
1489                 return err;
1490
1491         *ppos += cnt;
1492
1493         return cnt;
1494 }
1495
1496 static LIST_HEAD(event_subsystems);
1497
1498 static int subsystem_open(struct inode *inode, struct file *filp)
1499 {
1500         struct event_subsystem *system = NULL;
1501         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1502         struct trace_array *tr;
1503         int ret;
1504
1505         if (tracing_is_disabled())
1506                 return -ENODEV;
1507
1508         /* Make sure the system still exists */
1509         mutex_lock(&event_mutex);
1510         mutex_lock(&trace_types_lock);
1511         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1512                 list_for_each_entry(dir, &tr->systems, list) {
1513                         if (dir == inode->i_private) {
1514                                 /* Don't open systems with no events */
1515                                 if (dir->nr_events) {
1516                                         __get_system_dir(dir);
1517                                         system = dir->subsystem;
1518                                 }
1519                                 goto exit_loop;
1520                         }
1521                 }
1522         }
1523  exit_loop:
1524         mutex_unlock(&trace_types_lock);
1525         mutex_unlock(&event_mutex);
1526
1527         if (!system)
1528                 return -ENODEV;
1529
1530         /* Some versions of gcc think dir can be uninitialized here */
1531         WARN_ON(!dir);
1532
1533         /* Still need to increment the ref count of the system */
1534         if (trace_array_get(tr) < 0) {
1535                 put_system(dir);
1536                 return -ENODEV;
1537         }
1538
1539         ret = tracing_open_generic(inode, filp);
1540         if (ret < 0) {
1541                 trace_array_put(tr);
1542                 put_system(dir);
1543         }
1544
1545         return ret;
1546 }
1547
1548 static int system_tr_open(struct inode *inode, struct file *filp)
1549 {
1550         struct trace_subsystem_dir *dir;
1551         struct trace_array *tr = inode->i_private;
1552         int ret;
1553
1554         /* Make a temporary dir that has no system but points to tr */
1555         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1556         if (!dir)
1557                 return -ENOMEM;
1558
1559         ret = tracing_open_generic_tr(inode, filp);
1560         if (ret < 0) {
1561                 kfree(dir);
1562                 return ret;
1563         }
1564         dir->tr = tr;
1565         filp->private_data = dir;
1566
1567         return 0;
1568 }
1569
1570 static int subsystem_release(struct inode *inode, struct file *file)
1571 {
1572         struct trace_subsystem_dir *dir = file->private_data;
1573
1574         trace_array_put(dir->tr);
1575
1576         /*
1577          * If dir->subsystem is NULL, then this is a temporary
1578          * descriptor that was made for a trace_array to enable
1579          * all subsystems.
1580          */
1581         if (dir->subsystem)
1582                 put_system(dir);
1583         else
1584                 kfree(dir);
1585
1586         return 0;
1587 }
1588
1589 static ssize_t
1590 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1591                       loff_t *ppos)
1592 {
1593         struct trace_subsystem_dir *dir = filp->private_data;
1594         struct event_subsystem *system = dir->subsystem;
1595         struct trace_seq *s;
1596         int r;
1597
1598         if (*ppos)
1599                 return 0;
1600
1601         s = kmalloc(sizeof(*s), GFP_KERNEL);
1602         if (!s)
1603                 return -ENOMEM;
1604
1605         trace_seq_init(s);
1606
1607         print_subsystem_event_filter(system, s);
1608         r = simple_read_from_buffer(ubuf, cnt, ppos,
1609                                     s->buffer, trace_seq_used(s));
1610
1611         kfree(s);
1612
1613         return r;
1614 }
1615
1616 static ssize_t
1617 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1618                        loff_t *ppos)
1619 {
1620         struct trace_subsystem_dir *dir = filp->private_data;
1621         char *buf;
1622         int err;
1623
1624         if (cnt >= PAGE_SIZE)
1625                 return -EINVAL;
1626
1627         buf = memdup_user_nul(ubuf, cnt);
1628         if (IS_ERR(buf))
1629                 return PTR_ERR(buf);
1630
1631         err = apply_subsystem_event_filter(dir, buf);
1632         kfree(buf);
1633         if (err < 0)
1634                 return err;
1635
1636         *ppos += cnt;
1637
1638         return cnt;
1639 }
1640
1641 static ssize_t
1642 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1643 {
1644         int (*func)(struct trace_seq *s) = filp->private_data;
1645         struct trace_seq *s;
1646         int r;
1647
1648         if (*ppos)
1649                 return 0;
1650
1651         s = kmalloc(sizeof(*s), GFP_KERNEL);
1652         if (!s)
1653                 return -ENOMEM;
1654
1655         trace_seq_init(s);
1656
1657         func(s);
1658         r = simple_read_from_buffer(ubuf, cnt, ppos,
1659                                     s->buffer, trace_seq_used(s));
1660
1661         kfree(s);
1662
1663         return r;
1664 }
1665
1666 static void ignore_task_cpu(void *data)
1667 {
1668         struct trace_array *tr = data;
1669         struct trace_pid_list *pid_list;
1670         struct trace_pid_list *no_pid_list;
1671
1672         /*
1673          * This function is called by on_each_cpu() while the
1674          * event_mutex is held.
1675          */
1676         pid_list = rcu_dereference_protected(tr->filtered_pids,
1677                                              mutex_is_locked(&event_mutex));
1678         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1679                                              mutex_is_locked(&event_mutex));
1680
1681         this_cpu_write(tr->array_buffer.data->ignore_pid,
1682                        trace_ignore_this_task(pid_list, no_pid_list, current));
1683 }
1684
1685 static void register_pid_events(struct trace_array *tr)
1686 {
1687         /*
1688          * Register a probe that is called before all other probes
1689          * to set ignore_pid if next or prev do not match.
1690          * Register a probe this is called after all other probes
1691          * to only keep ignore_pid set if next pid matches.
1692          */
1693         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1694                                          tr, INT_MAX);
1695         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1696                                          tr, 0);
1697
1698         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1699                                          tr, INT_MAX);
1700         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1701                                          tr, 0);
1702
1703         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1704                                              tr, INT_MAX);
1705         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1706                                              tr, 0);
1707
1708         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1709                                          tr, INT_MAX);
1710         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1711                                          tr, 0);
1712 }
1713
1714 static ssize_t
1715 event_pid_write(struct file *filp, const char __user *ubuf,
1716                 size_t cnt, loff_t *ppos, int type)
1717 {
1718         struct seq_file *m = filp->private_data;
1719         struct trace_array *tr = m->private;
1720         struct trace_pid_list *filtered_pids = NULL;
1721         struct trace_pid_list *other_pids = NULL;
1722         struct trace_pid_list *pid_list;
1723         struct trace_event_file *file;
1724         ssize_t ret;
1725
1726         if (!cnt)
1727                 return 0;
1728
1729         ret = tracing_update_buffers();
1730         if (ret < 0)
1731                 return ret;
1732
1733         mutex_lock(&event_mutex);
1734
1735         if (type == TRACE_PIDS) {
1736                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1737                                                           lockdep_is_held(&event_mutex));
1738                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1739                                                           lockdep_is_held(&event_mutex));
1740         } else {
1741                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1742                                                           lockdep_is_held(&event_mutex));
1743                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1744                                                           lockdep_is_held(&event_mutex));
1745         }
1746
1747         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1748         if (ret < 0)
1749                 goto out;
1750
1751         if (type == TRACE_PIDS)
1752                 rcu_assign_pointer(tr->filtered_pids, pid_list);
1753         else
1754                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1755
1756         list_for_each_entry(file, &tr->events, list) {
1757                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1758         }
1759
1760         if (filtered_pids) {
1761                 tracepoint_synchronize_unregister();
1762                 trace_free_pid_list(filtered_pids);
1763         } else if (pid_list && !other_pids) {
1764                 register_pid_events(tr);
1765         }
1766
1767         /*
1768          * Ignoring of pids is done at task switch. But we have to
1769          * check for those tasks that are currently running.
1770          * Always do this in case a pid was appended or removed.
