kernel/trace/trace_events.c

   1 /*
   2  * event tracer
   3  *
   4  * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
   5  *
   6  *  - Added format output of fields of the trace point.
   7  *    This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
   8  *
   9  */
  10
  11 #include <linux/workqueue.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/kthread.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/module.h>
  17 #include <linux/ctype.h>
  18 #include <linux/slab.h>
  19 #include <linux/delay.h>
  20
  21 #include <asm/setup.h>
  22
  23 #include "trace_output.h"
  24
  25 #undef TRACE_SYSTEM
  26 #define TRACE_SYSTEM "TRACE_SYSTEM"
  27
  28 DEFINE_MUTEX(event_mutex);
  29
  30 DEFINE_MUTEX(event_storage_mutex);
  31 EXPORT_SYMBOL_GPL(event_storage_mutex);
  32
  33 char event_storage[EVENT_STORAGE_SIZE];
  34 EXPORT_SYMBOL_GPL(event_storage);
  35
  36 LIST_HEAD(ftrace_events);
  37 static LIST_HEAD(ftrace_common_fields);
  38
  39 #define GFP_TRACE (GFP_KERNEL | __GFP_ZERO)
  40
  41 static struct kmem_cache *field_cachep;
  42 static struct kmem_cache *file_cachep;
  43
  44 /* Double loops, do not use break, only goto's work */
  45 #define do_for_each_event_file(tr, file)                        \
  46         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  47                 list_for_each_entry(file, &tr->events, list)
  48
  49 #define do_for_each_event_file_safe(tr, file)                   \
  50         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  51                 struct ftrace_event_file *___n;                         \
  52                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  53
  54 #define while_for_each_event_file()             \
  55         }
  56
  57 static struct list_head *
  58 trace_get_fields(struct ftrace_event_call *event_call)
  59 {
  60         if (!event_call->class->get_fields)
  61                 return &event_call->class->fields;
  62         return event_call->class->get_fields(event_call);
  63 }
  64
  65 static struct ftrace_event_field *
  66 __find_event_field(struct list_head *head, char *name)
  67 {
  68         struct ftrace_event_field *field;
  69
  70         list_for_each_entry(field, head, link) {
  71                 if (!strcmp(field->name, name))
  72                         return field;
  73         }
  74
  75         return NULL;
  76 }
  77
  78 struct ftrace_event_field *
  79 trace_find_event_field(struct ftrace_event_call *call, char *name)
  80 {
  81         struct ftrace_event_field *field;
  82         struct list_head *head;
  83
  84         field = __find_event_field(&ftrace_common_fields, name);
  85         if (field)
  86                 return field;
  87
  88         head = trace_get_fields(call);
  89         return __find_event_field(head, name);
  90 }
  91
  92 static int __trace_define_field(struct list_head *head, const char *type,
  93                                 const char *name, int offset, int size,
  94                                 int is_signed, int filter_type)
  95 {
  96         struct ftrace_event_field *field;
  97
  98         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
  99         if (!field)
 100                 goto err;
 101
 102         field->name = name;
 103         field->type = type;
 104
 105         if (filter_type == FILTER_OTHER)
 106                 field->filter_type = filter_assign_type(type);
 107         else
 108                 field->filter_type = filter_type;
 109
 110         field->offset = offset;
 111         field->size = size;
 112         field->is_signed = is_signed;
 113
 114         list_add(&field->link, head);
 115
 116         return 0;
 117
 118 err:
 119         kmem_cache_free(field_cachep, field);
 120
 121         return -ENOMEM;
 122 }
 123
 124 int trace_define_field(struct ftrace_event_call *call, const char *type,
 125                        const char *name, int offset, int size, int is_signed,
 126                        int filter_type)
 127 {
 128         struct list_head *head;
 129
 130         if (WARN_ON(!call->class))
 131                 return 0;
 132
 133         head = trace_get_fields(call);
 134         return __trace_define_field(head, type, name, offset, size,
 135                                     is_signed, filter_type);
 136 }
 137 EXPORT_SYMBOL_GPL(trace_define_field);
 138
 139 #define __common_field(type, item)                                      \
 140         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 141                                    "common_" #item,                     \
 142                                    offsetof(typeof(ent), item),         \
 143                                    sizeof(ent.item),                    \
 144                                    is_signed_type(type), FILTER_OTHER); \
 145         if (ret)                                                        \
 146                 return ret;
 147
 148 static int trace_define_common_fields(void)
 149 {
 150         int ret;
 151         struct trace_entry ent;
 152
 153         __common_field(unsigned short, type);
 154         __common_field(unsigned char, flags);
 155         __common_field(unsigned char, preempt_count);
 156         __common_field(int, pid);
 157
 158         return ret;
 159 }
 160
 161 static void trace_destroy_fields(struct ftrace_event_call *call)
 162 {
 163         struct ftrace_event_field *field, *next;
 164         struct list_head *head;
 165
 166         head = trace_get_fields(call);
 167         list_for_each_entry_safe(field, next, head, link) {
 168                 list_del(&field->link);
 169                 kmem_cache_free(field_cachep, field);
 170         }
 171 }
 172
 173 int trace_event_raw_init(struct ftrace_event_call *call)
 174 {
 175         int id;
 176
 177         id = register_ftrace_event(&call->event);
 178         if (!id)
 179                 return -ENODEV;
 180
 181         return 0;
 182 }
 183 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 184
 185 int ftrace_event_reg(struct ftrace_event_call *call,
 186                      enum trace_reg type, void *data)
 187 {
 188         struct ftrace_event_file *file = data;
 189
 190         switch (type) {
 191         case TRACE_REG_REGISTER:
 192                 return tracepoint_probe_register(call->name,
 193                                                  call->class->probe,
 194                                                  file);
 195         case TRACE_REG_UNREGISTER:
 196                 tracepoint_probe_unregister(call->name,
 197                                             call->class->probe,
 198                                             file);
 199                 return 0;
 200
 201 #ifdef CONFIG_PERF_EVENTS
 202         case TRACE_REG_PERF_REGISTER:
 203                 return tracepoint_probe_register(call->name,
 204                                                  call->class->perf_probe,
 205                                                  call);
 206         case TRACE_REG_PERF_UNREGISTER:
 207                 tracepoint_probe_unregister(call->name,
 208                                             call->class->perf_probe,
 209                                             call);
 210                 return 0;
 211         case TRACE_REG_PERF_OPEN:
 212         case TRACE_REG_PERF_CLOSE:
 213         case TRACE_REG_PERF_ADD:
 214         case TRACE_REG_PERF_DEL:
 215                 return 0;
 216 #endif
 217         }
 218         return 0;
 219 }
 220 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 221
 222 void trace_event_enable_cmd_record(bool enable)
 223 {
 224         struct ftrace_event_file *file;
 225         struct trace_array *tr;
 226
 227         mutex_lock(&event_mutex);
 228         do_for_each_event_file(tr, file) {
 229
 230                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 231                         continue;
 232
 233                 if (enable) {
 234                         tracing_start_cmdline_record();
 235                         set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 236                 } else {
 237                         tracing_stop_cmdline_record();
 238                         clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 239                 }
 240         } while_for_each_event_file();
 241         mutex_unlock(&event_mutex);
 242 }
 243
 244 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
 245                                          int enable, int soft_disable)
 246 {
 247         struct ftrace_event_call *call = file->event_call;
 248         int ret = 0;
 249         int disable;
 250
 251         switch (enable) {
 252         case 0:
 253                 /*
 254                  * When soft_disable is set and enable is cleared, the sm_ref
 255                  * reference counter is decremented. If it reaches 0, we want
 256                  * to clear the SOFT_DISABLED flag but leave the event in the
 257                  * state that it was. That is, if the event was enabled and
 258                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 259                  * is set we do not want the event to be enabled before we
 260                  * clear the bit.
 261                  *
 262                  * When soft_disable is not set but the SOFT_MODE flag is,
 263                  * we do nothing. Do not disable the tracepoint, otherwise
 264                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 265                  */
 266                 if (soft_disable) {
 267                         if (atomic_dec_return(&file->sm_ref) > 0)
 268                                 break;
 269                         disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
 270                         clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 271                 } else
 272                         disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
 273
 274                 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
 275                         clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 276                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 277                                 tracing_stop_cmdline_record();
 278                                 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 279                         }
 280                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 281                 }
 282                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT */
 283                 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 284                         set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 285                 break;
 286         case 1:
 287                 /*
 288                  * When soft_disable is set and enable is set, we want to
 289                  * register the tracepoint for the event, but leave the event
 290                  * as is. That means, if the event was already enabled, we do
 291                  * nothing (but set SOFT_MODE). If the event is disabled, we
 292                  * set SOFT_DISABLED before enabling the event tracepoint, so
 293                  * it still seems to be disabled.
