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