1771          */
1772         on_each_cpu(ignore_task_cpu, tr, 1);
1773
1774  out:
1775         mutex_unlock(&event_mutex);
1776
1777         if (ret > 0)
1778                 *ppos += ret;
1779
1780         return ret;
1781 }
1782
1783 static ssize_t
1784 ftrace_event_pid_write(struct file *filp, const char __user *ubuf,
1785                        size_t cnt, loff_t *ppos)
1786 {
1787         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
1788 }
1789
1790 static ssize_t
1791 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
1792                         size_t cnt, loff_t *ppos)
1793 {
1794         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
1795 }
1796
1797 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1798 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1799 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1800 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
1801 static int ftrace_event_release(struct inode *inode, struct file *file);
1802
1803 static const struct seq_operations show_event_seq_ops = {
1804         .start = t_start,
1805         .next = t_next,
1806         .show = t_show,
1807         .stop = t_stop,
1808 };
1809
1810 static const struct seq_operations show_set_event_seq_ops = {
1811         .start = s_start,
1812         .next = s_next,
1813         .show = t_show,
1814         .stop = t_stop,
1815 };
1816
1817 static const struct seq_operations show_set_pid_seq_ops = {
1818         .start = p_start,
1819         .next = p_next,
1820         .show = trace_pid_show,
1821         .stop = p_stop,
1822 };
1823
1824 static const struct seq_operations show_set_no_pid_seq_ops = {
1825         .start = np_start,
1826         .next = np_next,
1827         .show = trace_pid_show,
1828         .stop = p_stop,
1829 };
1830
1831 static const struct file_operations ftrace_avail_fops = {
1832         .open = ftrace_event_avail_open,
1833         .read = seq_read,
1834         .llseek = seq_lseek,
1835         .release = seq_release,
1836 };
1837
1838 static const struct file_operations ftrace_set_event_fops = {
1839         .open = ftrace_event_set_open,
1840         .read = seq_read,
1841         .write = ftrace_event_write,
1842         .llseek = seq_lseek,
1843         .release = ftrace_event_release,
1844 };
1845
1846 static const struct file_operations ftrace_set_event_pid_fops = {
1847         .open = ftrace_event_set_pid_open,
1848         .read = seq_read,
1849         .write = ftrace_event_pid_write,
1850         .llseek = seq_lseek,
1851         .release = ftrace_event_release,
1852 };
1853
1854 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
1855         .open = ftrace_event_set_npid_open,
1856         .read = seq_read,
1857         .write = ftrace_event_npid_write,
1858         .llseek = seq_lseek,
1859         .release = ftrace_event_release,
1860 };
1861
1862 static const struct file_operations ftrace_enable_fops = {
1863         .open = tracing_open_generic,
1864         .read = event_enable_read,
1865         .write = event_enable_write,
1866         .llseek = default_llseek,
1867 };
1868
1869 static const struct file_operations ftrace_event_format_fops = {
1870         .open = trace_format_open,
1871         .read = seq_read,
1872         .llseek = seq_lseek,
1873         .release = seq_release,
1874 };
1875
1876 static const struct file_operations ftrace_event_id_fops = {
1877         .read = event_id_read,
1878         .llseek = default_llseek,
1879 };
1880
1881 static const struct file_operations ftrace_event_filter_fops = {
1882         .open = tracing_open_generic,
1883         .read = event_filter_read,
1884         .write = event_filter_write,
1885         .llseek = default_llseek,
1886 };
1887
1888 static const struct file_operations ftrace_subsystem_filter_fops = {
1889         .open = subsystem_open,
1890         .read = subsystem_filter_read,
1891         .write = subsystem_filter_write,
1892         .llseek = default_llseek,
1893         .release = subsystem_release,
1894 };
1895
1896 static const struct file_operations ftrace_system_enable_fops = {
1897         .open = subsystem_open,
1898         .read = system_enable_read,
1899         .write = system_enable_write,
1900         .llseek = default_llseek,
1901         .release = subsystem_release,
1902 };
1903
1904 static const struct file_operations ftrace_tr_enable_fops = {
1905         .open = system_tr_open,
1906         .read = system_enable_read,
1907         .write = system_enable_write,
1908         .llseek = default_llseek,
1909         .release = subsystem_release,
1910 };
1911
1912 static const struct file_operations ftrace_show_header_fops = {
1913         .open = tracing_open_generic,
1914         .read = show_header,
1915         .llseek = default_llseek,
1916 };
1917
1918 static int
1919 ftrace_event_open(struct inode *inode, struct file *file,
1920                   const struct seq_operations *seq_ops)
1921 {
1922         struct seq_file *m;
1923         int ret;
1924
1925         ret = security_locked_down(LOCKDOWN_TRACEFS);
1926         if (ret)
1927                 return ret;
1928
1929         ret = seq_open(file, seq_ops);
1930         if (ret < 0)
1931                 return ret;
1932         m = file->private_data;
1933         /* copy tr over to seq ops */
1934         m->private = inode->i_private;
1935
1936         return ret;
1937 }
1938
1939 static int ftrace_event_release(struct inode *inode, struct file *file)
1940 {
1941         struct trace_array *tr = inode->i_private;
1942
1943         trace_array_put(tr);
1944
1945         return seq_release(inode, file);
1946 }
1947
1948 static int
1949 ftrace_event_avail_open(struct inode *inode, struct file *file)
1950 {
1951         const struct seq_operations *seq_ops = &show_event_seq_ops;
1952
1953         /* Checks for tracefs lockdown */
1954         return ftrace_event_open(inode, file, seq_ops);
1955 }
1956
1957 static int
1958 ftrace_event_set_open(struct inode *inode, struct file *file)
1959 {
1960         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1961         struct trace_array *tr = inode->i_private;
1962         int ret;
1963
1964         ret = tracing_check_open_get_tr(tr);
1965         if (ret)
1966                 return ret;
1967
1968         if ((file->f_mode & FMODE_WRITE) &&
1969             (file->f_flags & O_TRUNC))
1970                 ftrace_clear_events(tr);
1971
1972         ret = ftrace_event_open(inode, file, seq_ops);
1973         if (ret < 0)
1974                 trace_array_put(tr);
1975         return ret;
1976 }
1977
1978 static int
1979 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1980 {
1981         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1982         struct trace_array *tr = inode->i_private;
1983         int ret;
1984
1985         ret = tracing_check_open_get_tr(tr);
1986         if (ret)
1987                 return ret;
1988
1989         if ((file->f_mode & FMODE_WRITE) &&
1990             (file->f_flags & O_TRUNC))
1991                 ftrace_clear_event_pids(tr, TRACE_PIDS);
1992
1993         ret = ftrace_event_open(inode, file, seq_ops);
1994         if (ret < 0)
1995                 trace_array_put(tr);
1996         return ret;
1997 }
1998
1999 static int
2000 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2001 {
2002         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2003         struct trace_array *tr = inode->i_private;
2004         int ret;
2005
2006         ret = tracing_check_open_get_tr(tr);
2007         if (ret)
2008                 return ret;
2009
2010         if ((file->f_mode & FMODE_WRITE) &&
2011             (file->f_flags & O_TRUNC))
2012                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2013
2014         ret = ftrace_event_open(inode, file, seq_ops);
2015         if (ret < 0)
2016                 trace_array_put(tr);
2017         return ret;
2018 }
2019
2020 static struct event_subsystem *
2021 create_new_subsystem(const char *name)
2022 {
2023         struct event_subsystem *system;
2024
2025         /* need to create new entry */
2026         system = kmalloc(sizeof(*system), GFP_KERNEL);
2027         if (!system)
2028                 return NULL;
2029
2030         system->ref_count = 1;
2031
2032         /* Only allocate if dynamic (kprobes and modules) */
2033         system->name = kstrdup_const(name, GFP_KERNEL);
2034         if (!system->name)
2035                 goto out_free;
2036
2037         system->filter = NULL;
2038
2039         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2040         if (!system->filter)
2041                 goto out_free;
2042
2043         list_add(&system->list, &event_subsystems);
2044
2045         return system;
2046
2047  out_free:
2048         kfree_const(system->name);
2049         kfree(system);
2050         return NULL;
2051 }
2052
2053 static struct dentry *
2054 event_subsystem_dir(struct trace_array *tr, const char *name,
2055                     struct trace_event_file *file, struct dentry *parent)
2056 {
2057         struct trace_subsystem_dir *dir;
2058         struct event_subsystem *system;
2059         struct dentry *entry;
2060
2061         /* First see if we did not already create this dir */
2062         list_for_each_entry(dir, &tr->systems, list) {
2063                 system = dir->subsystem;
2064                 if (strcmp(system->name, name) == 0) {
2065                         dir->nr_events++;
2066                         file->system = dir;
2067                         return dir->entry;
2068                 }
2069         }
2070
2071         /* Now see if the system itself exists. */
2072         list_for_each_entry(system, &event_subsystems, list) {
2073                 if (strcmp(system->name, name) == 0)
2074                         break;
2075         }
2076         /* Reset system variable when not found */
2077         if (&system->list == &event_subsystems)
2078                 system = NULL;
2079
2080         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2081         if (!dir)
2082                 goto out_fail;
2083
2084         if (!system) {
2085                 system = create_new_subsystem(name);
2086                 if (!system)
2087                         goto out_free;
2088         } else
2089                 __get_system(system);
2090
2091         dir->entry = tracefs_create_dir(name, parent);
2092         if (!dir->entry) {
2093                 pr_warn("Failed to create system directory %s\n", name);
2094                 __put_system(system);
2095                 goto out_free;
2096         }
2097
2098         dir->tr = tr;
2099         dir->ref_count = 1;
2100         dir->nr_events = 1;
2101         dir->subsystem = system;
2102         file->system = dir;
2103
2104         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2105                                     &ftrace_subsystem_filter_fops);
2106         if (!entry) {
2107                 kfree(system->filter);
2108                 system->filter = NULL;
2109                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2110         }
2111
2112         trace_create_file("enable", 0644, dir->entry, dir,
2113                           &ftrace_system_enable_fops);
2114
2115         list_add(&dir->list, &tr->systems);
2116
2117         return dir->entry;
2118
2119  out_free:
2120         kfree(dir);
2121  out_fail:
2122         /* Only print this message if failed on memory allocation */
2123         if (!dir || !system)
2124                 pr_warn("No memory to create event subsystem %s\n", name);
2125         return NULL;
2126 }
2127
2128 static int
2129 event_define_fields(struct trace_event_call *call)
2130 {
2131         struct list_head *head;
2132         int ret = 0;
2133
2134         /*
2135          * Other events may have the same class. Only update
2136          * the fields if they are not already defined.