 294                  */
 295                 if (!soft_disable)
 296                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 297                 else {
 298                         if (atomic_inc_return(&file->sm_ref) > 1)
 299                                 break;
 300                         set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 301                 }
 302
 303                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 304
 305                         /* Keep the event disabled, when going to SOFT_MODE. */
 306                         if (soft_disable)
 307                                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 308
 309                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 310                                 tracing_start_cmdline_record();
 311                                 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 312                         }
 313                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 314                         if (ret) {
 315                                 tracing_stop_cmdline_record();
 316                                 pr_info("event trace: Could not enable event "
 317                                         "%s\n", call->name);
 318                                 break;
 319                         }
 320                         set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 321
 322                         /* WAS_ENABLED gets set but never cleared. */
 323                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 324                 }
 325                 break;
 326         }
 327
 328         return ret;
 329 }
 330
 331 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 332                                        int enable)
 333 {
 334         return __ftrace_event_enable_disable(file, enable, 0);
 335 }
 336
 337 static void ftrace_clear_events(struct trace_array *tr)
 338 {
 339         struct ftrace_event_file *file;
 340
 341         mutex_lock(&event_mutex);
 342         list_for_each_entry(file, &tr->events, list) {
 343                 ftrace_event_enable_disable(file, 0);
 344         }
 345         mutex_unlock(&event_mutex);
 346 }
 347
 348 static void __put_system(struct event_subsystem *system)
 349 {
 350         struct event_filter *filter = system->filter;
 351
 352         WARN_ON_ONCE(system->ref_count == 0);
 353         if (--system->ref_count)
 354                 return;
 355
 356         list_del(&system->list);
 357
 358         if (filter) {
 359                 kfree(filter->filter_string);
 360                 kfree(filter);
 361         }
 362         kfree(system);
 363 }
 364
 365 static void __get_system(struct event_subsystem *system)
 366 {
 367         WARN_ON_ONCE(system->ref_count == 0);
 368         system->ref_count++;
 369 }
 370
 371 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 372 {
 373         WARN_ON_ONCE(dir->ref_count == 0);
 374         dir->ref_count++;
 375         __get_system(dir->subsystem);
 376 }
 377
 378 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 379 {
 380         WARN_ON_ONCE(dir->ref_count == 0);
 381         /* If the subsystem is about to be freed, the dir must be too */
 382         WARN_ON_ONCE(dir->subsystem->ref_count == 1 && dir->ref_count != 1);
 383
 384         __put_system(dir->subsystem);
 385         if (!--dir->ref_count)
 386                 kfree(dir);
 387 }
 388
 389 static void put_system(struct ftrace_subsystem_dir *dir)
 390 {
 391         mutex_lock(&event_mutex);
 392         __put_system_dir(dir);
 393         mutex_unlock(&event_mutex);
 394 }
 395
 396 /*
 397  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 398  */
 399 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 400                                   const char *sub, const char *event, int set)
 401 {
 402         struct ftrace_event_file *file;
 403         struct ftrace_event_call *call;
 404         int ret = -EINVAL;
 405
 406         mutex_lock(&event_mutex);
 407         list_for_each_entry(file, &tr->events, list) {
 408
 409                 call = file->event_call;
 410
 411                 if (!call->name || !call->class || !call->class->reg)
 412                         continue;
 413
 414                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 415                         continue;
 416
 417                 if (match &&
 418                     strcmp(match, call->name) != 0 &&
 419                     strcmp(match, call->class->system) != 0)
 420                         continue;
 421
 422                 if (sub && strcmp(sub, call->class->system) != 0)
 423                         continue;
 424
 425                 if (event && strcmp(event, call->name) != 0)
 426                         continue;
 427
 428                 ftrace_event_enable_disable(file, set);
 429
 430                 ret = 0;
 431         }
 432         mutex_unlock(&event_mutex);
 433
 434         return ret;
 435 }
 436
 437 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 438 {
 439         char *event = NULL, *sub = NULL, *match;
 440
 441         /*
 442          * The buf format can be <subsystem>:<event-name>
 443          *  *:<event-name> means any event by that name.
 444          *  :<event-name> is the same.
 445          *
 446          *  <subsystem>:* means all events in that subsystem
 447          *  <subsystem>: means the same.
 448          *
 449          *  <name> (no ':') means all events in a subsystem with
 450          *  the name <name> or any event that matches <name>
 451          */
 452
 453         match = strsep(&buf, ":");
 454         if (buf) {
 455                 sub = match;
 456                 event = buf;
 457                 match = NULL;
 458
 459                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 460                         sub = NULL;
 461                 if (!strlen(event) || strcmp(event, "*") == 0)
 462                         event = NULL;
 463         }
 464
 465         return __ftrace_set_clr_event(tr, match, sub, event, set);
 466 }
 467
 468 /**
 469  * trace_set_clr_event - enable or disable an event
 470  * @system: system name to match (NULL for any system)
 471  * @event: event name to match (NULL for all events, within system)
 472  * @set: 1 to enable, 0 to disable
 473  *
 474  * This is a way for other parts of the kernel to enable or disable
 475  * event recording.
 476  *
 477  * Returns 0 on success, -EINVAL if the parameters do not match any
 478  * registered events.
 479  */
 480 int trace_set_clr_event(const char *system, const char *event, int set)
 481 {
 482         struct trace_array *tr = top_trace_array();
 483
 484         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 485 }
 486 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 487
 488 /* 128 should be much more than enough */
 489 #define EVENT_BUF_SIZE          127
 490
 491 static ssize_t
 492 ftrace_event_write(struct file *file, const char __user *ubuf,
 493                    size_t cnt, loff_t *ppos)
 494 {
 495         struct trace_parser parser;
 496         struct seq_file *m = file->private_data;
 497         struct trace_array *tr = m->private;
 498         ssize_t read, ret;
 499
 500         if (!cnt)
 501                 return 0;
 502
 503         ret = tracing_update_buffers();
 504         if (ret < 0)
 505                 return ret;
 506
 507         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 508                 return -ENOMEM;
 509
 510         read = trace_get_user(&parser, ubuf, cnt, ppos);
 511
 512         if (read >= 0 && trace_parser_loaded((&parser))) {
 513                 int set = 1;
 514
 515                 if (*parser.buffer == '!')
 516                         set = 0;
 517
 518                 parser.buffer[parser.idx] = 0;
 519
 520                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 521                 if (ret)
 522                         goto out_put;
 523         }
 524
 525         ret = read;
 526
 527  out_put:
 528         trace_parser_put(&parser);
 529
 530         return ret;
 531 }
 532
 533 static void *
 534 t_next(struct seq_file *m, void *v, loff_t *pos)
 535 {
 536         struct ftrace_event_file *file = v;
 537         struct ftrace_event_call *call;
 538         struct trace_array *tr = m->private;
 539
 540         (*pos)++;
 541
 542         list_for_each_entry_continue(file, &tr->events, list) {
 543                 call = file->event_call;
 544                 /*
 545                  * The ftrace subsystem is for showing formats only.
 546                  * They can not be enabled or disabled via the event files.
 547                  */
 548                 if (call->class && call->class->reg)
 549                         return file;
 550         }
 551
 552         return NULL;
 553 }
 554
 555 static void *t_start(struct seq_file *m, loff_t *pos)
 556 {
 557         struct ftrace_event_file *file;
 558         struct trace_array *tr = m->private;
 559         loff_t l;
 560
 561         mutex_lock(&event_mutex);
 562
 563         file = list_entry(&tr->events, struct ftrace_event_file, list);
 564         for (l = 0; l <= *pos; ) {
 565                 file = t_next(m, file, &l);
 566                 if (!file)
 567                         break;
 568         }
 569         return file;
 570 }
 571
 572 static void *
 573 s_next(struct seq_file *m, void *v, loff_t *pos)
 574 {
 575         struct ftrace_event_file *file = v;
 576         struct trace_array *tr = m->private;
 577
 578         (*pos)++;
 579
 580         list_for_each_entry_continue(file, &tr->events, list) {
 581                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
 582                         return file;
 583         }
 584
 585         return NULL;
 586 }
 587
 588 static void *s_start(struct seq_file *m, loff_t *pos)
 589 {
 590         struct ftrace_event_file *file;
 591         struct trace_array *tr = m->private;
 592         loff_t l;
 593
 594         mutex_lock(&event_mutex);
 595
 596         file = list_entry(&tr->events, struct ftrace_event_file, list);
 597         for (l = 0; l <= *pos; ) {
 598                 file = s_next(m, file, &l);
 599                 if (!file)
 600                         break;
 601         }
 602         return file;
 603 }
 604
 605 static int t_show(struct seq_file *m, void *v)
 606 {
 607         struct ftrace_event_file *file = v;
 608         struct ftrace_event_call *call = file->event_call;
 609
 610         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 611                 seq_printf(m, "%s:", call->class->system);
 612         seq_printf(m, "%s\n", call->name);
 613
 614         return 0;
 615 }
 616
 617 static void t_stop(struct seq_file *m, void *p)
 618 {
 619         mutex_unlock(&event_mutex);
 620 }
 621
 622 static ssize_t
 623 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 624                   loff_t *ppos)
 625 {
 626         struct ftrace_event_file *file = filp->private_data;
 627         char *buf;
 628
 629         if (file->flags & FTRACE_EVENT_FL_ENABLED) {
 630                 if (file->flags & FTRACE_EVENT_FL_SOFT_DISABLED)
 631                         buf = "0*\n";
 632                 else if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 633                         buf = "1*\n";
 634                 else
 635                         buf = "1\n";
 636         } else
 637                 buf = "0\n";
 638
 639         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
 640 }
 641
 642 static ssize_t
 643 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 644                    loff_t *ppos)
 645 {
 646         struct ftrace_event_file *file = filp->private_data;
 647         unsigned long val;
 648         int ret;
 649
 650         if (!file)
 651                 return -EINVAL;
 652
 653         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 654         if (ret)
 655                 return ret;
 656
 657         ret = tracing_update_buffers();
 658         if (ret < 0)
 659                 return ret;
 660
 661         switch (val) {
 662         case 0:
 663         case 1:
 664                 mutex_lock(&event_mutex);
 665                 ret = ftrace_event_enable_disable(file, val);
 666                 mutex_unlock(&event_mutex);
 667                 break;
 668
 669         default:
 670                 return -EINVAL;
 671         }
 672
 673         *ppos += cnt;
 674
 675         return ret ? ret : cnt;
 676 }
 677
 678 static ssize_t
 679 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 680                    loff_t *ppos)
 681 {
 682         const char set_to_char[4] = { '?', '0', '1', 'X' };
 683         struct ftrace_subsystem_dir *dir = filp->private_data;
 684         struct event_subsystem *system = dir->subsystem;
 685         struct ftrace_event_call *call;
 686         struct ftrace_event_file *file;
 687         struct trace_array *tr = dir->tr;
 688         char buf[2];
 689         int set = 0;
 690         int ret;
 691
 692         mutex_lock(&event_mutex);
 693         list_for_each_entry(file, &tr->events, list) {
 694                 call = file->event_call;
 695                 if (!call->name || !call->class || !call->class->reg)
 696                         continue;
 697
 698                 if (system && strcmp(call->class->system, system->name) != 0)
 699                         continue;
 700
 701                 /*
 702                  * We need to find out if all the events are set
 703                  * or if all events or cleared, or if we have
 704                  * a mixture.
