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