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 #define SYSTEM_FL_FREE_NAME             (1 << 31)
  45
  46 static inline int system_refcount(struct event_subsystem *system)
  47 {
  48         return system->ref_count & ~SYSTEM_FL_FREE_NAME;
  49 }
  50
  51 static int system_refcount_inc(struct event_subsystem *system)
  52 {
  53         return (system->ref_count++) & ~SYSTEM_FL_FREE_NAME;
  54 }
  55
  56 static int system_refcount_dec(struct event_subsystem *system)
  57 {
  58         return (--system->ref_count) & ~SYSTEM_FL_FREE_NAME;
  59 }
  60
  61 /* Double loops, do not use break, only goto's work */
  62 #define do_for_each_event_file(tr, file)                        \
  63         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  64                 list_for_each_entry(file, &tr->events, list)
  65
  66 #define do_for_each_event_file_safe(tr, file)                   \
  67         list_for_each_entry(tr, &ftrace_trace_arrays, list) {   \
  68                 struct ftrace_event_file *___n;                         \
  69                 list_for_each_entry_safe(file, ___n, &tr->events, list)
  70
  71 #define while_for_each_event_file()             \
  72         }
  73
  74 static struct list_head *
  75 trace_get_fields(struct ftrace_event_call *event_call)
  76 {
  77         if (!event_call->class->get_fields)
  78                 return &event_call->class->fields;
  79         return event_call->class->get_fields(event_call);
  80 }
  81
  82 static struct ftrace_event_field *
  83 __find_event_field(struct list_head *head, char *name)
  84 {
  85         struct ftrace_event_field *field;
  86
  87         list_for_each_entry(field, head, link) {
  88                 if (!strcmp(field->name, name))
  89                         return field;
  90         }
  91
  92         return NULL;
  93 }
  94
  95 struct ftrace_event_field *
  96 trace_find_event_field(struct ftrace_event_call *call, char *name)
  97 {
  98         struct ftrace_event_field *field;
  99         struct list_head *head;
 100
 101         field = __find_event_field(&ftrace_common_fields, name);
 102         if (field)
 103                 return field;
 104
 105         head = trace_get_fields(call);
 106         return __find_event_field(head, name);
 107 }
 108
 109 static int __trace_define_field(struct list_head *head, const char *type,
 110                                 const char *name, int offset, int size,
 111                                 int is_signed, int filter_type)
 112 {
 113         struct ftrace_event_field *field;
 114
 115         field = kmem_cache_alloc(field_cachep, GFP_TRACE);
 116         if (!field)
 117                 return -ENOMEM;
 118
 119         field->name = name;
 120         field->type = type;
 121
 122         if (filter_type == FILTER_OTHER)
 123                 field->filter_type = filter_assign_type(type);
 124         else
 125                 field->filter_type = filter_type;
 126
 127         field->offset = offset;
 128         field->size = size;
 129         field->is_signed = is_signed;
 130
 131         list_add(&field->link, head);
 132
 133         return 0;
 134 }
 135
 136 int trace_define_field(struct ftrace_event_call *call, const char *type,
 137                        const char *name, int offset, int size, int is_signed,
 138                        int filter_type)
 139 {
 140         struct list_head *head;
 141
 142         if (WARN_ON(!call->class))
 143                 return 0;
 144
 145         head = trace_get_fields(call);
 146         return __trace_define_field(head, type, name, offset, size,
 147                                     is_signed, filter_type);
 148 }
 149 EXPORT_SYMBOL_GPL(trace_define_field);
 150
 151 #define __common_field(type, item)                                      \
 152         ret = __trace_define_field(&ftrace_common_fields, #type,        \
 153                                    "common_" #item,                     \
 154                                    offsetof(typeof(ent), item),         \
 155                                    sizeof(ent.item),                    \
 156                                    is_signed_type(type), FILTER_OTHER); \
 157         if (ret)                                                        \
 158                 return ret;
 159
 160 static int trace_define_common_fields(void)
 161 {
 162         int ret;
 163         struct trace_entry ent;
 164
 165         __common_field(unsigned short, type);
 166         __common_field(unsigned char, flags);
 167         __common_field(unsigned char, preempt_count);
 168         __common_field(int, pid);
 169
 170         return ret;
 171 }
 172
 173 static void trace_destroy_fields(struct ftrace_event_call *call)
 174 {
 175         struct ftrace_event_field *field, *next;
 176         struct list_head *head;
 177
 178         head = trace_get_fields(call);
 179         list_for_each_entry_safe(field, next, head, link) {
 180                 list_del(&field->link);
 181                 kmem_cache_free(field_cachep, field);
 182         }
 183 }
 184
 185 int trace_event_raw_init(struct ftrace_event_call *call)
 186 {
 187         int id;
 188
 189         id = register_ftrace_event(&call->event);
 190         if (!id)
 191                 return -ENODEV;
 192
 193         return 0;
 194 }
 195 EXPORT_SYMBOL_GPL(trace_event_raw_init);
 196
 197 int ftrace_event_reg(struct ftrace_event_call *call,
 198                      enum trace_reg type, void *data)
 199 {
 200         struct ftrace_event_file *file = data;
 201
 202         switch (type) {
 203         case TRACE_REG_REGISTER:
 204                 return tracepoint_probe_register(call->name,
 205                                                  call->class->probe,
 206                                                  file);
 207         case TRACE_REG_UNREGISTER:
 208                 tracepoint_probe_unregister(call->name,
 209                                             call->class->probe,
 210                                             file);
 211                 return 0;
 212
 213 #ifdef CONFIG_PERF_EVENTS
 214         case TRACE_REG_PERF_REGISTER:
 215                 return tracepoint_probe_register(call->name,
 216                                                  call->class->perf_probe,
 217                                                  call);
 218         case TRACE_REG_PERF_UNREGISTER:
 219                 tracepoint_probe_unregister(call->name,
 220                                             call->class->perf_probe,
 221                                             call);
 222                 return 0;
 223         case TRACE_REG_PERF_OPEN:
 224         case TRACE_REG_PERF_CLOSE:
 225         case TRACE_REG_PERF_ADD:
 226         case TRACE_REG_PERF_DEL:
 227                 return 0;
 228 #endif
 229         }
 230         return 0;
 231 }
 232 EXPORT_SYMBOL_GPL(ftrace_event_reg);
 233
 234 void trace_event_enable_cmd_record(bool enable)
 235 {
 236         struct ftrace_event_file *file;
 237         struct trace_array *tr;
 238
 239         mutex_lock(&event_mutex);
 240         do_for_each_event_file(tr, file) {
 241
 242                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED))
 243                         continue;
 244
 245                 if (enable) {
 246                         tracing_start_cmdline_record();
 247                         set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 248                 } else {
 249                         tracing_stop_cmdline_record();
 250                         clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 251                 }
 252         } while_for_each_event_file();
 253         mutex_unlock(&event_mutex);
 254 }
 255
 256 static int __ftrace_event_enable_disable(struct ftrace_event_file *file,
 257                                          int enable, int soft_disable)
 258 {
 259         struct ftrace_event_call *call = file->event_call;
 260         int ret = 0;
 261         int disable;
 262
 263         switch (enable) {
 264         case 0:
 265                 /*
 266                  * When soft_disable is set and enable is cleared, the sm_ref
 267                  * reference counter is decremented. If it reaches 0, we want
 268                  * to clear the SOFT_DISABLED flag but leave the event in the
 269                  * state that it was. That is, if the event was enabled and
 270                  * SOFT_DISABLED isn't set, then do nothing. But if SOFT_DISABLED
 271                  * is set we do not want the event to be enabled before we
 272                  * clear the bit.
 273                  *
 274                  * When soft_disable is not set but the SOFT_MODE flag is,
 275                  * we do nothing. Do not disable the tracepoint, otherwise
 276                  * "soft enable"s (clearing the SOFT_DISABLED bit) wont work.
 277                  */
 278                 if (soft_disable) {
 279                         if (atomic_dec_return(&file->sm_ref) > 0)
 280                                 break;
 281                         disable = file->flags & FTRACE_EVENT_FL_SOFT_DISABLED;
 282                         clear_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 283                 } else
 284                         disable = !(file->flags & FTRACE_EVENT_FL_SOFT_MODE);
 285
 286                 if (disable && (file->flags & FTRACE_EVENT_FL_ENABLED)) {
 287                         clear_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 288                         if (file->flags & FTRACE_EVENT_FL_RECORDED_CMD) {
 289                                 tracing_stop_cmdline_record();
 290                                 clear_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 291                         }
 292                         call->class->reg(call, TRACE_REG_UNREGISTER, file);
 293                 }
 294                 /* If in SOFT_MODE, just set the SOFT_DISABLE_BIT, else clear it */
 295                 if (file->flags & FTRACE_EVENT_FL_SOFT_MODE)
 296                         set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 297                 else
 298                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 299                 break;
 300         case 1:
 301                 /*
 302                  * When soft_disable is set and enable is set, we want to
 303                  * register the tracepoint for the event, but leave the event
 304                  * as is. That means, if the event was already enabled, we do
 305                  * nothing (but set SOFT_MODE). If the event is disabled, we
 306                  * set SOFT_DISABLED before enabling the event tracepoint, so
 307                  * it still seems to be disabled.