 705                  */
 706                 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
 707
 708                 /*
 709                  * If we have a mixture, no need to look further.
 710                  */
 711                 if (set == 3)
 712                         break;
 713         }
 714         mutex_unlock(&event_mutex);
 715
 716         buf[0] = set_to_char[set];
 717         buf[1] = '\n';
 718
 719         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 720
 721         return ret;
 722 }
 723
 724 static ssize_t
 725 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 726                     loff_t *ppos)
 727 {
 728         struct ftrace_subsystem_dir *dir = filp->private_data;
 729         struct event_subsystem *system = dir->subsystem;
 730         const char *name = NULL;
 731         unsigned long val;
 732         ssize_t ret;
 733
 734         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 735         if (ret)
 736                 return ret;
 737
 738         ret = tracing_update_buffers();
 739         if (ret < 0)
 740                 return ret;
 741
 742         if (val != 0 && val != 1)
 743                 return -EINVAL;
 744
 745         /*
 746          * Opening of "enable" adds a ref count to system,
 747          * so the name is safe to use.
 748          */
 749         if (system)
 750                 name = system->name;
 751
 752         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
 753         if (ret)
 754                 goto out;
 755
 756         ret = cnt;
 757
 758 out:
 759         *ppos += cnt;
 760
 761         return ret;
 762 }
 763
 764 enum {
 765         FORMAT_HEADER           = 1,
 766         FORMAT_FIELD_SEPERATOR  = 2,
 767         FORMAT_PRINTFMT         = 3,
 768 };
 769
 770 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 771 {
 772         struct ftrace_event_call *call = m->private;
 773         struct ftrace_event_field *field;
 774         struct list_head *common_head = &ftrace_common_fields;
 775         struct list_head *head = trace_get_fields(call);
 776
 777         (*pos)++;
 778
 779         switch ((unsigned long)v) {
 780         case FORMAT_HEADER:
 781                 if (unlikely(list_empty(common_head)))
 782                         return NULL;
 783
 784                 field = list_entry(common_head->prev,
 785                                    struct ftrace_event_field, link);
 786                 return field;
 787
 788         case FORMAT_FIELD_SEPERATOR:
 789                 if (unlikely(list_empty(head)))
 790                         return NULL;
 791
 792                 field = list_entry(head->prev, struct ftrace_event_field, link);
 793                 return field;
 794
 795         case FORMAT_PRINTFMT:
 796                 /* all done */
 797                 return NULL;
 798         }
 799
 800         field = v;
 801         if (field->link.prev == common_head)
 802                 return (void *)FORMAT_FIELD_SEPERATOR;
 803         else if (field->link.prev == head)
 804                 return (void *)FORMAT_PRINTFMT;
 805
 806         field = list_entry(field->link.prev, struct ftrace_event_field, link);
 807
 808         return field;
 809 }
 810
 811 static void *f_start(struct seq_file *m, loff_t *pos)
 812 {
 813         loff_t l = 0;
 814         void *p;
 815
 816         /* Start by showing the header */
 817         if (!*pos)
 818                 return (void *)FORMAT_HEADER;
 819
 820         p = (void *)FORMAT_HEADER;
 821         do {
 822                 p = f_next(m, p, &l);
 823         } while (p && l < *pos);
 824
 825         return p;
 826 }
 827
 828 static int f_show(struct seq_file *m, void *v)
 829 {
 830         struct ftrace_event_call *call = m->private;
 831         struct ftrace_event_field *field;
 832         const char *array_descriptor;
 833
 834         switch ((unsigned long)v) {
 835         case FORMAT_HEADER:
 836                 seq_printf(m, "name: %s\n", call->name);
 837                 seq_printf(m, "ID: %d\n", call->event.type);
 838                 seq_printf(m, "format:\n");
 839                 return 0;
 840
 841         case FORMAT_FIELD_SEPERATOR:
 842                 seq_putc(m, '\n');
 843                 return 0;
 844
 845         case FORMAT_PRINTFMT:
 846                 seq_printf(m, "\nprint fmt: %s\n",
 847                            call->print_fmt);
 848                 return 0;
 849         }
 850
 851         field = v;
 852
 853         /*
 854          * Smartly shows the array type(except dynamic array).
 855          * Normal:
 856          *      field:TYPE VAR
 857          * If TYPE := TYPE[LEN], it is shown:
 858          *      field:TYPE VAR[LEN]
 859          */
 860         array_descriptor = strchr(field->type, '[');
 861
 862         if (!strncmp(field->type, "__data_loc", 10))
 863                 array_descriptor = NULL;
 864
 865         if (!array_descriptor)
 866                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 867                            field->type, field->name, field->offset,
 868                            field->size, !!field->is_signed);
 869         else
 870                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 871                            (int)(array_descriptor - field->type),
 872                            field->type, field->name,
 873                            array_descriptor, field->offset,
 874                            field->size, !!field->is_signed);
 875
 876         return 0;
 877 }
 878
 879 static void f_stop(struct seq_file *m, void *p)
 880 {
 881 }
 882
 883 static const struct seq_operations trace_format_seq_ops = {
 884         .start          = f_start,
 885         .next           = f_next,
 886         .stop           = f_stop,
 887         .show           = f_show,
 888 };
 889
 890 static int trace_format_open(struct inode *inode, struct file *file)
 891 {
 892         struct ftrace_event_call *call = inode->i_private;
 893         struct seq_file *m;
 894         int ret;
 895
 896         ret = seq_open(file, &trace_format_seq_ops);
 897         if (ret < 0)
 898                 return ret;
 899
 900         m = file->private_data;
 901         m->private = call;
 902
 903         return 0;
 904 }
 905
 906 static ssize_t
 907 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 908 {
 909         struct ftrace_event_call *call = filp->private_data;
 910         struct trace_seq *s;
 911         int r;
 912
 913         if (*ppos)
 914                 return 0;
 915
 916         s = kmalloc(sizeof(*s), GFP_KERNEL);
 917         if (!s)
 918                 return -ENOMEM;
 919
 920         trace_seq_init(s);
 921         trace_seq_printf(s, "%d\n", call->event.type);
 922
 923         r = simple_read_from_buffer(ubuf, cnt, ppos,
 924                                     s->buffer, s->len);
 925         kfree(s);
 926         return r;
 927 }
 928
 929 static ssize_t
 930 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 931                   loff_t *ppos)
 932 {
 933         struct ftrace_event_call *call = filp->private_data;
 934         struct trace_seq *s;
 935         int r;
 936
 937         if (*ppos)
 938                 return 0;
 939
 940         s = kmalloc(sizeof(*s), GFP_KERNEL);
 941         if (!s)
 942                 return -ENOMEM;
 943
 944         trace_seq_init(s);
 945
 946         print_event_filter(call, s);
 947         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 948
 949         kfree(s);
 950
 951         return r;
 952 }
 953
 954 static ssize_t
 955 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 956                    loff_t *ppos)
 957 {
 958         struct ftrace_event_call *call = filp->private_data;
 959         char *buf;
 960         int err;
 961
 962         if (cnt >= PAGE_SIZE)
 963                 return -EINVAL;
 964
 965         buf = (char *)__get_free_page(GFP_TEMPORARY);
 966         if (!buf)
 967                 return -ENOMEM;
 968
 969         if (copy_from_user(buf, ubuf, cnt)) {
 970                 free_page((unsigned long) buf);
 971                 return -EFAULT;
 972         }
 973         buf[cnt] = '\0';
 974
 975         err = apply_event_filter(call, buf);
 976         free_page((unsigned long) buf);
 977         if (err < 0)
 978                 return err;
 979
 980         *ppos += cnt;
 981
 982         return cnt;
 983 }
 984
 985 static LIST_HEAD(event_subsystems);
 986
 987 static int subsystem_open(struct inode *inode, struct file *filp)
 988 {
 989         struct event_subsystem *system = NULL;
 990         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
 991         struct trace_array *tr;
 992         int ret;
 993
 994         /* Make sure the system still exists */
 995         mutex_lock(&event_mutex);
 996         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
 997                 list_for_each_entry(dir, &tr->systems, list) {
 998                         if (dir == inode->i_private) {
 999                                 /* Don't open systems with no events */
1000                                 if (dir->nr_events) {
1001                                         __get_system_dir(dir);
1002                                         system = dir->subsystem;
1003                                 }
1004                                 goto exit_loop;
1005                         }
1006                 }
1007         }
1008  exit_loop:
1009         mutex_unlock(&event_mutex);
1010
1011         if (!system)
1012                 return -ENODEV;
1013
1014         /* Some versions of gcc think dir can be uninitialized here */
1015         WARN_ON(!dir);
1016
1017         ret = tracing_open_generic(inode, filp);
1018         if (ret < 0)
1019                 put_system(dir);
1020
1021         return ret;
1022 }
1023
1024 static int system_tr_open(struct inode *inode, struct file *filp)
1025 {
1026         struct ftrace_subsystem_dir *dir;
1027         struct trace_array *tr = inode->i_private;
1028         int ret;
1029
1030         /* Make a temporary dir that has no system but points to tr */
1031         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1032         if (!dir)
1033                 return -ENOMEM;
1034
1035         dir->tr = tr;
1036
1037         ret = tracing_open_generic(inode, filp);
1038         if (ret < 0)
1039                 kfree(dir);
1040
1041         filp->private_data = dir;
1042
1043         return ret;
1044 }
1045
1046 static int subsystem_release(struct inode *inode, struct file *file)
1047 {
1048         struct ftrace_subsystem_dir *dir = file->private_data;
1049
1050         /*
1051          * If dir->subsystem is NULL, then this is a temporary
1052          * descriptor that was made for a trace_array to enable
1053          * all subsystems.
