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(int, common_cpu, FILTER_CPU);
 172         __generic_field(char *, COMM, FILTER_COMM);
 173         __generic_field(char *, comm, FILTER_COMM);
 174
 175         return ret;
 176 }
 177
 178 static int trace_define_common_fields(void)
 179 {
 180         int ret;
 181         struct trace_entry ent;
 182
 183         __common_field(unsigned short, type);
 184         __common_field(unsigned char, flags);
 185         __common_field(unsigned char, preempt_count);
 186         __common_field(int, pid);
 187
 188         return ret;
 189 }
 190
 191 static void trace_destroy_fields(struct trace_event_call *call)
 192 {
 193         struct ftrace_event_field *field, *next;
 194         struct list_head *head;
 195
 196         head = trace_get_fields(call);
 197         list_for_each_entry_safe(field, next, head, link) {
 198                 list_del(&field->link);
 199                 kmem_cache_free(field_cachep, field);
 200         }
 201 }
 202
 203 /*
 204  * run-time version of trace_event_get_offsets_<call>() that returns the last
 205  * accessible offset of trace fields excluding __dynamic_array bytes
 206  */
 207 int trace_event_get_offsets(struct trace_event_call *call)
 208 {
 209         struct ftrace_event_field *tail;
 210         struct list_head *head;
 211
 212         head = trace_get_fields(call);
 213         /*
 214          * head->next points to the last field with the largest offset,
 215          * since it was added last by trace_define_field()
 216          */
 217         tail = list_first_entry(head, struct ftrace_event_field, link);
 218         return tail->offset + tail->size;
 219 }
 220
 221 int trace_event_raw_init(struct trace_event_call *call)
 222 {
 223         int id;
 224
 225         id = register_trace_event(&call->event);
 226         if (!id)
 227                 return -ENODEV;
 228
 229         return 0;
 230 }
 231 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 232
 233 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file)
 234 {
 235         struct trace_array *tr = trace_file->tr;
 236         struct trace_array_cpu *data;
 237         struct trace_pid_list *no_pid_list;
 238         struct trace_pid_list *pid_list;
 239
 240         pid_list = rcu_dereference_raw(tr->filtered_pids);
 241         no_pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 242
 243         if (!pid_list && !no_pid_list)
 244                 return false;
 245
 246         data = this_cpu_ptr(tr->array_buffer.data);
 247
 248         return data->ignore_pid;
 249 }
 250 EXPORT_SYMBOL_GPL(trace_event_ignore_this_pid);
 251
 252 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 253                                  struct trace_event_file *trace_file,
 254                                  unsigned long len)
 255 {
 256         struct trace_event_call *event_call = trace_file->event_call;
 257
 258         if ((trace_file->flags & EVENT_FILE_FL_PID_FILTER) &&
 259             trace_event_ignore_this_pid(trace_file))
 260                 return NULL;
 261
 262         local_save_flags(fbuffer->flags);
 263         fbuffer->pc = preempt_count();
 264         /*
 265          * If CONFIG_PREEMPTION is enabled, then the tracepoint itself disables
 266          * preemption (adding one to the preempt_count). Since we are
 267          * interested in the preempt_count at the time the tracepoint was
 268          * hit, we need to subtract one to offset the increment.
 269          */
 270         if (IS_ENABLED(CONFIG_PREEMPTION))
 271                 fbuffer->pc--;
 272         fbuffer->trace_file = trace_file;
 273
 274         fbuffer->event =
 275                 trace_event_buffer_lock_reserve(&fbuffer->buffer, trace_file,
 276                                                 event_call->event.type, len,
 277                                                 fbuffer->flags, fbuffer->pc);
 278         if (!fbuffer->event)
 279                 return NULL;
 280
 281         fbuffer->regs = NULL;
 282         fbuffer->entry = ring_buffer_event_data(fbuffer->event);
 283         return fbuffer->entry;
 284 }
 285 EXPORT_SYMBOL_GPL(trace_event_buffer_reserve);
 286
 287 int trace_event_reg(struct trace_event_call *call,
 288                     enum trace_reg type, void *data)
 289 {
 290         struct trace_event_file *file = data;
 291
 292         WARN_ON(!(call->flags & TRACE_EVENT_FL_TRACEPOINT));
 293         switch (type) {
 294         case TRACE_REG_REGISTER:
 295                 return tracepoint_probe_register(call->tp,
 296                                                  call->class->probe,
 297                                                  file);
 298         case TRACE_REG_UNREGISTER:
 299                 tracepoint_probe_unregister(call->tp,
 300                                             call->class->probe,
 301                                             file);
 302                 return 0;
 303
 304 #ifdef CONFIG_PERF_EVENTS
 305         case TRACE_REG_PERF_REGISTER:
 306                 return tracepoint_probe_register(call->tp,
 307                                                  call->class->perf_probe,
 308                                                  call);
 309         case TRACE_REG_PERF_UNREGISTER:
 310                 tracepoint_probe_unregister(call->tp,
 311                                             call->class->perf_probe,
 312                                             call);
 313                 return 0;
 314         case TRACE_REG_PERF_OPEN:
 315         case TRACE_REG_PERF_CLOSE:
 316         case TRACE_REG_PERF_ADD:
 317         case TRACE_REG_PERF_DEL:
 318                 return 0;
 319 #endif
 320         }
 321         return 0;
 322 }
 323 EXPORT_SYMBOL_GPL(trace_event_reg);
 324
 325 void trace_event_enable_cmd_record(bool enable)
 326 {
 327         struct trace_event_file *file;
 328         struct trace_array *tr;
 329
 330         lockdep_assert_held(&event_mutex);
 331
 332         do_for_each_event_file(tr, file) {
 333
 334                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 335                         continue;
 336
 337                 if (enable) {
 338                         tracing_start_cmdline_record();
 339                         set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 340                 } else {
 341                         tracing_stop_cmdline_record();
 342                         clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 343                 }
 344         } while_for_each_event_file();
 345 }
 346
 347 void trace_event_enable_tgid_record(bool enable)
 348 {
 349         struct trace_event_file *file;
 350         struct trace_array *tr;
 351
 352         lockdep_assert_held(&event_mutex);
 353
 354         do_for_each_event_file(tr, file) {
 355                 if (!(file->flags & EVENT_FILE_FL_ENABLED))
 356                         continue;
 357
 358                 if (enable) {
 359                         tracing_start_tgid_record();
 360                         set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 361                 } else {
 362                         tracing_stop_tgid_record();
 363                         clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT,
 364                                   &file->flags);
 365                 }
 366         } while_for_each_event_file();
 367 }
 368
 369 static int __ftrace_event_enable_disable(struct trace_event_file *file,
 370                                          int enable, int soft_disable)
 371 {
 372         struct trace_event_call *call = file->event_call;
 373         struct trace_array *tr = file->tr;
 374         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                         /* Disable use of trace_buffered_event */
 398                         trace_buffered_event_disable();
 399                 } else
 400                         disable = !(file->flags & EVENT_FILE_FL_SOFT_MODE);
 401
 402                 if (disable && (file->flags & EVENT_FILE_FL_ENABLED)) {
 403                         clear_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 404                         if (file->flags & EVENT_FILE_FL_RECORDED_CMD) {
 405                                 tracing_stop_cmdline_record();
 406                                 clear_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 407                         }
 408
 409                         if (file->flags & EVENT_FILE_FL_RECORDED_TGID) {
 410                                 tracing_stop_tgid_record();
 411                                 clear_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 412                         }
 413
 414                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 415                 }
 416                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 417                 if (file->flags & EVENT_FILE_FL_SOFT_MODE)
 418                         set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 419                 else
 420                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 421                 break;
 422         case 1:
 423                 /*
 424                  * When soft_disable is set and enable is set, we want to
 425                  * register the tracepoint for the event, but leave the event
 426                  * as is. That means, if the event was already enabled, we do
 427                  * nothing (but set SOFT_MODE). If the event is disabled, we
 428                  * set SOFT_DISABLED before enabling the event tracepoint, so
 429                  * it still seems to be disabled.
 430                  */
 431                 if (!soft_disable)
 432                         clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 433                 else {
 434                         if (atomic_inc_return(&file->sm_ref) > 1)
 435                                 break;
 436                         set_bit(EVENT_FILE_FL_SOFT_MODE_BIT, &file->flags);
 437                         /* Enable use of trace_buffered_event */
 438                         trace_buffered_event_enable();
 439                 }
 440
 441                 if (!(file->flags & EVENT_FILE_FL_ENABLED)) {
 442                         bool cmd = false, tgid = false;
 443
 444                         /* Keep the event disabled, when going to SOFT_MODE. */
 445                         if (soft_disable)
 446                                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags);
 447
 448                         if (tr->trace_flags & TRACE_ITER_RECORD_CMD) {
 449                                 cmd = true;
 450                                 tracing_start_cmdline_record();
 451                                 set_bit(EVENT_FILE_FL_RECORDED_CMD_BIT, &file->flags);
 452                         }
 453
 454                         if (tr->trace_flags & TRACE_ITER_RECORD_TGID) {
 455                                 tgid = true;
 456                                 tracing_start_tgid_record();
 457                                 set_bit(EVENT_FILE_FL_RECORDED_TGID_BIT, &file->flags);
 458                         }
 459
 460                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 461                         if (ret) {
 462                                 if (cmd)
 463                                         tracing_stop_cmdline_record();
 464                                 if (tgid)
 465                                         tracing_stop_tgid_record();
 466                                 pr_info("event trace: Could not enable event "
 467                                         "%s\n", trace_event_name(call));
 468                                 break;
 469                         }
 470                         set_bit(EVENT_FILE_FL_ENABLED_BIT, &file->flags);
 471
 472                         /* WAS_ENABLED gets set but never cleared. */
 473                         set_bit(EVENT_FILE_FL_WAS_ENABLED_BIT, &file->flags);
 474                 }
 475                 break;
 476         }
 477
 478         return ret;
 479 }
 480
 481 int trace_event_enable_disable(struct trace_event_file *file,
 482                                int enable, int soft_disable)
 483 {
 484         return __ftrace_event_enable_disable(file, enable, soft_disable);
 485 }
 486
 487 static int ftrace_event_enable_disable(struct trace_event_file *file,
 488                                        int enable)
 489 {
 490         return __ftrace_event_enable_disable(file, enable, 0);
 491 }
 492
 493 static void ftrace_clear_events(struct trace_array *tr)
 494 {
 495         struct trace_event_file *file;
 496
 497         mutex_lock(&event_mutex);
 498         list_for_each_entry(file, &tr->events, list) {
 499                 ftrace_event_enable_disable(file, 0);
 500         }
 501         mutex_unlock(&event_mutex);
 502 }
 503
 504 static void
 505 event_filter_pid_sched_process_exit(void *data, struct task_struct *task)
 506 {
 507         struct trace_pid_list *pid_list;
 508         struct trace_array *tr = data;
 509
 510         pid_list = rcu_dereference_raw(tr->filtered_pids);
 511         trace_filter_add_remove_task(pid_list, NULL, task);
 512
 513         pid_list = rcu_dereference_raw(tr->filtered_no_pids);
 514         trace_filter_add_remove_task(pid_list, NULL, task);
 515 }
 516
 517 static void
 518 event_filter_pid_sched_process_fork(void *data,
 519                                     struct task_struct *self,
 520                                     struct task_struct *task)
 521 {
 522         struct trace_pid_list *pid_list;
 523         struct trace_array *tr = data;
 524
 525         pid_list = rcu_dereference_sched(tr->filtered_pids);
 526         trace_filter_add_remove_task(pid_list, self, task);
 527
 528         pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 529         trace_filter_add_remove_task(pid_list, self, task);
 530 }
 531
 532 void trace_event_follow_fork(struct trace_array *tr, bool enable)
 533 {
 534         if (enable) {
 535                 register_trace_prio_sched_process_fork(event_filter_pid_sched_process_fork,
 536                                                        tr, INT_MIN);
 537                 register_trace_prio_sched_process_free(event_filter_pid_sched_process_exit,
 538                                                        tr, INT_MAX);
 539         } else {
 540                 unregister_trace_sched_process_fork(event_filter_pid_sched_process_fork,
 541                                                     tr);
 542                 unregister_trace_sched_process_free(event_filter_pid_sched_process_exit,
 543                                                     tr);
 544         }
 545 }
 546
 547 static void
 548 event_filter_pid_sched_switch_probe_pre(void *data, bool preempt,
 549                     struct task_struct *prev, struct task_struct *next)
 550 {
 551         struct trace_array *tr = data;
 552         struct trace_pid_list *no_pid_list;
 553         struct trace_pid_list *pid_list;
 554         bool ret;
 555
 556         pid_list = rcu_dereference_sched(tr->filtered_pids);
 557         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 558
 559         /*
 560          * Sched switch is funny, as we only want to ignore it
 561          * in the notrace case if both prev and next should be ignored.