 308                  */
 309                 if (!soft_disable)
 310                         clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 311                 else {
 312                         if (atomic_inc_return(&file->sm_ref) > 1)
 313                                 break;
 314                         set_bit(FTRACE_EVENT_FL_SOFT_MODE_BIT, &file->flags);
 315                 }
 316
 317                 if (!(file->flags & FTRACE_EVENT_FL_ENABLED)) {
 318
 319                         /* Keep the event disabled, when going to SOFT_MODE. */
 320                         if (soft_disable)
 321                                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags);
 322
 323                         if (trace_flags & TRACE_ITER_RECORD_CMD) {
 324                                 tracing_start_cmdline_record();
 325                                 set_bit(FTRACE_EVENT_FL_RECORDED_CMD_BIT, &file->flags);
 326                         }
 327                         ret = call->class->reg(call, TRACE_REG_REGISTER, file);
 328                         if (ret) {
 329                                 tracing_stop_cmdline_record();
 330                                 pr_info("event trace: Could not enable event "
 331                                         "%s\n", call->name);
 332                                 break;
 333                         }
 334                         set_bit(FTRACE_EVENT_FL_ENABLED_BIT, &file->flags);
 335
 336                         /* WAS_ENABLED gets set but never cleared. */
 337                         call->flags |= TRACE_EVENT_FL_WAS_ENABLED;
 338                 }
 339                 break;
 340         }
 341
 342         return ret;
 343 }
 344
 345 static int ftrace_event_enable_disable(struct ftrace_event_file *file,
 346                                        int enable)
 347 {
 348         return __ftrace_event_enable_disable(file, enable, 0);
 349 }
 350
 351 static void ftrace_clear_events(struct trace_array *tr)
 352 {
 353         struct ftrace_event_file *file;
 354
 355         mutex_lock(&event_mutex);
 356         list_for_each_entry(file, &tr->events, list) {
 357                 ftrace_event_enable_disable(file, 0);
 358         }
 359         mutex_unlock(&event_mutex);
 360 }
 361
 362 static void __put_system(struct event_subsystem *system)
 363 {
 364         struct event_filter *filter = system->filter;
 365
 366         WARN_ON_ONCE(system_refcount(system) == 0);
 367         if (system_refcount_dec(system))
 368                 return;
 369
 370         list_del(&system->list);
 371
 372         if (filter) {
 373                 kfree(filter->filter_string);
 374                 kfree(filter);
 375         }
 376         if (system->ref_count & SYSTEM_FL_FREE_NAME)
 377                 kfree(system->name);
 378         kfree(system);
 379 }
 380
 381 static void __get_system(struct event_subsystem *system)
 382 {
 383         WARN_ON_ONCE(system_refcount(system) == 0);
 384         system_refcount_inc(system);
 385 }
 386
 387 static void __get_system_dir(struct ftrace_subsystem_dir *dir)
 388 {
 389         WARN_ON_ONCE(dir->ref_count == 0);
 390         dir->ref_count++;
 391         __get_system(dir->subsystem);
 392 }
 393
 394 static void __put_system_dir(struct ftrace_subsystem_dir *dir)
 395 {
 396         WARN_ON_ONCE(dir->ref_count == 0);
 397         /* If the subsystem is about to be freed, the dir must be too */
 398         WARN_ON_ONCE(system_refcount(dir->subsystem) == 1 && dir->ref_count != 1);
 399
 400         __put_system(dir->subsystem);
 401         if (!--dir->ref_count)
 402                 kfree(dir);
 403 }
 404
 405 static void put_system(struct ftrace_subsystem_dir *dir)
 406 {
 407         mutex_lock(&event_mutex);
 408         __put_system_dir(dir);
 409         mutex_unlock(&event_mutex);
 410 }
 411
 412 static void remove_subsystem(struct ftrace_subsystem_dir *dir)
 413 {
 414         if (!dir)
 415                 return;
 416
 417         if (!--dir->nr_events) {
 418                 debugfs_remove_recursive(dir->entry);
 419                 list_del(&dir->list);
 420                 __put_system_dir(dir);
 421         }
 422 }
 423
 424 static void *event_file_data(struct file *filp)
 425 {
 426         return ACCESS_ONCE(file_inode(filp)->i_private);
 427 }
 428
 429 static void remove_event_file_dir(struct ftrace_event_file *file)
 430 {
 431         struct dentry *dir = file->dir;
 432         struct dentry *child;
 433
 434         if (dir) {
 435                 spin_lock(&dir->d_lock);        /* probably unneeded */
 436                 list_for_each_entry(child, &dir->d_subdirs, d_u.d_child) {
 437                         if (child->d_inode)     /* probably unneeded */
 438                                 child->d_inode->i_private = NULL;
 439                 }
 440                 spin_unlock(&dir->d_lock);
 441
 442                 debugfs_remove_recursive(dir);
 443         }
 444
 445         list_del(&file->list);
 446         remove_subsystem(file->system);
 447         kmem_cache_free(file_cachep, file);
 448 }
 449
 450 /*
 451  * __ftrace_set_clr_event(NULL, NULL, NULL, set) will set/unset all events.
 452  */
 453 static int
 454 __ftrace_set_clr_event_nolock(struct trace_array *tr, const char *match,
 455                               const char *sub, const char *event, int set)
 456 {
 457         struct ftrace_event_file *file;
 458         struct ftrace_event_call *call;
 459         int ret = -EINVAL;
 460
 461         list_for_each_entry(file, &tr->events, list) {
 462
 463                 call = file->event_call;
 464
 465                 if (!call->name || !call->class || !call->class->reg)
 466                         continue;
 467
 468                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
 469                         continue;
 470
 471                 if (match &&
 472                     strcmp(match, call->name) != 0 &&
 473                     strcmp(match, call->class->system) != 0)
 474                         continue;
 475
 476                 if (sub && strcmp(sub, call->class->system) != 0)
 477                         continue;
 478
 479                 if (event && strcmp(event, call->name) != 0)
 480                         continue;
 481
 482                 ftrace_event_enable_disable(file, set);
 483
 484                 ret = 0;
 485         }
 486
 487         return ret;
 488 }
 489
 490 static int __ftrace_set_clr_event(struct trace_array *tr, const char *match,
 491                                   const char *sub, const char *event, int set)
 492 {
 493         int ret;
 494
 495         mutex_lock(&event_mutex);
 496         ret = __ftrace_set_clr_event_nolock(tr, match, sub, event, set);
 497         mutex_unlock(&event_mutex);
 498
 499         return ret;
 500 }
 501
 502 static int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set)
 503 {
 504         char *event = NULL, *sub = NULL, *match;
 505
 506         /*
 507          * The buf format can be <subsystem>:<event-name>
 508          *  *:<event-name> means any event by that name.
 509          *  :<event-name> is the same.
 510          *
 511          *  <subsystem>:* means all events in that subsystem
 512          *  <subsystem>: means the same.
 513          *
 514          *  <name> (no ':') means all events in a subsystem with
 515          *  the name <name> or any event that matches <name>
 516          */
 517
 518         match = strsep(&buf, ":");
 519         if (buf) {
 520                 sub = match;
 521                 event = buf;
 522                 match = NULL;
 523
 524                 if (!strlen(sub) || strcmp(sub, "*") == 0)
 525                         sub = NULL;
 526                 if (!strlen(event) || strcmp(event, "*") == 0)
 527                         event = NULL;
 528         }
 529
 530         return __ftrace_set_clr_event(tr, match, sub, event, set);
 531 }
 532
 533 /**
 534  * trace_set_clr_event - enable or disable an event
 535  * @system: system name to match (NULL for any system)
 536  * @event: event name to match (NULL for all events, within system)
 537  * @set: 1 to enable, 0 to disable
 538  *
 539  * This is a way for other parts of the kernel to enable or disable
 540  * event recording.
 541  *
 542  * Returns 0 on success, -EINVAL if the parameters do not match any
 543  * registered events.
 544  */
 545 int trace_set_clr_event(const char *system, const char *event, int set)
 546 {
 547         struct trace_array *tr = top_trace_array();
 548
 549         return __ftrace_set_clr_event(tr, NULL, system, event, set);
 550 }
 551 EXPORT_SYMBOL_GPL(trace_set_clr_event);
 552
 553 /* 128 should be much more than enough */
 554 #define EVENT_BUF_SIZE          127
 555
 556 static ssize_t
 557 ftrace_event_write(struct file *file, const char __user *ubuf,
 558                    size_t cnt, loff_t *ppos)
 559 {
 560         struct trace_parser parser;
 561         struct seq_file *m = file->private_data;
 562         struct trace_array *tr = m->private;
 563         ssize_t read, ret;
 564
 565         if (!cnt)
 566                 return 0;
 567
 568         ret = tracing_update_buffers();
 569         if (ret < 0)
 570                 return ret;
 571
 572         if (trace_parser_get_init(&parser, EVENT_BUF_SIZE + 1))
 573                 return -ENOMEM;
 574
 575         read = trace_get_user(&parser, ubuf, cnt, ppos);
 576
 577         if (read >= 0 && trace_parser_loaded((&parser))) {
 578                 int set = 1;
 579
 580                 if (*parser.buffer == '!')
 581                         set = 0;
 582
 583                 parser.buffer[parser.idx] = 0;
 584
 585                 ret = ftrace_set_clr_event(tr, parser.buffer + !set, set);
 586                 if (ret)
 587                         goto out_put;
 588         }
 589
 590         ret = read;
 591
 592  out_put:
 593         trace_parser_put(&parser);
 594
 595         return ret;
 596 }
 597
 598 static void *
 599 t_next(struct seq_file *m, void *v, loff_t *pos)
 600 {
 601         struct ftrace_event_file *file = v;
 602         struct ftrace_event_call *call;
 603         struct trace_array *tr = m->private;
 604
 605         (*pos)++;
 606
 607         list_for_each_entry_continue(file, &tr->events, list) {
 608                 call = file->event_call;
 609                 /*
 610                  * The ftrace subsystem is for showing formats only.
 611                  * They can not be enabled or disabled via the event files.
 612                  */
 613                 if (call->class && call->class->reg)
 614                         return file;
 615         }
 616
 617         return NULL;
 618 }
 619
 620 static void *t_start(struct seq_file *m, loff_t *pos)
 621 {
 622         struct ftrace_event_file *file;
 623         struct trace_array *tr = m->private;
 624         loff_t l;
 625
 626         mutex_lock(&event_mutex);
 627
 628         file = list_entry(&tr->events, struct ftrace_event_file, list);
 629         for (l = 0; l <= *pos; ) {
 630                 file = t_next(m, file, &l);
 631                 if (!file)
 632                         break;
 633         }
 634         return file;
 635 }
 636
 637 static void *
 638 s_next(struct seq_file *m, void *v, loff_t *pos)
 639 {
 640         struct ftrace_event_file *file = v;
 641         struct trace_array *tr = m->private;
 642
 643         (*pos)++;
 644
 645         list_for_each_entry_continue(file, &tr->events, list) {
 646                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
 647                         return file;
 648         }
 649
 650         return NULL;
 651 }
 652
 653 static void *s_start(struct seq_file *m, loff_t *pos)
 654 {
 655         struct ftrace_event_file *file;
 656         struct trace_array *tr = m->private;
 657         loff_t l;
 658
 659         mutex_lock(&event_mutex);
 660
 661         file = list_entry(&tr->events, struct ftrace_event_file, list);
 662         for (l = 0; l <= *pos; ) {
 663                 file = s_next(m, file, &l);
 664                 if (!file)
 665                         break;
 666         }
 667         return file;
 668 }
 669
 670 static int t_show(struct seq_file *m, void *v)
 671 {
 672         struct ftrace_event_file *file = v;
 673         struct ftrace_event_call *call = file->event_call;
 674
 675         if (strcmp(call->class->system, TRACE_SYSTEM) != 0)
 676                 seq_printf(m, "%s:", call->class->system);
 677         seq_printf(m, "%s\n", call->name);
 678
 679         return 0;
 680 }
 681
 682 static void t_stop(struct seq_file *m, void *p)
 683 {
 684         mutex_unlock(&event_mutex);
 685 }
 686
 687 static ssize_t
 688 event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 689                   loff_t *ppos)
 690 {
 691         struct ftrace_event_file *file;
 692         unsigned long flags;
 693         char buf[4] = "0";
 694
 695         mutex_lock(&event_mutex);
 696         file = event_file_data(filp);
 697         if (likely(file))
 698                 flags = file->flags;
 699         mutex_unlock(&event_mutex);
 700
 701         if (!file)
 702                 return -ENODEV;
 703
 704         if (flags & FTRACE_EVENT_FL_ENABLED &&
 705             !(flags & FTRACE_EVENT_FL_SOFT_DISABLED))
 706                 strcpy(buf, "1");
 707
 708         if (flags & FTRACE_EVENT_FL_SOFT_DISABLED ||
 709             flags & FTRACE_EVENT_FL_SOFT_MODE)
 710                 strcat(buf, "*");
 711
 712         strcat(buf, "\n");
 713
 714         return simple_read_from_buffer(ubuf, cnt, ppos, buf, strlen(buf));
 715 }
 716
 717 static ssize_t
 718 event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 719                    loff_t *ppos)
 720 {
 721         struct ftrace_event_file *file;
 722         unsigned long val;
 723         int ret;
 724
 725         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 726         if (ret)
 727                 return ret;
 728
 729         ret = tracing_update_buffers();
 730         if (ret < 0)
 731                 return ret;
 732
 733         switch (val) {
 734         case 0:
 735         case 1:
 736                 ret = -ENODEV;
 737                 mutex_lock(&event_mutex);
 738                 file = event_file_data(filp);
 739                 if (likely(file))
 740                         ret = ftrace_event_enable_disable(file, val);
 741                 mutex_unlock(&event_mutex);
 742                 break;
 743
 744         default:
 745                 return -EINVAL;
 746         }
 747
 748         *ppos += cnt;
 749
 750         return ret ? ret : cnt;
 751 }
 752
 753 static ssize_t
 754 system_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
 755                    loff_t *ppos)
 756 {
 757         const char set_to_char[4] = { '?', '0', '1', 'X' };
 758         struct ftrace_subsystem_dir *dir = filp->private_data;
 759         struct event_subsystem *system = dir->subsystem;
 760         struct ftrace_event_call *call;
 761         struct ftrace_event_file *file;
 762         struct trace_array *tr = dir->tr;
 763         char buf[2];
 764         int set = 0;
 765         int ret;
 766
 767         mutex_lock(&event_mutex);
 768         list_for_each_entry(file, &tr->events, list) {
 769                 call = file->event_call;
 770                 if (!call->name || !call->class || !call->class->reg)
 771                         continue;
 772
 773                 if (system && strcmp(call->class->system, system->name) != 0)
 774                         continue;
 775
 776                 /*
 777                  * We need to find out if all the events are set
 778                  * or if all events or cleared, or if we have
 779                  * a mixture.