1054          */
1055         if (dir->subsystem)
1056                 put_system(dir);
1057         else
1058                 kfree(dir);
1059
1060         return 0;
1061 }
1062
1063 static ssize_t
1064 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1065                       loff_t *ppos)
1066 {
1067         struct ftrace_subsystem_dir *dir = filp->private_data;
1068         struct event_subsystem *system = dir->subsystem;
1069         struct trace_seq *s;
1070         int r;
1071
1072         if (*ppos)
1073                 return 0;
1074
1075         s = kmalloc(sizeof(*s), GFP_KERNEL);
1076         if (!s)
1077                 return -ENOMEM;
1078
1079         trace_seq_init(s);
1080
1081         print_subsystem_event_filter(system, s);
1082         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1083
1084         kfree(s);
1085
1086         return r;
1087 }
1088
1089 static ssize_t
1090 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1091                        loff_t *ppos)
1092 {
1093         struct ftrace_subsystem_dir *dir = filp->private_data;
1094         char *buf;
1095         int err;
1096
1097         if (cnt >= PAGE_SIZE)
1098                 return -EINVAL;
1099
1100         buf = (char *)__get_free_page(GFP_TEMPORARY);
1101         if (!buf)
1102                 return -ENOMEM;
1103
1104         if (copy_from_user(buf, ubuf, cnt)) {
1105                 free_page((unsigned long) buf);
1106                 return -EFAULT;
1107         }
1108         buf[cnt] = '\0';
1109
1110         err = apply_subsystem_event_filter(dir, buf);
1111         free_page((unsigned long) buf);
1112         if (err < 0)
1113                 return err;
1114
1115         *ppos += cnt;
1116
1117         return cnt;
1118 }
1119
1120 static ssize_t
1121 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1122 {
1123         int (*func)(struct trace_seq *s) = filp->private_data;
1124         struct trace_seq *s;
1125         int r;
1126
1127         if (*ppos)
1128                 return 0;
1129
1130         s = kmalloc(sizeof(*s), GFP_KERNEL);
1131         if (!s)
1132                 return -ENOMEM;
1133
1134         trace_seq_init(s);
1135
1136         func(s);
1137         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1138
1139         kfree(s);
1140
1141         return r;
1142 }
1143
1144 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1145 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1146
1147 static const struct seq_operations show_event_seq_ops = {
1148         .start = t_start,
1149         .next = t_next,
1150         .show = t_show,
1151         .stop = t_stop,
1152 };
1153
1154 static const struct seq_operations show_set_event_seq_ops = {
1155         .start = s_start,
1156         .next = s_next,
1157         .show = t_show,
1158         .stop = t_stop,
1159 };
1160
1161 static const struct file_operations ftrace_avail_fops = {
1162         .open = ftrace_event_avail_open,
1163         .read = seq_read,
1164         .llseek = seq_lseek,
1165         .release = seq_release,
1166 };
1167
1168 static const struct file_operations ftrace_set_event_fops = {
1169         .open = ftrace_event_set_open,
1170         .read = seq_read,
1171         .write = ftrace_event_write,
1172         .llseek = seq_lseek,
1173         .release = seq_release,
1174 };
1175
1176 static const struct file_operations ftrace_enable_fops = {
1177         .open = tracing_open_generic,
1178         .read = event_enable_read,
1179         .write = event_enable_write,
1180         .llseek = default_llseek,
1181 };
1182
1183 static const struct file_operations ftrace_event_format_fops = {
1184         .open = trace_format_open,
1185         .read = seq_read,
1186         .llseek = seq_lseek,
1187         .release = seq_release,
1188 };
1189
1190 static const struct file_operations ftrace_event_id_fops = {
1191         .open = tracing_open_generic,
1192         .read = event_id_read,
1193         .llseek = default_llseek,
1194 };
1195
1196 static const struct file_operations ftrace_event_filter_fops = {
1197         .open = tracing_open_generic,
1198         .read = event_filter_read,
1199         .write = event_filter_write,
1200         .llseek = default_llseek,
1201 };
1202
1203 static const struct file_operations ftrace_subsystem_filter_fops = {
1204         .open = subsystem_open,
1205         .read = subsystem_filter_read,
1206         .write = subsystem_filter_write,
1207         .llseek = default_llseek,
1208         .release = subsystem_release,
1209 };
1210
1211 static const struct file_operations ftrace_system_enable_fops = {
1212         .open = subsystem_open,
1213         .read = system_enable_read,
1214         .write = system_enable_write,
1215         .llseek = default_llseek,
1216         .release = subsystem_release,
1217 };
1218
1219 static const struct file_operations ftrace_tr_enable_fops = {
1220         .open = system_tr_open,
1221         .read = system_enable_read,
1222         .write = system_enable_write,
1223         .llseek = default_llseek,
1224         .release = subsystem_release,
1225 };
1226
1227 static const struct file_operations ftrace_show_header_fops = {
1228         .open = tracing_open_generic,
1229         .read = show_header,
1230         .llseek = default_llseek,
1231 };
1232
1233 static int
1234 ftrace_event_open(struct inode *inode, struct file *file,
1235                   const struct seq_operations *seq_ops)
1236 {
1237         struct seq_file *m;
1238         int ret;
1239
1240         ret = seq_open(file, seq_ops);
1241         if (ret < 0)
1242                 return ret;
1243         m = file->private_data;
1244         /* copy tr over to seq ops */
1245         m->private = inode->i_private;
1246
1247         return ret;
1248 }
1249
1250 static int
1251 ftrace_event_avail_open(struct inode *inode, struct file *file)
1252 {
1253         const struct seq_operations *seq_ops = &show_event_seq_ops;
1254
1255         return ftrace_event_open(inode, file, seq_ops);
1256 }
1257
1258 static int
1259 ftrace_event_set_open(struct inode *inode, struct file *file)
1260 {
1261         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1262         struct trace_array *tr = inode->i_private;
1263
1264         if ((file->f_mode & FMODE_WRITE) &&
1265             (file->f_flags & O_TRUNC))
1266                 ftrace_clear_events(tr);
1267
1268         return ftrace_event_open(inode, file, seq_ops);
1269 }
1270
1271 static struct event_subsystem *
1272 create_new_subsystem(const char *name)
1273 {
1274         struct event_subsystem *system;
1275
1276         /* need to create new entry */
1277         system = kmalloc(sizeof(*system), GFP_KERNEL);
1278         if (!system)
1279                 return NULL;
1280
1281         system->ref_count = 1;
1282         system->name = name;
1283
1284         system->filter = NULL;
1285
1286         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1287         if (!system->filter)
1288                 goto out_free;
1289
1290         list_add(&system->list, &event_subsystems);
1291
1292         return system;
1293
1294  out_free:
1295         kfree(system);
1296         return NULL;
1297 }
1298
1299 static struct dentry *
1300 event_subsystem_dir(struct trace_array *tr, const char *name,
1301                     struct ftrace_event_file *file, struct dentry *parent)
1302 {
1303         struct ftrace_subsystem_dir *dir;
1304         struct event_subsystem *system;
1305         struct dentry *entry;
1306
1307         /* First see if we did not already create this dir */
1308         list_for_each_entry(dir, &tr->systems, list) {
1309                 system = dir->subsystem;
1310                 if (strcmp(system->name, name) == 0) {
1311                         dir->nr_events++;
1312                         file->system = dir;
1313                         return dir->entry;
1314                 }
1315         }
1316
1317         /* Now see if the system itself exists. */
1318         list_for_each_entry(system, &event_subsystems, list) {
1319                 if (strcmp(system->name, name) == 0)
1320                         break;
1321         }
1322         /* Reset system variable when not found */
1323         if (&system->list == &event_subsystems)
1324                 system = NULL;
1325
1326         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1327         if (!dir)
1328                 goto out_fail;
1329
1330         if (!system) {
1331                 system = create_new_subsystem(name);
1332                 if (!system)
1333                         goto out_free;
1334         } else
1335                 __get_system(system);
1336
1337         dir->entry = debugfs_create_dir(name, parent);
1338         if (!dir->entry) {
1339                 pr_warning("Failed to create system directory %s\n", name);
1340                 __put_system(system);
1341                 goto out_free;
1342         }
1343
1344         dir->tr = tr;
1345         dir->ref_count = 1;
1346         dir->nr_events = 1;
1347         dir->subsystem = system;
1348         file->system = dir;
1349
1350         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1351                                     &ftrace_subsystem_filter_fops);
1352         if (!entry) {
1353                 kfree(system->filter);
1354                 system->filter = NULL;
1355                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1356         }
1357
1358         trace_create_file("enable", 0644, dir->entry, dir,
1359                           &ftrace_system_enable_fops);
1360
1361         list_add(&dir->list, &tr->systems);
1362
1363         return dir->entry;
1364
1365  out_free:
1366         kfree(dir);
1367  out_fail:
1368         /* Only print this message if failed on memory allocation */
1369         if (!dir || !system)
1370                 pr_warning("No memory to create event subsystem %s\n",
1371                            name);
1372         return NULL;
1373 }
1374
1375 static int
1376 event_create_dir(struct dentry *parent,
1377                  struct ftrace_event_file *file,
1378                  const struct file_operations *id,
1379                  const struct file_operations *enable,
1380                  const struct file_operations *filter,
1381                  const struct file_operations *format)
1382 {
1383         struct ftrace_event_call *call = file->event_call;
1384         struct trace_array *tr = file->tr;
1385         struct list_head *head;
1386         struct dentry *d_events;
1387         int ret;
1388
1389         /*
1390          * If the trace point header did not define TRACE_SYSTEM
1391          * then the system would be called "TRACE_SYSTEM".