 562          */
 563         ret = trace_ignore_this_task(NULL, no_pid_list, prev) &&
 564                 trace_ignore_this_task(NULL, no_pid_list, next);
 565
 566         this_cpu_write(tr->array_buffer.data->ignore_pid, ret ||
 567                        (trace_ignore_this_task(pid_list, NULL, prev) &&
 568                         trace_ignore_this_task(pid_list, NULL, next)));
 569 }
 570
 571 static void
 572 event_filter_pid_sched_switch_probe_post(void *data, bool preempt,
 573                     struct task_struct *prev, struct task_struct *next)
 574 {
 575         struct trace_array *tr = data;
 576         struct trace_pid_list *no_pid_list;
 577         struct trace_pid_list *pid_list;
 578
 579         pid_list = rcu_dereference_sched(tr->filtered_pids);
 580         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 581
 582         this_cpu_write(tr->array_buffer.data->ignore_pid,
 583                        trace_ignore_this_task(pid_list, no_pid_list, next));
 584 }
 585
 586 static void
 587 event_filter_pid_sched_wakeup_probe_pre(void *data, struct task_struct *task)
 588 {
 589         struct trace_array *tr = data;
 590         struct trace_pid_list *no_pid_list;
 591         struct trace_pid_list *pid_list;
 592
 593         /* Nothing to do if we are already tracing */
 594         if (!this_cpu_read(tr->array_buffer.data->ignore_pid))
 595                 return;
 596
 597         pid_list = rcu_dereference_sched(tr->filtered_pids);
 598         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 599
 600         this_cpu_write(tr->array_buffer.data->ignore_pid,
 601                        trace_ignore_this_task(pid_list, no_pid_list, task));
 602 }
 603
 604 static void
 605 event_filter_pid_sched_wakeup_probe_post(void *data, struct task_struct *task)
 606 {
 607         struct trace_array *tr = data;
 608         struct trace_pid_list *no_pid_list;
 609         struct trace_pid_list *pid_list;
 610
 611         /* Nothing to do if we are not tracing */
 612         if (this_cpu_read(tr->array_buffer.data->ignore_pid))
 613                 return;
 614
 615         pid_list = rcu_dereference_sched(tr->filtered_pids);
 616         no_pid_list = rcu_dereference_sched(tr->filtered_no_pids);
 617
 618         /* Set tracing if current is enabled */
 619         this_cpu_write(tr->array_buffer.data->ignore_pid,
 620                        trace_ignore_this_task(pid_list, no_pid_list, current));
 621 }
 622
 623 static void unregister_pid_events(struct trace_array *tr)
 624 {
 625         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_pre, tr);
 626         unregister_trace_sched_switch(event_filter_pid_sched_switch_probe_post, tr);
 627
 628         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre, tr);
 629         unregister_trace_sched_wakeup(event_filter_pid_sched_wakeup_probe_post, tr);
 630
 631         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre, tr);
 632         unregister_trace_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post, tr);
 633
 634         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_pre, tr);
 635         unregister_trace_sched_waking(event_filter_pid_sched_wakeup_probe_post, tr);
 636 }
 637
 638 static void __ftrace_clear_event_pids(struct trace_array *tr, int type)
 639 {
 640         struct trace_pid_list *pid_list;
 641         struct trace_pid_list *no_pid_list;
 642         struct trace_event_file *file;
 643         int cpu;
 644
 645         pid_list = rcu_dereference_protected(tr->filtered_pids,
 646                                              lockdep_is_held(&event_mutex));
 647         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
 648                                              lockdep_is_held(&event_mutex));
 649
 650         /* Make sure there's something to do */
 651         if (!pid_type_enabled(type, pid_list, no_pid_list))
 652                 return;
 653
 654         if (!still_need_pid_events(type, pid_list, no_pid_list)) {
 655                 unregister_pid_events(tr);
 656
 657                 list_for_each_entry(file, &tr->events, list) {
 658                         clear_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
 659                 }
 660
 661                 for_each_possible_cpu(cpu)
 662                         per_cpu_ptr(tr->array_buffer.data, cpu)->ignore_pid = false;
 663         }
 664
 665         if (type & TRACE_PIDS)
 666                 rcu_assign_pointer(tr->filtered_pids, NULL);
 667
 668         if (type & TRACE_NO_PIDS)
 669                 rcu_assign_pointer(tr->filtered_no_pids, NULL);
 670
 671         /* Wait till all users are no longer using pid filtering */
 672         tracepoint_synchronize_unregister();
 673
 674         if ((type & TRACE_PIDS) && pid_list)
 675                 trace_free_pid_list(pid_list);
 676
 677         if ((type & TRACE_NO_PIDS) && no_pid_list)
 678                 trace_free_pid_list(no_pid_list);
 679 }
 680
 681 static void ftrace_clear_event_pids(struct trace_array *tr, int type)
 682 {
 683         mutex_lock(&event_mutex);
 684         __ftrace_clear_event_pids(tr, type);
 685         mutex_unlock(&event_mutex);
 686 }
 687
 688 static void __put_system(struct event_subsystem *system)
 689 {
 690         struct event_filter *filter = system->filter;
 691
 692         WARN_ON_ONCE(system_refcount(system) == 0);
 693         if (system_refcount_dec(system))
 694                 return;
 695
 696         list_del(&system->list);
 697
 698         if (filter) {
 699                 kfree(filter->filter_string);
 700                 kfree(filter);
 701         }
 702         kfree_const(system->name);
 703         kfree(system);
 704 }
 705
 706 static void __get_system(struct event_subsystem *system)
 707 {
 708         WARN_ON_ONCE(system_refcount(system) == 0);
 709         system_refcount_inc(system);
 710 }
 711
 712 static void __get_system_dir(struct trace_subsystem_dir *dir)
 713 {
 714         WARN_ON_ONCE(dir->ref_count == 0);
 715         dir->ref_count++;
 716         __get_system(dir->subsystem);
 717 }
 718
 719 static void __put_system_dir(struct trace_subsystem_dir *dir)
 720 {
 721         WARN_ON_ONCE(dir->ref_count == 0);
 722         /* If the subsystem is about to be freed, the dir must be too */
 723         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 724
 725         __put_system(dir->subsystem);
 726         if (!--dir->ref_count)
 727                 kfree(dir);
 728 }
 729
 730 static void put_system(struct trace_subsystem_dir *dir)
 731 {
 732         mutex_lock(&event_mutex);
 733         __put_system_dir(dir);
 734         mutex_unlock(&event_mutex);
 735 }
 736
 737 static void remove_subsystem(struct trace_subsystem_dir *dir)
 738 {
 739         if (!dir)
 740                 return;
 741
 742         if (!--dir->nr_events) {
 743                 tracefs_remove(dir->entry);
 744                 list_del(&dir->list);
 745                 __put_system_dir(dir);
 746         }
 747 }
 748
 749 void event_file_get(struct trace_event_file *file)
 750 {
 751         atomic_inc(&file->ref);
 752 }
 753
 754 void event_file_put(struct trace_event_file *file)
 755 {
 756         if (WARN_ON_ONCE(!atomic_read(&file->ref))) {
 757                 if (file->flags & EVENT_FILE_FL_FREED)
 758                         kmem_cache_free(file_cachep, file);
 759                 return;
 760         }
 761
 762         if (atomic_dec_and_test(&file->ref)) {
 763                 /* Count should only go to zero when it is freed */
 764                 if (WARN_ON_ONCE(!(file->flags & EVENT_FILE_FL_FREED)))
 765                         return;
 766                 kmem_cache_free(file_cachep, file);
 767         }
 768 }
 769
 770 static void remove_event_file_dir(struct trace_event_file *file)
 771 {
 772         struct dentry *dir = file->dir;
 773
 774         tracefs_remove(dir);
 775
 776         list_del(&file->list);
 777         remove_subsystem(file->system);
 778         free_event_filter(file->filter);
 779         file->flags |= EVENT_FILE_FL_FREED;
 780         event_file_put(file);
 781 }
 782
 783 /*
 784  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 785  */
 786 static int
 787 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 788                               const char *sub, const char *event, int set)
 789 {
 790         struct trace_event_file *file;
 791         struct trace_event_call *call;
 792         const char *name;
 793         int ret = -EINVAL;
 794         int eret = 0;
 795
 796         list_for_each_entry(file, &tr->events, list) {
 797
 798                 call = file->event_call;
 799                 name = trace_event_name(call);
 800
 801                 if (!name || !call->class || !call->class->reg)
 802                         continue;
 803
 804                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 805                         continue;
 806
 807                 if (match &&
 808                     strcmp(match, name) != 0 &&
 809                     strcmp(match, call->class->system) != 0)
 810                         continue;
 811
 812                 if (sub && strcmp(sub, call->class->system) != 0)
 813                         continue;
 814
 815                 if (event && strcmp(event, name) != 0)
 816                         continue;
 817
 818                 ret = ftrace_event_enable_disable(file, set);
 819
 820                 /*
 821                  * Save the first error and return that. Some events
 822                  * may still have been enabled, but let the user
 823                  * know that something went wrong.
 824                  */
 825                 if (ret && !eret)
 826                         eret = ret;
 827
 828                 ret = eret;
 829         }
 830
 831         return ret;
 832 }
 833
 834 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 835                                   const char *sub, const char *event, int set)
 836 {
 837         int ret;
 838
 839         mutex_lock(&event_mutex);
 840         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 841         mutex_unlock(&event_mutex);
 842
 843         return ret;
 844 }
 845
 846 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 847 {
 848         char *event = NULL, *sub = NULL, *match;
 849         int ret;
 850
 851         if (!tr)
 852                 return -ENOENT;
 853         /*
 854          * The buf format can be <subsystem>:<event-name>
 855          *  *:<event-name> means any event by that name.
 856          *  :<event-name> is the same.
 857          *
 858          *  <subsystem>:* means all events in that subsystem
 859          *  <subsystem>: means the same.
 860          *
 861          *  <name> (no ':') means all events in a subsystem with
 862          *  the name <name> or any event that matches <name>
 863          */
 864
 865         match = strsep(&buf, ":");
 866         if (buf) {
 867                 sub = match;
 868                 event = buf;
 869                 match = NULL;
 870
 871                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 872                         sub = NULL;
 873                 if (!strlen(event) || strcmp(event, "*") == 0)
 874                         event = NULL;
 875         }
 876
 877         ret = __ftrace_set_clr_event(tr, match, sub, event, set);
 878
 879         /* Put back the colon to allow this to be called again */
 880         if (buf)
 881                 *(buf - 1) = ':';
 882
 883         return ret;
 884 }
 885
 886 /**
 887  * trace_set_clr_event - enable or disable an event
 888  * @system: system name to match (NULL for any system)
 889  * @event: event name to match (NULL for all events, within system)
 890  * @set: 1 to enable, 0 to disable
 891  *
 892  * This is a way for other parts of the kernel to enable or disable
 893  * event recording.
 894  *
 895  * Returns 0 on success, -EINVAL if the parameters do not match any
 896  * registered events.
 897  */
 898 int trace_set_clr_event(const char *system, const char *event, int set)
 899 {
 900         struct trace_array *tr = top_trace_array();
 901
 902         if (!tr)
 903                 return -ENODEV;
 904
 905         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 906 }
 907 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 908
 909 /**
 910  * trace_array_set_clr_event - enable or disable an event for a trace array.
 911  * @tr: concerned trace array.
 912  * @system: system name to match (NULL for any system)
 913  * @event: event name to match (NULL for all events, within system)
 914  * @enable: true to enable, false to disable
 915  *
 916  * This is a way for other parts of the kernel to enable or disable
 917  * event recording.
 918  *
 919  * Returns 0 on success, -EINVAL if the parameters do not match any
 920  * registered events.
 921  */
 922 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
 923                 const char *event, bool enable)
 924 {
 925         int set;
 926
 927         if (!tr)
 928                 return -ENOENT;
 929
 930         set = (enable == true) ? 1 : 0;
 931         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 932 }
 933 EXPORT_SYMBOL_GPL(trace_array_set_clr_event);
 934
 935 /* 128 should be much more than enough */
 936 #define EVENT_BUF_SIZE          127
 937
 938 static ssize_t
 939 ftrace_event_write(struct file *file, const char __user *ubuf,
 940                    size_t cnt, loff_t *ppos)
 941 {
 942         struct trace_parser parser;
 943         struct seq_file *m = file->private_data;
 944         struct trace_array *tr = m->private;
 945         ssize_t read, ret;
 946
 947         if (!cnt)
 948                 return 0;
 949
 950         ret = tracing_update_buffers();
 951         if (ret < 0)
 952                 return ret;
 953
 954         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 955                 return -ENOMEM;
 956
 957         read = trace_get_user(&parser, ubuf, cnt, ppos);
 958
 959         if (read >= 0 && trace_parser_loaded((&parser))) {
 960                 int set = 1;
 961
 962                 if (*parser.buffer == '!')
 963                         set = 0;
 964
 965                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 966                 if (ret)
 967                         goto out_put;
 968         }
 969
 970         ret = read;
 971
 972  out_put:
 973         trace_parser_put(&parser);
 974
 975         return ret;
 976 }
 977
 978 static void *
 979 t_next(struct seq_file *m, void *v, loff_t *pos)
 980 {
 981         struct trace_event_file *file = v;
 982         struct trace_event_call *call;
 983         struct trace_array *tr = m->private;
 984
 985         (*pos)++;
 986
 987         list_for_each_entry_continue(file, &tr->events, list) {
 988                 call = file->event_call;
 989                 /*
 990                  * The ftrace subsystem is for showing formats only.
 991                  * They can not be enabled or disabled via the event files.