 780                  */
 781                 set |= (1 << !!(file->flags & FTRACE_EVENT_FL_ENABLED));
 782
 783                 /*
 784                  * If we have a mixture, no need to look further.
 785                  */
 786                 if (set == 3)
 787                         break;
 788         }
 789         mutex_unlock(&event_mutex);
 790
 791         buf[0] = set_to_char[set];
 792         buf[1] = '\n';
 793
 794         ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
 795
 796         return ret;
 797 }
 798
 799 static ssize_t
 800 system_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
 801                     loff_t *ppos)
 802 {
 803         struct ftrace_subsystem_dir *dir = filp->private_data;
 804         struct event_subsystem *system = dir->subsystem;
 805         const char *name = NULL;
 806         unsigned long val;
 807         ssize_t ret;
 808
 809         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 810         if (ret)
 811                 return ret;
 812
 813         ret = tracing_update_buffers();
 814         if (ret < 0)
 815                 return ret;
 816
 817         if (val != 0 && val != 1)
 818                 return -EINVAL;
 819
 820         /*
 821          * Opening of "enable" adds a ref count to system,
 822          * so the name is safe to use.
 823          */
 824         if (system)
 825                 name = system->name;
 826
 827         ret = __ftrace_set_clr_event(dir->tr, NULL, name, NULL, val);
 828         if (ret)
 829                 goto out;
 830
 831         ret = cnt;
 832
 833 out:
 834         *ppos += cnt;
 835
 836         return ret;
 837 }
 838
 839 enum {
 840         FORMAT_HEADER           = 1,
 841         FORMAT_FIELD_SEPERATOR  = 2,
 842         FORMAT_PRINTFMT         = 3,
 843 };
 844
 845 static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 846 {
 847         struct ftrace_event_call *call = event_file_data(m->private);
 848         struct list_head *common_head = &ftrace_common_fields;
 849         struct list_head *head = trace_get_fields(call);
 850         struct list_head *node = v;
 851
 852         (*pos)++;
 853
 854         switch ((unsigned long)v) {
 855         case FORMAT_HEADER:
 856                 node = common_head;
 857                 break;
 858
 859         case FORMAT_FIELD_SEPERATOR:
 860                 node = head;
 861                 break;
 862
 863         case FORMAT_PRINTFMT:
 864                 /* all done */
 865                 return NULL;
 866         }
 867
 868         node = node->prev;
 869         if (node == common_head)
 870                 return (void *)FORMAT_FIELD_SEPERATOR;
 871         else if (node == head)
 872                 return (void *)FORMAT_PRINTFMT;
 873         else
 874                 return node;
 875 }
 876
 877 static int f_show(struct seq_file *m, void *v)
 878 {
 879         struct ftrace_event_call *call = event_file_data(m->private);
 880         struct ftrace_event_field *field;
 881         const char *array_descriptor;
 882
 883         switch ((unsigned long)v) {
 884         case FORMAT_HEADER:
 885                 seq_printf(m, "name: %s\n", call->name);
 886                 seq_printf(m, "ID: %d\n", call->event.type);
 887                 seq_printf(m, "format:\n");
 888                 return 0;
 889
 890         case FORMAT_FIELD_SEPERATOR:
 891                 seq_putc(m, '\n');
 892                 return 0;
 893
 894         case FORMAT_PRINTFMT:
 895                 seq_printf(m, "\nprint fmt: %s\n",
 896                            call->print_fmt);
 897                 return 0;
 898         }
 899
 900         field = list_entry(v, struct ftrace_event_field, link);
 901         /*
 902          * Smartly shows the array type(except dynamic array).
 903          * Normal:
 904          *      field:TYPE VAR
 905          * If TYPE := TYPE[LEN], it is shown:
 906          *      field:TYPE VAR[LEN]
 907          */
 908         array_descriptor = strchr(field->type, '[');
 909
 910         if (!strncmp(field->type, "__data_loc", 10))
 911                 array_descriptor = NULL;
 912
 913         if (!array_descriptor)
 914                 seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 915                            field->type, field->name, field->offset,
 916                            field->size, !!field->is_signed);
 917         else
 918                 seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 919                            (int)(array_descriptor - field->type),
 920                            field->type, field->name,
 921                            array_descriptor, field->offset,
 922                            field->size, !!field->is_signed);
 923
 924         return 0;
 925 }
 926
 927 static void *f_start(struct seq_file *m, loff_t *pos)
 928 {
 929         void *p = (void *)FORMAT_HEADER;
 930         loff_t l = 0;
 931
 932         /* ->stop() is called even if ->start() fails */
 933         mutex_lock(&event_mutex);
 934         if (!event_file_data(m->private))
 935                 return ERR_PTR(-ENODEV);
 936
 937         while (l < *pos && p)
 938                 p = f_next(m, p, &l);
 939
 940         return p;
 941 }
 942
 943 static void f_stop(struct seq_file *m, void *p)
 944 {
 945         mutex_unlock(&event_mutex);
 946 }
 947
 948 static const struct seq_operations trace_format_seq_ops = {
 949         .start          = f_start,
 950         .next           = f_next,
 951         .stop           = f_stop,
 952         .show           = f_show,
 953 };
 954
 955 static int trace_format_open(struct inode *inode, struct file *file)
 956 {
 957         struct seq_file *m;
 958         int ret;
 959
 960         ret = seq_open(file, &trace_format_seq_ops);
 961         if (ret < 0)
 962                 return ret;
 963
 964         m = file->private_data;
 965         m->private = file;
 966
 967         return 0;
 968 }
 969
 970 static ssize_t
 971 event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
 972 {
 973         int id = (long)event_file_data(filp);
 974         char buf[32];
 975         int len;
 976
 977         if (*ppos)
 978                 return 0;
 979
 980         if (unlikely(!id))
 981                 return -ENODEV;
 982
 983         len = sprintf(buf, "%d\n", id);
 984
 985         return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
 986 }
 987
 988 static ssize_t
 989 event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 990                   loff_t *ppos)
 991 {
 992         struct ftrace_event_call *call;
 993         struct trace_seq *s;
 994         int r = -ENODEV;
 995
 996         if (*ppos)
 997                 return 0;
 998
 999         s = kmalloc(sizeof(*s), GFP_KERNEL);
1000
1001         if (!s)
1002                 return -ENOMEM;
1003
1004         trace_seq_init(s);
1005
1006         mutex_lock(&event_mutex);
1007         call = event_file_data(filp);
1008         if (call)
1009                 print_event_filter(call, s);
1010         mutex_unlock(&event_mutex);
1011
1012         if (call)
1013                 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1014
1015         kfree(s);
1016
1017         return r;
1018 }
1019
1020 static ssize_t
1021 event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1022                    loff_t *ppos)
1023 {
1024         struct ftrace_event_call *call;
1025         char *buf;
1026         int err = -ENODEV;
1027
1028         if (cnt >= PAGE_SIZE)
1029                 return -EINVAL;
1030
1031         buf = (char *)__get_free_page(GFP_TEMPORARY);
1032         if (!buf)
1033                 return -ENOMEM;
1034
1035         if (copy_from_user(buf, ubuf, cnt)) {
1036                 free_page((unsigned long) buf);
1037                 return -EFAULT;
1038         }
1039         buf[cnt] = '\0';
1040
1041         mutex_lock(&event_mutex);
1042         call = event_file_data(filp);
1043         if (call)
1044                 err = apply_event_filter(call, buf);
1045         mutex_unlock(&event_mutex);
1046
1047         free_page((unsigned long) buf);
1048         if (err < 0)
1049                 return err;
1050
1051         *ppos += cnt;
1052
1053         return cnt;
1054 }
1055
1056 static LIST_HEAD(event_subsystems);
1057
1058 static int subsystem_open(struct inode *inode, struct file *filp)
1059 {
1060         struct event_subsystem *system = NULL;
1061         struct ftrace_subsystem_dir *dir = NULL; /* Initialize for gcc */
1062         struct trace_array *tr;
1063         int ret;
1064
1065         /* Make sure the system still exists */
1066         mutex_lock(&trace_types_lock);
1067         mutex_lock(&event_mutex);
1068         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1069                 list_for_each_entry(dir, &tr->systems, list) {
1070                         if (dir == inode->i_private) {
1071                                 /* Don't open systems with no events */
1072                                 if (dir->nr_events) {
1073                                         __get_system_dir(dir);
1074                                         system = dir->subsystem;
1075                                 }
1076                                 goto exit_loop;
1077                         }
1078                 }
1079         }
1080  exit_loop:
1081         mutex_unlock(&event_mutex);
1082         mutex_unlock(&trace_types_lock);
1083
1084         if (!system)
1085                 return -ENODEV;
1086
1087         /* Some versions of gcc think dir can be uninitialized here */
1088         WARN_ON(!dir);
1089
1090         /* Still need to increment the ref count of the system */
1091         if (trace_array_get(tr) < 0) {
1092                 put_system(dir);
1093                 return -ENODEV;
1094         }
1095
1096         ret = tracing_open_generic(inode, filp);
1097         if (ret < 0) {
1098                 trace_array_put(tr);
1099                 put_system(dir);
1100         }
1101
1102         return ret;
1103 }
1104
1105 static int system_tr_open(struct inode *inode, struct file *filp)
1106 {
1107         struct ftrace_subsystem_dir *dir;
1108         struct trace_array *tr = inode->i_private;
1109         int ret;
1110
1111         if (trace_array_get(tr) < 0)
1112                 return -ENODEV;
1113
1114         /* Make a temporary dir that has no system but points to tr */
1115         dir = kzalloc(sizeof(*dir), GFP_KERNEL);
1116         if (!dir) {
1117                 trace_array_put(tr);
1118                 return -ENOMEM;
1119         }
1120
1121         dir->tr = tr;
1122
1123         ret = tracing_open_generic(inode, filp);
1124         if (ret < 0) {
1125                 trace_array_put(tr);
1126                 kfree(dir);
1127         }
1128
1129         filp->private_data = dir;
1130
1131         return ret;
1132 }
1133
1134 static int subsystem_release(struct inode *inode, struct file *file)
1135 {
1136         struct ftrace_subsystem_dir *dir = file->private_data;
1137
1138         trace_array_put(dir->tr);
1139
1140         /*
1141          * If dir->subsystem is NULL, then this is a temporary
1142          * descriptor that was made for a trace_array to enable
1143          * all subsystems.