1392          */
1393         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1394                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1395                 if (!d_events)
1396                         return -ENOMEM;
1397         } else
1398                 d_events = parent;
1399
1400         file->dir = debugfs_create_dir(call->name, d_events);
1401         if (!file->dir) {
1402                 pr_warning("Could not create debugfs '%s' directory\n",
1403                            call->name);
1404                 return -1;
1405         }
1406
1407         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1408                 trace_create_file("enable", 0644, file->dir, file,
1409                                   enable);
1410
1411 #ifdef CONFIG_PERF_EVENTS
1412         if (call->event.type && call->class->reg)
1413                 trace_create_file("id", 0444, file->dir, call,
1414                                   id);
1415 #endif
1416
1417         /*
1418          * Other events may have the same class. Only update
1419          * the fields if they are not already defined.
1420          */
1421         head = trace_get_fields(call);
1422         if (list_empty(head)) {
1423                 ret = call->class->define_fields(call);
1424                 if (ret < 0) {
1425                         pr_warning("Could not initialize trace point"
1426                                    " events/%s\n", call->name);
1427                         return -1;
1428                 }
1429         }
1430         trace_create_file("filter", 0644, file->dir, call,
1431                           filter);
1432
1433         trace_create_file("format", 0444, file->dir, call,
1434                           format);
1435
1436         return 0;
1437 }
1438
1439 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
1440 {
1441         if (!dir)
1442                 return;
1443
1444         if (!--dir->nr_events) {
1445                 debugfs_remove_recursive(dir->entry);
1446                 list_del(&dir->list);
1447                 __put_system_dir(dir);
1448         }
1449 }
1450
1451 static void remove_event_from_tracers(struct ftrace_event_call *call)
1452 {
1453         struct ftrace_event_file *file;
1454         struct trace_array *tr;
1455
1456         do_for_each_event_file_safe(tr, file) {
1457
1458                 if (file->event_call != call)
1459                         continue;
1460
1461                 list_del(&file->list);
1462                 debugfs_remove_recursive(file->dir);
1463                 remove_subsystem(file->system);
1464                 kmem_cache_free(file_cachep, file);
1465
1466                 /*
1467                  * The do_for_each_event_file_safe() is
1468                  * a double loop. After finding the call for this
1469                  * trace_array, we use break to jump to the next
1470                  * trace_array.
1471                  */
1472                 break;
1473         } while_for_each_event_file();
1474 }
1475
1476 static void event_remove(struct ftrace_event_call *call)
1477 {
1478         struct trace_array *tr;
1479         struct ftrace_event_file *file;
1480
1481         do_for_each_event_file(tr, file) {
1482                 if (file->event_call != call)
1483                         continue;
1484                 ftrace_event_enable_disable(file, 0);
1485                 /*
1486                  * The do_for_each_event_file() is
1487                  * a double loop. After finding the call for this
1488                  * trace_array, we use break to jump to the next
1489                  * trace_array.
1490                  */
1491                 break;
1492         } while_for_each_event_file();
1493
1494         if (call->event.funcs)
1495                 __unregister_ftrace_event(&call->event);
1496         remove_event_from_tracers(call);
1497         list_del(&call->list);
1498 }
1499
1500 static int event_init(struct ftrace_event_call *call)
1501 {
1502         int ret = 0;
1503
1504         if (WARN_ON(!call->name))
1505                 return -EINVAL;
1506
1507         if (call->class->raw_init) {
1508                 ret = call->class->raw_init(call);
1509                 if (ret < 0 && ret != -ENOSYS)
1510                         pr_warn("Could not initialize trace events/%s\n",
1511                                 call->name);
1512         }
1513
1514         return ret;
1515 }
1516
1517 static int
1518 __register_event(struct ftrace_event_call *call, struct module *mod)
1519 {
1520         int ret;
1521
1522         ret = event_init(call);
1523         if (ret < 0)
1524                 return ret;
1525
1526         list_add(&call->list, &ftrace_events);
1527         call->mod = mod;
1528
1529         return 0;
1530 }
1531
1532 static struct ftrace_event_file *
1533 trace_create_new_event(struct ftrace_event_call *call,
1534                        struct trace_array *tr)
1535 {
1536         struct ftrace_event_file *file;
1537
1538         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1539         if (!file)
1540                 return NULL;
1541
1542         file->event_call = call;
1543         file->tr = tr;
1544         atomic_set(&file->sm_ref, 0);
1545         list_add(&file->list, &tr->events);
1546
1547         return file;
1548 }
1549
1550 /* Add an event to a trace directory */
1551 static int
1552 __trace_add_new_event(struct ftrace_event_call *call,
1553                       struct trace_array *tr,
1554                       const struct file_operations *id,
1555                       const struct file_operations *enable,
1556                       const struct file_operations *filter,
1557                       const struct file_operations *format)
1558 {
1559         struct ftrace_event_file *file;
1560
1561         file = trace_create_new_event(call, tr);
1562         if (!file)
1563                 return -ENOMEM;
1564
1565         return event_create_dir(tr->event_dir, file, id, enable, filter, format);
1566 }
1567
1568 /*
1569  * Just create a decriptor for early init. A descriptor is required
1570  * for enabling events at boot. We want to enable events before
1571  * the filesystem is initialized.
1572  */
1573 static __init int
1574 __trace_early_add_new_event(struct ftrace_event_call *call,
1575                             struct trace_array *tr)
1576 {
1577         struct ftrace_event_file *file;
1578
1579         file = trace_create_new_event(call, tr);
1580         if (!file)
1581                 return -ENOMEM;
1582
1583         return 0;
1584 }
1585
1586 struct ftrace_module_file_ops;
1587 static void __add_event_to_tracers(struct ftrace_event_call *call,
1588                                    struct ftrace_module_file_ops *file_ops);
1589
1590 /* Add an additional event_call dynamically */
1591 int trace_add_event_call(struct ftrace_event_call *call)
1592 {
1593         int ret;
1594         mutex_lock(&event_mutex);
1595
1596         ret = __register_event(call, NULL);
1597         if (ret >= 0)
1598                 __add_event_to_tracers(call, NULL);
1599
1600         mutex_unlock(&event_mutex);
1601         return ret;
1602 }
1603
1604 /*
1605  * Must be called under locking both of event_mutex and trace_event_sem.
1606  */
1607 static void __trace_remove_event_call(struct ftrace_event_call *call)
1608 {
1609         event_remove(call);
1610         trace_destroy_fields(call);
1611         destroy_preds(call);
1612 }
1613
1614 /* Remove an event_call */
1615 void trace_remove_event_call(struct ftrace_event_call *call)
1616 {
1617         mutex_lock(&event_mutex);
1618         down_write(&trace_event_sem);
1619         __trace_remove_event_call(call);
1620         up_write(&trace_event_sem);
1621         mutex_unlock(&event_mutex);
1622 }
1623
1624 #define for_each_event(event, start, end)                       \
1625         for (event = start;                                     \
1626              (unsigned long)event < (unsigned long)end;         \
1627              event++)
1628
1629 #ifdef CONFIG_MODULES
1630
1631 static LIST_HEAD(ftrace_module_file_list);
1632
1633 /*
1634  * Modules must own their file_operations to keep up with
1635  * reference counting.
1636  */
1637 struct ftrace_module_file_ops {
1638         struct list_head                list;
1639         struct module                   *mod;
1640         struct file_operations          id;
1641         struct file_operations          enable;
1642         struct file_operations          format;
1643         struct file_operations          filter;
1644 };
1645
1646 static struct ftrace_module_file_ops *
1647 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1648 {
1649         /*
1650          * As event_calls are added in groups by module,
1651          * when we find one file_ops, we don't need to search for
1652          * each call in that module, as the rest should be the
1653          * same. Only search for a new one if the last one did
1654          * not match.
1655          */
1656         if (file_ops && mod == file_ops->mod)
1657                 return file_ops;
1658
1659         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1660                 if (file_ops->mod == mod)
1661                         return file_ops;
1662         }
1663         return NULL;
1664 }
1665
1666 static struct ftrace_module_file_ops *
1667 trace_create_file_ops(struct module *mod)
1668 {
1669         struct ftrace_module_file_ops *file_ops;
1670
1671         /*
1672          * This is a bit of a PITA. To allow for correct reference
1673          * counting, modules must "own" their file_operations.
1674          * To do this, we allocate the file operations that will be
1675          * used in the event directory.