 992                  */
 993                 if (call->class && call->class->reg &&
 994                     !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
 995                         return file;
 996         }
 997
 998         return NULL;
 999 }
1000
1001 static void *t_start(struct seq_file *m, loff_t *pos)
1002 {
1003         struct trace_event_file *file;
1004         struct trace_array *tr = m->private;
1005         loff_t l;
1006
1007         mutex_lock(&event_mutex);
1008
1009         file = list_entry(&tr->events, struct trace_event_file, list);
1010         for (l = 0; l <= *pos; ) {
1011                 file = t_next(m, file, &l);
1012                 if (!file)
1013                         break;
1014         }
1015         return file;
1016 }
1017
1018 static void *
1019 s_next(struct seq_file *m, void *v, loff_t *pos)
1020 {
1021         struct trace_event_file *file = v;
1022         struct trace_array *tr = m->private;
1023
1024         (*pos)++;
1025
1026         list_for_each_entry_continue(file, &tr->events, list) {
1027                 if (file->flags & EVENT_FILE_FL_ENABLED)
1028                         return file;
1029         }
1030
1031         return NULL;
1032 }
1033
1034 static void *s_start(struct seq_file *m, loff_t *pos)
1035 {
1036         struct trace_event_file *file;
1037         struct trace_array *tr = m->private;
1038         loff_t l;
1039
1040         mutex_lock(&event_mutex);
1041
1042         file = list_entry(&tr->events, struct trace_event_file, list);
1043         for (l = 0; l <= *pos; ) {
1044                 file = s_next(m, file, &l);
1045                 if (!file)
1046                         break;
1047         }
1048         return file;
1049 }
1050
1051 static int t_show(struct seq_file *m, void *v)
1052 {
1053         struct trace_event_file *file = v;
1054         struct trace_event_call *call = file->event_call;
1055
1056         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
1057                 seq_printf(m, "%s:", call->class->system);
1058         seq_printf(m, "%s\n", trace_event_name(call));
1059
1060         return 0;
1061 }
1062
1063 static void t_stop(struct seq_file *m, void *p)
1064 {
1065         mutex_unlock(&event_mutex);
1066 }
1067
1068 static void *
1069 __next(struct seq_file *m, void *v, loff_t *pos, int type)
1070 {
1071         struct trace_array *tr = m->private;
1072         struct trace_pid_list *pid_list;
1073
1074         if (type == TRACE_PIDS)
1075                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1076         else
1077                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1078
1079         return trace_pid_next(pid_list, v, pos);
1080 }
1081
1082 static void *
1083 p_next(struct seq_file *m, void *v, loff_t *pos)
1084 {
1085         return __next(m, v, pos, TRACE_PIDS);
1086 }
1087
1088 static void *
1089 np_next(struct seq_file *m, void *v, loff_t *pos)
1090 {
1091         return __next(m, v, pos, TRACE_NO_PIDS);
1092 }
1093
1094 static void *__start(struct seq_file *m, loff_t *pos, int type)
1095         __acquires(RCU)
1096 {
1097         struct trace_pid_list *pid_list;
1098         struct trace_array *tr = m->private;
1099
1100         /*
1101          * Grab the mutex, to keep calls to p_next() having the same
1102          * tr->filtered_pids as p_start() has.
1103          * If we just passed the tr->filtered_pids around, then RCU would
1104          * have been enough, but doing that makes things more complex.
1105          */
1106         mutex_lock(&event_mutex);
1107         rcu_read_lock_sched();
1108
1109         if (type == TRACE_PIDS)
1110                 pid_list = rcu_dereference_sched(tr->filtered_pids);
1111         else
1112                 pid_list = rcu_dereference_sched(tr->filtered_no_pids);
1113
1114         if (!pid_list)
1115                 return NULL;
1116
1117         return trace_pid_start(pid_list, pos);
1118 }
1119
1120 static void *p_start(struct seq_file *m, loff_t *pos)
1121         __acquires(RCU)
1122 {
1123         return __start(m, pos, TRACE_PIDS);
1124 }
1125
1126 static void *np_start(struct seq_file *m, loff_t *pos)
1127         __acquires(RCU)
1128 {
1129         return __start(m, pos, TRACE_NO_PIDS);
1130 }
1131
1132 static void p_stop(struct seq_file *m, void *p)
1133         __releases(RCU)
1134 {
1135         rcu_read_unlock_sched();
1136         mutex_unlock(&event_mutex);
1137 }
1138
1139 static ssize_t
1140 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1141                   loff_t *ppos)
1142 {
1143         struct trace_event_file *file;
1144         unsigned long flags;
1145         char buf[4] = "0";
1146
1147         mutex_lock(&event_mutex);
1148         file = event_file_data(filp);
1149         if (likely(file))
1150                 flags = file->flags;
1151         mutex_unlock(&event_mutex);
1152
1153         if (!file || flags & EVENT_FILE_FL_FREED)
1154                 return -ENODEV;
1155
1156         if (flags & EVENT_FILE_FL_ENABLED &&
1157             !(flags & EVENT_FILE_FL_SOFT_DISABLED))
1158                 strcpy(buf, "1");
1159
1160         if (flags & EVENT_FILE_FL_SOFT_DISABLED ||
1161             flags & EVENT_FILE_FL_SOFT_MODE)
1162                 strcat(buf, "*");
1163
1164         strcat(buf, "\n");
1165
1166         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
1167 }
1168
1169 static ssize_t
1170 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1171                    loff_t *ppos)
1172 {
1173         struct trace_event_file *file;
1174         unsigned long val;
1175         int ret;
1176
1177         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1178         if (ret)
1179                 return ret;
1180
1181         ret = tracing_update_buffers();
1182         if (ret < 0)
1183                 return ret;
1184
1185         switch (val) {
1186         case 0:
1187         case 1:
1188                 ret = -ENODEV;
1189                 mutex_lock(&event_mutex);
1190                 file = event_file_data(filp);
1191                 if (likely(file && !(file->flags & EVENT_FILE_FL_FREED)))
1192                         ret = ftrace_event_enable_disable(file, val);
1193                 mutex_unlock(&event_mutex);
1194                 break;
1195
1196         default:
1197                 return -EINVAL;
1198         }
1199
1200         *ppos += cnt;
1201
1202         return ret ? ret : cnt;
1203 }
1204
1205 static ssize_t
1206 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
1207                    loff_t *ppos)
1208 {
1209         const char set_to_char[4] = { '?', '0', '1', 'X' };
1210         struct trace_subsystem_dir *dir = filp->private_data;
1211         struct event_subsystem *system = dir->subsystem;
1212         struct trace_event_call *call;
1213         struct trace_event_file *file;
1214         struct trace_array *tr = dir->tr;
1215         char buf[2];
1216         int set = 0;
1217         int ret;
1218
1219         mutex_lock(&event_mutex);
1220         list_for_each_entry(file, &tr->events, list) {
1221                 call = file->event_call;
1222                 if ((call->flags & TRACE_EVENT_FL_IGNORE_ENABLE) ||
1223                     !trace_event_name(call) || !call->class || !call->class->reg)
1224                         continue;
1225
1226                 if (system && strcmp(call->class->system, system->name) != 0)
1227                         continue;
1228
1229                 /*
1230                  * We need to find out if all the events are set
1231                  * or if all events or cleared, or if we have
1232                  * a mixture.
1233                  */
1234                 set |= (1 << !!(file->flags & EVENT_FILE_FL_ENABLED));
1235
1236                 /*
1237                  * If we have a mixture, no need to look further.
1238                  */
1239                 if (set == 3)
1240                         break;
1241         }
1242         mutex_unlock(&event_mutex);
1243
1244         buf[0] = set_to_char[set];
1245         buf[1] = '\n';
1246
1247         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
1248
1249         return ret;
1250 }
1251
1252 static ssize_t
1253 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
1254                     loff_t *ppos)
1255 {
1256         struct trace_subsystem_dir *dir = filp->private_data;
1257         struct event_subsystem *system = dir->subsystem;
1258         const char *name = NULL;
1259         unsigned long val;
1260         ssize_t ret;
1261
1262         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1263         if (ret)
1264                 return ret;
1265
1266         ret = tracing_update_buffers();
1267         if (ret < 0)
1268                 return ret;
1269
1270         if (val != 0 && val != 1)
1271                 return -EINVAL;
1272
1273         /*
1274          * Opening of "enable" adds a ref count to system,
1275          * so the name is safe to use.
1276          */
1277         if (system)
1278                 name = system->name;
1279
1280         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
1281         if (ret)
1282                 goto out;
1283
1284         ret = cnt;
1285
1286 out:
1287         *ppos += cnt;
1288
1289         return ret;
1290 }
1291
1292 enum {
1293         FORMAT_HEADER           = 1,
1294         FORMAT_FIELD_SEPERATOR  = 2,
1295         FORMAT_PRINTFMT         = 3,
1296 };
1297
1298 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
1299 {
1300         struct trace_event_call *call = event_file_data(m->private);
1301         struct list_head *common_head = &ftrace_common_fields;
1302         struct list_head *head = trace_get_fields(call);
1303         struct list_head *node = v;
1304
1305         (*pos)++;
1306
1307         switch ((unsigned long)v) {
1308         case FORMAT_HEADER:
1309                 node = common_head;
1310                 break;
1311
1312         case FORMAT_FIELD_SEPERATOR:
1313                 node = head;
1314                 break;
1315
1316         case FORMAT_PRINTFMT:
1317                 /* all done */
1318                 return NULL;
1319         }
1320
1321         node = node->prev;
1322         if (node == common_head)
1323                 return (void *)FORMAT_FIELD_SEPERATOR;
1324         else if (node == head)
1325                 return (void *)FORMAT_PRINTFMT;
1326         else
1327                 return node;
1328 }
1329
1330 static int f_show(struct seq_file *m, void *v)
1331 {
1332         struct trace_event_call *call = event_file_data(m->private);
1333         struct ftrace_event_field *field;
1334         const char *array_descriptor;
1335
1336         switch ((unsigned long)v) {
1337         case FORMAT_HEADER:
1338                 seq_printf(m, "name: %s\n", trace_event_name(call));
1339                 seq_printf(m, "ID: %d\n", call->event.type);
1340                 seq_puts(m, "format:\n");
1341                 return 0;
1342
1343         case FORMAT_FIELD_SEPERATOR:
1344                 seq_putc(m, '\n');
1345                 return 0;
1346
1347         case FORMAT_PRINTFMT:
1348                 seq_printf(m, "\nprint fmt: %s\n",
1349                            call->print_fmt);
1350                 return 0;
1351         }
1352
1353         field = list_entry(v, struct ftrace_event_field, link);
1354         /*
1355          * Smartly shows the array type(except dynamic array).
1356          * Normal:
1357          *      field:TYPE VAR
1358          * If TYPE := TYPE[LEN], it is shown:
1359          *      field:TYPE VAR[LEN]
1360          */
1361         array_descriptor = strchr(field->type, '[');
1362
1363         if (str_has_prefix(field->type, "__data_loc"))
1364                 array_descriptor = NULL;
1365
1366         if (!array_descriptor)
1367                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1368                            field->type, field->name, field->offset,
1369                            field->size, !!field->is_signed);
1370         else
1371                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
1372                            (int)(array_descriptor - field->type),
1373                            field->type, field->name,
1374                            array_descriptor, field->offset,
1375                            field->size, !!field->is_signed);
1376
1377         return 0;
1378 }
1379
1380 static void *f_start(struct seq_file *m, loff_t *pos)
1381 {
1382         void *p = (void *)FORMAT_HEADER;
1383         loff_t l = 0;
1384
1385         /* ->stop() is called even if ->start() fails */
1386         mutex_lock(&event_mutex);
1387         if (!event_file_data(m->private))
1388                 return ERR_PTR(-ENODEV);
1389
1390         while (l < *pos && p)
1391                 p = f_next(m, p, &l);
1392
1393         return p;
1394 }
1395
1396 static void f_stop(struct seq_file *m, void *p)
1397 {
1398         mutex_unlock(&event_mutex);
1399 }
1400
1401 static const struct seq_operations trace_format_seq_ops = {
1402         .start          = f_start,
1403         .next           = f_next,
1404         .stop           = f_stop,
1405         .show           = f_show,
1406 };
1407
1408 static int trace_format_open(struct inode *inode, struct file *file)
1409 {
1410         struct seq_file *m;
1411         int ret;
1412
1413         /* Do we want to hide event format files on tracefs lockdown? */
1414
1415         ret = seq_open(file, &trace_format_seq_ops);
1416         if (ret < 0)
1417                 return ret;
1418
1419         m = file->private_data;
1420         m->private = file;
1421
1422         return 0;
1423 }
1424
1425 static ssize_t
1426 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1427 {
1428         int id = (long)event_file_data(filp);
1429         char buf[32];
1430         int len;
1431
1432         if (unlikely(!id))
1433                 return -ENODEV;
1434
1435         len = sprintf(buf, "%d\n", id);
1436
1437         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
1438 }
1439
1440 static ssize_t
1441 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1442                   loff_t *ppos)
1443 {
1444         struct trace_event_file *file;
1445         struct trace_seq *s;
1446         int r = -ENODEV;
1447
1448         if (*ppos)
1449                 return 0;
1450
1451         s = kmalloc(sizeof(*s), GFP_KERNEL);
1452
1453         if (!s)
1454                 return -ENOMEM;
1455
1456         trace_seq_init(s);
1457
1458         mutex_lock(&event_mutex);
1459         file = event_file_data(filp);
1460         if (file && !(file->flags & EVENT_FILE_FL_FREED))
1461                 print_event_filter(file, s);
1462         mutex_unlock(&event_mutex);
1463
1464         if (file)
1465                 r = simple_read_from_buffer(ubuf, cnt, ppos,
1466                                             s->buffer, trace_seq_used(s));
1467
1468         kfree(s);
1469
1470         return r;
1471 }
1472
1473 static ssize_t
1474 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1475                    loff_t *ppos)
1476 {
1477         struct trace_event_file *file;
1478         char *buf;
1479         int err = -ENODEV;
1480
1481         if (cnt >= PAGE_SIZE)
1482                 return -EINVAL;
1483
1484         buf = memdup_user_nul(ubuf, cnt);
1485         if (IS_ERR(buf))
1486                 return PTR_ERR(buf);
1487
1488         mutex_lock(&event_mutex);
1489         file = event_file_data(filp);
1490         if (file)
1491                 err = apply_event_filter(file, buf);
1492         mutex_unlock(&event_mutex);
1493
1494         kfree(buf);
1495         if (err < 0)
1496                 return err;
1497
1498         *ppos += cnt;
1499
1500         return cnt;
1501 }
1502
1503 static LIST_HEAD(event_subsystems);
1504
1505 static int subsystem_open(struct inode *inode, struct file *filp)
1506 {
1507         struct event_subsystem *system = NULL;
1508         struct trace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1509         struct trace_array *tr;
1510         int ret;
1511
1512         if (tracing_is_disabled())
1513                 return -ENODEV;
1514
1515         /* Make sure the system still exists */
1516         mutex_lock(&event_mutex);
1517         mutex_lock(&trace_types_lock);
1518         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1519                 list_for_each_entry(dir, &tr->systems, list) {
1520                         if (dir == inode->i_private) {
1521                                 /* Don't open systems with no events */
1522                                 if (dir->nr_events) {
1523                                         __get_system_dir(dir);
1524                                         system = dir->subsystem;
1525                                 }
1526                                 goto exit_loop;
1527                         }
1528                 }
1529         }
1530  exit_loop:
1531         mutex_unlock(&trace_types_lock);
1532         mutex_unlock(&event_mutex);
1533
1534         if (!system)
1535                 return -ENODEV;
1536
1537         /* Some versions of gcc think dir can be uninitialized here */
1538         WARN_ON(!dir);
1539
1540         /* Still need to increment the ref count of the system */
1541         if (trace_array_get(tr) < 0) {
1542                 put_system(dir);
1543                 return -ENODEV;
1544         }
1545
1546         ret = tracing_open_generic(inode, filp);
1547         if (ret < 0) {
1548                 trace_array_put(tr);
1549                 put_system(dir);
1550         }
1551
1552         return ret;
1553 }
1554
1555 static int system_tr_open(struct inode *inode, struct file *filp)
1556 {
1557         struct trace_subsystem_dir *dir;
1558         struct trace_array *tr = inode->i_private;
1559         int ret;
1560
1561         /* Make a temporary dir that has no system but points to tr */
1562         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1563         if (!dir)
1564                 return -ENOMEM;
1565
1566         ret = tracing_open_generic_tr(inode, filp);
1567         if (ret < 0) {
1568                 kfree(dir);
1569                 return ret;
1570         }
1571         dir->tr = tr;
1572         filp->private_data = dir;
1573
1574         return 0;
1575 }
1576
1577 static int subsystem_release(struct inode *inode, struct file *file)
1578 {
1579         struct trace_subsystem_dir *dir = file->private_data;
1580
1581         trace_array_put(dir->tr);
1582
1583         /*
1584          * If dir->subsystem is NULL, then this is a temporary
1585          * descriptor that was made for a trace_array to enable
1586          * all subsystems.