1144          */
1145         if (dir->subsystem)
1146                 put_system(dir);
1147         else
1148                 kfree(dir);
1149
1150         return 0;
1151 }
1152
1153 static ssize_t
1154 subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
1155                       loff_t *ppos)
1156 {
1157         struct ftrace_subsystem_dir *dir = filp->private_data;
1158         struct event_subsystem *system = dir->subsystem;
1159         struct trace_seq *s;
1160         int r;
1161
1162         if (*ppos)
1163                 return 0;
1164
1165         s = kmalloc(sizeof(*s), GFP_KERNEL);
1166         if (!s)
1167                 return -ENOMEM;
1168
1169         trace_seq_init(s);
1170
1171         print_subsystem_event_filter(system, s);
1172         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1173
1174         kfree(s);
1175
1176         return r;
1177 }
1178
1179 static ssize_t
1180 subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
1181                        loff_t *ppos)
1182 {
1183         struct ftrace_subsystem_dir *dir = filp->private_data;
1184         char *buf;
1185         int err;
1186
1187         if (cnt >= PAGE_SIZE)
1188                 return -EINVAL;
1189
1190         buf = (char *)__get_free_page(GFP_TEMPORARY);
1191         if (!buf)
1192                 return -ENOMEM;
1193
1194         if (copy_from_user(buf, ubuf, cnt)) {
1195                 free_page((unsigned long) buf);
1196                 return -EFAULT;
1197         }
1198         buf[cnt] = '\0';
1199
1200         err = apply_subsystem_event_filter(dir, buf);
1201         free_page((unsigned long) buf);
1202         if (err < 0)
1203                 return err;
1204
1205         *ppos += cnt;
1206
1207         return cnt;
1208 }
1209
1210 static ssize_t
1211 show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
1212 {
1213         int (*func)(struct trace_seq *s) = filp->private_data;
1214         struct trace_seq *s;
1215         int r;
1216
1217         if (*ppos)
1218                 return 0;
1219
1220         s = kmalloc(sizeof(*s), GFP_KERNEL);
1221         if (!s)
1222                 return -ENOMEM;
1223
1224         trace_seq_init(s);
1225
1226         func(s);
1227         r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
1228
1229         kfree(s);
1230
1231         return r;
1232 }
1233
1234 static int ftrace_event_avail_open(struct inode *inode, struct file *file);
1235 static int ftrace_event_set_open(struct inode *inode, struct file *file);
1236 static int ftrace_event_release(struct inode *inode, struct file *file);
1237
1238 static const struct seq_operations show_event_seq_ops = {
1239         .start = t_start,
1240         .next = t_next,
1241         .show = t_show,
1242         .stop = t_stop,
1243 };
1244
1245 static const struct seq_operations show_set_event_seq_ops = {
1246         .start = s_start,
1247         .next = s_next,
1248         .show = t_show,
1249         .stop = t_stop,
1250 };
1251
1252 static const struct file_operations ftrace_avail_fops = {
1253         .open = ftrace_event_avail_open,
1254         .read = seq_read,
1255         .llseek = seq_lseek,
1256         .release = seq_release,
1257 };
1258
1259 static const struct file_operations ftrace_set_event_fops = {
1260         .open = ftrace_event_set_open,
1261         .read = seq_read,
1262         .write = ftrace_event_write,
1263         .llseek = seq_lseek,
1264         .release = ftrace_event_release,
1265 };
1266
1267 static const struct file_operations ftrace_enable_fops = {
1268         .open = tracing_open_generic,
1269         .read = event_enable_read,
1270         .write = event_enable_write,
1271         .llseek = default_llseek,
1272 };
1273
1274 static const struct file_operations ftrace_event_format_fops = {
1275         .open = trace_format_open,
1276         .read = seq_read,
1277         .llseek = seq_lseek,
1278         .release = seq_release,
1279 };
1280
1281 static const struct file_operations ftrace_event_id_fops = {
1282         .read = event_id_read,
1283         .llseek = default_llseek,
1284 };
1285
1286 static const struct file_operations ftrace_event_filter_fops = {
1287         .open = tracing_open_generic,
1288         .read = event_filter_read,
1289         .write = event_filter_write,
1290         .llseek = default_llseek,
1291 };
1292
1293 static const struct file_operations ftrace_subsystem_filter_fops = {
1294         .open = subsystem_open,
1295         .read = subsystem_filter_read,
1296         .write = subsystem_filter_write,
1297         .llseek = default_llseek,
1298         .release = subsystem_release,
1299 };
1300
1301 static const struct file_operations ftrace_system_enable_fops = {
1302         .open = subsystem_open,
1303         .read = system_enable_read,
1304         .write = system_enable_write,
1305         .llseek = default_llseek,
1306         .release = subsystem_release,
1307 };
1308
1309 static const struct file_operations ftrace_tr_enable_fops = {
1310         .open = system_tr_open,
1311         .read = system_enable_read,
1312         .write = system_enable_write,
1313         .llseek = default_llseek,
1314         .release = subsystem_release,
1315 };
1316
1317 static const struct file_operations ftrace_show_header_fops = {
1318         .open = tracing_open_generic,
1319         .read = show_header,
1320         .llseek = default_llseek,
1321 };
1322
1323 static int
1324 ftrace_event_open(struct inode *inode, struct file *file,
1325                   const struct seq_operations *seq_ops)
1326 {
1327         struct seq_file *m;
1328         int ret;
1329
1330         ret = seq_open(file, seq_ops);
1331         if (ret < 0)
1332                 return ret;
1333         m = file->private_data;
1334         /* copy tr over to seq ops */
1335         m->private = inode->i_private;
1336
1337         return ret;
1338 }
1339
1340 static int ftrace_event_release(struct inode *inode, struct file *file)
1341 {
1342         struct trace_array *tr = inode->i_private;
1343
1344         trace_array_put(tr);
1345
1346         return seq_release(inode, file);
1347 }
1348
1349 static int
1350 ftrace_event_avail_open(struct inode *inode, struct file *file)
1351 {
1352         const struct seq_operations *seq_ops = &show_event_seq_ops;
1353
1354         return ftrace_event_open(inode, file, seq_ops);
1355 }
1356
1357 static int
1358 ftrace_event_set_open(struct inode *inode, struct file *file)
1359 {
1360         const struct seq_operations *seq_ops = &show_set_event_seq_ops;
1361         struct trace_array *tr = inode->i_private;
1362         int ret;
1363
1364         if (trace_array_get(tr) < 0)
1365                 return -ENODEV;
1366
1367         if ((file->f_mode & FMODE_WRITE) &&
1368             (file->f_flags & O_TRUNC))
1369                 ftrace_clear_events(tr);
1370
1371         ret = ftrace_event_open(inode, file, seq_ops);
1372         if (ret < 0)
1373                 trace_array_put(tr);
1374         return ret;
1375 }
1376
1377 static struct event_subsystem *
1378 create_new_subsystem(const char *name)
1379 {
1380         struct event_subsystem *system;
1381
1382         /* need to create new entry */
1383         system = kmalloc(sizeof(*system), GFP_KERNEL);
1384         if (!system)
1385                 return NULL;
1386
1387         system->ref_count = 1;
1388
1389         /* Only allocate if dynamic (kprobes and modules) */
1390         if (!core_kernel_data((unsigned long)name)) {
1391                 system->ref_count |= SYSTEM_FL_FREE_NAME;
1392                 system->name = kstrdup(name, GFP_KERNEL);
1393                 if (!system->name)
1394                         goto out_free;
1395         } else
1396                 system->name = name;
1397
1398         system->filter = NULL;
1399
1400         system->filter = kzalloc(sizeof(struct event_filter), GFP_KERNEL);
1401         if (!system->filter)
1402                 goto out_free;
1403
1404         list_add(&system->list, &event_subsystems);
1405
1406         return system;
1407
1408  out_free:
1409         if (system->ref_count & SYSTEM_FL_FREE_NAME)
1410                 kfree(system->name);
1411         kfree(system);
1412         return NULL;
1413 }
1414
1415 static struct dentry *
1416 event_subsystem_dir(struct trace_array *tr, const char *name,
1417                     struct ftrace_event_file *file, struct dentry *parent)
1418 {
1419         struct ftrace_subsystem_dir *dir;
1420         struct event_subsystem *system;
1421         struct dentry *entry;
1422
1423         /* First see if we did not already create this dir */
1424         list_for_each_entry(dir, &tr->systems, list) {
1425                 system = dir->subsystem;
1426                 if (strcmp(system->name, name) == 0) {
1427                         dir->nr_events++;
1428                         file->system = dir;
1429                         return dir->entry;
1430                 }
1431         }
1432
1433         /* Now see if the system itself exists. */
1434         list_for_each_entry(system, &event_subsystems, list) {
1435                 if (strcmp(system->name, name) == 0)
1436                         break;
1437         }
1438         /* Reset system variable when not found */
1439         if (&system->list == &event_subsystems)
1440                 system = NULL;
1441
1442         dir = kmalloc(sizeof(*dir), GFP_KERNEL);
1443         if (!dir)
1444                 goto out_fail;
1445
1446         if (!system) {
1447                 system = create_new_subsystem(name);
1448                 if (!system)
1449                         goto out_free;
1450         } else
1451                 __get_system(system);
1452
1453         dir->entry = debugfs_create_dir(name, parent);
1454         if (!dir->entry) {
1455                 pr_warning("Failed to create system directory %s\n", name);
1456                 __put_system(system);
1457                 goto out_free;
1458         }
1459
1460         dir->tr = tr;
1461         dir->ref_count = 1;
1462         dir->nr_events = 1;
1463         dir->subsystem = system;
1464         file->system = dir;
1465
1466         entry = debugfs_create_file("filter", 0644, dir->entry, dir,
1467                                     &ftrace_subsystem_filter_fops);
1468         if (!entry) {
1469                 kfree(system->filter);
1470                 system->filter = NULL;
1471                 pr_warning("Could not create debugfs '%s/filter' entry\n", name);
1472         }
1473
1474         trace_create_file("enable", 0644, dir->entry, dir,
1475                           &ftrace_system_enable_fops);
1476
1477         list_add(&dir->list, &tr->systems);
1478
1479         return dir->entry;
1480
1481  out_free:
1482         kfree(dir);
1483  out_fail:
1484         /* Only print this message if failed on memory allocation */
1485         if (!dir || !system)
1486                 pr_warning("No memory to create event subsystem %s\n",
1487                            name);
1488         return NULL;
1489 }
1490
1491 static int
1492 event_create_dir(struct dentry *parent,
1493                  struct ftrace_event_file *file,
1494                  const struct file_operations *id,
1495                  const struct file_operations *enable,
1496                  const struct file_operations *filter,
1497                  const struct file_operations *format)
1498 {
1499         struct ftrace_event_call *call = file->event_call;
1500         struct trace_array *tr = file->tr;
1501         struct list_head *head;
1502         struct dentry *d_events;
1503         int ret;
1504
1505         /*
1506          * If the trace point header did not define TRACE_SYSTEM
1507          * then the system would be called "TRACE_SYSTEM".