1676          */
1677
1678         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1679         if (!file_ops)
1680                 return NULL;
1681
1682         file_ops->mod = mod;
1683
1684         file_ops->id = ftrace_event_id_fops;
1685         file_ops->id.owner = mod;
1686
1687         file_ops->enable = ftrace_enable_fops;
1688         file_ops->enable.owner = mod;
1689
1690         file_ops->filter = ftrace_event_filter_fops;
1691         file_ops->filter.owner = mod;
1692
1693         file_ops->format = ftrace_event_format_fops;
1694         file_ops->format.owner = mod;
1695
1696         list_add(&file_ops->list, &ftrace_module_file_list);
1697
1698         return file_ops;
1699 }
1700
1701 static void trace_module_add_events(struct module *mod)
1702 {
1703         struct ftrace_module_file_ops *file_ops = NULL;
1704         struct ftrace_event_call **call, **start, **end;
1705
1706         start = mod->trace_events;
1707         end = mod->trace_events + mod->num_trace_events;
1708
1709         if (start == end)
1710                 return;
1711
1712         file_ops = trace_create_file_ops(mod);
1713         if (!file_ops)
1714                 return;
1715
1716         for_each_event(call, start, end) {
1717                 __register_event(*call, mod);
1718                 __add_event_to_tracers(*call, file_ops);
1719         }
1720 }
1721
1722 static void trace_module_remove_events(struct module *mod)
1723 {
1724         struct ftrace_module_file_ops *file_ops;
1725         struct ftrace_event_call *call, *p;
1726         bool clear_trace = false;
1727
1728         down_write(&trace_event_sem);
1729         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1730                 if (call->mod == mod) {
1731                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1732                                 clear_trace = true;
1733                         __trace_remove_event_call(call);
1734                 }
1735         }
1736
1737         /* Now free the file_operations */
1738         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1739                 if (file_ops->mod == mod)
1740                         break;
1741         }
1742         if (&file_ops->list != &ftrace_module_file_list) {
1743                 list_del(&file_ops->list);
1744                 kfree(file_ops);
1745         }
1746         up_write(&trace_event_sem);
1747
1748         /*
1749          * It is safest to reset the ring buffer if the module being unloaded
1750          * registered any events that were used. The only worry is if
1751          * a new module gets loaded, and takes on the same id as the events
1752          * of this module. When printing out the buffer, traced events left
1753          * over from this module may be passed to the new module events and
1754          * unexpected results may occur.
1755          */
1756         if (clear_trace)
1757                 tracing_reset_all_online_cpus();
1758 }
1759
1760 static int trace_module_notify(struct notifier_block *self,
1761                                unsigned long val, void *data)
1762 {
1763         struct module *mod = data;
1764
1765         mutex_lock(&event_mutex);
1766         switch (val) {
1767         case MODULE_STATE_COMING:
1768                 trace_module_add_events(mod);
1769                 break;
1770         case MODULE_STATE_GOING:
1771                 trace_module_remove_events(mod);
1772                 break;
1773         }
1774         mutex_unlock(&event_mutex);
1775
1776         return 0;
1777 }
1778
1779 static int
1780 __trace_add_new_mod_event(struct ftrace_event_call *call,
1781                           struct trace_array *tr,
1782                           struct ftrace_module_file_ops *file_ops)
1783 {
1784         return __trace_add_new_event(call, tr,
1785                                      &file_ops->id, &file_ops->enable,
1786                                      &file_ops->filter, &file_ops->format);
1787 }
1788
1789 #else
1790 static inline struct ftrace_module_file_ops *
1791 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1792 {
1793         return NULL;
1794 }
1795 static inline int trace_module_notify(struct notifier_block *self,
1796                                       unsigned long val, void *data)
1797 {
1798         return 0;
1799 }
1800 static inline int
1801 __trace_add_new_mod_event(struct ftrace_event_call *call,
1802                           struct trace_array *tr,
1803                           struct ftrace_module_file_ops *file_ops)
1804 {
1805         return -ENODEV;
1806 }
1807 #endif /* CONFIG_MODULES */
1808
1809 /* Create a new event directory structure for a trace directory. */
1810 static void
1811 __trace_add_event_dirs(struct trace_array *tr)
1812 {
1813         struct ftrace_module_file_ops *file_ops = NULL;
1814         struct ftrace_event_call *call;
1815         int ret;
1816
1817         list_for_each_entry(call, &ftrace_events, list) {
1818                 if (call->mod) {
1819                         /*
1820                          * Directories for events by modules need to
1821                          * keep module ref counts when opened (as we don't
1822                          * want the module to disappear when reading one
1823                          * of these files). The file_ops keep account of
1824                          * the module ref count.
1825                          */
1826                         file_ops = find_ftrace_file_ops(file_ops, call->mod);
1827                         if (!file_ops)
1828                                 continue; /* Warn? */
1829                         ret = __trace_add_new_mod_event(call, tr, file_ops);
1830                         if (ret < 0)
1831                                 pr_warning("Could not create directory for event %s\n",
1832                                            call->name);
1833                         continue;
1834                 }
1835                 ret = __trace_add_new_event(call, tr,
1836                                             &ftrace_event_id_fops,
1837                                             &ftrace_enable_fops,
1838                                             &ftrace_event_filter_fops,
1839                                             &ftrace_event_format_fops);
1840                 if (ret < 0)
1841                         pr_warning("Could not create directory for event %s\n",
1842                                    call->name);
1843         }
1844 }
1845
1846 #ifdef CONFIG_DYNAMIC_FTRACE
1847
1848 /* Avoid typos */
1849 #define ENABLE_EVENT_STR        "enable_event"
1850 #define DISABLE_EVENT_STR       "disable_event"
1851
1852 struct event_probe_data {
1853         struct ftrace_event_file        *file;
1854         unsigned long                   count;
1855         int                             ref;
1856         bool                            enable;
1857 };
1858
1859 static struct ftrace_event_file *
1860 find_event_file(struct trace_array *tr, const char *system,  const char *event)
1861 {
1862         struct ftrace_event_file *file;
1863         struct ftrace_event_call *call;
1864
1865         list_for_each_entry(file, &tr->events, list) {
1866
1867                 call = file->event_call;
1868
1869                 if (!call->name || !call->class || !call->class->reg)
1870                         continue;
1871
1872                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
1873                         continue;
1874
1875                 if (strcmp(event, call->name) == 0 &&
1876                     strcmp(system, call->class->system) == 0)
1877                         return file;
1878         }
1879         return NULL;
1880 }
1881
1882 static void
1883 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1884 {
1885         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1886         struct event_probe_data *data = *pdata;
1887
1888         if (!data)
1889                 return;
1890
1891         if (data->enable)
1892                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1893         else
1894                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
1895 }
1896
1897 static void
1898 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
1899 {
1900         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1901         struct event_probe_data *data = *pdata;
1902
1903         if (!data)
1904                 return;
1905
1906         if (!data->count)
1907                 return;
1908
1909         /* Skip if the event is in a state we want to switch to */
1910         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
1911                 return;
1912
1913         if (data->count != -1)
1914                 (data->count)--;
1915
1916         event_enable_probe(ip, parent_ip, _data);
1917 }
1918
1919 static int
1920 event_enable_print(struct seq_file *m, unsigned long ip,
1921                       struct ftrace_probe_ops *ops, void *_data)
1922 {
1923         struct event_probe_data *data = _data;
1924
1925         seq_printf(m, "%ps:", (void *)ip);
1926
1927         seq_printf(m, "%s:%s:%s",
1928                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
1929                    data->file->event_call->class->system,
1930                    data->file->event_call->name);
1931
1932         if (data->count == -1)
1933                 seq_printf(m, ":unlimited\n");
1934         else
1935                 seq_printf(m, ":count=%ld\n", data->count);
1936
1937         return 0;
1938 }
1939
1940 static int
1941 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
1942                   void **_data)
1943 {
1944         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1945         struct event_probe_data *data = *pdata;
1946
1947         data->ref++;
1948         return 0;
1949 }
1950
1951 static void
1952 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
1953                   void **_data)
1954 {
1955         struct event_probe_data **pdata = (struct event_probe_data **)_data;
1956         struct event_probe_data *data = *pdata;
1957
1958         if (WARN_ON_ONCE(data->ref <= 0))
1959                 return;
1960
1961         data->ref--;
1962         if (!data->ref) {
1963                 /* Remove the SOFT_MODE flag */
1964                 __ftrace_event_enable_disable(data->file, 0, 1);
1965                 module_put(data->file->event_call->mod);
1966                 kfree(data);
1967         }
1968         *pdata = NULL;
1969 }
1970
1971 static struct ftrace_probe_ops event_enable_probe_ops = {
1972         .func                   = event_enable_probe,
1973         .print                  = event_enable_print,
1974         .init                   = event_enable_init,
1975         .free                   = event_enable_free,
1976 };
1977
1978 static struct ftrace_probe_ops event_enable_count_probe_ops = {
1979         .func                   = event_enable_count_probe,
1980         .print                  = event_enable_print,
1981         .init                   = event_enable_init,
1982         .free                   = event_enable_free,
1983 };
1984
1985 static struct ftrace_probe_ops event_disable_probe_ops = {
1986         .func                   = event_enable_probe,
1987         .print                  = event_enable_print,
1988         .init                   = event_enable_init,
1989         .free                   = event_enable_free,
1990 };
1991
1992 static struct ftrace_probe_ops event_disable_count_probe_ops = {
1993         .func                   = event_enable_count_probe,
1994         .print                  = event_enable_print,
1995         .init                   = event_enable_init,
1996         .free                   = event_enable_free,
1997 };
1998
1999 static int
2000 event_enable_func(struct ftrace_hash *hash,
2001                   char *glob, char *cmd, char *param, int enabled)
2002 {
2003         struct trace_array *tr = top_trace_array();
2004         struct ftrace_event_file *file;
2005         struct ftrace_probe_ops *ops;
2006         struct event_probe_data *data;
2007         const char *system;
2008         const char *event;
2009         char *number;
2010         bool enable;
2011         int ret;
2012
2013         /* hash funcs only work with set_ftrace_filter */
2014         if (!enabled)
2015                 return -EINVAL;
2016
2017         if (!param)
2018                 return -EINVAL;
2019
2020         system = strsep(&param, ":");
2021         if (!param)
2022                 return -EINVAL;
2023
2024         event = strsep(&param, ":");
2025
2026         mutex_lock(&event_mutex);
2027
2028         ret = -EINVAL;
2029         file = find_event_file(tr, system, event);
2030         if (!file)
2031                 goto out;
2032
2033         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2034
2035         if (enable)
2036                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2037         else
2038                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2039
2040         if (glob[0] == '!') {
2041                 unregister_ftrace_function_probe_func(glob+1, ops);
2042                 ret = 0;
2043                 goto out;
2044         }
2045
2046         ret = -ENOMEM;
2047         data = kzalloc(sizeof(*data), GFP_KERNEL);
2048         if (!data)
2049                 goto out;
2050
2051         data->enable = enable;
2052         data->count = -1;
2053         data->file = file;
2054
2055         if (!param)
2056                 goto out_reg;
2057
2058         number = strsep(&param, ":");
2059
2060         ret = -EINVAL;
2061         if (!strlen(number))
2062                 goto out_free;
2063
2064         /*
2065          * We use the callback data field (which is a pointer)
2066          * as our counter.