1587          */
1588         if (dir->subsystem)
1589                 put_system(dir);
1590         else
1591                 kfree(dir);
1592
1593         return 0;
1594 }
1595
1596 static ssize_t
1597 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1598                       loff_t *ppos)
1599 {
1600         struct trace_subsystem_dir *dir = filp->private_data;
1601         struct event_subsystem *system = dir->subsystem;
1602         struct trace_seq *s;
1603         int r;
1604
1605         if (*ppos)
1606                 return 0;
1607
1608         s = kmalloc(sizeof(*s), GFP_KERNEL);
1609         if (!s)
1610                 return -ENOMEM;
1611
1612         trace_seq_init(s);
1613
1614         print_subsystem_event_filter(system, s);
1615         r = simple_read_from_buffer(ubuf, cnt, ppos,
1616                                     s->buffer, trace_seq_used(s));
1617
1618         kfree(s);
1619
1620         return r;
1621 }
1622
1623 static ssize_t
1624 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1625                        loff_t *ppos)
1626 {
1627         struct trace_subsystem_dir *dir = filp->private_data;
1628         char *buf;
1629         int err;
1630
1631         if (cnt >= PAGE_SIZE)
1632                 return -EINVAL;
1633
1634         buf = memdup_user_nul(ubuf, cnt);
1635         if (IS_ERR(buf))
1636                 return PTR_ERR(buf);
1637
1638         err = apply_subsystem_event_filter(dir, buf);
1639         kfree(buf);
1640         if (err < 0)
1641                 return err;
1642
1643         *ppos += cnt;
1644
1645         return cnt;
1646 }
1647
1648 static ssize_t
1649 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1650 {
1651         int (*func)(struct trace_seq *s) = filp->private_data;
1652         struct trace_seq *s;
1653         int r;
1654
1655         if (*ppos)
1656                 return 0;
1657
1658         s = kmalloc(sizeof(*s), GFP_KERNEL);
1659         if (!s)
1660                 return -ENOMEM;
1661
1662         trace_seq_init(s);
1663
1664         func(s);
1665         r = simple_read_from_buffer(ubuf, cnt, ppos,
1666                                     s->buffer, trace_seq_used(s));
1667
1668         kfree(s);
1669
1670         return r;
1671 }
1672
1673 static void ignore_task_cpu(void *data)
1674 {
1675         struct trace_array *tr = data;
1676         struct trace_pid_list *pid_list;
1677         struct trace_pid_list *no_pid_list;
1678
1679         /*
1680          * This function is called by on_each_cpu() while the
1681          * event_mutex is held.
1682          */
1683         pid_list = rcu_dereference_protected(tr->filtered_pids,
1684                                              mutex_is_locked(&event_mutex));
1685         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
1686                                              mutex_is_locked(&event_mutex));
1687
1688         this_cpu_write(tr->array_buffer.data->ignore_pid,
1689                        trace_ignore_this_task(pid_list, no_pid_list, current));
1690 }
1691
1692 static void register_pid_events(struct trace_array *tr)
1693 {
1694         /*
1695          * Register a probe that is called before all other probes
1696          * to set ignore_pid if next or prev do not match.
1697          * Register a probe this is called after all other probes
1698          * to only keep ignore_pid set if next pid matches.
1699          */
1700         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_pre,
1701                                          tr, INT_MAX);
1702         register_trace_prio_sched_switch(event_filter_pid_sched_switch_probe_post,
1703                                          tr, 0);
1704
1705         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_pre,
1706                                          tr, INT_MAX);
1707         register_trace_prio_sched_wakeup(event_filter_pid_sched_wakeup_probe_post,
1708                                          tr, 0);
1709
1710         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_pre,
1711                                              tr, INT_MAX);
1712         register_trace_prio_sched_wakeup_new(event_filter_pid_sched_wakeup_probe_post,
1713                                              tr, 0);
1714
1715         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_pre,
1716                                          tr, INT_MAX);
1717         register_trace_prio_sched_waking(event_filter_pid_sched_wakeup_probe_post,
1718                                          tr, 0);
1719 }
1720
1721 static ssize_t
1722 event_pid_write(struct file *filp, const char __user *ubuf,
1723                 size_t cnt, loff_t *ppos, int type)
1724 {
1725         struct seq_file *m = filp->private_data;
1726         struct trace_array *tr = m->private;
1727         struct trace_pid_list *filtered_pids = NULL;
1728         struct trace_pid_list *other_pids = NULL;
1729         struct trace_pid_list *pid_list;
1730         struct trace_event_file *file;
1731         ssize_t ret;
1732
1733         if (!cnt)
1734                 return 0;
1735
1736         ret = tracing_update_buffers();
1737         if (ret < 0)
1738                 return ret;
1739
1740         mutex_lock(&event_mutex);
1741
1742         if (type == TRACE_PIDS) {
1743                 filtered_pids = rcu_dereference_protected(tr->filtered_pids,
1744                                                           lockdep_is_held(&event_mutex));
1745                 other_pids = rcu_dereference_protected(tr->filtered_no_pids,
1746                                                           lockdep_is_held(&event_mutex));
1747         } else {
1748                 filtered_pids = rcu_dereference_protected(tr->filtered_no_pids,
1749                                                           lockdep_is_held(&event_mutex));
1750                 other_pids = rcu_dereference_protected(tr->filtered_pids,
1751                                                           lockdep_is_held(&event_mutex));
1752         }
1753
1754         ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
1755         if (ret < 0)
1756                 goto out;
1757
1758         if (type == TRACE_PIDS)
1759                 rcu_assign_pointer(tr->filtered_pids, pid_list);
1760         else
1761                 rcu_assign_pointer(tr->filtered_no_pids, pid_list);
1762
1763         list_for_each_entry(file, &tr->events, list) {
1764                 set_bit(EVENT_FILE_FL_PID_FILTER_BIT, &file->flags);
1765         }
1766
1767         if (filtered_pids) {
1768                 tracepoint_synchronize_unregister();
1769                 trace_free_pid_list(filtered_pids);
1770         } else if (pid_list && !other_pids) {
1771                 register_pid_events(tr);
1772         }
1773
1774         /*
1775          * Ignoring of pids is done at task switch. But we have to
1776          * check for those tasks that are currently running.
1777          * Always do this in case a pid was appended or removed.
1778          */
1779         on_each_cpu(ignore_task_cpu, tr, 1);
1780
1781  out:
1782         mutex_unlock(&event_mutex);
1783
1784         if (ret > 0)
1785                 *ppos += ret;
1786
1787         return ret;
1788 }
1789
1790 static ssize_t
1791 ftrace_event_pid_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_PIDS);
1795 }
1796
1797 static ssize_t
1798 ftrace_event_npid_write(struct file *filp, const char __user *ubuf,
1799                         size_t cnt, loff_t *ppos)
1800 {
1801         return event_pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
1802 }
1803
1804 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1805 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1806 static int ftrace_event_set_pid_open(struct inode *inode, struct file *file);
1807 static int ftrace_event_set_npid_open(struct inode *inode, struct file *file);
1808 static int ftrace_event_release(struct inode *inode, struct file *file);
1809
1810 static const struct seq_operations show_event_seq_ops = {
1811         .start = t_start,
1812         .next = t_next,
1813         .show = t_show,
1814         .stop = t_stop,
1815 };
1816
1817 static const struct seq_operations show_set_event_seq_ops = {
1818         .start = s_start,
1819         .next = s_next,
1820         .show = t_show,
1821         .stop = t_stop,
1822 };
1823
1824 static const struct seq_operations show_set_pid_seq_ops = {
1825         .start = p_start,
1826         .next = p_next,
1827         .show = trace_pid_show,
1828         .stop = p_stop,
1829 };
1830
1831 static const struct seq_operations show_set_no_pid_seq_ops = {
1832         .start = np_start,
1833         .next = np_next,
1834         .show = trace_pid_show,
1835         .stop = p_stop,
1836 };
1837
1838 static const struct file_operations ftrace_avail_fops = {
1839         .open = ftrace_event_avail_open,
1840         .read = seq_read,
1841         .llseek = seq_lseek,
1842         .release = seq_release,
1843 };
1844
1845 static const struct file_operations ftrace_set_event_fops = {
1846         .open = ftrace_event_set_open,
1847         .read = seq_read,
1848         .write = ftrace_event_write,
1849         .llseek = seq_lseek,
1850         .release = ftrace_event_release,
1851 };
1852
1853 static const struct file_operations ftrace_set_event_pid_fops = {
1854         .open = ftrace_event_set_pid_open,
1855         .read = seq_read,
1856         .write = ftrace_event_pid_write,
1857         .llseek = seq_lseek,
1858         .release = ftrace_event_release,
1859 };
1860
1861 static const struct file_operations ftrace_set_event_notrace_pid_fops = {
1862         .open = ftrace_event_set_npid_open,
1863         .read = seq_read,
1864         .write = ftrace_event_npid_write,
1865         .llseek = seq_lseek,
1866         .release = ftrace_event_release,
1867 };
1868
1869 static const struct file_operations ftrace_enable_fops = {
1870         .open = tracing_open_file_tr,
1871         .read = event_enable_read,
1872         .write = event_enable_write,
1873         .release = tracing_release_file_tr,
1874         .llseek = default_llseek,
1875 };
1876
1877 static const struct file_operations ftrace_event_format_fops = {
1878         .open = trace_format_open,
1879         .read = seq_read,
1880         .llseek = seq_lseek,
1881         .release = seq_release,
1882 };
1883
1884 static const struct file_operations ftrace_event_id_fops = {
1885         .read = event_id_read,
1886         .llseek = default_llseek,
1887 };
1888
1889 static const struct file_operations ftrace_event_filter_fops = {
1890         .open = tracing_open_file_tr,
1891         .read = event_filter_read,
1892         .write = event_filter_write,
1893         .release = tracing_release_file_tr,
1894         .llseek = default_llseek,
1895 };
1896
1897 static const struct file_operations ftrace_subsystem_filter_fops = {
1898         .open = subsystem_open,
1899         .read = subsystem_filter_read,
1900         .write = subsystem_filter_write,
1901         .llseek = default_llseek,
1902         .release = subsystem_release,
1903 };
1904
1905 static const struct file_operations ftrace_system_enable_fops = {
1906         .open = subsystem_open,
1907         .read = system_enable_read,
1908         .write = system_enable_write,
1909         .llseek = default_llseek,
1910         .release = subsystem_release,
1911 };
1912
1913 static const struct file_operations ftrace_tr_enable_fops = {
1914         .open = system_tr_open,
1915         .read = system_enable_read,
1916         .write = system_enable_write,
1917         .llseek = default_llseek,
1918         .release = subsystem_release,
1919 };
1920
1921 static const struct file_operations ftrace_show_header_fops = {
1922         .open = tracing_open_generic,
1923         .read = show_header,
1924         .llseek = default_llseek,
1925 };
1926
1927 static int
1928 ftrace_event_open(struct inode *inode, struct file *file,
1929                   const struct seq_operations *seq_ops)
1930 {
1931         struct seq_file *m;
1932         int ret;
1933
1934         ret = security_locked_down(LOCKDOWN_TRACEFS);
1935         if (ret)
1936                 return ret;
1937
1938         ret = seq_open(file, seq_ops);
1939         if (ret < 0)
1940                 return ret;
1941         m = file->private_data;
1942         /* copy tr over to seq ops */
1943         m->private = inode->i_private;
1944
1945         return ret;
1946 }
1947
1948 static int ftrace_event_release(struct inode *inode, struct file *file)
1949 {
1950         struct trace_array *tr = inode->i_private;
1951
1952         trace_array_put(tr);
1953
1954         return seq_release(inode, file);
1955 }
1956
1957 static int
1958 ftrace_event_avail_open(struct inode *inode, struct file *file)
1959 {
1960         const struct seq_operations *seq_ops = &show_event_seq_ops;
1961
1962         /* Checks for tracefs lockdown */
1963         return ftrace_event_open(inode, file, seq_ops);
1964 }
1965
1966 static int
1967 ftrace_event_set_open(struct inode *inode, struct file *file)
1968 {
1969         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1970         struct trace_array *tr = inode->i_private;
1971         int ret;
1972
1973         ret = tracing_check_open_get_tr(tr);
1974         if (ret)
1975                 return ret;
1976
1977         if ((file->f_mode & FMODE_WRITE) &&
1978             (file->f_flags & O_TRUNC))
1979                 ftrace_clear_events(tr);
1980
1981         ret = ftrace_event_open(inode, file, seq_ops);
1982         if (ret < 0)
1983                 trace_array_put(tr);
1984         return ret;
1985 }
1986
1987 static int
1988 ftrace_event_set_pid_open(struct inode *inode, struct file *file)
1989 {
1990         const struct seq_operations *seq_ops = &show_set_pid_seq_ops;
1991         struct trace_array *tr = inode->i_private;
1992         int ret;
1993
1994         ret = tracing_check_open_get_tr(tr);
1995         if (ret)
1996                 return ret;
1997
1998         if ((file->f_mode & FMODE_WRITE) &&
1999             (file->f_flags & O_TRUNC))
2000                 ftrace_clear_event_pids(tr, TRACE_PIDS);
2001
2002         ret = ftrace_event_open(inode, file, seq_ops);
2003         if (ret < 0)
2004                 trace_array_put(tr);
2005         return ret;
2006 }
2007
2008 static int
2009 ftrace_event_set_npid_open(struct inode *inode, struct file *file)
2010 {
2011         const struct seq_operations *seq_ops = &show_set_no_pid_seq_ops;
2012         struct trace_array *tr = inode->i_private;
2013         int ret;
2014
2015         ret = tracing_check_open_get_tr(tr);
2016         if (ret)
2017                 return ret;
2018
2019         if ((file->f_mode & FMODE_WRITE) &&