2137          */
2138         head = trace_get_fields(call);
2139         if (list_empty(head)) {
2140                 struct trace_event_fields *field = call->class->fields_array;
2141                 unsigned int offset = sizeof(struct trace_entry);
2142
2143                 for (; field->type; field++) {
2144                         if (field->type == TRACE_FUNCTION_TYPE) {
2145                                 field->define_fields(call);
2146                                 break;
2147                         }
2148
2149                         offset = ALIGN(offset, field->align);
2150                         ret = trace_define_field(call, field->type, field->name,
2151                                                  offset, field->size,
2152                                                  field->is_signed, field->filter_type);
2153                         if (WARN_ON_ONCE(ret)) {
2154                                 pr_err("error code is %d\n", ret);
2155                                 break;
2156                         }
2157
2158                         offset += field->size;
2159                 }
2160         }
2161
2162         return ret;
2163 }
2164
2165 static int
2166 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2167 {
2168         struct trace_event_call *call = file->event_call;
2169         struct trace_array *tr = file->tr;
2170         struct dentry *d_events;
2171         const char *name;
2172         int ret;
2173
2174         /*
2175          * If the trace point header did not define TRACE_SYSTEM
2176          * then the system would be called "TRACE_SYSTEM".
2177          */
2178         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2179                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2180                 if (!d_events)
2181                         return -ENOMEM;
2182         } else
2183                 d_events = parent;
2184
2185         name = trace_event_name(call);
2186         file->dir = tracefs_create_dir(name, d_events);
2187         if (!file->dir) {
2188                 pr_warn("Could not create tracefs '%s' directory\n", name);
2189                 return -1;
2190         }
2191
2192         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2193                 trace_create_file("enable", 0644, file->dir, file,
2194                                   &ftrace_enable_fops);
2195
2196 #ifdef CONFIG_PERF_EVENTS
2197         if (call->event.type && call->class->reg)
2198                 trace_create_file("id", 0444, file->dir,
2199                                   (void *)(long)call->event.type,
2200                                   &ftrace_event_id_fops);
2201 #endif
2202
2203         ret = event_define_fields(call);
2204         if (ret < 0) {
2205                 pr_warn("Could not initialize trace point events/%s\n", name);
2206                 return ret;
2207         }
2208
2209         /*
2210          * Only event directories that can be enabled should have
2211          * triggers or filters.
2212          */
2213         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2214                 trace_create_file("filter", 0644, file->dir, file,
2215                                   &ftrace_event_filter_fops);
2216
2217                 trace_create_file("trigger", 0644, file->dir, file,
2218                                   &event_trigger_fops);
2219         }
2220
2221 #ifdef CONFIG_HIST_TRIGGERS
2222         trace_create_file("hist", 0444, file->dir, file,
2223                           &event_hist_fops);
2224 #endif
2225 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2226         trace_create_file("hist_debug", 0444, file->dir, file,
2227                           &event_hist_debug_fops);
2228 #endif
2229         trace_create_file("format", 0444, file->dir, call,
2230                           &ftrace_event_format_fops);
2231
2232 #ifdef CONFIG_TRACE_EVENT_INJECT
2233         if (call->event.type && call->class->reg)
2234                 trace_create_file("inject", 0200, file->dir, file,
2235                                   &event_inject_fops);
2236 #endif
2237
2238         return 0;
2239 }
2240
2241 static void remove_event_from_tracers(struct trace_event_call *call)
2242 {
2243         struct trace_event_file *file;
2244         struct trace_array *tr;
2245
2246         do_for_each_event_file_safe(tr, file) {
2247                 if (file->event_call != call)
2248                         continue;
2249
2250                 remove_event_file_dir(file);
2251                 /*
2252                  * The do_for_each_event_file_safe() is
2253                  * a double loop. After finding the call for this
2254                  * trace_array, we use break to jump to the next
2255                  * trace_array.
2256                  */
2257                 break;
2258         } while_for_each_event_file();
2259 }
2260
2261 static void event_remove(struct trace_event_call *call)
2262 {
2263         struct trace_array *tr;
2264         struct trace_event_file *file;
2265
2266         do_for_each_event_file(tr, file) {
2267                 if (file->event_call != call)
2268                         continue;
2269
2270                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2271                         tr->clear_trace = true;
2272
2273                 ftrace_event_enable_disable(file, 0);
2274                 /*
2275                  * The do_for_each_event_file() is
2276                  * a double loop. After finding the call for this
2277                  * trace_array, we use break to jump to the next
2278                  * trace_array.
2279                  */
2280                 break;
2281         } while_for_each_event_file();
2282
2283         if (call->event.funcs)
2284                 __unregister_trace_event(&call->event);
2285         remove_event_from_tracers(call);
2286         list_del(&call->list);
2287 }
2288
2289 static int event_init(struct trace_event_call *call)
2290 {
2291         int ret = 0;
2292         const char *name;
2293
2294         name = trace_event_name(call);
2295         if (WARN_ON(!name))
2296                 return -EINVAL;
2297
2298         if (call->class->raw_init) {
2299                 ret = call->class->raw_init(call);
2300                 if (ret < 0 && ret != -ENOSYS)
2301                         pr_warn("Could not initialize trace events/%s\n", name);
2302         }
2303
2304         return ret;
2305 }
2306
2307 static int
2308 __register_event(struct trace_event_call *call, struct module *mod)
2309 {
2310         int ret;
2311
2312         ret = event_init(call);
2313         if (ret < 0)
2314                 return ret;
2315
2316         list_add(&call->list, &ftrace_events);
2317         call->mod = mod;
2318
2319         return 0;
2320 }
2321
2322 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2323 {
2324         int rlen;
2325         int elen;
2326
2327         /* Find the length of the eval value as a string */
2328         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2329         /* Make sure there's enough room to replace the string with the value */
2330         if (len < elen)
2331                 return NULL;
2332
2333         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2334
2335         /* Get the rest of the string of ptr */
2336         rlen = strlen(ptr + len);
2337         memmove(ptr + elen, ptr + len, rlen);
2338         /* Make sure we end the new string */
2339         ptr[elen + rlen] = 0;
2340
2341         return ptr + elen;
2342 }
2343
2344 static void update_event_printk(struct trace_event_call *call,
2345                                 struct trace_eval_map *map)
2346 {
2347         char *ptr;
2348         int quote = 0;
2349         int len = strlen(map->eval_string);
2350
2351         for (ptr = call->print_fmt; *ptr; ptr++) {
2352                 if (*ptr == '\\') {
2353                         ptr++;
2354                         /* paranoid */
2355                         if (!*ptr)
2356                                 break;
2357                         continue;
2358                 }
2359                 if (*ptr == '"') {
2360                         quote ^= 1;
2361                         continue;
2362                 }
2363                 if (quote)
2364                         continue;
2365                 if (isdigit(*ptr)) {
2366                         /* skip numbers */
2367                         do {
2368                                 ptr++;
2369                                 /* Check for alpha chars like ULL */
2370                         } while (isalnum(*ptr));
2371                         if (!*ptr)
2372                                 break;
2373                         /*
2374                          * A number must have some kind of delimiter after
2375                          * it, and we can ignore that too.