2067          */
2068         ret = kstrtoul(number, 0, &data->count);
2069         if (ret)
2070                 goto out_free;
2071
2072  out_reg:
2073         /* Don't let event modules unload while probe registered */
2074         ret = try_module_get(file->event_call->mod);
2075         if (!ret) {
2076                 ret = -EBUSY;
2077                 goto out_free;
2078         }
2079
2080         ret = __ftrace_event_enable_disable(file, 1, 1);
2081         if (ret < 0)
2082                 goto out_put;
2083         ret = register_ftrace_function_probe(glob, ops, data);
2084         /*
2085          * The above returns on success the # of functions enabled,
2086          * but if it didn't find any functions it returns zero.
2087          * Consider no functions a failure too.
2088          */
2089         if (!ret) {
2090                 ret = -ENOENT;
2091                 goto out_disable;
2092         } else if (ret < 0)
2093                 goto out_disable;
2094         /* Just return zero, not the number of enabled functions */
2095         ret = 0;
2096  out:
2097         mutex_unlock(&event_mutex);
2098         return ret;
2099
2100  out_disable:
2101         __ftrace_event_enable_disable(file, 0, 1);
2102  out_put:
2103         module_put(file->event_call->mod);
2104  out_free:
2105         kfree(data);
2106         goto out;
2107 }
2108
2109 static struct ftrace_func_command event_enable_cmd = {
2110         .name                   = ENABLE_EVENT_STR,
2111         .func                   = event_enable_func,
2112 };
2113
2114 static struct ftrace_func_command event_disable_cmd = {
2115         .name                   = DISABLE_EVENT_STR,
2116         .func                   = event_enable_func,
2117 };
2118
2119 static __init int register_event_cmds(void)
2120 {
2121         int ret;
2122
2123         ret = register_ftrace_command(&event_enable_cmd);
2124         if (WARN_ON(ret < 0))
2125                 return ret;
2126         ret = register_ftrace_command(&event_disable_cmd);
2127         if (WARN_ON(ret < 0))
2128                 unregister_ftrace_command(&event_enable_cmd);
2129         return ret;
2130 }
2131 #else
2132 static inline int register_event_cmds(void) { return 0; }
2133 #endif /* CONFIG_DYNAMIC_FTRACE */
2134
2135 /*
2136  * The top level array has already had its ftrace_event_file
2137  * descriptors created in order to allow for early events to
2138  * be recorded. This function is called after the debugfs has been
2139  * initialized, and we now have to create the files associated
2140  * to the events.
2141  */
2142 static __init void
2143 __trace_early_add_event_dirs(struct trace_array *tr)
2144 {
2145         struct ftrace_event_file *file;
2146         int ret;
2147
2148
2149         list_for_each_entry(file, &tr->events, list) {
2150                 ret = event_create_dir(tr->event_dir, file,
2151                                        &ftrace_event_id_fops,
2152                                        &ftrace_enable_fops,
2153                                        &ftrace_event_filter_fops,
2154                                        &ftrace_event_format_fops);
2155                 if (ret < 0)
2156                         pr_warning("Could not create directory for event %s\n",
2157                                    file->event_call->name);
2158         }
2159 }
2160
2161 /*
2162  * For early boot up, the top trace array requires to have
2163  * a list of events that can be enabled. This must be done before
2164  * the filesystem is set up in order to allow events to be traced
2165  * early.
2166  */
2167 static __init void
2168 __trace_early_add_events(struct trace_array *tr)
2169 {
2170         struct ftrace_event_call *call;
2171         int ret;
2172
2173         list_for_each_entry(call, &ftrace_events, list) {
2174                 /* Early boot up should not have any modules loaded */
2175                 if (WARN_ON_ONCE(call->mod))
2176                         continue;
2177
2178                 ret = __trace_early_add_new_event(call, tr);
2179                 if (ret < 0)
2180                         pr_warning("Could not create early event %s\n",
2181                                    call->name);
2182         }
2183 }
2184
2185 /* Remove the event directory structure for a trace directory. */
2186 static void
2187 __trace_remove_event_dirs(struct trace_array *tr)
2188 {
2189         struct ftrace_event_file *file, *next;
2190
2191         list_for_each_entry_safe(file, next, &tr->events, list) {
2192                 list_del(&file->list);
2193                 debugfs_remove_recursive(file->dir);
2194                 remove_subsystem(file->system);
2195                 kmem_cache_free(file_cachep, file);
2196         }
2197 }
2198
2199 static void
2200 __add_event_to_tracers(struct ftrace_event_call *call,
2201                        struct ftrace_module_file_ops *file_ops)
2202 {
2203         struct trace_array *tr;
2204
2205         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2206                 if (file_ops)
2207                         __trace_add_new_mod_event(call, tr, file_ops);
2208                 else
2209                         __trace_add_new_event(call, tr,
2210                                               &ftrace_event_id_fops,
2211                                               &ftrace_enable_fops,
2212                                               &ftrace_event_filter_fops,
2213                                               &ftrace_event_format_fops);
2214         }
2215 }
2216
2217 static struct notifier_block trace_module_nb = {
2218         .notifier_call = trace_module_notify,
2219         .priority = 0,
2220 };
2221
2222 extern struct ftrace_event_call *__start_ftrace_events[];
2223 extern struct ftrace_event_call *__stop_ftrace_events[];
2224
2225 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2226
2227 static __init int setup_trace_event(char *str)
2228 {
2229         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2230         ring_buffer_expanded = true;
2231         tracing_selftest_disabled = true;
2232
2233         return 1;
2234 }
2235 __setup("trace_event=", setup_trace_event);
2236
2237 /* Expects to have event_mutex held when called */
2238 static int
2239 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2240 {
2241         struct dentry *d_events;
2242         struct dentry *entry;
2243
2244         entry = debugfs_create_file("set_event", 0644, parent,
2245                                     tr, &ftrace_set_event_fops);
2246         if (!entry) {
2247                 pr_warning("Could not create debugfs 'set_event' entry\n");
2248                 return -ENOMEM;
2249         }
2250
2251         d_events = debugfs_create_dir("events", parent);
2252         if (!d_events) {
2253                 pr_warning("Could not create debugfs 'events' directory\n");
2254                 return -ENOMEM;
2255         }
2256
2257         /* ring buffer internal formats */
2258         trace_create_file("header_page", 0444, d_events,
2259                           ring_buffer_print_page_header,
2260                           &ftrace_show_header_fops);
2261
2262         trace_create_file("header_event", 0444, d_events,
2263                           ring_buffer_print_entry_header,
2264                           &ftrace_show_header_fops);
2265
2266         trace_create_file("enable", 0644, d_events,
2267                           tr, &ftrace_tr_enable_fops);
2268
2269         tr->event_dir = d_events;
2270
2271         return 0;
2272 }
2273
2274 /**
2275  * event_trace_add_tracer - add a instance of a trace_array to events
2276  * @parent: The parent dentry to place the files/directories for events in
2277  * @tr: The trace array associated with these events
2278  *
2279  * When a new instance is created, it needs to set up its events
2280  * directory, as well as other files associated with events. It also
2281  * creates the event hierachry in the @parent/events directory.
2282  *
2283  * Returns 0 on success.
2284  */
2285 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2286 {
2287         int ret;
2288
2289         mutex_lock(&event_mutex);
2290
2291         ret = create_event_toplevel_files(parent, tr);
2292         if (ret)
2293                 goto out_unlock;
2294
2295         down_write(&trace_event_sem);
2296         __trace_add_event_dirs(tr);
2297         up_write(&trace_event_sem);
2298
2299  out_unlock:
2300         mutex_unlock(&event_mutex);
2301
2302         return ret;
2303 }
2304
2305 /*
2306  * The top trace array already had its file descriptors created.
2307  * Now the files themselves need to be created.