2020             (file->f_flags & O_TRUNC))
2021                 ftrace_clear_event_pids(tr, TRACE_NO_PIDS);
2022
2023         ret = ftrace_event_open(inode, file, seq_ops);
2024         if (ret < 0)
2025                 trace_array_put(tr);
2026         return ret;
2027 }
2028
2029 static struct event_subsystem *
2030 create_new_subsystem(const char *name)
2031 {
2032         struct event_subsystem *system;
2033
2034         /* need to create new entry */
2035         system = kmalloc(sizeof(*system), GFP_KERNEL);
2036         if (!system)
2037                 return NULL;
2038
2039         system->ref_count = 1;
2040
2041         /* Only allocate if dynamic (kprobes and modules) */
2042         system->name = kstrdup_const(name, GFP_KERNEL);
2043         if (!system->name)
2044                 goto out_free;
2045
2046         system->filter = NULL;
2047
2048         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
2049         if (!system->filter)
2050                 goto out_free;
2051
2052         list_add(&system->list, &event_subsystems);
2053
2054         return system;
2055
2056  out_free:
2057         kfree_const(system->name);
2058         kfree(system);
2059         return NULL;
2060 }
2061
2062 static struct dentry *
2063 event_subsystem_dir(struct trace_array *tr, const char *name,
2064                     struct trace_event_file *file, struct dentry *parent)
2065 {
2066         struct trace_subsystem_dir *dir;
2067         struct event_subsystem *system;
2068         struct dentry *entry;
2069
2070         /* First see if we did not already create this dir */
2071         list_for_each_entry(dir, &tr->systems, list) {
2072                 system = dir->subsystem;
2073                 if (strcmp(system->name, name) == 0) {
2074                         dir->nr_events++;
2075                         file->system = dir;
2076                         return dir->entry;
2077                 }
2078         }
2079
2080         /* Now see if the system itself exists. */
2081         list_for_each_entry(system, &event_subsystems, list) {
2082                 if (strcmp(system->name, name) == 0)
2083                         break;
2084         }
2085         /* Reset system variable when not found */
2086         if (&system->list == &event_subsystems)
2087                 system = NULL;
2088
2089         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
2090         if (!dir)
2091                 goto out_fail;
2092
2093         if (!system) {
2094                 system = create_new_subsystem(name);
2095                 if (!system)
2096                         goto out_free;
2097         } else
2098                 __get_system(system);
2099
2100         dir->entry = tracefs_create_dir(name, parent);
2101         if (!dir->entry) {
2102                 pr_warn("Failed to create system directory %s\n", name);
2103                 __put_system(system);
2104                 goto out_free;
2105         }
2106
2107         dir->tr = tr;
2108         dir->ref_count = 1;
2109         dir->nr_events = 1;
2110         dir->subsystem = system;
2111         file->system = dir;
2112
2113         entry = tracefs_create_file("filter", 0644, dir->entry, dir,
2114                                     &ftrace_subsystem_filter_fops);
2115         if (!entry) {
2116                 kfree(system->filter);
2117                 system->filter = NULL;
2118                 pr_warn("Could not create tracefs '%s/filter' entry\n", name);
2119         }
2120
2121         trace_create_file("enable", 0644, dir->entry, dir,
2122                           &ftrace_system_enable_fops);
2123
2124         list_add(&dir->list, &tr->systems);
2125
2126         return dir->entry;
2127
2128  out_free:
2129         kfree(dir);
2130  out_fail:
2131         /* Only print this message if failed on memory allocation */
2132         if (!dir || !system)
2133                 pr_warn("No memory to create event subsystem %s\n", name);
2134         return NULL;
2135 }
2136
2137 static int
2138 event_define_fields(struct trace_event_call *call)
2139 {
2140         struct list_head *head;
2141         int ret = 0;
2142
2143         /*
2144          * Other events may have the same class. Only update
2145          * the fields if they are not already defined.
2146          */
2147         head = trace_get_fields(call);
2148         if (list_empty(head)) {
2149                 struct trace_event_fields *field = call->class->fields_array;
2150                 unsigned int offset = sizeof(struct trace_entry);
2151
2152                 for (; field->type; field++) {
2153                         if (field->type == TRACE_FUNCTION_TYPE) {
2154                                 field->define_fields(call);
2155                                 break;
2156                         }
2157
2158                         offset = ALIGN(offset, field->align);
2159                         ret = trace_define_field(call, field->type, field->name,
2160                                                  offset, field->size,
2161                                                  field->is_signed, field->filter_type);
2162                         if (WARN_ON_ONCE(ret)) {
2163                                 pr_err("error code is %d\n", ret);
2164                                 break;
2165                         }
2166
2167                         offset += field->size;
2168                 }
2169         }
2170
2171         return ret;
2172 }
2173
2174 static int
2175 event_create_dir(struct dentry *parent, struct trace_event_file *file)
2176 {
2177         struct trace_event_call *call = file->event_call;
2178         struct trace_array *tr = file->tr;
2179         struct dentry *d_events;
2180         const char *name;
2181         int ret;
2182
2183         /*
2184          * If the trace point header did not define TRACE_SYSTEM
2185          * then the system would be called "TRACE_SYSTEM".
2186          */
2187         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
2188                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
2189                 if (!d_events)
2190                         return -ENOMEM;
2191         } else
2192                 d_events = parent;
2193
2194         name = trace_event_name(call);
2195         file->dir = tracefs_create_dir(name, d_events);
2196         if (!file->dir) {
2197                 pr_warn("Could not create tracefs '%s' directory\n", name);
2198                 return -1;
2199         }
2200
2201         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
2202                 trace_create_file("enable", 0644, file->dir, file,
2203                                   &ftrace_enable_fops);
2204
2205 #ifdef CONFIG_PERF_EVENTS
2206         if (call->event.type && call->class->reg)
2207                 trace_create_file("id", 0444, file->dir,
2208                                   (void *)(long)call->event.type,
2209                                   &ftrace_event_id_fops);
2210 #endif
2211
2212         ret = event_define_fields(call);
2213         if (ret < 0) {
2214                 pr_warn("Could not initialize trace point events/%s\n", name);
2215                 return ret;
2216         }
2217
2218         /*
2219          * Only event directories that can be enabled should have
2220          * triggers or filters.
2221          */
2222         if (!(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)) {
2223                 trace_create_file("filter", 0644, file->dir, file,
2224                                   &ftrace_event_filter_fops);
2225
2226                 trace_create_file("trigger", 0644, file->dir, file,
2227                                   &event_trigger_fops);
2228         }
2229
2230 #ifdef CONFIG_HIST_TRIGGERS
2231         trace_create_file("hist", 0444, file->dir, file,
2232                           &event_hist_fops);
2233 #endif
2234 #ifdef CONFIG_HIST_TRIGGERS_DEBUG
2235         trace_create_file("hist_debug", 0444, file->dir, file,
2236                           &event_hist_debug_fops);
2237 #endif
2238         trace_create_file("format", 0444, file->dir, call,
2239                           &ftrace_event_format_fops);
2240
2241 #ifdef CONFIG_TRACE_EVENT_INJECT
2242         if (call->event.type && call->class->reg)
2243                 trace_create_file("inject", 0200, file->dir, file,
2244                                   &event_inject_fops);
2245 #endif
2246
2247         return 0;
2248 }
2249
2250 static void remove_event_from_tracers(struct trace_event_call *call)
2251 {
2252         struct trace_event_file *file;
2253         struct trace_array *tr;
2254
2255         do_for_each_event_file_safe(tr, file) {
2256                 if (file->event_call != call)
2257                         continue;
2258
2259                 remove_event_file_dir(file);
2260                 /*
2261                  * The do_for_each_event_file_safe() is
2262                  * a double loop. After finding the call for this
2263                  * trace_array, we use break to jump to the next
2264                  * trace_array.
2265                  */
2266                 break;
2267         } while_for_each_event_file();
2268 }
2269
2270 static void event_remove(struct trace_event_call *call)
2271 {
2272         struct trace_array *tr;
2273         struct trace_event_file *file;
2274
2275         do_for_each_event_file(tr, file) {
2276                 if (file->event_call != call)
2277                         continue;
2278
2279                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2280                         tr->clear_trace = true;
2281
2282                 ftrace_event_enable_disable(file, 0);
2283                 /*
2284                  * The do_for_each_event_file() is
2285                  * a double loop. After finding the call for this
2286                  * trace_array, we use break to jump to the next
2287                  * trace_array.
2288                  */
2289                 break;
2290         } while_for_each_event_file();
2291
2292         if (call->event.funcs)
2293                 __unregister_trace_event(&call->event);
2294         remove_event_from_tracers(call);
2295         list_del(&call->list);
2296 }
2297
2298 static int event_init(struct trace_event_call *call)
2299 {
2300         int ret = 0;
2301         const char *name;
2302
2303         name = trace_event_name(call);
2304         if (WARN_ON(!name))
2305                 return -EINVAL;
2306
2307         if (call->class->raw_init) {
2308                 ret = call->class->raw_init(call);
2309                 if (ret < 0 && ret != -ENOSYS)
2310                         pr_warn("Could not initialize trace events/%s\n", name);
2311         }
2312
2313         return ret;
2314 }
2315
2316 static int
2317 __register_event(struct trace_event_call *call, struct module *mod)
2318 {
2319         int ret;
2320
2321         ret = event_init(call);
2322         if (ret < 0)
2323                 return ret;
2324
2325         list_add(&call->list, &ftrace_events);
2326         call->mod = mod;
2327
2328         return 0;
2329 }
2330
2331 static char *eval_replace(char *ptr, struct trace_eval_map *map, int len)
2332 {
2333         int rlen;
2334         int elen;
2335
2336         /* Find the length of the eval value as a string */
2337         elen = snprintf(ptr, 0, "%ld", map->eval_value);
2338         /* Make sure there's enough room to replace the string with the value */
2339         if (len < elen)
2340                 return NULL;
2341
2342         snprintf(ptr, elen + 1, "%ld", map->eval_value);
2343
2344         /* Get the rest of the string of ptr */
2345         rlen = strlen(ptr + len);
2346         memmove(ptr + elen, ptr + len, rlen);
2347         /* Make sure we end the new string */
2348         ptr[elen + rlen] = 0;
2349
2350         return ptr + elen;
2351 }
2352
2353 static void update_event_printk(struct trace_event_call *call,
2354                                 struct trace_eval_map *map)
2355 {
2356         char *ptr;
2357         int quote = 0;
2358         int len = strlen(map->eval_string);
2359
2360         for (ptr = call->print_fmt; *ptr; ptr++) {
2361                 if (*ptr == '\\') {
2362                         ptr++;
2363                         /* paranoid */
2364                         if (!*ptr)
2365                                 break;
2366                         continue;
2367                 }
2368                 if (*ptr == '"') {
2369                         quote ^= 1;
2370                         continue;
2371                 }
2372                 if (quote)
2373                         continue;
2374                 if (isdigit(*ptr)) {
2375                         /* skip numbers */
2376                         do {
2377                                 ptr++;
2378                                 /* Check for alpha chars like ULL */
2379                         } while (isalnum(*ptr));
2380                         if (!*ptr)
2381                                 break;
2382                         /*
2383                          * A number must have some kind of delimiter after
2384                          * it, and we can ignore that too.
2385                          */
2386                         continue;
2387                 }
2388                 if (isalpha(*ptr) || *ptr == '_') {
2389                         if (strncmp(map->eval_string, ptr, len) == 0 &&
2390                             !isalnum(ptr[len]) && ptr[len] != '_') {
2391                                 ptr = eval_replace(ptr, map, len);
2392                                 /* enum/sizeof string smaller than value */
2393                                 if (WARN_ON_ONCE(!ptr))
2394                                         return;
2395                                 /*
2396                                  * No need to decrement here, as eval_replace()
2397                                  * returns the pointer to the character passed
2398                                  * the eval, and two evals can not be placed
2399                                  * back to back without something in between.
2400                                  * We can skip that something in between.
2401                                  */
2402                                 continue;
2403                         }
2404                 skip_more:
2405                         do {
2406                                 ptr++;
2407                         } while (isalnum(*ptr) || *ptr == '_');
2408                         if (!*ptr)
2409                                 break;
2410                         /*
2411                          * If what comes after this variable is a '.' or
2412                          * '->' then we can continue to ignore that string.