2376                          */
2377                         continue;
2378                 }
2379                 if (isalpha(*ptr) || *ptr == '_') {
2380                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2381                             !isalnum(ptr[len]) && ptr[len] != '_') {
2382                                 ptr = eval_replace(ptr, map, len);
2383                                 /* enum/sizeof string smaller than value */
2384                                 if (WARN_ON_ONCE(!ptr))
2385                                         return;
2386                                 /*
2387                                  * No need to decrement here, as eval_replace()
2388                                  * returns the pointer to the character passed
2389                                  * the eval, and two evals can not be placed
2390                                  * back to back without something in between.
2391                                  * We can skip that something in between.
2392                                  */
2393                                 continue;
2394                         }
2395                 skip_more:
2396                         do {
2397                                 ptr++;
2398                         } while (isalnum(*ptr) || *ptr == '_');
2399                         if (!*ptr)
2400                                 break;
2401                         /*
2402                          * If what comes after this variable is a '.' or
2403                          * '->' then we can continue to ignore that string.
2404                          */
2405                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2406                                 ptr += *ptr == '.' ? 1 : 2;
2407                                 if (!*ptr)
2408                                         break;
2409                                 goto skip_more;
2410                         }
2411                         /*
2412                          * Once again, we can skip the delimiter that came
2413                          * after the string.
2414                          */
2415                         continue;
2416                 }
2417         }
2418 }
2419
2420 void trace_event_eval_update(struct trace_eval_map **map, int len)
2421 {
2422         struct trace_event_call *call, *p;
2423         const char *last_system = NULL;
2424         bool first = false;
2425         int last_i;
2426         int i;
2427
2428         down_write(&trace_event_sem);
2429         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2430                 /* events are usually grouped together with systems */
2431                 if (!last_system || call->class->system != last_system) {
2432                         first = true;
2433                         last_i = 0;
2434                         last_system = call->class->system;
2435                 }
2436
2437                 /*
2438                  * Since calls are grouped by systems, the likelyhood that the
2439                  * next call in the iteration belongs to the same system as the
2440                  * previous call is high. As an optimization, we skip searching
2441                  * for a map[] that matches the call's system if the last call
2442                  * was from the same system. That's what last_i is for. If the
2443                  * call has the same system as the previous call, then last_i
2444                  * will be the index of the first map[] that has a matching
2445                  * system.
2446                  */
2447                 for (i = last_i; i < len; i++) {
2448                         if (call->class->system == map[i]->system) {
2449                                 /* Save the first system if need be */
2450                                 if (first) {
2451                                         last_i = i;
2452                                         first = false;
2453                                 }
2454                                 update_event_printk(call, map[i]);
2455                         }
2456                 }
2457         }
2458         up_write(&trace_event_sem);
2459 }
2460
2461 static struct trace_event_file *
2462 trace_create_new_event(struct trace_event_call *call,
2463                        struct trace_array *tr)
2464 {
2465         struct trace_event_file *file;
2466
2467         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2468         if (!file)
2469                 return NULL;
2470
2471         file->event_call = call;
2472         file->tr = tr;
2473         atomic_set(&file->sm_ref, 0);
2474         atomic_set(&file->tm_ref, 0);
2475         INIT_LIST_HEAD(&file->triggers);
2476         list_add(&file->list, &tr->events);
2477
2478         return file;
2479 }
2480
2481 /* Add an event to a trace directory */
2482 static int
2483 __trace_add_new_event(struct trace_event_call *call, struct trace_array *tr)
2484 {
2485         struct trace_event_file *file;
2486
2487         file = trace_create_new_event(call, tr);
2488         if (!file)
2489                 return -ENOMEM;
2490
2491         if (eventdir_initialized)
2492                 return event_create_dir(tr->event_dir, file);
2493         else
2494                 return event_define_fields(call);
2495 }
2496
2497 /*
2498  * Just create a decriptor for early init. A descriptor is required
2499  * for enabling events at boot. We want to enable events before
2500  * the filesystem is initialized.
2501  */
2502 static int
2503 __trace_early_add_new_event(struct trace_event_call *call,
2504                             struct trace_array *tr)
2505 {
2506         struct trace_event_file *file;
2507
2508         file = trace_create_new_event(call, tr);
2509         if (!file)
2510                 return -ENOMEM;
2511
2512         return event_define_fields(call);
2513 }
2514
2515 struct ftrace_module_file_ops;
2516 static void __add_event_to_tracers(struct trace_event_call *call);
2517
2518 /* Add an additional event_call dynamically */
2519 int trace_add_event_call(struct trace_event_call *call)
2520 {
2521         int ret;
2522         lockdep_assert_held(&event_mutex);
2523
2524         mutex_lock(&trace_types_lock);
2525
2526         ret = __register_event(call, NULL);
2527         if (ret >= 0)
2528                 __add_event_to_tracers(call);
2529
2530         mutex_unlock(&trace_types_lock);
2531         return ret;
2532 }
2533
2534 /*
2535  * Must be called under locking of trace_types_lock, event_mutex and
2536  * trace_event_sem.
2537  */
2538 static void __trace_remove_event_call(struct trace_event_call *call)
2539 {
2540         event_remove(call);
2541         trace_destroy_fields(call);
2542         free_event_filter(call->filter);
2543         call->filter = NULL;
2544 }
2545
2546 static int probe_remove_event_call(struct trace_event_call *call)
2547 {
2548         struct trace_array *tr;
2549         struct trace_event_file *file;
2550
2551 #ifdef CONFIG_PERF_EVENTS
2552         if (call->perf_refcount)
2553                 return -EBUSY;
2554 #endif
2555         do_for_each_event_file(tr, file) {
2556                 if (file->event_call != call)
2557                         continue;
2558                 /*
2559                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2560                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2561                  * TRACE_REG_UNREGISTER.
2562                  */
2563                 if (file->flags & EVENT_FILE_FL_ENABLED)
2564                         return -EBUSY;
2565                 /*
2566                  * The do_for_each_event_file_safe() is
2567                  * a double loop. After finding the call for this
2568                  * trace_array, we use break to jump to the next
2569                  * trace_array.
2570                  */
2571                 break;
2572         } while_for_each_event_file();
2573
2574         __trace_remove_event_call(call);
2575
2576         return 0;
2577 }
2578
2579 /* Remove an event_call */
2580 int trace_remove_event_call(struct trace_event_call *call)
2581 {
2582         int ret;
2583
2584         lockdep_assert_held(&event_mutex);
2585
2586         mutex_lock(&trace_types_lock);
2587         down_write(&trace_event_sem);
2588         ret = probe_remove_event_call(call);
2589         up_write(&trace_event_sem);
2590         mutex_unlock(&trace_types_lock);
2591
2592         return ret;
2593 }
2594
2595 #define for_each_event(event, start, end)                       \
2596         for (event = start;                                     \
2597              (unsigned long)event < (unsigned long)end;         \
2598              event++)
2599
2600 #ifdef CONFIG_MODULES
2601
2602 static void trace_module_add_events(struct module *mod)
2603 {
2604         struct trace_event_call **call, **start, **end;
2605
2606         if (!mod->num_trace_events)
2607                 return;
2608
2609         /* Don't add infrastructure for mods without tracepoints */
2610         if (trace_module_has_bad_taint(mod)) {
2611                 pr_err("%s: module has bad taint, not creating trace events\n",
2612                        mod->name);
2613                 return;
2614         }
2615
2616         start = mod->trace_events;
2617         end = mod->trace_events + mod->num_trace_events;
2618
2619         for_each_event(call, start, end) {
2620                 __register_event(*call, mod);
2621                 __add_event_to_tracers(*call);
2622         }
2623 }
2624
2625 static void trace_module_remove_events(struct module *mod)
2626 {
2627         struct trace_event_call *call, *p;
2628
2629         down_write(&trace_event_sem);
2630         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2631                 if (call->mod == mod)
2632                         __trace_remove_event_call(call);
2633         }
2634         up_write(&trace_event_sem);
2635
2636         /*
2637          * It is safest to reset the ring buffer if the module being unloaded
2638          * registered any events that were used. The only worry is if
2639          * a new module gets loaded, and takes on the same id as the events
2640          * of this module. When printing out the buffer, traced events left
2641          * over from this module may be passed to the new module events and
2642          * unexpected results may occur.