2308  */
2309 static __init int
2310 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2311 {
2312         int ret;
2313
2314         mutex_lock(&event_mutex);
2315
2316         ret = create_event_toplevel_files(parent, tr);
2317         if (ret)
2318                 goto out_unlock;
2319
2320         down_write(&trace_event_sem);
2321         __trace_early_add_event_dirs(tr);
2322         up_write(&trace_event_sem);
2323
2324  out_unlock:
2325         mutex_unlock(&event_mutex);
2326
2327         return ret;
2328 }
2329
2330 int event_trace_del_tracer(struct trace_array *tr)
2331 {
2332         /* Disable any running events */
2333         __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2334
2335         mutex_lock(&event_mutex);
2336
2337         down_write(&trace_event_sem);
2338         __trace_remove_event_dirs(tr);
2339         debugfs_remove_recursive(tr->event_dir);
2340         up_write(&trace_event_sem);
2341
2342         tr->event_dir = NULL;
2343
2344         mutex_unlock(&event_mutex);
2345
2346         return 0;
2347 }
2348
2349 static __init int event_trace_memsetup(void)
2350 {
2351         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2352         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2353         return 0;
2354 }
2355
2356 static __init int event_trace_enable(void)
2357 {
2358         struct trace_array *tr = top_trace_array();
2359         struct ftrace_event_call **iter, *call;
2360         char *buf = bootup_event_buf;
2361         char *token;
2362         int ret;
2363
2364         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2365
2366                 call = *iter;
2367                 ret = event_init(call);
2368                 if (!ret)
2369                         list_add(&call->list, &ftrace_events);
2370         }
2371
2372         /*
2373          * We need the top trace array to have a working set of trace
2374          * points at early init, before the debug files and directories
2375          * are created. Create the file entries now, and attach them
2376          * to the actual file dentries later.
2377          */
2378         __trace_early_add_events(tr);
2379
2380         while (true) {
2381                 token = strsep(&buf, ",");
2382
2383                 if (!token)
2384                         break;
2385                 if (!*token)
2386                         continue;
2387
2388                 ret = ftrace_set_clr_event(tr, token, 1);
2389                 if (ret)
2390                         pr_warn("Failed to enable trace event: %s\n", token);
2391         }
2392
2393         trace_printk_start_comm();
2394
2395         register_event_cmds();
2396
2397         return 0;
2398 }
2399
2400 static __init int event_trace_init(void)
2401 {
2402         struct trace_array *tr;
2403         struct dentry *d_tracer;
2404         struct dentry *entry;
2405         int ret;
2406
2407         tr = top_trace_array();
2408
2409         d_tracer = tracing_init_dentry();
2410         if (!d_tracer)
2411                 return 0;
2412
2413         entry = debugfs_create_file("available_events", 0444, d_tracer,
2414                                     tr, &ftrace_avail_fops);
2415         if (!entry)
2416                 pr_warning("Could not create debugfs "
2417                            "'available_events' entry\n");
2418
2419         if (trace_define_common_fields())
2420                 pr_warning("tracing: Failed to allocate common fields");
2421
2422         ret = early_event_add_tracer(d_tracer, tr);
2423         if (ret)
2424                 return ret;
2425
2426         ret = register_module_notifier(&trace_module_nb);
2427         if (ret)
2428                 pr_warning("Failed to register trace events module notifier\n");
2429
2430         return 0;
2431 }
2432 early_initcall(event_trace_memsetup);
2433 core_initcall(event_trace_enable);
2434 fs_initcall(event_trace_init);
2435
2436 #ifdef CONFIG_FTRACE_STARTUP_TEST
2437
2438 static DEFINE_SPINLOCK(test_spinlock);
2439 static DEFINE_SPINLOCK(test_spinlock_irq);
2440 static DEFINE_MUTEX(test_mutex);
2441
2442 static __init void test_work(struct work_struct *dummy)
2443 {
2444         spin_lock(&test_spinlock);
2445         spin_lock_irq(&test_spinlock_irq);
2446         udelay(1);
2447         spin_unlock_irq(&test_spinlock_irq);
2448         spin_unlock(&test_spinlock);
2449
2450         mutex_lock(&test_mutex);
2451         msleep(1);
2452         mutex_unlock(&test_mutex);
2453 }
2454
2455 static __init int event_test_thread(void *unused)
2456 {
2457         void *test_malloc;
2458
2459         test_malloc = kmalloc(1234, GFP_KERNEL);
2460         if (!test_malloc)
2461                 pr_info("failed to kmalloc\n");
2462
2463         schedule_on_each_cpu(test_work);
2464
2465         kfree(test_malloc);
2466
2467         set_current_state(TASK_INTERRUPTIBLE);
2468         while (!kthread_should_stop())
2469                 schedule();
2470
2471         return 0;
2472 }
2473
2474 /*
2475  * Do various things that may trigger events.
2476  */
2477 static __init void event_test_stuff(void)
2478 {
2479         struct task_struct *test_thread;
2480
2481         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2482         msleep(1);
2483         kthread_stop(test_thread);
2484 }
2485
2486 /*
2487  * For every trace event defined, we will test each trace point separately,
2488  * and then by groups, and finally all trace points.
2489  */
2490 static __init void event_trace_self_tests(void)
2491 {
2492         struct ftrace_subsystem_dir *dir;
2493         struct ftrace_event_file *file;
2494         struct ftrace_event_call *call;
2495         struct event_subsystem *system;
2496         struct trace_array *tr;
2497         int ret;
2498
2499         tr = top_trace_array();
2500
2501         pr_info("Running tests on trace events:\n");
2502
2503         list_for_each_entry(file, &tr->events, list) {
2504
2505                 call = file->event_call;
2506
2507                 /* Only test those that have a probe */
2508                 if (!call->class || !call->class->probe)
2509                         continue;
2510
2511 /*
2512  * Testing syscall events here is pretty useless, but
2513  * we still do it if configured. But this is time consuming.
2514  * What we really need is a user thread to perform the
2515  * syscalls as we test.
2516  */
2517 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2518                 if (call->class->system &&
2519                     strcmp(call->class->system, "syscalls") == 0)
2520                         continue;
2521 #endif
2522
2523                 pr_info("Testing event %s: ", call->name);
2524
2525                 /*
2526                  * If an event is already enabled, someone is using
2527                  * it and the self test should not be on.
2528                  */
2529                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2530                         pr_warning("Enabled event during self test!\n");
2531                         WARN_ON_ONCE(1);
2532                         continue;
2533                 }
2534
2535                 ftrace_event_enable_disable(file, 1);
2536                 event_test_stuff();
2537                 ftrace_event_enable_disable(file, 0);
2538
2539                 pr_cont("OK\n");
2540         }
2541
2542         /* Now test at the sub system level */
2543
2544         pr_info("Running tests on trace event systems:\n");
2545
2546         list_for_each_entry(dir, &tr->systems, list) {
2547
2548                 system = dir->subsystem;
2549
2550                 /* the ftrace system is special, skip it */
2551                 if (strcmp(system->name, "ftrace") == 0)
2552                         continue;
2553
2554                 pr_info("Testing event system %s: ", system->name);
2555
2556                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2557                 if (WARN_ON_ONCE(ret)) {
2558                         pr_warning("error enabling system %s\n",
2559                                    system->name);
2560                         continue;
2561                 }
2562
2563                 event_test_stuff();
2564
2565                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2566                 if (WARN_ON_ONCE(ret)) {
2567                         pr_warning("error disabling system %s\n",
2568                                    system->name);
2569                         continue;
2570                 }
2571
2572                 pr_cont("OK\n");
2573         }
2574
2575         /* Test with all events enabled */
2576
2577         pr_info("Running tests on all trace events:\n");
2578         pr_info("Testing all events: ");
2579
2580         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2581         if (WARN_ON_ONCE(ret)) {
2582                 pr_warning("error enabling all events\n");
2583                 return;
2584         }
2585
2586         event_test_stuff();
2587
2588         /* reset sysname */
2589         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2590         if (WARN_ON_ONCE(ret)) {
2591                 pr_warning("error disabling all events\n");
2592                 return;
2593         }
2594
2595         pr_cont("OK\n");
2596 }
2597
2598 #ifdef CONFIG_FUNCTION_TRACER
2599
2600 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2601
2602 static void
2603 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2604                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2605 {
2606         struct ring_buffer_event *event;
2607         struct ring_buffer *buffer;
2608         struct ftrace_entry *entry;
2609         unsigned long flags;
2610         long disabled;
2611         int cpu;
2612         int pc;
2613
2614         pc = preempt_count();
2615         preempt_disable_notrace();
2616         cpu = raw_smp_processor_id();
2617         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2618
2619         if (disabled != 1)
2620                 goto out;
2621
2622         local_save_flags(flags);
2623
2624         event = trace_current_buffer_lock_reserve(&buffer,
2625                                                   TRACE_FN, sizeof(*entry),
2626                                                   flags, pc);
2627         if (!event)
2628                 goto out;
2629         entry   = ring_buffer_event_data(event);
2630         entry->ip                       = ip;
2631         entry->parent_ip                = parent_ip;
2632
2633         trace_buffer_unlock_commit(buffer, event, flags, pc);
2634
2635  out:
2636         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2637         preempt_enable_notrace();
2638 }
2639
2640 static struct ftrace_ops trace_ops __initdata  =
2641 {
2642         .func = function_test_events_call,
2643         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2644 };
2645
2646 static __init void event_trace_self_test_with_function(void)
2647 {
2648         int ret;
2649         ret = register_ftrace_function(&trace_ops);
2650         if (WARN_ON(ret < 0)) {
2651                 pr_info("Failed to enable function tracer for event tests\n");
2652                 return;
2653         }
2654         pr_info("Running tests again, along with the function tracer\n");
2655         event_trace_self_tests();
2656         unregister_ftrace_function(&trace_ops);
2657 }
2658 #else
2659 static __init void event_trace_self_test_with_function(void)
2660 {
2661 }
2662 #endif
2663
2664 static __init int event_trace_self_tests_init(void)
2665 {
2666         if (!tracing_selftest_disabled) {
2667                 event_trace_self_tests();
2668                 event_trace_self_test_with_function();
2669         }
2670
2671         return 0;
2672 }
2673
2674 late_initcall(event_trace_self_tests_init);
2675
2676 #endif