2413                          */
2414                         if (*ptr == '.' || (ptr[0] == '-' && ptr[1] == '>')) {
2415                                 ptr += *ptr == '.' ? 1 : 2;
2416                                 if (!*ptr)
2417                                         break;
2418                                 goto skip_more;
2419                         }
2420                         /*
2421                          * Once again, we can skip the delimiter that came
2422                          * after the string.
2423                          */
2424                         continue;
2425                 }
2426         }
2427 }
2428
2429 void trace_event_eval_update(struct trace_eval_map **map, int len)
2430 {
2431         struct trace_event_call *call, *p;
2432         const char *last_system = NULL;
2433         bool first = false;
2434         int last_i;
2435         int i;
2436
2437         down_write(&trace_event_sem);
2438         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2439                 /* events are usually grouped together with systems */
2440                 if (!last_system || call->class->system != last_system) {
2441                         first = true;
2442                         last_i = 0;
2443                         last_system = call->class->system;
2444                 }
2445
2446                 /*
2447                  * Since calls are grouped by systems, the likelyhood that the
2448                  * next call in the iteration belongs to the same system as the
2449                  * previous call is high. As an optimization, we skip seaching
2450                  * for a map[] that matches the call's system if the last call
2451                  * was from the same system. That's what last_i is for. If the
2452                  * call has the same system as the previous call, then last_i
2453                  * will be the index of the first map[] that has a matching
2454                  * system.
2455                  */
2456                 for (i = last_i; i < len; i++) {
2457                         if (call->class->system == map[i]->system) {
2458                                 /* Save the first system if need be */
2459                                 if (first) {
2460                                         last_i = i;
2461                                         first = false;
2462                                 }
2463                                 update_event_printk(call, map[i]);
2464                         }
2465                 }
2466                 cond_resched();
2467         }
2468         up_write(&trace_event_sem);
2469 }
2470
2471 static struct trace_event_file *
2472 trace_create_new_event(struct trace_event_call *call,
2473                        struct trace_array *tr)
2474 {
2475         struct trace_pid_list *no_pid_list;
2476         struct trace_pid_list *pid_list;
2477         struct trace_event_file *file;
2478
2479         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
2480         if (!file)
2481                 return NULL;
2482
2483         pid_list = rcu_dereference_protected(tr->filtered_pids,
2484                                              lockdep_is_held(&event_mutex));
2485         no_pid_list = rcu_dereference_protected(tr->filtered_no_pids,
2486                                              lockdep_is_held(&event_mutex));
2487
2488         if (pid_list || no_pid_list)
2489                 file->flags |= EVENT_FILE_FL_PID_FILTER;
2490
2491         file->event_call = call;
2492         file->tr = tr;
2493         atomic_set(&file->sm_ref, 0);
2494         atomic_set(&file->tm_ref, 0);
2495         INIT_LIST_HEAD(&file->triggers);
2496         list_add(&file->list, &tr->events);
2497         event_file_get(file);
2498
2499         return file;
2500 }
2501
2502 /* Add an event to a trace directory */
2503 static int
2504 __trace_add_new_event(struct trace_event_call *call, 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         if (eventdir_initialized)
2513                 return event_create_dir(tr->event_dir, file);
2514         else
2515                 return event_define_fields(call);
2516 }
2517
2518 /*
2519  * Just create a decriptor for early init. A descriptor is required
2520  * for enabling events at boot. We want to enable events before
2521  * the filesystem is initialized.
2522  */
2523 static int
2524 __trace_early_add_new_event(struct trace_event_call *call,
2525                             struct trace_array *tr)
2526 {
2527         struct trace_event_file *file;
2528
2529         file = trace_create_new_event(call, tr);
2530         if (!file)
2531                 return -ENOMEM;
2532
2533         return event_define_fields(call);
2534 }
2535
2536 struct ftrace_module_file_ops;
2537 static void __add_event_to_tracers(struct trace_event_call *call);
2538
2539 /* Add an additional event_call dynamically */
2540 int trace_add_event_call(struct trace_event_call *call)
2541 {
2542         int ret;
2543         lockdep_assert_held(&event_mutex);
2544
2545         mutex_lock(&trace_types_lock);
2546
2547         ret = __register_event(call, NULL);
2548         if (ret >= 0)
2549                 __add_event_to_tracers(call);
2550
2551         mutex_unlock(&trace_types_lock);
2552         return ret;
2553 }
2554
2555 /*
2556  * Must be called under locking of trace_types_lock, event_mutex and
2557  * trace_event_sem.
2558  */
2559 static void __trace_remove_event_call(struct trace_event_call *call)
2560 {
2561         event_remove(call);
2562         trace_destroy_fields(call);
2563         free_event_filter(call->filter);
2564         call->filter = NULL;
2565 }
2566
2567 static int probe_remove_event_call(struct trace_event_call *call)
2568 {
2569         struct trace_array *tr;
2570         struct trace_event_file *file;
2571
2572 #ifdef CONFIG_PERF_EVENTS
2573         if (call->perf_refcount)
2574                 return -EBUSY;
2575 #endif
2576         do_for_each_event_file(tr, file) {
2577                 if (file->event_call != call)
2578                         continue;
2579                 /*
2580                  * We can't rely on ftrace_event_enable_disable(enable => 0)
2581                  * we are going to do, EVENT_FILE_FL_SOFT_MODE can suppress
2582                  * TRACE_REG_UNREGISTER.
2583                  */
2584                 if (file->flags & EVENT_FILE_FL_ENABLED)
2585                         goto busy;
2586
2587                 if (file->flags & EVENT_FILE_FL_WAS_ENABLED)
2588                         tr->clear_trace = true;
2589                 /*
2590                  * The do_for_each_event_file_safe() is
2591                  * a double loop. After finding the call for this
2592                  * trace_array, we use break to jump to the next
2593                  * trace_array.
2594                  */
2595                 break;
2596         } while_for_each_event_file();
2597
2598         __trace_remove_event_call(call);
2599
2600         return 0;
2601  busy:
2602         /* No need to clear the trace now */
2603         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2604                 tr->clear_trace = false;
2605         }
2606         return -EBUSY;
2607 }
2608
2609 /* Remove an event_call */
2610 int trace_remove_event_call(struct trace_event_call *call)
2611 {
2612         int ret;
2613
2614         lockdep_assert_held(&event_mutex);
2615
2616         mutex_lock(&trace_types_lock);
2617         down_write(&trace_event_sem);
2618         ret = probe_remove_event_call(call);
2619         up_write(&trace_event_sem);
2620         mutex_unlock(&trace_types_lock);
2621
2622         return ret;
2623 }
2624
2625 #define for_each_event(event, start, end)                       \
2626         for (event = start;                                     \
2627              (unsigned long)event < (unsigned long)end;         \
2628              event++)
2629
2630 #ifdef CONFIG_MODULES
2631
2632 static void trace_module_add_events(struct module *mod)
2633 {
2634         struct trace_event_call **call, **start, **end;
2635
2636         if (!mod->num_trace_events)
2637                 return;
2638
2639         /* Don't add infrastructure for mods without tracepoints */
2640         if (trace_module_has_bad_taint(mod)) {
2641                 pr_err("%s: module has bad taint, not creating trace events\n",
2642                        mod->name);
2643                 return;
2644         }
2645
2646         start = mod->trace_events;
2647         end = mod->trace_events + mod->num_trace_events;
2648
2649         for_each_event(call, start, end) {
2650                 __register_event(*call, mod);
2651                 __add_event_to_tracers(*call);
2652         }
2653 }
2654
2655 static void trace_module_remove_events(struct module *mod)
2656 {
2657         struct trace_event_call *call, *p;
2658
2659         down_write(&trace_event_sem);
2660         list_for_each_entry_safe(call, p, &ftrace_events, list) {
2661                 if (call->mod == mod)
2662                         __trace_remove_event_call(call);
2663         }
2664         up_write(&trace_event_sem);
2665
2666         /*
2667          * It is safest to reset the ring buffer if the module being unloaded
2668          * registered any events that were used. The only worry is if
2669          * a new module gets loaded, and takes on the same id as the events
2670          * of this module. When printing out the buffer, traced events left
2671          * over from this module may be passed to the new module events and
2672          * unexpected results may occur.
2673          */
2674         tracing_reset_all_online_cpus_unlocked();
2675 }
2676
2677 static int trace_module_notify(struct notifier_block *self,
2678                                unsigned long val, void *data)
2679 {
2680         struct module *mod = data;
2681
2682         mutex_lock(&event_mutex);
2683         mutex_lock(&trace_types_lock);
2684         switch (val) {
2685         case MODULE_STATE_COMING:
2686                 trace_module_add_events(mod);
2687                 break;
2688         case MODULE_STATE_GOING:
2689                 trace_module_remove_events(mod);
2690                 break;
2691         }
2692         mutex_unlock(&trace_types_lock);
2693         mutex_unlock(&event_mutex);
2694
2695         return NOTIFY_OK;
2696 }
2697
2698 static struct notifier_block trace_module_nb = {
2699         .notifier_call = trace_module_notify,
2700         .priority = 1, /* higher than trace.c module notify */
2701 };
2702 #endif /* CONFIG_MODULES */
2703
2704 /* Create a new event directory structure for a trace directory. */
2705 static void
2706 __trace_add_event_dirs(struct trace_array *tr)
2707 {
2708         struct trace_event_call *call;
2709         int ret;
2710
2711         list_for_each_entry(call, &ftrace_events, list) {
2712                 ret = __trace_add_new_event(call, tr);
2713                 if (ret < 0)
2714                         pr_warn("Could not create directory for event %s\n",
2715                                 trace_event_name(call));
2716         }
2717 }
2718
2719 /* Returns any file that matches the system and event */
2720 struct trace_event_file *
2721 __find_event_file(struct trace_array *tr, const char *system, const char *event)
2722 {
2723         struct trace_event_file *file;
2724         struct trace_event_call *call;
2725         const char *name;
2726
2727         list_for_each_entry(file, &tr->events, list) {
2728
2729                 call = file->event_call;
2730                 name = trace_event_name(call);
2731
2732                 if (!name || !call->class)
2733                         continue;
2734
2735                 if (strcmp(event, name) == 0 &&
2736                     strcmp(system, call->class->system) == 0)
2737                         return file;
2738         }
2739         return NULL;
2740 }
2741
2742 /* Returns valid trace event files that match system and event */
2743 struct trace_event_file *
2744 find_event_file(struct trace_array *tr, const char *system, const char *event)
2745 {
2746         struct trace_event_file *file;
2747
2748         file = __find_event_file(tr, system, event);
2749         if (!file || !file->event_call->class->reg ||
2750             file->event_call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2751                 return NULL;
2752
2753         return file;
2754 }
2755
2756 /**
2757  * trace_get_event_file - Find and return a trace event file
2758  * @instance: The name of the trace instance containing the event
2759  * @system: The name of the system containing the event
2760  * @event: The name of the event
2761  *
2762  * Return a trace event file given the trace instance name, trace
2763  * system, and trace event name.  If the instance name is NULL, it
2764  * refers to the top-level trace array.
2765  *
2766  * This function will look it up and return it if found, after calling
2767  * trace_array_get() to prevent the instance from going away, and
2768  * increment the event's module refcount to prevent it from being
2769  * removed.
2770  *
2771  * To release the file, call trace_put_event_file(), which will call
2772  * trace_array_put() and decrement the event's module refcount.
2773  *
2774  * Return: The trace event on success, ERR_PTR otherwise.
2775  */
2776 struct trace_event_file *trace_get_event_file(const char *instance,
2777                                               const char *system,
2778                                               const char *event)
2779 {
2780         struct trace_array *tr = top_trace_array();
2781         struct trace_event_file *file = NULL;
2782         int ret = -EINVAL;
2783
2784         if (instance) {
2785                 tr = trace_array_find_get(instance);
2786                 if (!tr)
2787                         return ERR_PTR(-ENOENT);
2788         } else {
2789                 ret = trace_array_get(tr);
2790                 if (ret)
2791                         return ERR_PTR(ret);
2792         }
2793
2794         mutex_lock(&event_mutex);
2795
2796         file = find_event_file(tr, system, event);
2797         if (!file) {
2798                 trace_array_put(tr);
2799                 ret = -EINVAL;
2800                 goto out;
2801         }
2802
2803         /* Don't let event modules unload while in use */
2804         ret = try_module_get(file->event_call->mod);
2805         if (!ret) {
2806                 trace_array_put(tr);
2807                 ret = -EBUSY;
2808                 goto out;
2809         }
2810
2811         ret = 0;
2812  out:
2813         mutex_unlock(&event_mutex);
2814
2815         if (ret)
2816                 file = ERR_PTR(ret);
2817
2818         return file;
2819 }
2820 EXPORT_SYMBOL_GPL(trace_get_event_file);
2821
2822 /**
2823  * trace_put_event_file - Release a file from trace_get_event_file()
2824  * @file: The trace event file
2825  *
2826  * If a file was retrieved using trace_get_event_file(), this should
2827  * be called when it's no longer needed.  It will cancel the previous
2828  * trace_array_get() called by that function, and decrement the
2829  * event's module refcount.