1508          */
1509         if (strcmp(call->class->system, TRACE_SYSTEM) != 0) {
1510                 d_events = event_subsystem_dir(tr, call->class->system, file, parent);
1511                 if (!d_events)
1512                         return -ENOMEM;
1513         } else
1514                 d_events = parent;
1515
1516         file->dir = debugfs_create_dir(call->name, d_events);
1517         if (!file->dir) {
1518                 pr_warning("Could not create debugfs '%s' directory\n",
1519                            call->name);
1520                 return -1;
1521         }
1522
1523         if (call->class->reg && !(call->flags & TRACE_EVENT_FL_IGNORE_ENABLE))
1524                 trace_create_file("enable", 0644, file->dir, file,
1525                                   enable);
1526
1527 #ifdef CONFIG_PERF_EVENTS
1528         if (call->event.type && call->class->reg)
1529                 trace_create_file("id", 0444, file->dir,
1530                                   (void *)(long)call->event.type, id);
1531 #endif
1532
1533         /*
1534          * Other events may have the same class. Only update
1535          * the fields if they are not already defined.
1536          */
1537         head = trace_get_fields(call);
1538         if (list_empty(head)) {
1539                 ret = call->class->define_fields(call);
1540                 if (ret < 0) {
1541                         pr_warning("Could not initialize trace point"
1542                                    " events/%s\n", call->name);
1543                         return -1;
1544                 }
1545         }
1546         trace_create_file("filter", 0644, file->dir, call,
1547                           filter);
1548
1549         trace_create_file("format", 0444, file->dir, call,
1550                           format);
1551
1552         return 0;
1553 }
1554
1555 static void remove_event_from_tracers(struct ftrace_event_call *call)
1556 {
1557         struct ftrace_event_file *file;
1558         struct trace_array *tr;
1559
1560         do_for_each_event_file_safe(tr, file) {
1561                 if (file->event_call != call)
1562                         continue;
1563
1564                 remove_event_file_dir(file);
1565                 /*
1566                  * The do_for_each_event_file_safe() is
1567                  * a double loop. After finding the call for this
1568                  * trace_array, we use break to jump to the next
1569                  * trace_array.
1570                  */
1571                 break;
1572         } while_for_each_event_file();
1573 }
1574
1575 static void event_remove(struct ftrace_event_call *call)
1576 {
1577         struct trace_array *tr;
1578         struct ftrace_event_file *file;
1579
1580         do_for_each_event_file(tr, file) {
1581                 if (file->event_call != call)
1582                         continue;
1583                 ftrace_event_enable_disable(file, 0);
1584                 /*
1585                  * The do_for_each_event_file() is
1586                  * a double loop. After finding the call for this
1587                  * trace_array, we use break to jump to the next
1588                  * trace_array.
1589                  */
1590                 break;
1591         } while_for_each_event_file();
1592
1593         if (call->event.funcs)
1594                 __unregister_ftrace_event(&call->event);
1595         remove_event_from_tracers(call);
1596         list_del(&call->list);
1597 }
1598
1599 static int event_init(struct ftrace_event_call *call)
1600 {
1601         int ret = 0;
1602
1603         if (WARN_ON(!call->name))
1604                 return -EINVAL;
1605
1606         if (call->class->raw_init) {
1607                 ret = call->class->raw_init(call);
1608                 if (ret < 0 && ret != -ENOSYS)
1609                         pr_warn("Could not initialize trace events/%s\n",
1610                                 call->name);
1611         }
1612
1613         return ret;
1614 }
1615
1616 static int
1617 __register_event(struct ftrace_event_call *call, struct module *mod)
1618 {
1619         int ret;
1620
1621         ret = event_init(call);
1622         if (ret < 0)
1623                 return ret;
1624
1625         list_add(&call->list, &ftrace_events);
1626         call->mod = mod;
1627
1628         return 0;
1629 }
1630
1631 static struct ftrace_event_file *
1632 trace_create_new_event(struct ftrace_event_call *call,
1633                        struct trace_array *tr)
1634 {
1635         struct ftrace_event_file *file;
1636
1637         file = kmem_cache_alloc(file_cachep, GFP_TRACE);
1638         if (!file)
1639                 return NULL;
1640
1641         file->event_call = call;
1642         file->tr = tr;
1643         atomic_set(&file->sm_ref, 0);
1644         list_add(&file->list, &tr->events);
1645
1646         return file;
1647 }
1648
1649 /* Add an event to a trace directory */
1650 static int
1651 __trace_add_new_event(struct ftrace_event_call *call,
1652                       struct trace_array *tr,
1653                       const struct file_operations *id,
1654                       const struct file_operations *enable,
1655                       const struct file_operations *filter,
1656                       const struct file_operations *format)
1657 {
1658         struct ftrace_event_file *file;
1659
1660         file = trace_create_new_event(call, tr);
1661         if (!file)
1662                 return -ENOMEM;
1663
1664         return event_create_dir(tr->event_dir, file, id, enable, filter, format);
1665 }
1666
1667 /*
1668  * Just create a decriptor for early init. A descriptor is required
1669  * for enabling events at boot. We want to enable events before
1670  * the filesystem is initialized.
1671  */
1672 static __init int
1673 __trace_early_add_new_event(struct ftrace_event_call *call,
1674                             struct trace_array *tr)
1675 {
1676         struct ftrace_event_file *file;
1677
1678         file = trace_create_new_event(call, tr);
1679         if (!file)
1680                 return -ENOMEM;
1681
1682         return 0;
1683 }
1684
1685 struct ftrace_module_file_ops;
1686 static void __add_event_to_tracers(struct ftrace_event_call *call,
1687                                    struct ftrace_module_file_ops *file_ops);
1688
1689 /* Add an additional event_call dynamically */
1690 int trace_add_event_call(struct ftrace_event_call *call)
1691 {
1692         int ret;
1693         mutex_lock(&trace_types_lock);
1694         mutex_lock(&event_mutex);
1695
1696         ret = __register_event(call, NULL);
1697         if (ret >= 0)
1698                 __add_event_to_tracers(call, NULL);
1699
1700         mutex_unlock(&event_mutex);
1701         mutex_unlock(&trace_types_lock);
1702         return ret;
1703 }
1704
1705 /*
1706  * Must be called under locking of trace_types_lock, event_mutex and
1707  * trace_event_sem.
1708  */
1709 static void __trace_remove_event_call(struct ftrace_event_call *call)
1710 {
1711         event_remove(call);
1712         trace_destroy_fields(call);
1713         destroy_preds(call);
1714 }
1715
1716 static int probe_remove_event_call(struct ftrace_event_call *call)
1717 {
1718         struct trace_array *tr;
1719         struct ftrace_event_file *file;
1720
1721 #ifdef CONFIG_PERF_EVENTS
1722         if (call->perf_refcount)
1723                 return -EBUSY;
1724 #endif
1725         do_for_each_event_file(tr, file) {
1726                 if (file->event_call != call)
1727                         continue;
1728                 /*
1729                  * We can't rely on ftrace_event_enable_disable(enable => 0)
1730                  * we are going to do, FTRACE_EVENT_FL_SOFT_MODE can suppress
1731                  * TRACE_REG_UNREGISTER.
1732                  */
1733                 if (file->flags & FTRACE_EVENT_FL_ENABLED)
1734                         return -EBUSY;
1735                 /*
1736                  * The do_for_each_event_file_safe() is
1737                  * a double loop. After finding the call for this
1738                  * trace_array, we use break to jump to the next
1739                  * trace_array.
1740                  */
1741                 break;
1742         } while_for_each_event_file();
1743
1744         __trace_remove_event_call(call);
1745
1746         return 0;
1747 }
1748
1749 /* Remove an event_call */
1750 int trace_remove_event_call(struct ftrace_event_call *call)
1751 {
1752         int ret;
1753
1754         mutex_lock(&trace_types_lock);
1755         mutex_lock(&event_mutex);
1756         down_write(&trace_event_sem);
1757         ret = probe_remove_event_call(call);
1758         up_write(&trace_event_sem);
1759         mutex_unlock(&event_mutex);
1760         mutex_unlock(&trace_types_lock);
1761
1762         return ret;
1763 }
1764
1765 #define for_each_event(event, start, end)                       \
1766         for (event = start;                                     \
1767              (unsigned long)event < (unsigned long)end;         \
1768              event++)
1769
1770 #ifdef CONFIG_MODULES
1771
1772 static LIST_HEAD(ftrace_module_file_list);
1773
1774 /*
1775  * Modules must own their file_operations to keep up with
1776  * reference counting.