2643          */
2644         tracing_reset_all_online_cpus();
2645 }
2646
2647 static int trace_module_notify(struct notifier_block *self,
2648                                unsigned long val, void *data)
2649 {
2650         struct module *mod = data;
2651
2652         mutex_lock(&event_mutex);
2653         mutex_lock(&trace_types_lock);
2654         switch (val) {
2655         case MODULE_STATE_COMING:
2656                 trace_module_add_events(mod);
2657                 break;
2658         case MODULE_STATE_GOING:
2659                 trace_module_remove_events(mod);
2660                 break;
2661         }
2662         mutex_unlock(&trace_types_lock);
2663         mutex_unlock(&event_mutex);
2664
2665         return NOTIFY_OK;
2666 }
2667
2668 static struct notifier_block trace_module_nb = {
2669         .notifier_call = trace_module_notify,
2670         .priority = 1, /* higher than trace.c module notify */
2671 };
2672 #endif /* CONFIG_MODULES */
2673
2674 /* Create a new event directory structure for a trace directory. */
2675 static void
2676 __trace_add_event_dirs(struct trace_array *tr)
2677 {
2678         struct trace_event_call *call;
2679         int ret;
2680
2681         list_for_each_entry(call, &ftrace_events, list) {
2682                 ret = __trace_add_new_event(call, tr);
2683                 if (ret < 0)
2684                         pr_warn("Could not create directory for event %s\n",
2685                                 trace_event_name(call));
2686         }
2687 }
2688
2689 /* Returns any file that matches the system and event */
2690 struct trace_event_file *
2691 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2692 {
2693         struct trace_event_file *file;
2694         struct trace_event_call *call;
2695         const char *name;
2696
2697         list_for_each_entry(file, &tr->events, list) {
2698
2699                 call = file->event_call;
2700                 name = trace_event_name(call);
2701
2702                 if (!name || !call->class)
2703                         continue;
2704
2705                 if (strcmp(event, name) == 0 &&
2706                     strcmp(system, call->class->system) == 0)
2707                         return file;
2708         }
2709         return NULL;
2710 }
2711
2712 /* Returns valid trace event files that match system and event */
2713 struct trace_event_file *
2714 find_event_file(struct trace_array *tr, const char *system, const char *event)
2715 {
2716         struct trace_event_file *file;
2717
2718         file = __find_event_file(tr, system, event);
2719         if (!file || !file->event_call->class->reg ||
2720             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2721                 return NULL;
2722
2723         return file;
2724 }
2725
2726 /**
2727  * trace_get_event_file - Find and return a trace event file
2728  * @instance: The name of the trace instance containing the event
2729  * @system: The name of the system containing the event
2730  * @event: The name of the event
2731  *
2732  * Return a trace event file given the trace instance name, trace
2733  * system, and trace event name.  If the instance name is NULL, it
2734  * refers to the top-level trace array.
2735  *
2736  * This function will look it up and return it if found, after calling
2737  * trace_array_get() to prevent the instance from going away, and
2738  * increment the event's module refcount to prevent it from being
2739  * removed.
2740  *
2741  * To release the file, call trace_put_event_file(), which will call
2742  * trace_array_put() and decrement the event's module refcount.
2743  *
2744  * Return: The trace event on success, ERR_PTR otherwise.
2745  */
2746 struct trace_event_file *trace_get_event_file(const char *instance,
2747                                               const char *system,
2748                                               const char *event)
2749 {
2750         struct trace_array *tr = top_trace_array();
2751         struct trace_event_file *file = NULL;
2752         int ret = -EINVAL;
2753
2754         if (instance) {
2755                 tr = trace_array_find_get(instance);
2756                 if (!tr)
2757                         return ERR_PTR(-ENOENT);
2758         } else {
2759                 ret = trace_array_get(tr);
2760                 if (ret)
2761                         return ERR_PTR(ret);
2762         }
2763
2764         mutex_lock(&event_mutex);
2765
2766         file = find_event_file(tr, system, event);
2767         if (!file) {
2768                 trace_array_put(tr);
2769                 ret = -EINVAL;
2770                 goto out;
2771         }
2772
2773         /* Don't let event modules unload while in use */
2774         ret = try_module_get(file->event_call->mod);
2775         if (!ret) {
2776                 trace_array_put(tr);
2777                 ret = -EBUSY;
2778                 goto out;
2779         }
2780
2781         ret = 0;
2782  out:
2783         mutex_unlock(&event_mutex);
2784
2785         if (ret)
2786                 file = ERR_PTR(ret);
2787
2788         return file;
2789 }
2790 EXPORT_SYMBOL_GPL(trace_get_event_file);
2791
2792 /**
2793  * trace_put_event_file - Release a file from trace_get_event_file()
2794  * @file: The trace event file
2795  *
2796  * If a file was retrieved using trace_get_event_file(), this should
2797  * be called when it's no longer needed.  It will cancel the previous
2798  * trace_array_get() called by that function, and decrement the
2799  * event's module refcount.
2800  */
2801 void trace_put_event_file(struct trace_event_file *file)
2802 {
2803         mutex_lock(&event_mutex);
2804         module_put(file->event_call->mod);
2805         mutex_unlock(&event_mutex);
2806
2807         trace_array_put(file->tr);
2808 }
2809 EXPORT_SYMBOL_GPL(trace_put_event_file);
2810
2811 #ifdef CONFIG_DYNAMIC_FTRACE
2812
2813 /* Avoid typos */
2814 #define ENABLE_EVENT_STR        "enable_event"
2815 #define DISABLE_EVENT_STR       "disable_event"
2816
2817 struct event_probe_data {
2818         struct trace_event_file *file;
2819         unsigned long                   count;
2820         int                             ref;
2821         bool                            enable;
2822 };
2823
2824 static void update_event_probe(struct event_probe_data *data)
2825 {
2826         if (data->enable)
2827                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2828         else
2829                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2830 }
2831
2832 static void
2833 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2834                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2835                    void *data)
2836 {
2837         struct ftrace_func_mapper *mapper = data;
2838         struct event_probe_data *edata;
2839         void **pdata;
2840
2841         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2842         if (!pdata || !*pdata)
2843                 return;
2844
2845         edata = *pdata;
2846         update_event_probe(edata);
2847 }
2848
2849 static void
2850 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2851                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2852                          void *data)
2853 {
2854         struct ftrace_func_mapper *mapper = data;
2855         struct event_probe_data *edata;
2856         void **pdata;
2857
2858         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2859         if (!pdata || !*pdata)
2860                 return;
2861
2862         edata = *pdata;
2863
2864         if (!edata->count)
2865                 return;
2866
2867         /* Skip if the event is in a state we want to switch to */
2868         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2869                 return;
2870
2871         if (edata->count != -1)
2872                 (edata->count)--;
2873
2874         update_event_probe(edata);
2875 }
2876
2877 static int
2878 event_enable_print(struct seq_file *m, unsigned long ip,
2879                    struct ftrace_probe_ops *ops, void *data)
2880 {
2881         struct ftrace_func_mapper *mapper = data;
2882         struct event_probe_data *edata;
2883         void **pdata;
2884
2885         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2886
2887         if (WARN_ON_ONCE(!pdata || !*pdata))
2888                 return 0;
2889
2890         edata = *pdata;
2891
2892         seq_printf(m, "%ps:", (void *)ip);
2893
2894         seq_printf(m, "%s:%s:%s",
2895                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2896                    edata->file->event_call->class->system,
2897                    trace_event_name(edata->file->event_call));
2898