2830  */
2831 void trace_put_event_file(struct trace_event_file *file)
2832 {
2833         mutex_lock(&event_mutex);
2834         module_put(file->event_call->mod);
2835         mutex_unlock(&event_mutex);
2836
2837         trace_array_put(file->tr);
2838 }
2839 EXPORT_SYMBOL_GPL(trace_put_event_file);
2840
2841 #ifdef CONFIG_DYNAMIC_FTRACE
2842
2843 /* Avoid typos */
2844 #define ENABLE_EVENT_STR        "enable_event"
2845 #define DISABLE_EVENT_STR       "disable_event"
2846
2847 struct event_probe_data {
2848         struct trace_event_file *file;
2849         unsigned long                   count;
2850         int                             ref;
2851         bool                            enable;
2852 };
2853
2854 static void update_event_probe(struct event_probe_data *data)
2855 {
2856         if (data->enable)
2857                 clear_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2858         else
2859                 set_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &data->file->flags);
2860 }
2861
2862 static void
2863 event_enable_probe(unsigned long ip, unsigned long parent_ip,
2864                    struct trace_array *tr, struct ftrace_probe_ops *ops,
2865                    void *data)
2866 {
2867         struct ftrace_func_mapper *mapper = data;
2868         struct event_probe_data *edata;
2869         void **pdata;
2870
2871         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2872         if (!pdata || !*pdata)
2873                 return;
2874
2875         edata = *pdata;
2876         update_event_probe(edata);
2877 }
2878
2879 static void
2880 event_enable_count_probe(unsigned long ip, unsigned long parent_ip,
2881                          struct trace_array *tr, struct ftrace_probe_ops *ops,
2882                          void *data)
2883 {
2884         struct ftrace_func_mapper *mapper = data;
2885         struct event_probe_data *edata;
2886         void **pdata;
2887
2888         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2889         if (!pdata || !*pdata)
2890                 return;
2891
2892         edata = *pdata;
2893
2894         if (!edata->count)
2895                 return;
2896
2897         /* Skip if the event is in a state we want to switch to */
2898         if (edata->enable == !(edata->file->flags & EVENT_FILE_FL_SOFT_DISABLED))
2899                 return;
2900
2901         if (edata->count != -1)
2902                 (edata->count)--;
2903
2904         update_event_probe(edata);
2905 }
2906
2907 static int
2908 event_enable_print(struct seq_file *m, unsigned long ip,
2909                    struct ftrace_probe_ops *ops, void *data)
2910 {
2911         struct ftrace_func_mapper *mapper = data;
2912         struct event_probe_data *edata;
2913         void **pdata;
2914
2915         pdata = ftrace_func_mapper_find_ip(mapper, ip);
2916
2917         if (WARN_ON_ONCE(!pdata || !*pdata))
2918                 return 0;
2919
2920         edata = *pdata;
2921
2922         seq_printf(m, "%ps:", (void *)ip);
2923
2924         seq_printf(m, "%s:%s:%s",
2925                    edata->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2926                    edata->file->event_call->class->system,
2927                    trace_event_name(edata->file->event_call));
2928
2929         if (edata->count == -1)
2930                 seq_puts(m, ":unlimited\n");
2931         else
2932                 seq_printf(m, ":count=%ld\n", edata->count);
2933
2934         return 0;
2935 }
2936
2937 static int
2938 event_enable_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
2939                   unsigned long ip, void *init_data, void **data)
2940 {
2941         struct ftrace_func_mapper *mapper = *data;
2942         struct event_probe_data *edata = init_data;
2943         int ret;
2944
2945         if (!mapper) {
2946                 mapper = allocate_ftrace_func_mapper();
2947                 if (!mapper)
2948                         return -ENODEV;
2949                 *data = mapper;
2950         }
2951
2952         ret = ftrace_func_mapper_add_ip(mapper, ip, edata);
2953         if (ret < 0)
2954                 return ret;
2955
2956         edata->ref++;
2957
2958         return 0;
2959 }
2960
2961 static int free_probe_data(void *data)
2962 {
2963         struct event_probe_data *edata = data;
2964
2965         edata->ref--;
2966         if (!edata->ref) {
2967                 /* Remove the SOFT_MODE flag */
2968                 __ftrace_event_enable_disable(edata->file, 0, 1);
2969                 module_put(edata->file->event_call->mod);
2970                 kfree(edata);
2971         }
2972         return 0;
2973 }
2974
2975 static void
2976 event_enable_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
2977                   unsigned long ip, void *data)
2978 {
2979         struct ftrace_func_mapper *mapper = data;
2980         struct event_probe_data *edata;
2981
2982         if (!ip) {
2983                 if (!mapper)
2984                         return;
2985                 free_ftrace_func_mapper(mapper, free_probe_data);
2986                 return;
2987         }
2988
2989         edata = ftrace_func_mapper_remove_ip(mapper, ip);
2990
2991         if (WARN_ON_ONCE(!edata))
2992                 return;
2993
2994         if (WARN_ON_ONCE(edata->ref <= 0))
2995                 return;
2996
2997         free_probe_data(edata);
2998 }
2999
3000 static struct ftrace_probe_ops event_enable_probe_ops = {
3001         .func                   = event_enable_probe,
3002         .print                  = event_enable_print,
3003         .init                   = event_enable_init,
3004         .free                   = event_enable_free,
3005 };
3006
3007 static struct ftrace_probe_ops event_enable_count_probe_ops = {
3008         .func                   = event_enable_count_probe,
3009         .print                  = event_enable_print,
3010         .init                   = event_enable_init,
3011         .free                   = event_enable_free,
3012 };
3013
3014 static struct ftrace_probe_ops event_disable_probe_ops = {
3015         .func                   = event_enable_probe,
3016         .print                  = event_enable_print,
3017         .init                   = event_enable_init,
3018         .free                   = event_enable_free,
3019 };
3020
3021 static struct ftrace_probe_ops event_disable_count_probe_ops = {
3022         .func                   = event_enable_count_probe,
3023         .print                  = event_enable_print,
3024         .init                   = event_enable_init,
3025         .free                   = event_enable_free,
3026 };
3027
3028 static int
3029 event_enable_func(struct trace_array *tr, struct ftrace_hash *hash,
3030                   char *glob, char *cmd, char *param, int enabled)
3031 {
3032         struct trace_event_file *file;
3033         struct ftrace_probe_ops *ops;
3034         struct event_probe_data *data;
3035         const char *system;
3036         const char *event;
3037         char *number;
3038         bool enable;
3039         int ret;
3040
3041         if (!tr)
3042                 return -ENODEV;
3043
3044         /* hash funcs only work with set_ftrace_filter */
3045         if (!enabled || !param)
3046                 return -EINVAL;
3047
3048         system = strsep(&param, ":");
3049         if (!param)
3050                 return -EINVAL;
3051
3052         event = strsep(&param, ":");
3053
3054         mutex_lock(&event_mutex);
3055
3056         ret = -EINVAL;
3057         file = find_event_file(tr, system, event);
3058         if (!file)
3059                 goto out;
3060
3061         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
3062
3063         if (enable)
3064                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
3065         else
3066                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
3067
3068         if (glob[0] == '!') {
3069                 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
3070                 goto out;
3071         }
3072
3073         ret = -ENOMEM;
3074
3075         data = kzalloc(sizeof(*data), GFP_KERNEL);
3076         if (!data)
3077                 goto out;
3078
3079         data->enable = enable;
3080         data->count = -1;
3081         data->file = file;
3082
3083         if (!param)
3084                 goto out_reg;
3085
3086         number = strsep(&param, ":");
3087
3088         ret = -EINVAL;
3089         if (!strlen(number))
3090                 goto out_free;
3091
3092         /*
3093          * We use the callback data field (which is a pointer)
3094          * as our counter.
3095          */
3096         ret = kstrtoul(number, 0, &data->count);
3097         if (ret)
3098                 goto out_free;
3099
3100  out_reg:
3101         /* Don't let event modules unload while probe registered */
3102         ret = try_module_get(file->event_call->mod);
3103         if (!ret) {
3104                 ret = -EBUSY;
3105                 goto out_free;
3106         }
3107
3108         ret = __ftrace_event_enable_disable(file, 1, 1);
3109         if (ret < 0)
3110                 goto out_put;
3111
3112         ret = register_ftrace_function_probe(glob, tr, ops, data);
3113         /*
3114          * The above returns on success the # of functions enabled,
3115          * but if it didn't find any functions it returns zero.
3116          * Consider no functions a failure too.
3117          */
3118         if (!ret) {
3119                 ret = -ENOENT;
3120                 goto out_disable;
3121         } else if (ret < 0)
3122                 goto out_disable;
3123         /* Just return zero, not the number of enabled functions */
3124         ret = 0;
3125  out:
3126         mutex_unlock(&event_mutex);
3127         return ret;
3128
3129  out_disable:
3130         __ftrace_event_enable_disable(file, 0, 1);
3131  out_put:
3132         module_put(file->event_call->mod);
3133  out_free:
3134         kfree(data);
3135         goto out;
3136 }
3137
3138 static struct ftrace_func_command event_enable_cmd = {
3139         .name                   = ENABLE_EVENT_STR,
3140         .func                   = event_enable_func,
3141 };
3142
3143 static struct ftrace_func_command event_disable_cmd = {
3144         .name                   = DISABLE_EVENT_STR,
3145         .func                   = event_enable_func,
3146 };
3147
3148 static __init int register_event_cmds(void)
3149 {
3150         int ret;
3151
3152         ret = register_ftrace_command(&event_enable_cmd);
3153         if (WARN_ON(ret < 0))
3154                 return ret;
3155         ret = register_ftrace_command(&event_disable_cmd);
3156         if (WARN_ON(ret < 0))
3157                 unregister_ftrace_command(&event_enable_cmd);
3158         return ret;
3159 }
3160 #else
3161 static inline int register_event_cmds(void) { return 0; }
3162 #endif /* CONFIG_DYNAMIC_FTRACE */
3163
3164 /*
3165  * The top level array and trace arrays created by boot-time tracing
3166  * have already had its trace_event_file descriptors created in order
3167  * to allow for early events to be recorded.
3168  * This function is called after the tracefs has been initialized,
3169  * and we now have to create the files associated to the events.
3170  */
3171 static void __trace_early_add_event_dirs(struct trace_array *tr)
3172 {
3173         struct trace_event_file *file;
3174         int ret;
3175
3176
3177         list_for_each_entry(file, &tr->events, list) {
3178                 ret = event_create_dir(tr->event_dir, file);
3179                 if (ret < 0)
3180                         pr_warn("Could not create directory for event %s\n",
3181                                 trace_event_name(file->event_call));
3182         }
3183 }
3184
3185 /*
3186  * For early boot up, the top trace array and the trace arrays created
3187  * by boot-time tracing require to have a list of events that can be
3188  * enabled. This must be done before the filesystem is set up in order
3189  * to allow events to be traced early.
3190  */
3191 void __trace_early_add_events(struct trace_array *tr)
3192 {
3193         struct trace_event_call *call;
3194         int ret;
3195
3196         list_for_each_entry(call, &ftrace_events, list) {
3197                 /* Early boot up should not have any modules loaded */
3198                 if (WARN_ON_ONCE(call->mod))
3199                         continue;
3200
3201                 ret = __trace_early_add_new_event(call, tr);
3202                 if (ret < 0)
3203                         pr_warn("Could not create early event %s\n",
3204                                 trace_event_name(call));
3205         }
3206 }
3207
3208 /* Remove the event directory structure for a trace directory. */
3209 static void
3210 __trace_remove_event_dirs(struct trace_array *tr)
3211 {
3212         struct trace_event_file *file, *next;
3213
3214         list_for_each_entry_safe(file, next, &tr->events, list)
3215                 remove_event_file_dir(file);
3216 }
3217
3218 static void __add_event_to_tracers(struct trace_event_call *call)
3219 {
3220         struct trace_array *tr;
3221
3222         list_for_each_entry(tr, &ftrace_trace_arrays, list)
3223                 __trace_add_new_event(call, tr);
3224 }
3225
3226 extern struct trace_event_call *__start_ftrace_events[];
3227 extern struct trace_event_call *__stop_ftrace_events[];
3228
3229 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
3230
3231 static __init int setup_trace_event(char *str)
3232 {
3233         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
3234         ring_buffer_expanded = true;
3235         disable_tracing_selftest("running event tracing");
3236
3237         return 1;
3238 }
3239 __setup("trace_event=", setup_trace_event);
3240
3241 /* Expects to have event_mutex held when called */
3242 static int
3243 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
3244 {
3245         struct dentry *d_events;
3246         struct dentry *entry;
3247
3248         entry = tracefs_create_file("set_event", 0644, parent,
3249                                     tr, &ftrace_set_event_fops);
3250         if (!entry) {
3251                 pr_warn("Could not create tracefs 'set_event' entry\n");
3252                 return -ENOMEM;
3253         }
3254
3255         d_events = tracefs_create_dir("events", parent);
3256         if (!d_events) {
3257                 pr_warn("Could not create tracefs 'events' directory\n");
3258                 return -ENOMEM;
3259         }
3260
3261         entry = trace_create_file("enable", 0644, d_events,
3262                                   tr, &ftrace_tr_enable_fops);
3263         if (!entry) {
3264                 pr_warn("Could not create tracefs 'enable' entry\n");
3265                 return -ENOMEM;
3266         }
3267
3268         /* There are not as crucial, just warn if they are not created */
3269
3270         entry = tracefs_create_file("set_event_pid", 0644, parent,
3271                                     tr, &ftrace_set_event_pid_fops);
3272         if (!entry)
3273                 pr_warn("Could not create tracefs 'set_event_pid' entry\n");
3274
3275         entry = tracefs_create_file("set_event_notrace_pid", 0644, parent,
3276                                     tr, &ftrace_set_event_notrace_pid_fops);
3277         if (!entry)
3278                 pr_warn("Could not create tracefs 'set_event_notrace_pid' entry\n");
3279
3280         /* ring buffer internal formats */
3281         entry = trace_create_file("header_page", 0444, d_events,
3282                                   ring_buffer_print_page_header,
3283                                   &ftrace_show_header_fops);
3284         if (!entry)
3285                 pr_warn("Could not create tracefs 'header_page' entry\n");
3286
3287         entry = trace_create_file("header_event", 0444, d_events,
3288                                   ring_buffer_print_entry_header,
3289                                   &ftrace_show_header_fops);
3290         if (!entry)
3291                 pr_warn("Could not create tracefs 'header_event' entry\n");
3292
3293         tr->event_dir = d_events;
3294
3295         return 0;
3296 }
3297
3298 /**
3299  * event_trace_add_tracer - add a instance of a trace_array to events
3300  * @parent: The parent dentry to place the files/directories for events in
3301  * @tr: The trace array associated with these events
3302  *
3303  * When a new instance is created, it needs to set up its events
3304  * directory, as well as other files associated with events. It also
3305  * creates the event hierachry in the @parent/events directory.