1777  */
1778 struct ftrace_module_file_ops {
1779         struct list_head                list;
1780         struct module                   *mod;
1781         struct file_operations          id;
1782         struct file_operations          enable;
1783         struct file_operations          format;
1784         struct file_operations          filter;
1785 };
1786
1787 static struct ftrace_module_file_ops *
1788 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1789 {
1790         /*
1791          * As event_calls are added in groups by module,
1792          * when we find one file_ops, we don't need to search for
1793          * each call in that module, as the rest should be the
1794          * same. Only search for a new one if the last one did
1795          * not match.
1796          */
1797         if (file_ops && mod == file_ops->mod)
1798                 return file_ops;
1799
1800         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1801                 if (file_ops->mod == mod)
1802                         return file_ops;
1803         }
1804         return NULL;
1805 }
1806
1807 static struct ftrace_module_file_ops *
1808 trace_create_file_ops(struct module *mod)
1809 {
1810         struct ftrace_module_file_ops *file_ops;
1811
1812         /*
1813          * This is a bit of a PITA. To allow for correct reference
1814          * counting, modules must "own" their file_operations.
1815          * To do this, we allocate the file operations that will be
1816          * used in the event directory.
1817          */
1818
1819         file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
1820         if (!file_ops)
1821                 return NULL;
1822
1823         file_ops->mod = mod;
1824
1825         file_ops->id = ftrace_event_id_fops;
1826         file_ops->id.owner = mod;
1827
1828         file_ops->enable = ftrace_enable_fops;
1829         file_ops->enable.owner = mod;
1830
1831         file_ops->filter = ftrace_event_filter_fops;
1832         file_ops->filter.owner = mod;
1833
1834         file_ops->format = ftrace_event_format_fops;
1835         file_ops->format.owner = mod;
1836
1837         list_add(&file_ops->list, &ftrace_module_file_list);
1838
1839         return file_ops;
1840 }
1841
1842 static void trace_module_add_events(struct module *mod)
1843 {
1844         struct ftrace_module_file_ops *file_ops = NULL;
1845         struct ftrace_event_call **call, **start, **end;
1846
1847         start = mod->trace_events;
1848         end = mod->trace_events + mod->num_trace_events;
1849
1850         if (start == end)
1851                 return;
1852
1853         file_ops = trace_create_file_ops(mod);
1854         if (!file_ops)
1855                 return;
1856
1857         for_each_event(call, start, end) {
1858                 __register_event(*call, mod);
1859                 __add_event_to_tracers(*call, file_ops);
1860         }
1861 }
1862
1863 static void trace_module_remove_events(struct module *mod)
1864 {
1865         struct ftrace_module_file_ops *file_ops;
1866         struct ftrace_event_call *call, *p;
1867         bool clear_trace = false;
1868
1869         down_write(&trace_event_sem);
1870         list_for_each_entry_safe(call, p, &ftrace_events, list) {
1871                 if (call->mod == mod) {
1872                         if (call->flags & TRACE_EVENT_FL_WAS_ENABLED)
1873                                 clear_trace = true;
1874                         __trace_remove_event_call(call);
1875                 }
1876         }
1877
1878         /* Now free the file_operations */
1879         list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
1880                 if (file_ops->mod == mod)
1881                         break;
1882         }
1883         if (&file_ops->list != &ftrace_module_file_list) {
1884                 list_del(&file_ops->list);
1885                 kfree(file_ops);
1886         }
1887         up_write(&trace_event_sem);
1888
1889         /*
1890          * It is safest to reset the ring buffer if the module being unloaded
1891          * registered any events that were used. The only worry is if
1892          * a new module gets loaded, and takes on the same id as the events
1893          * of this module. When printing out the buffer, traced events left
1894          * over from this module may be passed to the new module events and
1895          * unexpected results may occur.
1896          */
1897         if (clear_trace)
1898                 tracing_reset_all_online_cpus();
1899 }
1900
1901 static int trace_module_notify(struct notifier_block *self,
1902                                unsigned long val, void *data)
1903 {
1904         struct module *mod = data;
1905
1906         mutex_lock(&trace_types_lock);
1907         mutex_lock(&event_mutex);
1908         switch (val) {
1909         case MODULE_STATE_COMING:
1910                 trace_module_add_events(mod);
1911                 break;
1912         case MODULE_STATE_GOING:
1913                 trace_module_remove_events(mod);
1914                 break;
1915         }
1916         mutex_unlock(&event_mutex);
1917         mutex_unlock(&trace_types_lock);
1918
1919         return 0;
1920 }
1921
1922 static int
1923 __trace_add_new_mod_event(struct ftrace_event_call *call,
1924                           struct trace_array *tr,
1925                           struct ftrace_module_file_ops *file_ops)
1926 {
1927         return __trace_add_new_event(call, tr,
1928                                      &file_ops->id, &file_ops->enable,
1929                                      &file_ops->filter, &file_ops->format);
1930 }
1931
1932 #else
1933 static inline struct ftrace_module_file_ops *
1934 find_ftrace_file_ops(struct ftrace_module_file_ops *file_ops, struct module *mod)
1935 {
1936         return NULL;
1937 }
1938 static inline int trace_module_notify(struct notifier_block *self,
1939                                       unsigned long val, void *data)
1940 {
1941         return 0;
1942 }
1943 static inline int
1944 __trace_add_new_mod_event(struct ftrace_event_call *call,
1945                           struct trace_array *tr,
1946                           struct ftrace_module_file_ops *file_ops)
1947 {
1948         return -ENODEV;
1949 }
1950 #endif /* CONFIG_MODULES */
1951
1952 /* Create a new event directory structure for a trace directory. */
1953 static void
1954 __trace_add_event_dirs(struct trace_array *tr)
1955 {
1956         struct ftrace_module_file_ops *file_ops = NULL;
1957         struct ftrace_event_call *call;
1958         int ret;
1959
1960         list_for_each_entry(call, &ftrace_events, list) {
1961                 if (call->mod) {
1962                         /*
1963                          * Directories for events by modules need to
1964                          * keep module ref counts when opened (as we don't
1965                          * want the module to disappear when reading one
1966                          * of these files). The file_ops keep account of
1967                          * the module ref count.
1968                          */
1969                         file_ops = find_ftrace_file_ops(file_ops, call->mod);
1970                         if (!file_ops)
1971                                 continue; /* Warn? */
1972                         ret = __trace_add_new_mod_event(call, tr, file_ops);
1973                         if (ret < 0)
1974                                 pr_warning("Could not create directory for event %s\n",
1975                                            call->name);
1976                         continue;
1977                 }
1978                 ret = __trace_add_new_event(call, tr,
1979                                             &ftrace_event_id_fops,
1980                                             &ftrace_enable_fops,
1981                                             &ftrace_event_filter_fops,
1982                                             &ftrace_event_format_fops);
1983                 if (ret < 0)
1984                         pr_warning("Could not create directory for event %s\n",
1985                                    call->name);
1986         }
1987 }
1988
1989 #ifdef CONFIG_DYNAMIC_FTRACE
1990
1991 /* Avoid typos */
1992 #define ENABLE_EVENT_STR        "enable_event"
1993 #define DISABLE_EVENT_STR       "disable_event"
1994
1995 struct event_probe_data {
1996         struct ftrace_event_file        *file;
1997         unsigned long                   count;
1998         int                             ref;
1999         bool                            enable;
2000 };
2001
2002 static struct ftrace_event_file *
2003 find_event_file(struct trace_array *tr, const char *system,  const char *event)
2004 {
2005         struct ftrace_event_file *file;
2006         struct ftrace_event_call *call;
2007
2008         list_for_each_entry(file, &tr->events, list) {
2009
2010                 call = file->event_call;
2011
2012                 if (!call->name || !call->class || !call->class->reg)
2013                         continue;
2014
2015                 if (call->flags & TRACE_EVENT_FL_IGNORE_ENABLE)
2016                         continue;
2017
2018                 if (strcmp(event, call->name) == 0 &&
2019                     strcmp(system, call->class->system) == 0)
2020                         return file;
2021         }
2022         return NULL;
2023 }
2024
2025 static void
2026 event_enable_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2027 {
2028         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2029         struct event_probe_data *data = *pdata;
2030
2031         if (!data)
2032                 return;
2033
2034         if (data->enable)
2035                 clear_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
2036         else
2037                 set_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &data->file->flags);
2038 }
2039
2040 static void
2041 event_enable_count_probe(unsigned long ip, unsigned long parent_ip, void **_data)
2042 {
2043         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2044         struct event_probe_data *data = *pdata;
2045
2046         if (!data)
2047                 return;
2048
2049         if (!data->count)
2050                 return;
2051
2052         /* Skip if the event is in a state we want to switch to */
2053         if (data->enable == !(data->file->flags & FTRACE_EVENT_FL_SOFT_DISABLED))
2054                 return;
2055
2056         if (data->count != -1)
2057                 (data->count)--;
2058
2059         event_enable_probe(ip, parent_ip, _data);
2060 }
2061
2062 static int
2063 event_enable_print(struct seq_file *m, unsigned long ip,
2064                       struct ftrace_probe_ops *ops, void *_data)
2065 {
2066         struct event_probe_data *data = _data;
2067
2068         seq_printf(m, "%ps:", (void *)ip);
2069
2070         seq_printf(m, "%s:%s:%s",
2071                    data->enable ? ENABLE_EVENT_STR : DISABLE_EVENT_STR,
2072                    data->file->event_call->class->system,
2073                    data->file->event_call->name);
2074
2075         if (data->count == -1)
2076                 seq_printf(m, ":unlimited\n");
2077         else
2078                 seq_printf(m, ":count=%ld\n", data->count);
2079
2080         return 0;
2081 }
2082
2083 static int
2084 event_enable_init(struct ftrace_probe_ops *ops, unsigned long ip,
2085                   void **_data)
2086 {
2087         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2088         struct event_probe_data *data = *pdata;
2089
2090         data->ref++;
2091         return 0;
2092 }
2093
2094 static void
2095 event_enable_free(struct ftrace_probe_ops *ops, unsigned long ip,
2096                   void **_data)
2097 {
2098         struct event_probe_data **pdata = (struct event_probe_data **)_data;
2099         struct event_probe_data *data = *pdata;
2100
2101         if (WARN_ON_ONCE(data->ref <= 0))
2102                 return;
2103
2104         data->ref--;
2105         if (!data->ref) {
2106                 /* Remove the SOFT_MODE flag */
2107                 __ftrace_event_enable_disable(data->file, 0, 1);
2108                 module_put(data->file->event_call->mod);
2109                 kfree(data);
2110         }
2111         *pdata = NULL;
2112 }
2113
2114 static struct ftrace_probe_ops event_enable_probe_ops = {
2115         .func                   = event_enable_probe,
2116         .print                  = event_enable_print,
2117         .init                   = event_enable_init,
2118         .free                   = event_enable_free,
2119 };
2120
2121 static struct ftrace_probe_ops event_enable_count_probe_ops = {
2122         .func                   = event_enable_count_probe,
2123         .print                  = event_enable_print,
2124         .init                   = event_enable_init,
2125         .free                   = event_enable_free,
2126 };
2127
2128 static struct ftrace_probe_ops event_disable_probe_ops = {
2129         .func                   = event_enable_probe,
2130         .print                  = event_enable_print,
2131         .init                   = event_enable_init,
2132         .free                   = event_enable_free,
2133 };
2134
2135 static struct ftrace_probe_ops event_disable_count_probe_ops = {
2136         .func                   = event_enable_count_probe,
2137         .print                  = event_enable_print,
2138         .init                   = event_enable_init,
2139         .free                   = event_enable_free,
2140 };
2141
2142 static int
2143 event_enable_func(struct ftrace_hash *hash,
2144                   char *glob, char *cmd, char *param, int enabled)
2145 {
2146         struct trace_array *tr = top_trace_array();
2147         struct ftrace_event_file *file;
2148         struct ftrace_probe_ops *ops;
2149         struct event_probe_data *data;
2150         const char *system;
2151         const char *event;
2152         char *number;
2153         bool enable;
2154         int ret;
2155
2156         /* hash funcs only work with set_ftrace_filter */
2157         if (!enabled || !param)
2158                 return -EINVAL;
2159
2160         system = strsep(&param, ":");
2161         if (!param)
2162                 return -EINVAL;
2163
2164         event = strsep(&param, ":");
2165
2166         mutex_lock(&event_mutex);
2167
2168         ret = -EINVAL;
2169         file = find_event_file(tr, system, event);
2170         if (!file)
2171                 goto out;
2172
2173         enable = strcmp(cmd, ENABLE_EVENT_STR) == 0;
2174
2175         if (enable)
2176                 ops = param ? &event_enable_count_probe_ops : &event_enable_probe_ops;
2177         else
2178                 ops = param ? &event_disable_count_probe_ops : &event_disable_probe_ops;
2179
2180         if (glob[0] == '!') {
2181                 unregister_ftrace_function_probe_func(glob+1, ops);
2182                 ret = 0;
2183                 goto out;
2184         }
2185
2186         ret = -ENOMEM;
2187         data = kzalloc(sizeof(*data), GFP_KERNEL);
2188         if (!data)
2189                 goto out;
2190
2191         data->enable = enable;
2192         data->count = -1;
2193         data->file = file;
2194
2195         if (!param)
2196                 goto out_reg;
2197
2198         number = strsep(&param, ":");
2199
2200         ret = -EINVAL;
2201         if (!strlen(number))
2202                 goto out_free;
2203
2204         /*
2205          * We use the callback data field (which is a pointer)
2206          * as our counter.