2899         if (edata->count == -1)
2900                 seq_puts(m, ":unlimited\n");
2901         else
2902                 seq_printf(m, ":count=%ld\n", edata->count);
2903
2904         return 0;
2905 }
2906
2907 static int
2908 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2909                   unsigned long ip, void *init_data, void **data)
2910 {
2911         struct ftrace_func_mapper *mapper = *data;
2912         struct event_probe_data *edata = init_data;
2913         int ret;
2914
2915         if (!mapper) {
2916                 mapper = allocate_ftrace_func_mapper();
2917                 if (!mapper)
2918                         return -ENODEV;
2919                 *data = mapper;
2920         }
2921
2922         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2923         if (ret < 0)
2924                 return ret;
2925
2926         edata->ref++;
2927
2928         return 0;
2929 }
2930
2931 static int free_probe_data(void *data)
2932 {
2933         struct event_probe_data *edata = data;
2934
2935         edata->ref--;
2936         if (!edata->ref) {
2937                 /* Remove the SOFT_MODE flag */
2938                 __ftrace_event_enable_disable(edata->file, 0, 1);
2939                 module_put(edata->file->event_call->mod);
2940                 kfree(edata);
2941         }
2942         return 0;
2943 }
2944
2945 static void
2946 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2947                   unsigned long ip, void *data)
2948 {
2949         struct ftrace_func_mapper *mapper = data;
2950         struct event_probe_data *edata;
2951
2952         if (!ip) {
2953                 if (!mapper)
2954                         return;
2955                 free_ftrace_func_mapper(mapper, free_probe_data);
2956                 return;
2957         }
2958
2959         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2960
2961         if (WARN_ON_ONCE(!edata))
2962                 return;
2963
2964         if (WARN_ON_ONCE(edata->ref <= 0))
2965                 return;
2966
2967         free_probe_data(edata);
2968 }
2969
2970 static struct ftrace_probe_ops event_enable_probe_ops = {
2971         .func                   = event_enable_probe,
2972         .print                  = event_enable_print,
2973         .init                   = event_enable_init,
2974         .free                   = event_enable_free,
2975 };
2976
2977 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2978         .func                   = event_enable_count_probe,
2979         .print                  = event_enable_print,
2980         .init                   = event_enable_init,
2981         .free                   = event_enable_free,
2982 };
2983
2984 static struct ftrace_probe_ops event_disable_probe_ops = {
2985         .func                   = event_enable_probe,
2986         .print                  = event_enable_print,
2987         .init                   = event_enable_init,
2988         .free                   = event_enable_free,
2989 };
2990
2991 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2992         .func                   = event_enable_count_probe,
2993         .print                  = event_enable_print,
2994         .init                   = event_enable_init,
2995         .free                   = event_enable_free,
2996 };
2997
2998 static int
2999 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3000                   char *glob, char *cmd, char *param, int enabled)
3001 {
3002         struct trace_event_file *file;
3003         struct ftrace_probe_ops *ops;
3004         struct event_probe_data *data;
3005         const char *system;
3006         const char *event;
3007         char *number;
3008         bool enable;
3009         int ret;
3010
3011         if (!tr)
3012                 return -ENODEV;
3013
3014         /* hash funcs only work with set_ftrace_filter */
3015         if (!enabled || !param)
3016                 return -EINVAL;
3017
3018         system = strsep(&param, ":");
3019         if (!param)
3020                 return -EINVAL;
3021
3022         event = strsep(&param, ":");
3023
3024         mutex_lock(&event_mutex);
3025
3026         ret = -EINVAL;
3027         file = find_event_file(tr, system, event);
3028         if (!file)
3029                 goto out;
3030
3031         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3032
3033         if (enable)
3034                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3035         else
3036                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3037
3038         if (glob[0] == '!') {
3039                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3040                 goto out;
3041         }
3042
3043         ret = -ENOMEM;
3044
3045         data = kzalloc(sizeof(*data), GFP_KERNEL);
3046         if (!data)
3047                 goto out;
3048
3049         data->enable = enable;
3050         data->count = -1;
3051         data->file = file;
3052
3053         if (!param)
3054                 goto out_reg;
3055
3056         number = strsep(&param, ":");
3057
3058         ret = -EINVAL;
3059         if (!strlen(number))
3060                 goto out_free;
3061
3062         /*
3063          * We use the callback data field (which is a pointer)
3064          * as our counter.
3065          */
3066         ret = kstrtoul(number, 0, &data->count);
3067         if (ret)
3068                 goto out_free;
3069
3070  out_reg:
3071         /* Don't let event modules unload while probe registered */
3072         ret = try_module_get(file->event_call->mod);
3073         if (!ret) {
3074                 ret = -EBUSY;
3075                 goto out_free;
3076         }
3077
3078         ret = __ftrace_event_enable_disable(file, 1, 1);
3079         if (ret < 0)
3080                 goto out_put;
3081
3082         ret = register_ftrace_function_probe(glob, tr, ops, data);
3083         /*
3084          * The above returns on success the # of functions enabled,
3085          * but if it didn't find any functions it returns zero.
3086          * Consider no functions a failure too.
3087          */
3088         if (!ret) {
3089                 ret = -ENOENT;
3090                 goto out_disable;
3091         } else if (ret < 0)
3092                 goto out_disable;
3093         /* Just return zero, not the number of enabled functions */
3094         ret = 0;
3095  out:
3096         mutex_unlock(&event_mutex);
3097         return ret;
3098
3099  out_disable:
3100         __ftrace_event_enable_disable(file, 0, 1);
3101  out_put:
3102         module_put(file->event_call->mod);
3103  out_free:
3104         kfree(data);
3105         goto out;
3106 }
3107
3108 static struct ftrace_func_command event_enable_cmd = {
3109         .name                   = ENABLE_EVENT_STR,
3110         .func                   = event_enable_func,
3111 };
3112
3113 static struct ftrace_func_command event_disable_cmd = {
3114         .name                   = DISABLE_EVENT_STR,
3115         .func                   = event_enable_func,
3116 };
3117
3118 static __init int register_event_cmds(void)
3119 {
3120         int ret;
3121
3122         ret = register_ftrace_command(&event_enable_cmd);
3123         if (WARN_ON(ret < 0))
3124                 return ret;
3125         ret = register_ftrace_command(&event_disable_cmd);
3126         if (WARN_ON(ret < 0))
3127                 unregister_ftrace_command(&event_enable_cmd);
3128         return ret;
3129 }
3130 #else
3131 static inline int register_event_cmds(void) { return 0; }
3132 #endif /* CONFIG_DYNAMIC_FTRACE */
3133
3134 /*
3135  * The top level array and trace arrays created by boot-time tracing
3136  * have already had its trace_event_file descriptors created in order
3137  * to allow for early events to be recorded.
3138  * This function is called after the tracefs has been initialized,
3139  * and we now have to create the files associated to the events.
3140  */
3141 static void __trace_early_add_event_dirs(struct trace_array *tr)
3142 {
3143         struct trace_event_file *file;
3144         int ret;
3145
3146
3147         list_for_each_entry(file, &tr->events, list) {
3148                 ret = event_create_dir(tr->event_dir, file);
3149                 if (ret < 0)
3150                         pr_warn("Could not create directory for event %s\n",
3151                                 trace_event_name(file->event_call));
3152         }
3153 }
3154
3155 /*
3156  * For early boot up, the top trace array and the trace arrays created
3157  * by boot-time tracing require to have a list of events that can be
3158  * enabled. This must be done before the filesystem is set up in order
3159  * to allow events to be traced early.