3306  *
3307  * Returns 0 on success.
3308  *
3309  * Must be called with event_mutex held.
3310  */
3311 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
3312 {
3313         int ret;
3314
3315         lockdep_assert_held(&event_mutex);
3316
3317         ret = create_event_toplevel_files(parent, tr);
3318         if (ret)
3319                 goto out;
3320
3321         down_write(&trace_event_sem);
3322         /* If tr already has the event list, it is initialized in early boot. */
3323         if (unlikely(!list_empty(&tr->events)))
3324                 __trace_early_add_event_dirs(tr);
3325         else
3326                 __trace_add_event_dirs(tr);
3327         up_write(&trace_event_sem);
3328
3329  out:
3330         return ret;
3331 }
3332
3333 /*
3334  * The top trace array already had its file descriptors created.
3335  * Now the files themselves need to be created.
3336  */
3337 static __init int
3338 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
3339 {
3340         int ret;
3341
3342         mutex_lock(&event_mutex);
3343
3344         ret = create_event_toplevel_files(parent, tr);
3345         if (ret)
3346                 goto out_unlock;
3347
3348         down_write(&trace_event_sem);
3349         __trace_early_add_event_dirs(tr);
3350         up_write(&trace_event_sem);
3351
3352  out_unlock:
3353         mutex_unlock(&event_mutex);
3354
3355         return ret;
3356 }
3357
3358 /* Must be called with event_mutex held */
3359 int event_trace_del_tracer(struct trace_array *tr)
3360 {
3361         lockdep_assert_held(&event_mutex);
3362
3363         /* Disable any event triggers and associated soft-disabled events */
3364         clear_event_triggers(tr);
3365
3366         /* Clear the pid list */
3367         __ftrace_clear_event_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
3368
3369         /* Disable any running events */
3370         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
3371
3372         /* Make sure no more events are being executed */
3373         tracepoint_synchronize_unregister();
3374
3375         down_write(&trace_event_sem);
3376         __trace_remove_event_dirs(tr);
3377         tracefs_remove(tr->event_dir);
3378         up_write(&trace_event_sem);
3379
3380         tr->event_dir = NULL;
3381
3382         return 0;
3383 }
3384
3385 static __init int event_trace_memsetup(void)
3386 {
3387         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
3388         file_cachep = KMEM_CACHE(trace_event_file, SLAB_PANIC);
3389         return 0;
3390 }
3391
3392 static __init void
3393 early_enable_events(struct trace_array *tr, bool disable_first)
3394 {
3395         char *buf = bootup_event_buf;
3396         char *token;
3397         int ret;
3398
3399         while (true) {
3400                 token = strsep(&buf, ",");
3401
3402                 if (!token)
3403                         break;
3404
3405                 if (*token) {
3406                         /* Restarting syscalls requires that we stop them first */
3407                         if (disable_first)
3408                                 ftrace_set_clr_event(tr, token, 0);
3409
3410                         ret = ftrace_set_clr_event(tr, token, 1);
3411                         if (ret)
3412                                 pr_warn("Failed to enable trace event: %s\n", token);
3413                 }
3414
3415                 /* Put back the comma to allow this to be called again */
3416                 if (buf)
3417                         *(buf - 1) = ',';
3418         }
3419 }
3420
3421 static __init int event_trace_enable(void)
3422 {
3423         struct trace_array *tr = top_trace_array();
3424         struct trace_event_call **iter, *call;
3425         int ret;
3426
3427         if (!tr)
3428                 return -ENODEV;
3429
3430         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
3431
3432                 call = *iter;
3433                 ret = event_init(call);
3434                 if (!ret)
3435                         list_add(&call->list, &ftrace_events);
3436         }
3437
3438         /*
3439          * We need the top trace array to have a working set of trace
3440          * points at early init, before the debug files and directories
3441          * are created. Create the file entries now, and attach them
3442          * to the actual file dentries later.
3443          */
3444         __trace_early_add_events(tr);
3445
3446         early_enable_events(tr, false);
3447
3448         trace_printk_start_comm();
3449
3450         register_event_cmds();
3451
3452         register_trigger_cmds();
3453
3454         return 0;
3455 }
3456
3457 /*
3458  * event_trace_enable() is called from trace_event_init() first to
3459  * initialize events and perhaps start any events that are on the
3460  * command line. Unfortunately, there are some events that will not
3461  * start this early, like the system call tracepoints that need
3462  * to set the TIF_SYSCALL_TRACEPOINT flag of pid 1. But event_trace_enable()
3463  * is called before pid 1 starts, and this flag is never set, making
3464  * the syscall tracepoint never get reached, but the event is enabled
3465  * regardless (and not doing anything).
3466  */
3467 static __init int event_trace_enable_again(void)
3468 {
3469         struct trace_array *tr;
3470
3471         tr = top_trace_array();
3472         if (!tr)
3473                 return -ENODEV;
3474
3475         early_enable_events(tr, true);
3476
3477         return 0;
3478 }
3479
3480 early_initcall(event_trace_enable_again);
3481
3482 /* Init fields which doesn't related to the tracefs */
3483 static __init int event_trace_init_fields(void)
3484 {
3485         if (trace_define_generic_fields())
3486                 pr_warn("tracing: Failed to allocated generic fields");
3487
3488         if (trace_define_common_fields())
3489                 pr_warn("tracing: Failed to allocate common fields");
3490
3491         return 0;
3492 }
3493
3494 __init int event_trace_init(void)
3495 {
3496         struct trace_array *tr;
3497         struct dentry *entry;
3498         int ret;
3499
3500         tr = top_trace_array();
3501         if (!tr)
3502                 return -ENODEV;
3503
3504         entry = tracefs_create_file("available_events", 0444, NULL,
3505                                     tr, &ftrace_avail_fops);
3506         if (!entry)
3507                 pr_warn("Could not create tracefs 'available_events' entry\n");
3508
3509         ret = early_event_add_tracer(NULL, tr);
3510         if (ret)
3511                 return ret;
3512
3513 #ifdef CONFIG_MODULES
3514         ret = register_module_notifier(&trace_module_nb);
3515         if (ret)
3516                 pr_warn("Failed to register trace events module notifier\n");
3517 #endif
3518
3519         eventdir_initialized = true;
3520
3521         return 0;
3522 }
3523
3524 void __init trace_event_init(void)
3525 {
3526         event_trace_memsetup();
3527         init_ftrace_syscalls();
3528         event_trace_enable();
3529         event_trace_init_fields();
3530 }
3531
3532 #ifdef CONFIG_EVENT_TRACE_STARTUP_TEST
3533
3534 static DEFINE_SPINLOCK(test_spinlock);
3535 static DEFINE_SPINLOCK(test_spinlock_irq);
3536 static DEFINE_MUTEX(test_mutex);
3537
3538 static __init void test_work(struct work_struct *dummy)
3539 {
3540         spin_lock(&test_spinlock);
3541         spin_lock_irq(&test_spinlock_irq);
3542         udelay(1);
3543         spin_unlock_irq(&test_spinlock_irq);
3544         spin_unlock(&test_spinlock);
3545
3546         mutex_lock(&test_mutex);
3547         msleep(1);
3548         mutex_unlock(&test_mutex);
3549 }
3550
3551 static __init int event_test_thread(void *unused)
3552 {
3553         void *test_malloc;
3554
3555         test_malloc = kmalloc(1234, GFP_KERNEL);
3556         if (!test_malloc)
3557                 pr_info("failed to kmalloc\n");
3558
3559         schedule_on_each_cpu(test_work);
3560
3561         kfree(test_malloc);
3562
3563         set_current_state(TASK_INTERRUPTIBLE);
3564         while (!kthread_should_stop()) {
3565                 schedule();
3566                 set_current_state(TASK_INTERRUPTIBLE);
3567         }
3568         __set_current_state(TASK_RUNNING);
3569
3570         return 0;
3571 }
3572
3573 /*
3574  * Do various things that may trigger events.
3575  */
3576 static __init void event_test_stuff(void)
3577 {
3578         struct task_struct *test_thread;
3579
3580         test_thread = kthread_run(event_test_thread, NULL, "test-events");
3581         msleep(1);
3582         kthread_stop(test_thread);
3583 }
3584
3585 /*
3586  * For every trace event defined, we will test each trace point separately,
3587  * and then by groups, and finally all trace points.
3588  */
3589 static __init void event_trace_self_tests(void)
3590 {
3591         struct trace_subsystem_dir *dir;
3592         struct trace_event_file *file;
3593         struct trace_event_call *call;
3594         struct event_subsystem *system;
3595         struct trace_array *tr;
3596         int ret;
3597
3598         tr = top_trace_array();
3599         if (!tr)
3600                 return;
3601
3602         pr_info("Running tests on trace events:\n");
3603
3604         list_for_each_entry(file, &tr->events, list) {
3605
3606                 call = file->event_call;
3607
3608                 /* Only test those that have a probe */
3609                 if (!call->class || !call->class->probe)
3610                         continue;
3611
3612 /*
3613  * Testing syscall events here is pretty useless, but
3614  * we still do it if configured. But this is time consuming.
3615  * What we really need is a user thread to perform the
3616  * syscalls as we test.
3617  */
3618 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
3619                 if (call->class->system &&
3620                     strcmp(call->class->system, "syscalls") == 0)
3621                         continue;
3622 #endif
3623
3624                 pr_info("Testing event %s: ", trace_event_name(call));
3625
3626                 /*
3627                  * If an event is already enabled, someone is using
3628                  * it and the self test should not be on.
3629                  */
3630                 if (file->flags & EVENT_FILE_FL_ENABLED) {
3631                         pr_warn("Enabled event during self test!\n");
3632                         WARN_ON_ONCE(1);
3633                         continue;
3634                 }
3635
3636                 ftrace_event_enable_disable(file, 1);
3637                 event_test_stuff();
3638                 ftrace_event_enable_disable(file, 0);
3639
3640                 pr_cont("OK\n");
3641         }
3642
3643         /* Now test at the sub system level */
3644
3645         pr_info("Running tests on trace event systems:\n");
3646
3647         list_for_each_entry(dir, &tr->systems, list) {
3648
3649                 system = dir->subsystem;
3650
3651                 /* the ftrace system is special, skip it */
3652                 if (strcmp(system->name, "ftrace") == 0)
3653                         continue;
3654
3655                 pr_info("Testing event system %s: ", system->name);
3656
3657                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
3658                 if (WARN_ON_ONCE(ret)) {
3659                         pr_warn("error enabling system %s\n",
3660                                 system->name);
3661                         continue;
3662                 }
3663
3664                 event_test_stuff();
3665
3666                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
3667                 if (WARN_ON_ONCE(ret)) {
3668                         pr_warn("error disabling system %s\n",
3669                                 system->name);
3670                         continue;
3671                 }
3672
3673                 pr_cont("OK\n");
3674         }
3675
3676         /* Test with all events enabled */
3677
3678         pr_info("Running tests on all trace events:\n");
3679         pr_info("Testing all events: ");
3680
3681         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
3682         if (WARN_ON_ONCE(ret)) {
3683                 pr_warn("error enabling all events\n");
3684                 return;
3685         }
3686
3687         event_test_stuff();
3688
3689         /* reset sysname */
3690         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
3691         if (WARN_ON_ONCE(ret)) {
3692                 pr_warn("error disabling all events\n");
3693                 return;
3694         }
3695
3696         pr_cont("OK\n");
3697 }
3698
3699 #ifdef CONFIG_FUNCTION_TRACER
3700
3701 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
3702
3703 static struct trace_event_file event_trace_file __initdata;
3704
3705 static void __init
3706 function_test_events_call(unsigned long ip, unsigned long parent_ip,
3707                           struct ftrace_ops *op, struct pt_regs *pt_regs)
3708 {
3709         struct trace_buffer *buffer;
3710         struct ring_buffer_event *event;
3711         struct ftrace_entry *entry;
3712         unsigned long flags;
3713         long disabled;
3714         int cpu;
3715         int pc;
3716
3717         pc = preempt_count();
3718         preempt_disable_notrace();
3719         cpu = raw_smp_processor_id();
3720         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
3721
3722         if (disabled != 1)
3723                 goto out;
3724
3725         local_save_flags(flags);
3726
3727         event = trace_event_buffer_lock_reserve(&buffer, &event_trace_file,
3728                                                 TRACE_FN, sizeof(*entry),
3729                                                 flags, pc);
3730         if (!event)
3731                 goto out;
3732         entry   = ring_buffer_event_data(event);
3733         entry->ip                       = ip;
3734         entry->parent_ip                = parent_ip;
3735
3736         event_trigger_unlock_commit(&event_trace_file, buffer, event,
3737                                     entry, flags, pc);
3738  out:
3739         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
3740         preempt_enable_notrace();
3741 }
3742
3743 static struct ftrace_ops trace_ops __initdata  =
3744 {
3745         .func = function_test_events_call,
3746         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3747 };
3748
3749 static __init void event_trace_self_test_with_function(void)
3750 {
3751         int ret;
3752
3753         event_trace_file.tr = top_trace_array();
3754         if (WARN_ON(!event_trace_file.tr))
3755                 return;
3756
3757         ret = register_ftrace_function(&trace_ops);
3758         if (WARN_ON(ret < 0)) {
3759                 pr_info("Failed to enable function tracer for event tests\n");
3760                 return;
3761         }
3762         pr_info("Running tests again, along with the function tracer\n");
3763         event_trace_self_tests();
3764         unregister_ftrace_function(&trace_ops);
3765 }
3766 #else
3767 static __init void event_trace_self_test_with_function(void)
3768 {
3769 }
3770 #endif
3771
3772 static __init int event_trace_self_tests_init(void)
3773 {
3774         if (!tracing_selftest_disabled) {
3775                 event_trace_self_tests();
3776                 event_trace_self_test_with_function();
3777         }
3778
3779         return 0;
3780 }
3781
3782 late_initcall(event_trace_self_tests_init);
3783
3784 #endif