2207          */
2208         ret = kstrtoul(number, 0, &data->count);
2209         if (ret)
2210                 goto out_free;
2211
2212  out_reg:
2213         /* Don't let event modules unload while probe registered */
2214         ret = try_module_get(file->event_call->mod);
2215         if (!ret) {
2216                 ret = -EBUSY;
2217                 goto out_free;
2218         }
2219
2220         ret = __ftrace_event_enable_disable(file, 1, 1);
2221         if (ret < 0)
2222                 goto out_put;
2223         ret = register_ftrace_function_probe(glob, ops, data);
2224         /*
2225          * The above returns on success the # of functions enabled,
2226          * but if it didn't find any functions it returns zero.
2227          * Consider no functions a failure too.
2228          */
2229         if (!ret) {
2230                 ret = -ENOENT;
2231                 goto out_disable;
2232         } else if (ret < 0)
2233                 goto out_disable;
2234         /* Just return zero, not the number of enabled functions */
2235         ret = 0;
2236  out:
2237         mutex_unlock(&event_mutex);
2238         return ret;
2239
2240  out_disable:
2241         __ftrace_event_enable_disable(file, 0, 1);
2242  out_put:
2243         module_put(file->event_call->mod);
2244  out_free:
2245         kfree(data);
2246         goto out;
2247 }
2248
2249 static struct ftrace_func_command event_enable_cmd = {
2250         .name                   = ENABLE_EVENT_STR,
2251         .func                   = event_enable_func,
2252 };
2253
2254 static struct ftrace_func_command event_disable_cmd = {
2255         .name                   = DISABLE_EVENT_STR,
2256         .func                   = event_enable_func,
2257 };
2258
2259 static __init int register_event_cmds(void)
2260 {
2261         int ret;
2262
2263         ret = register_ftrace_command(&event_enable_cmd);
2264         if (WARN_ON(ret < 0))
2265                 return ret;
2266         ret = register_ftrace_command(&event_disable_cmd);
2267         if (WARN_ON(ret < 0))
2268                 unregister_ftrace_command(&event_enable_cmd);
2269         return ret;
2270 }
2271 #else
2272 static inline int register_event_cmds(void) { return 0; }
2273 #endif /* CONFIG_DYNAMIC_FTRACE */
2274
2275 /*
2276  * The top level array has already had its ftrace_event_file
2277  * descriptors created in order to allow for early events to
2278  * be recorded. This function is called after the debugfs has been
2279  * initialized, and we now have to create the files associated
2280  * to the events.
2281  */
2282 static __init void
2283 __trace_early_add_event_dirs(struct trace_array *tr)
2284 {
2285         struct ftrace_event_file *file;
2286         int ret;
2287
2288
2289         list_for_each_entry(file, &tr->events, list) {
2290                 ret = event_create_dir(tr->event_dir, file,
2291                                        &ftrace_event_id_fops,
2292                                        &ftrace_enable_fops,
2293                                        &ftrace_event_filter_fops,
2294                                        &ftrace_event_format_fops);
2295                 if (ret < 0)
2296                         pr_warning("Could not create directory for event %s\n",
2297                                    file->event_call->name);
2298         }
2299 }
2300
2301 /*
2302  * For early boot up, the top trace array requires to have
2303  * a list of events that can be enabled. This must be done before
2304  * the filesystem is set up in order to allow events to be traced
2305  * early.
2306  */
2307 static __init void
2308 __trace_early_add_events(struct trace_array *tr)
2309 {
2310         struct ftrace_event_call *call;
2311         int ret;
2312
2313         list_for_each_entry(call, &ftrace_events, list) {
2314                 /* Early boot up should not have any modules loaded */
2315                 if (WARN_ON_ONCE(call->mod))
2316                         continue;
2317
2318                 ret = __trace_early_add_new_event(call, tr);
2319                 if (ret < 0)
2320                         pr_warning("Could not create early event %s\n",
2321                                    call->name);
2322         }
2323 }
2324
2325 /* Remove the event directory structure for a trace directory. */
2326 static void
2327 __trace_remove_event_dirs(struct trace_array *tr)
2328 {
2329         struct ftrace_event_file *file, *next;
2330
2331         list_for_each_entry_safe(file, next, &tr->events, list)
2332                 remove_event_file_dir(file);
2333 }
2334
2335 static void
2336 __add_event_to_tracers(struct ftrace_event_call *call,
2337                        struct ftrace_module_file_ops *file_ops)
2338 {
2339         struct trace_array *tr;
2340
2341         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2342                 if (file_ops)
2343                         __trace_add_new_mod_event(call, tr, file_ops);
2344                 else
2345                         __trace_add_new_event(call, tr,
2346                                               &ftrace_event_id_fops,
2347                                               &ftrace_enable_fops,
2348                                               &ftrace_event_filter_fops,
2349                                               &ftrace_event_format_fops);
2350         }
2351 }
2352
2353 static struct notifier_block trace_module_nb = {
2354         .notifier_call = trace_module_notify,
2355         .priority = 0,
2356 };
2357
2358 extern struct ftrace_event_call *__start_ftrace_events[];
2359 extern struct ftrace_event_call *__stop_ftrace_events[];
2360
2361 static char bootup_event_buf[COMMAND_LINE_SIZE] __initdata;
2362
2363 static __init int setup_trace_event(char *str)
2364 {
2365         strlcpy(bootup_event_buf, str, COMMAND_LINE_SIZE);
2366         ring_buffer_expanded = true;
2367         tracing_selftest_disabled = true;
2368
2369         return 1;
2370 }
2371 __setup("trace_event=", setup_trace_event);
2372
2373 /* Expects to have event_mutex held when called */
2374 static int
2375 create_event_toplevel_files(struct dentry *parent, struct trace_array *tr)
2376 {
2377         struct dentry *d_events;
2378         struct dentry *entry;
2379
2380         entry = debugfs_create_file("set_event", 0644, parent,
2381                                     tr, &ftrace_set_event_fops);
2382         if (!entry) {
2383                 pr_warning("Could not create debugfs 'set_event' entry\n");
2384                 return -ENOMEM;
2385         }
2386
2387         d_events = debugfs_create_dir("events", parent);
2388         if (!d_events) {
2389                 pr_warning("Could not create debugfs 'events' directory\n");
2390                 return -ENOMEM;
2391         }
2392
2393         /* ring buffer internal formats */
2394         trace_create_file("header_page", 0444, d_events,
2395                           ring_buffer_print_page_header,
2396                           &ftrace_show_header_fops);
2397
2398         trace_create_file("header_event", 0444, d_events,
2399                           ring_buffer_print_entry_header,
2400                           &ftrace_show_header_fops);
2401
2402         trace_create_file("enable", 0644, d_events,
2403                           tr, &ftrace_tr_enable_fops);
2404
2405         tr->event_dir = d_events;
2406
2407         return 0;
2408 }
2409
2410 /**
2411  * event_trace_add_tracer - add a instance of a trace_array to events
2412  * @parent: The parent dentry to place the files/directories for events in
2413  * @tr: The trace array associated with these events
2414  *
2415  * When a new instance is created, it needs to set up its events
2416  * directory, as well as other files associated with events. It also
2417  * creates the event hierachry in the @parent/events directory.
2418  *
2419  * Returns 0 on success.
2420  */
2421 int event_trace_add_tracer(struct dentry *parent, struct trace_array *tr)
2422 {
2423         int ret;
2424
2425         mutex_lock(&event_mutex);
2426
2427         ret = create_event_toplevel_files(parent, tr);
2428         if (ret)
2429                 goto out_unlock;
2430
2431         down_write(&trace_event_sem);
2432         __trace_add_event_dirs(tr);
2433         up_write(&trace_event_sem);
2434
2435  out_unlock:
2436         mutex_unlock(&event_mutex);
2437
2438         return ret;
2439 }
2440
2441 /*
2442  * The top trace array already had its file descriptors created.
2443  * Now the files themselves need to be created.