3160  */
3161 void __trace_early_add_events(struct trace_array *tr)
3162 {
3163         struct trace_event_call *call;
3164         int ret;
3165
3166         list_for_each_entry(call, &ftrace_events, list) {
3167                 /* Early boot up should not have any modules loaded */
3168                 if (WARN_ON_ONCE(call->mod))
3169                         continue;
3170
3171                 ret = __trace_early_add_new_event(call, tr);
3172                 if (ret < 0)
3173                         pr_warn("Could not create early event %s\n",
3174                                 trace_event_name(call));
3175         }
3176 }
3177
3178 /* Remove the event directory structure for a trace directory. */
3179 static void
3180 __trace_remove_event_dirs(struct trace_array *tr)
3181 {
3182         struct trace_event_file *file, *next;
3183
3184         list_for_each_entry_safe(file, next, &tr->events, list)
3185                 remove_event_file_dir(file);
3186 }
3187
3188 static void __add_event_to_tracers(struct trace_event_call *call)
3189 {
3190         struct trace_array *tr;
3191
3192         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3193                 __trace_add_new_event(call, tr);
3194 }
3195
3196 extern struct trace_event_call *__start_ftrace_events[];
3197 extern struct trace_event_call *__stop_ftrace_events[];
3198
3199 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3200
3201 static __init int setup_trace_event(char *str)
3202 {
3203         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3204         ring_buffer_expanded = true;
3205         disable_tracing_selftest("running event tracing");
3206
3207         return 1;
3208 }
3209 __setup("trace_event=", setup_trace_event);
3210
3211 /* Expects to have event_mutex held when called */
3212 static int
3213 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3214 {
3215         struct dentry *d_events;
3216         struct dentry *entry;
3217
3218         entry = tracefs_create_file("set_event", 0644, parent,
3219                                     tr, &ftrace_set_event_fops);
3220         if (!entry) {
3221                 pr_warn("Could not create tracefs 'set_event' entry\n");
3222                 return -ENOMEM;
3223         }
3224
3225         d_events = tracefs_create_dir("events", parent);
3226         if (!d_events) {
3227                 pr_warn("Could not create tracefs 'events' directory\n");
3228                 return -ENOMEM;
3229         }
3230
3231         entry = trace_create_file("enable", 0644, d_events,
3232                                   tr, &ftrace_tr_enable_fops);
3233         if (!entry) {
3234                 pr_warn("Could not create tracefs 'enable' entry\n");
3235                 return -ENOMEM;
3236         }
3237
3238         /* There are not as crucial, just warn if they are not created */
3239
3240         entry = tracefs_create_file("set_event_pid", 0644, parent,
3241                                     tr, &ftrace_set_event_pid_fops);
3242         if (!entry)
3243                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3244
3245         entry = tracefs_create_file("set_event_notrace_pid", 0644, parent,
3246                                     tr, &ftrace_set_event_notrace_pid_fops);
3247         if (!entry)
3248                 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3249
3250         /* ring buffer internal formats */
3251         entry = trace_create_file("header_page", 0444, d_events,
3252                                   ring_buffer_print_page_header,
3253                                   &ftrace_show_header_fops);
3254         if (!entry)
3255                 pr_warn("Could not create tracefs 'header_page' entry\n");
3256
3257         entry = trace_create_file("header_event", 0444, d_events,
3258                                   ring_buffer_print_entry_header,
3259                                   &ftrace_show_header_fops);
3260         if (!entry)
3261                 pr_warn("Could not create tracefs 'header_event' entry\n");
3262
3263         tr->event_dir = d_events;
3264
3265         return 0;
3266 }
3267
3268 /**
3269  * event_trace_add_tracer - add a instance of a trace_array to events
3270  * @parent: The parent dentry to place the files/directories for events in
3271  * @tr: The trace array associated with these events
3272  *
3273  * When a new instance is created, it needs to set up its events
3274  * directory, as well as other files associated with events. It also
3275  * creates the event hierarchy in the @parent/events directory.
3276  *
3277  * Returns 0 on success.
3278  *
3279  * Must be called with event_mutex held.
3280  */
3281 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3282 {
3283         int ret;
3284
3285         lockdep_assert_held(&event_mutex);
3286
3287         ret = create_event_toplevel_files(parent, tr);
3288         if (ret)
3289                 goto out;
3290
3291         down_write(&trace_event_sem);
3292         /* If tr already has the event list, it is initialized in early boot. */
3293         if (unlikely(!list_empty(&tr->events)))
3294                 __trace_early_add_event_dirs(tr);
3295         else
3296                 __trace_add_event_dirs(tr);
3297         up_write(&trace_event_sem);
3298
3299  out:
3300         return ret;
3301 }
3302
3303 /*
3304  * The top trace array already had its file descriptors created.
3305  * Now the files themselves need to be created.
3306  */
3307 static __init int
3308 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3309 {
3310         int ret;
3311
3312         mutex_lock(&event_mutex);
3313
3314         ret = create_event_toplevel_files(parent, tr);
3315         if (ret)
3316                 goto out_unlock;
3317
3318         down_write(&trace_event_sem);
3319         __trace_early_add_event_dirs(tr);
3320         up_write(&trace_event_sem);
3321
3322  out_unlock:
3323         mutex_unlock(&event_mutex);
3324
3325         return ret;
3326 }
3327
3328 /* Must be called with event_mutex held */
3329 int event_trace_del_tracer(struct trace_array *tr)
3330 {
3331         lockdep_assert_held(&event_mutex);
3332
3333         /* Disable any event triggers and associated soft-disabled events */
3334         clear_event_triggers(tr);
3335
3336         /* Clear the pid list */
3337         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3338
3339         /* Disable any running events */
3340         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3341
3342         /* Make sure no more events are being executed */
3343         tracepoint_synchronize_unregister();
3344
3345         down_write(&trace_event_sem);
3346         __trace_remove_event_dirs(tr);
3347         tracefs_remove(tr->event_dir);
3348         up_write(&trace_event_sem);
3349
3350         tr->event_dir = NULL;
3351
3352         return 0;
3353 }
3354
3355 static __init int event_trace_memsetup(void)
3356 {
3357         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3358         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3359         return 0;
3360 }
3361
3362 static __init void
3363 early_enable_events(struct trace_array *tr, bool disable_first)
3364 {
3365         char *buf = bootup_event_buf;
3366         char *token;
3367         int ret;
3368
3369         while (true) {
3370                 token = strsep(&buf, ",");
3371
3372                 if (!token)
3373                         break;
3374
3375                 if (*token) {
3376                         /* Restarting syscalls requires that we stop them first */
3377                         if (disable_first)
3378                                 ftrace_set_clr_event(tr, token, 0);
3379
3380                         ret = ftrace_set_clr_event(tr, token, 1);
3381                         if (ret)
3382                                 pr_warn("Failed to enable trace event: %s\n", token);
3383                 }
3384
3385                 /* Put back the comma to allow this to be called again */
3386                 if (buf)
3387                         *(buf - 1) = ',';
3388         }
3389 }
3390
3391 static __init int event_trace_enable(void)
3392 {
3393         struct trace_array *tr = top_trace_array();
3394         struct trace_event_call **iter, *call;
3395         int ret;
3396
3397         if (!tr)
3398                 return -ENODEV;
3399
3400         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3401
3402                 call = *iter;
3403                 ret = event_init(call);
3404                 if (!ret)
3405                         list_add(&call->list, &ftrace_events);
3406         }
3407
3408         /*
3409          * We need the top trace array to have a working set of trace
3410          * points at early init, before the debug files and directories
3411          * are created. Create the file entries now, and attach them
3412          * to the actual file dentries later.
3413          */
3414         __trace_early_add_events(tr);
3415
3416         early_enable_events(tr, false);
3417
3418         trace_printk_start_comm();
3419
3420         register_event_cmds();
3421
3422         register_trigger_cmds();
3423
3424         return 0;
3425 }
3426
3427 /*
3428  * event_trace_enable() is called from trace_event_init() first to
3429  * initialize events and perhaps start any events that are on the
3430  * command line. Unfortunately, there are some events that will not
3431  * start this early, like the system call tracepoints that need
3432  * to set the %SYSCALL_WORK_SYSCALL_TRACEPOINT flag of pid 1. But
3433  * event_trace_enable() is called before pid 1 starts, and this flag
3434  * is never set, making the syscall tracepoint never get reached, but
3435  * the event is enabled regardless (and not doing anything).