2444  */
2445 static __init int
2446 early_event_add_tracer(struct dentry *parent, struct trace_array *tr)
2447 {
2448         int ret;
2449
2450         mutex_lock(&event_mutex);
2451
2452         ret = create_event_toplevel_files(parent, tr);
2453         if (ret)
2454                 goto out_unlock;
2455
2456         down_write(&trace_event_sem);
2457         __trace_early_add_event_dirs(tr);
2458         up_write(&trace_event_sem);
2459
2460  out_unlock:
2461         mutex_unlock(&event_mutex);
2462
2463         return ret;
2464 }
2465
2466 int event_trace_del_tracer(struct trace_array *tr)
2467 {
2468         mutex_lock(&event_mutex);
2469
2470         /* Disable any running events */
2471         __ftrace_set_clr_event_nolock(tr, NULL, NULL, NULL, 0);
2472
2473         down_write(&trace_event_sem);
2474         __trace_remove_event_dirs(tr);
2475         debugfs_remove_recursive(tr->event_dir);
2476         up_write(&trace_event_sem);
2477
2478         tr->event_dir = NULL;
2479
2480         mutex_unlock(&event_mutex);
2481
2482         return 0;
2483 }
2484
2485 static __init int event_trace_memsetup(void)
2486 {
2487         field_cachep = KMEM_CACHE(ftrace_event_field, SLAB_PANIC);
2488         file_cachep = KMEM_CACHE(ftrace_event_file, SLAB_PANIC);
2489         return 0;
2490 }
2491
2492 static __init int event_trace_enable(void)
2493 {
2494         struct trace_array *tr = top_trace_array();
2495         struct ftrace_event_call **iter, *call;
2496         char *buf = bootup_event_buf;
2497         char *token;
2498         int ret;
2499
2500         for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
2501
2502                 call = *iter;
2503                 ret = event_init(call);
2504                 if (!ret)
2505                         list_add(&call->list, &ftrace_events);
2506         }
2507
2508         /*
2509          * We need the top trace array to have a working set of trace
2510          * points at early init, before the debug files and directories
2511          * are created. Create the file entries now, and attach them
2512          * to the actual file dentries later.
2513          */
2514         __trace_early_add_events(tr);
2515
2516         while (true) {
2517                 token = strsep(&buf, ",");
2518
2519                 if (!token)
2520                         break;
2521                 if (!*token)
2522                         continue;
2523
2524                 ret = ftrace_set_clr_event(tr, token, 1);
2525                 if (ret)
2526                         pr_warn("Failed to enable trace event: %s\n", token);
2527         }
2528
2529         trace_printk_start_comm();
2530
2531         register_event_cmds();
2532
2533         return 0;
2534 }
2535
2536 static __init int event_trace_init(void)
2537 {
2538         struct trace_array *tr;
2539         struct dentry *d_tracer;
2540         struct dentry *entry;
2541         int ret;
2542
2543         tr = top_trace_array();
2544
2545         d_tracer = tracing_init_dentry();
2546         if (!d_tracer)
2547                 return 0;
2548
2549         entry = debugfs_create_file("available_events", 0444, d_tracer,
2550                                     tr, &ftrace_avail_fops);
2551         if (!entry)
2552                 pr_warning("Could not create debugfs "
2553                            "'available_events' entry\n");
2554
2555         if (trace_define_common_fields())
2556                 pr_warning("tracing: Failed to allocate common fields");
2557
2558         ret = early_event_add_tracer(d_tracer, tr);
2559         if (ret)
2560                 return ret;
2561
2562         ret = register_module_notifier(&trace_module_nb);
2563         if (ret)
2564                 pr_warning("Failed to register trace events module notifier\n");
2565
2566         return 0;
2567 }
2568 early_initcall(event_trace_memsetup);
2569 core_initcall(event_trace_enable);
2570 fs_initcall(event_trace_init);
2571
2572 #ifdef CONFIG_FTRACE_STARTUP_TEST
2573
2574 static DEFINE_SPINLOCK(test_spinlock);
2575 static DEFINE_SPINLOCK(test_spinlock_irq);
2576 static DEFINE_MUTEX(test_mutex);
2577
2578 static __init void test_work(struct work_struct *dummy)
2579 {
2580         spin_lock(&test_spinlock);
2581         spin_lock_irq(&test_spinlock_irq);
2582         udelay(1);
2583         spin_unlock_irq(&test_spinlock_irq);
2584         spin_unlock(&test_spinlock);
2585
2586         mutex_lock(&test_mutex);
2587         msleep(1);
2588         mutex_unlock(&test_mutex);
2589 }
2590
2591 static __init int event_test_thread(void *unused)
2592 {
2593         void *test_malloc;
2594
2595         test_malloc = kmalloc(1234, GFP_KERNEL);
2596         if (!test_malloc)
2597                 pr_info("failed to kmalloc\n");
2598
2599         schedule_on_each_cpu(test_work);
2600
2601         kfree(test_malloc);
2602
2603         set_current_state(TASK_INTERRUPTIBLE);
2604         while (!kthread_should_stop())
2605                 schedule();
2606
2607         return 0;
2608 }
2609
2610 /*
2611  * Do various things that may trigger events.
2612  */
2613 static __init void event_test_stuff(void)
2614 {
2615         struct task_struct *test_thread;
2616
2617         test_thread = kthread_run(event_test_thread, NULL, "test-events");
2618         msleep(1);
2619         kthread_stop(test_thread);
2620 }
2621
2622 /*
2623  * For every trace event defined, we will test each trace point separately,
2624  * and then by groups, and finally all trace points.
2625  */
2626 static __init void event_trace_self_tests(void)
2627 {
2628         struct ftrace_subsystem_dir *dir;
2629         struct ftrace_event_file *file;
2630         struct ftrace_event_call *call;
2631         struct event_subsystem *system;
2632         struct trace_array *tr;
2633         int ret;
2634
2635         tr = top_trace_array();
2636
2637         pr_info("Running tests on trace events:\n");
2638
2639         list_for_each_entry(file, &tr->events, list) {
2640
2641                 call = file->event_call;
2642
2643                 /* Only test those that have a probe */
2644                 if (!call->class || !call->class->probe)
2645                         continue;
2646
2647 /*
2648  * Testing syscall events here is pretty useless, but
2649  * we still do it if configured. But this is time consuming.
2650  * What we really need is a user thread to perform the
2651  * syscalls as we test.
2652  */
2653 #ifndef CONFIG_EVENT_TRACE_TEST_SYSCALLS
2654                 if (call->class->system &&
2655                     strcmp(call->class->system, "syscalls") == 0)
2656                         continue;
2657 #endif
2658
2659                 pr_info("Testing event %s: ", call->name);
2660
2661                 /*
2662                  * If an event is already enabled, someone is using
2663                  * it and the self test should not be on.
2664                  */
2665                 if (file->flags & FTRACE_EVENT_FL_ENABLED) {
2666                         pr_warning("Enabled event during self test!\n");
2667                         WARN_ON_ONCE(1);
2668                         continue;
2669                 }
2670
2671                 ftrace_event_enable_disable(file, 1);
2672                 event_test_stuff();
2673                 ftrace_event_enable_disable(file, 0);
2674
2675                 pr_cont("OK\n");
2676         }
2677
2678         /* Now test at the sub system level */
2679
2680         pr_info("Running tests on trace event systems:\n");
2681
2682         list_for_each_entry(dir, &tr->systems, list) {
2683
2684                 system = dir->subsystem;
2685
2686                 /* the ftrace system is special, skip it */
2687                 if (strcmp(system->name, "ftrace") == 0)
2688                         continue;
2689
2690                 pr_info("Testing event system %s: ", system->name);
2691
2692                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 1);
2693                 if (WARN_ON_ONCE(ret)) {
2694                         pr_warning("error enabling system %s\n",
2695                                    system->name);
2696                         continue;
2697                 }
2698
2699                 event_test_stuff();
2700
2701                 ret = __ftrace_set_clr_event(tr, NULL, system->name, NULL, 0);
2702                 if (WARN_ON_ONCE(ret)) {
2703                         pr_warning("error disabling system %s\n",
2704                                    system->name);
2705                         continue;
2706                 }
2707
2708                 pr_cont("OK\n");
2709         }
2710
2711         /* Test with all events enabled */
2712
2713         pr_info("Running tests on all trace events:\n");
2714         pr_info("Testing all events: ");
2715
2716         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 1);
2717         if (WARN_ON_ONCE(ret)) {
2718                 pr_warning("error enabling all events\n");
2719                 return;
2720         }
2721
2722         event_test_stuff();
2723
2724         /* reset sysname */
2725         ret = __ftrace_set_clr_event(tr, NULL, NULL, NULL, 0);
2726         if (WARN_ON_ONCE(ret)) {
2727                 pr_warning("error disabling all events\n");
2728                 return;
2729         }
2730
2731         pr_cont("OK\n");
2732 }
2733
2734 #ifdef CONFIG_FUNCTION_TRACER
2735
2736 static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
2737
2738 static void
2739 function_test_events_call(unsigned long ip, unsigned long parent_ip,
2740                           struct ftrace_ops *op, struct pt_regs *pt_regs)
2741 {
2742         struct ring_buffer_event *event;
2743         struct ring_buffer *buffer;
2744         struct ftrace_entry *entry;
2745         unsigned long flags;
2746         long disabled;
2747         int cpu;
2748         int pc;
2749
2750         pc = preempt_count();
2751         preempt_disable_notrace();
2752         cpu = raw_smp_processor_id();
2753         disabled = atomic_inc_return(&per_cpu(ftrace_test_event_disable, cpu));
2754
2755         if (disabled != 1)
2756                 goto out;
2757
2758         local_save_flags(flags);
2759
2760         event = trace_current_buffer_lock_reserve(&buffer,
2761                                                   TRACE_FN, sizeof(*entry),
2762                                                   flags, pc);
2763         if (!event)
2764                 goto out;
2765         entry   = ring_buffer_event_data(event);
2766         entry->ip                       = ip;
2767         entry->parent_ip                = parent_ip;
2768
2769         trace_buffer_unlock_commit(buffer, event, flags, pc);
2770
2771  out:
2772         atomic_dec(&per_cpu(ftrace_test_event_disable, cpu));
2773         preempt_enable_notrace();
2774 }
2775
2776 static struct ftrace_ops trace_ops __initdata  =
2777 {
2778         .func = function_test_events_call,
2779         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
2780 };
2781
2782 static __init void event_trace_self_test_with_function(void)
2783 {
2784         int ret;
2785         ret = register_ftrace_function(&trace_ops);
2786         if (WARN_ON(ret < 0)) {
2787                 pr_info("Failed to enable function tracer for event tests\n");
2788                 return;
2789         }
2790         pr_info("Running tests again, along with the function tracer\n");
2791         event_trace_self_tests();
2792         unregister_ftrace_function(&trace_ops);
2793 }
2794 #else
2795 static __init void event_trace_self_test_with_function(void)
2796 {
2797 }
2798 #endif
2799
2800 static __init int event_trace_self_tests_init(void)
2801 {
2802         if (!tracing_selftest_disabled) {
2803                 event_trace_self_tests();
2804                 event_trace_self_test_with_function();
2805         }
2806
2807         return 0;
2808 }
2809
2810 late_initcall(event_trace_self_tests_init);
2811
2812 #endif