3436  */
3437 static __init int event_trace_enable_again(void)
3438 {
3439         struct trace_array *tr;
3440
3441         tr = top_trace_array();
3442         if (!tr)
3443                 return -ENODEV;
3444
3445         early_enable_events(tr, true);
3446
3447         return 0;
3448 }
3449
3450 early_initcall(event_trace_enable_again);
3451
3452 /* Init fields which doesn't related to the tracefs */
3453 static __init int event_trace_init_fields(void)
3454 {
3455         if (trace_define_generic_fields())
3456                 pr_warn("tracing: Failed to allocated generic fields");
3457
3458         if (trace_define_common_fields())
3459                 pr_warn("tracing: Failed to allocate common fields");
3460
3461         return 0;
3462 }
3463
3464 __init int event_trace_init(void)
3465 {
3466         struct trace_array *tr;
3467         struct dentry *entry;
3468         int ret;
3469
3470         tr = top_trace_array();
3471         if (!tr)
3472                 return -ENODEV;
3473
3474         entry = tracefs_create_file("available_events", 0444, NULL,
3475                                     tr, &ftrace_avail_fops);
3476         if (!entry)
3477                 pr_warn("Could not create tracefs 'available_events' entry\n");
3478
3479         ret = early_event_add_tracer(NULL, tr);
3480         if (ret)
3481                 return ret;
3482
3483 #ifdef CONFIG_MODULES
3484         ret = register_module_notifier(&trace_module_nb);
3485         if (ret)
3486                 pr_warn("Failed to register trace events module notifier\n");
3487 #endif
3488
3489         eventdir_initialized = true;
3490
3491         return 0;
3492 }
3493
3494 void __init trace_event_init(void)
3495 {
3496         event_trace_memsetup();
3497         init_ftrace_syscalls();
3498         event_trace_enable();
3499         event_trace_init_fields();
3500 }
3501
3502 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3503
3504 static DEFINE_SPINLOCK(test_spinlock);
3505 static DEFINE_SPINLOCK(test_spinlock_irq);
3506 static DEFINE_MUTEX(test_mutex);
3507
3508 static __init void test_work(struct work_struct *dummy)
3509 {
3510         spin_lock(&test_spinlock);
3511         spin_lock_irq(&test_spinlock_irq);
3512         udelay(1);
3513         spin_unlock_irq(&test_spinlock_irq);
3514         spin_unlock(&test_spinlock);
3515
3516         mutex_lock(&test_mutex);
3517         msleep(1);
3518         mutex_unlock(&test_mutex);
3519 }
3520
3521 static __init int event_test_thread(void *unused)
3522 {
3523         void *test_malloc;
3524
3525         test_malloc = kmalloc(1234, GFP_KERNEL);
3526         if (!test_malloc)
3527                 pr_info("failed to kmalloc\n");
3528
3529         schedule_on_each_cpu(test_work);
3530
3531         kfree(test_malloc);
3532
3533         set_current_state(TASK_INTERRUPTIBLE);
3534         while (!kthread_should_stop()) {
3535                 schedule();
3536                 set_current_state(TASK_INTERRUPTIBLE);
3537         }
3538         __set_current_state(TASK_RUNNING);
3539
3540         return 0;
3541 }
3542
3543 /*
3544  * Do various things that may trigger events.
3545  */
3546 static __init void event_test_stuff(void)
3547 {
3548         struct task_struct *test_thread;
3549
3550         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3551         msleep(1);
3552         kthread_stop(test_thread);
3553 }
3554
3555 /*
3556  * For every trace event defined, we will test each trace point separately,
3557  * and then by groups, and finally all trace points.
3558  */
3559 static __init void event_trace_self_tests(void)
3560 {
3561         struct trace_subsystem_dir *dir;
3562         struct trace_event_file *file;
3563         struct trace_event_call *call;
3564         struct event_subsystem *system;
3565         struct trace_array *tr;
3566         int ret;
3567
3568         tr = top_trace_array();
3569         if (!tr)
3570                 return;
3571
3572         pr_info("Running tests on trace events:\n");
3573
3574         list_for_each_entry(file, &tr->events, list) {
3575
3576                 call = file->event_call;
3577
3578                 /* Only test those that have a probe */
3579                 if (!call->class || !call->class->probe)
3580                         continue;
3581
3582 /*
3583  * Testing syscall events here is pretty useless, but
3584  * we still do it if configured. But this is time consuming.
3585  * What we really need is a user thread to perform the
3586  * syscalls as we test.
3587  */
3588 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3589                 if (call->class->system &&
3590                     strcmp(call->class->system, "syscalls") == 0)
3591                         continue;
3592 #endif
3593
3594                 pr_info("Testing event %s: ", trace_event_name(call));
3595
3596                 /*
3597                  * If an event is already enabled, someone is using
3598                  * it and the self test should not be on.
3599                  */
3600                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3601                         pr_warn("Enabled event during self test!\n");
3602                         WARN_ON_ONCE(1);
3603                         continue;
3604                 }
3605
3606                 ftrace_event_enable_disable(file, 1);
3607                 event_test_stuff();
3608                 ftrace_event_enable_disable(file, 0);
3609
3610                 pr_cont("OK\n");
3611         }
3612
3613         /* Now test at the sub system level */
3614
3615         pr_info("Running tests on trace event systems:\n");
3616
3617         list_for_each_entry(dir, &tr->systems, list) {
3618
3619                 system = dir->subsystem;
3620
3621                 /* the ftrace system is special, skip it */
3622                 if (strcmp(system->name, "ftrace") == 0)
3623                         continue;
3624
3625                 pr_info("Testing event system %s: ", system->name);
3626
3627                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3628                 if (WARN_ON_ONCE(ret)) {
3629                         pr_warn("error enabling system %s\n",
3630                                 system->name);
3631                         continue;
3632                 }
3633
3634                 event_test_stuff();
3635
3636                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3637                 if (WARN_ON_ONCE(ret)) {
3638                         pr_warn("error disabling system %s\n",
3639                                 system->name);
3640                         continue;
3641                 }
3642
3643                 pr_cont("OK\n");
3644         }
3645
3646         /* Test with all events enabled */
3647
3648         pr_info("Running tests on all trace events:\n");
3649         pr_info("Testing all events: ");
3650
3651         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3652         if (WARN_ON_ONCE(ret)) {
3653                 pr_warn("error enabling all events\n");
3654                 return;
3655         }
3656
3657         event_test_stuff();
3658
3659         /* reset sysname */
3660         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3661         if (WARN_ON_ONCE(ret)) {
3662                 pr_warn("error disabling all events\n");
3663                 return;
3664         }
3665
3666         pr_cont("OK\n");
3667 }
3668
3669 #ifdef CONFIG_FUNCTION_TRACER
3670
3671 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3672
3673 static struct trace_event_file event_trace_file __initdata;
3674
3675 static void __init
3676 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3677                           struct ftrace_ops *op, struct ftrace_regs *regs)
3678 {
3679         struct trace_buffer *buffer;
3680         struct ring_buffer_event *event;
3681         struct ftrace_entry *entry;
3682         unsigned long flags;
3683         long disabled;
3684         int cpu;
3685         int pc;
3686
3687         pc = preempt_count();
3688         preempt_disable_notrace();
3689         cpu = raw_smp_processor_id();
3690         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3691
3692         if (disabled != 1)
3693                 goto out;
3694
3695         local_save_flags(flags);
3696
3697         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3698                                                 TRACE_FN, sizeof(*entry),
3699                                                 flags, pc);
3700         if (!event)
3701                 goto out;
3702         entry   = ring_buffer_event_data(event);
3703         entry->ip                       = ip;
3704         entry->parent_ip                = parent_ip;
3705
3706         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3707                                     entry, flags, pc);
3708  out:
3709         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3710         preempt_enable_notrace();
3711 }
3712
3713 static struct ftrace_ops trace_ops __initdata  =
3714 {
3715         .func = function_test_events_call,
3716 };
3717
3718 static __init void event_trace_self_test_with_function(void)
3719 {
3720         int ret;
3721
3722         event_trace_file.tr = top_trace_array();
3723         if (WARN_ON(!event_trace_file.tr))
3724                 return;
3725
3726         ret = register_ftrace_function(&trace_ops);
3727         if (WARN_ON(ret < 0)) {
3728                 pr_info("Failed to enable function tracer for event tests\n");
3729                 return;
3730         }
3731         pr_info("Running tests again, along with the function tracer\n");
3732         event_trace_self_tests();
3733         unregister_ftrace_function(&trace_ops);
3734 }
3735 #else
3736 static __init void event_trace_self_test_with_function(void)
3737 {
3738 }
3739 #endif
3740
3741 static __init int event_trace_self_tests_init(void)
3742 {
3743         if (!tracing_selftest_disabled) {
3744                 event_trace_self_tests();
3745                 event_trace_self_test_with_function();
3746         }
3747
3748         return 0;
3749 }
3750
3751 late_initcall(event_trace_self_tests_init);
3752
3753 #endif